Conference Agenda

Point cloud registration is a key facilitator for scan alignment in mapping, autonomous driving, and robotic applications. Current pipelines increasingly adopt neural-based paradigms, where most research focuses on learning view-consistent descriptors for correspondence matching. Due to outliers, matching is typically followed by a geometric verification phase that assesses correspondences by enforcing distance or angular consistency to support transformation estimation. Although effective, this verification stage scales quadratically, creating a computational bottleneck that hampers efficient registration. More importantly, since matching and verification are usually optimized separately, the verification stage cannot guide the learned descriptors or foster their geometric awareness. To address both limitations, we introduce a novel end-to-end neural registration framework that unifies correspondence learning and verification within a single differentiable formulation. Specifically, we propose a new consistency-driven cross-attention module that dynamically correlates cross-scan neighborhoods to suppress inconsistent matches and reinforce inter-scan feature coherence. In doing so, it produces robust and discriminative descriptors without incurring the quadratic cost of explicit pairwise verification. Our formulation is readily applicable, and we demonstrate its seamless integration into the GeoTransformer and RoITr state-of-the-art architectures without additional supervision or post-processing. Results show that our method excels in challenging low-overlap scenarios, where competing methods often yield few correct correspondences or fail entirely. It consistently achieves superior inlier ratios and the lowest registration errors on 3DMatch, 3DLoMatch, and KITTI, improving registration recall by up to 2.6%. Beyond accuracy, it converges faster during training and achieves the quickest inference among state-of-the-art methods.

Combining terrestrial Mobile Mapping System (MMS) point clouds with aerial photogrammetric data offers a practical route to comprehensive 3D urban models that integrate street-level geometric detail with wide-area coverage. However, direct 3D-to-3D registration between these data sources often fails because of large differences in viewpoint, point density, scale, and scene composition. This study presents an orthoimage-based registration framework that reformulates cross-source alignment in the Bird's-Eye-View (BEV) domain. After removing transient objects and extracting ground-level points from the MMS cloud, the data are rasterised into a synthetic orthoimage aligned in resolution and projection with a geo-referenced Unmanned Aerial Vehicle (UAV) orthoimage. A learned dense matcher establishes image correspondences, which are geometrically verified and lifted to 3D for coarse alignment, followed by tile-wise point-to-plane Iterative Closest Point (ICP) refinement and global trajectory regularisation via robust factor-graph optimisation. The aligned MMS and UAV point clouds are then integrated through reliability-driven voxel-level fusion. Experiments on a 3.7km urban corridor in Seoul demonstrate that the proposed framework achieves a 3D root-mean-square error of 6.19cm, indicating that BEV-domain orthoimage matching combined with local 3D refinement and trajectory regularisation provides a viable approach for large-scale MMS-UAV registration in dense urban environments.

In recent years, backpack mobile mapping systems (MMS) have shown great promise for under-canopy forest mapping. These systems integrate cameras, LiDAR sensors, and Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) units to provide multi-modal geospatial data essential for modern forest applications that require both geometric and spectral information. However, transportation logistics and improper handling can degrade the system calibration. Moreover, canopy-induced GNSS signal outages will cause trajectory errors. The resulting misalignments between the image-LiDAR data necessitate the application of image–LiDAR registration. Such algorithms can be broadly classified as 2D-3D, 3D-3D, or 2D-2D, depending on the domain in which image-LiDAR features are identified. Due to the inherent modality differences, 2D–3D methods often struggle with feature matching. These methods typically require manual feature selection (Habib et al., 2005) or the availability of prominent features in urban environments (Liao et al., 2023). In contrast, 3D-3D methods rely on generating 3D image point clouds, which imposes strict requirements on image overlap (Yang et al., 2015). Although 2D–2D approaches are less demanding on image data (Hu et al., 2023), none have been applied in under-canopy forests, where establishing multi-modal correspondences remains challenging. To overcome these limitations, this study introduces a post-processing framework for automated image–LiDAR alignment enhancement for backpack MMS in forest environments. This method utilizes a 2D–2D image–LiDAR registration approach based on semantic tree-trunk features.

In order to enable photogrammetric tracking of objects from a computed tomography (CT) dataset with a multi-camera system, a transformation between the CT data space and a photogrammetric reference frame is required, typically based on control points. To achieve a robust and precise registration between CT and photogrammetric datasets, this work proposes a marker-based approach. The main goal is to use a marker model that allows straightforward segmentation and control point estimation in CT voxel space, while also supporting reliable and precise control point estimation in the photogrammetric images. As a proof-of-concept, spherical markers were investigated, since they allow centre estimation in both domains. In the CT data, marker centres were determined by intensity-based thresholding followed by sphere fitting, while in the photogrammetric data they were estimated by intensity-based thresholding, edge detection, circle fitting, and multi-image spatial intersection. Two different marker models were tested. The results show that the proposed method is feasible and yields sub-millimetre standard deviations of unit weight for both marker types. However, since a sufficient stochastic model is not yet available, the reported accuracy measures may be optimistic and should therefore be interpreted with caution. Future work will address these limitations, in particular uncertainty modelling as well as remaining lighting and contrast issues.

Scanned aerial images acquired with film cameras (hereafter referred to as historical images) over the past century is a unique source for deriving Digital Elevation Models (DEMs) and orthoimage to reconstruct past Earth’s surface and quantify long-term changes from glacier to landscape and urban development.

The Historical Structure-from-Motion (HSfM) pipeline (Knuth et al., 2023) currently represents the state of the art to fully automatically generate these historical DEMs. However, struggles with inconsistent image quality, distortions, distinct geometries and above all is based on the commercial software Metahape. Therefore, we aim to: (1) develop a fully open-source solution in COLMAPs environment, (2) integrate learned models in different SfM-steps to better handle the complex properties that come with historical imagery, and (3) compare our output against HSfM.

Our work is based on 180 historical aerial images acquired above the challenging terrain of Gran Campo Nevado Glacier. The results show that our photogrammetric workflow leads to a 0.26 px smaller mean reprojection error as well as roughly 9-times more tie-points for the sparse point cloud compared to the HSfM. The mean DEM difference with a reference DEM on stable terrain and the 95%-quantile DEM difference are also smaller in our experiments (0.71m vs. 10.10 m and 73.62 m vs. 99.03 m).

Further tests of our workflow include employing alternative models for feature extraction, matching, and dense reconstruction as well as evaluating multitemporal approaches (as adopted in Knuth et al., 2023) to enable a more representative comparison.

In knee arthroplasty, computer-assisted navigation enhances the accuracy of prosthesis placement. However, current methods rely on invasively drilled locators to track the knee position during surgery, prolonging the healing process. For this reason, research is focused on markerless approaches capable of determining knee orientation and transferring preoperative planning into the surgical environment. This work presents a trinocular multi-object 3D reconstruction system designed for intraoperative acquisition of the knee surface, providing a foundation for marker less navigation. Due to the scarcity of real surgical data with ground truth, a synthetic dataset was created using Blender to simulate optical image acquisition of a virtual knee model under controlled camera and lighting conditions. The dataset enables a systematic evaluation of how camera motion and viewpoint affect pose estimation and 3D reconstruction accuracy. The results demonstrate that moderate camera deflection between 15° and 25° achieve the best balance between accurate camera pose estimation and surface reconstruction quality. The work confirms the potential of trinocular SLAM for robust bone surface tracking while also identifying the limitations of synthetic data, such as the absence of real-world visual variability. These results form the basis for future work on 3D reconstruction during dynamic knee movements and their tracking, as well as on the integration of markerless optical navigation systems into surgery.

This study statistically analyzes fire events in the Atlantic Forest, seeking to understand their particularities in relation to the national scenario. The biome, historically pressured by deforestation, fragmentation, and anthropogenic activities, also suffers from agricultural, livestock, and accidental fires, which increase its vulnerability. The research used data from Censipam's Fire Panel, obtained by MODIS and VIIRS orbital sensors, considering records from 2020 onwards and specific sections for the Atlantic Forest. Variables such as area, severity, persistence, speed of expansion, number of outbreaks, Fire Radiative Power (FRP), and detections were analyzed. The results indicate that, compared to the national pattern, fires in the Atlantic Forest are less intense and shorter in duration, a phenomenon associated with higher humidity, landscape fragmentation, and management conditions. It is concluded that the dynamics of fire in the biome differ significantly from the national average, reinforcing the importance of regional monitoring and firefighting strategies aimed at preserving its ecological integrity.

The MODIS burned area product is widely used to extract ignition locations and delineate individual fires for wildfire probabilistic loss modeling. However, limited studies have systematically evaluated the accuracy of these derived fire events through detailed spatial and temporal comparisons with reference datasets. This study addresses this gap by developing a robust framework to assess the accuracy of MODIS-derived individual fires across the United States. In this study, the MODIS Collection 6 MCD64 burned area product was used to extract ignition locations and individual fire events using the Fire Events Delineation (FIRED) algorithm. A comprehensive evaluation framework was then implemented to assess the delineated fire events against the Monitoring Trends in Burn Severity (MTBS) reference dataset, accounting for both spatial overlap and temporal consistency. The results show that the proposed approach achieved an average Intersection over Union (IoU) score of 0.54, an F-score of 0.701, an overall accuracy of 0.77, a precision of 0.90, and a recall of 0.57. These metrics represent averages across the period 2001–2020. Collectively, the results highlight the strengths and limitations of the event detection system and provide a quantitative assessment of its performance. This comprehensive evaluation offers valuable insights into the reliability of MODIS-derived individual fire events and improves understanding of their suitability for wildfire probabilistic loss modeling and related applications.

Wildfires cause economic, social, and environmental consequences, as they affect ecosystems, public safety, biodiversity and natural resources. They pose challenges to various world regions, particularly Mediterranean areas such as Spain. Numerous fire prediction and detection systems were introduced to detect and predict fires as well as prevent their risks and damage. Statistical methods and classical machine learning models were often employed to estimate and predict fire risk, showing their efficiency in generating fire risk maps. However, they fail to accurately capture complex temporal and spatial characteristics related to fire ignition. To address this challenge, a novel Convolutional Neural Network (CNN) model, namely CFMap, was introduced for predicting and generating detailed wildfire risk maps covering Spain regions. Comprehensive analyses were performed using data between 2008 and 2024, including fire history, geographical location information, land usage features, human activity indices, topography data, meteorological features, and vegetation indices from Spain regions, collected from the IberFire dataset. CFMap showed a superior performance with an accuracy of 0.8028 ± 0.0440, an AUC (Area Under the Curve) of 0.9354 ± 0.0088, and an F1-score of 0.7787 ± 0.0623, outperforming classical machine learning methods (XGBoost, LightGBM, and RandomForest) and deep learning models including ResNet and a simple CNN. These results demonstrate its reliability in predicting fire events and generating monthly fire risk maps for different Spain regions. Consequently, it helps to identify high fire risk zones, improve fire management strategies, and efficiently deploy firefighting resources, thereby reducing the potential risk and impact of fires.

In the immediate aftermath of a hurricane, the rapid, accurate assessment of building damage is paramount for effective emergency response and the allocation of resources. Traditional methods of damage assessment, which rely on ground-based surveys, are often slow, hazardous, and subjective. While the advent of remote sensing (RS), through Unmanned Aerial Vehicles (UAVs) and the application of Convolutional Neural Networks (CNNs), has significantly advanced the automation of this process, these models operate on a pixel-level or object-level basis, failing to capture the inherent spatial relationships and contextual information within a disaster zone. Damage patterns are not spatially random; they exhibit strong spatial autocorrelation, a principle encapsulated by Tobler's First Law of Geography. This paper introduces a novel approach that leverages Graph Attention Networks (GATs) to explicitly model spatial dependencies when evaluating building damage. By representing damaged buildings and their surroundings as nodes and edges in a graph, our model can learn and weigh the influence of neighboring structures and the local environment when assessing their damage level. This spatially-aware methodology moves beyond simple image classification to a more holistic scene understanding. We evaluate the method on DoriaNET, a geo-referenced UAV dataset collected after Hurricane Dorian (2019) that provides masked building patches, GPS centroids, structural metadata, and ordinal FEMA/HAZUS-style damage labels. By incorporating spatial context via a graph-based framework, our GAT model achieves superior performance in building damage classification compared to state-of-the-art CNN-based approaches, producing more coherent and accurate damage maps better suited to real-world disaster management scenarios.

This work presents an innovative image-based method for measuring wind speed and direction in urban environment using video

footage. Wind dynamics are traditionally investigated at multiple spatial scales, including pollutant dispersion at the canopy level

(Allwine, 2000), architectural design and outdoor comfort at the building scale (Allard, 2012; Holst, 2011) and the convection heat

transfer coefficient ℎ [Wm-²K-1] used to define the boundary conditions of numerical simulations (Oke, 2017). In 1997, Gary Settles

showed that image measurement could provide non-invasive and high-resolution measurements of fluid motion. This paper presents a

method for extracting anemometric data from images at the urban scale. We process freely accessible videos from the internet in which

air masses are identified at the canopy level. Motion extraction technique is used to isolate elements of the video that are in motion.

This information is fed into an optical flow algorithm that estimates an apparent velocity in [pixels/frame]. To convert the data to

[km/h], the view’s perspective is considered to ensure the conversion is accurate across the entire image. Distance mapping is

performed by projecting the image onto a 3D model of the scene, and the camera's recording parameters are estimated by simulating

the illumination of the scene. The anemometric data obtained are evaluated in relation to meteorological data recorded at a nearby

weather station. Innovative and simple to implement, this approach provides estimates of wind speeds and directions that are both

reliable and directly usable for architectural design and climate studies.

Rapid urbanization and the loss of vegetative cover in Indian cities have raised serious concerns about environmental sustainability and public health. This study focuses on analyzing and forecasting Urban Heat Island (UHI) patterns in Jaipur, India, by examining both Surface UHI (SUHI) and Atmospheric UHI (AUHI). Using Google Earth Engine, the research integrates diverse spatio-temporal datasets—including Landsat-derived indices (such as LULC, NDVI, NDWI, NDBI, NDMI, albedo, and emissivity), geospatial features (building density, sky view factor, and population density), and meteorological data (air temperature, humidity, wind speed, and solar radiation) from 2000 to 2024—to train a Random Forest Regression model. The model demonstrated strong performance (R² = 0.806; RMSE = 0.059), surpassing linear and generalized additive models by effectively capturing complex, non-linear relationships. It also helped identify high-risk areas like Transport Nagar and Budhsinghpura. Projections for 2030 and 2035 indicate increasing heat stress, particularly in Jaipur’s expanding urban periphery. This GIS-integrated machine learning framework presents a replicable approach for UHI prediction in other fast-growing Indian cities.

Southwestern China’s complex terrain and climate make landslides frequent, especially along highways where numerous high, steep artificial slopes are formed during construction. These slopes often deform or fail within 1–2 rainy seasons due to intricate geology, severely affecting construction and infrastructure safety. An automated, real-time monitoring and early-warning system is therefore urgently needed.

Conventional techniques (leveling, GPS, crack meters) are limited by small coverage, low efficiency, high cost, and inability to detect regional or hidden deformations. Spaceborne InSAR offers wide-area, high-precision, all-weather monitoring but faces severe decorrelation noise from dense vegetation and atmospheric delay errors in mountainous regions.

This study developed advanced InSAR methods for artificial slopes along the Huali Highway (G4216) in Yunnan Province. Using TCPInSAR and >240 Sentinel-1 images (2015–2025), we retrieved surface deformation throughout pre-construction, construction, and post-construction phases. To overcome local challenges, two novel correction approaches were proposed: (1) a noise-reduction method based on spatial correlation estimation of deformation signals, effectively suppressing vegetation-induced decorrelation; and (2) an atmospheric correction technique using Singular Spectrum Analysis (SSA), significantly reducing delays caused by complex weather.

Results show the improved InSAR system successfully detected multiple deformation zones along the corridor and provided reliable early warnings for safety management. By addressing key technical bottlenecks, this work validates the practicality and effectiveness of advanced InSAR for automated slope stability monitoring in geologically and environmentally complex regions, offering valuable reference for similar large-scale infrastructure projects.

Meteorological control of L-band forest backscatter from the BorealScat-2 radar tower

How strongly do weather conditions imprint on L-band radar signals from forests at sub-daily time scales? This question is investigated using the BorealScat-2 tower experiment in the Svartberget Experimental Forest (northern Sweden). The system acquires fully polarimetric, tomographic radar data at P, UHF and L band every 30 minutes, providing height-resolved backscatter profiles from the ground, through the trunk zone, into the upper canopy. Within the shared footprint, an ICOS flux mast delivers continuous measurements of CO₂, water vapour and energy fluxes, together with radiation, vapour pressure deficit (VPD), temperature, wind and precipitation. Sap-flow sensors, dendrometers and soil water probes further characterise water storage and transport in trees and soils, offering an unusually detailed description of forest water dynamics.

The study will focus on L-band backscatter during late spring and summer, quantifying how diurnal amplitude, phase and vertical centre-of-mass in different height zones and polarisations relate to VPD, temperature, radiation and rainfall. It will specifically assess the relative roles of atmospheric demand, canopy wetness and soil water status in driving sub-daily L-band variability, and examine differences between co- and cross-polarised channels and between structural layers. Overall, the study aims to provide process-based insight into how specific meteorological drivers control sub-daily L-band radar variability in boreal forests, supporting the interpretation and modelling of future vegetation radar missions.

The digital elevation model (DEM) based on synthetic aperture radar interferometry (SAR, InSAR) technology have become an important data source for large-scale topographic mapping, but their characteristics vary with satellite systems and methodologies. In this paper, we conduct the cross-validation for the first time to compare the LuTan-1 raw DEM (LT-1 RDEM) and GaoFen-7 (GF-7) satellite laser altimetry data. Besides, we compared the penetration capabilities of SAR satellites including C-band SRTM and X-band TanDEM-X. The optically derived ZiYuan-3 (ZY-3) DEM was also included for multi-source cross-validation. Taking Guyuan County, Hebei Province, China (including four landform types: plains, tablelands, hills, and mountains) as the study area, we introduced GF-7 laser altimetry points (LAPs) as the verification benchmark to cross-validate the elevation accuracy of LT-1 RDEM, SRTM, TanDEM-X DEM (TanDEM), and ZY-3 DEM. The results indicate that: (1) Topographic relief has a significant impact on accuracy, and the RMSE of the DEMs in the study area generally increases sequentially with the intensification of topographic relief; (2) Benefiting from the 10-meter spatial resolution, LT-1 RDEM performs best in detail representation; (3) In terms of mean height error, LT-1 RDEM exhibits a general negative bias, confirming the stronger penetration capability of the L-band, and its elevation values may be closer to the true ground surface; (4) The RMSE of LT-1 RDEM in the study area is 1.958m, slightly larger than TanDEM’s 1.65m, but in fact, the accuracy of TanDEM as a digital surface model (DSM) may be systematically overestimated by laser altimetry data.

This study evaluates the operational capability of the Chinese LuTan-1 (LT-1) L-band SAR constellation for monitoring the Hong Kong–Zhuhai–Macao Bridge (HZMB), a representative sea-crossing bridge under a complex subtropical marine climate. Leveraging the advantages of L-band SAR—including strong resistance to decorrelation and a spatial resolution of up to 3 meters—we applied the Small Baseline Subset (SBAS) technique to 47 ascending and descending orbital images. To the best of our knowledge, this represents one of the first comprehensive deformation studies of the HZMB using the LT-1 constellation. A key aspect of our methodology is the cross-validation between multi-orbit datasets, which confirmed both the reliability of the measurements and the complementary distribution of coherent points due to SAR imaging geometry. The results indicate overall structural stability of the HZMB, with the maximum deformation localized at the Jianghai Navigation Bridge, showing a Line-of-Sight (LOS) displacement rate of –4.3 mm/yr. In contrast, the two artificial islands exhibited minor deformation, with LOS rates not exceeding –3.0 mm/yr. These findings validate LT-1 as a powerful and reliable tool for the operational health monitoring of large-scale coastal infrastructure.

LuTan-1 (LT-1) satellites have been launched for about 4 years. About 771,282 images have been distributed to the users of China till 29th October, 2025. Main application purpose of LT-1 is geohazard monitoring and geoinformation production. Interferometric capability is the primary consideration for LT-1. In this paper, we assessed the interferometric applications in the natural resource monitoring industry. First, we overviewed the status of LT-1, the main interferometric products were introduced as S2A, S2B, S3A, S4A, S5A, S5B and S5C. They are geometrically calibrated single look complex (SLC) image, interferometrically calibrated SLC, differential interferometric synthetic aperture radar (SAR, InSAR, DInSAR) products, stacking, MTInSAR, digital orthorectified image, and digital surface model, respectively. S2A are generated after geometric calibration, the geometric accuracy is about 1.53 after calibration. The baseline is then calibrated for helix bistatic formation data and generate S2B whose accuracy is better than 0.96 m. S3A, S4A and S5A are all used for deformation monitoring, the accuracy values of them are 2.7 mm, 8.6 mm/yr, and 3.7 mm. Geometric accuracy of S5B is 12.5 m, and the height accuracy of S5C is better than 4.7 m. More than 330 geohazards were detected in Guangdong province. The geohazards recognition rate in the field working stage increase from 28% to 47.24%. Even a prediction has been made to avoid disaster for a family and saved 3 people. The application effectiveness has been validated through those years.

The influence of soil moisture change on interference phase information is fully taken into account for accurate deformation monitoring in this research. Especially the effects have more prominent contribution to L-band SAR data. In order to obtain high-precision surface deformation information over agricultural area, the interference phase component caused by soil moisture change should be considered, and the optimal processing of interference phase information is achieved. The reliable interference phase information characterizing the surface deformation details is obtained, thus the natural surface deformation information with high precision can be achieved. Firstly, the penetration depth of different band SAR for agricultural soil was analyzed and simulated. And the sensitivity between penetration depth variation and different band SAR signals were discussed. The fact of soil moisture changes for interferometric phase contribution is confirmed, which provided the foundation for reliable deformation montoring considering the soil moisture variation effects, especially for L-band SAR data. The periodic irrigation for the wheat fields will induce soil moisture variation, which may result in the penetration depth change for radar electromagnetic wave. Therefore, the phase component was derived by the variation of soil moisture over the wheat fields. Multi-temporal Lutan-1 SAR data were acquired over ShanDong agricultural plain in China. The obvious ‘deformation details’ induced by the soil moisture change were acquired over the agricultural area, which demonstrated the effect of soil moisture variation for interference phase. Therefore, the accurate deformation details over agricultural area can be obtained by the combination of soil moisture information.

Multi-temporal remote sensing is crucial for crop monitoring, but existing mapping methods struggle with incomplete time series due to data missingness. Current models often assume consistent data, leading to performance degradation when faced with irregular or missing observations. To address this, we propose an enhanced approach combining random temporal masking with neural Ordinary Differential Equation (ODE) optimization, designed to be embedded into existing models.

Our method first employs a random temporal masking strategy during training, forcing the model to learn effective temporal dependencies from sparse, incomplete sequences, thereby boosting its adaptability to diverse missing data scenarios. Second, a time-smoothing regularization term, based on neural ODE, guides the model to learn a continuous, smooth feature trajectory from discrete observations, effectively mitigating temporal inconsistencies and abrupt fluctuations caused by missing data. We also incorporate sine-cosine positional encoding with slight perturbations for precise time representation.

We integrated our approach into the state-of-the-art TSViT model and evaluated it on the PASTIS dataset. Experiments show that while the original TSViT’s accuracy (OA and mIoU) sharply declines with increasing missing frames, our enhanced model maintains significantly better performance. At 80% missing data, our method improves OA by approximately 8% and mIoU by about 12% compared to the baseline. Qualitative results further demonstrate our model’s ability to preserve coherent, smooth spatiotemporal predictions, enhancing robustness and generalization in real-world applications.

Real-world agricultural distributions often suffer from severe class imbalance, typically following a long-tailed distribution. Labeled datasets for crop-type classification are inherently scarce and remain costly to obtain. When working with such limited data, training sets are frequently constructed to be artificially balanced—in particular in the case of few-shot learning—failing to reflect real-world conditions. This mismatch induces a shift between training and test label distributions, degrading real-world generalization. To address this, we propose Dirichlet Prior Augmentation (DirPA), a novel method that simulates an unknown label distribution skew of the target domain proactively during model training. Specifically, we model the real-world distribution as Dirichlet-distributed random variables, effectively performing a prior augmentation during few-shot learning. Our experiments show that DirPA successfully shifts the decision boundary and stabilizes the training process by acting as a dynamic feature regularizer.

Accurate discrimination of cereal crops in heterogeneous agroecosystems requires methods that integrate both spectral and temporal information. This study proposes a compact spectral–temporal framework that combines Optimal Hyperspectral Narrowbands (OHNB) selected from EnMAP imagery using a Spectral Attention Module (SAM) with a Dynamic Time Warping (DTW)-derived phenological distance computed from Sentinel-2 EVI time series. The analysis was conducted in the Saïss region of Morocco, one of the country’s major cereal-producing areas.

SAM identified 29 physiologically meaningful narrowbands spanning the visible, red-edge, near-infrared, and shortwave-infrared regions (429–2438 nm), capturing key pigment, structural, and moisture-related vegetation properties. EVI time series were preprocessed through 10-day median compositing, linear interpolation, and Savitzky–Golay smoothing to generate stable phenological profiles. DTW quantified the temporal similarity of each field’s EVI trajectory to a cereal reference curve, producing a phenology-driven distance feature.

Three classifiers—Random Forest, SVM, and TabPFN—were evaluated under a nested standard and spatial cross-validation strategy. Using only hyperspectral bands, SVM and TabPFN achieved the highest accuracies (ROC-AUC = 0.95–0.93). Incorporating the DTW feature consistently improved performance under spatial CV, especially for RF (ROC-AUC increase: 0.89→0.91), and reduced the performance gap between validation schemes.

Overall, the fusion of SAM-selected hyperspectral bands with DTW-based phenological information enhanced spatial robustness and improved discrimination between cereal and non-cereal fields. The proposed approach offers an efficient and transferable solution for operational crop mapping in semi-arid agricultural landscapes.

Mapping banana crops in heterogeneous tropical landscapes remains challenging due to spectral similarity with surrounding vegetation, fragmented smallholder systems, and complex land-use mosaics. This study applies a deep learning approach, using a U-Net model, on high spatial resolution CBERS-4A imagery to map banana crops in Brazil’s Ribeira Valley, a subtropical region with high rainfall and heterogeneous land cover. Reference data were created through manual interpretation of satellite imagery supported by field knowledge. Representative image tiles were selected and divided into smaller patches for model training, validation, and testing. The U-Net model was trained with standard optimization techniques and evaluated using common semantic segmentation metrics. On the validation set, it achieved strong performance (accuracy 0.91, F1-score 0.84, AUC-ROC 0.96, AUC-PR 0.92). Performance was maintained or improved on the independent test set (accuracy 0.91, F1-score 0.86, AUC-ROC 0.97, AUC-PR 0.93), indicating good generalization. with high agreement between predicted and reference data. Most errors occurred at boundaries between crops and natural vegetation. Additional validation using official agricultural statistics confirmed consistency at the municipal scale. The approach demonstrates that high-resolution imagery combined with deep learning can effectively map banana crops in the region and offers a promising tool for agricultural monitoring and land-use planning in complex environments. The code, trained models, and data are publicly available at https://github.com/hnbendini/banana-unet-mapping.

This study is based on the best crop classification result generated by the proposed unsupervised Machine Learning (ML) method in Li et al., 2025a, using the spectral indices calculated by the formula with spectral bands from Sentinel-2 image products, Normalized Difference Vegetation Index (NDVI), Soil Adjusted Vegetation Index (SAVI), Enhanced Vegetation Index (EVI) and Normalized Difference Moisture Index (NDMI). The patterns and characteristics of these spectral indices, across arable fields with different crop types following the winter growing seasons, have not yet been analyzed in detail. This research aims to provide a comprehensive study of each input spectral index and its impact on the crop classification model. Each spectral index is analyzed across a series of crop fields, using Sentinel-2 images, carefully selected to follow the patterns of winter crop phenology, and the results of unsupervised classification for each crop type in Norfolk, UK are successfully generated and analyzed. The different growing rates between winter barley and wheat have been classified found on a monthly basis using Sentinel-2 RGB images and thus the images during the harvest time, May and June, can support crop classifications. Wild grasses or other plants on the fields led to some crop misclassification from November to March in the Sentinel-2 RGB images. Similarity between winter barley and wheat and the different sowing time among the same type of crop also led to misclassification. In future these misclassifications could be avoided through better understanding of the relation between spectral indices and crop planting cycles.

Current pest monitoring relies on labor-intensive manual scouting, often leading to preventive insecticide use, highlighting the need for automated surveillance. This study presents low-cost RGB camera sensors integrated with edge artificial intelligence (AI) for real-time aphid detection, enabling timely and targeted interventions. Using field images, we trained the YOLO11-n model and evaluated its performance under commercial farming conditions, achieving an average precision of 85 % for apterous aphids. The complex structure of lettuce, with overlapping leaves and shaded areas, limits detection accuracy, particularly for nymphal stages. Nevertheless, these results pave the way for affordable precision agriculture solutions to sustainably improve pest management.

We present a residual-aware, multi-sensor 3-D coseismic displacement decomposition applied to the 2025 Mw 7.7 Myanmar earthquake. The workflow combines multi-track Sentinel-1 SAR pixel offsets (range and azimuth) with Sentinel-2 optical pixel offsets, using only the north–south component where the signal clearly exceeds the optical noise level.

The key innovation is to handle sensor- and mosaicking-related residuals within a robust inversion framework rather than as ad hoc preprocessing. Strip-wise and inter-track trends are removed by MAD–Tukey IRLS plane fitting that suppresses long-wavelength orbital and viewing-geometry errors while preserving sharp near-fault steps in overlap zones. A residual-aware weighted least-squares inversion is then performed per pixel to recover east–west, north–south and vertical displacements and their fault-parallel projection. The resulting fields provide spatially continuous, cross-sensor-consistent constraints on fault-parallel slip along this exceptionally long rupture.

In Kolkata, potential land subsidence occurred primarily due to excessive groundwater extraction, which has been one of the major environmental crises, along with rapid urbanization and soft soil characteristics. This study investigates Kolkata's land surface deformation patterns from 2017 to 2023 using Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) technology. The study comprehensively analyzes deformation scenarios from 2017 to 2022; additionally, a detailed examination of the 2023 deformation scenario reveals continued trends and localized changes in subsidence patterns. The result shows that the mean ground velocity between 2017 and 2022 varies between -2.8 and -5.5 mm/year, and the area under the subsidence zone shows an increasing trend. Predictive models for 2024 and 2025 are developed based on historical data, providing forecasts of future subsidence trends. The prediction indicates that in 2024, the area under the high deformation class will be relatively higher compared with 2025. Spatial association analyses explore correlations between subsidence patterns of different years in Kolkata. The findings of this study may facilitate the assessment of the possible effects of ground-level movement on resource management, safety, and economics in this densely populated city.

Rapid and accurate detection of landslides after extreme climate events, such as heavy rainfalls or hurricanes, is essential for hazard response and mitigation. Traditional mapping methods rely on manual interpretation or labelled datasets, limiting scalability. This paper presents an integrated workflow combining unsupervised autoencoder-based + KMeans change detection and deep learning semantic segmentation to improve landslide identification in Western North Carolina following Hurricane Helene (September 2024). The approach leverages Planetscope RGB-NIR imagery at 3 m spatial resolution and North Carolina Department of Environmental Quality post-event landslide inventory points. The unsupervised autoencoder extracts latent features and highlights change zones, while segmentation models such as UNet learn spatial–contextual patterns from semi-automated labels. Results demonstrate high detection accuracy with segmentation models achieving strong overlap with ground-truth inventories and minimal false positives with an F1-score of 92%. This hybrid pipeline bridges rapid unsupervised detection and precise pixel-level segmentation, enabling scalable, near-real-time landslide mapping.

Rapid and reliable landslide mapping is critical for post-disaster response, yet Synthetic Aperture Radar (SAR)-based detection remains challenging due to speckle noise, geometric distortions, and complex terrain. This study develops an operational post-event landslide extraction framework using a UNet segmentation architecture with multimodal geospatial data fusion. High-resolution COSMO- SkyMed SAR imagery is combined with terrain representations derived from Digital Elevation Models (DEM), Red Relief Image Maps (RRIM), and rainfall indices to evaluate the contribution of complementary geospatial information to segmentation performance.

Experiments were conducted across three major landslide-triggering events in Japan (Kyushu, Hokkaido, and Kumamoto), comparing SAR-only and multimodal configurations. Results demonstrate that integrating terrain information and rainfall data improves landslide detection performance compared with SAR-only inputs. RRIM consistently outperforms DEM as a topographic descriptor, particularly in steep or heterogeneous terrain, while rainfall information provides moderate gains in recall. Boundary-based metrics further indicate improved geometric fidelity of mapped landslides when multimodal inputs are incorporated. The framework requires only a single post-event SAR acquisition supplemented with publicly available ancillary datasets, enabling rapid and scalable generation of landslide inventories without reliance on pre-disaster imagery. These findings establish a reproducible baseline for SAR-driven landslide segmentation and highlight the potential of multimodal geospatial data fusion for operational disaster response and hazard monitoring.

In this study, we integrated information from the French avalanche database, high-resolution digital elevation models (DEMs), and Sentinel-1 SAR images to model avalanche extents for events occurring across three distinct time periods in three French massifs. The modelled avalanche extents were compared with manually delineated polygons mapped over SAR RGB composites generated using the principle applied in colour-based change detection algorithms. The comparison revealed a strong correspondence between the two independent approaches, with IoU values ranging from 0.42 to 0.47 and F1 scores between 0.58 and 0.63 across the different massifs. We further analyzed the distribution of SAR backscatter values in pre- and post-event images across different zones of the avalanche paths. The results indicated that a fixed 3 dB threshold would most likely be insufficient to capture the complete avalanche extent, as certain zones exhibited backscatter increases of less than 3 dB in post-event SAR imagery. As a result, a multi-threshold approach based on different avalanche zones is recommended. Finally, we assessed the potential of Sentinel-2 optical imagery to detect surface changes and characterize the physical behaviour of avalanche-affected paths following intense avalanche

events. However, the results were inconsistent, exhibiting the expected trends in one study area but nearly opposite patterns in the other, indicating that the integration of optical data for automated avalanche mapping may not always be reliable.

To meet the growing demand for 3D digital map applications and to better understand the multi-level spatial structure of cities, some cities have implemented citywide 3D digital map programs. In 3D digital map production, vehicle-mounted mobile surveying is a key component. Drawing with a practical project, this paper proposes a technical scheme for road data acquisition and processing based on the SSW VMMS (Vehicle-mounted Mobile Mapping System). Through integrated processing steps, including combined navigation solution, point cloud correction, image coordinate calculation, image deblurring, point cloud coloring, point cloud denoising, and Orbit GT data preparation, the rapid production of colored point cloud data with georeferenced coordinates, 360° panoramic image data, and individual image data is achieved. A technical scheme suitable for 3D digital map production along urban roads was developed and validated. The results produced by this scheme have passed inspection and acceptance, and were released to the public free of charge as the first batch of visualized 3D map data on the Common Spatial Data Infrastructure Portal (portal.csdi.gov.hk), receiving widespread attention and positive recognition from various sectors of society. This scheme not only promotes the broader application of the SSW VMMS but also provides effective reference for similar urban vehicle-mounted mobile mapping projects.

We present an automatic algorithm for graph-based road network extraction from remote sensing images. While existing works mostly focus on improving accuracy, we address the problem of the geometric quality of the output graphs. The state-of-the-art methods largely overlook this aspect by generating graphs without strong geometric guarantees, regularity preservation and low-complexity, which, ultimately, reduces their impact in many application scenarios. Our algorithm relies upon foundation models that analyze road networks with pixel-based representations, as well as geometric algorithms and data structures in charge of connecting geometric primitives into planar graphs. This hybrid strategy allows us to strongly enforce the geometric quality of the output graphs while bringing a high level of generalization. We show the potential of our algorithm and its advantages over existing methods on two datasets commonly-used in the field using both the conventional accuracy metrics and new metrics introduced to measure the geometric quality of the output graphs.

High-resolution monitoring of road infrastructure is essential for the early detection of geomorphological instabilities such as landslides and erosion. This study evaluates the performance of handled MMS under different vehicle-mounted configurations: a 2-meter survey pole versus a suction-cup mount, and varying acquisition speeds (10 and 20 km/h). Furthermore, a GNSS-denied scenario was simulated to test the robustness of SLAM-based processing. Initial results revealed significant geometric discrepancies (double-points artifacts and drift), particularly in the SLAM-only and high-speed datasets. To address this, an automated segment-based refinement workflow was developed using a ICP algorithm. The refinement successfully reduced the standard deviation to the level of the point cloud´s mean point spacing (5 cm). Comparative multitemporal analysis against UAV-LiDAR reference data confirms that the proposed refinement renders even SLAM-processed data viable for detecting centimetric terrain displacements. The findings demonstrate that while suction-cup mounting at 10 km/h is optimal, algorithmic refinement allows for reliable road slopes monitoring and change detection across all tested configurations

Underground utilities such as water pipelines and sewers are critical for urban systems, yet their management is challenging due to limited accessibility and uncertain positional data. Current inspection practices rely on robotic crawlers equipped with CCTV cameras or man-entry inspections, enabling visual documentation of structural conditions but lacking accurate georeferencing of internal points. Advanced solutions relying on panoramic imaging and IMUs offer partial 3D measurements and trajectory estimation, though accuracy remains limited by drift and environmental variability.

This study investigates the feasibility of multi-camera photogrammetry for mapping pipelines and confined underground environments and improving positional accuracy. Preliminary experiments were conducted using the Atom-Ant3D system on two test sets: (i) five pipelines of varying materials (concrete, PVC, fiberglass) and diameters (60–110 cm); and (ii) a 1.3 km water-distribution tunnel (~2 m diameter) prepared with 28 fixed targets measured via total station for accuracy evaluation. Data were acquired using robotic and handheld configurations and processed through two workflows: Structure-from-Motion (SfM) and multi-view V-SLAM.

Accuracy assessment focused on the tunnel test, comparing unconstrained and constrained trajectories against a reference solution. Results provide insights into the potential of photogrammetric approaches for precise pipeline reconstruction and georeferencing, supporting improved subsurface utility management and planning.

3D multi-view, multi-object tracking (3D MV-MOT) makes use of multiple cameras to reduce the number of missed detections and to mitigate occlusions. Most current 3D MV-MOT methods suffer from information loss when associating 3D locations with 2D image features via a 3D-to-2D projection, as they use a discrete grid in 3D and sample image features only at the projected centers of each grid cell. Thus, all other feature information is lost. An additional information loss commonly arises during cross-view aggregation when applying max or average pooling: these methods either overemphasize a single view or treat conflicting views, that depict different entities, e.g., due to occlusions, equally.

In this work, we introduce two novel modules for 3D MV-MOT, employed to pedestrian tracking, that target these limitations:

(i) VoxROI aggregates all image features that fall within the bounding box around a voxel's projection into each respective image, instead of only sampling features at the projected voxel center.

(ii) SimFuse aggregates per-view voxel features into one coherent feature representation per voxel, using similarity weights computed from re-identification (Re-ID) features.

Subsequently, they are used to measure cross-view identity similarity. Views with higher Re-ID feature similarity receive larger weights, while inconsistent views are suppressed.

Experimental results on the WildTrack dataset confirm our method's effectiveness for multi-view pedestrian detection and tracking, reaching, and in particular in cross-view scenarios improving, the general state-of-the-art. The approach maintains strong performance across different camera configurations, demonstrating its generalization capability when training and testing on different camera setups.

With the rapid growth of urban low-altitude applications, traditional airspace management approaches based on simple altitude limits and static no-fly zones can no longer meet the demands of high-density and highly dynamic operations. To address this issue, this study proposes a fine-grained gridding method for urban low-altitude airspace with dynamic event response and vertical flight corridor constraints. First, a unified three-dimensional grid model is constructed on the basis of an urban 3D digital twin platform, and the grid scale and update cycle are determined by jointly considering clearance requirements and safety separation. Second, a method for injecting static and dynamic attributes is established to achieve the unified representation and continuous updating of terrain, buildings, no-fly and restricted zones, wind fields, temporary restrictions, as well as occupancy and release information within the grid. Third, fixed-geometry and dynamically open vertical flight corridors are designed to support controlled cross-layer flight transitions and reduce the risk of vertical conflict propagation. An experimental system is developed using a typical high-density urban area in Yuehai Subdistrict, Nanshan District, Shenzhen, as the case study. The results show that the proposed method can achieve stable spatial discretization, accurate attribute loading and updating, and clear organization of cross-layer flight. The proposed method provides a unified technical framework for low-altitude airspace representation, state management, and operational governance in complex urban environments.

Megacities—defined as Functional Urban Areas (FUAs) of more than 10 million inhabitants—are global hotspots of population growth, economic activity, and environmental pressure. Their development trajectories shape regional and global emission patterns, yet a comprehensive understanding of how urban expansion, air pollution, and economic development interact over time has remained limited. While prior research has examined either urban growth or atmospheric pollution trends, an integrated analysis linking both dimensions within a socio-economic framework is still lacking. This study addresses this gap by leveraging long-term Earth Observation (EO) datasets to systematically analyze settlement growth and tropospheric nitrogen dioxide (NO₂) pollution across 38 megacities between 1996 and 2015. Using the World Bank income classification, we evaluate whether observed environmental and urbanization patterns align with the Environmental Kuznets Curve (EKC)—a hypothesis that posits a non-linear (inverted U-shaped) relationship between environmental degradation and economic development.

Urban green space accessibility represents a critical dimension of sustainable planning and public health outcomes. This research quantifies compliance with the "3-30-300" framework - requiring residents to view three trees from home, neighborhoods to maintain 30% canopy coverage, and proximity to public green space within 300 meters across Montreal Island and Quebec City. While this policy has gained substantial theoretical traction, empirical implementation assessment in Canadian urban contexts remains limited. Employing high-resolution remote sensing imagery, deep learning-based land cover classification, and LiDAR-derived canopy data, we conducted comprehensive spatial analysis of both municipalities. Road network data from OpenStreetMap enabled walkability assessment. Integrated compliance metrics (I330300) revealed stark disparities: Montreal achieved 20.93% compliance, while Quebec City registered merely 2.69%. These findings underscore substantial green space accessibility deficits across both municipalities, with particular concentration in peripheral neighborhoods. Spatial statistical analysis identified pronounced clustering of non-compliance zones, demonstrating heterogeneous distribution of environmental amenities. Population demographic analysis revealed significant correlations between socioeconomic indicators and green space availability, suggesting environmental inequity patterns. Such disparities raise critical equity concerns regarding differential access to environmental services and associated health benefits. These results directly advance United Nations Sustainable Development Goal 11 objectives for establishing inclusive, sustainable cities. The quantitative assessment methodology demonstrates the efficacy of integrating remote sensing, machine learning, and spatial analysis for evidence-based urban environmental policy evaluation. Findings provide empirical foundations for targeted interventions addressing green space deficits in underserved urban communities, enabling data-driven municipal planning strategies that prioritize equitable environmental resource distribution and enhanced public health outcomes.

This presentation examines heat stress and mitigation capacity around transit stops during an extreme heat wave in Vancouver, Canada. Using hyperlocal microclimate modelling and high-resolution urban geometry data, we estimate “feels-like” Mean Radiant Temperature and shade availability to develop two new indicators: the Transit Stop Heat Stress Index and the Transit Stop Mitigation Capacity Index. Results reveal strong spatial and socio-economic disparities, with higher heat exposure and fewer mitigation features in lower-income neighbourhoods. The study demonstrates how microclimate data can guide climate-responsive, equitable transit planning under intensifying heat conditions.

Landfills are among the largest anthropogenic sources of methane, yet accurately detecting and quantifying their emissions remains challenging due to diffuse release patterns, complex terrain, and weather-driven variability. This presentation introduces a drone-based monitoring approach that uses a path-integrated Tunable Diode Laser Absorption Spectroscopy (TDLAS) sensor to detect and quantify methane emissions at a municipal landfill in London, Canada. Methane measurements collected throughout the year, together with on-site meteorological observations, were integrated into an inverse atmospheric plume-dispersion model to estimate emission rates. This contribution demonstrates the potential of drone-based TDLAS measurements to provide practical, high-resolution landfill methane monitoring and to reduce uncertainties in greenhouse gas reporting and mitigation efforts.

Climate change and urbanization transform life globally, with direct impacts on each other, yet they are rarely studied together across disciplines. The Synergy Grant urbisphere (https://urbisphere.eu), funded by the European Research Council (ERC), aims to forecast feedbacks between climate and cities. With new synergies between four disciplines (spatial planning, remote sensing, modelling and ground-based observations), urbisphere incorporates city dynamics and human behaviour into climate forecasts/projections, focusing on within-city dynamics of peoples’ activities and how these can be up-scaled to cities globally. urbisphere is studying inter/intra-city form and function (demographics, mobility, climate adaptation and vulnerability planning typologies), exploring human/socio-economic vulnerability, exposure, risk perception, coping/adaptive measures to climatic stressors and settlement/building typologies.

urbisphere is developing new ways to represent city dynamics for weather/climate models. These models are informed by the urbisphere developed Earth Observation system, using space-borne/airborne and ground based sensors with near real-time data transmission, processing and visualization of data from 500+ sensors, including a network of ceilometers, scintillometers, Doppler wind lidars, flux towers combined with street-level and indoor sensors. Combined these measure the 3-dimensional state of the atmosphere and the surface. These observations are providing both new understanding of urban surface-atmosphere processes and datasets for model evaluation at unprecedented detail.

Open-vocabulary 3D segmentation offers an attractive alternative to closed-set scene parsing, yet directly transferring 2D vision-language models to outdoor point clouds remains difficult because projection disrupts geometric continuity and sparse sampling weakens mask quality. This paper presents GeoOpen3D, a geometry-guided and training-free framework for open-vocabulary 3D point cloud segmentation. GeoOpen3D constructs a geometry-preserving RGB-D representation through projection, super-sampling, and depth enhancement to improve alignment between 3D structure and 2D foundation models. It then combines GroundingDINO for language-driven proposal generation with SAM for mask extraction, while introducing depth-aware regularisation to favour structurally coherent regions and clearer boundaries. The selected masks are back-projected to the original point cloud through pixel-to-point correspondence, yielding point-wise semantic labels without any 3D model training. Experiments on the SensatUrban dataset show that GeoOpen3D achieves 42.1\% mIoU, including 98.5\% IoU for buildings and 97.3\% IoU for vegetation, outperforming existing training-free open-vocabulary baselines. Additional experiments on a custom island dataset further demonstrate promising transferability to unseen categories. These results indicate that geometry-guided 2D-to-3D transfer provides an effective and scalable path towards open-vocabulary understanding of large-scale outdoor scenes.

Large-scale 3D panoptic segmentation is critical for digital twins and geospatial analysis, demanding models that process massive point clouds while distinguishing instances across highly diverse spatial scales. However, prevailing graph-based approaches rely on one-shot optimization, suffering from \textit{short-sighted decisions} where irreversible local errors propagate globally, leading to severe under-segmentation at boundaries between objects of disparate scales. To overcome this short-sightedness, we present \textbf{SPARC}, a scalable framework that reframes graph clustering as a sequential, self-correcting decision process driven by hierarchical reinforcement learning. Specifically, SPARC employs a dual-level agent where a meta-controller adaptively determines instance completeness while a low-level policy iteratively refines edge affinities, enabling the model to revise early mistakes based on long-horizon rewards rather than greedy local cues. Complementing this, we introduce Semantic Voxel Partitioning (SVP) to generate semantically coherent superpoints, ensuring robust primitives that mitigate noise before clustering begins. Extensive experiments demonstrate that SPARC achieves state-of-the-art performance on the DALES dataset with a Panoptic Quality of 62.4\%, surpassing existing methods by 9.8\% and effectively resolving multi-scale segmentation ambiguities.

3D Referring Expression Segmentation (3DRES) aims to segment point cloud scenes based on a given expression. However, existing 3DRES methods face three main challenges: (1) significant progress has been made in indoor scenes, yet large-scale and complex outdoor scenes, captured by airborne or mobile LiDAR, remain fully unexplored; (2) traditional methods often suffer from inefficiency and mis-segmentation due to insufficient attention to the spatial information of instances during query generation; and (3) existing models treat all queries equally in the decoder and predict the final mask in one step, which is inefficient in outdoor road scenes dominated by background point clouds, where objects are sparse and small. To address these challenges, a new outdoor 3DRES benchmark, named UrbanRefer, is introduced. The dataset consists of 100 large-scale outdoor scenes and 1,100 specially designed long textual descriptions, emphasizing geospatial relationships and multi-object contexts unique to outdoor environments. Additionally, the Language-guided Spatial Anchoring Network (LaSA-Net) is proposed for the directional segmentation task in outdoor scenes. Specifically, the Local-Global Aggregation (LGA) module is incorporated into the backbone to enhance local and global context awareness, effectively optimizing point features. Furthermore, a Text-driven Localization (TL) module is introduced, which directly predicts the 3D positions of all entities mentioned in the text, providing robust spatial priors for the decoder. Finally, a Hierarchical Prompt-aware Decoder (HPAD) is designed to locate rough regions by extracting task-driven signals from the interaction between expressions and visual features. Extensive experiments demonstrate that the mIoU metric of LaSA-Net outperforms state-of-the-art methods by 0.9%.

Most existing 3D vision-language models focus on object-level or single-room understanding and perform poorly in large-scale,

multi-room indoor environments where task-relevant objects constitute only a small fraction of the total point cloud. When multi-

room point clouds are fed directly into an LLM, critical semantic signals are diluted by the vast amount of redundant background,

making it difficult for the model to focus on truly relevant regions. We propose SceneReasoner, a decoupled spatial tokenisation

framework that addresses this challenge through three core designs: (1) pre-tokenisation text-guided feature weighting that leverages

the shared CLIP embedding space between OpenScene point features and text queries to amplify question-relevant point features

before any spatial compression occurs; (2) 2D–3D feature fusion that integrates top-down 2D CLIP features with 3D sparse tokens,

supplying the model with appearance semantics—such as texture, material, and room layout—absent from raw point clouds; and

(3) layer-wise dense feature injection that inserts local dense features into the LLM attention mechanism layer by layer for fine-

grained perception of key regions. We evaluate on the XR-Scene benchmark, which covers cross-room question answering and

scene captioning over HM3D indoor environments with an average area of 132 m2. SceneReasoner achieves the best CIDEr on

XR-SceneCaption (+0.33 over LSceneLLM), the highest METEOR on XR-QA, and competitive ROUGE-L across all three tasks,

demonstrating the effectiveness of task-guided spatial tokenisation for large-scene understanding.

Understanding and manipulating 3D spatial environments remains a fundamental challenge in geospatial sciences, with applications spanning digital twins, facility management, urban planning, and autonomous systems. While point cloud acquisition technologies have matured significantly, the semantic interpretation and interactive manipulation of captured 3D scenes continue to require extensive manual intervention and domain expertise. This paper presents a novel LLM framework that bridges unsupervised graph-based 3D scene understanding with natural language-driven interactive manipulation, enabling context-aware spatial intelligence at scale.

Accurate road inventories are crucial for maintenance, safety, and resource allocation, with automation improving efficiency but often lacking user-friendly human-machine interaction. This paper evaluates how non-photorealistic rendering of 3D point clouds impacts Multimodal Large Language Models (MLLMs) interpretation for road inventory, testing three methods on real road data in Santarém, Portugal. From 3D point clouds coloured with RGB information, non-photorealistic techniques are implemented and compared: Ambient Occlusion (AO), Eye-Dome Lighting (EDL) and Multi Feature-Rich Synthetic Color (MFRSC). Several state-of-the-art MLLMs are also tested: GPT5, Gemini2.5-Pro, Gemini2.5-Flash, CogVLM2, MiniCPM-V, Llama4-scout-17b, Mistral-Small3.2, Qwen2.5vl and Gemma3. The results indicate that non-photorealistic techniques do not hinder the identification of road elements by MLLMs, indicating their potential for 3D point cloud classification tasks even when true RGB colour is not available. Furthermore, the overall performance metrics, with F-scores over 80% for proprietary, state-of-the-art models (GPT5, Sonnet 4.5 and Gemini) show that 2D captures of 3D point clouds can be a suitable data source for zero-shot object classification.

Light Detection and Ranging (lidar) has become an essential tool to quantify forest structure in three dimensions, allowing extraction of tree-level metrics such as height, crown volume, diameter at breast height (DBH), and biomass. Accurate forest structure quantification supports applications such as wildfire management, biodiversity assessment, forest health monitoring, and timber management. This is particularly urgent in countries such as Canada, where wildfires pose a significant challenge to forest management due to their increasing frequency and severity; advanced fire behavior models aid wildfire preparedness by predicting fire behaviour at fine-scale in 3D but require detailed 3D fuel information including canopy and ladder fuels.

Terrestrial Laser Scanning (TLS) and Uncrewed Aerial Vehicle (UAV) lidar provide dense point clouds that allow highly accurate characterization of individual trees, critical for assessing forest attributes and wildfire fuel characteristics. However, their limited spatial coverage makes them neither time- nor cost-effective for mapping extensive forested regions. Airborne Laser Scanning (ALS), in contrast, covers broad areas efficiently by collecting data from higher altitudes, but at the cost of lower point densities (typically 1–100 points/m²), insufficient for precise individual tree characterization.

To address this challenge, this study reconstructs densified tree point clouds from low-resolution ALS data using an upsampling framework based on a deep generative network trained on real and synthetic datasets. This approach bridges the gap between ALS’s extensive coverage and the detailed structural information provided by TLS and UAV lidar, enabling accurate, large-scale quantification of forest structure for applications such as wildfire management and monitoring.

Accurate tree positions play a vital role in precision forestry and environmental sciences. In this study, we propose an accurate, efficient, and adaptable method for tree localization by integrating heading, diameter at breast height (DBH), and ultra-wideband technology. The proposed method is simple to implement in different forest environments and can determine the position of each tree within a few seconds. Compared with traditional field measures, such as laser rangefinders and inclinometers, the proposed approach is more efficient. In comparison with commonly used measures, such as terrestrial laser scanning (TLS) and mobile laser scanning (MLS), the proposed method is more cost-effective and easier to implement, making it particularly suitable for natural forests that are remote from roads yet require accurate measurements. Field experiments were conducted in a managed boreal forest in southern Finland, characterized by minimal understory vegetation and good visibility, where a total of 50 trees were mapped. Experimental results indicate that the proposed method can accurately determine tree positions with an RMSE of 0.12 m and an MAE of 0.11 m.

Plant phenotyping has become a fundamental tool in modern agronomic research, enabling quantitative analysis of morphological characteristics that can be collected in three dimensions using photogrammetric techniques or point clouds obtained by LiDAR systems. However, automatic segmentation of plants, especially the main stem and its branches, still poses a challenge for certain crops. This work proposes a non-destructive, geometry-based methodology for morphological phenotyping of tomato plants (Solanum lycopersicum) using photogrammetric point clouds. The proposed methodology consists of the following steps: stratification of the plant into horizontal sections; clustering of each stratum using the DBSCAN algorithm; selection of clusters based on the linearity tensor derived from eigenvalue analysis; and the fitting of a 3D cylinder to the linear clusters to approximate the main stem. The method was validated using manually labeled point clouds from nine tomato cultivars, achieving accuracy between 88% and 97%, with average F1-scores of 63.6% for the stem and 96% for the branches

Linking remotely sensed forest backscatter with fine-scale tree crown structural dynamics provides insights into tree growth strategies under varying conditions. In this study, we investigate whether branch-scale tree growth allocation dynamics, derived from multi-temporal TreeQSM models, are reflected in SAR and ALS observations. We analyzed branch organization dynamics of silver birch (Betula pendula) using terrestrial laser scanning data from 2021, 2023 and 2025 at a boreal forest site in southern Finland. Branch allocation metrics, including volume-weighted mean diameter (VWMD), small branch fraction (SBF), distal volume fraction, relative branch height, and top canopy volume, were quantified to capture shifts between structural reinforcement and exploratory growth. These metrics were compared with Sentinel-1 SAR features (α, entropy, C11, C22) and ALS-derived canopy metrics (plant area index, vertical complexity index) alongside local structural variables. Results show a consistent trade-off between coarse and fine branching, with strong negative correlations between ΔVWMD and ΔSBF across both periods (ρ = –0.92). SAR-derived α exhibits strong associations with these allocation shifts during 2021–2023 (ρ = –0.81 with ΔVWMD; ρ = 0.75 with ΔSBF), indicating sensitivity to internal redistribution of branch material. ALS metrics from 2021 reflect initial canopy structure and are associated with subsequent allocation shifts. Despite the small magnitude of observed changes, consistent monotonic relationships across datasets suggest that subtle within-crown branch allocation is detectable from satellite and aerial observations, reflecting the surrounding canopy context. However, weakened correlations in 2023–2025 highlight the influence of external factors on SAR signals.

Tree instance segmentation of airborne laser scanning (ALS) data is of utmost importance for forest monitoring but remains challenging due to variations in the data caused by factors such as sensor resolution, vegetation state at acquisition time, terrain characteristics, etc. Moreover, obtaining a sufficient amount of precisely labeled data to train fully supervised instance segmentation methods is expensive. To address these challenges, we propose a weakly supervised approach where labels of an initial segmentation result obtained either by a non-finetuned model or a closed-form algorithm are rated by a human operator. The labels produced during the quality assessment are then used to train a rating model, whose task is to classify a segmentation output into the same classes as specified by the human operator. Finally, the segmentation model is finetuned using feedback from the rating model. This in turn improves the original segmentation model by 34% in terms of correctly identified tree instances while considerably reducing the number of non-tree instances predicted.

Trees play a key role in our planet. They regulate climate, preserve biodiversity, and contribute to human well-being. Each species has different contributions to our globe and a specific carbon storage potential. Identify tree species enable better measurement of global carbone, help authorities for better manage forests and green spaces. Unmanned Aerial System (UAS) LiDAR has become a powerful source of 3D point cloud for vegetation analysis, given its ability to captured large area in a short time and its capacity to penetrate canopy layers. Deep learning methods extract discriminative features directly from raw point clouds and generalize well to unseen datasets. This study optimises PointNet++ deep learning architecture for tree species classification by analysing the influence of sampling configurations on the performance of model detection, by using an open-source dataset “FOR-species20K”.Three-point cloud sampling configurations (4 096, 8 192, and 16 384 points per tree) were tested with three random seeds (0,42 and 123) to assess their impact on classification accuracy and ensure stability of prediction. Results on a separate test set of 508 trees show a consistent improvement in performance of PointNet++ with a sampling configuration of 8 192 points per tree, reaching a macro-average F1-score of 89.65%, surpassing the 74.9 % reported by (Puliti et al., 2025) for evaluating the same architecture. Dominant species such as Fagus sylvatica, Picea abies, and Pinus sylvestris achieve F1-scores exceeding 90%, indicating high model robustness.

Semantic segmentation deep learning algorithms still depend on large quantities of high-quality annotated samples. Because remote sensing imagery spans vast areas and highly variable land surface environments, annotation demands substantial expertise and is both time-consuming and labour-intensive, leaving the field with an acute shortage of first-rate training samples. Moreover, object categories in land cover data are inherently imbalanced. Models trained under imbalance often underperform in small sample categories.

This study proposed a decoupled latent diffusion framework for RS semantic segmentation sample generation, and a proportion-aware loss to optimize balance of sample classes. We tested the proposed method on the ISPRS Potsdam dataset and compared it with two classic image generation baselines. The results show that our approach outperforms the baselines, producing synthetic samples with superior visual quality and semantic consistency. To verify downstream utility, we trained DeeplabV3+, PSPNet, and SegFormer segmentation models with the synthesized data. Across all three networks, overall segmentation accuracy and class balance metrics improved markedly; gains were especially pronounced for the rare “Clutter” and “Car” categories, underscoring the proposed method’s generality and robustness. We further analysed how the proportion of synthetic samples affects performance. As the ratio of synthetic to real samples increased, mIoU and mF1 first rose and then declined; the best results were obtained when the proportion of synthesized samples approached 40%. This indicates that a moderate amount of synthetic sample can significantly boost segmentation performance, whereas excessive synthetic data risks over-fitting or misclassification.

Existing remote sensing change captioning methods are limited to optical-only data, precluding all-weather, all-day monitoring. To address this, we introduce Bright-CC, the first large-scale benchmark for cross-modal (Optical-to-SAR) change captioning. Curated from the newly-proposed BRIGHT dataset, Bright-CC comprises 9,953 paired images focused on building damage assessment. It features dense four-class semantic labels (intact, damaged, destroyed) and a rich corpus of 49,765 GPT-4O-generated sentences (5 per pair), moving beyond simple binary change labels. Furthermore, we propose the Hybrid Feature Alignment Network (HFA-Net) as a robust baseline for this new task. HFA-Net is specifically designed to tackle the significant domain shift between heterogeneous sensors. Its architecture features: (1) a pseudo-siamese alignment module (HFEA) to project features into a common space; (2) a multi-scale atrous convolution module (CSTDF) to refine change context; and (3) a novel Lightweight Caption Generator (LCG), which is a parameter-efficient Transformer trained from scratch to avoid overfitting. Experiments show HFA-Net substantially outperforms adapted optical-only baselines (RSICCFormer, Chg2Cap) on all standard metrics. This work provides the community with a critical dataset and a strong baseline for future cross-modal spatio-temporal intelligence.

The scarcity and high cost of acquiring high-quality post-event remote sensing images (due to cloud cover, satellite limitations, and security risks) severely constrain the development and accuracy of change/damage detection models. This data gap is especially critical in disaster or military conflict scenarios. Existing cross-temporal image generation methods often lack precise spatial and semantic control, leading to inconsistent or unrealistic synthetic results. To address this core challenge, this paper introduces the Remote Sensing Change/Damage Generator (RSCDG), a novel method based on the Latent Diffusion Model for high-fidelity simulation of post-event satellite imagery. The RSCDG’s core innovation lies in its multimodal condition embedding framework, which integrates three specialized control pathways:The Pre-event Visual Prompt Adapter (built on PrithviModel) ensures high structural consistency between the pre-event and generated post-event images.The Spatial Location Control Pathway (using a ControlNet architecture and change/damage masks) precisely dictates the geometric location of the simulated change.The Generation Content Controller (using a CLIP Text Encoder) enhances semantic realism by guiding the model with natural language descriptions of the change/damage.Furthermore, we introduce a Mask Alignment Loss to enforce spatial and semantic adherence to detection rules. Results demonstrate that RSCDG accurately simulates complex scenarios like new urban construction and catastrophic building collapse. RSCDG is a powerful new tool designed to augment training data and significantly accelerate high-precision disaster response and urban monitoring.

Deep-learning methods dominate remote-sensing change detection (CD), yet state-of-the-art models remain parameter-heavy and struggle with crisp boundaries, limiting their use on edge devices. We present LEDGNet, a Lightweight, Edge-knowledge-Distillation-Guided CD Network, that reconciles accuracy, boundary fidelity, and efficiency. LEDGNet integrates three purpose-built components: 1) an Edge Distillation Module that mines multi-scale boundary cues from a high-capacity teacher and transfers them to a compact student through an edge-aware loss; 2) StarLite, a depth-wise separable encoder that preserves fine spatial detail while minimizing floating-point operations; and 3) LiteDecoder, an inexpensive feature-fusion head that restores full resolution without bulky up-sampling. This design halves the parameters and inference time of mainstream fine-grained CD networks while enhancing edge sharpness. On the CDD and LEVIR-CD benchmarks, LEDGNet achieves competitive F1 performance while maintaining a compact footprint of 20.58 M parameters and 35.18 G FLOPs. With an inference time of 255 ms, it strikes a balance between resource consumption and detection efficiency, making it well-suited for high-efficiency remote sensing monitoring.

Semantic segmentation of remote sensing imagery (RSI) is essential for urban mapping, land-use monitoring, and many areas. However, pixel-level annotation is expensive, making weakly supervised semantic segmentation (WSSS) that relies on image-level labels an attractive alternative. Leveraging pre-trained models offers strong priors from large-scale learned representations can help the WSSS, yet frozen models often yield sparse and misaligned class activation maps (CAMs) due to domain gaps and static inference. We propose a lightweight and efficient framework that integrates CLIP and DINO to address three challenges: semantic misalignment between generic text prompts and RSI-specific visuals, static CAM quality, and incomplete object coverage. Our design includes: (1) a Textual Prototype-Aware Enrichment (TPE) module that builds an RS-specific knowledge base using LLM generated descriptions to enrich text prompts; (2) a Unified Semantic Relation Mining (USR) module that fuses learnable adapter features with CLIP attention and DINO affinity for online CAM refinement; and (3) a Visual Prototype-Aware Enrichment (VPE) modulemaintains momentumvisualprototypes to complete regions and sharpen boundaries. Using frozen priors while only training a lightweight decoder ensures efficiency and consistently improves segmentation accuracy across diverse remote sensing scenes. Experimental results on the iSAID and ISPRS Potsdam datasets demonstrate the effectiveness of the proposed framework, achieving 38.01% mIoU on iSAID dataset and 47.01% mIoU with 66.89% overall accuracy on Potsdam dataset.

Shadows cast by terrain and tall structures remain a major obstacle for high-resolution satellite image analysis. Public resources offering geometry-consistent paired shadow/shadow-free satellite imagery are essentially missing, even though there is a growing body of work on shadow removal in remote sensing, and most large-scale Earth-observation datasets are designed for shadow detection or 3D modelling rather than shadow removal. Existing deep shadow-removal datasets either target ground-level or aerial scenes or rely on unpaired and weakly supervised formulations rather than explicit satellite pairs. We address this gap with deSEO, a geometry-aware and physics-informed methodology that, to the best of our knowledge, is the first to derive paired supervision for satellite shadow removal from the S-EO shadow detection dataset through a fully replicable pipeline. For each tile, deSEO selects a minimally shadowed acquisition as a weak reference and pairs it with shadowed counterparts using temporal and geometric filtering, Jacobian-based orientation normalisation, and LoFTR–RANSAC registration. A per-pixel validity mask restricts learning to reliably aligned regions, enabling supervision despite residual off-nadir parallax. In addition to this paired dataset, we develop a DSM-aware deshadowing model that combines residual translation, perceptual objectives, and mask-constrained adversarial learning. In contrast, a direct adaptation of a UAV-based SRNet/pix2pix architecture fails to converge under satellite viewpoint variability. Our model consistently reduces the visual impact of cast shadows across diverse illumination and viewing conditions, achieving improved structural and perceptual fidelity on held-out scenes. deSEO therefore provides the first reproducible, geometry-aware paired dataset and baseline for shadow removal in satellite Earth observation.

Persistent Scatterer Interferometry (PSI) is a consolidated active remote sensing technique to measure and monitor land deformation. The technique has experienced an intense development in the last 25 years. PSI techniques use large stacks of SAR images that cover a given observation period. The outcome of any PSI processing is a cloud of geocoded measurement points that contain the estimated deformation time series over the observation period. If the analysed area is wide, the corresponding point cloud can be huge. In these cases, the potential users often experience problem in analysing such huge point clouds, and this can limit the PSI exploitation. In this paper we present a set of products that address specific application needs or that offer higher-level products with respect to the standard PSI products, which can facilitate the interpretation and exploitation of the PSI results.

Interferometric SAR (InSAR) is a well-established remote sensing technique to measure and monitor land deformation. We focus in this paper on Persistent Scatterer Interferometry (PSI) techniques based on large stacks of SAR images. Several PSI approached have been proposed in the last three decades, see for a review Crosetto et al. (2016). In this paper, we describe an approach the is based on the direct integration of the interferometric phases (back-to-back approach).

The Persistent Scatterer Interferometry (PSI) is a multi-temporal InSAR approach that allows to monitor displacement time series of the Earth's surface. The method identifies and analyzes Persistent Scatterers (PSs) which are phase stable scattering points which dominate the backscatter of their resolution cell. Standard PSI techniques only identify and analyze PSs which are coherent throughout the whole considered SAR time series. However, PSs can fade, appear or be occluded during the time series, forming so called Temporary PSs (TPSs), which should be integrated into the PSI to establish optimal measurement point networks. Previous research has proposed methods to integrate such TPSs into the PSI, however these were exclusively evaluated for construction-related TPSs. In this work, we evaluate the performance of a TPS integration method in handling recurringly occured PSs, and compare the performance of individual components of the algorithm against alternative methods. We evaluate the methods using simulated TPSs with temporal and spatial baseline settings taken from real Sentinel-1 data stacks. Furthermore, we present and discuss the application of the methods to a Sentinel-1 data stack acquired over the Oetztal Alps, which are seasonally covered by snow. We show that the integration of ROPSs significantly increases the measurement pixel density at many locations across the study area, compared to results from the European Ground Motion Service. Even if most of the ROPS did not have identified coherent segments in each covered summer with the current analysis algorithm, their integration leads to a significant information gain compared to standard PSI approaches.

Deformation monitoring in open‑pit mining and tailings‑dam operations is critical for operational safety, yet conventional in situ geodetic techniques provide only sparse, point‑based measurements. InSAR offers many displacement measurements, but its operational uptake is limited by complex workflows and the difficulty of interpreting analysis‑ready products such as EGMS. In cold regions, seasonal data gaps can introduce phase‑unwrapping artefacts that appear as winter‑only displacement offsets of approximately half the Sentinel‑1 wavelength. We propose a semi‑automated workflow to post‑process EGMS displacement time series, including pre‑filtering to identify and remove points affected by phase‑unwrapping errors and subsequent time‑series clustering in either a reduced‑dimensional representation or the full feature space. Cluster selection is automated using heuristic criteria and a custom metric based on temporal homogeneity and consistency. The findings show that the semi‑automatically detected clusters are plausible with regards to a visual interpretation of the EGMS data. The workflow supports improved interpretation of EGMS time series and avoids hard‑coded thresholds or reliance on velocity‑based estimates.

The Inglewood Oil Field, located in the Los Angeles Basin, California, is a major urban hydrocarbon production site with a documented history of ground deformation linked to oil extraction. To assess ongoing deformation and validate previous monitoring results, Interferometric Synthetic Aperture Radar (InSAR) analysis was conducted using Sentinel-1 SAR data processed through the Alaska Satellite Facility’s HyP3 platform and the Miami InSAR Time-series software in Python (MintPy). The study analysed ascending and descending datasets acquired between 2020 and 2025 to derive high-resolution deformation time series and velocity maps. Results reveal a localized deformation pattern characterized by low-magnitude vertical motion, with maximum uplift and subsidence rates of approximately +0.8 cm/yr and –1.6 cm/yr, respectively. Minor horizontal displacements (±1.0 cm/yr) suggest limited lateral strain associated with reservoir compaction and stress redistribution. Compared with previous assessments conducted up to 2024, the current findings indicate a marked reduction in deformation magnitude, implying progressive stabilization of reservoir pressure and improved subsurface management. These results demonstrate the effectiveness of InSAR for long-term monitoring of urban oilfields, providing critical insights into the behaviour and contributing to risk mitigation in densely populated environments.

Ground deformation caused by excessive groundwater extraction has become a major environmental concern in agricultural regions worldwide. Interferometric Synthetic Aperture Radar (InSAR) enables large-scale monitoring of ground deformation. However, its performance often decreases in low-coherence areas affected by vegetation growth and irrigation. In this study, we conducted a comparative evaluation of three multi-temporal SBAS-InSAR processing frameworks, MintPy, LiCSBAS, and SARvey, to assess their consistency in monitoring ground deformation across Golestan Province, Iran, using Sentinel-1 data acquired between 2014 and 2024. The analysis included deformation velocity fields, cross-sectional profiles, and time-series displacements, which were compared with temperature and precipitation variations. All three frameworks identified a pronounced deformation zone in the Gorgan Plain, with maximum line-of-sight deformation rates up to 13 cm/year. Quantitative comparisons showed strong correlations among the frameworks (r = 0.80 to 0.89), confirming their mutual reliability even under low coherence conditions. The time-series analysis revealed clear seasonal deformation patterns, with summer subsidence and winter uplift closely related to hydroclimatic fluctuations. Overall, this study demonstrates that multi-temporal SBAS-InSAR approaches can provide consistent and physically meaningful deformation estimates in challenging agricultural environments, offering valuable insights for subsidence monitoring and water resource management.

Asteroid exploration is crucial for understanding the solar system’s origin, but establishing a precise body-fixed coordinate system—relying on accurate rotation parameters—remains challenging. Conventional methods like ground-based light curve inversion often lack precision: for example, it yielded ±2° errors for Ceres’ pole and ±10° for Vesta’s, failing to meet demands for topographic mapping and navigation. This study proposes a refinement method combining stereo intersection angle optimization and grayscale threshold masking. First, using the camera’s interior orientation parameters and tie point coordinates, relative orientation of stereo image pairs is conducted to build a stereo model, followed by forward intersection to calculate intersection angles. Only pairs with favorable geometry (intersection angle >5°) are retained to avoid large position errors from nearly parallel sightlines. Second, a grayscale-based binary mask is created to separate the asteroid from the deep-space background, eliminating spurious edge features that cause mismatches; the SIFT algorithm then extracts and matches features exclusively within the masked region. Finally, an “exhaustive search” iteratively refines rotation parameters using optimized matched points. Validated on 127 Hayabusa2 ONC-T images of asteroid Ryugu (captured July 10, 2018, 2.11m/pixel), the method reduced 5,174 initial candidate pairs to 1,454 valid ones (137,191 matched points). After 4 iterations, refined parameters were RA=96.5° and Dec=-66.4°, with minimal errors (δRA=0.069°, δDec=0.0126°) against reference values (RA=96.431°, Dec=-66.387°). Compared to methods without the two strategies, mismatches dropped from 14,949 to 7,369, and forward intersection residuals decreased. Future work will integrate initial parameters into a bundle adjustment model for further refinement.

The lunar polar regions have emerged as core targets in lunar exploration, primarily due to the potential water ice resources stored within their permanently shadowed areas. However, the complex terrain and extreme illumination conditions in these polar regions present significant challenges to the navigation of lunar rovers—systems that previously relied on dead reckoning and visual matching techniques. To address this, active 3D sensors such as LiDAR will be integrated into future exploration missions.Simultaneous Localization and Mapping (SLAM) based on multi-sensor fusion via factor graphs can significantly enhance the localization robustness of rovers on the lunar surface. In this context, we propose the Lunar Scene Recognition Adaptive SLAM (LSRA-SLAM) method: a framework that leverages environment-aware pre-training to dynamically adjust factor-graph weights, thereby achieving more consistent fusion of stereo camera, LiDAR, and IMU measurements across diverse lunar scenarios. We also introduce a reinforcement learning-based online training strategy, which enables the network to robustly learn from the system's dynamic behaviors. Simulated experiments validate the effectiveness of the proposed LSRA-SLAM method.

Accurate crater mapping in lunar permanently shadowed regions (PSRs) is hindered by extreme low-light and low-resolution imagery. We present YOLOLens2.0, a unified, end-to-end deep learning framework designed for high-fidelity crater detection and terrain reconstruction in these challenging environments. The architecture integrates a Dense-Residual-Connected Transformer

(DRCT) for multimodal super-resolution (SR) with a YOLO-derived detection module and an affine calibrator to ensure geometric consistency at meter scale. Our framework exploits a bidirectional synergy where SR enhances feature discriminability for detection, while detection-driven supervision refines structural reconstruction. Validation on Kaguya data demonstrates a significant performance leap, achieving a Recall of 89.20% and an mAP@50 of 0.844 an improvement of over 33 percentage points in recall compared to the original YOLOLens. Out-of-distribution validation on ShadowCam imagery, performed without fine-tuning, confirms the model’s robustness and scalability. The framework successfully preserves quantitative elevation fidelity and supports detailed morphometric analyses, including the extraction of the crater size-frequency distributions (SFDs) that align with theoretical lunar production functions. YOLOLens2.0 provides a scalable, high-precision methodology for planetary mapping, offering critical insights for lunar surface evolution studies and future exploration missions.

Reliable localization across the full orbit-descent-ground chain in planetary exploration remains difficult because extreme differences in altitude, viewing geometry, resolution, and illumination cause cross-view image matching to fail. Traditional keypoint pipelines and unified Structure-from-Motion (SfM) struggle to establish robust correspondences across these heterogeneous Satellite-Descent-Ground datasets due to severe domain gaps.

To overcome these limitations, we propose a novel framework based on a joint semantic-geometric optimization paradigm. Rather than forcing a unified SfM pipeline across drastically different viewpoints, our method leverages independent intra-domain SfM outputs and telemetry data as structural priors. We introduce a differentiable rendering approach that tightly couples the optimization of 3D Gaussian Splatting (3DGS) scene parameters with learnable camera extrinsics. Furthermore, by integrating high-level semantic epipolar constraints derived from foundation models, our method dynamically refines initial cross-domain pose estimates during the rasterization loop. This joint formulation effectively bypasses the fragility of low-level pixel matching, enabling accurate and robust alignment across the vast baselines inherent to multi-stage planetary exploration image sequences.

Mars topographic data are crucial for quantitatively characterizing the Martian surface, supporting exploration missions, and enabling scientific study of surface processes. Photogrammetric processing of Mars orbital imagery is the most representative method for generating 3D terrain models, with bundle adjustment (BA) serving as the key step for mitigating inconsistencies in overlapping regions of different orbital images and further improving the spatial accuracy of the resulting DTMs. However, due to the texture-less surface of Mars and the absence of ground control points, the stability of BA is often compromised. Impact craters, which are prevalent on the Marian surface, have been utilized as an important semantic prior in various image analysis applications. They can also be used to assist the BA process for precision topographic mapping of the Martian surface.

This study introduces a novel BA method assisted by robust crater graph features to address this. The approach involves: (1) extracting craters using a deep learning model (YOLOv5) and constructing a stable graph structure via a minimum spanning tree; (2) establishing crater correspondences across different images based on graph features to generate robust tie points; and (3) formulating a strengthened BA equation with constraints from the graph's angular and edge relationships to mitigate geometric inconsistencies.

Experimental results indicate that the proposed method provides an effective solution for high-precision 3D mapping from Martian surface imagery with limited textures and significant illumination variation. By incorporating crater graph features, it enhances the precision and stability of BA, yielding high-precision topographic mapping results for various applications.

Surface roughness influences image contrast by altering illumination, which depends on the surface slope. We conducted Monte Carlo simulations of rough surfaces under both directly illuminated and secondary-illuminated lunar conditions. Our results indicate that PSR secondary illumination yields significantly lower contrast, characterized by soft, diffuse shading and negligible shadow fraction.

Accurate land cover classification requires both spatial details and temporal information of remote sensing data. While publicly available satellite image time series (SITS) offer short revisit times, they suffer from limited spatial resolution. In contrast, aerial imagery provides fine-grained spatial details, but its temporal coverage is limited. Thus, combining data from those sensors is of interest as their properties are complementary w.r.t. the problem domain. However, the large gap in spatial resolution between these two sensors makes their integration challenging. Generating super-resolution-SITS (SR-SITS) before fusion can help to reduce this gap. In this work, we propose a new approach that integrates diffusion models for generating SR-SITS into a method for the joint pixel-wise classification of aerial and SITS data. Specifically, we employ a diffusion model to generate SR-SITS at an intermediate resolution from the raw SITS and aerial imagery of the same observed area. The SR-SITS are temporally encoded and fused with the aerial features using a cross attention module to produce pixel-wise classification at the geometrical resolution of aerial image. Experimental results on the existing FLAIR benchmark dataset indicate that our approach achieves state-of-the-art results, with a mean Intersection over Union score of 64.0% and an overall accuracy of 76.6%.

We present Sky-NeRF, a novel method for cloud topography estimation based on Dynamic Neural Radiance Fields. Similar to NeRF, we propose to model the 3D structure of clouds as a radiance field, encoded in the parameters of a neural representation. Our goal is to reconstruct the 3D geometry, appearance, and motion of the cloud using a stereo-video of high-resolution top of the atmosphere radiance images. In this paper, we evaluate a novel way of modeling the dynamic behavior of clouds, with the goal of extracting added-value physical information regarding the cloud such as advection speed and direction, velocity field and cloud trajectories. We investigate how to include a simple physical prior, advection, into the learning system and evaluate its impact. Our results show that Sky-NeRF is able to provide a more complete 4D reconstruction than traditional stereo-matching-based algorithms. Moreover, thanks to a physics-based interpolation, Sky-NeRF is able to generate coherent new images from unseen viewing angles, and at any time between the observed frames.

3D topographic change estimation is a fundamental task for understanding Earth surface dynamics in fields of photogrammetry and laser scanning. However, at the current state of research, it is still challenging to accurately separate and quantify various components of topographic surface changes (i.e., rigid spatial movement and non-rigid morphological deformation). In this paper, we conceptualize a surface change tensor to describe 3D surface change based on the displacement field, considering contribution of neighboring points to their center point on the surface. With this concept, we design a new method that is able to quantitatively separate rigid and non-rigid topographic change components from the mixed topographic change. Experiments on synthetic datasets demonstrate that our method is accurate and robust to quantify rigid and non-rigid surface changes, with superiority to the baseline method (M3C2). Additionally, real-world experiments on 3D point clouds collected at four epochs show the effectiveness of the proposed method for monitoring topographic dynamics and identifying geomorphological processes in complex large-scale mountain environments.

Monitoring surface change in dynamic environments is essential to preserve the integrity of human infrastructure and livelihoods from natural hazard consequences. With the advent of 4D remote sensing, near-continuous monitoring of dynamic scenes is unlocked. However, the unordered and irregular nature of point clouds, compounded by temporally variable occlusions and diverse acquisition conditions, hinders the accurate analysis of highly information-rich 4D data. This work addresses the challenge of irregular spatiotemporal sampling in time series of 3D point clouds for the case study of a dynamic sandy beach at different time scales. We explore the use of implicit neural representations (INRs) to model 4D data as continuous spatiotemporal functions that are optimised to estimate the beach topography continuously through space and time. By comparing four model variants and assessing their performance to reconstruct spatially and temporally subsampled data, we evaluate the applicability of INRs to high-frequency topographic monitoring, especially in the context of 4D change analysis. Our results show the ability to reconstruct missing epochs from time series of 3D point clouds with centimetric to decimetric accuracy at time scales ranging from seasonal to daily observations. Our findings highlight the importance of hyperparameter tuning to enable the capture of local details in complex spatiotemporal datasets. Through this, our work lays the foundation for continuous spatiotemporal representation of dynamic scenes, supporting a potentially broad range of change analysis applications.

Spatiotemporal analysis of geospatial time series data has gained increasing attention with the emergence of 4D point clouds and automatic acquisition technologies such as permanent laser scanning (PLS), time-lapse photogrammetry, and uncrewed aerial vehicle (UAV) platforms, enabling near-continuous monitoring of Earth surface dynamics for change detection and process characterization. However, facing massive data volumes through the temporal domain, current topographic data curation practices often rely on empirically determined data processing and management, which may significantly affect reusability, interoperability, and hence processing efficiency due to the absence or heterogeneous nature of metadata. The need for standardized approaches to manage time-dependent metadata has become critical as the demands for sharing data and reproducing analysis across tools and application domains increase. We propose a topographic 4D extension (topo4d) to the SpatioTemporal Asset Catalog (STAC) framework, which provides an open and extensible specification for automatic metadata curation and FAIR data management practices. This paper demonstrates how the topo4d extension facilitates the interoperability and reusability of 4D datasets and presents the corresponding metadata curation workflows applied to two real-world environmental monitoring applications.

This study presents an integrated remote sensing and deep learning approach for large-scale detection of asbestos-cement roofing in urban environments. Asbestos remains a major environmental health concern across Europe, where asbestos-cement materials persist in the built environment despite regulatory bans. Accurate identification and quantification of these materials are critical for effective remediation planning and equitable health protection.

The research focused on Poland’s two largest metropolitan areas—Warsaw and Kraków—which differ markedly in morphology and historical development, providing contrasting case studies for model validation. High-resolution orthophotomaps (5 cm and 25 cm) from 2023–2024, combined with national building footprint datasets and field-verified information, were used to train and validate a convolutional neural network (CNN) for binary classification of asbestos and non-asbestos roofs.

The highest producer accuracy (90.4%) and overall accuracy (92.9%) were achieved using 128×128-pixel image windows, confirming that broader spatial context enhances classification precision in dense urban settings. The CNN model demonstrated consistent performance across both cities, highlighting its robustness and scalability.

By integrating open orthophotos with open-source analytical frameworks, the method supports the creation of spatially detailed asbestos inventories aligned with the EU INSPIRE Directive and the 2023 Asbestos Directive (EU 2023/2668). The approach enables cost-effective, standardized monitoring applicable to metropolitan and smaller urban contexts alike.

This study advances data-driven environmental health management by demonstrating that deep learning applied to national aerial imagery can deliver operational tools for mapping asbestos exposure risks and informing sustainable, equitable remediation strategies across Europe.

Urban greenspace is essential for ecological resilience, climate regulation, and human well-being, yet long-term, fine-scale assessments of its spatiotemporal dynamics and the extent to which residents benefit from green exposure remain limited. This study develops a 30-m resolution Fractional Vegetation Cover (FVC) dataset to monitor interannual and seasonal variations in urban greenspace across twelve representative Chinese cities from 2000 to 2020. To capture temporal exposure, we introduce the “greendays” metric, defined as the number of days per year that residents experience visible greenery. A population-weighted exposure model was applied to quantify both the magnitude and equality of greenspace exposure. Results show that greenspace increased across all cities over the two decades, with peri-urban areas exhibiting the most substantial gains due to ecological restoration and park development, while core urban areas experienced moderate but consistent improvements linked to renewal and localized greening efforts. Greendays displayed a slight upward trend, indicating an extended duration of annual greenery exposure for residents. Exposure equality remained high and improved in most cities, suggesting that greening initiatives increasingly benefited diverse population groups. Overall, this study provides a robust and scalable remote-sensing-based framework for tracking urban greenspace and exposure equity, offering critical evidence to support nature-based solutions, environmental justice, and sustainable urban planning in alignment with global development goals.

This study develops a generalized method of moments (GMM) framework to quantitatively assess the integrated relationships among climate, vegetation, and lake water volume. Using GSOD precipitation data, SSEBop evapotranspiration, Nino3.4 and MEI indices, and NDVI, we analyzed monthly variations of climatic and vegetation conditions in the Lake Victoria basin from 2000 to 2020. The associations between these factors and lake water-volume changes were first examined, and dynamic GMM was then applied to remove mutual influences among climate variables, allowing for a more reliable attribution of dominant drivers.Results show that precipitation is the primary driver of seasonal to interannual water-volume variations, while evapotranspiration imposes a consistent negative effect on lake storage. ENSO significantly modulates multi-year water anomalies. Vegetation dynamics respond to both climatic variability and lake water-volume changes, with water-level fluctuations providing additional positive feedback after controlling for climate effects.

Posidonia oceanica (Linnaeus) Delile (referred from now on also as P. oceanica) is a marine flowering plant endemic to the Mediterranean Sea, forming extensive underwater meadows that play vital ecological roles, especially as blue carbon reservoirs. Its distribution spans from Gibraltar to Turkey and North Africa to the Adriatic down to 40-50 m of depth (Cocozza et al., 2024). Human impacts, such as pollution, urbanization, and global warming, have reduced its extent by up to 56% in some regions (Robello et al., 2024). Monitoring these meadows is essential, and remote sensing data such as Neo-Pléiades satellite imagery enable their accurate mapping and health assessment. This study applies object-based classification to orthorectified Neo-Pléiades images to evaluate Posidonia oceanica distribution along Sardinia’s eastern coast.

Soil is a crucial component of the ecosystem, affected by climate change, and is often overlooked by remote sensing experts and insufficiently considered while discussing sustainable development projects. To enhance the use of soil related datasets based on earth observation during the planning phase of participatory processes, a specific analysis workflow was piloted during community consultations in Dodoma, Central Tanzania. In order to enhance the integration of the soil conditions during the design of a new community development plan Landsat 8 data from 2023 and 2024 was processed and prepared to make soil information more accessible to non-technical staff and the local communities in Chamwino district. Results confirm the suitability of the SBI as soil indicator thanks to its high resolution, easy interpretability, and context specificity. Preprocessing through experts was identified as viable solution for preparing the data. In addition, field truthing exercises and conversations with the local community members further confirm the accuracy of this dataset for highlighting areas affected by soil salinity or fertility loss and for the final use during participatory planning processes.

Automatic extraction of vectorized roof structures above LOD2.0 remains challenging due to their geometric complexity and the presence of small and occluded elements over the roofs. Detecting fine-scale roof objects such as chimneys and dormer windows in very high resolution aerial imagery is still an active research topic. This study presents a workflow for automated detection and vectorization roof structures at city scale using Line Segment Detection (LSD) networks. Compared to model-based building reconstruction approaches, LSD networks do not rely on pre-defined roof typologies and are able to extract complex roof structures and small objects over the building roofs. For this purpose, a dataset comprising approximately 139,000 buildings with LOD2.2 roof structures and more than 2.2 million roof segments is generated using 8 cm GSD aerial imagery. An automated end-to-end workflow is developed, trained and tested from the available data. Experimental results indicate that roof structures suitable for LOD2.2 3D roofs can be extracted and vectorized with high accuracy, achieving 58.4% msAP and 73.1% mAPJ with ULSD network. Robustness is further assessed by visual inspection in areas affected by roof-blocking objects such as trees and cast shadows.

Topographic maps provide structured, polygonal representations of the Earth’s surface, delineating land-cover classes such as buildings, roads, water bodies, and vegetation.

They form the foundation of national geospatial data infrastructures and support a wide range of applications, including urban planning, environmental monitoring, and cadastral management. However, the production and maintenance of such large-scale topographic maps still rely heavily on manual photo-interpretation and vector editing.

While such human-in-the-loop workflows ensure geometric accuracy, they are labor-intensive, costly, and non-reproducible, limiting scalability and update frequency. However, most existing polygonal outline extraction methods are restricted to single-class, which typically leads to overlaps, gaps, and inconsistent shared boundaries when extended to multi-class mapping. Moreover, few studies have demonstrated nationwide implementation or validation, leaving the scalability and generalization of current methods largely unexplored. To address these challenges, this study develops a fully automated framework for large-scale topographic mapping directly from high-resolution aerial imagery. The framework aims to produce seamless, multi-class topographic maps in a single run that remain topologically consistent across diverse urban and rural regions in the Netherlands and beyond.

Instance segmentation of individual trees from UAV-derived orthomosaics and DSMs remains challenging in dense planted forests in Japan because SfM-derived DSMs often have blurred crown boundaries and unstable quality. We propose a PFFT-based method that encodes the local DSM shape around treetop candidates and integrates it into Mask2Former to suppress unreliable candidates and improve crown separation.

Experiments on Abies sachalinensis plantation (Todo fir) data from two sites in Hokkaido showed that the method improved mAP75 from 52.18% to 55.47% and F1 at a confidence threshold of 0.5 from 89.86% to 92.08%, while reducing false positives by 41% without increasing false negatives. The results indicate that treetop-centered local shape cues are useful for instance segmentation in densely planted forests.

This study presents an integrated framework combining YOLOv9e-Seg and photogrammetric geometric analysis for detecting road-safety assets and assessing their condition using UAV imagery. Traffic cones and tubular markers, which define construction-zone boundaries, are difficult to detect due to their small size in high-resolution images. To address this, a crop-tiling strategy (512×512 pixels) was applied to enhance the representation of small objects. Polygon-based labeling was used to preserve fine object geometry, and YOLOv9e-Seg was trained to output instance masks and polygon coordinates. During testing, tiled predictions were restored to the global coordinate frame, and duplicate detections were removed by retaining only the highest-confidence results. Geometric analysis utilized segmentation-derived polygons to compute centroids and principal axes, distinguishing intact and damaged tubular markers through vector angle difference analysis. For traffic cones, convex hulls constructed from centroid positions accurately delineated construction-zone boundaries. The proposed approach achieved the highest F1 score at a 512-pixel tile size, improving detection and segmentation of small, slender objects. These results demonstrate that the framework goes beyond basic detection and segmentation by enabling quantitative geometric interpretation and reliable construction-zone reconstruction from UAV data.

Satellite imagery enables large-scale remote sensing applications by providing frequent and large-scale coverage. However, its limited spatial resolution often restricts the use of satellite images in tasks that require detailed, fine-scale information. In contrast, aerial images offer a much higher spatial resolution, allowing the extraction of fine-grained features, but typically cover smaller, more localized areas. In this work, we investigate whether super-resolution (SR) methods can bridge the gap between aerial and high-resolution satellite imagery, enabling a label-free model transfer without additional manual annotations. The idea is to enhance the spatial resolution of high-resolution satellite images, allowing models trained on aerial data to be directly applied to satellite images. Towards this goal, a state-of-the-art SR algorithm is used to upscale three high-resolution satellite images, matching the resolution of the aerial training data. Then, a segmentation network trained on an aerial image dataset is applied to segment roads and parking areas in the super-resolved satellite images. The approach is evaluated on an annotated dataset and compared to the results in the original satellite images. Additionally, we investigate its performance on a low-resolution aerial image. Our results demonstrate that SR facilitates the utilization of models trained on aerial image datasets for large-scale satellite applications without requiring new labels.

This study tackles multilingual and explicit-implicit gaps in Visual Grounding (VG) for UAV imagery, focusing on real-world UAV needs (e.g., disaster response) that require implicit reference understanding. It evaluates Qwen2.5-VL-7B’s cross-linguistic robustness via Acc@0.5% across nine languages (Chinese, English, Japanese, Russian, Korean, German, French, Spanish, Portuguese). Key results: Explicit VG (using visual attributes) outperforms implicit VG (needing context/common sense) universally. East Asian languages lead in both tasks; Indo-European languages (e.g., Portuguese, 48.63% implicit accuracy drop) lag. Attention analysis shows the model better aligns with East Asian linguistic structures. This work informs LVLM optimization for multilingual UAV applications, guiding future cross-model comparisons.

This invited talk showcases two linked advances in Canadian urban remote sensing from the University of Waterloo. The first work presents large-scale 3D urban scene reconstruction and point-cloud densification using Gaussian Splatting with Google Earth Studio imagery. It recovers geometry and photorealistic radiance for the Kitchener–Waterloo region, benchmarking against NeRF baselines and achieving higher view-synthesis quality with faster training. The study demonstrates practical pipelines for city-scale digital twins and urban analytics.

The second study advances building-level reconstruction through the Gaussian Building Mesh (GBM) framework. GBM automatically extracts metrically accurate 3D building meshes from open-access imagery using segmentation models such as SAM2 and GroundingDINO, combined with Gaussian Splatting for dense, photorealistic surface generation. This pipeline enables efficient, data-driven modeling of urban structures, supporting applications from municipal infrastructure documentation to heritage reconstruction.

Together these contributions deliver scalable 3D reconstruction, object-level meshing, and data-driven urban modeling. They strengthen Canada’s leadership in remote sensing research and support resilient urban planning, infrastructure monitoring, and Earth observation–driven decision systems for Canadian cities.

Fine-resolution urban land-cover (ULC) mapping is essential for understanding intra-urban heterogeneity and monitoring rapid land-use change. However, large-area mosaics from CubeSat constellations such as PlanetScope often suffer from strong radiometric inconsistencies caused by varying sensor calibration, viewing geometry, and illumination, leading to unreliable classification and visual artifacts. This study introduces a Semantic-Aware Harmonization Model (SAHM) that jointly addresses spectral and semantic inconsistencies across multi-source imagery. SAHM integrates two synergistic components: a Spectral Harmonization Module (SHM) for radiometric alignment between PlanetScope and Sentinel-2 imagery, and a Semantic Consistency Module (SCM) inspired by prompt-based architectures to enforce category-level coherence. Through bidirectional interaction, semantic features guide spectral correction, while harmonized representations improve segmentation reliability. Applied to the Toronto and Region Conservation Authority area (TRCA), SAHM achieved an overall accuracy of 91.9%, with F1-scores exceeding 94% for impervious surfaces and 97% for agriculture. Harmonized PlanetScope mosaics demonstrated high spectral fidelity (PSNR = 34.2 dB, SSIM = 0.93) and reduced inter-scene NDVI/NDWI bias (< 0.05). The results highlight SAHM’s capability to produce spatially coherent, semantically reliable urban maps from radiometrically inconsistent high-resolution imagery. This framework offers a scalable solution for consistent urban monitoring across CubeSat constellations, paving the way toward semantic-driven harmonization in next-generation Earth observation.

Urban trees enhance the urban environment through various ecosystem services. Individual tree delineation and species classification provide information on the location, structure, and species of each tree from remote sensing datasets, offering valuable data support for efficient and refined urban greening management. However, existing individual tree delineation algorithms developed based on single-source remote sensing datasets struggle to address the complexity of urban green landscapes, such as conifer-broadleaf mixtures, tree-shrub complexes. Additionally, the relationship between classification accuracy and individual tree delineation quality remains unclear.

This study integrates the Segment Anything Model (SAM) and Marker-Controlled Watershed Segmentation (MCWS), combining imagery and LiDAR features, to optimize individual tree delineation in complex urban landscapes. Species classification was then performed on crown datasets from different algorithms to investigate how classification accuracy responds to varying crown qualities.

The results demonstrate that the proposed SAM-WS algorithm effectively enhances individual tree delineation accuracy, achieving the highest F1-Score of 0.75, with improvements of 0.20 and 0.27 over SAM and MCWS, respectively. The classification accuracy based on SAM-WS crowns was the highest among all algorithm-derived crown datasets, with an Overall Accuracy (OA) of 0.79 and a Kappa of 0.64. As the average F1-Score of crown delineation dropped from 1.00 to 0.48, the OA for classification decreased from 0.86 to 0.74, and Kappa from 0.77 to 0.38. Additionally, the classification accuracy of conifers and shrubs was more sensitive to the crown quality. This research offers new methodologies and insights into the application of remote sensing-based urban vegetation monitoring.

This study develops a satellite-based framework for detecting invasive shrub presence, focusing on common buckthorn (Rhamnus cathartica), across urban woodland environments in southern Ontario. Invasive shrubs exhibit extended leaf phenology compared to native understory species, leafing out earlier in spring and retaining foliage later into fall. Leveraging this phenological contrast, the workflow integrates multi-season Sentinel-2 MSI composites with higher-resolution PlanetScope imagery, combined with 2025 field observations collected across mixed-canopy woodlands in the Greater Toronto Area.

Spectral features (NDVI, EVI, NDWI, red-edge indices, Tasseled Cap transformations) and contextual variables (distance to woodland edges, canopy-openness metrics) are incorporated into a Random Forest classifier designed to distinguish buckthorn presence under complex understory conditions. A presence-background training strategy and spatially blocked cross-validation are implemented to reduce label uncertainty and spatial autocorrelation.

Preliminary results show that early-spring and late-fall imagery substantially improve detection sensitivity, with late-season spectral indices supporting the hypothesis that extended leaf persistence is a reliable cue for invasive shrub identification. This cost-effective workflow highlights the potential of multi-sensor satellite data to support early warning, invasion-risk mapping, and more efficient allocation of ground-validation efforts in urban conservation planning.

Drone remote sensing offers unique potential for capturing fine-scale variations in land surface temperature and vegetation condition, two tightly coupled variables that jointly regulate surface energy balance, evapotranspiration, and local microclimates. Understanding their interactions is crucial for assessing ecosystem function, evaluating the impacts of land use, and informing nature-based climate adaptation strategies. Yet, despite growing interest, UAV-based thermal and multispectral data have largely been used individually, and their integration for quantifying coupled seasonal dynamics in vegetation function and surface temperature remains limited. To address this gap, this study introduces a commercial off-the-shelf dual-drone multisensory data collection framework. The system integrates thermal infrared and multispectral imaging to analyze seasonal variations in surface temperature and vegetation health. The study area is a suburban-naturalized mixed landscape located at the University of Toronto Mississauga, Canada. Ten monthly drone flights were conducted from August 2024 to August 2025, with thermal and Normalized Difference Red Edge (NDRE) indices mosaiced for analysis. Results revealed distinct seasonal patterns, with impervious surfaces consistently exhibiting the highest surface temperatures, followed by vegetation and water, which were the coolest. NDRE values exhibited summer maxima and winter minima, aligning with the expected phenological cycles of vegetation. Regression analyses indicated that higher NDRE generally corresponded to lower surface temperatures, particularly for maintained trees and evergreen vegetation, highlighting the role of vegetation in moderating local heat. The developed workflow demonstrates the potential of drone-based remote sensing for cost-effective, fine-scale, multi-temporal environmental monitoring. It provides an adaptable framework for future applications in microclimate assessments.

Aging bridge infrastructure requires efficient, network-scale monitoring, especially in remote areas where traditional in-situ sensors are costly and logistically challenging. This paper presents a remote sensing framework for structural health monitoring based on spaceborne Synthetic Aperture Radar (SAR). The approach combines Persistent Scatterer Interferometry (PSI) and Least Squares Collocation (LSC), implemented through the PHASE open-source MATLAB software, to derive a millimeter-level spatio-temporal displacement model.

The methodology is applied to a reinforced-concrete viaduct in the Alpine foothills of Lombardy, Italy, using five years of Copernicus Sentinel-1 data. A custom elevation-based spatial filtering strategy enables the isolation of structural displacements from the surrounding topography. The resulting spatio-temporal displacement model captures the expected seasonal thermal behavior of the structure and highlights localized deviations from the dominant cyclic response.

Finally, the SAR-derived model is integrated with UAV photogrammetry and official inspection reports within the P.O.N.T.I. 3D viewer. This multi-source, Digital Twin-like environment facilitates the joint interpretation of remote sensing observations and in-situ evidence, providing a scalable framework to support infrastructure monitoring and management.

Large-scale land surface deformation monitoring using Interferometric Synthetic Aperture Radar (InSAR) requires robust detection of changes in long-term deformation trends. However, accurate change point (CP) detection remains challenging due to complex time series characteristics, including seasonal and quasi-periodic components and noise. Classical methods and many existing deep learning approaches rely on restrictive assumptions and training data that do not fully represent real-world InSAR time series, limiting their generalization and scalability in large-scale, real-world applications. In this study, we propose an integrated, fully supervised framework for CP detection in InSAR displacement time series based on Temporal Convolutional Networks (TCNs). The proposed TCN model employs dilated convolutions with multi-scale receptive fields to capture long-term temporal dependencies and complex deformation patterns, enabling robust identification of significant trend changes under noisy conditions. To effectively train this model, we introduce a deep learning-based InSAR time series simulation framework trained on real time series. This simulation framework produces physically consistent InSAR time series that retain essential temporal characteristics while introducing predefined, credible trend changes. Finally, we integrate the trained model into a large-scale anomalous change-detection pipeline that aggregates detected CPs from individual time series into spatially coherent deformation heatmaps suitable for operational monitoring. The proposed framework is evaluated using simulated data and real InSAR time series from the Phlegraean Fields caldera (Campi Flegrei), Italy. The results show clusters of anomalous behavior in the central Campi Flegrei–Pozzuoli area and in parts of Ischia and Procida, consistent with known unrest zones, associated periods, and independent measurements.

Electro-optical (EO) imagery is intuitive but highly dependent on weather and illumination, whereas synthetic aperture radar (SAR) imagery provides reliable all-weather observations yet offers limited spectral information. To complement these modalities, recent studies have applied cGAN-based image-to-image translation for SAR-to-EO translation. However, side-looking SAR introduces spatial distortions such as foreshortening and layover that cause relative misalignment with EO imagery, undermining pixelwise supervision and yielding structural discrepancies between translated outputs and reference EO imagery. In this study, we propose Geometry-Conditioned Pix2Pix (GC-Pix2Pix), which explicitly conditions on projected local incidence angle (PLIA) information derived from SAR imagery to better preserve structure and alignment in translated EO imagery. The method is based on Pix2Pix and comprises a 2-branch generator and a PatchGAN discriminator. The generator consists of a main network that processes SAR polarimetric channels (VV, VH) and a conditioning subnetwork that extracts PLIA features. The subnetwork uses multi-layer convolutional blocks to capture local PLIA patterns, and the extracted features are then fused with features from the main branch and emphasized via a spatial attention module. For training and evaluation, we assembled a dataset over South Korea that combines Sentinel-1A GRD VV/VH with PLIA and Sentinel-2B Level-2A RGB imagery. We compared GC-Pix2Pix against representative baselines. Across multiple image quality assessment metrics and complementary qualitative analyses, the proposed approach consistently improved SAR-to-EO translation performance.

Cloud cover in multispectral imagery (MSI) significantly hinders early-season crop mapping by corrupting spectral information. Existing Vision Transformer(ViT)-based time-series reconstruction methods, like SMTS-ViT, often employ coarse temporal embeddings that aggregate entire sequences, causing substantial information loss and reducing reconstruction accuracy. To address these limitations, a Video Vision Transformer (ViViT)-based framework with temporal-spatial fusion embedding for MSI reconstruction in cloud-covered regions is proposed in this study. Non-overlapping tubelets are extracted via 3D convolution with constrained temporal span t=2, ensuring local temporal coherence while reducing cross-day information degradation. Both MSI-only and SAR-MSI fusion scenarios are considered during the experiments. Comprehensive experiments on 2020 Traill County data demonstrate notable performance improvements: MTS-ViViT achieves a 2.23% reduction in MSE compared to the MTS-ViT baseline, while SMTS-ViViT achieves a 10.33% improvement with SAR integration over the SMTS-ViT baseline. The proposed framework effectively enhances spectral reconstruction quality for robust agricultural monitoring.

This paper focuses on cross-modal image matching between Synthetic Aperture Radar (SAR) and optical imagery, a long-standing challenge due to disparate sensing physics, radiometric behaviour and geometric distortions. Beyond applicational needs in satellite data fusion and downstream mapping, the study is additionally motivated by the rapid advances of feature matching in the field of Computer Vision. Under a unified, lightweight pipeline, the authors evaluate a classical handcrafted baseline (SIFT) against modern deep matchers, including a modality-invariant approach (MINIMA), as well as a SuperPoint+LightGlue pipeline, using the SpaceNet9 dataset with provided ground truth. The aim is to assess each models' ability to establish reliable correspondences without retraining or modality-specific adaptation, aiming to provide practical guidance for other researchers working with SAR-optical fusion. The paper highlights where pretrained multimodal models already yield consistent correspondences, where they still struggle and outlines possible next steps.

Marine pollution, including Marine Debris and Oil Spills, poses a serious environmental threat that requires systematic monitoring. While satellite observations and machine learning models have been widely applied in this domain, the use of advanced deep learning techniques remains limited. To support progress in this area, we construct a new annotated Sentinel-1 SAR dataset derived from the MADOS Sentinel-2 marine pollution dataset, including labels for oil spills, sea surface, look-alikes, ships, and offshore platforms. We evaluate several deep learning architectures on this dataset, including traditional models such as U-Net, state-of-the-art segmentation models such as SegNeXt and domain-specific frameworks such as MariNeXt. Our results show that MariNeXt achieves the best performance with an F₁-macro score of 92.7%, significantly outperforming U-Net and SegNeXt. Qualitative analysis using paired Sentinel-2 imagery further validates these findings. The study also highlights the persistent difficulty of detecting marine debris in SAR imagery, particularly when complementary optical data are unavailable.

This paper introduces a coarse-to-fine cross-view localization framework that unifies image-level retrieval and geometry-level alignment within a single pipeline. The proposed approach first employs a Bird’s-Eye-View (BEV)-guided retrieval module to establish a perspective-consistent intermediary space, enhancing cross-view consistency and retrieval precision. In the fine stage, a geometry-aware alignment module estimates the 3-DoF pose through interpretable point-plane matching based on BEV correspondences. This hierarchical design bridges global retrieval and local geometric reasoning, achieving both scalability and high localization accuracy. Extensive experiments on the VIGOR benchmark demonstrate that the proposed framework achieves state-of-the-art performance in both retrieval and alignment, significantly improving end-to-end localization precision while maintaining computational efficiency.

Urban rooftop solar planning relies on solar radiation inputs, yet estimates vary across models and measurement methods. This study compares radiation estimates from ArcGIS Solar Analyst, NASA solar radiation values, in-situ observations from research-grade and personal weather stations, and SDT-based Solar Radiation Modelling. We derive hourly global horizontal irradiance (GHI) values from these solar radiation data centres, model building-level estimates, harmonise all sources through temporal alignment, and then evaluate the values. The comparison reveals the hourly modelling of solar radiation models and common solar radiation centres, highlighting where an urban-adjusted local sensor provides lower solar radiation values because of the limited representation of the built and urban environment. Results show that utilising gridded or terrain-based models over urban-adjusted solar radiation values overrepresent due to the uncaptured localised shadings, roof placement effects, and increasing systemic errors for downstream rooftop PV terrain-based assessments. The cross-validated workflow of sensor-based city-scale solar radiation modelling is reproducible and scalable, offering local governments a more nuanced understanding of their solar capacity, and paves the way for carbon emission budget management.

Deep learning, in particular, has driven hundreds of new studies in remote sensing each year. However, ensuring the reliability of these models requires robust uncertainty quantification, an aspect that remains insufficiently explored. Current remote sensing deep learning models typically yield single, deterministic predictions, such as a class label for each pixel or a single biomass value for a given location or region. While commonly used metrics such as RMSE or classification accuracy summarize overall model performance, they fail to convey the reliability of individual predictions, leaving users without guidance on how much confidence to place in each output. Uncertainty estimation addresses this critical gap by quantifying the variability or confidence associated with model predictions. This enables practitioners to interpret not only what the model predicts but also how confident it is in those predictions, providing a more nuanced understanding that is essential for informed decision-making.

We address aleatoric uncertainty using Sentinel-1 and Sentinel-2 time series, proposing two approaches: (i) Gaussian UC, which predicts mean and standard deviation, and (ii) Quantile UC, which estimates the 10th, 50th, and 90th quantiles to capture asymmetric errors. We evaluate these approaches on three representative EO tasks: building height, canopy height, and aboveground biomass estimation. Our results (ID and OOD) show that both models achieve accuracy comparable to deterministic benchmarks while providing well-calibrated, interpretable, and operationally useful confidence estimates. Notably, both models outperform existing global canopy height products on evaluated sites, including the recent 1 m canopy height maps produced by vision transformers.

Recent advances in remote sensing have shifted the focus from the analysis of individual image scenes to the understanding of complex earth systems. This means that the analysis of dynamic evolutions replaces previous static examinations for fixed time points. Furthermore, interdisciplinary research and the integration of heterogeneous data sources are characterizing this transformation process. Digital Elevation Models (DEMs) are predestined for supporting this process by supplementing orthophotos and map data. Promising applications include city planning, landslide analysis, and flood risk assessment where spatio-temporal change detection is a central concept to be applied. Concerning map data, the OpenStreetMap project, based on the idea of Volunteered Geographic Information, has revolutionized the effective production and update of digital maps. However, OSM data does not include elevation information and often contains incorrect geometric information. In this paper, we introduce a self-training framework for validating OSM building footprints with the aid of high-resolution DEMs. The framework supports building segmentation with a self-supervised approach to improve the representation of OSM building footprints. The availability of Digital Elevation Models is used to check the quality of OSM data. The applicability of the approach is demonstrated by a case study conducted in Karlsruhe, Germany. The promising results are described in detail. With our approach, change detection of OSM data can also be carried out using different temporal versions of DEM and OSM data.

Accurate and reliable road extraction from LiDAR data remains a major challenge when spectral cues are limited or spatial heterogeneity increases model uncertainty. This study introduces a comparative, entropy-driven framework for evaluating the performance and reliability of road asset detection using three supervised machine learning algorithms—XGBoost, Random Forest (RF), and Support Vector Machine (SVM). Using a high-density aerial point cloud, a reproducible computational pipeline was implemented, to help practitioners in real-world scenarios for selecting the most robust and reliable machine learning methods for large-scale road assets mapping. Beyond traditional accuracy metrics (Overall Accuracy, F1-score, and Kappa coefficient), uncertainty-based evaluation of the outputs has been conducted using KPIs of entropy and sensitivity to training sets to quantify model reliability and spatial instability. Results reveal that the inclusion of RGB significantly reduces entropy across all models. XGBoost achieved the lowest mean entropy (0.084–0.143) and the most consistent probabilistic behaviour, reflecting confident and well-calibrated model. SVM, while statistically the most accurate (OA and Kappa > 0.97), exhibited higher local entropy (≈ 0.23–0.26), implying precise yet less certain classification. RF demonstrated the highest entropy (≈ 0.65–0.70) and the greatest variability, underscoring its sensitivity to feature noise. Under the WOR configuration, mean entropy rose markedly—most for RF_WOR (≈ 0.93) and moderately for SVM_WOR (≈ 0.39)—while XGBoost retained low uncertainty.

Spatial entropy maps further highlighted that uncertainty concentrates along road edges with RGB data but expands diffusely under WOR conditions, emphasizing the critical role of spectral–spatial synergy in constraining ambiguity. entropy-based evaluation provided insights beyond conventional accuracy metrics, revealing paradoxes between correctness and confidence.

Efficient processing of large-scale 3D point clouds acquired from Terrestrial or Airborne Laser Scanning (TLS/ALS), presents a significant computational challenge. While transformer-based architectures excel at modeling the global context crucial for interpreting these complex scenes, their quadratic computational complexity makes them infeasible for direct application on massive point sets. To address this scalability bottleneck, we propose the Spatio-Sequential Point Transformer (S2PT), a novel hierarchical architecture for efficient and effective large-scale point cloud processing. Our approach begins by serializing the point cloud into an ordered sequence, which enables the use of attention with linear complexity. This not only circumvents the quadratic bottleneck of standard transformers but also establishes a global receptive field at every layer. To compensate for potential information loss during serialization, we further introduce a novel Spatio-sequential Positional Encoding (S2PE) that synergistically combines 3D local geometric features with 1D sequential order information, enhancing the model’s spatial awareness. Experiments on multiple benchmarks demonstrate that S2PT achieves performance comparable to state-of-the-art methods while being significantly more efficient during training and inference, offering a promising path towards scalable representation learning for large-scale 3D scenes.

Accurate semantic segmentation of 3D point clouds is a fundamental task in photogrammetry, robotics, and large-scale scene understanding. Despite recent advances in point-based architectures such as PointNeXt, segmentation performance remains limited near semantic boundaries, where local neighborhoods often contain points from multiple classes, leading to feature ambiguity and oversmoothing.

In this paper, we propose a lightweight boundary-aware learning framework that explicitly models boundary regions during training. Boundary supervision is automatically derived from local semantic label disagreement, eliminating the need for additional annotations. An auxiliary boundary prediction head is introduced to learn boundary-sensitive features, which are subsequently integrated into the segmentation process through a late-stage feature fusion mechanism. In addition, a boundary-aware loss formulation emphasizes boundary regions during optimization, encouraging improved feature discrimination at class transitions.

Experimental results on the S3DIS dataset using the standard 6-fold cross-validation protocol demonstrate consistent improvements over the PointNeXt baseline. The proposed method achieves gains of 3.22% in mean Intersection over Union (mIoU) and 2.85% in mean class accuracy (mACC), with notably improved segmentation quality at object boundaries. Importantly, these improvements are obtained without modifying the backbone architecture or increasing inference complexity.

The results indicate that incorporating boundary-aware supervision provides an effective and efficient strategy for improving segmentation performance in challenging regions.

China is actively pursuing synergistic governance to address air pollution and carbon mitigation issues. This study, focusing on concentration as a key feature, assesses the synergy performance in the Yangtze River Delta Urban Agglomeration (YRDUA), revealing fluctuating trends with only seven cities showing improvement. To further understand the influences from local drivers, we employed an explainable spatial machine learning approach, integrating Geographical Weighted Regression (GWR), Random Forest (RF), and Shapley Additive Explanation (SHAP) to capture nonlinear, threshold, and interaction effects among explanatory variables. The analysis identifies longitude, SO2 emissions from industrial sources, wind speed, latitude, and the proportion of GDP from tertiary sector as the top five influencing factors, emphasizing the importance of geographical position, local air pollution emission, and meteorological condition. Most drivers exhibit nonlinear impacts and interactions with clear thresholds. Such as, wind speeds, exceeding 9.3 m/s negatively impact synergy. Furthermore, spatial heterogeneity of drivers' influence is evident across cities and regions. Specifically, cities along the eastern coast benefit from geographical advantages that enhance synergy in air quality improvement and carbon mitigation. Meteorological conditions, especially wind speed, significantly influence synergy, with notable differences between northern and southern coastal cities. These findings underscore the need for locally tailored governance strategies that leverage each city's unique geographical and socioeconomic attributes to enhance synergistic governance effectiveness. This research contributes to understanding the complex interplay of local drivers influencing synergistic governance in the YRDUA, providing valuable insights for policymakers aiming to improve air quality and promote sustainable development in rapidly urbanizing regions.

With the acceleration of global urbanization, dynamic change detection of urban buildings is vital for urban planning, resource management, and public safety. Traditional bi-temporal remote sensing-based methods fail to capture gradual building evolution and are prone to noise-induced missed detections and false alarms. While multi-temporal imagery provides continuous temporal information, its sequential and high-dimensional nature poses greater challenges. Existing deep learning models like CNNs excel at spatial feature extraction but lack temporal modeling, while LSTM/ConvLSTM struggles with spatial detail preservation and small-target recognition.

To address issues including insufficient temporal modeling, channel redundancy, weakened spatial attention, and small-target loss, this study proposes the Dual Attention-coupled ConvLSTM Network (DAL-UNet). Its encoder embeds a dual attention module: channel attention selects change-related features and suppresses redundancy, while spatial attention enhances key region responses to improve building edge and small-target discrimination. A fully convolutional LSTM module models temporal evolution while preserving spatial topology. The decoder adopts a dual-branch multi-task framework to optimize change feature upsampling and semantic segmentation, enhancing subtle change perception and spatial detail restoration.

Experiments on the SpaceNet7 dataset show DAL-UNet outperforms state-of-the-art methods, with maximum improvements of 13.04% in F1-score, 1.32% in Precision, and 16.52% in Kappa coefficient. It performs exceptionally in high-rise shadow areas and dense small-target regions, reducing shadow interference via attention mechanisms and alleviating class imbalance through class-weighted loss.

In the aftermath of a disaster, whether natural, industrial, or war-related, a rapid and accurate assessment of building damage is crucial for rescue forces to conduct an effective emergency response. Very high-resolution satellite imagery enables such assessments and serves as an important indicator for understanding the scale of destruction, supporting time-critical rescue operations, and guiding resource allocation. While deep learning models have shown promising results in automating building damage assessment (BDA) from pre- and post-disaster optical satellite imagery, they often fail to generalize to new disasters due to domain shifts. This paper studies the challenge of rapid domain adaptation for BDA in the context of the war in Ukraine. We create a new, challenging dataset annotated with damage grades across six cities in Ukraine, using pre- and post-disaster optical imagery. To facilitate rapid adaptation, we propose an efficient fine-tuning workflow using Low-Rank Adaptation. Our experiments show that this approach substantially improves performance in both out-of-domain and in-domain settings, presenting a practical and data-efficient study for deploying BDA models in time-critical emergency scenarios.

This study examines district-level urban expansion and ecological change in the Hills Shire LGA using multitemporal Landsat imagery, Shannon’s entropy, RSEI, and hotspot analysis to identify spatial patterns of growth and environmental stress.

Urban sprawl, characterized by land discontinuity, low population density, and inefficient land use, hinders sustainable urbanization, particularly in rapidly growing regions such as Bulacan, Philippines. This phenomenon places strain on existing infrastructure, contributes to environmental degradation, and exacerbates socio-economic disparities. While previous studies have analyzed urban sprawl, these often neglect the integration of socio-economic factors, thereby reducing the accuracy of their analysis and policy relevance for developing regions. This research seeks to analyze urban sprawl patterns within Bulacan through the integration of socio-economic variables and identify key factors driving this sprawl. The study employs urban sprawl analysis, using the Multidimensional Urban Sprawl Index (USI) to assess land discontinuity, population density, and land use efficiency. Additional analysis using fractal analysis and factor analysis through Geodetector was also employed. The study found a positive shift toward more efficient, compact growth in Bulacan from 2005 to 2020, though mild and severe sprawl remain ongoing challenges. Fractal analysis revealed that complex urban forms encourage infill, while open areas are prone to leapfrog development. Land use benefit and road access consistently drove sprawl, with key factors like population and proximity to the city center changing over time. The study recommends stricter enforcement of zoning regulations to mitigate fragmented growth and the integration of additional socio-economic indicators (e.g., GDP, employment rates, and land values) into future analysis.

This study demonstrated that a two-stage architecture—comprising a segmentation model followed by a classification model—is effective for embankment extraction. By constructing a large, wide-area training corpus from medium-resolution SPOT imagery, transfer learning to higher-resolution satellites (e.g., Pleiades) was readily achieved. For operational use, exhaustively proposing candidates with the AI model and inserting a brief human check (embankment/non-embankment) per candidate can reduce false positives while limiting missed detections, making the approach sufficiently practical for deployment.

Land-sea segmentation is crucial for tasks such as marine target detection and coastline extraction in remote sensing imagery. However, complex and diverse background environments and land-sea boundaries can easily lead to inaccurate segmentation. To address this issue, a high-precision land-sea segmentation model based on the deep Otsu method is proposed. This method first utilizes our proposed remote sensing image texture enhancement algorithm based on Retinex theory and the Canny operator to enhance the remote sensing image and its edge information, further improving the segmentation accuracy of the land-sea boundary. Then, we combine deep learning concepts, the maximum inter-class variance method, and our proposed density space clustering method based on the difference innovation optimization algorithm to propose a deep maximum inter-class variance method for segmenting the ocean and land in the image. Simultaneously, an adaptive multi-scale fragmentation region removal method is proposed to remove small, fragmented regions extracted during the segmentation process. Experimental results show that the proposed method achieves an overall prediction accuracy of 98.41% and an average intersection-union ratio of 96.07%, demonstrating its ability to effectively perform land-sea segmentation tasks.

Low-resolution imagery is a major constraint for remote sensing tasks (e.g., urban land cover detection) where accurate classification of buildings, roads, vegetation, and small objects is required. Deep learning-based segmentation models are highly sensitive to image quality, resulting in degraded performance on low-resolution inputs. Super-Resolution (SR) techniques offer a promising solution by enhancing image fidelity to support downstream tasks. This work applied MAPSRNet, a Multi-Attention Pyramid SR Network to aerial images used for multi-class land cover detection. Evaluated on the ISPRS Potsdam dataset, MAPSRNet achieves state-of-the-art SR performance with PSNR of 32.92 dB and SSIM of 0.87, outperforming existing methods such as SRCNN (31.54 dB, 0.83) and DRRN (31.03 dB, 0.82) while maintaining competitive inference speed. Beyond image quality, MAPSRNet significantly improves multi-class land cover segmentation when integrated with a ConvNeXtV2-based U-Net, achieving an overall accuracy of 80.60%, mean IoU of 62.54%, and FwIoU of 68.34%, surpassing not only low-resolution inputs (Overall Accuracy: 65.28%, mIoU: 40.20%, FwIoU: 50.12%) but also high-resolution(HR) ones (Overall Accuracy: 80.50%, mIoU: 62.40%, FwIoU: 68.01%), especially in certain classes such as impervious surface and clutter. These results demonstrate that perceptual and structural fidelity, rather than pixel-level similarity, can drive superior performance in urban land cover segmentation. MAPSRNet offers a practical solution for scenarios where HR imagery is limited or unavailable, highlighting its potential for large-scale remote sensing applications.

This paper proposes the development of a multi-agent system (MAS) for Building Information Modeling (BIM) environments, where users interact with a 3D model and a chat-bot to query, validate, and analyze building elements. By leveraging conversational AI and modular agents capable of semantic understanding and geometric computation, this system allows users to retrieve data, perform quality checks, and visualize computed results directly using the BIM information. The approach supports diverse tasks, from attribute completion and filtering to volumetric calculations, thus enabling a more intelligent and accessible BIM experience for analytical purposes.

The adoption of innovative digital heritage workflows in the Architecture, Engineering, and Construction (AEC) sector faces significant challenges, particularly in integrating digital survey data with Building Information Modeling (BIM) into a unified model. This paper begins with a literature review that outlines the geometric and software-environment constraints complicating such integration and examines various proposed solutions, with particular attention to open-source tools and standard formats.

Building on this foundation, the paper introduces an innovative two-stage method: (1) segmenting, classifying, and enriching digital survey data into a BIM model; and (2) developing a web viewer that hybridizes this BIM model with the original survey data. The proposed workflow relies exclusively on open-source tools and open standards, with Industry Foundation Classes (IFC) used as the native editing format. A seamless continuity is established between the Bonsai add-on for Blender, used as a BIM authoring environment, and the web library That Open Engine, which serves as a dissemination tool enabling interactive querying of BIM data within a web browser. This library shares a common dependency on Three.js with 3DTilesRendererJS, allowing the overlay of a tiled photomesh of the asset. This integration enables the combination of an accurate geometric and visual representation with structured metadata interaction within a unified web environment. Overall, the proposed approach provides a robust and flexible framework for supporting practical applications such as dissemination, documentation, and diagnostic studies of heritage assets.

This study presents a comprehensive and semi-automated framework for generating multi-Level of Detail (LoD) 3D building models using multi-source LiDAR point clouds to support digital twin development. By integrating airborne, drone-based, mobile, and terrestrial LiDAR platforms, the framework addresses limitations of single-source datasets and enables scalable reconstruction across urban and building scales. A robust preprocessing workflow—encompassing subsampling, denoising, colorization, and two-stage registration—significantly enhances point-cloud quality and achieves seamless fusion of heterogeneous datasets with millimetre-level accuracy.

The framework supports outputs ranging from city-scale footprints (LoD0) to detailed parametric building models (LoD4), enabling applications in smart city planning, facility management, and heritage documentation. A knowledge-based segmentation layer further enables the creation of “Smart Point Clouds,” facilitating component-level querying and efficient generation of floor plans, elevations, and façade models. Real-world evaluations in downtown Toronto demonstrate high accuracy and strong computational performance, with LoD0–LoD2 models produced in minutes on a standard workstation.

By ensuring compatibility with CityGML and IFC standards, the framework enhances interoperability within digital twin ecosystems and supports integration with simulation and decision-support systems. While detailed LoD3–LoD4 modeling still requires manual refinement, the workflow establishes a foundation for future automation through AI-driven segmentation and cloud-based parallel processing. Overall, this research advances scalable 3D modeling practices and provides a practical pathway toward comprehensive, data-rich digital twins for smart cities.

With the Czech Republic's impending "BIM Act" driving the creation of a basic built environment model, the study proposes a compliant workflow for incorporating 3D models of two key vegetation feature types from the fundamental geographic vector database: "Forest ground with trees" and "Significant or lonely tree, grove." Modelling relies on nationwide datasets, the digital terrain model, the digital surface model based on image matching of aerial imagery, and supplementary aerial laser scanning data. For the forest features, the process comprised optical point cloud filtration and constrained triangulation, resulting in height-extruded forest base polygons with canopy cover tops. The 3D representation uses MultiSurface geometry, recorded as a PlantCover object in CityGML/3DCityDB, and is in line with the LoD2 standard for buildings. For solitary trees, predefined prototypes were scaled and positioned based on individual tree detection and parameters extracted from point clouds. Features were mapped to the CityGML/3DCityDB SolitaryVegetationObjects class, utilizing Implicit geometry to optimize for data volume and visualization speed. While the digital surface model, which can be easily generated from periodically acquired optical imagery, was sufficient for the forest features, aerial laser scanning data was superior in individual tree modelling. The number of extractable parameters increases with point density and is dependent on the platform used. However, the availability of such higher-density laser scanning data in Europe is limited and varies across countries and regions. The results demonstrate the generation of LoD2 compliant 3D models from nationwide datasets for both vegetation features, visually enriching the basic built environment model.

Despite the rapidly evolving and widely adopted tools in the Architecture, Engineering, Construction, and Operations (AECO) industry, Construction Supply Chain Management (CSCM) remains a fragmented practice with poor integration and interoperability between Building Information Modelling (BIM), Geographical Information Systems (GIS), and logistics systems. This research aims to bridge the gap between BIM, GIS, and logistics information by developing a unified, data-informed Digital Twins (DT) framework necessary to support multi-criteria decision-making (MCDM) in CSCM. They key characteristics of this work include: (1) a repeatable integration for heterogenous BIM-GIS environments powered by IoT networks; (2) a short-horizon predictive module optimized for construction logistics and Just-in-Time (JIT) delivery; and (3) a democratized analytics interface.

This study introduces an automated decision-support tool implemented on Google Earth Engine for mapping vegetated shorelines using Sentinel-2 NDVI. The tool enables reproducible diachronic coastline extraction, rapid processing of large datasets, and supports coastal change monitoring and management applications.

Average rates of shoreline change are key indicators for assessing coastal evolution. The study area is located in Bragança, on the northeast coast of Pará, Brazil, covering urban, estuarine and natural areas. Between 1984 and 2024, despite a general trend of increasing coastline, areas with increasing human occupation experienced significant coastal erosion, causing building retreat, partial loss of homes, and damage to beach access roads. Using the Digital Shoreline Analysis System (DSAS) and time series of dense satellite images processed in Google Earth Engine, the coastline was analyzed in the study area. As a result, the average linear rate of variation showed a slight general retreat of the coastline, accompanied by high morphodynamic variability and low statistical consistency in linear trends. Urbanized sectors exposed to ocean forces were the most vulnerable to erosion, while estuarine and mangrove areas were more stable. The high supply of sediments from the estuaries contributed positively to the addition of the coastline in several regions. These findings emphasize the importance of strategic coastal management considering natural and human influences on shoreline dynamics.

This study proposes a practical pipeline for cross-sensor harmonization and short-term temporal estimation of mangrove leaf reflectance using multi-platform hyperspectral data. We combine laboratory (HySpex VNIR-1800; Days 1/3/7), field (Specim IQ; Day 1), and UAV (Cubert X20 Plus; Day 1) measurements over 400–900 nm for three species (Ceriops tagal, Avicennia marina, A. germinans). Field and UAV spectra are interpolated to the HySpex grid, and multiplicative change factors derived from HySpex Day-1→Day-3/7 trends are used to estimate later-day reflectance for non-lab sensors. Accuracy is assessed via RMSE and Pearson’s r, with focus on chlorophyll-sensitive regions (~450, 680, 720–750 nm). Systematic platform effects appear: in-field spectra exceed HySpex by ~2.5% (A. germinans), ~5.7% (A. marina), and ~11.5% (C. tagal), while HySpex exceeds UAV by ~4.38%, ~7.89%, and ~11.5%, respectively. After harmonization, temporal consistency is strong for A. germinans (RMSE ≈0.047–0.050; r ≈0.958–0.981) and solid for A. marina (Specim RMSE ≈0.066–0.081; r ≈0.943–0.970), with higher UAV variability. Spectral trajectories track post-harvest stress: ~15–20% decline near 680 nm for C. tagal and ~10% for A. germinans, alongside expected green and red-edge/NIR shifts. The workflow enables comparable, temporally resolved spectra across instruments, supporting scalable vegetation phenotyping and long-term mangrove monitoring where single-sensor continuity is limited.

Coastal vegetation provides essential protection against shoreline erosion, wave action, storm surge, and supports biodiversity in low-lying tidal environments. This research discusses methods of using UAS based hyperspectral and multispectral sensors and a deterministic Spectral Information Divergence approach to monitor and preserve the ecosystem in coastal environments. The work focusses on implementing the methodology for monitoring different species of mangrove in a protected natural area located in Florida, USA. The achieved accuracy of 90% proves the ability of UAS based remote sensing system to support a resilience-based restoration and long-term monitoring.

Deforestation remains a critical global challenge with profound implications for food security, ecosystem resilience, and disaster risk reduction. In Sierra Leone, the Western Area Peninsula National Park (WAPNP), one of the country’s last remaining tracts of primary tropical moist forest, faces increasing pressures from illegal logging, mining, and land encroachment despite legal protection since 2012. These activities threaten essential ecosystem services, including water provision, fertile soils, and local climate regulation, while exacerbating vulnerability to floods, landslides, and droughts.

This study evaluates the extent of WAPNP’s closed-canopy forest cover using Sentinel-2 imagery from 2020 to 2024, complemented by very-high-resolution (VHR) data and ground-truth observations for validation. The analysis identifies the main human drivers of forest loss and maps the spatial distribution and remaining extent of forest cover within the park. The results highlight the power of combining Copernicus Sentinel-2 imagery with open-access forest datasets to provide a reproducible, and cost-effective monitoring of forest cover in data-limited tropical regions, offering a valuable tool for conservation planning and management.

The expansion of irrigated crops such as almond and pistachio in arid and semi-arid regions poses a challenge in a context of water resource scarcity. Understanding crop water requirements across large areas has become feasible thanks to remote sensing techniques and the growing availability of satellite imagery with increasingly higher spatial and temporal resolution. However, models have shortcomings that lead to uncertainties in their estimates. In this study, we introduce the model ensemble technique as a method to constrain uncertainty in crop water requirements, with a particular focus on woody crops. This study is centered in the province of Albacete, for the period 2022–2024, and combines two surface energy balance models, METRIC and SenET_TSEB, with a water balance model asssited by NDVI imagery to obtain time series of daily actual crop evapotranspiration (ETa), with a spatial resolution of 20–30 meters. Comparison with in situ measurements recorded at two eddy-covariance towers located in almond and pistachio orchards shows better correlation of the results using the ensemble. At a weekly scale, an average error of 4.9 mm d⁻¹ and 2.8 mm d⁻¹ are obtained for the almond and pistachio crops. Accumulated ETa values over the growing season are consistent and provide confidence to assist in irrigation scheduling, detect stress conditions, and/or quantify water needs at a plot scale. These results reinforce the role of satellite remote sensing in water resources management, in particularly relevant crops for our region such as almond and pistachio orchards.

This study verifies the effectiveness and advantages of spaceborne GNSS-R technology for VWC retrieval, and clarifies that the intercept feature of vegetation observations and Γpeak reflectivity are the core components for constructing high-precision models. The proposed method provides a new technical means for large-scale and efficient VWC monitoring, and has positive significance for improving the assessment of vegetation health and disaster risks.

Aerosol size information, typically represented by fine- and coarse-mode aerosol optical depth (fAOD and cAOD), is crucial for understanding anthropogenic emissions and radiative effects. However, satellite-based retrievals suffer from limited labeled data and high uncertainty over land. To address these challenges, we developed a novel deep pretraining framework capable of mining latent representations from unlabeled satellite pixels, thereby enhancing the accuracy and generalization of aerosol size information retrieval. The framework leverages a self-supervised pretraining stage to capture intrinsic spatiotemporal correlations in multispectral satellite data and transfers these latent features to a supervised fine-tuning model. Using MODIS data combined with AERONET observations, our pretrained model achieved a 10% improvement in correlation and a 15% enhancement in regions without ground observations compared to conventional deep-learning models. The retrieved global fAOD from 2001–2020 reveals a significant decreasing trend (−1.39 × 10⁻³ yr⁻¹), with regional differences—most notably, a threefold stronger decline over China than the global average. These results demonstrate that deep pretraining can effectively exploit unlabeled satellite information, bridging the gap between sparse ground networks and dense global observations, and offering a transformative approach for large-scale aerosol characterization and climate studies.

Atmospheric aerosols are a key factor influencing the Earth's radiation balance and climate change, and the accuracy of their retrieval is crucial for environmental monitoring and climate research. FY-4A AGRI, with its high-frequency observation capability, can provide aerosol data at high temporal resolution. Combined with deep learning technology, it enables efficient monitoring of dynamic aerosol variations. This study develops a retrieval algorithm for aerosol optical and microphysical parameters based on the Transformer deep learning model, specifically designed for the FY-4A geostationary satellite. The algorithm achieves multi-parameter collaborative retrieval of aerosol optical depth (AOD), fine/coarse-mode aerosol optical depth (FAOD/CAOD), and single scattering albedo (SSA). This research overcomes the reliance on prior assumptions inherent in traditional physical retrieval methods. By integrating multi-band spectral features, geometric observation parameters, and data from 104 AERONET sites, it significantly enhances retrieval accuracy under the complex surface conditions of East Asia. Experimental results demonstrate high accuracy in validation against AERONET sites, with correlation coefficients of R=0.915 for AOD, R=0.897 for FAOD, R=0.851 for CAOD, and R=0.536 for SSA. Comparative validation of various aerosol product spatial distributions highlights the advantages of the proposed algorithm in capturing aerosol diurnal variations (such as haze dissipation processes) and extreme events (e.g., dust storms and biomass burning). This study provides a new technical approach for regional air quality monitoring and climate effect assessment, advancing the application of China’s geostationary meteorological satellites in aerosol monitoring.

Heavy metal pollution presents a formidable challenge to global ecosystems, threatening biodiversity, soil and water quality, and human health. The atmosphere serves as both a source and long-range conveyor of bioaerosols, complex particles that include bacteria, fungal spores, and dust-bound heavy metals, profoundly influencing biosphere health and ecosystem function. In this study, we investigate atmosphere-biosphere interactions in Pin Valley National Park, a cold desert ecosystem in the Western Himalayas, by analyzing how bioaerosol-mediated deposition of heavy metals shapes vegetation stress and phytoremediation dynamics. Integrating field spectroscopy, in-situ chemical analysis (ICP-MS), and multi-temporal satellite data, we mapped heavy metal hotspots (Pb, Cd, Ni, Cr) and linked them to shifts in vegetation health and thermal indices. We observed significant spatial overlap between elevated metal concentrations likely introduced via long-range atmospheric transport and suppressed vegetation indices. Phytoremediator species such as Brassica juncea and Populus exhibited strong metal uptake, revealing natural biospheric buffering capacity against airborne contaminants. Additionally, iron oxide and hydrothermal indices indicated that soil mineral conditions, modulated by deposition, may influence microbial and root zone dynamics. This multidisciplinary assessment underscores the role of the atmosphere not merely as a depositor but as a dynamic bioreactor influencing terrestrial microbiomes and plant stress responses. By offering a scalable, remote sensing–assisted framework for monitoring ecosystem health and contaminant transport, our work directly supports SDG 13 by identifying atmospheric pathways of pollutant stress under warming trends, contributes to SDG 15 by protecting fragile alpine ecosystems through phytoremediation, and aligns with SDG 17 as an interdisciplinary approach.

The relationship between aerosol optical depth (AOD) and surface PM2.5 concentrations remains a significant difficulty in remote sensing-based air quality assessments due to meteorological conditions and aerosol vertical structure. This relationship is investigated using daily observations from 2021 to 2024 in Sherbrooke, Quebec, Canada.

Ground-based AERONET AOD500 and satellite-based MAIAC AOD at 550 nm are analyzed separately, together with surface PM2.5 measurements from a local PurpleAir sensor. Meteorological parameters such as relative humidity, boundary layer height, temperature, and wind speed are available from ERA5 reanalysis. Vertically resolved aerosol information from MPLNET lidar is used to identify elevated aerosol layers associated with transported wildfire smoke.

The approach combines Pearson and Spearman correlations, partial correlation analysis, multivariate regression, and Random Forest (RF) modeling to capture nonlinear interactions. Results indicate weak but statistically significant correlations between AOD and PM2.5 (r ≈ 0.26-0.30), with stronger monotonic relationships. A pronounced seasonal dependence is observed, with the strongest coupling in autumn and weak or insignificant relationships in winter. Partial correlation analysis suggests that a residual association between AOD and PM2.5 remains after accounting for meteorological influences.

RF models improve predictive performance (R² ≈ 0.39), although performance degrades in winter. Sensitivity analysis indicates that transported smoke plumes can influence the AOD-PM2.5 relationship, particularly when partial mixing into the boundary layer occurs.

Over the past decade, nearly ten satellites dedicated to atmospheric CO2 concentration monitoring have been launched, significantly advancing our understanding of the global carbon cycle. In 2022, China launched the DaQi-1 (DQ-1) satellite, which carries the Aerosol and Carbon Dioxide Lidar (ACDL)—the first spaceborne lidar sensor for CO2 monitoring. Relying on laser-based active sensing, ACDL can detect global XCO2 at nighttime, serving as an important complement to existing passive optical CO2 satellite missions. This study aims to introduce the scientific community to the XCO2 retrieval methodology of ACDL and its initial XCO2 product. The first version of ACDL XCO2 products scheduled for release is called “v1.0”. This paper presents a comparison between XCO2 at daytime and nighttime. Nonetheless, challenges remain, including reliance on meteorological reanalysis data and uncertainties in spectroscopic parameters. In future product versions, we plan to improve data quality through enhanced denoising techniques and signal processing methods for low signal-to-noise ratio (SNR) cases. We hope that this initial ACDL XCO2 product will spark broader interest and participation from the scientific community, thereby contributing fresh momentum to climate change research.

Traditional forecasting methods of air pollutants show intrinsic limitations due to the complexity of atmospheric interactions. Recent research has moved toward the employment of artificial intelligence (AI)-based approaches and satellite data processing. The framework proposed in this study is a transfer learning (TL) model to estimate surface-level NO2 concentrations across multiple locations by using satellite and environmental data.

The approach integrates Sentinel-5P TROPOMI-derived tropospheric NO2 columns, meteorological variables (temperature, precipitation etc), spatial coordinates and temporal features.

A CatBoost regression model is implemented, leveraging a Leave-One-City-Out (LOCO) TL framework across five cities (Berlin, London, Madrid, Paris and Toronto) in the world. This enables the model transfer from multiple source domains to a new target city with minimal ground-based data. Experimental results are outperforming city-specific baseline models, by showing an increased prediction accuracy, a reduced Root Mean Square Error (RMSE) by approximately 7% and a Coefficient of Determination (R2) higher by 2.7%. Toronto, which represents an environment with a low monitoring density, benefits most from TL, with R2 improving from 0.58 (baseline) to 0.66 (transfer) and RMSE dropping from 6.44 µg/m3 to 5.84 µg/m3.

A detailed Leave-One-Block-Out (LOBO) ablation study shows how each group of features contributes to the performance of the model. Spatial coordinates and meteorological features are the most influential predictors of NO2 concentration, while the satellite NO2 data increase model generalization. These results highlight the potential of cross-city TL and remote sensing synergy for scalable urban air pollution monitoring, especially in limited ground-based monitoring scenarios.

High-resolution mapping of tropospheric ozone is essential for urban environmental assessment; however, satellite-derived ozone products are generally too coarse to capture neighborhood-scale variability in complex megacities such as Tehran. This study introduces an interpretable deep-learning framework that downscales coarse Sentinel-5P ozone observations to a 30-m spatial grid by integrating a U-Net convolutional architecture with SHapley Additive exPlanations (SHAP). A diverse suite of predictors—including land-surface indicators, meteorological parameters, terrain morphology, and chemical precursors—was harmonized and resampled to a unified spatial resolution. SHAP analysis was applied to quantify each predictor’s contribution, enabling the removal of redundant or low-impact variables before model training. Using spring 2020 as the evaluation period, the optimized U-Net successfully reconstructed fine-scale ozone gradients and reproduced Tehran’s characteristic north–south pattern driven by topography and emission density. Comparative analysis with preliminary outputs demonstrates that feature optimization enhances spatial coherence, reduces noise artifacts, and improves the representation of localized hotspots. Statistical evaluation further showed strong agreement between the downscaled ozone estimates and observational data at both station and district scales, demonstrating effective generalization across heterogeneous urban environments. Overall, the findings highlight the potential of combining deep learning with interpretability techniques to refine coarse satellite ozone observations and provide a scalable, high-resolution framework for urban air-quality monitoring and exposure assessment.

This talk will highlight the future Canadian radar satellite mission, currently named the Terrestrial Snow Mission, under development by Environment and Climate Change Canada, in partnership with the Canadian Space Agency and Natural Resources Canada.

The mission concept will be presented, as well as recent scientific advancements made in the field of snow radar remote sensing, modeling and data assimilation, to continue the advancement of the mission's science readiness level.

This Canadian radar mission will provide weekly coverage of the northern hemisphere with Ku-band SAR data, and, coupled with modeled data, will provide daily snow water equivalent data, to assist hydrological applications and decision-making.

River systems originating from the Southern Alberta Canadian Rocky Mountains provide snowpack meltwater to an extensive downstream reservoir and irrigation network. Future water supplies have the potential to be significantly decreased due to changes in climate and reduced winter snowpack melt regimes. Estimating accurate water volumes in mountain regions is especially challenging. Current practices for estimating snow water equivalent (SWE) over a large mountain region use single point field-based snow measurements generally at valley or sub-alpine elevations. These field measurements are not spatially representative of basin-wide snowpack variability. The Alberta River Forecast Centre uses the Raven hydrological modelling framework to estimate daily winter snow water equivalent (SWE). To address the need for more accurate simulations of spatially explicit SWE, a combined airborne lidar and field snowpack sampling and modelling framework was compared with a Raven Model simulation. “Single point in time” SWE estimates were obtained between 2014 to 2021 using a combination of a) airborne lidar snow depth models, and b) public field sampled snow density. However, annual water yields cannot be generated from this type of snow sampling. The goal of this study was to improve spatialized Raven modelled SWE using the spatially-explicit lidar-based gridded SWE estimates across the West Castle Watershed (WCW, approximately 100 km^2). Results indicated Raven modelled SWE outputs were underestimated in comparison to the lidar-derived SWE with the largest deviation in the sub-alpine forested and grassland areas. Further research aims to use these comparative data to improve Raven-simulated wintertime headwater SWE estimates.

This study evaluates water surface elevation (WSE) retrievals from the Surface Water and Ocean Topography (SWOT) mission and investigates lake dynamics in the karst influenced environment of Bruce Peninsula National Park, Ontario. SWOT derived WSE measurements are validated against high frequency in situ depth logger data referenced to a consistent vertical datum using GNSS. The analysis compares multiple SWOT products, quality filtering approaches, and pixel aggregation methods to determine optimal workflows and assess performance under varying surface conditions, including open water, small surface area (<1km2), vegetation, and ice cover. Results demonstrate that SWOT accuracy is strongly dependent on surface conditions and lake characteristics, with reduced performance in smaller or vegetated systems. The study also examines spatial correlations in lake level variability to identify potential karst influences on hydrological connectivity. These findings provide guidance for the effective use of SWOT in monitoring inland water systems and highlight its potential and limitations for hydrological applications in complex environments.

Comparing airborne lidar datasets collected during snow-free and snow-covered ground conditions enables snow depth mapping at high accuracy and resolution (Hopkinson et al. 2004, Deems et al. 2013). Imputation of snow depth samples combined with field-based or modeled density can produce SWE for small to meso-scale (~100 km2) watersheds (Barnes et al, Submitted, Cartwright et al. 2020, Hopkinson et al. 2012). The goal of this study was to test lidar-based sampling and imputation in an operational regional (>20,000 km2) basin-scale SWE and runoff forecasting framework. Following initial tests in the winter of 2023, two lidar sensors were flown in March (Teledyne Optech Galaxy) and April (Teledyne Optech Titan) 2024 (and again in 2025 and 2026 – results not reported here), to collect 76 snow depth transects (~1 km wide, >2,000 km2) over the Bow and Oldman River Basin headwaters (>400 km north-south, >50 km east-west) near coincident with field samples at 28 sites. For 85 transect intersections, snow depth covariance was high (r2 0.70, RMSE 0.12m), with a small but acceptable bias of -0.04m or -5% (r2 0.94, n 198). An online digital twin platform is being developed to host the snow depth modeling results, as well as real-time weather telemetry and landscape change for public education and data dissemination purposes.

Surface soil moisture (SSM), representing the moisture content within the top layer of soil, provides valuable information and plays an important role in agricultural management. This study presents a deep learning (DL)-based method to estimate field-scale SSM time series over vegetated agricultural areas in Manitoba, Canada, by combining microwave and optical remote sensing (RS) data with auxiliary information. The input dataset was built using Sentinel-1 Synthetic Aperture Radar (SAR) and Harmonized Landsat Sentinel-2 (HLS S30) optical imagery, together with meteorological variables, soil temperature, crop type, topography, and soil texture. Since Sentinel-1 and HLS images were not acquired simultaneously, temporal interpolation was applied to align optical feature values with SAR acquisition times. Features were extracted at 30 m around nine Real-Time In-Situ Soil Monitoring for Agriculture (RISMA) stations.

A one-dimensional convolutional neural network (1D-CNN) was developed to learn local temporal patterns from the multi-source input dataset. The model was trained on multi-year data from 2016 to 2024 and externally validated on 2017 and 2021. On the validation dataset, the model achieved strong accuracy, with R² = 0.815, RMSE = 0.036 m³/m³, and MAE = 0.026 m³/m³. Model interpretation using Shapley additive explanations (SHAP) highlighted a physically coherent set of predictors, including vegetation cover and structure indices, radar backscatter features, solar radiation, minimum air temperature, and precipitation. Overall, the proposed DL framework provides accurate and interpretable field-scale SSM estimates suitable for agricultural monitoring and downstream water-management applications.

Surface water monitoring is critical due to increasing climate impacts, yet small reservoirs (0.01–1 km²) often lack the in-situ infrastructure required for consistent observation. This study evaluates the reliability of the Surface Water and Ocean Topography (SWOT) satellite mission for monitoring such water bodies by integrating UAV-based Digital Elevation Models (DEMs) and traditional gauge station data. A UAV survey was conducted at Recentino Lake (Umbria, Italy) in December 2024 to generate a high-resolution DEM (1.56 cm/pixel) with a vertical accuracy of 3.4 cm. Parallelly, SWOT data were processed by strictly retaining high-quality flags and applying a temporal outlier removal filter based on water level change velocity. The water surface elevation (WSE) derived from the DEM was compared with the processed SWOT data and in-situ gauge records. Results indicated high consistency between the UAV-DEM and SWOT-derived levels (110.78 m and 110.76 m, respectively) after harmonizing height reference frames. Conversely, comparisons with the gauge station revealed significant systematic biases (+18 cm vs. DEM; +44 cm vs. SWOT), attributed to the gauge’s undefined vertical datum. Despite this bias, the SWOT and gauge time series showed a reasonable correlation. These findings demonstrate the applicability of SWOT data for monitoring small reservoirs but underscore the critical challenge of vertical inconsistency across observing systems. Also, the study highlights the urgent need for unified vertical reference frames to ensure the accurate integration of heterogeneous hydrological data from different sources (satellite, aerial, and ground).

This presentation focuses on the challenge of atmospheric correction for high-resolution optical satellites (Sentinel-2) in the presence of adjacency effects, a major source of radiometric bias over small inland water bodies. Because water reflectance is extremely low in the visible and near-infrared, even small contributions of photons scattered from surrounding land surfaces can distort surface reflectance estimates of the observed water body. Traditional physics-based models such as 6SV offer radiative consistency but are limited by assumptions of atmospheric homogeneity and Lambertian surfaces, while empirical and semi-empirical approaches struggle to generalize across diverse atmospheric and geometric conditions.

This project addresses these limitations by developing a Physics-Informed Machine Learning (PIML) pipeline. We emulate heavy 3D Monte Carlo simulations to generate synthetic point-spread function (PSF) datasets. These datasets feed a tabular foundation model (TabPFN), leveraging In-Context Learning to capture the adjacency effect's non-linear dynamics without architectural retraining. We compare TabPFN against classical machine learning (XGBoost) using Sentinel-2 and in situ data. Results demonstrate TabPFN's superiority in resolving complex higher-order scattering, offering a rapid, physically consistent operational pipeline.

Detecting objects hidden beneath a forest cover with Synthetic Aperture Radar (SAR) is challenging due to strong vegetation scattering, canopy attenuation, and ground returns. This work investigates two methods for detecting concealed targets using Polarimetric tomographic SAR (PolTomoSAR). The first approach exploits full-rank polarimetric tomographic focusing to achieve high-resolution separation of scattering sources and estimate their polarimetric responses. Target detection is then carried out using descriptors derived from decomposition techniques, such as the polarimetric entropy, and double-bounce scattering intensity, enabling the identification of man-made objects embedded within a dense vegetation layer. The second approach considers a compact configuration using only two interferometric SAR (InSAR) images. Coherent ground-notching suppresses the dominant ground scattering contribution, while preserving responses from above-ground scatterers. It is demonstrated that the baseline value plays a significant role in the detection process, and an optimum value is selected. Both methods are evaluated using L-band data set acquired by the DLR F-SAR over Dornstetten, Germany. Results demonstrate successful detection of concealed objects for varying baseline configurations.

This study aimed to develop a highly accurate crop classification framework using multi-temporal Landsat 9 imagery and advanced deep learning architectures for the Tokachi Plain, a major agricultural region in Japan. Six time-series scenes, acquired between May 2 and September 16, 2024, were used to classify six crop categories: beans, beetroot, grassland, maize, potato, and winter wheat. Three temporal models with attention mechanisms were evaluated: long short-term memory (LSTM), bidirectional gated recurrent unit (Bi-GRU), and temporal convolutional network (TCN). Of the models tested, the TCN + Attention architecture achieved the highest overall accuracy (81.3%), significantly outperforming LSTM and Bi-GRU (p < 0.001). The Near-Infrared (NIR) band (Band 5) consistently exhibited the highest importance, highlighting its sensitivity to vegetation structure and chlorophyll content. Despite relying on only six optical scenes, the proposed model demonstrated robust performance comparable to or exceeding previous multi-sensor studies. These results underscore the potential of combining freely available Landsat 9 time-series data with attention-enhanced deep learning methods for efficient and scalable crop classification. The findings emphasize the important role of NIR reflectance during key growth stages and the effectiveness of TCN architectures in modeling temporal spectral variations for agricultural monitoring applications.

Optical satellite imagery plays a vital role in agricultural monitoring but is often constrained by cloud cover and illumination conditions. Synthetic aperture radar (SAR) offers an all-weather alternative, and recent advances in deep generative models provide opportunities to reconstruct optical-like imagery directly from SAR data. In this study, we investigated the potential of generating realistic red-green-blue (RGB) images of croplands using generative adversarial networks (GANs) trained on ALOS-2/PALSAR-2 quad-polarimetric (quad-pol) data. A distinctive feature of our work is the evaluation of not only backscatter coefficients (Gamma nought) but also polarimetric parameters derived from quad-pol decompositions, including the generalised Freeman–Durden, H/A/Alpha, and Yamaguchi four-component methods. Our results showed that paired image-to-image translation methods, such as feature-guiding GAN and pix2pixHD, achieved high similarity to PlanetScope reference imagery, with mean structural similarity index values exceeding 0.98 across all SAR inputs. In contrast, unpaired approaches demonstrated more variable performance depending on the input features. Notably, PUT showed significant improvement when H/A/Alpha or Yamaguchi decompositions were used, whereas Freeman–Durden produced results comparable to Gamma nought. The performance gap between paired and unpaired frameworks was most evident in heterogeneous landscapes, such as areas with adjacent grasslands and forests. These findings demonstrate the effectiveness of GAN-based translation from polarimetric SAR to RGB imagery for agricultural monitoring. The integration of polarimetric information adds value to unpaired learning schemes, and the ability to generate optical-like imagery under challenging observation conditions has strong potential for practical use in crop monitoring and assessment.

The performance of the Mask R-CNN model for instance segmentation of ceramic rooftops was evaluated using a high-resolution orthomosaic generated from UAV-based photogrammetry. Model training and inference were performed in ArcGIS Pro 3.5.3 with a ResNet-50 backbone. The model demonstrated high detection reliability, achieving a Precision of 96.62%, a Recall of 78.81%, and an F1-score of 86.81% at an Intersection over Union (IoU) threshold of 0.5. Most omission errors were associated with light-colored, elongated rooftops, highlighting limitations in the representativeness of the training sample and morphological variability. Fragmentation of larger rooftops into multiple segments was also observed, which affected accuracy metrics. To address this, a topological post-processing step was implemented to merge overlapping polygons, thereby improving segmentation consistency. These results indicate that Mask R-CNN is effective for high-resolution rooftop mapping, especially in applications requiring high precision. The approach is operationally feasible and transferable to similar datasets, enabling scalable analyses. It serves as a complementary tool for urban mapping, supporting the monitoring of urban dynamics and the analysis of construction patterns related to building standards and socioeconomic conditions.

Individual tree species classification from LiDAR (Light Detection And Ranging) point clouds has significant potential to support forest inventory and management, yet remains challenging due to complex three-dimensional canopy structures and the limited availability of labelled ground truth data. This study investigates self-supervised learning for tree species classification from LiDAR point clouds by comparing the PointMAE, a masked autoencoder-based model, with two supervised baselines, PointNet and PointNet++. Using the FOR-species20k dataset, two xperiments were conducted: a 33-species classification and a 6-species classification, each evaluated with point cloud sizes of 2048 and 8192 points. Using 2048 points, the PointMAE achieved the highest overall accuracy in both experiments (0.67 and 0.89 respectively), utperforming PointNet++ (0.63 and 0.84) and PointNet (0.39 and 0.75). Across all models, performance decreased when using 8192 points, indicating sensitivity to point cloud density

and sampling. Per-species analysis showed that coniferous species with distinctive crown geometries were the easiest to classify, while broadleaf species with similar crown forms, particularly Carpinus betulus, were the most challenging. These results show that self-supervised pretraining can improve classification accuracy over fully supervised approaches, highlighting its value for forestry applications where labelled data are limited.

This paper introduces the POD-HAR framework, a novel approach for deriving latent space dynamics in land surface modeling. The framework leverages Proper Orthogonal Decomposition (POD) to reduce data dimensionality by extracting dominant orthogonal modes and their temporal coefficients. It then applies Harmonic Analysis Regression with Sparsity (HAR) to identify sparse, interpretable nonlinear dynamical systems from this low-dimensional representation. By integrating these methods, POD-HAR establishes a regression-based technique for discovering parsimonious, often nonlinear, models that efficiently represent high-dimensional land surface evolution.

To address inefficiencies in manual inspection and color/geometric inconsistencies in tile production for web map services, this study develops an automated intelligent post-processing workflow. It integrates three core modules: automatic metadata quality inspection, computer vision-based image quality inspection (targeting invalid regions and color anomalies), and intelligent color uniformity adjustment with seamless edge fusion. By combining rule engines and image processing algorithms, automatic quality control and consistent fusion of produced/online tiles are achieved, significantly improving tile production automation and product reliability.

Classification of vessel types in Synthetic Aperture Radar (SAR) imagery is essential for maritime surveillance, yet distinguishing between ships with similar geometric characteristics—such as cargo and tanker vessels—remains challenging, particularly in medium-resolution images. This study investigates the role of wake patterns in improving ship-type classification using NovaSAR S-band imagery with 6 m spatial resolution. A dataset comprising 319 image patches (205 cargo, 114 tanker) was curated, including both centered ship patches and extended patches capturing wake structures. Experimental results demonstrate that incorporating wake information yields a 2–9% improvement across multiple evaluation metrics compared to ship-only scenarios. These findings highlight the potential of wake patterns as complementary features for enhancing classification accuracy in SAR-based maritime applications.

The growing demand for renewable energy has emphasized the importance of detailed geospatial information for photovoltaic (PV) site assessment and planning. Sentinel-2 imagery provides a valuable and widely accessible resource, yet its native 10-meter spatial resolution limits the ability to identify small structures such as rooftops, narrow roads, and compact built-up zones. This constraint affects the accuracy of solar suitability analyses and highlights the need for enhanced-resolution imagery capable of capturing finer spatial details.

This paper presents a photovoltaic (PV) assessment and optimization framework that integrates a resolution enhancement module based on latent diffusion models. This module operates in the latent space and relies on an iterative diffusion process to reconstruct fine urban and peri-urban structures, leading to higher-resolution products that support more accurate PV potential analysis and solar deployment.

Cloud-filtered Sentinel-2 L2A scenes are processed through this framework to produce ×4 enhanced imagery with an effective 2.5-meter resolution. Pretraining on

cross-sensor datasets can support realistic recovery of buildings, roads, and other small features while maintaining spectral coherence. The enhanced imagery enables more accurate rooftop segmentation, which serves as input for comprehensive photovoltaic potential assessment.

The installation optimization integrates multiple factors including solar radiation data, atmospheric conditions, shading analysis, rooftop orientation, tilt angles,

and panel layout efficiency to maximize energy generation capacity while considering technical and economic constraints. Qualitative evaluation demonstrates high-quality visual enhancement, confirming the relevance of this resolutionenhancement step within the overall workflow dedicated to PV site suitability analysis and installation optimization under real-world environmental conditions.

The given contribution describes an efficient artificial intelligence (AI) workflow for the detection and segmentation of ground-mounted photovoltaic (PV) systems in Bavaria (Germany), which can be transferred to any region. A two-stage approach was developed based on digital orthophotos (DOP) with a resolution of 20 cm (DOP20) or 100 cm (DOP100). Two different AI models, U-Net and YOLO, are used to identify and segment PV systems. The combined approach, which first analyses low-resolution DOP100 images and then uses targeted high-resolution DOP20 tiles, increases efficiency, by processing only relevant image areas with high resolution. Initial tests in three Bavarian districts show a high level of accuracy for both AI models. The approach is designed to be used for area-wide segmentation in Bavaria and thus contribute to change detection and quality assurance of the Digital Basic Landscape Model (ATKIS Base-DLM). Furthermore, the generalisation capability of the workflow was validated using an independent high-resolution dataset from the Piedmont region in Italy, where the models achieved promising recognition rates even without applying the post-processing pipeline.

We present a novel deep learning framework for above-ground biomass (AGB) estimation that produces high-resolution and multi-resolution biomass maps from coarse labels. The method is designed for the cases where dense pixel-level labels are unavailable. Using only 100 m scalar AGB values as supervision, our model predicts spatially detailed AGB maps at 100 m, 10 m, 3 m, and 1 m resolutions from PlanetScope imagery.

The task is formulated as a mass-conserving super-resolution problem, where each low-resolution label is reallocated over a high-resolution patch via learnable spatial weights. Our architecture is a lightweight encoder-decoder with four output heads, one per resolution scale. The final prediction is constrained to preserve total biomass per patch.

To guide spatial distribution without dense ground truth, we incorporate self-supervised learning (contrastive and equivariant losses), learnable pooling modules, and ecological priors such as NDVI/SAVI to suppress model hallucinations.

Trained on PlanetScope mosaics and ESA CCI-derived 100 m AGB maps, the model is evaluated on independent LiDAR-derived field plots. It explains 86% of the observed AGB variance (R² = 0.86) with only 2% bias, outperforming the baseline AGB map and recent CHM-based models in fine-scale detail.

This work demonstrates that both high-resolution and multi-resolution biomass mapping can be achieved from coarse supervision alone. It opens new opportunities for scalable AGB monitoring especially in data-scarce landscapes, with applications in ecological modeling, carbon stock estimation, and resolution-adaptive remote sensing.

Shadow correction is an important preprocessing step not only for visual enhancement but also for improving object recognition performance in remote sensing imagery. Although many datasets and deep learning models have been proposed for shadow detection and removal, most of them focus on natural images. In contrast, high-resolution aerial orthophotos contain large continuous shadows caused by tall buildings, especially in urban areas, and existing models often fail to handle such large-scale structures effectively.

In this study, we construct a new shadow annotation dataset specifically designed for aerial orthophotos with spatial resolutions of 20 cm/pixel and 5 cm/pixel. Furthermore, we propose a three-stage multi-resolution segmentation framework that progressively refines shadow predictions from low to high resolution. Predictions from lower-resolution stages are used as auxiliary information to guide higher-resolution prediction. Experimental results demonstrate that the proposed approach improves fuzzy Intersection over Union (IoU) by approximately 0.05 compared with a previously published shadow detection model, and also outperforms a single-stage baseline, particularly for large continuous shadow regions. The framework is also applicable to other large-scale segmentation tasks requiring extensive receptive fields.

With the in-depth penetration of UAV technology in fields such as geographic information surveying and mapping, the urban low-altitude economy has ushered in a critical period of rapid development. Among these fields, the safety issues in geographic surveying and mapping are particularly prominent. UAVs in this field are mainly used for field data collection of geographic information products such as digital orthophoto maps (DOM) and 3D oblique models. They realize fully automated flight mode through pre-set route planning, which significantly improves operational efficiency and operational convenience. However, they are confronted with the core technical challenge of "how to accurately determine the safety of flight routes within the survey area". This issue has become a key bottleneck restricting the safe and efficient operation of surveying and mapping UAVs.

This study takes remote sensing images and 3D geographic data as core supports, and combines multi-source data fusion technology and related algorithms to construct the "low-altitude flight safety field for urban surveying and mapping UAVs", drawing on the concept of “low-altitude safety corridors”. In essence, this field is a standardized digital 3D spatial grid system that covers the airworthy area of urban surveying and mapping UAVs, features three-dimensional connectivity, and supports intelligent coding. Shanghai was selected as a typical research area for data testing and verification. The test results show that the data achievements of this system can efficiently provide flight safety guarantees for the operation of surveying and mapping UAVs.

Precise monitoring of water resources is crucial for addressing global climate change. Water body extraction based on remote sensing imagery constitutes a core technical approach. Existing methods which based on CNN or Transformer (Chen et al., 2018; Gu et al., 2022; Lu et al., 2024), still encounter challenges when processing high-resolution imagery, including blurred boundaries, significant scale variations, and low computational efficiency. This makes it difficult to achieve a high degree of balance between accuracy and efficiency in water body extraction. To address these restrictions, this study proposes a residual network model integrating multi-scale contextual attention, called as MSCTFormer. It provides a novel approach for achieving high-precision and high-efficiency water extraction.

This paper proposes the MCAM model to address key challenges in hyperspectral image (HSI) classification. The core of the model comprises a cyclic adaptive scanning module, which achieves multi-view feature fusion through dynamic weights, and a multi-scale convolutional block, designed to extract hierarchical spatial features. Combined with an improved loss function, the model significantly enhances the discriminative capability for confusing land-cover categories. Experimental results on several public datasets demonstrate that MCAM outperforms existing methods in classification accuracy.

This paper proposes a modular fusion framework for delineating individual tree crowns from airborne LiDAR-derived canopy height models in a temperate mixed-wood forest in Ontario, Canada. Current instance segmentation models require expensive polygon annotations and tightly couple detection with segmentation, making cross-architecture fusion difficult. Limited forestry training data further causes transformer detectors to collapse on small datasets. The proposed framework decouples detection, fusion, and segmentation into independent stages. Two detectors, Faster R-CNN and DINO, are implemented with both ResNet-50 and domain-specific Masked Autoencoder backbones, with supplementary Finnish Taiga data stabilizing transformer training. A threshold-anchored score normalization maps each detector's confidence to a common scale before Weighted Box Fusion, enabling fair combination of architectures with incompatible confidence distributions. The fused bounding boxes prompt the Segment Anything Model (SAM) to generate per-tree polygon masks without domain-specific mask annotations. SAM's automatic mask generator additionally fills gaps where both detectors missed trees; SAM 1 is preferred over SAM 2, which produced fewer than half the automatic masks and missed smaller understory crowns. On two test plots with 233 and 107 ground truth trees, the framework achieves mask F1 scores of 0.79 and 0.61 at IoU thresholds of 0.25 and 0.50, matching 193 of 233 trees on the primary plot. Visual inspection indicates that many SAM-generated boundaries align more closely with canopy structure than the reference polygons. The modular design allows components to be independently replaced or upgraded, providing a practical pathway from LiDAR-derived CHMs to polygon-level crown delineation in data-limited forestry applications.

Remote sensing image captioning (RSIC) aims to describe key objects in remote sensing images using natural language, with significant applications in disaster assessment, land-use identification, and scene understanding. Existing methods face two critical challenges: insufficient cross-modal alignment due to the domain gap between generic visual representations and remote sensing semantics, and inadequate spatial relation modeling among regions in complex scenes, which compromises the semantic precision and logical coherence of generated descriptions. To address these issues, this paper proposes the Dual-Stream Relation-Aware Transformer (DSRAT) for remote sensing image captioning. On the visual encoding side, multi-scale CNN features serve as the foundation, fused with domain-specific semantic priors from RemoteCLIP through a gated dual-stream fusion module to achieve adaptive alignment of multi-source visual information. Subsequently, a spatial relation-aware mechanism is introduced into the encoder self-attention, which explicitly encodes geometric relationships such as relative position, distance, and orientation between regions as attention biases, enhancing the model’s capability for structured representation of complex spatial layouts and multi-object interaction scenarios. Finally, adaptive weighted aggregation of multi-layer encoder outputs generates discriminative cross-modal memory representations for the decoder. Experiments on the RSICD and NWPU-Captions datasets demonstrate that DSRAT achieves state-of-the-art performance across six metrics on RSICD and all seven metrics on NWPU-Captions. In particular, DSRAT achieves a significant performance improvement of +14.45 CIDEr on NWPU-Captions compared to the state-of-the-art method, validating the effectiveness of the proposed approach.

This study investigates a weighted ensemble framework for individual tree crown (ITC) delineation using LiDAR-derived canopy height models (CHMs). Three deep learning models, Mask R-CNN, U-Net, and YOLO were first independently evaluated to establish the baseline performance under consistent training and evaluation conditions. A weighted ensemble was then constructed by combining model outputs through a voting‑based fusion scheme, with an exhaustive search performed across multiple weight configurations to identify the ones that maximize common evaluation metrics. While certain weighting configurations yielded improvements in quantitative measures such as intersection over union (IoU), recall, F1 score, and accuracy relative to individual models, qualitative analysis revealed that these gains often coincided with substantial under segmentation, manifested as large, merged crown regions. This discrepancy highlights the limitations of binary map voting for instance level delineation and indicates that metric driven ensemble optimization may not reliably reflect instance level segmentation quality. The findings suggest that more expressive fusion strategies may be necessary for effective ensemble based ITC delineation in future work.

This paper presents an integrated approach using hyperspectral spectroscopy, granulometric analysis, and Sentinel-2 multispectral imagery for detailed mapping of carbonate beach sediments on Nogas Island, Philippines. By constructing a spectral library from field and laboratory data and employing the Spectral Angle Mapper (SAM) algorithm alongside the Grain Index, this study characterizes spatial variability in sediment grain size and carbonate composition. The methodology combines field sampling with remote sensing to generate maps that reveal sediment texture patterns influenced by hydrodynamics and depositional environments. The findings demonstrate that finer carbonate sediments exhibit higher reflectance and distinct spectral absorption features, enabling differentiation from coarser grains. This research highlights the potential of integrating multispectral satellite data with hyperspectral spectral libraries to provide rapid, reliable coastal sediment assessments critical for environmental monitoring, biodiversity conservation, and sustainable management of vulnerable tropical island beach systems.

For an increasing number of applications, land cover maps can be generated from remote sensing imagery using conventional and deep-learning-based semantic segmentation models.

Relying on a large pool of training data, the networks struggle with the spatial-temporal-spectral heterogeneity in the complex and diverse remote sensing imageries, leading to a significant number of errors in the model predictions.

This paper presents a workflow comprising domain adaptation and classification.

In particular, we analyze two domain adaptation techniques: First, a conventional histogram-matching method, which has turned out to be a surprisingly fast and reliable tool in a previous study, and second, a CycleGAN, which we applied both in its standard form and with the perceptual loss, thereby penalizing style inconsistencies on deeper layers.

By applying the workflow to three remote sensing datasets and six directions of domain adaptation, we show that there is ``no free lunch'' in the sense that all domain adaptation methods have their advantages.

Depending on the dataset, classification method, and especially on the availability of 3D data, the performance gap can be reduced to up to 1.5\% of the mean F1 score, demonstrating the soundness of the proposed method.

Road network is an important infrastructure of urban spatial structure and traffic system. Its accurate acquisition is of great significance for urban traffic analysis, automatic driving map construction and disaster emergency response. With the wide acquisition of high-resolution remote sensing images, automatic extraction of road masks from remote sensing images has become an important research direction in the field of remote sensing image understanding. However, the existing deep learning methods still face the problems of obvious modal differences and insufficient modeling of road structure continuity in remote sensing scenes. To solve the above problems, this paper proposes a remote sensing image road segmentation model LR-SAM based on SAM (Segment Anything Model). In this model, the LoRA (Low Rank Adaptation) fine-tuning strategy is introduced to achieve efficient parameter updating, and the MS multi-scale feature interaction module is designed in the coding phase to enhance the expression ability of the linear structure and fine-grained information of the road. At the same time, the original prompt encoder is removed and a lightweight ad decoder is constructed to achieve multi-scale feature fusion. In the reasoning stage, TTA (Test Time Augmentation) strategy is introduced to improve the stability and segmentation accuracy of the model. Experimental results based on chn6-cug and SAT-MTB datasets show that the proposed method achieves 97.20% and 85.06% mIoU and 96.67% and 84.94% F1-score, respectively, which is significantly better than the mainstream road segmentation method, and verifies the effectiveness of the proposed improvement points.

With the development of Earth observation technology, China's domestic high-resolution satellite remote sensing technique has achieved high-quality development. Currently, in orbit land resource satellites can obtain over 4500 images globally every day. With the explosive growth of data volume, traditional image processing method can not meet users demand for spatial information services with high frequency and large area. How to achieve automated image processing with massive data volume, and to provide real-time image services to users efficiently has become an urgent problem to be solved.

The recent volcanic unrest on Iceland's Reykjanes Peninsula was an excellent opportunity to better understand volcanic processes and develop hazard mitigation strategies. The eruption was studied using various direct and remote-sensing techniques. Here, we present an innovative UAV-based TomoSAR approach application, combined with photogrammetry, to explore the external and internal structures of an active volcanic vent within the Sundhnúkur crater row, where nine eruptions have occurred since December 2023. The surveys were conducted on 20 May 2024 (12 days after the end of the March–May eruption) and on 1 August 2024 (40 days after the May–June eruption). For optical data collection, we used a DJI Mavic 3T quadcopter, equipped with an RGB camera and an infrared sensor. The radar data were acquired using a UAV-based interferometric SAR system, Explorer RD350, which is capable of collecting P-band data in helical-trajectory mode. The optical data were processed using the standard photogrammetric workflow, and the SAR data were processed using the Refractive Back Projection algorithm, which enabled the extraction of amplitude images as slices at given depths with a ground penetration of up to 20 m. Our results show that the higher-intensity areas in the subsurface images correspond to the vent's crater center, while the lower-intensity areas correspond to the slopes of its cinder cone, composed of loose volcanic material. We assume that the higher-intensity areas in the amplitude images represent structures of denser material at depth, e.g., a lava conduit within the volcanic cone.

Analysis of the Phoenix area between 1992 to 2024, using DMSP-OLS and VIIRS Data.

Image super-resolution (SR) techniques have achieved significant performance improvements with the advancement of deep learning. Accordingly, deep learning-based SR methods have become the mainstream approach in SR research and are widely applied across various fields, including remote sensing. However, most state-of-the-art SR studies are primarily driven by computer vision research and tend to focus on generating visually realistic images rather than preserving structural fidelity with respect to the input images. In remote sensing applications, maintaining structural fidelity is particularly important because SR outputs are often used in downstream analytical tasks such as object detection.

In this study, we investigate the use of edge loss to enhance the structural fidelity of SR images for remote sensing imagery. The effectiveness of edge loss was evaluated using multiple benchmark datasets on both convolutional neural network (CNN)- and generative adversarial network (GAN)-based SR models. Several representative SR network architectures and GAN training frameworks were employed to assess the impact of integrating edge loss into the training objective. The experimental results demonstrate that incorporating edge loss improves both the structural fidelity and perceptual quality of SR images. Among the evaluated edge operators, the Prewitt-based edge loss showed the most consistent improvements compared with the Sobel- and Laplacian-based edge losses. These results indicate that edge loss is an effective and easily implementable strategy for improving SR reconstruction quality in remote sensing imagery. Furthermore, it can be combined with other edge-aware techniques to further enhance perceptual quality.

InSAR parallel processing become very attractive in recent years with the exponential growth of SAR data volume. Many InSAR parallel algorithms are deployed on cloud platforms with fixed hardware and network environments, or adopt a single granularity (e.g., scene-level or pixel-level), leading that the computing resources are not fully explored.

This research proposes a novel multi-granularity distributed parallel processing framework for time-series InSAR (TS-InSAR). The framework integrates three granularity levels (data granularity, task granularity, and algorithm granularity) and designs an adaptive scheduling strategy to dynamically adjust granularity based on task characteristics and computing resource status.

The proposed proposed multi-granularity parallel TS-InSAR processing framework has been employed to map ground deformation of the whole China territory annually since 2022, facilitating national-scale geohazard assessment.

Machine learning has become an increasingly important tool for quantitative prediction of complex mineralization patterns, offering new opportunities for improving mineral prospectivity mapping. Recent studies have shown that algorithms such as neural networks, support vector machines, and gradient boosting can capture nonlinear relationships and integrate diverse geoscientific variables with high predictive power. At the same time, traditional knowledge driven approaches such as the fuzzy weights of evidence method continue to demonstrate competitive performance, especially in geologically heterogeneous regions. This study provides a comparative evaluation of four machine learning models including logistic regression, support vector machine, backpropagation neural network, and extreme gradient boosting, together with the fuzzy weights of evidence method. The analysis is applied to a distinct environmental and geological predictor dataset from Labrador, Canada, a region characterized by complex lithological variation and limited historical exploration data. The goal of the study is to assess the robustness, stability, and generalization ability of these methods when transferred to previously unused datasets and differing geological conditions. Model evaluation is performed using cross validation, feature importance analysis, and spatially aware performance metrics. The resulting prospectivity maps highlight similarities and differences among the algorithms and identify areas with high potential for gold mineralization. The findings provide insight into the strengths and limitations of machine learning and knowledge based methods for mineral exploration and support the development of reproducible and interpretable workflows for regional scale mineral prediction.

Accurate segmentation of indoor 3D point clouds is essential for applications such as autonomous navigation, robotic interaction, and augmented reality mapping. Indoor scenes, however, remain difficult to segment due to clutter, occlusions, and repetitive structural patterns that often mislead conventional geometric or rule-based approaches. While deep learning models have improved segmentation accuracy by learning features directly from raw points, they typically require large annotated datasets and significant computational resources.

This paper presents SAMNet++, a hybrid segmentation framework that combines unsupervised segment generation with supervised refinement to achieve high accuracy while reducing annotation effort. In the first stage, a SAM-based LiDAR module—adapted from the Segment Anything Model—produces coarse, label-free segment proposals by leveraging fused LiDAR–RGB data. These proposals capture object boundaries and structural regions without manual labelling. In the second stage, a refined PointNet++ network enhances semantic precision and class consistency through targeted supervised learning.

To develop and evaluate the system, a dedicated indoor dataset was collected using a UAV equipped with a LiDAR sensor and an RGB camera, covering multiple rooms and corridor environments. Experimental results demonstrate that SAMNet++ outperforms state-of-the-art baselines in precision and F1-score, particularly when segmenting fine architectural details or navigating cluttered indoor spaces. With its balanced accuracy, efficiency, and reduced dependence on annotations, SAMNet++ offers a practical solution for real-time indoor mapping and scene understanding.

The research presented in this paper focuses on monitoring the phenomenon of urban heat islands (UHI) and provides local authorities with decision-making assistance in preventing their occurrence or mitigating the consequences of existing ones.

This paper proposes the design of a prototype design data warehouse for structured management, integration and analysis of multi-source geospatial data related to UHI detection and mitigation, focusing on two major Croatian cities: Zagreb and Split. Research in this area is the result of two started projects about UHI.

The proposed system is expected to provide a consistent and scalable framework for managing the heterogeneous geospatial datasets needed to understand urban climatic conditions. By standardising data handling and building on open data sources, the system creates the conditions for robust analysis of UHI patterns and for the development of tools that can support both research activities and the operational needs of local authorities.

Designed as a foundation for future monitoring mechanisms, planning tools and mitigation strategies, the system also aims to encourage broader use of open geospatial data in environmental and urban-climate studies. Its reproducibility and transparency should contribute to establishing a stable framework for further research and for practical applications in climate-resilient urban development.

Unauthorized land-use parcels are key targets in territorial spatial governance. Featuring diverse types, scattered distribution, strong concealment, traditional monitoring—satellite remote sensing with time lag and manual inspections with limited coverage—fails to meet the demand for rapid localization and verification.

This study proposes an automated verification framework driven by UAV hangars, integrating five links: intelligent scheduling, automatic data collection, real-time transmission, semantic interpretation, result dissemination. Adopting a "cloud-edge-terminal" architecture, it incorporates direct georeferencing, parcel segmentation, and improved A*+ algorithm-based path planning, achieving closed-loop automation of "detection-verification-evidence collection."

Field tests in Shanghai with 6 UAV hangar stations and 120 parcels showed 100% coverage, 75% less manual work, and adaptability to diverse scenarios. It addresses "slowness, omission, inaccuracy" in traditional workflows, providing a technical paradigm for data-driven territorial governance.

Accurate precipitation estimation is essential for hydrological applications and hazard monitoring in coastal regions, where complex terrain and strong land–sea interactions pose major challenges. This study investigates long-term rainfall variability and evaluates the performance of six gridded precipitation products—PERSIANN, IMERG, CHIRPS, ERA5-Land, GSMaP, and MSWEP—over the Guangdong–Hong Kong–Macao Greater Bay Area during 2001–2023. The results reveal pronounced spatial heterogeneity in precipitation trends: coastal subregions show a clear drying tendency, whereas the inland mountainous region remains comparatively stable. Despite these spatial differences, all regions exhibit synchronized interannual variability, suggesting the dominant influence of large-scale climatic drivers. All evaluated products successfully capture the unimodal seasonal cycle associated with the South China Monsoon, but notable discrepancies emerge during the peak rainy season, when intense convective rainfall leads to greater uncertainty. Among the six datasets, GSMaP and IMERG consistently outperform the others, showing higher correlation coefficients and lower RMSE across most months. In contrast, PERSIANN performs less reliably during low-intensity rainfall periods, while ERA5-Land systematically underestimates peak rainfall intensity. Overall, this study highlights the importance of region-specific evaluation of precipitation products in complex coastal environments and provides practical guidance for hydrological applications, hazard assessment, and disaster risk management.

This study presents an operational framework for the early detection of heavy rainfall based on the temporal dynamics of Cloud-Top Temperature (CTT) observed by geostationary meteorological satellites. The central hypothesis is that a characteristic “rapid rise followed by a sharp fall” in CTT serves as a precursor signature of subsequent convective intensification, as verified by radar-observed rainfall surges. The temporal pattern is analytically decomposed into the rise–peak–fall–trough phases, and the temperature drop amplitude (swing) between the peak and trough is quantified to define the WATCH (Warning and Threshold-based Convective Hotspot) window that indicates potential heavy-rain development. Two categories of lead time are formulated: the observed lead time, representing the exact temporal offset between the onset of CTT cooling and radar-detected rainfall intensification; and the estimated lead time, inferred from the gradient of the CTT decrease when radar data are unavailable or delayed. An edge-enhancement algorithm is implemented to minimize omission at the temporal boundaries, while adaptive thresholding and regional calibration enhance the algorithm’s transferability across diverse climatic and topographical environments. The proposed method is designed for real-time satellite operations and can be seamlessly integrated into existing satellite-radar hybrid nowcasting systems. By detecting convective growth phases preceding radar reflectivity increases, the method extends the effective warning lead time and improves the reliability of short-term rainfall forecasts. The findings demonstrate that CTT-based dynamic monitoring provides a physically consistent and computationally efficient tool for flash-flood preparedness, early warning, and rapid situational awareness in operational meteorological and hydrological applications.

this paper, we define a new wildfire risk prediction task from the perspective of wildfire state transition of next year, and hence, propose a novel approach named Wildfire State Transition Discrete Diffusion Model (WildfireSTDDM), that can directly capture the high-dimensional distribution of wildfire risk only through available and on hand historical wildfire events, with the following characteristics: (1) A 25-year-long-term daily wildfire historical record for British Columbia (BC) province, Canada is built deriving from the Fire Information for Resource Management System (FIRMS) with $10\text{km} \times 10\text{km}$ spatial resolution, using spatial aggregation. We define four wildfire state transition types based on the presence or absence of fire in a three-year historical period versus the fourth year: Persistent no-fire, New ignition, Fire cessation, and Persistent fire. (2) The proposed model can capture the categorical distribution of wildfire state transition type conditioning on the historical records and is trained in an end-to-end fashion, contributing to less cumulative error. (3) The proposed model can generate a high confidence map of next year's wildfire risk only through the long-term daily historical wildfire event without any other driving factors, and also correlate with the complex and stochastic wildfire pattern. (4) Since our model depicts the discrete wildfire state of each pixel forward as a discrete-time-inhomogeneous stochastic process, making it well-suited for characterizing next year's wildfire state transition uncertainty in model projections by performing multiple posterior sampling through Monte Carlo.

This study proposes a stripe removal method for remote sensing grayscale images based on ultralytics. First, we have got images from GEE, and stripes were annotated via Label Studio. Second,we have trained the ultralytics model with the annotated dataset, and adopting the best weights combined with pre-trained model for new image annotation. Finally, for stripe removal, the trained model detected stripe regions in remote sensing images and located their bounding box coordinates. Non-stripe areas were marked, with the largest normal area selected as the reference. Stripe region pixel data were segmented using detected bounding boxes, followed by histogram matching between stripe regions and the reference area to align grayscale distribution. Corrected stripe regions were replaced back to original positions to generate and save stripe-free images. This method achieves accurate stripe detection and effective grayscale correction, providing a reliable solution for remote sensing image preprocessing.

This contribution presents GEMAUT, a robust and open-source tool dedicated to the automatic generation of Digital Terrain Models (DTMs) from high-resolution satellite-based Digital Surface Models (DSMs). The paper provides a historical overview of the methods used for DTM extraction over the past twenty years, from early morphology-based filters to physically based optimization models and recent deep learning approaches. This retrospective is complemented by an analysis of the evolution of Earth-observation sensors, whose increasing spatial resolution now enables the application of LiDAR-oriented ground-filtering techniques directly to satellite DSMs.

The latest version of GEMAUT removes one of the main limitations of earlier implementations by eliminating the need for an external ground mask. Ground points are automatically extracted from the DSM using either the slope-based filter implemented in SAGA or the Cloth Simulation Filter available in PDAL. The terrain is then reconstructed through an energy-based surface optimization approach that combines robust data fidelity terms with curvature-based regularization.

A second major contribution is the introduction of a fully automatic quality assessment module. By analysing local DSM–DTM elevation differences, GEMAUT produces a spatialized precision mask that estimates the relative vertical accuracy at pixel level. This capability supports reliable quality control in operational and industrial workflows.

The tool has been fully refactored, relies exclusively on open-source libraries, and is publicly released on GitHub to encourage transparency, reproducibility, and collaboration within the ISPRS community.

The formation of canopy gaps alters forest microclimates, influencing understory regeneration, soil organic matter decomposition, and nutrient cycling, thereby playing a crucial role in forest ecology. Traditional methods for detecting canopy gaps typically rely on multispectral imagery or LiDAR data, which are accurate but costly and technically demanding. In recent years, several studies have explored the feasibility of using UAV-based RGB imagery for gap detection. This study utilized UAV RGB imagery to analyze the temporal dynamics of canopy gaps to assess the feasibility of employing RGB-based vegetation indices for canopy gap detection. The Normalized Green–Red Difference Index (NGRDI) combined with DSM differencing was used for analysis. Results show that when NGRDI < 0.03, forest areas can be effectively categorized into two classes: “canopy gaps” and “canopy cover.” The overall classification accuracy reached 93% with a Kappa coefficient of 0.68. However, the omission error was 44.44%, which suggesting that the model requires improvement in detecting small or edge gaps. It is recommended that identified threshold is used as a preliminary criterion for “canopy versus non-canopy” classification, supplemented with DSM or CHM data to improve detection accuracy.

Imagery from small satellites has been available for decades, yet automatic and accurate geometric correction remains a persistent challenge, especially when dealing with imagery which exhibit higher radiometric variability and a lower signal-to-noise ratio. This study introduces an enhanced version of the geometric processing module within the STORM processing chain, designed to perform fully automated orthorectification of images from small satellites. The module leverages publicly available ancillary data and deep learning-based road extraction techniques to eliminate the need for manual data collection and preprocessing. Ground Control Points (GCPs) are automatically generated by matching roads extracted from satellite imagery with corresponding vector roads obtained from open-access web databases. The orthorectification pipeline integrates several key components: ancillary data preparation, road extraction, GCP extraction, and final orthorectification using a digital elevation model. Experimental results on NEMO-HD small satellite imagery demonstrate that the proposed method can achieve accuracies of less than two pixel. The integration of deep learning for road detection provides a novel and effective approach for the fully automated orthorectification of satellite data of various types.

Digital Elevation Models (DEMs) are essential for terrain analysis and environmental applications, yet freely available global DEMs such as the Shuttle Radar Topography Mission (SRTM) DEM often contain noticeable elevation errors. Recent advances in space-borne LiDAR, particularly Ice, Cloud, and land Elevation Satellite-2 (ICESat-2), provide highly accurate elevation observations for DEM correction. However, most existing studies treat DEM correction as a single regression task and pay limited attention to correction direction, although direction errors may further degrade the corrected DEM. To address this issue, this study proposes a dual-task optimization framework for DEM correction using ICESat-2 data and auxiliary topographic and environmental variables. The network includes a shared feature extraction backbone, a regression branch for estimating correction values, and a classification branch for predicting whether DEM elevation should be increased or decreased. Kent County, New Brunswick, Canada, was selected as the study area, where 35,823 ICESat-2 elevation points were used for model training and validation. Results show that the proposed method outperforms Random Forest, XGBoost, and a conventional deep neural network, achieving a root mean square error (RMSE) of 1.76 m, a mean absolute error (MAE) of 1.37 m, and a direction consistency rate (DCR) of 75.05%. Compared with the original SRTM DEM, the corrected DEM reduces RMSE and MAE by approximately 27.6% and 25.9%, respectively, and improves DCR by 1.66% over the conventional deep neural network (DNN). These results demonstrate that incorporating correction direction into the learning process can effectively improve DEM correction accuracy and directional reliability.

Recent advances in UAV-based remote sensing have made high-resolution 2D imagery widely available, however the extraction of 3D tree structure from such data remains a primary challenge. This paper presents a novel framework for deriving True Tree Crown (TTC) geometry from Pseudo Tree Crown (PTC) representations, through a graph-based learning model. The PTC is generated from single nadir RGB images by interpreting grayscale intensity as height. This serves as an intermediate 2.5D representation that bridges the gap between conventional imagery and full 3D structure. We establish a spatial correlation between PTC and LiDAR-derived TTC meshes using geometric feature extraction and correspondence analysis. Preliminary results on synthetic data demonstrate a strong correlation between PTC and TTC height distributions, confirming that PTC encodes meaningful structural information. To learn the mapping from PTC to TTC, we propose a Graph Neural Network architecture with three GraphConv layers (64 – 128 – 256 channels), residual connections, and a composite loss function combining Chamfer distance with Laplacian and edge regularization. This framework enables the estimation of complete 3D tree crowns from single RGB images, transforming vast historical 2D image archives into valuable 3D forest data for ecological monitoring, carbon accounting, and sustainable forest management.

Advancements in unmanned aerial vehicle (UAV) technology have enhanced remote sensing and photogrammetry, enabling high-resolution mapping of terrain. This study evaluated the accuracy of digital elevation models (DEMs) derived from UAV-based structure-from-motion (SfM) photogrammetry by comparing them with real-time kinematic global navigation satellite system (RTK GNSS) survey data in the Steenbras Lower Dam wetland catchment, Cape Town, South Africa. High-resolution RGB imagery was captured using a DJI Phantom 3 UAV at an altitude of 35 meters above the highest terrain point, with a ground control network shared with the GNSS survey. Pix4D software was used to reconstruct the terrain, producing digital surface models, orthophotos, and ultra-high-resolution point clouds. Accuracy was assessed using 1,502 corresponding points. Initial metrics were affected by tall vegetation in the northern and southern periphery of the wetland. After filtering out absolute differences exceeding 0.5 m, the median elevation difference decreased from 0.464 m to 0.222 m, the median difference reduced from 0.344 m to 0.217 m, and the RMSE dropped from 0.605 m to 0.260 m. These results demonstrate that UAV-derived DEMs provide reliable and precise topographic information for wetland catchment mapping.

Dholavira, India’s second-largest Harappan site after Rakhigarhi, dating from 3000–1500 BCE, is renowned for its sophisticated water management system and has attracted significant archaeological interest since its discovery in 1968. Despite decades of conventional surveys, many structures remain unidentified, constraining spatial understanding of the site. This study develops a multi-sensor, multi-platform framework using active and passive datasets (optical, microwave, and LiDAR) from satellite, UAV, and ground-based sources to support improved documentation and analysis of archaeological features. Earth Observation (EO) datasets were processed to identify surface anomalies using multi-sensor analysis, while Synthetic Aperture Radar (SAR) data were used to delineate potential subsurface zones for subsequent GPR investigations. UAV-LiDAR data were utilized to enhance high-resolution 3D surface mapping of the site. Guided by satellite-derived anomalies, Ground Penetrating Radar (GPR) surveys were conducted at selected locations to investigate subsurface features. The GPR results revealed shallow hyperbolic reflections and stratigraphic discontinuities up to ~1.5 m depth, indicative of buried structures and disturbed ground conditions, with depth estimates derived using an assumed velocity model for dry sandy soils. Terrestrial Laser Scanning (TLS) enabled high-resolution three-dimensional reconstruction of excavated structures, showing close agreement with Archaeological Survey of India (ASI) records. The results demonstrate an effective and interpretable framework for archaeological prospection and multi-scale analysis, with future potential for integrating machine learning to advance systematic site analysis and digital heritage conservation.

Runway surface deterioration poses critical challenges for aviation safety and maintenance planning. Traditional inspection techniques are often labor-intensive and localized, lacking temporal continuity for assessing long-term degradation. Previous studies have primarily focused on pavement visual distress or thermal imaging, leaving a significant gap in non-destructive, satellite-based monitoring of runway condition using multispectral data.This study addresses that gap by employing multi-year Sentinel-2 Surface Reflectance imagery (2021–2025) to evaluate surface degradation of the Deoghar Airport runway. Six spectral bands (B2, B3, B4, B8, B11, B12) were analyzed to compute four spectral indices—Aggregate Degradation Index (ADI), Composite Condition Index (CCI), Surface Reflectance Index (SRI), and Thermal Stability Index (TSI). Temporal mean composites for each January were generated and analyzed for pixel-wise trends. Results revealed from 2021 to 2025, ADI decreased from 0.0876 to 0.0789, CCI increased from -0.2069 to -0.1718, SRI rose from 1.5171 to 1.6484, and TSI improved from -0.0158 to -0.0059, indicating overall runway surface stabilization with gradual roughness increase. A mean degradation rate of 0.010 year⁻¹, with 93.5% of pixels in the moderate class, 4.3% in high, and 2.2% in critical condition. The B12 band showed the maximum mean change (289.73), while B2 exhibited the most statistically significant trends (p < 0.05 for 72.1% pixels). The findings confirm that spectral reflectance indices effectively capture physical and chemical surface transformations. This method provides a scalable, non-destructive framework for continuous monitoring of runway health and supports predictive maintenance decision-making for sustainable infrastructure management.

The ecological analysis of desertification requires knowledge of post fire regeneration in the mid-step, influenced by topographic conditions and climate parameters. The North West regions of Algeria are affected each summer by violent forest fires which last over several days and affects woodlands, natural forests and reforestation. Usually NDVI is used, other derived index from radiometric data in remote sensing are widely used to monitor vegetation dynamics. The aim of this study is to determine the fire severity and monitor vegetation recovery with using multitemporal spectral indices together with topographical factors, and to recognise the different regeneration patterns of each burnt area. Several variables (such as climat, lithology, slope, aspect) were considered in order to analyse their possible relationship with the recovery process. Some of these variables showed a significant effect over the regeneration time, although further analyses seem still needed. Pre-fire and post-fire Landsat images and Alsat, were obtained to assess the related fire severity with using the widely-used Normalized Vegetation Index (NDVI) and modified Soil Adjusted Vegetation Index (MSAVI); Ratio vegetation index (RVI), and the index of regeneration (RI), to determine vegetation regeneration dynamics for period (2005-2007-2009 and 2015). Analysis showed that north-facing and east-facing slopes have higher regeneration rates in compared to other aspects. In addition, analysis of NDVI and RI stratified by pre-fire vegetation conditions and post-fire burn severity estimates could also be beneficial. And in this context post fire regeneration and topographics aspects are most important to ecological analysis of desertification in semi arids areas.

Urbanisation is accelerating globally and is a defining feature of modern cities. In 2016, 55% of the global population lived in cities, projected to reach nearly 70% by 2050. Rapid urban and population growth pose major challenges for sustainable development. By 2030, global urban land cover is expected to reach 1.2 million km²—three times that of 2000. This transformation involves significant Land Use Land Cover (LULC) changes, often converting natural vegetation into impervious surfaces like buildings and roads.

Urbanisation strongly correlates with rising Land Surface Temperature (LST) and intensified Urban Heat Island (UHI). Despite global attention to UHI, few studies have examined the spatio-temporal dynamics of LST in relation to recent urbanisation trends in Perth, Australia. As the city undergoes rapid suburban expansion and faces increasingly hotter summers, it is vital to understand how new urban development affects thermal patterns.

This study aims to address this gap by:

1. Identifying and delineating the areas of new development in Perth between 2005 and 2024,

2. Analysing and comparing LST patterns between long-established older and newly developed areas

3. Investigating the relationship between LST and its contributing factors, such as building and population density, tree canopy cover, surface moisture, albedo, and proximity to rivers

To achieve these aims, the study evaluates urban expansion between 2005 and 2024 and quantifies thermal differences using multi-temporal Landsat-derived LST.

Multimodal Remote Sensing (RS) methodologies have been increasingly studied in recent years due to their capacity to analyze multimodal RS data acquired from different sensors, thereby providing improved temporal resolution and extracting richer information than single-modal RS data. Deep Learning (DL) methodologies have accelerated the study of multimodal RS methods, thanks to their ability to learn features during training automatically. Many multimodal DL methods exploit this capability to learn a shared domain across modalities. However, most of them struggle to align heterogeneous modalities in a common representation. For this reason, we propose a supervised multimodal DL method that analyzes image time series acquired by different sensors to perform semantic segmentation. The proposed DL method is based on a Variational Autoencoder (VAE) that models the spatio-temporal information of the multimodal input image time series, with encoders and decoders composed of 3D convolutional layers, and learns the probability distributions for each modality. The probability distributions are combined to derive a joint distribution used for semantic segmentation. Learning the joint probabilistic distribution is achieved by combining the probabilistic parameters across modalities using a Product of Experts (PoE) approach. The feature maps derived from the obtained latent space are processed through three decoders. Two decoders aim to reconstruct the input multimodal image time series. The third decoder performs a semantic segmentation based on the inputs. Experiments conducted on the MultiSenGE and Austria datasets, which comprise Sentinel-1 and Sentinel-2 image time series acquired in France and Austria and representing heterogeneous classes, yielded promising results.

Rapid urbanization and climate change have the potential to negatively affect water availability in the coming decades. The Caribbean region is particularly at risk since, among other factors, large water storage facilities are not as abundant as in larger nations. It is imperative therefore, that water resources in the small island nations of this region are efficiently managed and monitored. Recent open-source, satellite earth-observation capabilities and data have presented additional tools for managers of this critical resource to better manage water and water infrastructure. In this study, we demonstrate the capacity of utilizing Synthetic Aperture Radar (SAR) data from the Sentinel-1 satellite for mapping surface area changes in three reservoirs on the island of Trinidad using a Google Earth Engine (GEE) framework. Sentinel-1 data was processed using GEE to produce average reservoir surface area calculations for each season (wet and dry) of each year for the period 2017-2023. The resultant reservoir surface area values were cross referenced against average seasonal precipitation values obtained from the CHIRPS (Climate Hazards Group Infra-Red Precipitation with Station data) database. The approach used in this study can be integrated into existing water resource monitoring frameworks to improve efficiency at little to no additional cost.

Mikania micrantha is one of the most aggressive invasive alien plant species in low-elevation landscapes of Taiwan. This study used fixed-wing UAV imagery to monitor its flowering-season distribution in a primary monitoring area north of Liyu Lake, Hualien County, eastern Taiwan. Rather than treating the dataset as a continuous annual time series, the analysis was based on three flowering-season observation epochs acquired on 14 January 2021, 7 December 2021, and 4 January 2024. UAV imagery was collected using an eBee X platform and processed in Pix4Dmapper Pro to generate high-resolution RGB orthomosaics with an average ground sampling distance of 3.08 cm/pixel. M. micrantha patches were delineated through manual image interpretation, and kernel density estimation (KDE) was applied to evaluate changes in spatial concentration and hotspot distribution. The interpreted infestation area decreased from 2,094.74 m² in the first epoch to 1,361.94 m² in the second, then increased to 1,799.09 m² in the third. KDE results showed a similar pattern, with persistent core infestation zones and renewed expansion in surrounding areas, including a new hotspot in the southeastern part of the monitoring area. These findings demonstrate the practical value of UAV-based monitoring for adaptive invasive plant management.

This presentation examines how to improve automatic road extraction from small-satellite images, where image quality is often limited by lower SNR and higher radiometric variation. The study tests whether data augmentation with noise and blur during pretraining can make deep-learning models more robust under these challenging conditions. Using a two-stage transfer-learning setup, a U-Net with a ResNet-50 encoder was first pretrained on PlanetScope RGB imagery and then fine-tuned on data from NEMO-HD, a Slovenian microsatellite mission. Several types of synthetic noise and blur were evaluated at different intensity levels.

The availability of high-resolution LiDAR data and advances in machine learning have opened new possibilities for automating coastal infrastructure mapping. This work presents a streamlined workflow for detecting marine docks using LiDAR intensity data and Detectron2, a state-of-the-art convolutional neural network framework. The approach integrates intensity normalization, scan-angle correction, and transfer learning to improve detection accuracy across diverse environments. Applied to LiDAR tiles from British Columbia’s Sunshine Coast, the method achieved detection rates of 70–80%, significantly reducing manual digitization effort. While recall remained high, variability in precision and segmentation accuracy highlights challenges in geometric alignment. The proposed workflow offers a scalable, data-driven solution for marine infrastructure mapping, supporting applications in coastal planning, environmental monitoring, and emergency response. Future work will explore 3D kernel point convolutions to enhance spatial accuracy and leverage elevation gradients directly from point clouds.

The aerospace sector urgently requires high-fidelity, real-world simulation environments that are both current and reactive, a challenge traditional workflows fail to meet. We present a fully automated, cloud-based pipeline developed by Airbus Defence & Space to produce trustworthy, reality-based aeronautical simulation data at a global scale.

Our core innovation is the automated co-extraction of a complete object stack—including precise building footprints, vegetation, and road networks—from the same Very High Resolution (VHR) satellite imagery source. This process, leveraging a multi-model deep learning approach based on foundation model paradigms, guarantees absolute spatial and temporal coherence across all extracted features.

The extracted features are then processed to generate high-fidelity LoD 2.1 3D geometry. This is achieved using a robust geometric framework and RANSAC-based plane fitting to reconstruct complex roof structures, delivering watertight volumes and filtering out photogrammetric noise. The pipeline is fuelled by the agile Pléiades Neo constellation and will be further reinforced by the four-satellite CO3D constellation, drastically improving revisit rates and ensuring data currency.

Operational validation on a 1000 km² diverse test area confirmed the system’s scalability, achieving full Digital Twin dataset generation in under 24 hours. This workflow effectively bridges the gap between raw satellite acquisition and actionable, high-fidelity simulation environments.

Accurate terrain elevation information is fundamental for geospatial analysis and environmental monitoring. Traditional 3D survey

methods such as LiDAR and photogrammetry provide high accuracy, but are costly, time-consuming, and limited in temporal

coverage. This work introduces Finding DEM0, a zero-shot framework that converts monocular depth predictions from foundation

models into metrically calibrated Digital Elevation Models (DEMs) without requiring supervised training. The approach leverages

the geometric consistency of DepthAnything V2 and anchors it to global elevation references from the Copernicus DEM and GEDI

LiDAR data through a linear regression-based calibration. Experiments conducted on around 2,500 tiles throughout the French

territory show consistent improvements over resampled Copernicus DEM baselines (approximately 1.5 m in vegetated areas and

more than 2.0 m in urban regions). The framework thus enables frequent, low-cost DEM updates using only high-resolution optical

imagery, eliminating the need for repeated airborne LiDAR/photogrammetric acquisitions and facilitating continuous and precise

elevation monitoring.

Wildfires are among the most destructive natural hazards, posing significant threats to ecosystems, infrastructure, and human life. While satellites provide objective information for burned-area assessment, their temporal resolution is insufficient for immediate response. Conversely, social media offers rapid eyewitness reports but suffers from limited reliability, vague descriptions, and spatial inaccuracy.

To bridge this gap, this study presents a hybrid verification framework that integrates social-media-derived event information with remote-sensing imagery and deep learning. The aim is to automatically confirm fire occurrence and refine coarse social-media coordinates to pixel-level accuracy. The major innovations include:

A large-scale GEE hierarchical search to locate possible burned regions.

A dual-branch deep learning model that performs change detection with pre- and post-fire Sentinel-2 patches.

A centroid regression mechanism enabling precise geolocation correction.

With the increasing global attention to clean energy, wind turbines (WTs) play a vital role in addressing both the greenhouse gas emissions and long-term energy sustainability. Nevertheless, accurately detecting the WT installations form remote sensing images remains a challenge. Existing data sources, such as the WT points of interest (POI) from OpenStreetMap (OSM), rely primarily from volunteer contributions are often incomplete or inconsistent, limiting their reliability for scientific assessment. This study proposes a global WT detection method form high-resolution remote sensing imagery via yolov8 deep learning model. The key contribution lies in constructing a WT installation suitability map based on multi-source spatial data, which reduce the search area by 38.99%, and improve the efficiency of global WT identification. In addition, to mitigate the challenges of small-target recognition in high-resolution remote sensing images, a method incorporating projection deformation of image regions is introduced. Using this method, more than 400,000 WT targets worldwide were successfully identified. Compared with OSM records, the method achieved an accuracy of 91.67% and revealed 48,688 newly installed WTs. This work provides a valuable tool for evaluating both the current status and future potential of global wind energy development, thereby supporting sustainable energy transitions.

Reliable, comparable measures of urban green cover are essential for a sustainable urban future. We construct a global, annual 30-m fractional green vegetation cover (FGVC) dataset covering over 60,000 urban areas from 1984 onward. Using Landsat imagery in a cloud environment, the workflow adapts to each image by learning local endmember spectral signatures before applying constrained spectral mixture analysis, mitigating the influence of endmember spectral variability. Accuracy against reference maps is high (r > 0.8; MAE < 10%; RMSE < 13%), and agreement with a widely used product at 500 m is strong (r > 0.7; MAE < 12%; RMSE < 15%). We will provide pixel layers, city/regional indicators, and validity metrics to support applications including SDG monitoring, climate-adaptation planning, and equity-minded urban greening.

Cloud masking has been a critical processing step in earth observation (EO) satellite systems. Its applicability in polar regions remains difficult due to the significant challenges in the differentiation between cloud and snow areas. Despite diverse EO satellite imagery, it lacks a general approach to leverage them jointly due to the sensor dependency of most cloud masking frameworks. Vision foundation models (VFMs) offer new perspectives in realizing towards sensor-agnostic frameworks for cloud masking, however it remains under-explored and merits further investigation.

In this contribution, we propose a solution that leverages the strong feature extraction capabilities of novel foundation models for cloud masking in polar regions, building on prior works of the developed cloud masking models and the subsequent cross-sensor transferability study. The architecture mainly utilizes the pretrained self-supervised backbone from mainstream foundation models (i.e. DINOv3) and effectively adapts to downstream tasks through fine-tuning with the adaptable decoder. It also investigates text-aligned DINOv3 by incorporating pretrained text encoders to enable multimodal understanding for additional EO applications, including text-prompted identification and object query of geographic features in satellite imagery.

Compared to the prior works on the developed transformer-based cloud masking models, the VFM-based approach offers several key contributions of model capabilities, in terms of foundational backbone, sensor-agnosticity, multimodality, etc. The VFM-based multimodal approach employs advanced spectral-spatial encoding strategies compared to vision baselines for the assessment of text-alignment strategies for improved semantic tasks, establishing foundations for emerging vision-language tasks that enable trustworthy EO applications.

The rapid expansion of Earth Observation (EO) data - from multispectral/hyperspectral to SAR, LiDAR, and dense time series - offers unprecedented opportunities to understand and monitor our changing planet. Concurrently, advances in artificial intelligence (AI) are transforming how these massive, multimodal datasets can be processed, interpreted, and translated into science-based decision support. Aligned with GEO’s Earth Intelligence for All Strategy, this work presents an integrated vision for accelerating global geospatial intelligence through AI-driven EO.

The GEO AI4EO Enabler plays a central role in realizing this vision. Designed to embed AI within GEO’s broader Earth intelligence ecosystem, it brings together a global network of AI and EO experts to foster cross-disciplinary collaboration, support capacity building, and develop and disseminate reproducible, accessible AI tools. The Enabler provides a framework to standardize AI-in-EO methodologies, promote responsible and ethical AI practices, and strengthen data-driven decision-making across diverse applications. As environmental and societal pressures intensify, this coordinated approach aims to make Earth intelligence more inclusive, scalable, and impactful.

Building on this foundation, we showcase transformational AI-driven EO applications: geospatial foundation model development and benchmarking; large-scale 2D and 3D urban mapping and continuous change detection; rapid flood and wildfire monitoring using satellite time series; multi-hazard building-damage assessment; and generative AI techniques that synthesize fine-resolution observations from coarse sensors for high-frequency operational monitoring.

By coupling the GEO AI4EO Enabler’s collaborative agenda with cutting-edge AI-driven EO, this work charts a clear pathway toward democratizing Earth intelligence and enabling informed decisions for a more sustainable and resilient future.

The rocky outcrops around the Horne copper smelter in Rouyn-Noranda (Quebec, Canada) exhibit highly variable surface conditions due to a century of atmospheric emissions. These surfaces act as passive archives of heavy metal deposits, but they remain poorly mapped due to their small size, spectral heterogeneity, and frequent mixing with vegetation or anthropogenic materials. This study presents a deep learning approach for high-resolution mapping of rock outcrops and their surface condition using multisensor remote sensing data.

We combined 0.2 m orthophotos (Vexcel UltraCam Eagle), Sentinel-1 SAR, Sentinel-2 multispectral imagery, and 1 m LiDAR derivatives to classify seven surface cover types: vegetation-covered rock, degraded soil mixed with till, smooth black-coated rock, anthropogenic surfaces, smooth uncoated rock, eroded till, and rough bare rock. The training data was created from a systematic 5 × 5 m annotation grid and field observations.

A U-Net convolutional neural network was trained for semantic segmentation using RGB orthophotos and features derived from LiDAR (slope, roughness, relief shading). The model achieved an overall accuracy of 86%, with high separability between bare rock classes and moderate confusion between degraded soils and eroded moraines. Probability and uncertainty maps with a resolution of 0.2 m were created from the softmax outputs to facilitate spatial interpretation.

The resulting maps reveal distinct spatial patterns of black coatings induced by pollution and erosion processes around the smelter. This work demonstrates the potential of multisensor fusion and deep learning for detailed environmental mapping in contaminated industrial landscapes.

Intelligent optimization algorithms are powerful tools for complex geospatial computing, focusing on the exploration of key regions in the solution space. A primary application is the automated identification of optimal parameters for classifiers like SVMs, which is crucial for remote sensing. Traditional penalty methods are hindered by their empirical penalty factors: overly small values cause the search to remain trapped in infeasible regions, while excessive values divert it from the true optima, particularly under equality constraints. To address this, we reconstruct the fitness function based on the Karush–Kuhn–Tucker (KKT) optimality conditions. This formulation inherently ensures convergence to the feasible region and explicitly leverages the inverse collinearity between the objective and active constraint gradients at the boundary. Consequently, infeasible solutions are guided efficiently along a composite gradient direction toward the boundary, enabling high-precision, adaptive tracking. Our approach improves convergence efficiency and substantially reduces reliance on penalty parameters.

Polarimetric synthetic aperture radar (PolSAR) is gaining increasing attention for monitoring and analyzing urban areas and their changes, such as area extraction (Wang et al., 2024) and mapping (Wu et al., 2021). A critical foundation for the studies is the accurate characterization of urban scattering mechanisms. This task can be accomplished using polarimetric decomposition methods (Quan et al., 2023).

PolSAR systems are undergoing rapid technological developments, aiming for fine spatial resolution, wide swath, and multiple wavelengths. The development or variation of system parameters leads to changes in both the geometric and physical interaction (mechanism) of the imaging process for a radar target in urban areas in Earth observation. Then, understanding urban backscatter is challenging. In this study, we focused on the wavelength effect on the scattering mechanisms of urban targets in PolSAR data. An alteration approach has been proposed to achieve an equivalence in the decomposition results using PolSAR data across different wavelengths. After the approach, urban targets in the decomposed results exhibit consistency across the three bands , qualitatively and quantitatively. The approach is viable in reducing the impact of radar wavelength on the PolSAR decomposition result.

This study presents an integrated approach for identifying potential large-scale rock fall areas using high-resolution UAV LiDAR data collected over the Stawamus Chief, British Columbia, Canada. The methodology couples UAV-derived morphometric and structural analysis with software-based block detection and stability evaluation to delineate unstable areas in rock masses and quantify their potential failure modes. A comparison study with terrestrial laser scanner (TLS) data was also conducted to compare different remote sensing dataset resolutions and accuracy.

This study presents an integrated and user-accessible framework for SAR imagery analysis that bridges SAR data processing and AI applications. The framework focuses on three objectives: (1) establishing a standardized pre-processing pipeline for harmonized cross-sensor Level-2 products, (2) enhancing usability through a streamlined interface, and (3) demonstrating practical applications through three AI modules—oil-tank detection and geometric measurement, shoreline extraction and change analysis, and ship detection. Experimental results show that the system achieves 1.5–2× faster processing compared to manual workflows and enables consistent analysis across multi-sensor SAR data, including TerraSAR-X and ICEYE. The oil-tank module achieves 86.5% detection accuracy with sub-pixel height estimation, while the ship detection module achieves up to 100% detection accuracy under high-resolution conditions and 90% overall accuracy. Shoreline analysis demonstrates consistent detection of temporal coastal changes. These results demonstrate that the proposed framework provides a practical and scalable solution for integrating multi-sensor SAR data into AI-based operational monitoring.

The Kumtag Desert in northwestern China hosts one of the world’s most extensive fields of feathered dunes, whose continuous migration poses a direct threat to downstream oases, farmland and water resources. Yet, monitoring dune mobility in this hyper-arid environment is challenging. In this study, we develop a multi-sensor remote sensing framework that combines Sentinel-2 optical imagery and Sentinel-1 SAR data with a dense optical flow algorithm to derive high-resolution, spatially continuous displacement fields for 2017–2022. Sub-pixel displacements from COSI-Corr are used as an independent benchmark, and time series of dune migration rates are reconstructed through least-squares inversion. We further couple the remotely sensed migration rates with regional wind data to estimate sand flux and invert dune heights based on sediment mass conservation. The results reveal a persistent northeast–southwest migration of feathered dunes, with typical velocities of ~5–8 m/yr and a clear negative correlation between dune height and migration rate. The proposed framework overcomes key limitations of traditional methods and provides a transferable tool for two-dimensional kinematic analysis, aeolian hazard assessment and desertification control in complex dune systems worldwide

Earth Observing (EO) satellites are an invaluable tool in remote sensing and have various applications. Spatial resolution is often

crucial to those applications. The current work focuses on sharpening Sentinel-2 images. Moreover, a new application/program has

been developed towards this goal. The application sharpens Sentinel-2 lower resolution bands (20m, 60m) and creates a 12-band

image in 10m resolution. To run the program, one needs to load a Sentinel-2 L2A product, select one or more pansharpening methods

and click the fuse button. This process will fuse the whole scene, but it is possible to crop areas of interest and process them instead.

To validate the process, 14 pansharpening methods were employed and tested against well-known image quality metrics. On all areas

of interest, the quality indices agree with each other. However, the indices tend to penalize methods who fail spectrally, which is

correct, but they also tend to favor images with poor performance in the spatial domain. MS-SSIM seems to rank better the algorithm

images and is closer to the visual comparison assessment. HPF is one of the best performing methods for sharpening a L2A product

of Sentinel-2. ATWT, AWLP, HCS and LMM are good alternatives according to our results. The application, S-2 Sharpy (A

Sentinel-2 Image Sharpening GUI) is made available on Github. Furthermore, its generic counterpart, PanFusion (Image

pansharpening GUI for various sensors) is also made available on the mentioned platform, since it was the application that set the

foundation for the current application and study.

Thermal simulation in urban digital twins enables effective monitoring of surface urban heat islands and supports climate adaptation planning. This paper evaluates machine learning and physics based approaches for this task through a unified validation framework based on 3D point clouds applied to an urban region in Berlin. The framework enables comparison of RandLA Net for 3D point cloud processing, InfraGAN for 2D texture synthesis, and physics based simulation on triangulated mesh geometries. RandLA Net architecture is adapted for thermal prediction and tested with two feature sets: RGB only and RGB with physics derived material parameters. Deep learning methods demonstrate severe spatial overfitting: training errors are minimal (MAE less than 1 K), but test performance degrades significantly on unseen regions with MAE increasing by factors of 1.9 to 2.5. Unexpectedly, augmenting with material parameters worsens generalization, indicating inadequate feature integration. Physics based simulation maintains consistent predictions (MAE approximately 8 K) with systematic bias addressable through calibration. These results motivate hybrid approaches embedding physical constraints into neural architectures for robust urban thermal modeling.

Land surface temperature (LST) obtained from satellite observations is a key parameter for understanding Earth surface-atmosphere energy exchange and urban thermal environments. However, the use of existing satellite-derived LST datasets for urban applications is limited by the coarse spatial resolution and the mixed-pixel problem. By integrating both two-dimensional (2D) surface properties and three-dimensional (3D) urban morphological characteristics, this study proposes a machine learning-based framework for high-resolution downscaling of satellite-based urban land surface temperature (SULST). A Random Forest model was developed to generate a 1-m downscaled SULST (DSULST) map. The model demonstrates strong performance, with a Pearson correlation coefficient of 0.89, RMSE of 1.15 K, NRMSE of 0.095, MAE of 0.56 K, and an index of agreement of 0.95. The 1-m DSULST maps reveal substantial sub-pixel thermal heterogeneity that is not captured by conventional 30-m LST data. Fine-scale spatial patterns associated with vegetation, building structures, and roads are clearly resolved in the downscaled 1-m temperature maps. These results highlight a critical limitation of satellite-derived LST in representing intra-urban thermal variability. The findings demonstrate that enhancing the spatial resolution of urban LST is essential for urban applications, including modeling surface energy fluxes, pedestrian-level heat exposure, and energy consumption, all of which benefit from higher spatial resolution.

Multi-object tracking in satellite videos faces unique challenges including small object sizes, low spatial resolution, frequent cloud-induced occlusions, and dramatic scene variations across geographic regions. Existing trackers, predominantly designed for ground-based scenarios, struggle to maintain reliable identity associations when satellite imagery exhibits long temporal gaps, transient visibility losses, and shifting appearance distributions. To address these challenges, we develop AURORA-Track, an end-to-end tracking framework that builds upon the Multiple Object Tracking as ID Prediction (MOTIP) backbone tailored for satellite video analytics. AURORA-Track introduces three key innovations: (1) an uncertainty-aware ID prediction module that augments the MOTIP decoder with calibrated confidence estimation, enabling robust handling of ambiguous associations and reducing false re-identifications; (2) a cloud/shadow-aware trajectory model that explicitly detects visibility degradations and leverages historical motion context to sustain tracking under partial or prolonged occlusions; and (3) a cross-scene knowledge transfer branch that meta-learns priors across diverse urban, maritime, and rural environments and rapidly adapts to new regions with minimal supervision. Extensive experiments on public satellite video datasets, including SatSOT and SatVideoDT, demonstrate that AURORA-Track achieves state-of-the-art performance, improving HOTA and reducing ID switches compared to leading baselines. These results validate the effectiveness of combining the MOTIP backbone with uncertainty-centric, occlusion-robust, and scene-adaptive enhancements for reliable satellite video tracking.

Urban heat islands (UHIs) present a critical challenge to sustainable urban development, demanding high-resolution monitoring tools for effective climate adaptation. We address this need by implementing a machine learning framework for downscaling Land Surface Temperature (LST) data, demonstrating its ability to capture fine-scale thermal variations. The methodology leverages multi-sensor remote sensing data fusion, integrating high-resolution optical observations from Sentinel-2 with thermal imagery from Landsat 9 (daytime LST reference) and ASTER (nighttime LST reference). Random Forest (RF) regression is employed, utilizing Sentinel-2 multispectral bands, derived spectral indices (e.g., NDVI, NDBI) to characterize land cover, and a Digital Elevation Model (DEM) to account for topographic effects. The RF model was rigorously trained and its hyperparameters optimized via randomized cross-validation to predict LST at a 10-meter resolution. Results demonstrate robust performance, achieving a high R2 of 0.75 (Mean Absolute Error, MAE: 1.7°C) for daytime LST and R2 of 0.50 (MAE: 0.6°C) for nighttime LST. The resulting downscaled maps delineate pronounced heat accumulation in dense built-up areas, notably its historic center and large commercial zones, contrasting sharply with cooler vegetated areas and green urban corridors. A comparative assessment against bilinear interpolation, TsHARP thermal sharpening, and linear regression confirms that the RF framework achieves the best balance between predictive accuracy, spatial coherence with the source thermal data, and meaningful sub-pixel detail, effectively preserving the critical fine-scale thermal patterns. Ultimately, this study advances UHI monitoring by enabling the precise identification of heat-vulnerable areas, thereby supporting targeted mitigation strategies even in small and medium-sized cities.

This study aims to assess the seasonal impact of land use on daytime normalized urban heat island (DNUHI) and nighttime normalized UHI (NNUHI) in Ahvaz, one of the hottest cities in Iran. To this end, 63 corrected Landsat images acquired in 2024 were used, and daytime land surface temperature (DLST) and nighttime land surface temperature (NLST) were derived for the four seasons. Thereafter, DNUHI, NNUHI, and normalized UHI (NUHI) indices were derived by normalizing the temperature differences between urban and non-urban areas. A land use layer consisting of 14 classes was overlaid with the thermal data to investigate the role of land use type in controlling thermal patterns. The results showed that the highest DNUHI values were observed in industrial (0.14-0.20) and oil (0.12-0.19) areas, which generated the highest daytime heating. At night, the highest NNUHI values were recorded in industrial (0.12-0.24), military (0.07-0.20), and oil (0.08-0.18) land uses, indicating the strong heat storage capacity of these areas. In contrast, green spaces, orchards, and agricultural lands showed the lowest DNUHI and NNUHI values (about 0.01-0.06). These findings can inform the design of sustainable climate strategies, the development of green spaces, and land use management to reduce urban heating.

Urbanisation in Johannesburg is significantly altering local climate conditions, yet long-term, satellite-based analyses of the Urban Heat Island (UHI) effect remain limited. This study addresses this gap through a ten-year (2014–2024) spatio-temporal assessment of Land Surface Temperature (LST) patterns and their socio-economic drivers. Landsat 8 imagery processed in Google Earth Engine (GEE) provided high-resolution LST data, which were integrated with regional socio-economic indicators, including population density and poverty metrics, and analysed using Ordinary Least Squares regression to examine their statistical relationships.

Findings indicate an apparent intensification of the UHI effect, with Johannesburg’s average LST in 2024 0.79°C higher than in 2014, and a 28% increase in population. Spatial analysis identified Regions D and G as persistent heat islands. At the same time, Region B consistently remained a cool zone, reflecting the significant role of land use and land cover in shaping intra-urban temperature variations. Poverty consistently correlated with higher surface temperatures, whereas population density showed a weak or negative relationship, suggesting that factors such as vegetation cover, construction materials, and surface permeability exert a greater influence on local temperatures than population density alone. Comparative analysis with other South African cities indicates that these patterns are systemic and socio-economically driven, highlighting broader issues of environmental inequality.

The study concludes that Johannesburg’s UHI effect is intensifying and raising urgent environmental justice concerns. It recommends targeted, socially equitable interventions, including urban greening programmes, cool roofing and paving materials, and thermal resilience strategies in informal settlements, to promote climate-adaptive and inclusive urban development.

Understanding the relationship between land surface temperature (LST) and near-surface air temperature is critical for urban microclimate research, especially for fine-scale thermal assessment in heterogeneous urban environments. This study investigates the spatial and temporal coupling between satellite-derived LST and in-situ air temperature during the summer of 2024 (June–August) on a university campus characterized by mixed building forms, surface materials, vegetation, and water bodies. High-resolution LST data were derived from Landsat-8 imagery, while near-surface air temperature was measured using a dense IoT-based monitoring network consisting of 19 observation sites.

Instead of treating LST as a direct proxy for air temperature, the analysis focuses on comparing spatial rankings, diurnal variations, and surface–air temperature differences across monitoring sites to identify systematic patterns of thermal consistency and divergence. The results show that LST presents stronger spatial differentiation than near-surface air temperature, whereas air temperature exhibits smoother spatial patterns and clear nighttime convergence. Surface–air temperature differences vary systematically across environmental settings, indicating heterogeneous coupling relationships between surface and atmospheric thermal conditions.

To further examine spatial correspondences, a convolutional neural network combined with Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to evaluate whether spatially reweighted LST information better explains observed air temperature variability. The results indicate that emphasizing thermally relevant surface regions improves the consistency between satellite-derived thermal signals and in-situ air temperature observations.

Overall, this study provides an interpretable framework for analyzing surface–air temperature relationships at the micro-scale and supports more reliable urban thermal environment assessment by integrating satellite observations with ground-based measurements.

This work presents a Machine Learning approach for the automatic recognition and mapping of rare earths in Brazil. While the country holds the world's second-largest reserves, identifying these valuable elements remains a challenge. By combining reflectance spectroscopy measured in the laboratory with open-access hyperspectral satellite imagery, a specific rare earths dataset is compiled. This dataset is used to train, validate and test neural networks to correctly classifiy rare earths by their spectral signatures.This method provides a novel and efficient tool for mineral prospecting and supports the geological community in assessing the national potential of these critical resources.

Urban vegetation is essential for mitigating the Urban Heat Island effect, yet its cooling performance depends on its three-dimensional structure. This study combines high-resolution Unmanned Aerial Vehicle - based LiDAR (Zenmuse L2) and thermal imaging (Zenmuse H20) to analyze vegetation structure and surface temperature across 4 urban parks in San Nicolás de los Garza, Mexico. LiDAR data were processed to generate Digital Terrain Model, Digital Surface Model and Canopy Height Model models, enabling the segmentation of individual trees and extraction of structural metrics such as canopy height, crown area and point density. Thermal orthomosaics were co-registered with LiDAR models to quantify temperature contrasts between vegetated and impervious areas. Results reveal consistent cooling effects in all parks, with vegetated zones showing 8–15 °C lower surface temperatures depending on canopy density and maturity. Larger parks with continuous canopies displayed the strongest thermal regulation. This integrated LiDAR–thermal approach provides a precise and scalable framework for assessing microclimatic benefits of urban vegetation, supporting climate-resilient planning in rapidly urbanizing regions.

This study examines land surface temperature (LST) trends and future projections in the Metropolitan Regions of Campinas and Piracicaba, São Paulo, Brazil, from 2002 to 2022. A time series of 15,091 MODIS LST images (MOD11A1 and MYD11A1 products, v6.1) was processed using Google Earth Engine to generate monthly composites, which were subsequently analyzed in ArcGIS Pro. Harmonic regression modeling identified seasonal and interannual temperature trends and simulated monthly temperatures through 2033. Eight municipalities, grouped by urban density, were selected for detailed comparison. The results indicate persistently higher LST values in highly urbanized areas, while municipalities with initially lower urbanization levels exhibited steeper warming trends over time. Projected January temperature increases between 2023 and 2033 range from 0.4°C to 1.0°C, with the most pronounced changes occurring in areas experiencing rapid land-use transformation. These findings are consistent with broader patterns of urban heat island intensification, emphasizing the combined effects of vegetation loss, impervious surface expansion, and urban densification. While the projections are statistical estimates based on historical trends, they provide valuable guidance for climate adaptation strategies and urban planning. This study demonstrates the utility of MODIS time series and multidimensional GIS analysis for monitoring and forecasting thermal dynamics in rapidly urbanizing regions.

Urban development is often accompanied by anthropogenic activities, changes in land morphology and serious damage to natural areas. Consequently, the urban climate is also affected, as temperatures are higher in urban centers and because of the presence of the urban heat island phenomenon, which poses a health threat to local citizens. The Monterrey Metropolitan Area (MMA) is the second-largest urban area in Mexico and is characterized for rapid urbanization and industrialization processes, steep climate conditions and the presence of urban heat islands. This combination makes living conditions rough for its inhabitants, especially for vulnerable groups. In order to quantify and compare heat vulnerability in urban areas, metrics such as the Heat Index measure the heat exposure and its effects on the human body. This study interpolated both relative humidity and temperature information from 15 local climate monitoring stations to determine the Heat Index for the six hottest weeks of the 2023 summer in the Monterrey Metropolitan Area. The interpolation methodologies used (IDW, Kriging and Spline) were later compared in order to cross-validate the results and define the most accurate performance base on both MAE and RMSE statistical analysis.

Operational monitoring increasingly depends on UAV imagery for safety, environmental, and infrastructure applications. Yet detecting unexpected objects remains challenging when targets blend into the background or operations extend to low-light and night conditions. Modern UAV platforms with integrated sensors now make high-resolution RGB, co-registered multispectral, and longwave infrared data more and more readily available, motivating methods that exploit complementary reflectance and thermal cues. In this paper, we address the detection of camouflaged objects by multispectral anomaly detection. We study 15 different three-channel stacks deviated from several image modalities, including real imagery and simulated longwave infrared images that encode the expected scene. This allows us to recast anomaly definition as reality–simulation discrepancy, as alternative to the conventional anomaly definition. We separately apply four detectors of differing categories to these image stacks: the classical Reed–Xiaoli Detector, a Region-of-Interest extractor, the Isolation Forest as convenctional machine learning approach, and a finetuned deep learning model. Evaluation is based on well-established metrics including precision, recall, and the F1-Score. Results reveal that combinations of near- and longwave infrared offers the best accuracy, longwave infrared alone is competitive, and simulated infrared imagery generally reduces performance, most likely due to a rather significant reality–simulation gap. We conclude that combining reflectance and thermal channels is critical for robust anomaly detection and that compact deep models currently provide the best trade-off for operational deployment.

The Siwana Ring Complex (SRC), located in western Rajasthan, India, is a distinctive geological formation characterized by its elliptical configuration. It primarily consists of rocks from the Neoproterozoic-era Malani Igneous Suite, reflecting its ancient volcanic origins. Peralkaline granitic rocks attract attention due to their potential to host valuable mineral deposits, particularly rare earth elements (REEs) and niobium (Nb). This study explores the potential of spaceborne imaging spectrometer (EMIT) and multispectral (Sentinel-2 MSI and ASTER Thermal) datasets for demarcation of REEs-bearing peralkaline granites, along with the potential sites of REEs. Silica and feldspar mapping was performed through the ASTER TIR dataset for targeting the potential sites of alteration zones within the peralkaline granites. Support Vector Machine (SVM) and Random Forest (RF) machine learning algorithms were applied on the EMIT and Sentinel-2 datasets for targeting the peralkaline granites of the region, which are the host rock for the REEs. The accuracy achieved through the EMIT and Sentinel-2 classified image varies. SVM and RF accuracies for EMIT are 93% and 96% respectively, while for Sentinel-2 are 95% and 99% respectively. Integrating the results from ASTER TIR with Sentinel-2 and EMIT highlighted the REEs-enriched zones within the peralkaline granites. This study demonstrated the potential of synergic use of thermal with spectral datasets for REEs delineation.

This contribution presents a novel method for infrared image stripe remover that addresses the limitations of current approaches in difficulty of extracting structural information of stripes. we design a progressive structure that sequentially aggregates contextual

information from intra-strip, inter-strip, to global levels. Specifically, a strip attention unit is proposed to harvest the contextual information for each pixel from its adjacent pixels in

the same row or column, while row attention and global attention are combined with their wide-ranging feature representation. This multi-scale attention mechanism address

local stripe artifacts and progressively incorporate broader image context in challenging conditions and provides more reliable results for practical applications. Experimental evaluations on multiple datasets demonstrate significant improvements compared to state-of-the-art methods. The method is designed to be computationally efficient and suitable for real-world deployment in fields.

The Landsat program provides over five decades of moderate-resolution satellite imagery, offering an invaluable record for monitoring land cover and land use changes. Despite its consistent calibration and open-access policy, Landsat’s low temporal resolution and frequent cloud contamination lead to sparse and irregular time series, limiting its usefulness for temporally continuous analyses. Reconstructing these missing observations is essential to improve temporal consistency and enable more accurate environmental monitoring. Previous studies, including our earlier work with the closed-form continuous-depth neural network (CFC-mmRNN), have shown promising results in modelling irregular Landsat time series. While the CFC-mmRNN achieved higher accuracy and lower computational cost than traditional methods such as continuous change detection (CCD), its performance declined under extremely sparse conditions, highlighting the need for more robust approaches. To address these limitations, this study introduces two transformer-based models for reconstructing very sparse historical Landsat time series: a one-dimensional Transformer and an enhanced three-dimensional variant that integrates a convolutional neural network (ResNet) with the Transformer architecture. The 1D Transformer processes individual sparse time series as input, whereas the 3D Transformer employs image patches (spatiotemporal cubes) to capture both spatial and temporal dependencies. Both models were applied to Landsat data (1985–2023) across the Canadian Prairies and evaluated against the CFC-mmRNN under varying spectral bands, data densities, and seasonal conditions. The results demonstrate that the Transformer-based models consistently outperform CFC-mmRNN, providing more accurate and temporally consistent reconstructions, particularly under extremely sparse observation scenarios.

Rapid Arctic warming has accelerated sea ice decline, intensifying interest in the Northern Sea Route (NSR) and the demand for reliable short-term forecasts. This study benchmarks non-recurrent deep learning models for daily sea ice concentration (SIC) forecasting over the NSR using the NSIDC-0051 SIC record (1988–2023). For each forecast, models ingest the previous 30 days of SIC on a 64 × 128 grid and predict the subsequent 10 days. Models are trained and validated with a five-fold walk-forward scheme over 1988–2020 and tested on 2021–2023. Two deployable architecture families are evaluated: CNN-based and Transformer-based backbones. To align with NSR operations, evaluation focuses on navigation-centric metrics. SIC fields are thresholded at 15% to define ice masks, and forecast skill for 1–10-day leads is assessed using Integrated Ice Edge Error, Mean Boundary Error, Intersection over Union and Anomaly Correlation Coefficient. CNN-based backbones consistently outperform Transformer-based backbones for boundary and overlap metrics across all lead times, with PoolFormer achieving the lowest errors and highest overlaps and leading short-term anomaly skill. However, the family-mean boundary error for the CNN group exceeds 30 km at a 7-day lead and 35 km at a 10-day lead, indicating that the practical utility of current models for NSR route planning is limited beyond about one week. These findings support modern CNN-based architectures for operational short-term Arctic sea ice forecasting and highlight the need for hybrid designs that preserve strong spatial feature encoding while better representing multivariate temporal dependencies.

Over the past decades, snow cover trends in North America have been analyzed, providing vital information to the Global Climate Observing System and other stakeholders about the looming signals of climate-driven snow declines. Detecting daily changes in snow parameters (e.g., snow depth, snow cover extent, and snow water equivalent) is, however, fraught with challenges, including internal variability unrelated to climate signals. We used GlobSnow's passive microwave remote sensing data and a Siamese U-Net model to compare patterns of daily changes in snow water equivalent (SWE) over the mid- and high-latitude regions of North America. The model detected changes in SWE with an F1-score of 94.8% and 100.0% in locations where it was not trained, and 99.3% at the location where it was primarily trained; this suggests the model's generalization potential to different climatologies and geographic locations. Using the model, we computed a similarity vector to compare SWE trends. We found that although lake-effect snowfall may be prevalent in the Great Lakes Basin during the winter months, the region consistently records the highest frequency of daily changes in SWE. Alaska, Yukon, and the Northwest Territories tended to have minimal daily changes in SWE, suggesting that latitudinal gradients may dominate changes in the snow regime and cryosphere's processes in the warming climate scenarios.

This contribution introduces OPTIG, an open-source Python tool for modeling glacier ice thickness and volume using Glen's Flow Law. The tool integrates geospatial inputs including DEMs, surface velocity raster, and flowline data to perform subglacial bed inversion and identify potential glacial lake outburst flood (GLOF) hazard sites. Validation against GPR measurements demonstrates ±22% uncertainty ranges. OPTIG empowers data-scarce regions with accessible, high-fidelity glaciological analysis for climate adaptation and hazard resilience.

The remote sensing of biophysical parameters in aquatic systems, such as water constituent concentrations, depends strongly on the quality of the spectral data. Sun glint, specular reflection from the water surface, is a major artifact that can substantially contaminate the remote sensing reflectance (Rrs). Accurate modeling of glint is essential, particularly in multi sensor analyses, to ensure seamless Rrs and water constituent products. We build upon the recently developed WASI AI model to mitigate sun glint effects. WASI AI is an AI assisted physical inversion framework that offers key advantages over traditional physics only approaches, including improved handling of spectral ambiguities and significantly faster inversions. We evaluate the effectiveness of WASI AI’s glint correction capability through an inter sensor consistency analysis between Landsat 9 and Sentinel 2. The analysis uses near simultaneous acquisitions over optically complex coastal waters of the Adriatic Sea. The two overpasses are only a few minutes apart, which allows to assume stable bio-optical conditions. However, sun glint can vary rapidly because it is sensitive to viewing and illumination geometry as well as wind driven surface roughness and currents. These factors may affect the data from the two sensors differently. Our results show that the WASI-AI glint correction identifies substantial differences in magnitude and spatial patterns of glint between the near simultaneous Landsat 9 and Sentinel 2 acquisitions. The Rrs consistency analysis demonstrates that, after glint correction, agreement between corresponding bands of the two sensors improves on average by 6% in R^2 and by 5% in NRMSD.

This study presents a near-real-time flood-mapping approach that integrates satellite-based Earth observation (EO) data, digital elevation models (DEMs), and machine-learning (ML) techniques. Several publicly available flood datasets were evaluated; however, none fully met the requirements for spatial coverage, data quality, and thematic diversity needed for robust model development. To address these limitations, a dedicated training dataset was constructed using Copernicus Emergency Management Service (EMS) Rapid Mapping products, comprising 38 flood events from 2022 to 2025. A modular workflow was developed to generate ML-ready datasets from satellite imagery, including data acquisition, advanced preprocessing, flood mask generation, and image tiling. Additional steps, such as co-registration, rescaling, data fusion, and masking irrelevant regions, were implemented to ensure spatial and temporal consistency across heterogeneous inputs. The developed model demonstrates reliable performance in delineating flood extents, achieving an average IoU of 0.70 on the validation dataset. Although the system remains under active development, the results indicate strong potential for operational deployment in near-real-time flood monitoring.

Rivers play a key role in the functioning of ecosystems, and their hydromorphological condition is essential for environmental assessments and water management. In France, field measurements used to evaluate channel geometry, such as bankfull width and slope, remain limited in spatial coverage due to logistical constraints. However, with the nationwide availability of high-density LiDAR data (>10 points per square metre), new opportunities have emerged for the large-scale, reproducible and automated characterisation of river morphology.

This paper introduces Bf3D, a fully automated 3D workflow designed to extract hydromorphological metrics from pre-classified LiDAR point clouds. Unlike traditional approaches based on manually placed cross-sections or 2D analyses, Bf3D relies on a continuous 3D representation of channel topography. The workflow includes automated river delineation, irregular digital terrain model (DTM) reconstruction, detrending, and a volumetric adaptation of the hydraulic-depth method to estimate bankfull stage and width.

Bf3D has been applied to over 1,400 river reaches across France. The results demonstrate accurate centreline delineation and bankfull width estimates that are close to field measurements. This approach removes user-dependent biases and enables rapid processing at a national scale. This approach introduces a new paradigm for hydromorphological monitoring by enabling the consistent, automated computation of key indicators across extensive river networks.

The present work focused on the synergistic utilization of Gravity Recovery and Climate Experiment (GRACE) and GRACE-Follow On (FO) derived Ensemble Terrestrial Water Storage Anomaly (TWSA), downscaled (GOU) TWSA and a gap filled JPL TWSA data and other hydrological variables to assess variability in groundwater storage (GWS) change in the last two decades (2002-2022) over Indus and Ganga river basins. The Indus basin witnessed a significant decline in groundwater with a rate of change between -2.11 to -3.0 cm/yr. Ganga basin also witnessed a significant decline in GWSA with ensemble dataset indicating a decline of -1.27 cm/yr and gap-filled JPL dataset indicating a decline of -1.88 cm/yr after removing soil moisture estimates respectively. Groundwater Storage Anomaly (GWSA) obtained from downscaled TWSA also indicated a significant negative trend. However, the magnitude of trend was considerably lower (0.9 cm/yr) than the ensemble (-1.37 cm/yr) and gap filled (-1.88 cm/yr) datasets. Ground observations also indicated a decline in GWSA in Ganga basin with a rate of change -0.26 cm/yr. GWSA computed from downscaled TWSA and satellite derived soil moisture showed highest positive corelation (R = 0.78) and least RMSE (17.5 cm) with in-situ GWSA in the Indus basin during 2002-2022. Similar results were observed for Ganga basin where downscaled TWSA (R=0.87) showed satisfactory corelation and low RMSE (8.11 cm) indicating that GOU TWSA and European Space Agency (ESA) soil moisture derived GWSA was able to capture the localized groundwater storage change effectively.

The Indus River Basin, a major Himalayan river system, has complex topography limiting availability of in situ measurements, which obstruct reliable characterization of precipitation patterns, thereby negatively affecting climate impact assessments and water resource management strategies. Understanding hydrological processes and efficiently managing water resources and dangers in Himalayan river basins depends on accurate high-altitude precipitation estimation. In this study, we have used the satellite-based precipitation reanalysis dataset (ERA5) and gauge-based data of IMD to overcome this issue. To conduct the analysis, we have used statistical methodologies, which include correlation analysis, root mean square error, probability of detection, and critical success ratio. We assessed the performance and detection of precipitation from ERA5 in comparison to IMD for the high altitudes. The performance evaluation of ERA5 precipitation against IMD bservations indicates a reasonably good agreement between IMD and ERA5 datasets in representing precipitation patterns over the study region, with R² = 0.793 and RMSE = 47.831 mm. The POD = 0.9686 and CSI = 0.7507. These results suggest that ERA5 provides a reliable representation of rainfall variability over the study area and can be effectively used for regional climate and hydrological applications. Further, we evaluated the performance of a gauge-merged precipitation dataset (GSMaP_ISRO) to highlight the significance of gauge merging over the study area. It was observed that the dataset outperformed in all the statistical indices. This study affirms the reliability of satellite-based precipitation datasets in high-altitude Himalayan regions and provides critical insights for sustainable water resource management in the face of evolving climatic conditions.

This study focuses on assessing the physical and chemical parameters of Indus River water

collected from a single sampling location, with special emphasis on seasonal variations and

sample preparation for ICP–MS analysis. The objective is to evaluate how water quality and

sediment inflow vary across different seasons and to determine the concentration of dissolved

and particulate matter in the river system. Water samples were collected regularly from the same

site of the Indus River during the summer, monsoon, and winter seasons. The analyzed physical

parameters include temperature, pH, oxidation-reduction potential (ORP), dissolved oxygen (DO

in mg/L and % saturation), electrical conductivity (EC), total dissolved solids (TDS), and

salinity. These parameters help in understanding the physicochemical condition of the river and

its environmental status. Temperature and DO show seasonal dependency due to changing flow

and temperature conditions, while EC, TDS, and salinity indicate variations in ionic

concentration and evaporation rate.

Spectrometry) analysis to estimate trace and heavy metal concentrations. In addition to field

observations, Remote Sensing and GIS techniques were used to analyze spatial variations in land

use, vegetation cover, and watershed characteristics influencing the Indus River. Satellite data

(Landsat and Sentinel) were processed in QGIS and Google Earth Engine to detect seasonal

changes in turbidity, surface temperature and land cover. The study concludes that the Indus

River water exhibits clear seasonal variations in its physical parameters and sediment load.

The cryosphere is essential for maintaining the balance of Earth's climate; however, it faces growing threats from increasing anthropogenic activities, including industrial emissions, biomass burning, and vehicular pollution, which have led to significant deposition of pollutants like ash on snow surfaces. These pollutants, originating from local industries, forest fires, and traditional wood-burning practices in the region, are altering the natural snow properties and accelerating disasters, snowmelt processes, potentially affecting climate, water resources, and local ecosystems. This research examines the effects of ash contamination on snow reflectance in the Himachal mountainous region of India, utilizing hyperspectral data collected through an XHR 1024i spectro-radiometer. The analysis involved a detailed examination of prominent absorption features, first derivative assessments, calculations of relative absorption strength, albedo evaluations, and the application of Principal Component Analysis (PCA) to thoroughly investigate the spectral alterations resulting from ash deposition. The need for this study arises from the growing concern over the accelerated melting of snow and glaciers due to reduced albedo caused by impurities like ash. The analysis indicates that the absorption feature at 1025 nm exhibits a pronounced sensitivity to ash contamination, demonstrating a reliable decline in relative absorption strength as ash concentration increases. The first derivative analysis highlighted rapid changes in reflectance, aiding in the identification of absorption features, while principal component analysis indicated that more than 99% of the spectral variance can be attributed to ash concentration. Albedo analysis supported the observed spectral alterations by confirming a notable decrease in snow reflectance.

The Chad Basin is a major water source for more than 30million people across four countries in the arid Sahel. Understanding long-term groundwater changes in the Chad Basin is necessary for water security, abstraction management and transboundary cooperation. In this study, we employed GRACE satellite and GRACE-FO satellite data (Total Water Storage Anomaly, TWSA) along with GLDAS land surface modeling to determine Groundwater Storage Anomaly (GWSA) trend between year period 2002 and 2024. The findings reveal water hydrological paradox as the basin shows a significant TWSA increasing trend of +5.91 mm/year (R² = 0.70). But, the gain is decoupled from replenishable reserves which are declining for the Surface Water/Soil Moisture (-1.04 mm/year) and near GWSA stagnation (+0.24mm/year, R² = 0.02). The rainfall shows a weak association (+1.65 mm per year trend) with GWSA (r = -0176). From this, it appears increasing rainfall is ineffective for recharging the deep aquifer. The excessive use of humans contributes to the localized depletion of the severe GWSA in the western margins, primarily in northeastern Nigeria. The present findings indicate that rather than climate variability, it is the failure of governance. That water scarcity is due to our unsustainable human activities and the inefficient water recharge pathways. In order to implement spatially-explicit abstraction quotas and prioritise effective high efficiency Managed aquifer recharge schemes, the data is essential for LCBC.

Monitoring dynamic alluvial rivers is essential for safe inland navigation, yet traditional bathymetric surveys are often costly and infrequent. This study presents an automated, cost-effective methodology for detecting and monitoring migrating sandbars by integrating Sentinel-2 satellite imagery with daily water gauge data.

Implemented within Google Earth Engine (GEE), the algorithm matches specific river water levels with cloud-optimized satellite scenes. It utilizes the Sentinel Water Mask (SWM) index to separate water from sediments, applying a 30-meter internal channel buffer to mitigate mixed-pixel errors along the shorelines.

The automated extraction was validated against high-resolution (3-meter) PlanetScope imagery. The results demonstrated high geometric agreement (mean Intersection over Union = 0.71) and a strong area correlation (R² = 0.97). While the 10-meter spatial resolution of Sentinel-2 introduces a systematic 26% overestimation of the sandbar areas , this over-segmentation serves as a beneficial safety margin in a navigational context, preventing the underestimation of submerged obstacles.

By correlating specific gauge levels with the emergence of sandbars, this method provides a vital 2D spatial baseline that enables the estimation of available water columns over specific bottlenecks. Ultimately, this approach supports the continuous generation of spatial databases, offering a practical foundation for dynamic relative depth mapping within River Information Services (RIS).

Snow cover plays a critical role in the hydrology and climate of the Himalayas, serving as a vital water reservoir for millions of people. Most previous studies often placed limited emphasis on recent country-scale assessments along with detailed snow variability. This study assessed the spatio-temporal dynamics of snow cover in Nepal during 2024 using 8-day MODIS snow cover products at 500 m resolution. Monthly maximum snow composites were generated to quantify snow cover fraction, seasonal trends, persistence, and variability. Results show distinct seasonal variation, with mean snow extent highest in winter (42.97%) and lowest in autumn (26.55%). Monthly snow cover peaked in April (50.01%) and reached a minimum in November (22.31%), reflecting strong intra-annual variability. Snow persistence mapping revealed that 32.32% of Nepal experienced no snow throughout the year, whereas 6.91% remained snow-covered year-round, corresponding to high-altitude permanent snow regions. The snow status change analysis highlighted dynamic snow behavior, with over 60% of pixels experiencing one or more transitions, underscoring the sensitivity of transitional snow zones. These findings improve understanding of snow variability in complex terrain and provide a scientific basis for hydrological modeling, water resource planning, and climate change adaptation in Nepal, where snowmelt-driven runoff is a key contributor to river discharge.

Due to global warming, Himalayan glaciers are retreating rapidly by several metres annually leading to the expansion of glacial lakes and increased risk of glacial lake outburst floods (GLOFs). These changes pose serious threats to downstream communities, highlighting the urgent need for climate adaptation and disaster preparedness. Gya Glacier, in particular, forms a moraine-dammed lake that experienced a significant outburst in 2014. The lake’s area expanded 1.25% in between 2018 to 2024, indicating a gradual increase and sustained hazard potential. To assess this risk, an integrated approach was employed using remote sensing, geographic information systems (GIS), and two-dimensional dam-break modelling with HEC-RAS. Multi-temporal satellite data from Sentinel-2 and High-resolution images were used to monitor changes in lake area, volume, and surrounding land use/land cover. High-resolution topographic data supported hydrodynamic modelling, allowing simulation of flood propagation and identification of vulnerable zones. The simulation revealed that a sudden lake breach could inundate approximately 1.71 ha of agricultural land, 1.28 ha of built-up area, 1.04 ha of fallow land, and 0.06 ha of a national highway. The greatest flood depths and velocities were recorded in the upper reaches due to steep gradients, with major damage concentrated downstream. To mitigate such risks, establishing an early warning system is crucial. This can include installation of Wireless Remote Terminal Units (WRTUs), Automatic Weather Stations (AWS), and GLOF detection systems at the lake site. Key sensors may include radar level sensors for monitoring water levels, and meteorological sensors to track climatic and hydrological changes in real time.

Rapid and accurate acquisition of the bathymetry of large-scale nearshore shallow sea is of great significance for coastal economic development, safe navigation of ships and coastal ecological protection. Beihai and Fangchenggang of the Beibu Gulf of Guangxi as the research area. Three inverse algorithms are firstly using for the bathymetric inversion experiments, which are one-band model, two-band-ratio model and multi-band-combination model, based on Landsat-9 images and electronic chart data. After that these three inverse algorithms of water depth are compared and then analyse the accuracy of the bathymetric inversion between the unzoned and zoned ones. The results of the experimental results that the multi-band-combination model exhibit the highest inversion accuracies in both experimental areas among the MAE and RMSE are 1.3843 m and 1.7611 m in Beihai and that of Fangchenggang is 1.8609 m and 2.4599m; following the bathymetric stratification, the average weighted errors of water depths are reduced, which mean MAE and RMSE reduced in the Beihai region by 0.6414 m and 0.8031 m and the mean MAE and the RMSE decreased by 1.6788 m and 1.9163 m The multiband combined regression model had a superior effect after the bathymetric layered inversion.

Monitoring global water storage dynamics is essential for understanding the impacts of climate change on hydrological systems. The Gravity Recovery and Climate Experiment (GRACE) and its successor GRACE Follow-On (GRACE-FO) missions have provided a unique opportunity to quantify Terrestrial Water Storage (TWS) variations at large spatial and temporal scales. However, differences among GRACE solutions from various processing centers, such as CSR, JPL, and GFZ, can lead to uncertainties that must be carefully assessed for reliable interpretations (Wang and Li, 2016).

This study aims to provide a comprehensive analysis of global TWS changes from 2002 to 2025 by integrating multiple GRACE-derived TWS products. Spatial trends of TWS were calculated to identify regions and countries experiencing significant water gain or depletion. Furthermore, monthly TWS variations were extracted to construct time series for individual countries, enabling the detection of long-term hydrological patterns and seasonal fluctuations. An uncertainty assessment was also performed to evaluate the robustness of the estimated trends and temporal variations.

Mount Kilimanjaro, an East African water tower, is undergoing hydro-climatic and land use changes with uncertain impacts on water availability along its elevation gradient. This ongoing study integrates satellite remote sensing, physically based hydrological modelling, and community knowledge to characterise spatial patterns of water availability and compare them with local experiences. Land use and land cover (LULC) are mapped using the European Space Agency (ESA) WorldCover 10-m product; vegetation dynamics are analysed with leaf area index (LAI); and climate forcings are derived from Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) precipitation and ECMWF Reanalysis v5 (ERA5) temperature. We implement the Soil and Water Assessment Tool (SWAT) to simulate water yield by elevation band, in the absence of streamflow, model evaluation uses independent remotely sensed constraints from the Global Land Evaporation Amsterdam Model (GLEAM) evapotranspiration (ET) and ESA Climate Change Initiative (CCI) soil moisture. Semi-structured interviews and surveys across three elevation zones capture perceived change and adaptation strategies. Preliminary analyses indicate heterogeneous trends, with the largest declines in lowland catchments and more variable responses at mid- and high elevations. Ongoing work will quantify uncertainties (forcings/LULC/parameters) and translate findings into elevation-specific measures for climate-resilient water planning.

This research focuses on exploring the capabilities of the SWOT satellite mission’s Ka-band Interferometric Radar (KaRIn) sensor for retrieving lake ice and overlying snow properties. SWOT KaRIn Version D Pixel Cloud Data Products are compared to in-situ snow and ice measurements on Łù'àn Män (Kluane Lake) during the Calibration and Validation phase that took place over a three month period in 2023. The Snow Microwave Radiative Transfer (SMRT) model is used to simulate backscatter for varying snow and ice scenarios to better understand variances in observed backscatter across the lake. Optical satellite acquisitions are also utilized to extract and compare backscatter to surface reflectance to analyze seasonal lake ice phenology trends. Preliminary results indicate that KaRIn-retrieved heights are inconsistent during the winter season. Additionally, the contrast in backscatter for ice and open water allow for effective ice cover mapping. During the winter season, backscatter values exhibit a general negative pattern, with SMRT simulations indicating a correlation to snow cover variability.

This study investigates the use of Landsat data for monitoring water clarity, expressed as Secchi Disk Depth (SDD), across the Finger Lakes region in New York. SDD, a long-established indicator of water clarity, is measured using a Secchi disk and widely applied in limnological research. Recent advances have enabled remote sensing-based estimation of SDD, with Landsat imagery frequently used alongside band ratios to mitigate atmospheric effects. Cloud-computing platforms such as Google Earth Engine (GEE) further support large-scale water clarity assessments by providing accessible Top-of-Atmosphere (TOA) and Surface Reflectance (SR) products.

The study uses citizen-science SDD measurements from the NY-DEC CSLAP program (2017–2023) across all 11 Finger Lakes. Corresponding Landsat 8 TOA and SR reflectance values are extracted from GEE using a 3×3 mean around sampling points and filtered for clouds and shadows. A Random Forest model is trained using both original bands and band ratios to estimate SDD under multiple evaluation schemes, including 80:20 train–test splits and 5-fold cross-validation with both random and stratified sampling.

Results show that stratified sampling yields more reliable predictions due to variability among lakes, and TOA performs slightly better than SR in this case. Feature-importance analysis indicates consistent influential band ratios across products. The study provides the first Landsat-based assessment of water clarity for all Finger Lakes and supports improved understanding of water quality trends in these socioeconomically important freshwater systems.

This work presented a data-driven and scalable approach for performing inland water bathymetry by integrating SAR-derived shoreline dynamics with water-level observations. The method leverages the high temporal resolution of Sentinel-1 imagery and diverse water-level data sources to infer relative elevation and uncertainty estimates.

By exploiting non-uniform sampling theory and regression-based interpolation, the method establishes a foundation for automated, reproducible bathymetry using globally accessible data. Future work will address error modeling and validation against high-resolution reference datasets.

Based on the integration of multisource satellite observations, including GRACE/GRACE-FO gravimetry, altimetry, steric, and sediment datasets, this study provides a comprehensive analysis of sea level changes and their driving mechanisms in the East China Sea (ECS) over the periods 1993–2022 and 2002–2022. The findings reveal that the regional mean sea level rise is predominantly driven by manometric changes (mass addition), contributing approximately 87% (3.06 mm/yr during 2002–2022), while steric effects account for only about 12.6%. A pivotal discovery is the critical role of substantial sediment deposition from major rivers like the Yangtze. This deposition introduces a net bias of –0.35 mm/year in GRACE-derived mass trends, and correcting for this "sediment effect" is proven essential for accurately closing the regional sea level budget. Decadal analysis further reveals significant variability: the ECS sea level rise rate was notably high at 6.51 mm/year (1993–2002), sharply decreased to 2.45 mm/year (2003–2012) primarily due to a strong negative thermosteric contribution (–1.53 mm/year), and subsequently recovered to 4.19 mm/year (2013–2022). At the seasonal scale, annual variations are dominated by steric effects, whereas semiannual signals are primarily controlled by manometric changes. This study successfully demonstrates that the ECS sea level budget can be closed within uncertainty when sediment corrections are applied, providing a robust methodological framework that is highly applicable to other sediment-rich coastal regions globally for improved sea level budget assessment.

Accurate, timely, and explainable Pan-Arctic sea ice concentration (SIC) maps are essential for climate change studies, Arctic sea route navigation, and climate adaptation of Northern communities. Every day, a large amount of active and passive microwave satellite imagery are collected by remote sensing systems over the Pan-Arctic region, including Synthetic Aperture Radar (SAR) from the RADARSAT Constellation Mission (RCM) and Sentinel-1, and Passive Microwave (PM) radiometry from the Advanced Microwave Scanning Radiometer 2 (AMSR2). While advanced DL-based data fusion models leverage extensive SAR and PM imagery to produce high-resolution SIC estimates, their decision making process is opaque and difficult to interpret. This study provides the first feature importance evaluation of SAR and PM inputs to improve the efficiency and transparency of using an advanced Transformer architecture for Pan-Arctic SIC mapping during the melting season.

This study evaluates the performance of five deep learning (DL) segmentation algorithms for retrieving lake ice cover from dual-polarization Sentinel-1 SAR imagery. Lake Hazen, located in the Canadian High Arctic, was selected as a representative site due to its strong climate sensitivity and variable ice conditions. A six-year dataset (2015–2021) comprising over 1,100 dual-polarization EW-mode SAR images was used to train and validate U-Net, U-Net++, SegFormer, DeepLab v3+, and PSPNet models. Binary ice–water labels were manually annotated to support model development. Temporal cross-validation using independent test years (2015, 2018, and 2021) was conducted to assess model generalization across different ice phenology periods, including ice-on, break-up, ice-free, and freeze-up phases. Results show that all models achieved high accuracy (>98% overall accuracy) during stable ice and open-water periods, while segmentation performance decreased during freeze-up due to mixed ice-water backscatter signatures. Visual analysis confirmed that each architecture successfully captured the spatial distribution of lake ice, though some misclassifications were observed in noisy or low-backscatter regions. The findings demonstrate the potential of segmentation-based DL models for automated lake ice monitoring and highlight the need for further model refinement to improve performance during transitional periods. Future work will extend the framework to additional lakes and multi-year datasets to enhance operational monitoring of lake ice.

The Surface Water and Ocean Topography (SWOT) mission

represents a major advance in Earth observation by providing the

first global two-dimensional measurements of surface water

extent and elevation. Its potential for hydrology, climate

monitoring, and water resource management is widely

recognized; however, recent studies indicate that its performance

varies across hydrological contexts.

This study presents a review of SWOT’s capabilities for inland

water monitoring based on a synthesis of published validation

studies, simulation experiments, and case applications. To

support a structured interpretation of these findings, a Strategic

Assessment Framework (SAF) is applied. The SAF is an

analytical framework that organizes the evaluation across four

components: strengths, limitations, opportunities, and risks,

enabling a systematic comparison of SWOT performance under

different environmental and observational conditions (Figure 1).

For large rivers and lakes (≥1 km²), SWOT meets its design

accuracy targets (Bazzi et al., 2025). However, in fragmented

wetlands and narrow channels, retrieval errors increase

significantly, with reported RMSE values of 30–70 cm in

simulation studies (Bergeron et al., 2020). Environmental

heterogeneity, including shoreline complexity and wind-induced

surface roughness, further increases uncertainty in elevation

retrieval (Bergeron et al., 2020), while vegetation and turbidity

reduce water–land separability and limit effective pixel

availability (Frasson et al., 2021).

The SAF highlights performance variability and identifies the

role of multi-sensor integration (Sentinel-1/2, Landsat, Planet

Scope) in improving the reliability of SWOT-based inland

water monitoring

Accurate monitoring of the Arctic Marginal Ice Zone (MIZ) is critical due to rapid Arctic Amplification. This study evaluates discrepancies between two widely used Level-4 reanalysis datasets—NOAA OISST and CMEMS L4 Arctic Ocean—over the Arctic (>58°N) from 1988 to 2022, specifically focusing on the MIZ (SIC 0–50%). After spatial reprojection to a common 0.25° grid, the comparison revealed significant discrepancies, particularly in the transition zone (SIC 15–50%). While both datasets exhibit long-term warming, CMEMS-L4 shows a much stronger warming trend (+1.173°C/decade) compared to OISST (+0.215°C/decade). This divergence is primarily attributed to algorithmic disparities: CMEMS-L4 incorporates Ice Surface Temperature (IST), resulting in higher variability, whereas OISST relies on proxy SST estimates. Crucially, a distinct temporal discontinuity was identified in OISST around 2005, coinciding with a change in its sea ice input source from NASA to NCEP. This structural break caused abrupt shifts in SIC values and even resulted in contradictory cooling trends in parts of the Greenland Sea, whereas CMEMS-L4 indicated widespread warming. These findings highlight that data processing methodologies induce non-negligible uncertainties. We recommend caution when utilizing OISST for long-term analysis in the MIZ due to its 2005 discontinuity.

Decades of fire suppression and exclusion in Jasper National Park (JNP), Alberta, Canada have altered forest conditions. Previous plot-level fire history analyses indicate a mixed-severity fire regime was disrupted after 1915 (Chavardès et al., 2018). Biomass (fuel) has accumulated, and stand connectivity and homogeneity have increased (Chavardès & Daniels, 2016). Furthermore, a mountain pine beetle epidemic has killed a significant portion of lodgepole pine within the park, shifting biomass distribution from the canopy as needles and branches drop (Talucci & Krawchuk, 2019) Under these conditions, fires can burn more intensely, with more high severity impacts, including substantial biomass loss (Hagmann et al., 2021; Harris & Taylor, 2015; Kreider et al., 2024). Understanding how altered fuel structures correspond to biomass loss is important for predicting future fire impacts, and informing forest management decisions (Schoennagel et al., 2004).

The 2022 Chetamon Fire in JNP provides an opportunity to study biomass loss using available pre- and post-fire airborne laser scanning (ALS) data. Fuel structures are determined following LidarForFuel protocol (Martin-Ducup et al., 2024). Pre- and post-fire outputs are differenced to determine spatial variability of biomass loss. Pre-fire ALS is further used to map pre-fire environmental conditions that influence fire intensity, and thus, biomass loss. This includes topography characteristics, and forest metrics such as density (Kane et al., 2007; Parks et al., 2012). These factors are analyzed as predictor variables of biomass loss in Random Forest analyses.

Warming air temperatures and prolonged periods of drought have increased fuel availability and fire activity across northern boreal forest regions. Modelling fuel structures, such as canopy fuel load, vertical distribution and spatial connectivity, is important for providing inputs in fire behavior models, as well as furthering our understanding of the environment. The overall aim of the project was to determine the efficacy and accuracy of three standard fuel modelling methodologies at high- (>30 pt/m2) and low- (<10 pt/m2) point densities and resolutions (5m, 10m, 20m, and 30m) in a dense forested environment near Fort Simpson, Northwest Territories. All metrics are compared to fuels measured in situ. This study highlights both the potential and limitations of scalable lidar-based fuel mapping and can help inform management practices, fire behavior applications, and future operational fuel hazard-mapping and risk-mitigation strategies.

A precise evaluation of positional uncertainty is crucial to maintaining the reliability of geospatial data, as well as supporting high-quality outcomes in professional surveying and mapping projects. This paper thoroughly examines the origins of error and the statistical and geodetic principles underlying accuracy assessment for technologies such as photogrammetry, airborne LiDAR, and mobile mapping systems. Building on these foundations, the study outlines a robust, methodical framework that enables practitioners to rigorously quantify the positional accuracy of their geospatial products. The approach is aligned with the most recent edition of the ASPRS Positional Accuracy Standards for Digital Geospatial Data, ensuring compliance with current industry benchmarks.

Urban green spaces including green roofs, parks, urban forests, community gardens and private green spaces are integral to city landscapes, offering ecosystem services and enhancing urban aesthetics. By leveraging data captured from satellite or aerial imagery, spectral analysis using indices such as Normalized Difference Vegetation Index (NDVI) enables effective mapping of vegetated surfaces in such urban green spaces. However, topographic views alone present certain limitations in this context, particularly for applications requiring the differentiation of vegetation based on vertical stratification. This study presents a novel approach that enables two-dimensional (2D) and three-dimensional (3D) visualization of rooftop vegetation using a combination of multispectral and digital elevation data.

Canada's boreal forest accounts for 28% of the world's boreal forest ecosystem and is a large carbon sink. Under climate change, the severity and frequency of wildland fires in this area is increasing. This is resulting in large amounts of carbon being released in to the atmosphere, affecting the rate at which climate change occurs. LiDAR is being used more frequently for studying wildland fires and has shown some success in measuring fuel consumption, providing insight into the amount of carbon emitted. This research aims to refine the methods used to process LiDAR data collected before and after a fire in the boreal forest. Different ground point filtering algorithms, methods of spatial alignment, downsampling values and DTM resolutions are explored. Findings demonstrate how the choice in data processing can influence how well LiDAR-based DoB estimates agree with field-based observations and highlight considerations to be accounted for in similar future work.

In this study, we compared existing national wetland maps with ground-truth polygons in four areas of interest located in Eastern Canada. By comparing the methods used for each map, we identified important elements to consider when producing a wetland map using remotely sensed data: 1) the five Canadian Wetland Classification System (CWCS) classes (bog, fen, swamp, marsh, shallow water) are broad and can create spectral confusion. It is preferable to use wetland subclasses and then merge them into the broad classes. 2) It is important to add SAR imagery to the classification, given that this imagery can detect many wetland characteristics related to the site's wetness and vegetation structure. 3) Ancillary data such as DEM, topographic metrics, and canopy height model are a valuable addition to the classification. 4) It is recommended to use multi-seasonal images to consider the seasonal and temporal variation in the vegetation phenology and in both surface and groundwater levels. 5) Images used should have a spatial resolution small enough to have a minimum mapping unit to be able to detect small landscape features; and 6) it is recommended to have a dense network of ground-truth sites representative of the AOI. Our study showed that mapping wetlands at the scale of Canada is very challenging, due in part to the diversity of wetland types, which complicates the definition of standardised wetland classes, as well as to the logistical challenges related to obtaining data at the Canadian scale.

This presentation will review past and recent progress in iceberg detection from space and motivate the development of a large iceberg database for future testing and comparison of the new detection techniques. The presentation also review work done to leverage ESAs Sentinel missions to build such a database.

test

test

This contribution presents a practical workflow for accurate and cost-effective forest terrain mapping in Japanese forests using a UAV equipped with low-cost LiDAR and GNSS. Instead of relying on a local reference station, we exploit the Centimeter-Level Augmentation Service (CLAS) of the Quasi-Zenith Satellite System "Michibiki" and integrate it with LiDAR-based SLAM to obtain dense terrain information with absolute coordinates.

In the proposed pipeline, LiDAR odometry estimated by FAST-LIO is aligned with the CLAS-based GNSS trajectory and fused in a pose graph on SE(3). The resulting optimization problem is solved in GTSAM using prior, odometry, and GNSS position constraints to compensate for the drift that accumulates when SLAM is used alone during large-scale flights.

Field experiments were conducted in real forest environments on multiple days and flight routes using a UAV-LiDAR system. Ground control points measured by post-processed kinematic GNSS were used as references to evaluate mapping accuracy. The results show that the integrated optimization reduces horizontal drift and improves terrain reconstruction to sub-metre accuracy, while keeping the system setup simple and low cost. The proposed approach is a promising option for operational forest surveys and other environmental applications that require frequent, wide-area terrain monitoring.

This abstract, authored by researchers from the University of Study of Pavia, DICAr, Laboratory of Geomatics, presents a comparative analysis of the geometric quality and cloud noise of four SLAM scanners. The study compares systems from different price points

Current deep learning models for UAV object detection often suffer from "context-blindness," leading to high false positives (logical fallacies, like misidentifying building features as vehicles) and low-confidence false negatives for occluded objects. To address this, this paper proposes the innovative Geo-Visual Fusion (GVF) enhancement strategy, which leverages the rich, deterministic geo-spatial prior knowledge embedded within High-Definition (HD) city maps.

The GVF approach is implemented as a lightweight, plug-and-play framework featuring a Geo-spatial Contextual Reasoning (GCR) Module. First, a Real-time Geo-spatial Registration module accurately projects initial 2D detections onto the city's unified geographic coordinate system using UAV GPS/IMU data and camera parameters. The GCR Module then performs two key functions: Logical Error Elimination, which uses a Semantic Compatibility Matrix to suppress detections that violate real-world spatial constraints (e.g., vehicles detected on building facades); and Low-Confidence Boosting, which employs a Bayesian approach to significantly raise the confidence scores of reasonable detections located in compatible geo-spatial contexts (e.g., partially occluded vehicles on a road).

Validated on a high-resolution urban dataset, the proposed framework (Baseline + GCR) consistently demonstrates improved mean Average Precision (mAP), successfully eliminating geographically implausible false positives and enhancing the True Positive Rate for low-confidence targets. This method offers a practical solution to transition from purely data-driven feature matching to context-aware semantic understanding in urban aerial perception.

Localization within a global reference frame is critical for the safe operation of UAVs. It is typically realized through GNSS measurements, however when signals are jammed, spoofed, occluded or reflected, this approach can lead to errors or fail. As most UAVs are equipped with cameras, absolute visual localization using georeferenced map representations offers a promising alternative. The recent invention of Gaussian Splatting introduces new opportunities for this task, leveraging real-time rendering from novel views to establish 2D-3D correspondences for pose estimation. In this work, we investigate the use of Gaussian Splatting maps for absolute visual localization of UAVs with a particular focus on geometric accuracy and its impact on the accuracy of position estimation. Through experiments with real-world data, we show that an initialization with dense Structure from Motion point clouds does not improve geometric accuracy compared to sparse initialization under the current training scheme. Additionally, constraining the position optimization of Gaussian Splats shows potential for improved pose estimation but introduces challenges during training. Despite these limitations, our results demonstrate the feasibility of Gaussian Splatting-based absolute visual localization for UAVs.

Uncrewed Aerial Systems (UAS, drones) are rapidly evolving technologies, with growing expectations for fully autonomous operations, enabling flights without onsite human control and Beyond Visual Line of Sight (BVLOS). A recent innovation is technology of ‘Drone-in-a-Box’ (DiaB) a.k.a. drone docks. DiaB systems provide an automated solution that integrates robust drones hosted in weather-resistant docks with typically also with cloud integration to data processing. Such connectivity enables utilization of real-time data products using both onboard and cloud processing workflows. This combination of robotics, AI, and data management holds the potential to deliver significant breakthroughs across diverse application scenarios. Objective of this study is to calibrate and assess the geometric performance of a novel multispectral (MS) DiaB system for environmental monitoring applications. The results indicated that the MS DiaB system delivers reliable performance without ground control points. For applications requiring cm-level accuracy, the post-processed georeferencing workflow was essential, whereas the direct georeferencing approach provided adequate accuracy for many operational scenarios. Our future work will extend this methodology to environmental applications.

This study investigates cross-domain generalization, adaptation behavior, robustness under visual degradation, and adaptive model selection for image-based object detection in autonomous driving scenarios. Two state-of-the-art detectors, YOLO11 and RT-DETR, are analyzed due to their complementary architectural paradigms, representing convolutional and transformer-based approaches, respectively.

The proposed framework consists of four stages: (1) zero-shot evaluation of COCO-pretrained models on the KITTI dataset to assess domain shift, (2) fine-tuning under short and extended training regimes to analyze adaptation dynamics, (3) robustness evaluation using synthetically degraded images simulating real-world perception challenges, and (4) the development of an image-based selection model for adaptive detector arbitration.

Experimental results show that YOLO11 demonstrates stronger zero-shot generalization and faster early adaptation, while RT-DETR achieves higher performance after extended fine-tuning, indicating superior long-term representation capacity. Under visual degradations, model performance varies depending on distortion type and training regime, confirming that no single detector consistently outperforms the other.

To address this, a lightweight selection model based on image quality features (brightness, blur, entropy, and edge density) is proposed to select the most suitable detector per image. The results demonstrate consistent performance improvements over individual models, achieving higher mAP without increasing computational cost.

This work highlights the effectiveness of adaptive, context-aware perception pipelines and demonstrates that exploiting model complementarity is a practical strategy for improving robustness in real-world autonomous driving systems.

This paper presents a novel methodological framework for georeferenced traffic monitoring that bridges the gap between image-based vehicle detection and geospatial analysis. Traditional video-based traffic monitoring systems operate exclusively in image space, limiting their utility for applications requiring physical measurements and integration with geospatial datasets. We address this limitation by developing a comprehensive camera calibration approach that leverages readily available geospatial data including smartphone video, drone-derived orthophotos, and 3D point cloud data.

The methodology establishes precise mathematical relationships between image coordinates and real-world geographic coordinates through a hierarchical calibration algorithm for camera parameter estimation. Ground control points are strategically selected from orthophoto and point cloud data, emphasizing features that are precisely identifiable and geometrically advantageous for calibration. The framework enables transformation of image-space vehicle detections to geographic coordinates, facilitating physical measurements, spatial analysis, and direct comparison with simulated traffic data.

Experimental results demonstrate the effectiveness of our approach, achieving a mean reprojection error of 2.94 pixels across calibration points. A case study of multi-lane traffic monitoring showcases the practical utility, where vehicle detections are successfully transformed from image to geographic coordinates, enabling lane-specific traffic analysis and potential integration with traffic simulation models. The proposed methodology offers a robust workflow for urban planning by connecting conventional video surveillance with geographic information systems, using only commonly available data sources and equipment, making it accessible for widespread implementation in intelligent transportation systems.

Reliable geometric verification is essential in the construction industry, particularly for large-scale maritime infrastructures where deviations can critically affect functionality and safety. The emerging Scan-vs-BIM approach enables automated quality assessment by comparing as-built point clouds with as-designed BIM models. It allows evaluation of the entire structure, rather than just specific points, but relies heavily on accurate spatial registration. This paper presents an evaluation of several Iterative Closest Point (ICP) variants for fine registration within a Scan-vs-BIM framework dedicated to construction tolerance verification. Three ICP variants are compared in terms of convergence behavior, robustness to noise, and stability using synthetic point clouds derived from maritime structures. The methods are then tested on real datasets, each acquired under different conditions, leading to varying data quality. Based on the results, a hybrid method is proposed to improve registration reliability. The results show that the proposed approach improves the inlier rate by 8–9% while reducing the mean deviation by approximately 1 cm on the noisiest datasets, compared to the classical point-to-plane ICP.

The automatic generation of detailed Level of Detail 3 (LOD3) building models (Gröger et al., 2012), featuring wall surface features such as windows, doors, and balconies, remains a significant challenge within urban 3D modeling. This challenge is particularly pronounced in high-density urban environments like Hong Kong, where complex building geometries, severe data occlusion from dense high-rise structures, and diverse architectural styles collectively create exceptionally difficult conditions for automated processes. In response to these challenges, this study proposes and develops a novel, adaptive workflow designed to efficiently generate semantically rich and geometrically accurate LOD3 models. Our methodology leverages multi-source data, including a large-scale repository of existing LOD2 models, Airborne Laser Scanning (ALS) data, and Mobile Laser Scanning (MLS) data, to overcome the limitations of any single data source.

This paper presents a methodology for automated semantic segmentation and 3D reconstruction of industrial building elements from unstructured point clouds. It addresses components such as roof panels, floors, rafters, purlins, and columns by combining orientation-based filtering, projection onto characteristic planes, morphological analysis, and optimization-based I-profile fitting. The workflow includes preprocessing with axis alignment and outlier removal, surface-orientation-based subdivision, contour extraction from binary projections, and automatic estimation of roof slopes and panel inclinations to guide structural reconstruction. The approach provides a systematic framework for precise digital modeling of industrial buildings, enabling efficient structural analysis, documentation, and planning.

Accurate damage localization is essential for infrastructure inspection, but conventional segmentation methods rely on dense pixel-level annotations that are costly to obtain and difficult to scale. This paper presents a foundation model-based pipeline for data-efficient damage localization in built infrastructure. The proposed workflow combines DINOv3 features for image-level classification, Grad-CAM for weak localization, and the Segment Anything Model (SAM) for prompt-guided pixel-level segmentation. The resulting masks are further transferred into 3D space for spatially contextualized visualization.

The pipeline is evaluated on two case studies. On a subset of Sewer-ML, three representative sewer defect classes are used to compare pretrained backbones and to qualitatively assess downstream localization. The DINOv3-based classifier achieves a higher average F2-score than a Google ViT baseline, reaching about 0.72 versus 0.64. On a custom historic masonry dataset, the method is quantitatively evaluated for material-loss segmentation using manually annotated test masks. The proposed heatmap-guided prompting strategy achieves a mean Dice score of 0.69 and a mean IoU of 0.53, while the classification stage reaches an F2-score of 0.99. A proof-of-concept experiment further demonstrates that segmented damage regions can be visualized within a larger local 3D scene. Overall, the results show that the proposed foundation-model based pipeline can support data-efficient and spatially meaningful damage localization across different infrastructure domains.

3D point clouds have been widely used in civil engineering, providing comprehensive geometric data for structural health monitoring, scene understanding, surface defect assessment, and more. However, the mainstream point cloud data acquisition sources, i.e., TLS and MLS, are superior for large-scale scene understanding and analysis but challenging for fine-scale analysis, particularly in laboratory testing, due to their low resolution.

This study proposes a close-range photogrammetry-based workflow for the 3D reconstruction and visual inspection of rubberised concrete (RuC) beams in an indoor-lab environment. High-resolution image sets were captured with both a Canon 5D Mark IV DSLR camera and an iPhone 14 Pro Max, and 3D models were generated in Agisoft Metashape. The comparison between reconstructed models revealed that the DSLR-based reconstruction achieved sub-millimetre resolution and texture, demonstrating satisfactory performance for fine-scale surface monitoring. An RGB-guided crack extraction method was developed to enhance the identification of surface defects to isolate the potential crack area from the background. The extracted crack regions were visually distinguishable and provided a well-structured geometrical representation of defect morphology. Furthermore, a before-and-after deformation analysis was conducted, which provides a sub-millimetre level comparison in different stages. The results confirm that the proposed workflow based on close-range photogrammetry is a flexible, intuitive, and high-resolution alternative to LiDAR-based methods for surface inspection and deformation monitoring in laboratory environmental concrete specimens. This workflow provides another aspect of structural assessment and establishes a foundation for future high-accuracy 3D feature characterisation, which can be integrated with material design and mechanical performance evaluation.

This contribution presents a Scan vs BIM–based framework for geometric quality monitoring that integrates large-scale site-acquired point clouds with design BIM models in a distributed processing environment. The approach targets both vertical structural components and complex mechanical, electrical, and plumbing (MEP) systems on active building sites. Large point clouds from terrestrial laser scanners are indexed using an octree structure, while structural columns and MEP objects are extracted from IFC-based BIM and converted into mesh representations that serve as analysis units. For each component, nearby scan points are clipped, filtered, and locally registered to the corresponding BIM mesh to compute horizontal deviations, verticality, and installation discrepancies without assuming specific cross-sectional shapes or component types. The workflow is parallelized across multiple nodes and threads so that the same procedure can be consistently applied to thousands of objects in project-scale datasets. By automating component extraction, point-cloud preprocessing, and deviation calculation, the framework enables quantitative tolerance checks and systematic identification of elements requiring inspection or rework during construction.