Conference Agenda

This paper introduces the Appearance-guided Scaling Diffusion Model (AGDM), a novel diffusion-based framework designed to densify sparse airborne laser scanning (ALS) point clouds while preserving fine geometric detail. Traditional diffusion models for 3D upsampling, such as LiDiff and PUDM, operate solely on intrinsic 3D information and struggle to reconstruct sharp edges and continuous surfaces when input data are extremely sparse. AGDM addresses these limitations by integrating two complementary conditional priors: multi-view appearance cues and geometry-aware 3D features.

Sparse point clouds are first rendered into ten synthetic viewpoints, and a Vision Transformer extracts high-level visual embeddings that encode surface appearance and boundary structures. In parallel, a Minkowski-based encoder processes the input geometry to capture spatial continuity and local shape characteristics. A cross-attention fusion module aligns and combines these modalities, producing a unified conditioning signal that guides a scaling diffusion network during iterative denoising.

AGDM is trained and evaluated on the YUTO dataset, where dense ground-truth scenes are reconstructed from multi-mission ALS data. Experiments demonstrate that AGDM achieves superior performance across Chamfer Distance, Jensen–Shannon Divergence, F1 score, and multi-scale IoU metrics. Qualitative results further show that the model produces more uniform, edge-preserving, and structurally coherent point clouds than existing diffusion approaches.

By leveraging appearance guidance alongside geometric priors, AGDM significantly improves the fidelity and practicality of LiDAR point-cloud upsampling, offering an effective pathway for scalable and cost-efficient 3D digital-twin generation.

Accurate 3D surface reconstruction and change analysis relies on point clouds representing persistent solid surfaces and should neglect very small (< laser footprint size) and temporary objects that create outliers. Terrestrial and Permanent Laser Scanning (TLS/PLS) data often contains transient or detached points, which violate assumptions of common cloud-, mesh-, and surface-based 3D change analysis methods. Those points cause wrong correspondences and change values in multi-temporal point cloud comparison. We address this with the Scan Outlier Ratio (ScOR) filter, a LiDAR scanning and survey-aware descriptor designed to identify points unsuitable for most point cloud-based change analysis methods. ScOR compares the measured point spacing with the expected spacing, assuming the surface is locally planar and orthogonal to the incoming laser beam. ScOR works with a single scan or multiple scans acquired from the same position, enabling multi-temporal neighborhoods for filtering. Using data from natural and urban environments, we analyze ScOR across different surfaces, neighborhood sizes, temporal neighborhoods, and compare it with the Statistical Outlier Removal (SOR) algorithm. Results show that ScOR successfully removes non-surface points, while preserving surface information. In our experiments, the true positive rate exceeds 95% in all but one case, while the false positive remains below 10% throughout. With neighborhoods from subsequent and aggregated epochs, the method automatically detects and removes large temporary objects (e.g., a person). Due to its interpretability, efficiency, and range-aware design, ScOR provides an effective pre-processing method for automated and near real-time 3D surface change analysis with TLS/PLS.

Autonomous positioning and scene reconstruction are crucial to the exploration and scientific research tasks of planetary rovers. 3D Gaussian splatting (3DGS) provides a new paradigm for dense reconstruction. However, the reconstruction method that relies only on monocular images will cause scale blur and insufficient geometric consistency. These problems are more prominent in planetary scenes that lack geometric constraints and weak textures. In order to overcome these limitations, we proposed a lidar-enhanced 3DGS-SLAM pipeline. By introducing sparse lidar measurements as prior information to improve depth prediction and ensuring consistent Gaussian initialization on the physical scale. Optimize the camera poses and Gaussian parameters through differentiable rendering to achieve robust localization and photometric-geometric consistency. Experiments on the Erfoud, a planetary similarity dataset, show that our method is superior to the advanced 3DGS-based SLAM system. The ATE has reduced by more than 50%. The PSNR, SSIM, and LPIPS have all improved significantly.

LiDAR-based perception models’ performance can degrade sharply when applied to data from sensors different to those they were trained on. LiDAR super-resolution aims to enhance sparse point clouds from low-cost sensors. This can help to bridge the sensor domain gap to higher resolution LiDAR. Prior work has primarily focused on reconstruction quality metrics for super-resolution with limited evaluation of downstream perception tasks. We address this gap by conducting a systematic analysis of how super-resolution quality impacts 3D object detection performance. We evaluate detection capability through zero-shot transfer experiments on the KITTI object dataset. Four representative detectors (SECOND, PointPillars, PV-RCNN, PointRCNN) trained on high-resolution data are directly applied to super-resolved low-resolution data without fine-tuning. Results reveal a critical insight: reconstruction improvements yield vastly different detection gains across architectures. PointPillars shows minimal improvement until reaching high reconstruction quality, then performance improves significantly. In contrast, PV-RCNN exhibits steady gains throughout. The highest-quality reconstruction closes up to 86% of the performance gap and enables detection in safety-critical scenarios, including distant vehicles and small pedestrians, where lower-quality methods fail entirely. This work establishes that LiDAR super-resolution effectiveness depends on both reconstruction quality and detector architecture.

Retrieving information from point clouds for analysis and visualization has gained ever-increasing interest. A growing niche in this regard is ray queries, commonly used for image synthesis. Ray tracing is widely used in computer graphics, with a multitude of solutions based on bounding volume hierarchies. However, these solutions are rarely straightforward to integrate with raw point cloud data and geospatial analytical workflows. To overcome this, we present a novel approach to ray tracing in raw point clouds that builds upon and extends existing geospatial indices. The solution is exemplified by a fast octree implementation that supports versatile query semantics, such as neighborhood queries with constraints on k and radius for both points and rays, while offering configurable data organization schemes, including layered, fixed, and adaptive depth. The evaluation demonstrates satisfactory speed and capabilities for many scientific use cases, while simultaneously exhibiting low implementation costs, high flexibility, and simplicity in integrating ray tracing into analytical point cloud workflows.

Accuracy analysis of free elevation models TDX-EDEM, AW3D30, SRTM and ASTER GDEM-3. Determination of systematic elevation model errors by Z-shift, model tilt and systematic errors as function of X and Y. Comparison with ICESat-2 data, determination of the systematic elevation model errors by ICESat-2 ATL08 data and correcting the free elevation models. Accuracy analysis of the corrected elevation models by airborne LiDAR data.

The corrections based on the ICESat-2 data significantly improved the free elevation models.

Accurate and spatially detailed soil information is essential for sustainable land management, agriculture, and environmental monitoring, yet existing soil maps often lack the resolution required to represent fine-scale soil texture patterns. This study investigates a hierarchical modelling framework that integrates high-resolution remote sensing data, including Sentinel-2 imagery and LiDAR-derived terrain attributes, with soil texture predictions from the provincial SIIGSOL dataset. The approach is evaluated across three contrasting regions in Quebec, eastern Canada, selected for their diverse landscape conditions and soil variability. Two modelling strategies were compared: a model based solely on Sentinel-2 and LiDAR predictors, and a hierarchical model that incorporates SIIGSOL covariates to examine their added value. The findings show that integrating multi-source information improves the representation of soil texture patterns and enhances model stability. This work highlights the potential of hierarchical, multi-scale approaches for producing more accurate digital soil maps. Future efforts will extend this modelling framework across the broader landscape to support high-resolution soil mapping for land management applications.

Crop type mapping is essential for agricultural monitoring, food security assessment, and regional management, yet large-scale operational mapping remains challenging. Reliance on a single modality and the absence of explicit spatio-temporal constraints limit existing methods from fully capturing diverse crop-rotation patterns and phenological trajectories over the growing season. To address this limitation, we propose a multi-source, multi-temporal crop mapping framework. Multi-epoch Sentinel-2 and Sentinel-1 observations are preprocessed in Google Earth Engine to produce co-registered optical and SAR time series, including spectral and vegetation indices as well as radar backscatter descriptors. The proposed model couples cross-sensor interaction with seasonal dynamics: an intermodal Mamba fusion mechanism exploits the complementarity between optical vegetation signals and SAR structural information to strengthen parcel boundaries and reduce sensor-specific artefacts, while a temporal Mamba module explicitly models crop development over time, capturing phenological evolution and differences in the diagnostic value of individual observation dates. Decoding the spatiotemporal representation yields the final crop type map. We evaluate our framework for the Federal State of Brandenburg in Germany, where results demonstrate field-aligned, spatially coherent predictions and robust suppression of speckle- and cloud-induced artifacts, validating joint multi-sensor, multi-temporal modeling for operational crop mapping.

Urban green infrastructure plays a crucial role in supporting ecological connectivity, enhancing climate resilience, and promoting human well-being. As cities densify, maintaining functional green networks increasingly depends on understanding the structural continuity of vegetation within complex urban fabrics. Morphological Spatial Pattern Analysis (MSPA) provides a practical framework for quantifying green infrastructure structure; however, its sensitivity to spatial resolution remains insufficiently examined—particularly at metropolitan scales, where high-resolution data are becoming increasingly available. This study examines the impact of spatial resolution on MSPA outputs for mapping and interpreting urban green connectivity in Miami-Dade County, USA. Two scenarios were compared using 10-m canopy data and 2-m high-resolution canopy data processed across 23 tiles. The workflow integrated vegetation preprocessing, MSPA classification, and quantitative and visual comparisons of structural classes to assess scale effects. Results demonstrate that fine-resolution MSPA (2 m) preserves continuous canopy structures and narrow vegetated corridors that the 10-m analysis tends to fragment or omit. High-resolution outputs provide a more realistic representation of neighborhood-scale connectivity, especially in tree-dense areas such as Coral Gables, while also revealing the computational demands of metropolitan-scale MSPA processing. The findings confirm that MSPA results are inherently scale-dependent and that the choice of resolution critically shapes the interpretation of connectivity. This research provides an operational foundation for incorporating high-resolution morphological analyses into urban resilience planning, nature-based solutions, and socio-ecological equity assessments.

Accurate Land Use and Land Cover (LULC) data are vital for effective land planning and management. This study evaluates the U-Net model for LULC mapping using high-spatial-resolution (2 m) imagery from the WPM sensor on the CBERS 04A satellite. The research focuses on the Servidão River Basin in Rio Claro, Brazil, an urban watershed susceptible to flooding. A pseudo-labeling framework is proposed to reduce reliance on manually annotated training data. Training samples were automatically generated by integrating spectral indices (NDVI, NDWI, SOCI, CI, NISI), Principal Component Analysis, and unsupervised Iso-Cluster classification. Several U-Net configurations were evaluated, with a ResNet-34 backbone with class weighting achieving the highest performance. The model was then retrained using a manually refined reference dataset to enhance the representation of spectrally complex classes. Accuracy assessment resulted in an Overall Accuracy of 0.93, average Precision and Recall of 0.92, and a mean Intersection over Union (IoU) of 0.86. These findings indicate that the proposed pseudo-labeling strategy, combined with a U-Net, offers a robust approach for LULC mapping in complex urban environments using freely available CBERS 04A imagery.

A first-order branch modelling method based on bidirectional searching was proposed, the key steps included skeletonization using local separators, trunk extraction based on path straightness and first-order branch extraction using bidirectional searching. The method was tested on ForestSemantic dataset, and results showed that the extraction precision was 80.29%, and RMSE of the pitch angle estimation was 9.74°, indicating that the method can effectively recover the topological structure of branches.

Long-term and high-resolution terrestrial water storage (TWS) monitoring is critical for water-resource management, climate adaptation, and understanding hydroclimatic variability. Satellite gravimetry missions such as GRACE and GRACE-FO provide unprecedented observations of TWS but are limited by coarse spatial resolution, short observational records, and temporal gaps. This study presents an integrated deep-learning framework for reconstructing and downscaling GRACE/GRACE-FO data to produce century-scale, high-resolution TWS datasets. We apply RecNet and an enhanced RecNet (ERecNet) to reconstruct historical TWS anomalies in the Sudd Wetland, Lake Victoria Basin, and Nile River Basin, leveraging climate variables and lake-level observations. To overcome spatial limitations, we develop DownGAN, a novel generative adversarial network with a high-to-high downscaling strategy, producing fine-scale TWS patterns while maintaining mass consistency. The fusion of reconstruction and downscaling enables detailed, long-term monitoring of wetland dynamics, droughts, and hydroclimatic variability. Reconstructed datasets reveal multi-decadal wetting/drying phases and strong links between TWS fluctuations and climate teleconnections such as ENSO and the Indian Ocean Dipole. This framework advances the application of GRACE/GRACE-FO for climate resilience, ecosystem monitoring, and water-resource management in data-scarce regions, demonstrating the potential of deep learning to extend satellite-based hydrological observations both spatially and temporally.

Methane (CH4) is a major greenhouse gas; however, large-scale monitoring remains challenging due to the high costs and spatial limitations of ground-based and airborne observations. In contrast, Sentinel-2 shortwave infrared (SWIR)–based plume detection is hindered by its coarse spectral resolution, surface artifacts, and limited real-world annotations. This study proposes a self-supervised learning (SSL) framework based on the Simple Framework for Contrastive Learning of Visual Representations (SimCLR) to learn transferable CH4 plume representations from unlabeled Sentinel-2 data. A real-world dataset of 456 Sentinel-2 image tiles was manually annotated using the multi-band–multi-pass (MBMP) approach and utilized to evaluate six encoder backbones. Across five labeled-data portions ranging from 20% to 100%, SimCLR pretraining improved plume segmentation compared to ImageNet-only initialization. In the full-data scenario, MobileNet achieved an F1-score of 0.90 with an Intersection over Union (IoU) of 0.80, while Shifted Window Transformer (SwinT) reached an F1-score of 0.85 with an IoU of 0.75. The benefit of self-supervised pretraining was most evident with limited labeled data, where ImageNet-only models degraded substantially, while SimCLR-pretrained encoders achieved higher accuracy. Moreover, the Integrated Mass Enhancement (IME) method was employed for quantifying the emission flux rate. MobileNet provided the strongest agreement with reference emission estimates, achieving an RMSE of 1690 kg/h. Finally, the results demonstrate that SimCLR-based SSL substantially enhances CH4 plume detection from Sentinel-2 imagery and supports more reliable emission quantification for large-scale CH4 monitoring.

Column-averaged methane (XCH4) data spanning 2018-2023 from the European Space Agency (ESA) Tropospheric Monitoring Instrument (TROPOMI) on the Copernicus Sentinel-5 Precursor satellite are assessed for evidence of methane (CH4) emissions from permafrost. We generated bi-monthly anomaly maps of XCH4 from TROPOMI and soil temperature (Tsoil) from reanalysis data for all land north of 50°N. Considering the XCH4 anomalies in the contexts of soil carbon content and wind variability led to a focus on Canada’s Hudson Bay Lowlands (HBL), Earth’s second largest peatland complex (~325,000 km2), which is underlain by continuous to isolated permafrost. This sub-Arctic region is vulnerable to rapid climatic warming and exhibits wind conditions favorable for emission detection from space. HBL XCH4 anomalies strongly correlate with soil temperature anomalies (R = 0.626 to 0.866), consistent with wetlands as the primary CH4 emission source; however, the strong increase in CH4 emissions over 2018-2023 may also suggest a contribution from permafrost thaw and expansion of thermokarst fens.

This study leverages TROPOMI satellite observations and atmospheric inversion modelling to quantify methane emissions from urban wetlands across six major North American cities, including Toronto, Montreal, New York, Los Angeles, Houston, and Mexico City. By coupling high-resolution column-averaged methane measurements with the GEOS-Chem chemical transport model via the Integrated Methane Inversion (IMI) platform, the research distinguishes emissions from both natural wetland and anthropogenic urban sectors. Results indicate that prior inventories substantially underestimate urban wetland methane emissions in most cities. Posterior wetland emissions are resolved alongside dominant anthropogenic sources such as landfills, energy systems, and wastewater, revealing spatially distinct patterns and highlighting seasonal wetland flux variability. The findings demonstrate that urban wetlands, although representing a relatively smaller source compared to anthropogenic emissions, display considerable underrepresented contributions to local methane budgets, underscoring the need for robust, integrated monitoring in urban environments. This methodology provides a scalable framework for routine urban wetland methane flux quantification and supports evidence-based climate mitigation and land management strategies.

Methane plume detection from medium-resolution multispectral satellites such as Sentinel-2 remains challenging due to weak methane signals and strong background variability across land cover, illumination conditions, and atmospheric states. To advance automated detection capabilities, we develop a large-scale benchmark dataset that combines simulated methane plume enhancements with real Sentinel-2 imagery, covering a wide range of emission magnitudes and diverse environmental scenarios. The dataset includes over 64,000 samples and incorporates methane-sensitive inputs derived from the MBMP retrieval workflow, providing a comprehensive foundation for robust model training and evaluation. Building on this dataset, a hybrid transformer–U-Net architecture is proposed, integrating global self-attention with Grouped Attention Gates to enhance feature fusion and improve segmentation of methane structures. The model achieves high accuracy on the benchmark dataset and demonstrates strong generalization to real emission events in complex environments. The combined contributions of the benchmark dataset and hybrid model offer a promising path toward reliable, scalable methane plume monitoring using widely available multispectral satellite observations.

Methane (CH₄) is a potent short-lived climate pollutant, making the detection of major point sources (“super-emitters”) crucial for mitigation. The Sentinel-5 Precursor (S5P) mission, with the TROPOMI instrument, captures global methane concentrations at ~7 × 5.5 km resolution with near-daily coverage. While this resolution is too coarse to identify emissions from individual facilities, its revisit frequency allows effective regional monitoring. Conversely, high-resolution (HR) imaging spectrometers like Carbon Mapper’s Tanager (~30 m) and NASA’s EMIT (~60 m) provide detailed plume mapping but have limited spatial and temporal coverage.

Carbon Mapper releases open-access, high-resolution plume products including georeferenced rasters and metadata. In this study, these HR detections serve as reference events to assess their visibility in coarser Sentinel-5P observations. The workflow includes curating HR events, summarizing their emission context, and inspecting nearby Sentinel-5P data for consistent methane enhancements. The method is exploratory and avoids presupposing Sentinel-5P’s success or failure in detecting plumes at this scale.

This analysis bridges the gap between frequent global monitoring and targeted HR observations. It establishes a path for future cross-sensor integration, combining HR spatial precision with Sentinel-5P’s temporal continuity. With additional labeled data, this approach could inform machine-learning tools for methane anomaly detection and plume segmentation, improving operational methane monitoring across scales.

The development of high-performance, affordable UAVs has transformed vegetation monitoring, enabling observation of forest canopies at an unprecedented level of detail. UAV-derived datasets now provide high-fidelity structural and physiological information at the individual tree level across entire forest stands, offering novel insights into forest dynamics. In the context of increasing tree mortality, such data are becoming essential for understanding forest resilience and adaptation. However, exploiting this data requires effective individual tree crown segmentation algorithms (ITCS) at the forest scale, capable of tackling large-scale data and variability introduced by the environment. In this paper, we developed a new workflow designed to process UAV hyperspectral imagery at the forest scale, enabling automated ITCS and analysis. Our pipeline integrates hyperspectral-to-RGB conversion, ITCS, and centroid-based mask fusion. To assess the performance of our pipeline, we evaluated the model on two replicated white spruce common gardens in Canada, each comprising approximately 6,000 trees of similar age and structure. The experiments rely on a large multi-temporal dataset of hyperspectral imagery acquired during 60 UAV missions between 2022 and 2024, allowing us to evaluate the robustness of the proposed pipeline across a wide range of seasonal and acquisition conditions. Results show that the proposed pipeline achieves a mean segmentation performance of 0.536 mAP (0.885 mAP50) on the annotated dataset. At the forest scale, the system demonstrates strong detection capability with F1-scores of 0.948 at the Pintendre site and 0.863 at the Pickering site, successfully detecting most trees while maintaining stable performance across varying environmental conditions.

Individual tree detection and crown delineation (ITDCD) in dehesa landscapes is complicated by geometric distortions from steep terrain, varying tree densities, and the partly multi-crown 'broccoli-like' structure of holm and cork oaks. This study evaluates the usability of a modified StarDist deep learning model, which has recently shown effectiveness for ITDCD in Canadian forests. Moreover, this study develops a workflow transforming the original StarDist, designed for microscopy images, into an ITDCD solution, taking the georeferencing of geospatial data into account. The tile-wise organized ground truth dataset is created with the pretrained Tree Segmentation model available in the ArcGIS Living Atlas, combined with manual revision. Several augmentation methods are applied, resulting in 960 images, which are split into 85 % for training and 15 % for validation. Following the approach of the Canadian forest study, the StarDist implementation is modified by introducing a constraint to the probability loss function. Rather than computing loss across all pixels, the modified loss function considers only pixels explicitly annotated as objects, while background pixels are excluded. An additional dataset of 1,200 trees serves as ground truth for testing the prediction across the entire study area. Using an Intersection over Union of 0.5, this test demonstrates good performance (Accuracy: 87.50 %; F1-score: 0.85). The accuracy varies with tree density: in areas with sparse tree cover, nearly all tree crowns are detected; in moderately dense areas, a number of tree crowns are missed; whereas in very dense tree layers, the frequency of missed detections increases.

Individual tree crowns detection and delineation from airborne LiDAR data is essential for forest inventory, carbon stock estimation, and ecosystem monitoring. In mixed-wood forests, however, this task remains difficult due to high stand density, multi-layered canopy structure, and the wide variation in crown size and shape across coniferous and deciduous species.

This study addresses two core limitations of existing deep learning methods for individual tree crown delineation. Standard instance segmentation models rely on blind anchor-based proposals that frequently miss small understorey trees in dense canopies, and their pixel-based mask representations struggle to accurately capture crown boundaries for small or irregular crowns.

We propose a multi-task learning framework that jointly trains a structure-aware treetop detection head and a crown segmentation head on a shared backbone network. The treetop detection head generates spatially precise crown seeds guided by canopy height and allometric relationships, replacing blind anchor proposals with data-driven initialisation. Two segmentation strategies are evaluated within this framework: a Mask R-CNN pixel-based approach and a StarDist contour-based approach.

Experiments are conducted on a high-density airborne LiDAR dataset acquired over a mixed-wood forest in Ontario, Canada, comprising 4,417 manually delineated reference crowns. Results demonstrate improved detection completeness for small crowns and more accurate boundary delineation for overlapping larger crowns compared to single-task baselines.

This study examines the use of multi-temporal UAV hyperspectral and LiDAR data to identify tree species in a mixed deciduous forest in southern Ontario, Canada. Weekly UAV flights were conducted from summer through spring to capture structural and spectral changes associated with leaf development, senescence, and leaf drop. Field measurements were collected to provide species labels and biometric information for individual trees.

LiDAR data are processed to delineate individual tree crowns and to derive structural metrics such as crown height, width, density, and vertical canopy profile. Hyperspectral imagery, consisting of more than 300 bands, is co-registered with the LiDAR-derived crowns to extract spectral signatures and compute vegetation indices. These data support the development of a spectral library for the main species in the study area.

The multi-temporal dataset allows evaluation of how phenological changes influence separability among species. Early leaf loss in autumn and differences in budburst timing in spring are expected to produce temporary structural and spectral contrasts that aid classification. Machine learning models, including random forest and neural networks, are applied to assess the contribution of structural, spectral, and seasonal features to species discrimination.

Accurate 3D reconstruction and instance segmentation of soybean seedlings are crucial for early phenotyping and precision agriculture. This study presents a UAV-based sparse-view 3D reconstruction and plant-level segmentation framework that integrates 3D Gaussian Splatting (3DGS) with Mobile-SAM, enabling efficient and high-fidelity modeling under routine field conditions. Traditional LiDAR and MVS approaches, while detailed, are constrained by cost, acquisition density, and computational complexity. By contrast, 3DGS offers explicit Gaussian primitives for fast rendering and direct geometric access but often fails under sparse-view UAV imagery due to weak multi-view constraints and repetitive canopy structures.

To overcome these limitations, the proposed method introduces a mask–geometry co-optimization mechanism: YOLO-generated bounding-box prompts guide Mobile-SAM to produce accurate single-view plant masks, which serve as semantic priors to associate 2D observations with 3D Gaussian primitives. Iterative refinement aligns rendered and observed masks, ensuring spatial consistency and coherent 3D plant boundaries.

Field experiments on a soybean plot demonstrated the method’s effectiveness, achieving high reconstruction quality and visually precise seedling segmentation. The resulting 3D models capture fine structural details and distinct plant instances even under sparse-view UAV data.

This work highlights the potential of combining explicit geometric modeling and lightweight semantic segmentation to achieve robust, scalable, and field-deployable 3D crop reconstruction, offering a promising pathway for high-throughput plant phenotyping and yield estimation in real-world agricultural applications.

In this study, we propose an upscaling approach based on 8-band PlanetScope SuperDove imagery (Coastal Blue, Blue, Green I, Green, Yellow, Red, Red Edge, NIR) combined with UAV data. We employed an evidential Dirichlet neural network to estimate the fractional cover of 13 herbaceous and shrub species typical of Mediterranean coastal dunes, previously mapped at 3 cm using a traditional Random Forest classifier trained on UAV multispectral samples. The overall goal is to enable large-scale mapping of coastal vegetation using high-resolution satellite imagery.

Climate change is intensifying droughts, creating an urgent need to understand these events and take necessary mitigation actions. This work focuses on the Bundelkhand region of Uttar Pradesh, India, an area that frequently experiences severe water stress and is highly susceptible to drought. We used multi-source remote sensing datasets to monitor drought conditions through established drought indices. The analysis period spans from 2000 to 2021. CHIRPS data were used to calculate SPI and RAI, while PKU GIMMS NDVI data were used to calculate VCI. ERA-5 Land was used for soil moisture data to derive SMCI. To track how drought propagates, we performed a correlation analysis between the indices representing meteorological and agricultural drought. The results show that the datasets and the methods are suitable for identifying droughts in the region. Historical drought episodes were accurately detected, and the analysis of the 2015 drought revealed its onset from June to September, which aligns with the monsoon season in Bundelkhand. The datasets and indices used provide a practical and reliable output for sparse ground-based observations for regional drought monitoring and management.

The Near-global Combined Drought Monitoring (NEC-DROMO) dataset provides a comprehensive depiction of drought conditions by integrating multiple agro-climatological variables across global land areas. Spanning 2002–2021 at a monthly time step and 0.25° spatial resolution, NEC-DROMO combines soil moisture, vegetation water content (VWC), rainfall, and temperature to capture both agricultural and meteorological drought signals. A key feature of the dataset is the use of Principal Component Analysis (PCA) to derive dynamic, month-specific weights for each variable, allowing the Combined Drought Indicator (CDI) to reflect seasonal and regional variability in drought drivers.

The dataset is built primarily on inputs from the ECoHydrological Land Reanalysis (ECHLA), which provides soil moisture, VWC, and temperature derived from passive microwave observations assimilated through a land–vegetation model. Rainfall fields are obtained from the ERA5 reanalysis, ensuring consistency across atmospheric and land-surface conditions. Validation against satellite-based indicators, ground observations, and event-based disaster datasets demonstrates NEC-DROMO’s strong capability to reproduce observed drought patterns globally.

With its multi-variable foundation and long-term coverage, NEC-DROMO serves as a valuable resource for drought monitoring, climate analysis, food-security assessment, and agricultural risk management. It supports detailed historical analyses and offers an integrated perspective for users seeking reliable, spatially consistent drought information.

Tropical terrestrial vegetation is critical to the global carbon cycle but faces escalating drought threats. Traditional assessments using fixed climate thresholds often ignore actual physiological responses and non-moisture disturbances. To address this, we developed a novel framework that isolates the true physiological impacts of atmospheric and soil moisture (SM) deficits to identify growing-season vegetation droughts (1982–2019). Results reveal pantropical increases in drought intensity, with tropical forests experiencing significantly sharper intensifications than other biomes. Regionally, African forests exhibit the most severe expansions in drought intensity and area. Interpretable machine learning attributes this intensifying drought predominantly to declining SM (NDVI: 52.1%; LAI: 53%). Finally, while reliable historical reconstruction is vital for future projections, CMIP6 models fail to reproduce these observed trends. These findings highlight mounting drought pressures on tropical forests and underscore the critical need for improved climate models to inform mitigation strategies.

Extreme heat events (EHEs) pose growing risks to densely populated subtropical cities such as Hong Kong, yet there remains a need for lightweight, interpretable tools that can provide multi-day forecasts based on readily available observations. This study develops a multi-source data driven framework that integrates aerosol optical depth (AOD), land surface temperature (LST), precipitable water (PW), and precipitation (Precip), together with ARIMA-based anomaly features, to predict EHEs over Hong Kong. Using a seven-day sliding window, independent XGBoost classifiers are trained to forecast daily EHE occurrence probabilities for the next 1–5 days over ten climate years (March 2015–February 2025). A lead-specific threshold optimization on a validation subset is applied to maximize F1-score. Test results show that AUC values for Lead 1–Lead 5 remain between 0.935 and 0.883, with F1-scores between 0.738 and 0.639, indicating robust predictability up to five days in advance. A process-scale duration inference method based on the leading continuous segment of the predicted sequence achieves 67.08% exact-match accuracy, 77.69% accuracy within ±1 day, and a mean absolute error of 0.75 days. The proposed framework is computationally efficient and operationally relevant, offering practical support for urban heat early warning and risk management.

Anthropogenic climate change and land use transformations are interactively reshaping environmental risks. This study investigates the critical feedback between Land Use/Land Cover (LULC) change and shifts in Köppen-Geiger (KG) climate zones in Taiwan from 2001–2020, and their combined impact on disaster hotspots.

Using MODIS and CHIRPS data alongside a comprehensive disaster inventory, we quantified the spatial co-occurrence of LULC change and climate zone transitions. Our preliminary results reveal a significant climatic shift, with over 10,500 km² transitioning from tropical monsoon (Am) to a drier tropical savanna (Aw) climate, alongside substantial wetland loss and urban expansion. We hypothesize that these dynamic "climate transition zones" are emerging fronts of heightened disaster risk.

Our analysis tests whether areas undergoing active climate reclassification concentrate a disproportionate share of historical landslides and floods. The expected outcome is a novel, dynamic risk assessment framework that moves beyond static models. By identifying these emerging hotspots, this research provides a critical tool for proactive land-use planning and climate-resilient disaster risk reduction, with methodologies applicable to other complex, hazard-prone regions.

Climate change is amplifying the occurrence and intensity of Extreme Event Induced Disasters (EEID), such as floods and wildfires, which increasingly threaten societies and ecosystems. Fast and accurate monitoring tools are therefore essential for damage assessment and emergency response. Remotely sensed data, particularly from the Copernicus Sentinel-2 mission, provide valuable multispectral information for large-scale environmental monitoring, but their manual analysis remains time-consuming.

Recent advances in Deep Learning (DL) have enhanced classification, segmentation, and change detection of geospatial data. New multimodal Prompt-Based (PB) architectures integrate image and text inputs via Text Encoders (TEs), enabling zero-shot detection of previously unseen objects. These models promise flexible, prompt-driven analysis but often underperform compared to Object-Specific (OS) models optimized for particular tasks.

In Earth Observation (EO), foundation models such as Prithvi-EO and TerraFM mark a major step forward, offering generalized pre-training across vast multi-sensor datasets to support downstream OS tasks with limited data.

While DL traditionally requires coding expertise, commercial GIS platforms now integrate DL tools accessible through Graphical User Interfaces (GUIs), allowing inference and limited fine-tuning of pre-trained models. This democratizes DL access for GIS users but shifts expertise toward model evaluation and interpretability.

This study systematically compares PB and OS models executed through both GUI-based and Python environments using Sentinel-2 flood and wildfire imagery, assessing accuracy, flexibility, and processing efficiency to evaluate the balance between accessibility and performance in the democratization of DL for EEID monitoring.

Real-time depth reconstruction from ultra-high-resolution UAV imagery is essential for time-critical geospatial tasks such as disaster response, yet remains challenging due to wide-baseline parallax, large image sizes, low-texture or specular surfaces, occlusions, and strict computational constraints. Recent zero-shot diffusion models offer fast per-image dense predictions without task-specific retraining, and require fewer labelled datasets than transformer-based predictors while avoiding the rigid capture geometry requirement of classical multi-view stereo. However, their probabilistic inference prevents reliable metric accuracy and temporal consistency across sequential frames and overlapping tiles. We present ZeD-MAP, a cluster-level framework that converts a test-time diffusion depth model into a metrically consistent, SLAM-like mapping pipeline by integrating incremental cluster-based bundle adjustment (BA). Streamed UAV frames are grouped into overlapping clusters; periodic BA produces metrically consistent poses and sparse 3D tie-points, which are reprojected into selected frames and used as metric guidance for diffusion-based depth estimation. Validation on ground-marker flights captured at approximately 50 m altitude (GSD ≈ 0.85 cm/px, ~2,650 m² ground coverage per frame) with the DLR Modular Aerial Camera System (MACS) shows that our method achieves sub-meter accuracy, with approximately 0.87 m error in the horizontal (XY) plane and 0.12 m in the vertical (Z) direction, while maintaining per-image runtimes between 1.47 and 4.91 seconds. Results are subject to minor noise from manual point-cloud annotation. These findings show that BA-based metric guidance provides consistency comparable to classical photogrammetric methods while significantly accelerating processing, enabling real-time 3D map generation.

Satellite-based 3D reconstruction has gained prominence with the advancement of Earth Observation techniques. Recent work on Earth Observation Gaussian Splatting (EOGS) demonstrated the potential of adapting 3D Gaussian Splatting to satellite imagery, enabling rapid Digital Surface Model (DSM) generation from multiple images using Rational Polynomial Coefficients (RPCs) as camera models. However, EOGS suffers from critical inefficiencies: it randomly initializes a large number of Gaussians in volumetric space and relies on opacity-based pruning, resulting in unstable memory footprints and premature loss of fine details—particularly problematic for low-resolution satellite data.

This work presents an improved Gaussian Splatting framework for satellite imagery that addresses these limitations through two key contributions. First, we introduce bundle-adjusted initialization, which leverages geometrically precise points from the bundle adjustment process as initialization seeds rather than random placement. This approach ensures Gaussians are anchored to accurate geometric positions from the outset, significantly improving convergence stability. Second, we propose densification-included optimization, which strategically adds Gaussians in regions requiring detailed reconstruction while maintaining computational efficiency. This selective densification preserves fine-scale features without the memory overhead of EOGS's initial over-allocation strategy.

Our method achieves faster processing times and maintains more consistent memory usage while producing higher-quality DSMs, particularly in challenging low-resolution scenarios. By combining geometric priors from bundle adjustment with adaptive densification, we enable more practical and efficient satellite-based 3D reconstruction suitable for large-scale Earth observation applications.

This contribution presents a comparative evaluation of classical and deep learning–based keypoint detectors for Structure-from-Motion (SfM) reconstruction in challenging Arctic UAV imagery. Snow-covered environments pose difficulties for standard feature matching due to low texture, repetitive patterns, and specular surfaces. While deep keypoint pipelines have shown strong performance on indoor and urban benchmarks, their effectiveness in winter aerial domains remains largely unexplored.

Using multi-view UAV datasets collected across several Alaskan sites, we benchmark three feature-extraction front-ends within a uniform pycolmap-based SfM pipeline: (i) classical SIFT with nearest-neighbor matching; (ii) SuperPoint, a self-supervised convolutional detector–descriptor; and (iii) DISK, a reinforcement-learning–based feature extractor. A simple hybrid approach combining SuperPoint and DISK matches is also tested. All methods share identical geometric verification and bundle-adjustment settings to ensure consistency.

Results show that SIFT remains highly robust on moderately textured Arctic scenes, registering all images and producing the most complete point clouds. SuperPoint and DISK achieve similar reprojection accuracy but struggle with image registration and keypoint coverage on some sequences. Conversely, on extremely low-texture scenes where SIFT fails almost entirely, both deep methods still enable partial reconstructions. Persistent failure cases for all techniques include dense canopy and homogeneous snow.

The study highlights a domain gap between existing deep keypoint models and Arctic aerial imagery, suggesting that domain-specific training and improved spatial keypoint diversity could substantially enhance deep SfM performance in polar regions.

Accurate 3D reconstruction is a key enabler for construction progress monitoring and digital-twin maintenance. However, in tower-crane imagery, persistent dynamic occluders such as hooks and slings violate the static-scene assumption of conventional Structure-from-Motion (SfM), leading to feature mismatches and degraded reconstruction consistency. In this paper, we present a geometry-guided occlusion-handling pipeline for crane-mounted construction-site SfM. Our approach leverages geometric cues from reprojection errors and depth inconsistencies to identify outlier observations, clusters them into spatially coherent prompts, and uses these to guide a foundation segmentation model (SAM2). The resulting per-frame masks are integrated into mask-constrained SfM optimization, ensuring that only static background contributes to reconstruction. Experiments on three real-world crane-mounted sequences (30m, 45m, and 120m) show consistent reductions in mean reprojection error relative to the unmasked baseline. In the most challenging case, the error decreases from 0.962 to 0.872 pixels (9.4%). Compared with a fixed rectangular masking strategy, the proposed masks yield similar reprojection errors while better preserving valid observations and sparse-point completeness. These results indicate that the proposed framework provides a practical geometry-guided strategy for improving internal reconstruction consistency in crane-mounted construction environments.

Video panoptic segmentation (VPS) unifies panoptic segmentation and object tracking by assigning each pixel a semantic class label, or for thing classes, an instance identifier that is consistent across frames. Addressing this task, we propose a novel online VPS method for processing stereoscopic image sequences, which is based on depth-aware kernel-based panoptic segmentation. Specifically, we introduce a geometrical constraint based on predicted bounding boxes into the segmentation of thing instances to overcome the fundamental limitation of kernel-based panoptic segmentation that only appearance information is considered in this step; this regularly leads to panoptic segmentation results in which distinct instances are erroneously merged into one mask. To link detected instances across frames, we propose to extend the commonly employed appearance-based association with a motion-related constraint based on optical flow; this resolves ambiguities in case of instances of similar appearance and, thus, reduces the number of incorrect associations. We experimentally evaluate our method on the publicly available Cityscapes-VPS dataset and compare our results to those of several related methods from the literature. The results demonstrate that our method improves the panoptic quality for a single frame and enhances the instance association across frames, leading to an overall improvement of 3.5% in Video Panoptic Quality on thing classes compared to the employed baseline.

This study addresses the morphological evolution of the Belvedere Glacier (Monte Rosa, Macugnaga – Italy) over the period 2009–2023, using a photogrammetric methodology based on Pleiades (2017) and Pleiades Neo (2023) Very-High Resolution (VHR) satellite imagery, integrated with historical aerial data from 2009. The main objective was to quantify altimetric and volumetric variations of the glacier, assess the intensity of ice mass loss, and analyze the geomorphological effects of the flood event that occurred on August 27, 2023, which generated a major debris flow.

Raster differencing between Digital Elevation Models (DEMs) revealed a significant lowering of the glacier surface. Between 2009 and 2017, the glacier lost approximately 19.3 × 10⁶ m³ of ice (about 2.4 × 10⁶ m³/year), while in the following period (2017–2023) the loss reached 16.9 × 10⁶ m³, with an increased average annual rate of 2.8 × 10⁶ m³/year. These values confirm an acceleration in the ablation process, consistent with other studies (De Gaetani 2021; Ioli 2023) and with the general retreat trend observed in Alpine glaciers due to climate warming.

Wetland characteristics such as size, inundation permanence and timing, and surface hydrological connectivity substantially impact wetland processes and functions. The ability to monitor these types of wetland characteristics, and changes through time, is dependent on the spatial and temporal resolution of the imagery data used to map wetland locations. Existing inventories of surface water features have largely been limited to permanently open water features such as lakes and ponds larger than 1km2 at monthly or annual intervals. To address these limitations a random forest model was trained to predict sub-pixel water fraction (SWF) in Sentinel-2 imagery at 10m and 20m spatial resolution. This approach facilitated the detection of small surface water features, including water features interspersed with vegetation such as wetlands, at a sub-monthly temporal scale. Overall, in the 10m SWF data, small and narrow water features were detected that were not apparent at the 20m scale, the shape of feature boundaries was more precise, and the continuity of narrow channels was better maintained compared to the 20m SWF data. Improved detection of small features and narrow channels supports improved wetland inventories, particularly regarding the inclusion of small wetlands which are important biogeochemical hotspots, and automated surface water connectivity classification. The temporal resolution of Sentinel-2 facilitates the detection of ephemeral inundation and wetland surface hydrologic connections, as well as monitoring changes in inundation and connectivity through time.

Salt marshes are ecologically significant ecosystems increasingly threatened by sea level rise, climate change, sediment disruption, and human pressure. Accurate delineation of marsh boundaries is essential for monitoring spatial and temporal change and informing conservation strategies. Remote sensing imagery provides an efficient means to map these boundaries over large areas. This study used high-resolution WorldView-3 imagery (0.3 m after pan-sharpening) to evaluate two methodological categories for mapping marsh extent in the Fraser River Delta, Canada: index-based thresholding (Global Otsu and Adaptive Otsu) and machine learning classification (Random Forest, K Nearest Neighbors, and Support Vector Machine). Each method produced binary marsh maps that were converted to boundary vectors and validated against field-surveyed marsh edges using spatial accuracy metrics, including mean distance error and RMSE. Adaptive Otsu achieved the highest accuracy (mean distance 0.42 m; RMSE 0.53 m) and effectively delineated boundaries across contrasting marsh conditions. Global Otsu performed moderately (mean distance 0.47 m; RMSE 0.62 m). Machine learning models showed lower accuracy overall: Random Forest (0.56 m; 0.73 m), K Nearest Neighbors (0.57 m; 0.76 m), and Support Vector Machine (0.71 m; 0.90 m). These findings demonstrate that locally adaptive thresholding outperforms traditional thresholding and machine learning classifiers for fine-scale marsh boundary extraction in heterogeneous coastal environments, offering a practical approach for remote sensing-based marsh monitoring.

Multispectral drone sensors enable fine-scale ecological mapping, but added bands can inflate processing costs. We evaluated the MicaSense RedEdge-MX Red and Blue cameras (5 bands each) versus the Dual Camera System (10 bands) for vegetation mapping in two salt marsh sites in Aulac, New Brunswick, Canada (24 classes at the reference site; 15 at the restoration site). Pixel-based Random Forest (RF) classifications were used to compare validation accuracy, variable importance, and processing time for stitching and classification. Five-band maps achieved up to 95% validation accuracy; the 10-band configuration improved accuracy by ≤2%. Band contributions were site dependent: the near-infrared (NIR) band from the Red camera aided classification at the reference site, whereas additional red-edge bands in the Blue/Dual setups improved performance at the restoration site. However, stitching time rose sharply for the Blue and Dual systems, and RF classification time scaled with data volume and class complexity. Overall, the 5-band Red camera provided a cost-effective balance of accuracy and efficiency, offering practical guidance for sensor selection in drone-based salt marsh monitoring.

Protection of wetlands in Canada is becoming increasingly important as the ecological services they provide become more well understood and simultaneously, as the advance of human settlement and impacts of climate change imperil them. Rapid and effective identification of wetland areas is crucial for this protection. While there is an estimated 1.2 million km2 of wetland area across the country, only a very small portion of this area is currently mapped and classified in accordance with the 5 major classes and 9 subclasses of the Canadian National Wetland Inventory (CNWI). Additionally, the mapping that has already been completed in some areas is of limited accuracy. To increase accuracy and reduce the cost of wetland mapping we use a combination of Sentinel-1 SAR and Sentinel-2 Multispectral images with topographical data (an SAR-derived DEM). Seasonal variations in water level and vegetation were accounted for through the acquisition of imagery from both satellites in May, July, and September. Using the Montérégie region of southern Quebec as a case study we use a combination of the images and DEM metrics for the entire study area to classify landcover into 21 classes with the Random Forest classifier. The initial Random Forest classification produced an overall classification accuracy of 96.3%. Our study shows that classifying Sentinel-1 and 2 images allows us to determine the location and type of wetlands with a high degree of accuracy. This will allow for more efficient conservation strategies in the mapped areas.

Coastal marshes provide critical ecosystem services, including habitat for diverse plant, fish, and bird communities, shoreline protection, and carbon storage. These low-lying ecosystems are increasingly threatened by sea-level rise and human pressures, necessitating systematic monitoring to inform conservation and restoration efforts. Marsh vegetation characteristics, such as species composition and leaf area index (LAI), are key indicators of ecosystem condition, yet traditional field surveys are often labor-intensive, costly, and spatially limited. High-resolution remote sensing offers a powerful alternative, providing extensive spatial coverage and repeated observations for long-term monitoring. In this study, 30 cm WorldView-3 imagery of the Sturgeon Bank Wildlife Management Area in southern British Columbia, Canada, was combined with machine learning (Random Forest) and deep learning models (2D CNN and Vision Transformer, ViT) to map marsh vegetation species and estimate LAI. Extensive field surveys were conducted at selected sampling points along 24 transects to document species composition and measure LAI, which datasets were used for model training and validation. Results show that the ViT model achieved the highest classification performance (Overall Accuracy 94.05%, Kappa 93.44%), outperforming CNN and RF, and was used to generate a species distribution map. Random Forest, while less effective for classification, accurately estimated LAI (R² ~0.85), producing an LAI map that, combined with the species map, revealed species-specific growth patterns. These results demonstrate the effectiveness of high-resolution remote sensing and advanced analytical models for detailed characterization of complex coastal marsh ecosystems, supporting both ecological understanding and local conservation planning.

The intensity, frequency, and duration of wildland fires are growing in Canada and around the world. Timely fire intelligence products from remote sensing platforms can assist fire managers and lead to fewer impacts. New onboard processing techniques using machine learning allow greater levels of analysis and refinement on edge devices like aircraft and satellites, reducing bandwidth and latencies. Our Fire Band Analysis Network approach brings together wildfire science and management experts and academics, an Indigenous owned business that specializes in satellite communication and community outreach, and a Canadian space company with expertise in deploying machine learning models to spacecraft. We show initial results with onboard segmentation models and present a path to prototype this onboard processing model on a cubesat currently in orbit and on drones equipped with infrared sensors, ultimately bringing the derived data products to user communities on the ground.

This presentation will provide an overview and update on the WildFireSat mission and its data product algorithm development. Specifically, we will summarize the 2025 Science and Applications Plan and share updates from the Tier 2 stage of products and algorithms. The Tier 2 products that will be shown include the multi-source fire events, time of arrival outputs, and satellite-derived fire behaviour products (e.g., satellite-observed rate and direction of spread, fireline intensity). Ongoing science-development activities include algorithmic validation, uncertainty characterization, and completion of algorithmic theoretical basis documents.

Built through Canadian and international partnerships, WildFireSat will support fire monitoring and management while enabling major scientific advancements for the global fire monitoring community. The scientific applications of WildFireSat are broad, covering all stages of a fire event’s life cycle. By prioritizing the needs of wildfire managers and a broad range of end-users, the WildFireSat mission is a strong model for future satellite missions to integrate user engagement throughout all phases of the mission timeline.

Normalized Hotspot Indices (NHI)—originally developed for volcanic hotspot detection—has emerged as a powerful, flexible tool for the identification and characterization of high-temperature sources using Sentinel-2 MSI and Landsat-8/9 OLI/OLI-2 observations. By exploiting the combined radiance information from the Near Infrared (NIR) and Short-Wave Infrared (SWIR) spectral bands, the NHI algorithm leverages the multispectral capabilities to identify and characterize hotspots of various origins.

A specific configuration of the NHI algorithm has recently been developed for wildfire mapping. This improved version demonstrated strong performance in complex environments such as California, Hawaii, Canada, Greece, Spain, and Australia, significantly improving the delineation of flame fronts and substantially reducing omission and commission errors.

In this work, we present the results of applying NHI-F to various wildfire events, including the wildfires in Canada in May 2025. Our analysis focuses on two main dimensions essential for modern fire science: (i) the spatial characterization of active flaming fronts and burned-area dynamics at 20–30 m scale and (ii) the quantification of fire intensity through Fire Radiative Power (FRP) and SWIR-based radiance metrics.

In the initial response to wildfires, securing rapid and accurate geographic information is essential. However, helicopter imagery acquired on-site often lacks precise sensor metadata, such as camera pose and internal parameters, making the application of georeferencing difficult. In particular, obliquely captured wildfire imagery presents additional registration challenges due to severe viewpoint changes, scale variations, and low-texture environments. This study proposes an automated georeferencing pipeline capable of operating under these constraints. The proposed method consists of five stages: preprocessing, image retrieval, feature extraction and matching, Exterior Orientation Parameters (EOP) estimation, and orthomosaic generation. An initial Area of Interest (AOI) is defined using inaccurate initial position data, and the Region of Interest (ROI) within the reference map is obtained through a ResNet50-based image retrieval approach. Subsequently, virtual Ground Control Points (GCPs) are generated through deep learning-based feature matching. Elevation data is then assigned using a Digital Elevation Model (DEM), and EOP are estimated via Perspective-n-Point (PnP) and RANSAC algorithms. Intermediate frames are initialized via interpolation and refined through bundle adjustment to produce the final orthomosaic. Experimental results demonstrated that utilizing SuperGlue and LightGlue complementarily increased the number of successfully georeferenced intervals from 5 to 9. Furthermore, a minimum RMSE of 28.30 m was achieved in the most accurate interval. This method proves that by automating the feature-based georeferencing process, practical geographic information can be rapidly provided for initial disaster response, even in sensor-limited environments.

California experiences globally intense wildfire activity with accelerating human casualties and economic losses. Existing research quantifies anthropogenic and climatic contributions to wildland-urban interface (WUI) fires at aggregate levels, yet overlooks heterogeneity arising from differences in ignition locations and dominant spread areas. Using multi-source data from California (2002–2023), we classified WUI fires into four behavioral modes based on ignition site and primary spread zone: I-I (WUI ignition, WUI spread), I-W (WUI ignition, wildland spread), W-I (wildland ignition, WUI spread), and W-W (wildland ignition, wildland spread). We systematically analyzed size characteristics, inter-annual trends, fuel composition, and climate sensitivity across modes. Key findings include: (1) WUI fires accounted for 95.6% of total burned area from large fires, with only 12.2% of burned area within the WUI; both total and mean burned area increased significantly over two decades. (2) Lightning-caused WUI fires showed significantly delayed ignition dates, whereas human-caused fires occurred significantly earlier, with elevated fire frequency observed during Independence Day, Labor Day, and Thanksgiving. (3) I-I fires were predominantly driven by anthropogenic factors with the highest proportion of shrub/grass fuel and the smallest mean size; W-W and I-W fires exhibited significant climate sensitivity, with I-W showing a higher rate of increase than W-W over the study period. These findings reveal differentiated driving mechanisms across WUI fire behavioral types, providing scientific evidence for targeted fire management strategies.

This study examines how street-view scenes relate to urban perception using a scene-graph-driven modeling method. Each image is parsed into subject–predicate–object triplets; entity appearance from a CNN backbone and relation semantics from a Transformer detector are fused at node level via a learnable gate. A relation-aware graph neural network performs message passing and attentive readout to predict six perception dimensions (beautiful, boring, depressing, lively, safe, wealthy). Taking Place Pulse 2.0 dataset as benchmark, we convert pairwise votes to binary labels per dimension with standard train/validation/test splits. Experiments compare the graph approach against CNN+SVM and Transformer+SVM baselines under identical protocols. Results show consistently higher accuracy across all six dimensions, with notable gains for beautiful and wealthy. Gradient and integrated-gradient analyses offer node- and edge-level attributions, highlighting elements such as trees, facades, and overhead wires. The method balances accuracy with clarity, and the results point to practical cues that can support human-centered urban design.

Road surface conditions decline due to heavy traffic volumes, severe weather, and recurring utility works, yet still, many road agencies still rely on manual windshield surveys and semi-automated inspections. Not only are these methods time-consuming, but also difficult to scale and labour-intensive. With the help of recent advances in deep learning and the widespread availability of built-in vehicle dashcams, they offer new opportunities for low-cost, automated pavement assessments. This contribution presents a mobile, dashcam-based framework for detecting road-surface defects using the latest YOLOv11, which is combined with geolocation tagging for spatial visualization.

To test out our YOLOv11 training model, we conducted the initial dataset at the University of the Fraser Valley campus and manually annotated it to identify crack fillings, crosswalk markings, speed bumps, lane markings, and other surface conditions. This was just a prototype, which would later be trained to detect all road conditions, such as gravel, potholes, and uneven roads, as well. To address variations in lighting and motion, augmentation techniques were applied. YOLOv11 acquired a mean average precision above 90% across all tested categories.

This prototype demonstrates a practical, low-cost approach for real-time pavement monitoring. Future work includes expanding data collection, developing an operational dashboard for road authorities, having exact GPS coordinates pinned on maps with damaged road images, and evaluating model performance across different data sources, including models trained through Google Images. By producing actionable geospatial information, this system supports more efficient maintenance workflows and offers a scalable pathway for municipalities seeking to modernize road-condition assessment.

Global population is projected to increase by about 70% by 2050, with a growing proportion of people living in urban areas. This trend highlights the importance of accurately assessing urban expansion. Automatic building detection from remotely sensed imagery using deep learning (DL) has demonstrated considerable potential for applications, including sustainable urban planning and infrastructure monitoring. However, the inherent black-box nature of DL models limits their transparency and reduces trust in model-driven decisions. Although various Explainable Artificial Intelligence (XAI) approaches have been proposed to highlight image regions influencing model predictions, qualitative visual inspection alone is insufficient for reliably evaluating the credibility of these explanations. This study evaluates several XAI techniques for building footprint extraction using a U-Net model trained on a refined Massachusetts Buildings Dataset. The segmentation model achieved precision, recall, F1-score, IoU, and overall accuracy values of 89.68%, 85.69%, 87.53%, 79.03%, and 94.35%, respectively. To investigate the model’s decision-making process, three explanation methods, namely Saliency, GradientSHAP, and GuidedGradCAM, were applied. The quality of the generated explanations was then quantitatively assessed using 16 evaluation metrics. Beyond single-image analysis, a dataset-level evaluation was conducted using 547 image patches containing building coverage greater than 20%. The results indicate that GuidedGradCAM produces more consistent and reliable explanations. Furthermore, dataset-level analysis using dense-building samples provides a statistically more robust representation of overall model behaviour compared to evaluations based on individual images. These findings highlight the importance of quantitative assessment in validating the interpretability of DL models for building footprint extraction.

Urban rooftops represent a critical vertical resource for sustainable development, yet comprehensive assessment of their utilization patterns and available capacity remains constrained by inadequate datasets and limited algorithmic capabilities. This study introduces the Urban Rooftop Utilization Dataset (URUD), the first multi-city, pixel-level semantic segmentation dataset encompassing 1,560 high-resolution satellite images from four Chinese cities. URUD establishes eight semantic categories including a novel "available area" class to address ambiguous regions that existing classification schemes fail to capture. The study further proposes MaskRoof, a transformer-based deep learning framework specifically designed for fine-grained rooftop analysis. The model integrates two task-specific modules, Hierarchical Zoom-in Attention (HZA) and Prior-Guided Cross-Attention (PGCA), to address challenges of small-scale target detection and class imbalance. Experimental results demonstrate that MaskRoof achieves superior performance with 94.46% accuracy and 47.29% mIoU, outperforming existing segmentation architectures. Application to Shanghai's outer ring area reveals that 60.74% of rooftop space remains available for utilization, with significant spatial heterogeneity across building types. Industrial and warehouse structures retain substantially greater unutilized areas compared to office and residential buildings. These findings provide quantitative evidence for differentiated urban planning strategies and demonstrate the framework's capability for large-scale rooftop potential assessment in complex urban environments.

Real-time defect detection in solar farms is critical for profitability and safety. This paper compares state-of-the-art (SOTA) object detection architectures for deployment on edge computing platforms for the purpose of thermal PV defect detection using UAV imagery. We systematically evaluated Closed-Set (YOLOv10, YOLOv12, RT-DETR, RF-DETR) and Open-Vocabulary (YOLO-World, OWL-ViT) models on a public thermal dataset. Our results highlight two leading candidates. The transformer-based RF-DETR sets a new accuracy record at 82.6% mAP@0.50, driven by its self-supervised backbone, but its inference speed is low (12.6 FPS). Conversely, the CNN-based YOLO-World integrates language semantics to reach a competitive 78.1% mAP@0.50 while operating at a real-time speed of 31.3 FPS. We conclude that both RF-DETR and YOLO-World are promising for embedded thermal fault detection. The final selection will depend on on-platform inference performance.

This study investigates the suitability of Distributed Scatterers (DS) for satellite-based bridge monitoring in very high-resolution (VHR) Synthetic Aperture Radar (SAR) data. While Persistent Scatterer Interferometry (PSI) relies on isolated, temporally stable reflectors, the DS concept extends the analysis to statistically homogeneous areas. In bridge monitoring, however, elevated and narrow structures challenge the assumption of spatial homogeneity due to signal contributions from both the bridge deck and the underlying terrain in side-looking SAR geometry.

Using 23 TerraSAR-X Staring Spotlight acquisitions (September 2022 - September 2023) over two highway bridges near Regensburg, Germany, the study analyses the effects of layover and partial pixel mixing on height correction and deformation estimation. The DS identification is based on statistical homogeneity testing and covariance estimation, with coherence thresholds applied to ensure phase stability.

Results demonstrate that bridge decks exhibit variable coherence depending on surface roughness and illumination geometry. In some cases, overlayed signals from bridge and ground surfaces produce erroneous elevation and deformation values. The analysis highlights the need for careful interpretation of DS results in VHR data and provides insights into the limitations and potential of DS-based InSAR for linear infrastructure monitoring.

Taipei has experienced an important urban development in the recent years with the expansion of its Taipei Mass Rapid system (MRT). This expansion is currently taking place in the Tamsui-Xinyi Line (Red Line) with one new metro station, the Guangci Fengtian Temple Station. This station connects the east part of the Xinyi district as the continuation of the Xiangshan Station. This project extension has been claimed to be one of the most difficult ones in the metro line development due to its complex geological setting going from very soft sediments to hard rock in a few meters. We have employed Sentinel-1 SAR images to measure the tunnel excavation settlement utilizing ascending and descending tracks and estimating vertical and horizontal time series deformations.

Structural health monitoring is essential for building safety. While SAR provides all-weather, non-contact imaging, it is often affected by geometric distortions like layover and foreshortening, making it difficult to extract accurate 3D structural information from complex targets like buildings. Inspired by stereo vision, we propose a stereo SAR mode that acquires two images via a single rotation. By transforming Cartesian to polar coordinates, the disparity is constrained to the angular direction, significantly simplifying the matching process. We derive the nonlinear relationship between height and disparity and apply Newton’s iterative method for accurate 3D reconstruction. Real data collected by a millimetre-wave radar system validate the effectiveness of the proposed approach.

3D city models have become an important piece of geoinformation. They are available in different Levels of Detail (LoD), which determine the amount of complexity provided in the model. LoD1 city models represent simple prismatic building volumes and are typically produced by means of remote sensing. In this article, we investigate the possibility for country-wide reconstruction of LoD1 city models from TanDEM-X intensity images by utilizing deep learning-based single-image height and building footprint reconstruction. As study area, we use the land surface of the country of Denmark. Our results show the general potential of this AI-based approach of country-wide city model reconstruction, which can serve as a data-efficient pipeline that is particularly well-suited in time-critical scenarios or for the exploitation of archive imagery of satellite missions with global data coverage.

Interferometric Synthetic Aperture Radar (InSAR) is widely used to measure millimetre-level deformation of bridges and other struc-tures. However, retrieving multi-dimensional displacements of a bridge and integrating these measurements with structural stress for coupled analysis remains a major challenge. To tackle this issue, we propose an integrated framework and demonstrate its application on the Hutiaohe extra-large bridge in Guizhou Province. First, a two-dimensional E-PS-InSAR time-series processing chain is de-veloped to derive the bridge’s bi-directional deformation. Next, structural temperatures are obtained through the ANUSPLIN interpo-lation scheme, allowing the accurate isolation of the thermal response. Finally, the finite-element model (FEM) of the bridge is con-structed to interpret the observed deformation and thermal signatures within the structural context. The results show that, compared to conventional InSAR approaches, the proposed framework yields a richer set of insights by conducting a joint analysis mul-ti-dimensional deformation, structural behavior and thermal effects.

Monitoring linear infrastructures such as railways and highways with Multitemporal Interferometric Synthetic Aperture Radar (MTInSAR) requires to identify spatial displacement gradients to assess and mitigate the related hazard. During conventional MTInSAR, the majority of the processed pixels are not directly relevant to the linear infrastructure. However, these pixels are required to aid the phase unwrapping and to remove the atmospheric phase contribution. To overcome this limitation, we propose a new method that directly estimates the spatial gradient from the Synthetic Aperture Radar (SAR) images solely along the linear infrastructure avoiding costly phase unwrapping, error propagation from pixels outside the linear infrastructure and atmospheric filtering. Our experiments based on high and medium resolution images from TerraSAR-X and Sentinel-1, respectively, demonstrate that the estimated spatial gradients agree well with the MTInSAR results with a maximum Root Mean Square Error (RMSE) of 3.5 mm/year. Applying our method on Sentinel-1 images enables computationally efficient monitoring of linear infrastructures exploiting the wide area coverage and availability of the SAR images.

This study proposes a deep learning–based enhancement of feature tracking to improve Sea Ice Drift (SID) estimation using Sentinel-1 Synthetic Aperture Radar (SAR) imagery. Traditional computer vision methods, such as Oriented FAST and Rotated BRIEF (ORB), are commonly used for generating initial drift vectors within the Nansen Environmental and Remote Sensing Center (NERSC) workflow; however, their performance declines under rotational variations, low-texture surfaces, and the fluid-like, short-term dynamics of sea ice. To address these limitations, this study evaluates two deep learning–based methods—SuperGlue and the Local Feature Transformer (LoFTR)—to enhance the robustness and accuracy of feature matching between consecutive SAR scenes. Furthermore, to effectively utilize multi-polarization information, a multi-polarization strategy was applied across both the feature tracking and pattern matching stages. Performance was evaluated using in-situ drift observations from Ice-Tethered Profiler (ITP) buoys, with feature matching assessed by the number of matched keypoints and estimated SID vectors, and drift accuracy evaluated using RMSE and the coefficient of determination (R²). Experimental results demonstrate that polarization integration significantly improves performance, reducing RMSE and increasing R². Among the methods, LoFTR achieved the best performance, followed by SuperGlue and ORB, with notable reductions in speed and directional errors. Overall, the findings demonstrate that deep learning–based methods substantially improve the stability and accuracy of SAR-derived SID estimation. These methods enable more stable and reliable performance in the Arctic environment, which is characterized by sea ice reduction, strong seasonal variability, and highly dynamic drift patterns.

global sea ice distributions on a daily basis. Ice concentration (IC) is one of the most important sea ice parameters derived from brightness temperatures measured by the microwave radiometers. However, even at microwave frequencies, the brightness

temperature data over open ocean areas are affected by the presence of adverse weather conditions, including elevated atmospheric water vapor, cloud liquid water, and abnormal surface roughness conditions. The net result is the retrieval of moderate sea ice concentration values in the open ocean where sea ice is not expected. The current sea ice algorithms make use of what is called a “weather filter” to correct such false retrieval of sea ice, but significant areas in the ice-free water that have the false ice cover remain in some areas. In this study, an improved weather filter, namely the Advanced Weather Filter (AWF), that minimizes, if not eliminates, this problem, developed by Cho et al. (2023), was implemented to produce JAXA/AMSR2 sea ice concentration products

of the Arctic for verification. The AWF was validated and shown to be very effective in selected study regions in the Arctic during the summer time from 30 June to 3 July 2014 and the winter time from 15 December to 18 December 2014, thereby supporting the

integration of the AWF into the standard AMSR2 sea ice concentration product. The AWF should be broadly applicable and can be implemented in other satellite passive microwave ice concentration datasets.

Seasonal soil freeze-thaw (FT) transitions govern critical hydrological and biogeochemical processes across northern landscapes. The physical state of freezing soil exists on a thermodynamic continuum influenced by the zero-curtain effect, a period where latent heat exchange stabilizes temperatures near 0°C. Despite this, operational passive microwave algorithms, such as FT-SMAP and FT-ESDR, enforce discrete binary classifications that mask this biogeochemically active partially frozen period. To address this limitation, this study establishes a probabilistic, non-binary FT detection framework using a parsimonious L1-regularized logistic regression model driven by multi-frequency passive microwave observations. To isolate dynamic phase changes from static landscape noise, the model integrates two locally standardized indices: the Normalized Polarization Ratio (NPR) from SMAP L-band to track soil liquid water permittivity, and the Normalized Difference V-Pol (NDV) from AMSR2 Ka/Ku-bands to capture volume scattering within canopies and snowpack. The model was trained using topsoil temperatures from North American networks, employing a probabilistic Soil Freezing Characteristic Curve to isolate high-confidence training end-members and a density-based spatial clustering approach to prevent spatial data leakage. The logistic framework demonstrated robust geographic generalizability, achieving an F1-score of 0.957 in Tundra environments. Crucially, it significantly mitigated false alarms in complex forested canopies, suppressing false positive rates in Mixed Forests to 12.6%, compared to 44.3% for FT-ESDR and 33.5% for FT-SMAP. By mathematically isolating the zero-curtain transition, this scalable approach provides the continuous baseline data necessary for advancing seasonal carbon respiration modeling in rapidly warming northern environments.

This work presents instrument development and calibration of a terrestrial L-band radiometer designed to support satellite retrieval validation and radiation transfer model parameter refinement in the Arctic. As satellite-based retrievals of key geophysical variables such as snow density and ground temperature continue to improve, their accuracy remains limited by scarce ground-truth data. Our refined radiometer addresses this gap by providing targeted, high-resolution terrestrial measurements capable of characterizing surface heterogeneity across Arctic land and water environments.

The instrument was redesigned from an existing model, and was improved based on lessons from earlier field campaigns, focusing on robustness, simplified operation, and enhanced radio-frequency isolation. Calibration procedure focused on measuring the night sky over several nights in cold temperatures to accurately characterize the operation in very cold conditions.

Initial calibration experiments show stable performance and improved consistency compared to earlier instrument versions. While some challenges remain, the system is expected to be field ready and able to capture brightness temperatures accurately over long time periods and varying conditions.

Future campaigns will extend these measurements to lake and sea ice, supported by ground-penetrating radar enabled surface roughness characterization. These efforts will ultimately contribute to improved radiative transfer modeling and more accurate satellite retrievals of key Arctic geophysical variables.

Accurate mapping of retrogressive thaw slumps (RTSs) in permafrost regions remains challenging due to their irregular morphology, blurred boundaries, and strong spatial correlation. This paper proposes a lightweight multi-level self-modulation (MLSM) module embedded into the UNet++ backbone to enhance non-local feature modeling for high-resolution image segmentation. The overall framework is built upon a UNet++ backbone with dense skip connections, where the proposed MLSM module adaptively fuses multi-scale contextual information to enhance feature coherence across spatially correlated regions. By incorporating low-rank regularization through a soft nuclear norm, MLSM dynamically modulates feature responses according to structural variations, allowing attention to adapt to spatially complex RTS regions. The integration of depth-wise convolution and channel recalibration further refines feature aggregation efficiency. Experimental evaluations on Maxar dataset demonstrate that the proposed method achieves superior segmentation accuracy and smoother boundary delineation compared with existing models. The proposed framework provides a robust and computationally efficient approach for RTS mapping, contributing to improved understanding of local geomorphic patterns.

This study investigates long-term snow persistence dynamics across the North-Western Himalaya (NWH) spanning 2000–2024 using MODIS Terra and Aqua daily snow products. Snow persistence—defined as the number of days a location remains snow-covered—is a crucial indicator of climatic variability and hydrological behaviour in high-mountain environments. Annual snow persistence was derived from daily CGF_NDSI_Snow_Cover layers after mosaicking, clipping to the study region, reclassifying snow pixels, and summing snow days at 500 m resolution. Pixel-wise trend analysis was conducted using the Mann–Kendall test, supported by Kendall’s Tau, p-values, and variability metrics.

The results show clear spatial contrasts: high-elevation zones (>4000 m) maintain persistent snow cover (>300 days/year), while mid-altitude regions (1500–3000 m) exhibit moderate persistence but significant negative trends. Low-elevation areas display minimal snow longevity and rapid decline over the 25-year period. The region recorded maximum snow-covered area in 2019 and a notably reduced extent in 2016. Approximately 29% of the NWH shows statistically significant trends, predominantly negative, with an overall mean decline of −3.2 snow days per year. Variability is highest in mid-elevation transition zones, which appear particularly sensitive to warming.These findings highlight ongoing reductions in seasonal snow cover in the NWH and their implications for glacier mass balance, water resource availability, and hydrological timing. The study underscores the value of long-term satellite-based monitoring to understand cryospheric response under changing climate conditions.

This work addresses the emerging challenge of achieving proactive spatial cognition for embodied and spatial AI systems operating in dynamic real-world environments. Conventional mapping and reasoning approaches are largely passive and task-dependent, limiting their ability to build persistent understanding beyond immediate goals. We introduce the Proactive Cognitive Map (PCM), a unified framework that enables agents to autonomously construct, verify, and refine their spatial knowledge through continual perception, self-questioning, and mental simulation. PCM integrates a grid-based perceptual map with a semantic, object-centric memory, forming an explicit and interpretable representation of the environment. A self-questioning module identifies uncertain or ambiguous regions and generates targeted queries, while a simulation module emulates human imagination to perform counterfactual reasoning and lightweight geometric self-verification across time and viewpoints.

We evaluate PCM across episodic-memory embodied QA tasks and the long-horizon, multi-task benchmarks, GOAT-Bench, covering episodic reasoning, continual understanding, and cross-task generalization. Results show that PCM’s self-driven graph construction and proactive refinement outperform goal-specific exploration methods.

By transforming mapping from static perception into a continual cognitive process of questioning, imagining, and verifying, this study provides a step toward lifelong, interpretable, and self-improving spatial intelligence.

As a spatial digital foundation for digital twins and smart cities, the timeliness and accuracy of realistic 3D models are of critical importance. Intelligent and automated data acquisition and update workflows form the core infrastructure that sustains this digital foundation. Current modeling techniques relying on a single data source face inherent limitations: UAV(Unmanned aerial vehicle)-based oblique photogrammetry struggles to capture lower facade details, often leading to geometric distortions and blurred textures, while conventional terrestrial surveying methods suffer from low efficiency and limited automation as well as intelligence. Moreover, the substantial viewpoint differences between aerial and ground data hinder effective fusion. However, recent technological advances in 3D Gaussian Splatting (3DGS), large vision model, multi-sensor SLAM and robotic systems, open up new opportunities to significantly improve the fidelity, efficiency, completeness and automation of 3D reconstruction through the cooperation of UGVs and UAVs.To address the current challenges from 3D reconstruction, this study proposes a novel framework which seamlessly integrates autonomous unmanned systems, state-of-the-art large visual models, multi-sensor SLAM (simultaneous localization and mapping) and cutting-edge 3D Gaussian rendering technology. The framework realizes an integrated workflow for automatic updating building facade and high-fidelity 3D GS rendering using air to ground fusion algorithms with autonomous systems. The primary focus is to advance the automation and intelligence of building 3D reconstruction, thereby enabling efficient updates of urban 3D models.

High-precision digital terrain models (DTMs) are important for Mars explorations and research. However, traditional terrain reconstruction methods suffer from limitations in coverage and resolution. To enhance the model's ability to recover fine-grained topography, we present a diffusion-based monocular terrain reconstruction method, which progressively recovers Martian terrains from single-view high-resolution optical images. We employed a multi-scale U-Net denoising network with attention mechanisms and introduced an additional end-to-end depth constraint. To improve terrain reconstruction efficiency, we implemented a diffusion model in the latent space and adopted a skipping sampling mechanism. We employed the proposed method to reconstruct terrain in different regions. Experimental results demonstrate that the reconstructed terrain achieves an accuracy of 2 m. Furthermore, compared to photogrammetric terrain, the shaded relief generated by our method exhibits greater similarity to the input imagery.

GESM is a Gaussian-mixture sonar mapping pipeline that converts 2D imaging sonar into a continuous 3D probabilistic map for navigation. We estimate posterior occupancy with Gamma-CFAR, cluster occupied and free space along beams, encode them with weighted EM/MPPCA and moment-matched Gaussians, and incrementally merge local mixtures into a globally consistent map. Loop closure is handled by in-place edits of mixture parameters. On simulation and pool/harbour data, GESM yields dense, navigation-ready structure and free water while reducing map memory by ~99% compared with a comparable voxel grid.

This paper investigates how the provenance and resolution of geospatial data used to construct mesh maps affect the accuracy and robustness of mesh-based visual localisation. Mesh-based approaches offer significant advantages over traditional pipelines reliant on Structure from Motion (SfM) models, including the ability to scale to city-sized scenes---by leveraging large-scale data sources such as national mapping databases--- and on-demand generation of arbitrary synthetic views. While prior work has focused on algorithmic improvements to mesh-based localisation, none has systematically analysed how different input data affect localisation outcomes. In this work, we evaluate three meshes---derived from aerial oblique imagery, combined aerial and ground mobile mapping data, and close-range ground imagery---across the egenioussBench Extended and House of Science query sets and four image matchers. We show that mesh quality is the dominant factor governing localisation performance. In the House of Science experiments, aerial meshes lack the resolution required to resolve façade detail, causing near-total localisation failure regardless of matcher. In the egenioussBench Extended experiments, augmenting an aerial mesh with ground data yields consistent but less dramatic improvements. We further introduce the Perceptual Detail Score (PDS), a viewing-condition-aware metric that proves to be a strong predictor of downstream pose accuracy across all experimental configurations.

The accurate recognition of segment joints serves as a critical step for capturing joint anomaly information, evaluating segment assembly quality, diagnosing structural health status, and determining the loosening of connecting bolts. It holds significant importance for the operation and maintenance of shield tunnels. However, existing studies on joint recognition based on Rail-borne Mobile Laser Scanning (RMLS) suffers from insufficient comprehensiveness in feature representation, leading to notably poor accuracy and robustness under complex scenarios such as noise interference, data loss due to object occlusion, and uneven point cloud density. To address this issue, this study proposes a shield tunnel segment joint recognition method based on the Joint Composite

Feature Index (JCFI). The proposed method first employs a cross-sectional ellipse fitting approach to filter out obvious non-lining points. Subsequently, a composite index JCFI, which integrates curvature, left-right density ratio, and relative depth, is designed to

quantitatively characterize the feature differences of segment joints. Finally, based on the constructed JCFI indicator, the recognition of circumferential and longitudinal joints is sequentially achieved. Validation tests using RMLS point cloud data from the Guangzhou Metro Line 8 tunnel demonstrate that the proposed method, by constructing the JCFI that comprehensively characterizes joint features, effectively handles complex scenarios including noise interference, joint missing, and uneven point cloud density. The

joint recognition achieves a recall rate of 90.14%, a precision rate of 99.04%, and an IoU of 89.36%, providing a reliable technical solution for the accurate identification of shield tunnel segment joints.

Multi-modal Satellite Image Time Series (SITS) analysis faces significant computational challenges for live land monitoring applications. While Transformer architectures excel at capturing temporal dependencies and fusing multi-modal data, their quadratic computational complexity and the need to reprocess entire sequences for each new acquisition limit their deployment for regular, large-area monitoring. This paper studies various dual-form attention mechanisms for efficient multi-modal SITS analysis, that enable parallel training while supporting recurrent inference for incremental processing.

We compare linear attention and retention mechanisms within a multi-modal spectro-temporal encoder. To address SITS-specific challenges of temporal irregularity and unalignment, we develop temporal adaptations of dual-form mechanisms that compute token distances based on actual acquisition dates rather than sequence indices.

Our approach is evaluated on two tasks using Sentinel-1 and Sentinel-2 data: multi-modal SITS forecasting as a proxy task, and real-world solar panel construction monitoring. Experimental results demonstrate that dual-form mechanisms achieve performance comparable to standard Transformers while enabling efficient recurrent inference. The multi-modal framework consistently outperforms mono-modal approaches across both tasks, demonstrating the effectiveness of dual mechanisms for sensor fusion. The results presented in this work open new opportunities for operational land monitoring systems requiring regular updates over large geographic areas.

Sensor harmonization is required to produce consistent Landsat imagery for long-term change detection. This study investigated the effect of seasonality and aerosol optical thickness (AOT) on linear harmonization functions, which are frequently used to create consistent Landsat 7 ETM+ and Landsat 8 OLI time series data. We found that training harmonization functions with pixels that have low or average AOT can greatly reduce the difference between near-coincidental Landsat 7 and Landsat 8 observations, and that seasonally trained harmonization models outperform models trained on year-round data. We assessed the effect of ETM+/OLI sensor harmonization on forest type classification with a Random Forest model, and found that seasonally harmonized imagery provided more consistent classification maps than the alternatives. This study illustrates important details related to the creation of harmonized datasets and is a significant step toward creating more consistent Landsat 7 and Landsat 8 data for long-term change detection analysis.

Urban growth in Ecuador's inter-Andean valleys has accelerated the territory's transformation, driven by the expansion of road infrastructure and the occupation of environmentally fragile areas. In this context, the Ruta Viva highway has reconfigured urbanisation patterns in the parishes of Cumbaya and Tumbaco, advancing the urban frontier into agricultural areas and moderate slopes. The objective of this study is to evaluate the dynamics of urban expansion in the parishes of Cumbaya and Tumbaco during the period 2002-2032, using a multitemporal and predictive approach to project future urbanisation scenarios and generate inputs for sustainable territorial planning and land management. The methodology integrated multitemporal analysis of land use and land cover data from MapBiomas (2002-2022), predictive modelling using CA-Markov-MOLUSCE, and urban expansion analysis. The results show a 3% increase in urban coverage during the 2002-2022 period and a projected 12% growth by 2032, concentrated south of the Ruta Viva corridor and within the agricultural mosaic. Simulations show that slopes below 25° are more susceptible to urbanisation, while vegetation cover loss reaches 30% on the slopes of Ilalo Hill. This study provides a robust, replicable tool for anticipating urbanisation scenarios in Andean environments, guiding land management and environmental conservation strategies in regions of high urban pressure.

The Area-Based Approach (ABA) for modelling forest biomass with ALS data assumes perfect co-registration, but operational inventories often have spatiotemporal misalignments. This study isolates and quantifies the independent error contributions from temporal gaps and spatial co-location errors. The analysis uses a unique dataset from the Taiga Plains, Canada, featuring 163 re-measured field plots paired with repeated ALS acquisitions from the same sensor.

To assess temporal effects, we constructed scenarios with varying time-gap distributions. Symmetrical time gaps (SD 1.1 vs 2.5 years) increased RMSE by ~1 percentage point but did not add bias. In contrast, skewed distributions introduced significant systematic biases of 8.0 % (6.8 Mg ha⁻¹).

To assess spatial effects, we linked co-location uncertainty directly to plot-level neighbourhood heterogeneity. This was done by shifting the 20x20m ALS footprint over a 1m lattice and recalculating predictors. The resulting predictor variability (RMS(CV) 12.7%) was propagated through the model, implying a positional sigma of 10-15%. Monte Carlo simulations confirmed this spatial component is the dominant error source, contributing 2–4 percentage points to the ~22% baseline %RMSE.

Our findings show that while balanced temporal gaps are manageable, spatial co-location affected by the local heterogeneity is the most critical factor for robust ABA models.

Tropospheric nitrogen dioxide (NO2) is a crucial trace gas for air quality assessment, yet satellite observations often suffer from spatial gaps (e.g., cloud cover) and temporal limitations. While the geostationary satellite GEMS provides hourly data over Asia, its short historical record and missing data restrict long-term studies. Therefore, a physics-guided adapted normalization network (PhysNorm-Net) is designed to reconstruct a gapless, hourly, and high-resolution (0.05°) tropospheric NO2 dataset over Asia from 2019 to 2024.

The model features an asymmetric U-Net architecture. It handles irregular data gaps using Partial Convolution with a dynamic mask and extracts spatiotemporal representations from meteorological and chemical priors. A novel Physics-Aware Normalization (PhysNorm) module bridges the modality gap by dynamically modulating satellite feature maps using physical backgrounds, ensuring adherence to atmospheric diffusion laws.

Extensive evaluations show that PhysNorm-Net achieves high prediction accuracy (R2 = 0.886). It robustly recovers spatial morphologies and pollution plumes even under extreme missing data scenarios. The generated 2019-2024 dataset accurately captures complex diurnal variations and localized hotspots, providing valuable insights into human activities and pollution policies in Asia.

Accurate estimation of nitrogen oxide (NOx) emissions is essential for understanding their role in atmospheric chemistry and managing air pollution. This study presents a novel approach using Physics-Informed Neural Networks (PINNs) to invert NOx emissions from TROPOspheric Monitoring Instrument (TROPOMI) satellite data. By coupling the physical laws of atmospheric processes, effectively bridging traditional data assimilation techniques with the computational efficiency of deep learning. Unlike purely data-driven models, it directly integrates physical constraints from atmospheric mass continuity equation into the model training process, eliminating the need for inputs or outputs from computationally intensive chemical transport models. Application to the Yangtze River Delta region of China (2018–2023) revealed detailed spatiotemporal NOx emission trends, including the impacts of the COVID-19 pandemic and subsequent recovery. Uncertainty quantification through Monte Carlo dropout provides robust error estimates. This physics-informed approach demonstrates strong potential for efficient NOx emission inversion and offers a versatile foundation for broader quantitative remote sensing applications.

With improvements in the spatial resolution of satellite spectrometers such as TROPOMI, Sentinel-4 and Sentinel-5, more homogeneous cloudy scenes can be resolved at the pixel scale. Therefore, it is worthwhile to use a scattering cloud model in cloud retrieval algorithms. DISAMAR (Determining Instrument Specifications and Analysing Methods for Atmospheric Retrieval) is a computer model developed to simulate the retrieval of atmospheric trace gases, aerosols, clouds, and land-surface properties from passive remote-sensing observations in the 270–2400 nm wavelength range. As a line-by-line radiative transfer model, DISAMAR provides accurate simulations but is computationally expensive. Machine learning techniques can improve the speed of cloud retrieval, because a neural network trained with detailed radiative transfer calculations for scattering clouds can replace the most time-consuming part of the retrieval algorithm.

In this study, we plan to build a cloud retrieval algorithm based on DISAMAR and accelerate it using neural network methods. The algorithm uses TROPOMI observations in the O₂-A band and supports the joint retrieval of cloud optical thickness (COT) and cloud-top pressure (CTP). The neural network models are trained offline using a large, high-resolution spectral data set in the O₂-A band generated by the DISAMAR forward model. All neural networks share the same set of input features but predict different targets, including reflectance and the derivatives of reflectance with respect to cloud pressure and cloud optical thickness. These predictions are then used within an optimal estimation framework to retrieve the cloud parameters.

This study involves remote sensing and artificial intelligence technologies. The study proposed a deep learning-based algorithm to generate the nighttime visible bands for Advanced Himawari Imager geostationary satellite.

Atmospheric aerosols play a vital role in regulating air quality, ecosystems, and climate. Owing to their short atmospheric lifetime, aerosols exhibit strong spatial and temporal variability. Accurate global and regional monitoring of aerosol properties is essential for ecological processes, and radiative forcing. Satellite remote sensing has become a key tool for monitoring aerosol optical thickness (AOT) because of its broad spatial coverage. Traditional physical approaches rely on radiative transfer models (RTMs) to simulate top-of-atmosphere radiances. However, RTMs simplify the real atmosphere, and their accuracy depends strongly on assumed aerosol optical properties and surface reflectance, leading to major uncertainties and inter-algorithm discrepancies. In recent years, data-driven methods have rapidly advanced, driven by developments in machine learning and the increasing availability of collocated satellite and ground-based AOT datasets. The data-driven methods exclusively rely on the data pairs of satellite observations and ground-measured aerosol properties. It learns empirical relationships between satellite observations and measured aerosol properties, and it is more flexible to incorporate more diverse information. However, the AERONET ground stations, commonly used for training, are unevenly distributed and concentrated in urban regions, leaving other surface types such as forests and barren lands underrepresented. Besides, extreme pollution events (e.g., dust storms) are often misclassified as clouds and masked out in AERONET records, introducing bias into training datasets. To mitigate these limitations, this study proposes integrating simulated RTM data into the inversion framework to enhance the robustness and generalization of data-driven AOT retrieval models.

Short annotation如下

Reliable satellite-based estimation of near-surface air pollutants increasingly relies on data-driven models, yet their credibility is hindered by biased generalization assessment and unverified uncertainty estimates. Spatially sparse and unevenly distributed monitoring networks together with strong spatial autocorrelation cause conventional cross-validation approaches to substantially overestimate predictive skill, especially in regions lacking in situ observations. At the same time, although many models produce pixel-level uncertainty estimates, the degree to which these uncertainties reflect true prediction error remains largely unexplored.

This study introduces a controlled, model-agnostic evaluation framework to rigorously examine both spatial generalization and uncertainty reliability in air-quality remote sensing models. A chemical transport model provides a continuous, full-coverage nitrogen dioxide field that serves as an idealized truth. Sampling this field at actual monitoring locations reproduces real observational sparsity while preserving an unbiased reference for domain-wide evaluation. Multiple machine learning models are assessed using sample-based, site-based, and spatially optimized cross-validation to quantify evaluation bias and its dependence on spatial structure.

A dual-path uncertainty strategy is implemented to separately characterize aleatoric and epistemic components, complemented by diagnostic metrics assessing calibration, interval coverage, and sharpness. The framework provides a rigorous pathway for diagnosing reliability in data-driven atmospheric estimation models and supports the development of robust, trustworthy applications in quantitative remote sensing.

The formation and stability of river ice covers in regulated waterways are critical for uninterrupted hydro-electric operations. This study investigates the use of remote sensing, including satellite imagery, aerial surveys, and near-surface observations, to monitor ice cover development in the Beauharnois Canal along the St. Lawrence River. Ice booms are deployed in this canal to promote the rapid formation of a stable ice cover during freezing events, minimizing disruptions to dam operations. Remote sensing data were used to assess the spatial extent and temporal evolution of an ice cover and to calibrate the river ice model RIVICE. The model was applied to simulate ice formation for the 2019-2020 ice season, first for the canal with a series of three ice booms and then rerun under a scenario without booms . Comparative analysis reveals that the presence of ice booms facilitates the development of a relatively thinner and more uniform ice cover. In contrast, the absence of booms leads to thicker ice accumulations and increased risk of ice jamming, which could impact water management and hydroelectric generation operations. These findings demonstrate the value of remote sensing in river ice modelling and potential applications to support operational decision-making for regulated river systems.

When heavy snow or thinning ice allows seawater to intrude into the snow–ice interface, a saline slush layer forms, softening the surface and reducing traction. Because flooding is often invisible, travelers risk becoming stuck in remote areas, creating hazardous conditions. Saline slush also alters the snowpack’s physical and electromagnetic properties. Increased liquid water and salinity affect microwave signal interactions, complicating the estimation of ice properties using remote sensing. Depending on snow depth, temperature, and salinity, slush may refreeze or remain unfrozen, influencing ice thickness and heat transfer. Synthetic Aperture Radar (SAR) is widely used to monitor sea ice under all weather and light conditions. Its signal penetrates the dry snowpack and respond to changes at the snow base, making SAR suitable for detecting seawater flooding. However, SAR observations are sensitive to the target dielectric properties, surface roughness, frequency, incidence angle, and environmental variability. L-band coherence has shown sensitivity to flooding, but its behaviour on snow-covered ice remains poorly understood. This study examines the relationship between seawater flooding and SAR coherence using X- and L-band data collected alongside 2024–2025 field measurements in Qikiqtarjuaq, Nunavut. This research will show how SAR coherence can reveal flooded ice, supporting safer travel in northern communities.

River ice is a key variable in northern regions, with impacts on transportation, infrastructure and flood events. There is increasing emphasis on using remote sensing data to assist operational monitoring. This study investigates the use of synthetic aperture radar (SAR) data for this purpose. The main goal is to provide an open, accessible and scalable approach for accurate semantic segmentation of SAR data into ice and water classes.

Snow Water Equivalent (SWE) refers to the quantity of water contained within the snowpack, which is a critical component of the seasonal water cycle in cold regions, notably Canada. The Gravity Recovery and Climate Experiment (GRACE) mission primarily focuses on quantifying Terrestrial Water Storage Anomalies (TWSA),

which is the sum of anomalies in groundwater, soil moisture, surface water, and snow/ice. Separating the individual components with high precision is a challenging task due to the complex interactions of these parameters and their uncertainties involved. This study proposes an enhanced estimator which is modified based on the

spectral combination theory, to extract the SWE component from GRACE/GRACE-FO (Follow-On) TWS measurements. This estimator uses a hydrological model and its uncertainty to optimally extract the SWE component from the GRACE monthly models in spectral domain. The approach was applied in eight selected basins across

Canada, covering a diverse range of climatic and geographical conditions. Different winter seasons of each basin were considered, including the peak accumulation and ablation phases of the snowpack, from January 2003 to the end of 2022.

During the winters of 2023-2024 and 2024-2025, machine learning (ML) based models were implemented to predict ice thickness at eight sites on the Churchill River and Lake Melville, Labrador for one- and three-day horizons. The forecast ice thicknesses were fed into the Churchill River Flood Forecasting System (CRFFS) operated by the Newfoundland and Labrador (NL) provincial government’s Water Resources Management Division (WRMD). The models were trained on measured ice thickness data from 2017-2023, with the 2024-2025 models additionally trained with data from the 2023-2024 ice season. The 2023-2024 models were deep learning models that used Long Short-Term Memory (LSTM) Recurrent Neural Networks (RNNs), and the 2024-2025 models were ML models that used a simpler gradient boosting regression (GBR) algorithm. The LSTM (2023-2024) models used a running time-series of local meteorological observations as predictor variables to directly forecast ice thickness, and the GBR (2024-2025) models mainly used forecast surface energy balance variables to predict changes in ice thickness. The average performance of the models across the eight sites was comparable between the two ice seasons; however, the 2024-2025 season models improved performance at key sites on the Churchill River that are critical to ice jam flood forecasting. This paper describes the development of the models and their operation and comparative performance over the 2023-2025 ice seasons.

Accurate, reliable, and early-warning forecasts of river ice breakup are essential for flood risk mitigation and public safety, particularly in relation to river transportation and ice road operations. Synthetic Aperture Radar (SAR) satellite imagery has been widely utilized for monitoring river ice conditions due to its sensitivity to surface roughness and dielectric properties. This study advances traditional SAR applications and, to our knowledge, presents the first model that directly incorporates SAR data as input within a machine learning (ML) framework for river ice breakup prediction. The method leverages the correlation between SAR backscatter dynamics and the onset of surface melt. The model was evaluated using leave-one-out cross-validation, achieving an overall accuracy of 92%, an F1-score of 0.91, a Kappa coefficient of 0.84, and a mean absolute error (MAE) of less than 6 days for both the two- and three-week forecasts. The algorithm was also implemented in near-real-time operational settings, demonstrating strong performance with MAE values ranging from zero to four days across different river segments. The approach was further tested on an independent site, where it maintained robust predictive skill. The newly developed method shows strong potential for two- and three-week forecasting of river ice breakup, offering a scalable, cost-effective, and operationally viable tool for management and early warning applications.

Advancing the level of digitalization and automation in concrete manufacturing can substantially contribute to lowering CO2 emissions associated with the concrete production. This work introduces a new methodology for predicting the time-dependent properties of fresh concrete during mixing. For the prediction, a deep learning network is created which uses stereoscopic image sequences of the flowing material together with tabular data as input. Besides mix design parameters and process state data, like energy consumption, moisture and fresh concrete temperature, temporal information is included in the tabular data. The temporal information represents the time interval between image acquisition and the time for which the properties should be predicted. During training, this interval corresponds to the difference between the image acquisition and the time at which reference measurements are taken, allowing the network to implicitly learn the temporal evolution of the material properties, namely the slump flow diameter, yield stress, and plastic viscosity. Incorporating time-dependent prediction enables the forecasting of property changes throughout the mixing process, offering a valuable tool for real-time process control. This capability allows timely adjustments whenever deviations from the desired material behavior are detected. The experimental investigations presented in this paper demonstrate the feasibility of this method under realistic conditions.

Effective integration of geometric and semantic data within Building Information Models (BIM) is essential for the efficient life cycle management of modern facilities. However, maintaining accurate as-is BIM models for existing buildings remains a significant challenge, as manual updates are labour-intensive and full 3D reconstruction is often impractical for incremental changes. In such cases, image-based approaches offer a fast and flexible alternative, but require reliable alignment of 2D imagery with existing BIM geometry. To address this challenge, this study introduces a streamlined pipeline for semantic enrichment that uses a single-image visual localisation approach to directly align 2D imagery with existing BIM geometry. The proposed method integrates transformer-based panoptic segmentation (Mask2Former) with a closed-form Perspective-n-Line solver to estimate 6-degrees-of-freedom (6-DoF) camera poses. The novelty of the proposed approach lies in the explicit use of semantic information as a geometric constraint to guide the selection of 2D–3D correspondences for pose estimation. Semantic labels are utilised to filter line correspondences, ensuring that only stable architectural boundaries (e.g., walls, floors, and ceilings) are used in the registration process. Such semantic filtering stabilises correspondence selection, effectively mitigating pose ambiguity in repetitive indoor layouts or scenes where structural elements are partially obscured by furniture and clutter. Experimental results confirm high accuracy, achieving a median position error of 9.84 cm and an orientation error of 1.05° in complex indoor environments. This robust registration framework provides a reliable foundation for the downstream semantic enrichment and digital twin updates.

Road markings are vital for HD maps and autonomous driving, yet LiDAR-based extraction is difficult due to missing RGB information, severe class imbalance, and thin, elongated geometry under sparse and noisy returns (Ma et al., 2020). We propose LSTNet, which performs local-shape tokenization by grouping points on tangent planes and encoding tokens from relative coordinates, normals, curvature, and intensity contrast. A geometry-aware transformer aggregates these tokens across multiple scales with attention biased by relative position and normal similarity, capturing long and thin structures while preserving edges. Our contributions can be summarized as follows: (1) We present LSTNet, which directly segments road marking from 3D point clouds, avoiding image conversion and preserving geometric fidelity. (2) We introduce a dedicated point-cloud dataset for road marking extraction to enable training and fair evaluation. (3) We design a task-specific and boundary-aware training objective that improves thin road marking recall and robustness under class imbalance.

The concept of Digital Twins (DTs) in Architecture, Engineering and Construction (AEC) domain encompasses a wide range of applications and scales, from single buildings to entire cities, spanning monitoring, simulation, energy management and operational control. Regardless of the specific application, a valid Digital Twin (DT) is a dynamic, data-driven model that stays continuously synchronized with its physical counterpart in both time and state via sensors and the Internet of Things (IoT). It must receive real-world input and provide feedback for analysis or control, ultimately progressing toward a self-operational DT. In the energy domain, an Energy Digital Twin (EDT) must be designed to (i) include sufficient geometric information (ii) support continuous monitoring, (iii) assist scenario-based simulation and (iv) enable operational maintenance and decision support. To achieve these objectives, the EDT’s geometry should be managed through two complementary representations: (i) a watertight solid volumetric model for physics-based simulation and (ii) a boundary representation (B-Rep) model for precise topology, semantics and data exchange. A mapping layer keeps the two representations consistent, preserving identity and topology across states and linking to the graph. Consequently, the EDT should adopt a multi-level architecture defining both geometric and data structures. This work introduces a robust Scan-to-Energy Digital Twins (Scan-to-EDTs) framework that generates multi-level building EDTs by integrating geometric, semantic and simulation layers to enable interoperable energy analyses.

Evaluating uncertainty is critical for reliable use of Mobile Laser Scanning (MLS) point clouds in many high-precision applications such as Scan-to-BIM, deformation analysis, and 3D modeling. However, obtaining the ground truth (GT) for evaluation is often costly and infeasible in many real-world applications. To reduce this long-standing reliance on GT in uncertainty evaluation research, this study presents a learning-based framework for MLS point clouds that integrates optimal neighborhood estimation with geometric feature extraction. Experiments on a real-world dataset show that the proposed framework is feasible and the XGBoost model delivers fully comparable accuracy to Random Forest while achieving substantially higher efficiency (about 3 times faster), providing initial evidence that geometric features can be used to predict point-level uncertainty quantified by the C2C distance. In summary, this study shows that MLS point clouds' uncertainty is learnable, offering a novel learning-based viewpoint towards uncertainty evaluation research.

The automation of Scan-to-BIM remains a major challenge within the Architecture, Engineering, and Construction industry, particularly in the detection and geometric characterisation of architectural openings such as doors and windows. Although recent advances in 3D semantic segmentation have improved the classification of architectural elements, the effect of segmentation accuracy on downstream geometric detection and reconstruction is still under study. This work compares five state-of-the-art deep learning models, PointNeXt, PointMetaBase, Point Transformer V1, Point Transformer V3, and Swin3D, on opening detection in Scan-to-BIM. A unified evaluation framework integrating DBSCAN clustering with axis-aligned bounding box fitting is introduced to generate per-instance geometric representations. The models are assessed using semantic metrics and geometric reliability indicators, including centroid error, dimensional deviation and 3D IoU. Experiments on the S3DIS Area 5 dataset, reveal notable performance differences across models. Swin3D achieved the highest door detection rate of 96.9%, followed by PointMetaBase at 92.9%, PointNeXt at 87.4%, PTV3 at 85.0%, and PTV1 at 81.9%. Window detection proved more challenging, with Swin3D and PTV3 both achieving 75.0%, PTV1 at 71.2%, and PointNeXt and PointMetaBase at 67.3%. Notably, PointMetaBase produced strong geometric accuracy for doors despite lower semantic scores. These results suggest that high segmentation accuracy does not always lead to precise geometric reconstruction. To assess generalisation, the trained models were applied to 11 Matterport3D rooms, confirming that the observed patterns extend across different scanning environments. This study concludes that in Scan-to-BIM workflows, greater emphasis should be placed on geometric reconstruction algorithms than segmentation performance alone.

The digitalization of the architecture, construction and subsurface utility engineering sectors demands efficient, accurate and flexible 3D point surveying methods. Established ones based on Global Navigation Satellite System (GNSS) rovers or total stations suffer from significant limitations, such as requiring open-sky visibility, high costs and complex setups. This paper introduces a novel method for georeferencing 3D points using the PIX4Dcatch mobile application coupled with an external Real-Time Kinematic (RTK) GNSS receiver. The method enables to survey a point of interest by just aiming the smartphone and tapping on the screen during a capture. A lightweight, modified Bundle Adjustment algorithm runs on the device, delivering accurate 3D coordinates in seconds without any post-processing. We evaluated the method by surveying several known cadaster points for hundreds of times across diverse field conditions, achieving a mean planimetry error norm of approximately 3 cm and 97% of errors below 10 cm. Similar statistics are achieved with single-point measurements using an RTK rover. Although not intended to replace millimeter-precision instruments, the accuracy profile of our method is perfectly suited for many applications, such as subsurface utility mapping, which often have decimeter-level regulatory requirements. Given its high efficiency, low cost and ease of use, we believe that our method has the potential to transform as-built documentation workflows in diverse engineering sectors.

Jamming and spoofing of GNSS signals have become common practice in war zones and areas of political tension. The unavailability of reliable GNSS signals has a major impact on mapping services. Airborne laser scanning is one type of aerial survey that depends on GNSS. In this presentation, we propose a concept for airborne laser scanning surveys without using GNSS. We also present the results of an initial feasibility study.

Accurate cadastral maps are essential for effective land administration, supporting tenure security, land management, and socio-

economic planning. Automating cadastral boundary extraction can accelerate mapping in regions with incomplete or absent cadas-

tral information, but deploying pretrained models in data-scarce areas is challenging due to limited reference data and heterogeneous

landscapes. In this study, we investigate cross-region transfer learning for delineating cadastral boundaries using high-resolution

aerial imagery. We employ CadNet, a U-shaped deep learning model with a Swin Transformer backbone pretrained on the Dutch

CadastreVision dataset, and fine-tune it using Polish cadastral reference data selected for landscape similarity to a data-scarce region

in northern Moldova. Evaluation on Moldovan test tiles demonstrates substantial quantitative improvements: recall for visually dis-

cernible boundaries increases from 0.310 to 0.624, total vector-based discrepancy via Normalized Discrepant Area decreases from

7.898 to 7.051. Qualitatively, fine-tuning produces more continuous and coherent boundaries, recovers interior parcel divisions, and

better aligns predicted parcel structures with ground truth, compared to the pretrained model, which generates fragmented and in-

complete boundaries. These results highlight the importance of landscape similarity and reference data quality for transfer learning

and demonstrate a scalable framework for automated cadastral mapping in regions with similar landscape characteristics.

This study presents Siemens star studies in Norway, Finland, and Denmark during 2023-2025. The preliminary results demonstrate a significant and expected difference between GSD and GRD, highlighting that the GRD is a critical parameter when planning and procuring aerial imagery services. GRD relates to the smallest objects that can be reliably mapped. Incorporating GRD into planning ensures that expectations better match the final outcome. The study provides valuable insight into the practical use of Siemens star considering size, frequency, design, material selection, including comparisons between Bayer pattern and pan-sharpened sensors. The Nordic countries have different strategies for evaluating GSD considering prequalification, national calibration fields and field installations on individual projects. This study provides an overall assessment of the different approaches. The project aims to establish common requirements and methodologies for aerial image quality assessment, ultimately contributing to a European-wide GRD based resolution standard

Benefiting of EU funds coming from National Plan for Recovery and Resilience after the covid-19 pandemic, Italian Ministry for the Environment and Energy Security, in coordination with Istituto Geografico Militare and Istituto Nazionale di Geofisica e Vulcanologia, is currently implementing a national project for the acquisition and processing of airborne LiDAR and gravimetric data covering the entire Italian territory. The goal is to overcome the heterogeneity of existing digital terrain and surface models and gravimetric dataset, which suffer from inconsistencies in spatial coverage, temporal epoch, accuracy, and metadata completeness. The project will produce homogeneous, high-resolution Digital Terrain and Surface Models (DTM and DSM) and a new airborne gravimetric database, enabling to estimate a refined gravimetric geoid and significantly improving the Italian geospatial reference infrastructure. All the collected data and realized products will be publicly available. The main features of the project, and a selection of the already available results are hereafter presented.

High-altitude aerial image national mosaics often exhibit visible colour and tone differences caused by atmospheric variability, illumination changes, sensor differences and post-processing workflows. These radiometric inconsistencies negatively influence both visual quality and the comparability of image data across sensors, time and campaigns. This work presents an empirical two-step colour adjustment and radiometric normalisation method for imagery acquired from 2000–3000~m altitude using a large multi-colour ground target designed to provide stable, spatially robust reference statistics. Field reflectance values are measured with a handheld spectrometer and converted to CIELAB coordinates. A global 3D similarity (Helmert) transform aligns measured image colours to ground-truth CIELAB values, followed by local residual chromatic correction using inverse distance weighting. Experiments on aerial datasets demonstrate that the method significantly reduces colour discrepancies at the calibration site.

Frequent cloud cover is a common impediment deterring many countries from employing optical earth observation data for the purposes of national map updates. A decision-level data fusion approach allows the use of satellite imagery in such locations and therefore has potential to assist in this task. In this study we test the use of cloud penetrating Sentinel-1 to enhance the delineation of impervious surfaces from other land cover types, impervious surfaces being a key component of hydro-climatological models in urban and semi-urbanised areas. Using machine learning techniques and leveraging the full Copernicus archive in the Google Earth Engine (GEE) platform, a post-classification change detection approach was developed to assess impervious surface expansion between 2017 and 2023 across the urban centre of Dublin, Ireland. Image classification, conducted using a random forest classifier, achieved overall accuracies of 93% and 91% and kappa coefficients of 0.91 and 0.89 for 2017 and 2023 data, respectively. The addition of multispectral and RADAR indices such as NDVI, NDBI and PRISI was tested and proved generally effective, but showed limitations in areas adjacent to the coast and inland water bodies, with indications of confusion between land cover types. The inclusion of NDWI in data fusion was shown to help differentiate waterbodies from impervious surfaces, particularly highlighting the importance of integrating a water-specific index. NDVI also outperformed other indices in feature importance, though PRISI was shown to helpfully cluster impervious surfaces

Digital twins (DTs) are digital representations of physical entities with data connections synchronising the physical and digital states. While DTs originated in manufacturing and aerospace, they are increasingly applied at geographic scales addressing urban issues. As a result, DTs must utilise geographic information (GI) to represent the built environment, though this is often an implicit aspect. Public sector geographic information (PSGI), typically produced by National Mapping and Cadastral Agencies (NMCA) is a particular type of GI that serves as an authoritative, foundational component to geospatial applications. However, the value of this PSGI as foundation component of DTs is not well understood. Existing GI valuation methodologies do not account for the unique characteristics of foundational PSGI, or its role within DTs , leaving NMCAs unable to justify investment, and adapt their contributions, to emerging DTs. To address this gap, this study applies Jabareen's (2009) conceptual framework analysis methodology to define what value means in the context of PSGI in DTs. The analysis identifies seven value enablers and five value dimensions that characterise PSGI value in DTs and provide the basis for future quantitative valuation methodologies. These concepts are integrated through an urban infrastructure DT example and synthesised through boundary case analysis. The resulting conceptual understanding provides a foundation for NMCAs to systematically articulate and evidence their contributions to DTs.

Digital Twins of Territories (DTTs) are increasingly adopted by municipalities to support ecological transition, crisis resilience, and participatory decision-making. Designing a DTT that fits local needs requires engineers to combine multiple areas of expertise (data discovery, integration, modeling, visualization, and stakeholder interaction) while working with heterogeneous geospatial datasets of varying quality. Nation-wide DTT frameworks aim to assist these efforts, yet they currently lack mechanisms to consolidate the expertise produced during local DTT developments.

This paper introduces dttrecipe, a model designed to capture, structure, and share DTT engineers' feedback and decision-making processes. Building on the prov, wfdesc and wfprov ontologies, and inspired by the OGC Geospatial User Feedback standard, dttrecipe formalizes the description of territorial stakes, data workflows, encountered problems, and the rationale behind design choices. It supports both complete and partial workflow descriptions, encouraging collaboration, reproducibility, and cross-territorial knowledge reuse.

The model is qualitatively evaluated via a case study focused on bicycle-mobility planning and citizen engagement in a rural city. The resulting recipe highlights recurrent categories of DTT engineering challenges, including data discoverability and usability issues, multi-source misalignment, documentation accessibility, and limited local expertise. Explicit documentation of these challenges shows how engineers' often implicit expertise can be converted into reusable knowledge for other territories facing similar constraints.

The work shows that structured documentation of DTT engineering practices can strengthen national DTT frameworks by improving interoperability and enabling efficient knowledge transfer. Future work will address querying mechanisms and evaluate the reuse of shared recipes at scale.

This study takes into account the heterogeneity of regional development stages and the local development context, systematically explores the relationship between localization and sustainable development, and constructs a quantifiable localized SDGs assessment model for China. An empirical analysis of the multi-level SDGs evaluation system was conducted. To address the challenges posed by heterogeneous multi-source data in the assessment process, a composite Key Performance Indicator (KPI) screening model based on Random Forest and Hyperlink-Induced Topic Search (HITS) was proposed, enhancing the scientific rigor and efficiency of localized SDGs monitoring and evaluation.

This paper is mainly based on the experience of Jiaxing City, which has done a good job in the investigation and monitoring work in China, to inspire the investigation and monitoring work in Beijing, and to provide technical support for supporting the investigation and monitoring work in Beijing to achieve the goal of "one inspection, multi-purpose, integration and sharing". Through research, the four databases of new basic surveying and mapping, land change survey, urban land space monitoring, and land space planning are integrated, and the integration of content indicators, survey methods, collection and storage, management and sharing is realized.

Accurate extraction of water bodies from remote sensing imagery is essential for environmental monitoring, water resource management, and hydrological applications. However, the performance of deep learning models for water segmentation depends on the availability of representative datasets that capture diverse environmental and spectral conditions, particularly in tropical and semi-arid regions that remain underrepresented in existing datasets. This study presents the Pernambuco Waterbody Dataset (PWD), a multi-source dataset comprising aerial and satellite remote sensing imagery for water-body segmentation. The dataset covers the state of Pernambuco, Brazil, including tropical and semi-arid environments associated with the Atlantic Forest and Caatinga biomes. The dataset includes high-resolution aerial imagery (0.5 m) from the Pernambuco Tridimensional Program (PE3D) and Sentinel-2A imagery (10 m), with manually annotated water bodies generated by cartographic specialists. The dataset was constructed through data acquisition, preprocessing, manual annotation, mask generation, patch extraction (512 × 512 pixels), and division into training, validation, and test subsets. The first version includes 51,743 aerial patches and 15,321 Sentinel-2A patches. To validate the dataset, U-Net, U-Net++, and DeepLabV3+ architectures with ResNet and EfficientNet backbones were evaluated using Recall, Precision, F1-score, and IoU metrics. The best performance was achieved by U-Net++ (ResNet34) for aerial imagery (IoU 0.946) and U-Net (ResNet34) for Sentinel-2A imagery (IoU 0.871). Overall, the proposed dataset provides a robust benchmark for advancing deep learning-based water body extraction using multi-source remote sensing data.

Multispectral satellite imagery provides an essential source of information for monitoring environmental transformations, yet robust unsupervised change detection remains challenging due to radiometric variability and seasonal dynamics. At the same time, supervised approaches based on Deep Learning are often constrained by the need for computationally expensive accelerated hardware and the limited availability of high-quality annotated datasets. This work introduces a framework based on the Bitemporal Spatial Autocorrelation (BSAC) matrix, that rather than relying on pixel-wise spectral differencing or data-intensive Deep Learning models, it is designed to quantify structural changes by evaluating the symmetry properties of the spatial autocorrelation across multiple spatial lags. Three complementary metrics are derived from the BSAC representation: a binary change/no-change trigger that identifies structural discontinuities, an asymmetry magnitude that measures the intensity of change, and a normalized Symmetry Index obtained via singular value decomposition to characterize the geometric coherence of the correlation structure.

The methodology is applied to Sentinel-2 imagery of Lake Fanaco (Sicily, Italy), which experienced severe desiccation during the 2024 drought. Experiments conducted using NDWI and NDVI confirm the index-agnostic nature of the framework, capturing both

hydrological contraction and vegetation stress. Comparison with an unsupervised K-means segmentation baseline shows strong spatial agreement in identifying the affected areas. Thanks to its unsupervised formulation and near-linear computational complexity, the BSAC framework represents a scalable and interpretable approach for operational change monitoring in Earth observation.

Accurate and rapid burned-area mapping is essential for assessing the ecological impacts of forest fires and supporting post-fire recovery efforts. Traditional pixel-based methods often suffer from limited accuracy due to spectral confusion, topographic effects, and reliance on empirical thresholds. Although deep learning models such as U-Net, DeepLab, and SegFormer improve spatial precision, their operational scalability is constrained by the need for extensive labeled data and regional retraining.

This study introduces a zero-shot burned area mapping approach using the Segment Anything Model (SAM) with Sentinel-2 imagery. SAM, trained on over a billion masks, enables prompt-based segmentation without task-specific training. Composite inputs derived from NBR, NBR2, and NDVI indices were generated and fed into SAM, followed by testing multiple pre-processing, post-processing, and hyperparameter configurations.

Results show that multi-scale settings (crop_n_layers = 2) significantly enhance boundary continuity and geometric accuracy. The method achieved IoU values of 0.89 (Bursa) and 0.87 (Çanakkale), with corresponding F1 scores of 0.94 and 0.92 performances comparable to, and in some cases exceeding, supervised models. Integrating spectral index composites further reduced boundary fragmentation and improved discrimination between burned and unburned surfaces.

Overall, the proposed framework eliminates dependence on manual labeling, offering a fast, scalable, and cost-effective solution adaptable to diverse ecosystems and sensor conditions. The study demonstrates one of the first systematic applications of SAM for burned-area detection, highlighting its strong potential for zero-shot environmental monitoring and rapid post-fire assessment.

In many regions of the world, especially those affected by armed conflicts, urbanization, or intensive environmental transformations, a high dynamic of land cover and land use changes is observed. Reliable monitoring of these processes requires the application of classification methods that ensure both high thematic accuracy and temporal consistency. This paper presents a multitemporal classification methodology based on Sentinel-2 optical data and machine learning models.

The research was conducted for the city of Sievierodonetsk (Luhansk Oblast, Ukraine) – an area that suffered significant destruction in 2022 as a result of military operations. The aim of the analysis was to identify land use changes in the years 2021-2025 using three classifiers: k-Nearest Neighbors (kNN), Random Forest (RF), and Gradient Boosting Classifier (GBC), combined into an ensemble system based on dynamic confidence weighting.

Quality assessment using the recall metric showed that the fusion method outperformed individual classifiers, achieving average values of 0.87-0.96, while classical models obtained 0.81-0.89. The largest changes (39%) occurred in the years 2022-2023, coinciding with the period of greatest military activity. The proposed method achieved the highest classification quality indices (F1 = 0.93, Acc = 0.98 for 2021), surpassing global products and models based on AlphaEarth. In subsequent years, high stability was maintained (F1 ≥ 0.88), confirming the effectiveness and robustness of the approach under various environmental conditions

This study examines the landscape dynamics in the region surrounding Comandante Ferraz Antarctic Station, Keller Peninsula, King George Island, focusing on the quantification of land cover changes over 23 years. Emphasis is placed on the integration of a multitemporal Landsat time series (2001–2024) within a standardized spatio-temporal data cube framework, coupled with a Random Forest (RF) classification approach. This methodology enables consistent pixel-wise trajectory analysis across seven distinct epochs. The RF models achieved robust performance, with F1-scores for dominant classes like water and soil typically exceeding 0.90, although seasonal snow and ice showed greater spectral ambiguity in transitional months. Quantitative results from the transition matrices reveal a significant landscape reconfiguration: while ice (85.3%) and soil (81.2%) showed high persistence, a prominent trend of deglaciation was identified, characterized by the transition of ice and snow into exposed soil and the emergence of pioneer vegetation communities detected from 2014 onwards. The study demonstrates that the integration of machine learning and data cubes provides a powerful tool for monitoring environmental shifts in high-latitude maritime Antarctica, supporting long-term ecological assessments and climate impact modeling.

Severe convective rainfall, triggered by multi-scale atmospheric interactions, poses a critical forecasting challenge in coastal cities like Shanghai, where monsoon, topography, and sea-land breeze amplify extremes. Conventional methods, constrained by scale separation and model biases, struggle to predict convection. This study develops the Synergistic Framework for Convective Rainfall Forecasting (SSF-CRF) by integrating three modules: (1) Adaptive S/X-band radar remote sensing, dynamically capturing mesoscale convective structures; (2) Gated Vertical Information Propagation (GVIP) network, machine learning on vertical energy propagation to capture convection; (3) Precipitation Ordinal Distribution Autoencoder (PODA), correcting numerical weather prediction (NWP) biases with ordinal precipitation classification. Verification against Radar data and European Centre for Medium-Range Weather Forecasts (ECMWF) model indicates that SSF-CRF improves heavy rainfall (≥50 mm/h) Critical Success Index (CSI) by 33% versus operational forecasts. It offers a potential solution for convective forecasting in climate-vulnerable coastal regions, advancing remote sensing-driven atmospheric applications.

Monitoring atmospheric water vapour is essential for weather forecasting and climate studies. GNSS networks can retrieve Precipitable Water Vapour (PWV) continuously at each station location, but the accuracy depends on the quality of the satellite orbit and clock corrections used in the processing. This study evaluates PWV retrieval from 478 stations of the French Centipede low-cost GNSS network using four levels of correction products with decreasing latency: GFZ Final ($\sim$2 weeks), Rapid ($\sim$1 day), Ultra-rapid (3--9 hours), and broadcast ephemerides (real-time). Validation against ERA5 reanalysis shows that the Final and Rapid products achieve similar performance (RMSE $\approx$ 2~mm, $r^2$ = 0.84), confirming that near-real-time processing introduces no significant accuracy loss. Ultra-rapid products remain usable (RMSE = 3.4~mm), while broadcast ephemerides show larger errors (RMSE = 5.8~mm) but still capture the spatial moisture pattern. In addition, a real-time experiment using the freely available Galileo High Accuracy Service (HAS) demonstrates that stable tropospheric estimates (ZTD $\pm$ 1.4~mm, PWV $\pm$ 0.2~mm) can be obtained in real time, even before the positioning solution has fully converged. These results suggest that combining the spatial density of low-cost networks with real-time HAS corrections could enable high-resolution PWV monitoring that is not achievable with existing systems.

Fengyun-3G, launched in 2023, carries Ku- Ka dual-frequency precipitation measurement radar (PMR) providing new opportunities for monitoring the fine three-dimensional structure of typhoon precipitation over the ocean. This study first validate the FY-3GPMR data by using the ground-based data, then utilizes PMR to analyze the precipitation during the rapid intensification phase of Super Typhoon Yagi in the year of 2024. The analysis reveals the horizontal and vertical distribution characteristics of precipitation during Yagi's RI phase based on the FY-3G PMR data, and discusses the associated dynamical-microphysical coupling mechanism. Overall, FY-3G PMR offers critical insights for understanding cloud and precipitation process involved in the RI.

Atmospheric water vapor plays a fundamental role in regional climate regulation and precipitation formation, yet its vertical structure and spatiotemporal evolution over mainland China remain insufficiently understood due to complex terrain and diverse climatic conditions. Using Fengyun-4A layered precipitable water (LPW) products from 2020 to 2023, this study provides a comprehensive assessment of the vertical distribution, spatiotemporal variability, and key environmental drivers of water vapor across China. Results show pronounced spatial gradients and seasonal contrasts: total precipitable water (TPW) exhibits a slight overall decline, primarily driven by reductions in low layer; spatially, TPW is highest in the southeast and lowest in the northwest; seasonally, water vapor peaks in summer and reaches its minimum in winter, with spring and autumn representing monsoon-transition phases. Vertically, approximately 75% of atmospheric water vapor is concentrated within the lowest 4 km, with the middle layer contributing most to regional differences, while high layer remains relatively uniform and minimally influenced by terrain. Environmental correlations indicate that TPW is positively associated with 2m temperature, relative humidity, surface pressure, total cloud cover, and precipitation, but negatively associated with DEM and evaporation. Layer-dependent responses indicate that the lower layer is strongly influenced by surface processes, the middle layer by both surface moisture transport and large-scale circulation, and the high layer primarily by thermodynamic structure and synoptic background. These findings, derived from high-resolution satellite observations, enhance understanding of atmospheric water vapor stratification and its controlling mechanisms, providing essential support for water vapor transport diagnosis, precipitation evolution, and operational forecasting improvement.

In recent decades, global climate change has triggered a rise in extreme environmental phenomena, including prolonged droughts, intensified precipitation events, and shifts in tidal patterns. This study focuses on the application of the observations from Global Navigation Satellite System (GNSS) signals for monitoring and classifying climatic conditions, with particular emphasis on drought.

Using daily vertical displacement data from a GNSS station in California (2005–2023), we developed a robust analysis framework. It includes data cleaning (removing outliers, filling gaps, detecting offsets, and modeling noise), trend and seasonal pattern extraction through Singular Spectrum Analysis (SSA), feature generation (like amplitude, energy, and dominant frequency), labeling

based on the Standardized Precipitation-Evapotranspiration Index (SPEI), and classification using a hybrid CNN-Transformer model. The results demonstrate the model’s capability to accurately detect drought periods (SPEI > -1) characterized by diminished amplitudes in seasonal components and heightened noisy fluctuations, as well as wet periods (SPEI < 1) marked by elevated energy in semi-annual signals. The model was evaluated with an overall accuracy of 83.3 percent, an F1-score of 0.90 for the drought class, and successful application to future data (2024–2029). This approach, independent of traditional meteorological data, underscores the potential of GNSS as a geodetic tool for environmental monitoring, albeit with limitations such as reliance on single stations and the need for supplementary datasets. The methodology holds promise for enhancing early warning systems and climate models.

Carbon dioxide (CO₂) and methane emissions (CH₄) are primary greenhouse gases whose rising atmospheric levels intensify global climate change. Wetlands, despite covering only 5–8% of Earth’s land area, contribute nearly 30% of global methane emission while storing up to 30% of global soil organic carbon. This makes wetlands both sinks and sources of greenhouse gases, though their seasonal CO₂ and CH₄ dynamics in the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) remain poorly understood.

Ground-based instruments offer high accuracy but limited spatial coverage, whereas satellite missions, such as Sentinel-5P/TROPOMI for XCH₄ and OCO-2 for XCO₂, enable wide-area monitoring. This study investigates the seasonal dynamics of CH₄ and CO₂ across different wetland ecosystems in the GBA using satellite observations and ERA5-Land climate variables. Seasonal means were computed in Google Earth Engine for Winter, Spring, Summer, and Autumn from 2019 to 2025.

Results show a consistent rise in atmospheric CH₄ from 1856 ppb (2019) to 1939 ppb (2025), with the highest levels in Autumn and Winter. CO₂ increased from 404 ppm to 424 ppm, peaking in Winter and Spring. Non-wetland regions and mangroves emerged as the primary contributors to greenhouse gas accumulation, while salt marshes and other wetlands showed lower values.

Pearson correlation analysis indicated strong influence of temperature, dew point, and precipitation on CO₂, while CH₄ showed variable sensitivity to rainfall and wind. Findings emphasize the impact of land-cover type and climate in shaping seasonal greenhouse gas dynamics, supporting SDG 13 and SDG 15, and necessitating hyperspectral data integration for climate policies.

Urban air quality remains a critical concern, as NO₂ emissions from transport and industrial activities frequently exceed healthy limits in major cities. Urban vegetation can help reduce pollution by enhancing natural filtration and cooling, while higher land surface temperatures (LST) tend to intensify pollutant accumulation. Using satellite-based remote sensing, this study investigates how vegetation health (NDVI) and surface temperature influence NO₂ levels in two distinct urban environments: Blackburn/Arlington Road in England and District No. 3 in Tehran, Iran, across pre-, during-, and post-COVID-19 lockdown periods. Both cities experienced notable environmental improvements in 2020: NDVI increased from approximately 0.45–0.48 to around 0.54–0.61, while NO₂ dropped significantly from about 0.46–0.50 to roughly 0.13–0.35. LST also declined from pre-lockdown values near 0.46–0.48 to as low as 0.12–0.38. During the lockdown, vegetation levels showed a clear negative relationship with NO₂ concentrations, and pollution trends displayed a strong positive association with higher temperatures, highlighting the linked benefits of greener and cooler environments. However, as human activities resumed after 2021, these relationships became inconsistent or weakened, with occasional shifts in direction depending on seasonal conditions and external drivers such as traffic recovery and industrial intensity. Overall, the results reinforce that increasing vegetation coverage and mitigating urban heating can meaningfully reduce NO₂ levels. By revealing how urban form, vegetation dynamics, and thermal conditions collectively shape pollution patterns, this research provides insights for city planners, environmental managers, and public health authorities working to design more sustainable and healthier urban environments.

Bathymetry information around islands, reefs, and shallow-water regions is critical for both navigation safety and environmental management. However, these areas often feature diverse substrate types and strong spatial heterogeneity, which makes it challenging to accurately retrieve fine-scale bathymetry from traditional medium-resolution satellite imagery. High-spatial-resolution (HSR) sensors, such as PlanetScope SuperDove (~ 3.7 m spatial resolution), offer the potential to capture more detailed spatial features, yet their relatively low signal-to-noise ratio (SNR) can lead to noisy retrievals, particularly over low-reflectance waters. To mitigate this issue, incorporating the spatial context of neighboring pixels while jointly utilizing the spectral information offered by low- and high-resolution sensors can enhance the stability and accuracy of HSR-based bathymetry retrievals. In this study, a UNet++ neural network with the spatial and channel squeeze & excitation (scSE) attention mechanisms (BathyUNet++) was employed to retrieve bathymetry from SuperDove imagery. To satisfy the patch-based input requirement of UNet++, the model was fully trained using two sources of data: clear-sky SuperDove image patches paired with Landsat-8-derived bathymetry and a limited set of ALB data. Validation results demonstrated that the model accurately retrieved bathymetry in regions independent of the training set.The proposed model and framework can be readily adapted to other HSR sensors, offering a promising approach for global HSR shallow-water bathymetry retrieval using multi-source satellite observations.

Despite its potential in forestry, agriculture, environmental monitoring, safety surveillance, and defence, real-time hyperspectral imaging (HSI) remains challenging in practice because of the high dimensionality of the data and limited onboard computational resources.

This work introduces a distributed HSI classification framework that integrates federated learning, a Self-learning Minimal Learning Machine classifier (SL-MLM), adaptive Kalman filter-based model fusion, and lightweight MQTT-based communication on Raspberry Pi edge devices and a laptop serving as the base station. Acting as local nodes, Raspberry Pis process HSI data row by row, update their models recursively, and only exchange compact model parameters and classification results with the base station. HSI data in its raw form remains local.

The findings suggest that the proposed local learning workflow can be implemented on Raspberry Pi devices, and Kalman-based fusion improves stability and consistency in comparison to individual local models. The method is feasible in scenarios where the number of labelled data points is restricted, as the SL-MLM classifier can be initialized with a mere handful of class-specific reference points. The research demonstrates a feasible, low-cost approach to distributed

embedded HSI classification and sensing.

Monitoring inland Water Quality Parameters (WQPs) is essential for managing freshwater ecosystems and assessing anthropogenic impacts (Mishra et al., 2017). Satellite remote sensing provides a cost-effective and large-scale approach for monitoring inland WQPs. However, most existing satellite sensors have limited spectral resolution, restricting their ability to capture subtle optical variations expressed by inland WQPs, and/or insufficient spatial resolution to yield valid water-only pixels in narrow rivers or nearshore zones (Ansari et al., 2025).

Recent advances in hyperspectral satellite technology have created new opportunities for inland WQP monitoring. The Wyvern Dragonette-001, launched in April 2023, provides hyperspectral imagery with a spatial resolution of 5.3 m and 23 spectral bands within the visible to near-infrared range (500–800 nm) (Ansari et al., 2025; Wyvern Dragonette, 2023). Given its novelty, the potential of such imagery for assessing WQPs in inland water remains largely unexplored. A recent review (Ansari et al., 2025) evaluating the sensor’s spectral resolution and signal-to-noise ratio for retrieving inland WQPs indicated that Dragonette-001 is suitable for estimating non-algal particles (NAP) and shows potential for chlorophyll-a mapping, although it is likely unsuitable for retrieving Colored Dissolved Organic Matter (CDOM).

This study reports on a practical test that assessed the feasibility of using Wyvern Dragonette-001 imagery to retrieve turbidity, Suspended Sediments (SS), and Dissolved Organic Carbon (DOC) in a portion of the St. Lawrence River, Québec, Canada.

Hyperspectral imaging produces vast data volumes that often exceed storage and transmission capacities on airborne and satellite platforms. This study systematically investigates the effects of lossy hyperspectral data compression on the scientific usability of the resulting data products. Using UAV-based HySpex acquisitions from the HyperThun’22 campaign, several state-of-the-art learning-based compression models were evaluated, including spectral, spatial, and spatio-spectral architectures. The analysis quantifies how compression-induced reconstruction errors propagate through the full pre-processing workflow, from raw digital numbers through radiometric calibration, geometric correction, and atmospheric correction to the final surface reflectance domain.

Results show that spectral models such as the Adaptive 1D Convolutional Autoencoder (A1D-CAE) achieve the highest fidelity, maintaining sub-degree spectral deviations and near-perfect structural similarity. In contrast, purely spatial or 3D convolutional models exhibit severe distortions that persist across all pre-processing levels. The findings demonstrate that lossy compression can be applied at the raw stage without compromising the integrity of reflectance products, provided that spectral correlations are explicitly modeled.

This work highlights the importance of selecting compression architectures consistent with sensor characteristics and pre-processing workflows and provides a quantitative foundation for future operational implementations of onboard hyperspectral compression in Earth observation missions.

Efficient nitrogen (N) management remains a major challenge for sustainable potato production, particularly on coarse-textured soils prone to nutrient leaching. This study investigates the use of Visible–Near Infrared to Short-Wave Infrared (VNIR–SWIR, 350–2500 nm) hyperspectral spectroscopy for non-destructive, in-season estimation of petiole nitrate nitrogen (PNN) under both field and laboratory conditions. Spectral data were collected using an ASD FieldSpec Pro spectroradiometer and processed through Savitzky–Golay smoothing, Standard Normal Variate normalization, and first-derivative transformation. Variable Importance in Projection (VIP) analysis was employed to identify N-sensitive wavelengths, and three predictive approaches—One-Dimensional Convolutional Neural Network (1D-CNN), Support Vector Regression (SVR), and Partial Least Squares Regression (PLSR)—were compared for their predictive accuracy. Calibration transfer using Piecewise Direct Standardization (PDS) was applied to harmonize field spectra with laboratory measurements. Results showed that the 1D-CNN achieved the highest predictive performance (R² = 0.90, RMSE = 0.22%), outperforming SVR and PLSR. PDS improved field-based predictions by reducing spectral discrepancies caused by illumination and canopy variability. The findings highlight the potential of hyperspectral spectroscopy combined with deep learning and calibration transfer techniques to provide accurate and scalable diagnostics of plant nitrogen status. This research supports the integration of proximal sensing and data-driven models for precision nutrient management in potato systems and broader agricultural applications.

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. The proposed framework integrates strip convlution layers with multi-size kernels in a dense connection to enhance stripe structural information expression 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.

This paper aims to evaluate the performance of ISODATA clustering for tree species classification using ultra-high-resolution multispectral data collected with Unmanned Aerial Vehicle. The study focuses on two sites in Żednia forest district near the city of Bialystok, northeastern Poland. The input data consist of 10-band multispectral orthomosaics with a resolution of 10 cm, acquired from an UAV platform equipped with a MicaSense RedEdge-MX dual camera and image-based Canopy Height Model. The classifications were conducted at two levels of forest detail: forest types, including two classes (broadleaf and conifer), and tree species, comprising four classes in Study Area 1 and ten species in Study Area 2. Multiple classifications were generated, testing different input parameters such as the number of clusters and various combinations of input data. For the first level of classification (forest type), overall accuracies range from 84,09% to 97,57% in Study Area 1 and from 82,31% to 92,74% in Study Area 2. At the second level of classification (tree species), overall accuracies vary from 70.73% to 91.77% in Study Area 1 and from 36,51% to 72,33% in Study Area 2. Overall, ISODATA demonstrates robust performance in classifying forest types in both study areas. However, performance in classifying tree species varies across different classes, with relatively high accuracies observed for certain species such as spruce, pine, oak, larch, and birch. The results underscore the potential of multispectral UAV data and unsupervised classification methods for accurately classifying tree species.

This work presents a multi-camera mobile mapping solution designed to deliver accurate and efficient 3D reconstructions across a wide variety of challenging environments, ranging from confined indoor spaces to complex urban outdoor settings. Traditional photogrammetric and terrestrial laser scanning approaches, while capable of high accuracy, often suffer from limitations related to acquisition speed, logistical complexity, and significant post-processing effort—especially in large, occluded, or hard-to-access sites. Mobile Mapping Systems (MMS) based on Visual SLAM (V-SLAM) offer a compelling alternative, thanks to their ability to acquire high-frequency imagery in continuous motion and estimate sensor trajectories in real-time. However, MMS outputs frequently face issues such as reduced geometric accuracy, scale drift in monocular sequences, and the need for extensive optimisation to reach survey-grade results.

To address these limitations, the study extends an existing multi-camera V-SLAM pipeline by tightly integrating monocular estimates with multi-stereo trajectories within the ATOM-ANT3D fisheye multi-camera system. A novel monocular scale-recovery strategy is introduced, based on path-length ratios derived from concurrently recorded stereo tracks. This metrized monocular trajectory is then fused with stereo estimates through a robust pose graph optimisation, followed by a multi-view, feature-based refinement leveraging pre-calibrated camera geometry.

The proposed method is evaluated across four real-world scenarios—spiral tower staircases, dark underground caves, narrow urban corridors, and constrained industrial pipelines. Accuracy is assessed against reference 3D point clouds, while efficiency is compared to a standard multi-view stereo photogrammetric pipeline. Results demonstrate that the integrated approach significantly improves reconstruction consistency, robustness, and end-to-end throughput.

Traditional image matching methods rely heavily on gradient or intensity information. However, the severe nonlinear radiometric distortion (NRD) between infrared and visible images hinders the extraction of repeatable feature points, leading to poor matching performance. To address this, we propose Shape2Match, a novel framework that replaces point features with more consistent, modality-invariant shape features. Specifically, the method utilizes EfficientSAM to extract shape contours and employs elliptic fourier descriptors (EFD) to parameterize and normalize them, creating shape descriptor that is invariant to translation, rotation, and scale. Shape2Match adopts a coarse-to-fine hierarchical strategy: it first performs robust global shape matching using a weighted EFD distance, followed by precise keypoint matching—using Shape Context—within the coarsely aligned shape pairs. We validated Shape2Match on 153 image pairs from 6 datasets, comparing it against methods like SIFT, RIFT, and MS-HLMO. Experimental results demonstrate that Shape2Match achieves a 100\% success rate (SR) across all datasets and significantly outperforms other methods in the number of correct matches (NCM), proving its effectiveness and robustness against NRD, rotation, and scale variations.

Historical film-based images, acquired by aerial sensors since the 1930s and by satellite platforms since the 1960s, provide a unique opportunity to document changes in the Earth surface over the 20th century. Yet, they present significant and specific challenges, including complex distortion in the scanned image pixel grid and poorly known camera exterior and interior orientation. In recent years, semi- or fully-automated approaches, based on photogrammetric and computer vision methods, have emerged, but the performance and limitations of these methods have yet to be directly compared.

The objectives of the Historical Images for Surface Topography Reconstruction Intercomparison eXperiment (Historix) project are to compare existing methods for processing stereoscopic historical images and harmonize processing tools. Here we present the study site and dataset selected for this comparison, the design of the intercomparison and evaluation metrics, as well as preliminary results. Full evaluation will be presented at the conference.

The location accuracy of an image acquired with a space borne camera relies on the knowledge of the orbit of the spacecraft and the orientation of the camera. The a posteriori estimation of a satellite orbit has been a well mastered technique for a long time. Sub-meter accuracy is achievable with a reasonable effort. The geolocation, with a similar accuracy, of the line of sight of an optical instrument flying at 500km or above is a much more challenging task..

On the other hand, the geolocation of a synthetic aperture radar (SAR) image depends only on the orbit of the spacecraft. It is, therefore, easy to acquire space borne SAR images with a sub-metric native geolocation. The Airbus SAR constellation (TerraSAR-X, TanDEM-X and PAZ) provides, on a commercial basis, images with a (better than) 0.2m geolocation accuracy.

The ability to find, through image matching, homologous points in SAR and optical images would transfer the native accuracy of SAR to optical observations, using classical photogrammetric bundle adjustment.

This paper describes an operational way to perform this SAR/Optic images matching and a validation of the absolute location accuracy achieved.

High-precision mapping of planetary surfaces, such as Mars, relies on matched control points derived from existing georeferenced data, as ground control points (GCPs) cannot be obtained through field measurement. However, the handcrafted image matchers like SIFT limit the robustness of this approach, particularly on texture-scarce and self-similar Martian terrain. While deep learning-based matchers offer a new paradigm, their performance gain for bundle adjustment remains inadequately quantified. This paper systematically evaluates four matchers (hand-crafted SIFT and deep learning-based DOG+HardNet+LightGlue, DISK+LightGlue, and LoFTR), assessing their impact on georeferencing tasks using Tianwen-1 high-resolution imagery. Deep learning methods, such as LoFTR, generate more correspondence points with a more uniform spatial distribution, halving the outlier rate and improving bundle adjustment accuracy by 10%. Our study provides a benchmark for planetary mapping and shows that powerful, learning-based image matchers are pivotal for next-generation automated mapping systems.

This work investigates the solutions developed at Natural Resources Canada to produce analysis-ready mapping products from Canada's national air photo library including two main workflows: 1) The photogrammetric processing of historical photos with an emphasis on the more challenging automated georeferencing component; 2) Enhancing interpretability through generative artificial intelligence models for super-resolution and deep colorization.

The accuracy of the aerial triangulation (AT) performed in the map projection for a GNSS-INS-supported image block consisting of 4342 vertical images, GSD 20 cm, with 22 main strips and 5 cross strips is investigated. Using 169 check points the obtained results are compared with the accuracy achieved by running the AT in an undistorted tangential system. It turns out, that in both systems the same accuracies can be achieved, with RMSE in (X, Y, Z) of (7, 10, 11) cm, if Earth curvature and scale distortion are correctly modelled in the map projection. If the scale distortion is not considered, then the RMSE in Z increases by 100% to 300% (depending on the height distribution of the GCPs). In AT software packages, that do not consider the scale distortion, a partial compensation is possible by either adapting the height of the projection centres or the principal distance leading to RMSE of around (10, 11, 15) cm.

Forest structural complexity (FSC) reflects the three-dimensional arrangement and distribution of forest elements and serves as a key ecological indicator of biodiversity and forest productivity. Decades of overharvesting have transformed many temperate rainforests into young, homogeneous stands. Given the central role of FSC in ecosystem functioning, silvicultural strategies increasingly aim to retain or enhance structural complexity and mitigate the ecological impacts of timber harvesting. Monitoring structural development across silvicultural treatments, environmental gradients, and disturbance regimes is therefore essential. However, large-scale assessments of FSC remain limited. In this study, we evaluate the scalability of canopy entropy (CE), a LiDAR-derived FSC index, using deep learning applied to multisensor radar and optical imagery. We trained a U-Net convolutional neural network using airborne LiDAR-derived CE as the reference variable and Sentinel-1 and Sentinel-2 data as wall-to-wall predictors. The model demonstrated strong overall predictive performance (R² = 0.80, MAE = 0.09, bias = 0.02, nRMSE = 12.2%). However, the horizontal complexity component of CE (CExy) exhibited substantially lower accuracy. Although aspects of horizontal complexity may be indirectly inferred from vertical structure or canopy cover, CE should be interpreted with caution. Future work should focus on improving the representation of horizontal complexity. Despite these limitations, the resulting CE map provides a foundation for evaluating silvicultural practices and identifying structurally complex forests with high conservation value.

Wildfire severity, often correlated with biomass loss, has increased since the 1980s, driving greater biomass depletion across landscapes. Canada's 2023 wildfire season burned over 15 million hectares and released 647 TgC of carbon, surpassing most nations' annual emissions. This trend underscores the need for scalable aboveground biomass (AGB) monitoring for greenhouse gas estimation. While LiDAR has improved AGB estimation, airborne systems remain costly with limited spatial coverage. Researchers have addressed this by scaling LiDAR-derived estimates to satellite imagery for broader monitoring. However, current scaling paradigms are developed predominantly for closed-canopy forests, with limited evaluation in open-canopy ecosystems like peatlands, despite their high fire severity and disproportionate carbon contributions when burned. Peatlands pose unique challenges: low and spatially heterogeneous AGB, open canopies that allow soil and water to obfuscate satellite signals, and non-linear structural-biomass relationships in sparse vegetation.

This systematic review and meta-analysis examines the accuracy of scaling LiDAR-derived AGB estimates to optical and radar satellite imagery across peatlands and structurally analogous ecosystems, including tropical savannas, floodplain forests, mangroves, and arctic shrublands. We searched Scopus and Google Scholar using a four-block query, yielding 271 peer-reviewed studies. Using a random-effects model, R² values were transformed to Fisher's Z scores, and heterogeneity was quantified using the I² statistic.

Preliminary analysis revealed no significant difference between modelling approaches and target ecosystem. Heterogeneity was minimal, indicating model type and ecosystem type exert limited influence on accuracy outcomes. Full dataset analysis is ongoing.

Habitat suitability models are central to conservation

planning, species management, and landscape-level decision

support. Continuous Landscape Inventory (CLI)

datasets provide stand-level forest attributes (species mix,

height, basal area, crown closure, age, disturbance history)

that are rarely used at scale together with satellitederived

biophysical indicators for operational habitat mapping.

This work proposes a replicable workflow that fuses

provincial CLI with multisensor satellite data (Sentinel-

2 MSI, Landsat series, and SAR-derived structure proxies)

and environmental layers (elevation, distance-to-water,

road density) to produce fine-scale habitat suitability surfaces

across New Brunswick, Canada.

Wildfires significantly impact forest ecosystems by disrupting carbon cycles, with effects varying based on fire intensity and forest bio-physical characteristics such as vegetation types, structures, topography, and climate. These factors collectively influence fire spread, biomass reduction, and post-fire vegetation regrowth, making it crucial to accurately quantify wildfire impacts on forest carbon dynamics for understanding terrestrial-atmosphere interactions and global climate implications.

This study uses wildfires in California's mountainous forests as a case study, employing two aboveground biomass (AGB) datasets—one derived from remote sensing data and the other from process-based ecological models—to assess wildfire impacts on forest carbon stocks. Remote sensing-based indices, while effective in detecting spectral changes, often fall short in quantifying biophysical alterations, particularly carbon dynamics. Conversely, process-based models adhere to ecological principles but may not fully capture fire-induced carbon changes. Our analysis reveals significant variations in post-fire disturbance and recovery patterns based on fire severity, elevation, and forest type. The remote sensing dataset showed faster initial recovery, likely due to herbaceous vegetation greening, while the ecological model dataset indicated slower, more stable recovery, reflecting delayed tree regeneration. These findings underscore the necessity of integrating multi-source datasets to accurately capture post-fire carbon dynamics.

Wildfire Early Warning Systems (EWS) are increasingly essential as climate-driven extreme fire events grow in frequency

and severity. Yet their maturity and operational robustness vary widely across countries due to differences in

resources, data infrastructure, and institutional capacity. This study conducts a systematic global assessment of wildfire

EWS across high-, middle-, and low-income nations, evaluating how multisensor Earth Observation (EO) data and predictive

intelligence are integrated into functional early warning and decision-support systems. A transparent benchmarking

framework is introduced with two core pillars: (i) multisensor geospatial monitoring—assessing temporal resolution,

spectral sensitivity, spatial detail, and GEO–LEO fusion; and (ii) hotspot intelligence and predictive modeling—evaluating

model class, forecast range, validation practices, and real-time operational performance. These pillars are complemented

by an impact-readiness layer aligned with the Sendai Framework, linking hazard detection to exposure, vulnerability, and

alert dissemination.

Results show strong stratification by income. High-income countries achieve near–real-time hotspot detection, GEO–LEO

data fusion, and validated multi-day behaviour forecasts. Middle-income nations display transitional but uneven progress,

while low-income countries rely almost exclusively on global detection platforms, highlighting institutional, not technological,

bottlenecks. Canada’s EWS landscape is evaluated, highlighting gaps in accessibility, standardization, and

timeliness of EO-derived intelligence. Opportunities for strengthening Canada’s system include adoption of emerging EO

technologies, improved fuel characterization, next-generation hybrid physics–ML/QML behaviour modeling, integrated

national decision-support platforms, and enhanced FireSmart community interfaces.

Overall, the study provides a scalable global framework for comparing national wildfire EWS maturity, identifying

investment priorities, and guiding future improvements.

This project explores how drones and deep learning can help monitor the recovery of former aggregate and mining sites. Traditional methods for assessing land restoration such as field surveys and satellite imagery are often time-consuming, expensive, and limited in detail. Using high-resolution drone imagery and a compact deep learning model, this study offers a faster and more flexible way to track how vegetation and land cover change over time.

The approach classifies ground surfaces into three simple categories: healthy vegetation, stressed vegetation, and bare soil or rock - providing clear indicators of how well a site is recovering after extraction/rehabilitation. Tested at two rehabilitated sites in southern Ontario, the model showed strong and consistent results across different months of the growing season, even using only standard colour drone imagery.

This work highlights how drone-based monitoring can make ecological restoration assessment more efficient, objective, and repeatable. Once trained, the model can quickly analyze new imagery without the need for extensive fieldwork, allowing land managers and regulators to identify problem areas and track recovery in near real time. Ultimately, this research points toward a future where rapid, data-driven drone assessments play a role in supporting sustainable land rehabilitation and environmental stewardship.

Deep, persistent drought in 2023 in the Canadian Boreal Plains was associated with wildfires that persisted underground and re-emerged the following spring, a process known as "overwintering" and sometimes called "zombie fires". We analyzed pre-fire Sentinel-2 multispectral imagery of paired 2023-2024 fires to extract any spectral anomalies, with the goal of characterizing conditions conducive to wildfire overwintering. We assessed several spectral indices, including NDVI, GNDVI, EVI, NDMI, TCW, and others relative to a 2016-2022 baseline using the npphen R package.

We found that sites of overwintering fires exhibited moisture anomalies in the spring of 2024, indicating drought conditions that were conducive to the reemergence of overwintering fires. We show how these anomalies were co-located with early season wildfire with an apparent absence of ignition events. Furthermore, we show how in 2024, 25 overwintering wildfires burned 22.8% of the total area burned, while comprising only 1.3% of the total fire count.

The EPSRC Centre for Doctoral Training (CDT) in Geospatial Systems was established in 2019 with a vision to establish an internationally recognised centre of excellence and an ambition to graduate 50 doctoral students across five annual cohort intakes. Since that time, the CDT has been delivered through a strategic partnership between Newcastle University and the University of Nottingham in the UK, together with c. 40 external partners from global academia, international industry and UK Government. The first doctoral students graduated from the CDT in July 2024, with the final students expected to complete their PhD studies in 2028. This paper provides an overview of the training structure and skills development initiatives implemented and offers critical reflections on the experiences and challenges encountered throughout the CDT’s lifetime to date (February 2026). While the content will be of particular interest to academics and stakeholders involved in any branch of geospatial doctoral training, many of the findings are transferable. As such, the insights presented may also be of value to the wider academic community, particularly those considering the establishment of similar cohort-based doctoral training models.

The CO3D mission, launched in July 2025, aims to reconstruct the Earth’s continental surface in 3D using pairs of synchronous satellite images, generating a Digital Surface Model (DSM) at 1 m Ground Sampling Distance (GSD). Assessing the quality of these DSMs requires an inter-DSM comparison tool, leading CNES to collaborate with the GlacioHack collective and join the governance of their open-source software xDEM.

Originally developed for glacier research, xDEM already offered valuable features for DEM analysis including coregistration, uncertainty analysis, geomorphological terrain attributes computation, etc. Recognizing its potential, CNES made the strategic choice to no longer maintain its own tool and instead contribute to xDEM. The main contributions include the ability to rapidly obtain statistics, scalability improvements through tiling, and the introduction of a command-line interface.

This collaboration has created a more robust tool that benefits both the CO3D mission and the broader scientific community. By combining resources and expertise, the project demonstrates how open-source development can drive innovation while reducing duplication of effort.

DINAMIS is a French national initiative designed to provide streamlined, cost-effective access to very high-resolution satellite imagery for research, public policy, and innovation. Coordinated by major public institutions—including CNES, IGN, INRAE, and several academic partners—DINAMIS acts as a single entry point for users who need high-quality Earth-observation data to support scientific studies, environmental monitoring, land-use analysis, and operational pu-blic-sector missions.

The platform facilitates access to a range of commercial satellite constellations, most notably Pléiades, Pléiades Neo, and SPOT 6/7, which offer imagery with spatial resolutions from sub-meter to a few meters. Users can request both archived scenes and new acquisitions, enabling them to obtain data tailored to their geographic area and temporal needs. DINAMIS also pro-vides standardized licensing conditions that simplify data sharing within research teams and public organizations.

A key objective of DINAMIS is to democratize the use of very high-resolution imagery by re-ducing financial barriers. Academic and public-interest projects often benefit from free or highly subsidized access, encouraging the development of innovative applications in fields such as agriculture, forestry, natural hazards, coastal management, and urban planning.

By centralizing requests, ensuring data quality, and supporting users throughout the process, DINAMIS strengthens France’s Earth-observation ecosystem and fosters collaboration between scientists, government agencies, and technology developers. Ultimately, DINAMIS contributes to a more informed understanding of the environment and helps public authorities make evi-dence-based decisions for sustainable territory management.

The “Green & Pink for Sustainable Education” (TNE-GPSEducation) project strengthens international cooperation between ten Italian universities and partner institutions worldwide, promoting multidisciplinary training in sustainability. The initiative integrates expertise in natural resource monitoring, socio-environmental resilience, innovative teaching, health, and gender equality. Partner countries—including Brazil, Argentina, Cambodia, Thailand, Palestine, Georgia, Morocco, China, and Vietnam—play strategic cultural and academic roles and are central to recent international efforts to foster joint education, research, and innovation. Through mobility and capacity-building actions, lecturers, staff, and students enhance their skills while acquiring transferable competencies usable across institutions. Italy’s broader cooperation policies, aligned with UN, EU, and CRUI–CUCS strategies, further support partnerships such as the MoUs signed by POLIMI with ENSMR and UIR in Morocco.

Within this framework, WP4 “Advanced Skills” represents the project’s core, merging socioeconomics, Earth Observation (EO), Nature-Based Solutions (NBS), and health. Five Long Life Learning Courses have been modularised to establish an EO Hub at ENSMR, serving as a regional network node. A Call for Applications invites professors and researchers to attend AS-LLLC programmes at POLIMI, covering EO techniques, BIM–XR workflows, NBS design, LULUCF-based EO monitoring, and decarbonisation methods.

The EO Hub Pathway links global-to-local scales through (a) the systematic use of global EO programmes; (b) LULUCF-aligned indicators and multi-decadal satellite analyses; (c) site-specific phenological monitoring for regenerative agriculture; (d) carbon-removal computation through NBS; and (e) XR/VR tools for immersive awareness raising. Together, these elements support adaptive strategies, MRV systems, regenerative practices, and innovative land-management approaches for regions facing degradation and climate challenges.

Exploration of socio-spatial pertinent dimensions of human population and its geo-spatial distribution in Nepal has been a foremost concern of planners and researchers for development. An input data matrix of 75 X 88 representing Nepal’s demographic, socio-economic, and environmental variables were used to investigate spatial pattern of latent fundamental characteristics and to examine their geo-spatial variability by integrated use of RS, GIS, GPS, Factor Analysis, and ANOVA. Six fundamental socio-spatial dimensions of human population explaining 74.0 percent of total variance were investigated. Demographic was the most prominent and significant dimension accounting for 27.0 percent of the total variance spatially clustered in Terai region indicating demographic pressure: old dependency and family size and also evident by Factorial Areas Analysis (FAA). Facility-Education Dimension was the second most dominant accounting for 19.62 percent of total variance spatially having insignificant geographic variability. Maize production and Ethnic Dimension was found as the third dominant dimension and was significantly concentrated in eastern mountain and hill districts, characterized as high dominancy in ethnic and language issue. Mother Tongue- Marriage age was the fourth accounting for 9.47 percent of total variance spatially clustered on EDR significantly spatial variability among development regions. Kathmandu district locating lower-left corner of both axes indicating the free from both pressure of old dependency and large family size. Family size- Wheat production was the least important dimension, significantly different and spatially distributed in Terai Region. The study demonstrates the usefulness of geospatial technology for demographic, and production planning, and sustainable regional policy in Nepal.

Black Carbon (BC) and Organic Carbon (OC) are among the most influential aerosol species affecting air quality, radiative forcing, and climate interactions, especially in regions dominated by coal-based industries. Understanding their temporal behaviour and associated deposition processes is critical for assessing pollution dynamics and guiding regional mitigation measures. Korba, located in Chhattisgarh, India, is widely known as the “Power Hub of India” due to its dense cluster of coal-fired thermal power plants, aluminium smelters, and mining activities, making it an ideal location to examine carbonaceous aerosol loading. The primary objective of this study is to quantify monthly variations in BC and OC and evaluate their atmospheric interactions and deposition characteristics during the study period. Methodology involved extracting BC and OC fractions, including hydrophilic (BCPI, OCPI) and hydrophobic (BCPO, OCPO) components, along with dry and wet deposition fluxes and meteorological drivers such as relative humidity, temperature, pressure, and precipitation. The results show that BC ranged from 3.97×10⁻⁹ to 1.00×10⁻⁸, while OC exhibited higher values between 7.68×10⁻⁹ and 2.24×10⁻⁸, indicating dominance of organic aerosols over black carbon. Dry deposition of BC was significantly high (up to 2.29×10⁹), whereas wet deposition remained several orders lower (≈1.75×10⁻¹² to 1.19×10⁻¹¹). Meteorological conditions, including RH (23–87%) and temperature (290–308 K), modulated concentrations and deposition behaviour. Overall, the study highlights substantial BC–OC loading driven by industrial and combustion sources in Korba. The conclusion emphasizes the need for cleaner combustion practices, while future work may integrate chemical transport modelling to identify precise source contributions.

The school of urban construction in Beijing City University (BCU) is committed to cultivating application-oriented talents who serve for urban planning, urban construction and urban management. The Geographic Information Science (GIS) program in our university began in 2019. It is carried out on the basis of the investigation of the national needs, the industry development, and the actual situation of our university and other universities in Beijing. Based on the above analysis, we have explored Unmanned Aerial Vehicle (UAV) remote sensing technology and LiDAR as two of the training orientations, focusing on the training of data acquisition and processing capabilities using UAV and LiDAR. We have carried out a lot of explorations and practice in curriculum structure and practical teaching. Student's professional ability is obviously improved. Their competitiveness is significantly enhanced.

This is an abstract for a paper on best pedagogical practices in engineering education. In particular, the paper will focus on a project-based course involving group work. Post pandemic, the course has been run twice. In both iterations there were serious breaches of academic integrity. This happened even though reasonable measures to prevent cheating had been put into place. The aim for future offerings of the course would be to preventatively tighten those measures and in the unfortunate scenario that cheating happens again to explore tools for its early detection.

Due to rapidly advancing technology and increasing societal needs, there is a significant demand for capacity building in the geospatial domain, which involves developing the skills, knowledge, and resources of individuals and organisations. The European association EuroSDR, a not-for-profit international network organisation linking National Mapping and Cadastral Agencies (NMCA) with research institutes and universities in Europe, recognised the challenges of skills development in the geospatial domain more than two decades ago. The EduServ annual training programme, organised under the EuroSDR umbrella, is a well-established and internationally recognised series of e-learning courses in photogrammetry, remote sensing, and geospatial information (GI) science. Since its inception in 2002, it has primarily aimed to transfer knowledge from EuroSDR research projects to the wider GI community. In recent years, interest in EduServ courses has increased significantly, and the topics have expanded to address new geospatial technologies and growing societal needs. This paper aims to share EuroSDR’s experience in distance education with the wider scientific community. Rather than limiting EuroSDR expertise to the European GI community, European mapping agencies can share their knowledge and experience with the international GI community.

The discussion on this topic is particularly relevant, as the changing geopolitical landscape has impacted the dissemination of geospatial data globally, as evidenced by reduced NASA funding for Earth and atmospheric studies, as well as the recent US government shutdown. The political alliances of countries also restrict the data availability during critical periods, such as war or disaster. This reminds countries to invest on sovereign geospatial dissemination portals to sustain research, innovation, and public discourse. At the same time, the emerging global conflicts open a new window of opportunity for India’s “Unified Geospatial Portal,” which is under development to become a predominant source not just for India but for the global community to leverage datasets generated by India's satellites, covering India and beyond.

This paper presents a tailored pedagogical approach to digital heritage documentation in contexts where heritage is under threat. It was developed during the July–August 2024 UNESCO/ICOMOS mission to Kyiv, Ukraine, within the UNESCO/Japan Funds-in-Trust project “Support for Ukraine in Culture and Education through UNESCO / Emergency response for World Heritage and cultural property: damage assessment and protection,” in relation to the UNESCO World Heritage property “Kyiv: Saint-Sophia Cathedral and Related Monastic Buildings, Kyiv-Pechersk Lavra.” The mission focused on the Metropolitan’s Residence and the Bell Tower of the Saint-Sophia Cathedral Complex. In parallel with the production of documentation for emergency preparedness and future conservation assessment, the mission implemented a dedicated capacity-building programme for the staff of the National Conservation Area “Sophia of Kyiv.” The paper discusses five interconnected components of this training programme: preparation before the mission, structure and content of the sessions, training activities and didactic material, learning outcomes and targeted competencies, and adaptive responses to a conflict-affected environment. The case study shows that integrating training within an active documentation workflow can strengthen both the immediate value of the records produced and the longer-term capacity of local professionals to support conservation, monitoring, and risk preparedness at World Heritage sites under threat.

Processing massive archives of satellite imagery has historically paralyzed traditional desktop-based remote sensing laboratories. The sheer volume of computationally heavy tasks-from bulk atmospheric correction to long-term radiometric calibration-frequently exceeds the hardware capacity of local campus networks and student laptops. To bypass these severe limitations, this study presents a dual-cloud pedagogical architecture that integrates Google Earth Engine (GEE) and Alibaba's AI Earth. This hybrid framework allows students to instantly access petabytes of analysis-ready data while maintaining low-latency processing for complex modelling via domestic servers. We operationalized this setup through four core practical modules: urbanization monitoring, urban heat island analysis, nighttime light assessment, and AI-driven road extraction. By entirely eliminating the overhead of raw data management and environment configuration, students can finally redirect their cognitive focus toward the actual physics and algorithmic logic of remote sensing—such as parameterizing radiative transfer equations and interpreting radiometric time-series. Furthermore, in light of AI Earth's recent policy shift in March 2026, which heavily restricts free access for educational usage, we critically evaluate the long-term sustainability of this curriculum. To maintain unhindered access to cloud-native geoprocessing, our future instructional designs will assess alternative localized platforms and open-source AI frameworks, ensuring the uninterrupted evolution of rigorous Earth observation education.

Web-based tools for synthetic spatial data generation offer flexibility and accessibility to students and educators alike. This abstract takes a brief overview of some of the existing and developing tools to this end.

The COVID-19 pandemic reshaped higher education and accelerated the shift toward blended learning (BL). In GIS education, however, most BL practices have emphasized technologies rather than pedagogical foundations. This study introduces a Complex Adaptive Blended Learning System for GIS education (CABLS-GIS) — a theory-driven framework that conceptualizes BL as an interdependent system comprising the learner, teacher, content, technology, learning support, and institutional environment. The framework was implemented in an introductory GIS course at the National University of Singapore through a flexible-mode BL design integrating face-to-face and online components. Survey results from undergraduate and graduate students revealed positive perceptions of the CABLS-GIS approach, particularly regarding learning flexibility, motivation, and conceptual understanding. The findings highlight how theoretically grounded BL design can enhance pedagogical coherence, technological integration, and educational resilience in the post-pandemic era. CABLS-GIS thus provides a holistic and adaptive model for advancing GIS education and serves as a foundation for developing future personalized and data-driven learning strategies.

Remote sensing imagery is a crucial resource for evaluating flood-affected areas following inundation events. The integration of optical satellite data and UAV-based drone surveillance enables the development of precise flood extent maps. This research determined inundated areas by applying spectral water indices and classification methods to both Landsat and Sentinel-2 imagery, supplemented by UAV-based damage assessment. To delineate flooded regions, the study utilized the Normalized Difference Water Index (NDWI), the Modified NDWI (MNDWI), and the Water Ratio Index (WRI). Additionally, land use and land cover analysis were conducted using supervised classification with the maximum likelihood algorithm, enabling effective identification and comparison of flood extents across the indices. The flood coverage was estimated at approximately 107 km² via Landsat, 111 km² through MNDWI, and 115 km² using NDWI. By leveraging classification insights from each index, a targeted correction process was implemented to address misclassifications and enhance delineation accuracy. Notably, both MNDWI and NDWI yielded accuracy rates surpassing 90%, reinforcing the reliability of the results. The proposed remote sensing techniques offer a reliable and innovative approach for detecting flood-affected areas, contributing significantly to timely disaster response and targeted relief efforts.

The author aims to introduce some concepts and practical tools, which were usefully applied in the curriculum development influenced by the Bologna process and successfully used in the quality improvement practice.

The first part of the paper is dealing with the definition of education/training needs and involvement of stakeholder’s curriculum planning. One of the most important outcomes from these activities is the definition of skills and competences; and stakeholder management plan. The curriculum is a crucial component of any education/training activities, it is a road map to knowledge, and it builds knowledge topology. The implementation of new curricula often needs capacity building for faculty delivering education or training. Faculty of Geoinformatics (GEO) at Tribhuvan University of Kathmandu, Nepal participated or managed in many relevant international projects. The author will share some good educational practices.

The second part is focusing on curriculum and learning material development methods. The competency matrix will be introduced as a tool used to document and compare the required competencies for graduates. It is used in a gap analysis for determining where critical overlaps between courses are or which skills/competencies are not taught deeply enough.

Quality is omnipresent, ubiquitous – like the cloud of computers. Understanding and evaluating the quality of education requires a comprehensive picture of the unique and complex characters of the system that produced them. The third part of the paper is dealing briefly with quality impro issues.

Photogrammetry has been known for a little less than a century as the art and science of making precise measurements from optical images. In the last few decades, photogrammetry was taught with traditional approaches focusing on using images captured by metric cameras. Recently, new sensors have been adopted in the surveying and mapping communities. Employers are now looking for graduates with the knowledge and skills required to extract accurate and reliable data from these sensors. Therefore, novel approaches are needed to blend essential principles and cutting-edge technologies in the photogrammetric courses. This article outlines the design and implementation of a new syllabus for a photogrammetry class, the experience delivering the material, and student feedback. The new curriculum introduces students to non-metric camera calibration; laser scanning; and satellite image rectification. sUAV flight planning and data processing were the core of the redevelopment; hence, the article focus on blending sUAV in the curriculum. Topics are taught in lectures and then practiced in labs. Comments received from students and academic and industry experts supported the new design and recommended it as part of renovating new surveying programs.

The geometric monitoring of historic masonry towers is a central topic in heritage preservation, where structural safety must be ensured despite complex geometries, heterogeneous materials and deformation processes that evolve over time. This contribution presents an integrated surveying workflow developed by the DICAM Geomatics Laboratory and tested on the Garisenda Tower in Bologna, one of the most emblematic slender structures in Italy. The tower, built in the early 12th century and today inclined by more than 3 m, represents a challenging case study due to its ongoing deformation, dense urban context and the impossibility of establishing forced-centering stations.

The proposed methodology combines the high-precision capabilities of a Leica TS30 total station with the geometric completeness of a Leica RTC360 terrestrial laser scanner. The total station defines a stable local reference system and ensures accurate vertical alignment of the scanning instrument, while the TLS provides detailed three-dimensional reconstruction of the tower’s surfaces. The resulting 3D model enabled the computation of out-of-plumb parameters, wall flatness and local deformation patterns. Validation against TS30 control points confirmed the metric reliability of the integrated approach.

Cultural heritage is increasingly reinterpreted and experienced through digital and immersive environments, including Extended Reality (XR) and Augmented Reality (AR) technologies. While these engage visitors in novel ways, the trend raises questions about what constitutes an “authentic” digital experience. This study examines perceptions of authenticity in an AR experience at the Twin Pagodas, a small-scale heritage site in Suzhou, China.

Building on a framework that distinguishes between objective authenticity (the accuracy of content), constructive authenticity (the interpretive meaning conveyed through stories), and subjective authenticity (the personal and emotional experience), the study explores how these dimensions interrelate and are mediated during digital engagement. Data were collected from 108 participants (ages 8–67, Chinese and international visitors) via pre- and post-experience surveys and 20 semi-structured interviews. Participants rated statements capturing each authenticity dimension, and Pearson correlation analysis examined relationships among them. Ethical approval was obtained prior to data collection.

Findings indicate that authenticity in mobile AR heritage experiences operates across multiple interacting layers. Cognitive judgments about historical accuracy shape interpretive meaning-making, while affective engagement forms a relatively independent experiential dimension. This pattern aligns with existing scholarship that emphasizes the interpretive and experiential nature of authenticity in heritage contexts, while providing empirical evidence from a mobile AR implementation at a modest urban heritage site. Limitations include the focus on a single site and AR design, indicating the need for further research across diverse contexts to strengthen generalizability.

Digital archiving of cultural heritage using 3D scanned point cloud data requires effective edge-highlighting visualization to analyze fine structures. However, conventional methods often produce edges that are too thick, obscuring fine details.

This study proposes a method for adaptively optimizing the scale (range) used for local statistical analysis. This allows for the extraction of both sharp and rounded soft edges with high visibility. The core idea is to automatically determine the optimal scale for the analysis.

First, an eigenvalue-based feature value is calculated at multiple scales. Next, the scale that yields the minimum sample variance of this feature value across the entire point cloud is found and selected as the optimal scale.

Using this optimal scale, edge regions are extracted using another feature value. Opacity gradation is applied to emphasize soft edges as well. When this method was applied to a complex cultural heritage relief, fine structures such as ship hulls and human figures, which were indistinct with conventional methods, were clearly visible in the visualization results of the proposed method.

Coastal bathymetry and water-clarity assessment using multispectral remote sensing is essential for understanding nearshore dynamics, sediment transport, and environmental variability. Optical indices such as the Normalized Difference Water Index (NDWI) and satellite-derived depth models provide a rapid means of monitoring these changes. This study focuses on the coastal region of Goa, located along the central western coast of India, an area influenced by strong monsoonal cycles, tidal fluctuations, and high sediment exchange from estuarine systems and open-sea interactions. The objective of this work is to evaluate monthly and annual variations in NDWI and satellite-derived bathymetric depth from 2022 to 2024 and to assess their seasonal and statistical relationships. Sentinel-2 imagery was processed to generate monthly median composites, from which NDWI and bathymetry were extracted; monthly mean NDWI and median depth values were calculated to represent surface water conditions and subsurface optical penetration, respectively. Results show clear seasonal contrasts, with NDWI values ranging from –0.02 to 0.33 and depth values varying between –8.5 m (deep, clear water) and +8.4 m (high turbidity). Annual mean NDWI remained relatively stable (~0.15), whereas median depth became progressively shallower from –2.01 m in 2022 to –0.52 m in 2024, indicating declining optical water clarity. Seasonal correlations between NDWI and depth shifted from strongly positive in winter (r = 0.70) to strongly negative during the pre-monsoon period (r = –0.83), reflecting the influence of sediment resuspension and monsoonal turbidity. Future work may integrate turbidity, wave climate, and machine-learning models for enhanced depth estimation.

The concept of Levels of Detail (LoDs) plays a critical role in 3D semantic modelling by balancing geometric and semantic complexity with application needs. In our earlier work, we proposed a four-level LoD framework tailored to Buddhist statues, ranging from symbolic representation to detailed geometry, aiming to fulfil the needs for about 60 applications. However, when implementing this concept to applications in the cultural heritage domain, it is suggested to introduce an intermediate LoD between LoD2 and LoD3 because some applications need geometries coarser than the LoD3 but more detail than LoD2. In this paper, we present the analysis of these requirements and propose a new LoD for the 3D modelling of Buddhist statues. To verify the updated concept, we conducted a questionnaire among experts in geomatics and archaeology. Feedback from 170 participants confirmed that the five-level LoD concept is more appropriate and the revised framework provides a more comprehensive alignment with tasks in archaeology, conservation, museum exhibition, and risk management, and demonstrates strong potential for standardization within CityGML ADE.

In recent years, digital archives using VR technology have been actively created, but most are intended for viewing culutual properties, with few designed for analysis. In this study, we create a VR system for understanding the 3D structure of cultural properties, using the 3D point cloud data of Tamaki Shrine, a World Heritage site in Nara Prefecture, Japan, as an example. As a feature enhancement method, we performed feature enhancement using principal component analysis. Furthermore, by applying it to a transparent VR environment, we aimed to improve the visibility of 3D structures.

Multispectral data improves instance segmentation in digital agriculture by combining geometric and spectral information to distinguish complex natural features. While geometric information captures structural details, it often falls short when dealing with complex natural features that exhibit high spectral similarity, rather than due to limitations inherent to geometric representation itself. This work presents a feasibility analysis of instance segmentation using a spectral point cloud. A combination of spectral bands is selected based on class separability and proximity to a normal distribution as estimated by the Shapiro–Wilk test. The aim is to identify the minimum number of bands required to produce optimum results. For the normality analysis, Euclidean magnitude normalisation was applied, and it was also used alongside standard scaling to support the Multilayer Perceptron (MLP) for classification and segmentation. To refine the MLP predictions and consolidate instance labels, a graph-based post-processing step was applied, linking each point to its nearest neighbours and using a majority-voting scheme, resulting in spatially coherent clusters and refining the MLP predictions. The results demonstrate that multispectral data can reliably segment individual objects, with ten spectral bands being sufficient to achieve highly satisfactory segmentation and accurately delineate natural features such as leaves and tree trunks. Further increasing the number of bands improved spectral definition even more, with 14 bands achieving the highest performance across all metrics (mIoU: 96.59%; AP50: 96.14%). These findings highlight the strong potential of multispectral point clouds for precise and scalable object-level segmentation in agricultural environments.

This study presents a UAV-based framework for early detection of Bacterial Leaf Blight (BLB) in rice using multi-temporal and multi-modal data. Conducted at the International Rice Research Institute (IRRI) during the 2023 wet season, the experiment integrated multispectral, thermal, and RGB imagery with crop physiological measurements from both healthy and artificially inoculated fields. Spectral (NDVI, NDRE), thermal (canopy temperature), and textural features were extracted and analyzed using a Random Forest classifier to identify early indicators of BLB infection. Results demonstrated that combining spectral and thermal data enhances early disease detection before visible symptoms appear, supporting precision agriculture and sustainable rice disease management.

National mapping agencies (NMAs) provide authoritative and authoritative geospatial data for their respective countries. All geospatial agencies face significant challenges, including rapid technological advancements, societal expectations, and environmental pressures.

To produce high-quality geospatial information that meets user needs, NMAs combine image data acquisition from various sensors, field data collection, and manual interpretation and processing.

The use of geospatial artificial intelligence (GeoAI) offers opportunities to optimize workflows and reduce manual workload.

This article presents preliminary results from a study on the applications of GeoAI in the activities of National Mapping Agencies, along with key challenges and ethical considerations.

Detecting early-stage vegetation stress at the individual tree scale is a pivotal remote sensing application. The ``green shoulder'' band at 530 nm serves as a key signal for early stress detection due to its sensitivity to carotenoid changes. However, existing remote sensing systems often struggle to simultaneously capture fine-scale canopy structures and stress-sensitive spectral data, making heterogeneous fusion a promising topic. Unlike mainstream supervised methods that rely on prescribed degradation models and high-quality samples, an unsupervised blind fusion framework based on Implicit Neural Representation and low-rank decomposition is proposed in this paper. Guided by orthorectified aerial images, the framework performs per-band super-resolution on PlanetScope SuperDove data to achieve a 0.16-meter resolution. It employs Sinusoidal Representation Networks to learn a continuous joint implicit representation of spatio-spectral information, effectively modeling the non-linear relationship between canopy structure and spectral response.To mitigate high-dimensional feature redundancy during heterogeneous data fusion, low-rank decomposition is integrated to reduce computation overhead. Experimental results show that the proposed method can fuse heterogeneous images effectively, providing a solid solution with practical guidance for subsequent early stress monitoring at the individual tree level.

Urban LiDAR point clouds provide rich geometric information but pose significant challenges for automated interpretation due to their scale, noise, and semantic complexity. Traditional convolutional and graph-based networks (e.g., PointNet++, RandLA-Net) have made significant strides by focusing on local geometric feature learning. However, they often lack the ability to incorporate high-level, global semantic context. This limitation leads to persistent errors in object boundary delineation and category confusion, particularly for semantically or geometrically similar classes (e.g., 'road' vs. 'sidewalk',or 'low-wall' vs. 'curb').Recent advances in large language models (LLMs) have demonstrated remarkable capabilities in contextual understanding, reasoning, and knowledge retrieval. Inspired by these developments, and motivated by the growing trend of cross-modal alignment in vision-language models, we propose an LLM-enhanced segmentation framework that integrates linguistic priors into the 3D perception pipeline. Our key contribution is the use of contextual prompts—textual descriptions generated or retrieved by an LLM based on 3D scene content—to guide the segmentation network. These prompts provide disambiguating cues, enabling the model to better distinguish between challenging classes and to recognize objects that are rare in the training data.The main contributions of this work are:1.A novel framework that synergistically combines a geometric point cloud encoder with an LLM-based contextual prompter for semantic segmentation.2.A methodology for generating and fusing contextual prompts from point cloud data, bridging the gap between geometric perception and linguistic reasoning.3.Extensive experiments demonstrating superior performance over state-of-the-art methods, particularly on semantically ambiguous and long-tailed object categories.

This contribution presents a new multi-sensor dataset of the urban road network of Turin, designed to support research in Digital Twins, AI-based road monitoring, and semantic 3D modelling. The dataset integrates mobile mapping (MMS), aerial LiDAR, imagery, and BIM/IFC models into a unified spatial and semantic framework. It includes detailed point cloud classifications, pavement defect annotations, and metadata to ensure full reproducibility. By combining geometric precision with semantic labelling, the dataset enables applications such as automated defect detection, semantic segmentation, 3D reconstruction, and predictive maintenance. Compared to existing benchmarks, it offers a unique focus on road surface condition and DT interoperability. The contribution outlines the methodology used to structure, validate, and document the dataset, positioning it as a valuable resource for both academic research and operational urban infrastructure management.

Structural inspection is crucial for comprehensive risk management, especially given the accelerated deterioration caused by factors such as climate change and obsolescence. The accurate determination of the percentage of surface damage is fundamental for optimizing maintenance decision-making and the administration of resources for infrastructure preservation. This work presents a methodological exploration to assess the superficial condition of a concrete pedestrian bridge located over an urban river. The study focuses on determining the structural conditions by calculating the percentage of surface damage to evaluate maintenance needs. For data acquisition, Light Detection and Ranging (LiDAR) technology is employed using a Terrestrial Laser Scanner (TLS) Trimble X7 laser scanner, generating a 3D point cloud that models the bridge surface with precise spatial coordinates. The methodology utilizes the reflective intensity of the laser pulses to obtain quantitative information about the surface. This approach allows for the precise identification, demarcation, and quantification of deteriorated areas. The application of this methodology facilitates a non-invasive and detailed diagnosis of the surface condition, providing quantitative and visual information that can enhance the maintenance planning of critical infrastructure such as pedestrian bridges.

Understanding public experiences in urban greenspace is essential for supporting more human-centric design and management. While traditional survey methods are often time- and labor-intensive, user-generated content (UGC) offers a rapid and scalable alternative for capturing public experiential insights. However, extracting detailed user experience information from this data remains methodologically challenging. This study proposes a novel multimodal analytical framework based on online review data and natural language processing techniques, combining LoRA fine-tuned RoBERTa model with CLIP vision-language model to analyze multidimensional ecosystem service experience patterns in urban greenspace from user-generated text and image reviews. Results demonstrate that the proposed approach achieves more robust extraction and analysis of user experience insights compared to conventional deep learning and lexicon-based methods, exhibiting greater capacity to process contextually embedded experiential information. The multimodal framework enables more comprehensive capture of user experiences than either text or image data alone, with particular gains on dimensions that are difficult to represent through a single modality. Applying the analytical framework to Amsterdam and Rotterdam as case studies, statistical and spatial analysis reveals heterogeneity in user urban greenspace experiences and identifies key experiential bundles alongside their associated synergies and trade-offs. This study offers a novel approach to quantifying urban greenspace experiences from a user perspective, and provides insights for evidence-based urban greening practices.

This paper describes the development of a 3D mesh product by Ordnance Survey, Britain's National Mapping Agency. The work originated in the research team and was then taken up by a multi-disciplinary cross-business team which used product development techniques and extensive customer interviews to determine the feasibility (could it be made) and viability (would it generate sufficient revenue) of a potential 3D mesh product. The 3D mesh, generated from nadir aerial imagery already captured for topographic map update, was introduced as a beta product and is currently being tested by potential users.

Photorealistic 3D reconstruction fundamentally requires recovering the intrinsic optical properties of object surfaces. Traditional multi-view photogrammetry, based on Structure-from-Motion (SfM) and Multi-View Stereo (MVS), effectively reconstructs geometry and texture but assumes Lambertian reflectance, failing on non-Lambertian materials with specular highlights and subsurface scattering. While recent implicit representations like NeRF and its extensions have advanced novel view synthesis, their effectiveness is constrained by the inherent coupling of geometric, material, and luminous properties. To overcome these issues, we propose a differentiable rendering method for photorealisitic material reconstruction in close-range photogrammetry, enabling physically accurate forward and inverse rendering of PBR parameters. Experimental results demonstrate that our method achieves high-fidelity reconstruction of object geometry and multi-channel SVBRDF/BSSRDF materials, robustly recovers HDR environment maps under complex indoor and outdoor illumination, can effectively removes indirect illumination artifacts through Monte Carlo ray tracing, and produces editable assets that enable realistic relighting and material editing.

The emergence of vision–language models (VLMs) has enabled joint multimodal understanding beyond traditional visual-only approaches. However, transferring VLMs from natural images to remote sensing (RS) segmentation remains challenging due to limited category diversity and significant domain gaps. We propose a training-free framework that decouples visual and textual inputs and performs multi-scale visual–language alignment for RS segmentation. At the global–local decoupling module, we separate text into local class nouns and global modifiers, while images are partitioned into class-agnostic mask proposals via unsupervised mask generation. At visual–textual alignment module, we introduce a context-aware cropping strategy and a knowledge-guided prompt engineering method to enhance text representations, enabling mask classification for open-vocabulary semantic segmentation (OVSS). A Cross-Scale Grad-CAM module refines activation maps using contextual cues from global modifiers, facilitating accurate and interpretable alignment for referring expression segmentation (RES). Evaluations on the benchmarks demonstrate strong performance, highlighting the potential of training-free VLM transfer to the RS domain.

Retrogressive thaw slumps (RTSs) are key indicators of permafrost degradation in Arctic regions. Yet, their detection remains challenging due to spectral similarity with surrounding terrain and the limited generalization of conventional deep learning approaches. This study presents a Geo-Foundation framework that integrates pretrained Clay embeddings with high-resolution PlanetScope multispectral imagery, spectral indices, and ArcticDEM data for RTS detection in the Northwest Territory (NWT), Canada. The proposed dual-branch architecture combines high-level geospatial representations with physically meaningful environmental features to improve segmentation performance. The model achieved an F1-score of 0.83 and a mean Intersection-over-Union (mIoU) of 0.75 on the validation dataset. Analysis of patch size indicates that intermediate spatial context provides optimal performance, while feature importance results highlight the dominant role of vegetation-sensitive spectral bands and indices. Qualitative evaluation further confirms accurate boundary delineation and spatial consistency across diverse terrain conditions. The results demonstrate that Geo-Foundation models enhance detection accuracy, reduce dependence on large labeled datasets, and improve generalization across heterogeneous Arctic landscapes. This approach provides a scalable and efficient solution for monitoring permafrost-related disturbances under a changing climate.

Forests provide essential ecosystem services beyond timber, yet locating high-yield areas for non-wood forest products such as bilberries (Vaccinium myrtillus) remains a challenge for both recreational and commercial pickers. By integrating Airborne Laser Scanning (ALS), Geographical Information System (GIS) data, and crowdsourced terrestrial imagery analyzed via deep learning (YOLO), we developed a predictive system optimized for identifying high-yield hotspots. We demonstrate that YOLO detection remains highly accurate, but plant height significantly contributes to berry omission. However, this limitation can be mitigated by selecting the maximum berry count from multi-angle terrestrial images. Using a Random Forest classifier across a 36-km² study area in Nuuksio, Finland, we achieved a precision of 58% for the highest yield category. This represents a 20-fold increase in the probability of encountering a high-yield area compared to random searching. Extensive user testing over two years validated the practical utility of the system, showing a 22.5% increase in harvested yield and a 36.5% reduction in time required to locate hotspots. Furthermore, 97% of users reported that the platform provided an accurate big picture of bilberry yield. These results highlight the potential of combining crowdsourced citizen science with advanced LiDAR metrics to create digital twins of forest ecosystems that enhance human interaction with nature and optimize the sustainable harvest of wild food resources.

Cultural heritage (CH) digitization currently suffers from fragmented multi-source heterogeneous data, insufficient knowledge organization, and limited semantic expression in 3D CH models. Existing knowledge graphs and HBIM in CH field lack unified semantic representation and effective GIS integration, thus restricting intelligent knowledge services. To overcome these issues, a knowledge service approach integrating knowledge graph and semantic 3D models is proposed, enabling the transformation from data process to knowledge-driven services. An extension model for CH (CHADE) is developed using the CityGML ADE mechanism to support the construction of semantically enriched 3D geospatial scenes. Meanwhile, A domain ontology (CHOnto) based on CIDOC CRM is constructed to formalize CH knowledge, and multi-source heterogeneous data are organized into a Cultural Heritage Knowledge Graph (CHKG). By establishing semantic connections between knowledge graph and 3D models, the proposed method achieves integrated representation of geometry, spatial context, and domain knowledge. A prototype system (3DCHKS) is implemented and validated through multiple heritage scenarios. Results demonstrate that the approach enhances semantic connectivity, knowledge organization, and scenario-based representation, supporting intuitive visualization and intelligent application. Although limitations remain in generalizability and knowledge extraction robustness, this study provides a novel framework for integrated CH knowledge services and lays a foundation for scalable, knowledge-driven heritage applications.

Aerosols play a critical role in the development of boundary layer and build-up of air pollution in urban environments. Their presence in the atmosphere is calculated and represented by Aerosol Optical Depth (AOD). Satellite sensors observe aerosol quantities and different algorithms are applied to retrieve AOD at varied spatial and temporal resolutions. In air quality monitoring, satellite-based AOD products are useful in modelling particulate matter (PM). This study evaluates AOD products observed by Moderate Resolution Imaging Spectroradiometer (MODIS), Visible Infrared Imaging Radiometer Suite (VIIRS) and Advanced Himawari Imager of Himawari-8 in predicting inland daytime PM2.5 for test sites in Japan and South Korea. Prediction models are constructed using eXtreme Gradient Boosting (XGBoost) regression with input variables from observation datasets matched on PM2.5 station locations. In addition to AOD, seventeen (17) predictor variables were considered to account topographic and meteorological parameters that can influence the formation and transport of PM2.5 near the ground surface. Overall results show that prediction model using MODIS MAIAC AOD generate relatively higher accuracy for daily estimates considering both spatial coverage and prediction skill metrics. For future work, model improvements will be done by exploring additional predictor variables to reduce overfitting and additional statistical tests to generate more accurate estimates of PM2.5.

Geospatial embeddings represent a promising paradigm for encoding geospatial information into compact and learnable representations that support scalable downstream tasks in remote sensing. Among recent developments, Google’s AlphaEarth embeddings are a dataset of 64-dimensional embeddings, made available globally at 10 m resolution, derived from multimodal inputs, including multispectral and SAR imagery, elevation, gravity and text data. In this study, we explore the feasibility of inferring surface height from AlphaEarth embeddings within a deep learning framework. The analysis focuses on an 8000 km² area in Nouvelle-Aquitaine, France, where a 5 m resolution Digital Surface Model (DSM) is available. A U-Net architecture with a ResNet34 encoder was trained to predict surface heights from the 64 embedding channels using a spatial cross-validation strategy to ensure independence between training and testing subsets. For computational efficiency in this preliminary experiment, both the embeddings (input) and DSM (target) were resampled to 100 m. Results indicate promising agreement between predicted and reference heights, achieving an R² of 0.83 and a Pearson correlation of 0.93 on the test set. However, a systematic bias was observed. These findings highlight the potential of AlphaEarth embeddings to capture height-related features, despite being trained on a broader geospatial domain. Future work will address bias investigation, increase inference spatial resolution, and expand the analysis across diverse geographical regions. Additionally, comparisons with alternative embedding datasets, such as Tessera, will be conducted to better evaluate the strengths and limitations of embedding-based surface height estimation.

With the rapid advancement of large language models, natural-language-based understanding of satellite task requests is becoming increasingly important for improving the accessibility of remote-sensing services. However, satellite commands issued by non-expert users are often conversational, ambiguous, and terminologically inconsistent, leading to a substantial gap between free-form expressions and structured task representations. To address this challenge, we propose a hierarchy-aware framework for intent recognition and task-oriented text generation from non-expert satellite instructions. Specifically, we design a hierarchical annotation scheme that models intent levels, parameter structures, inter-element relations, and execution complexity, and we further construct a hierarchical sequence representation for learning. We then introduce a boundary-aware sample organization method based on semantic similarity and structural proximity, together with a retrieval-augmented multi-type negative-sample reorganization strategy to enhance robustness. Finally, we adopt Qwen3-8B with LoRA for parameter-efficient domain adaptation and unified generation of top-level intents and task-oriented outputs. Experiments on a manually curated dataset of 4,025 non-expert satellite instructions show that the proposed method consistently outperforms multiple baselines on both intent classification and task-oriented generation, demonstrating a resource-efficient and scalable solution for natural-language satellite task interfaces.

China’s Low-Altitude Economy (LAE) requires robust airspace surveillance for the safe integration of Vertical Take-off and Landing (VTOL) aircraft and Unmanned Aerial Systems (UAS). Traditional radar Automatic Target Recognition (ATR) approaches—both micro-Doppler-based and tracking-based—depend on track accumulation, introducing Detection Response Times (DRT) exceeding 3–5 seconds that are incompatible with real-time low-altitude operations. This paper proposes a tracking-free ATR methodology that restructures the conventional serial “Detection–Tracking–Recognition” chain into a parallel “Integrated Detection and Recognition” (IDR) architecture. The classifier operates independently of the tracker, extracting target attributes from single-dwell echoes within one Coherent Processing Interval (CPI), achieving a DRT below 100 milliseconds—more than an order-of-magnitude improvement over existing systems. The methodology is validated through field trials using a X-band radar, demonstrating reliable identification of VTOL at ranges exceeding 12 km. We further clarify the precise definition of DRT and argue for NATO ATR hierarchy level T3 (Recognition) or above as the minimum performance standard for low-altitude radar sensors.

The Amazon rainforest faces threats from anthropogenic disturbances, which also increase greenhouse gas emissions and contribute to global climate change. In 2004, a system to detect disturbance for the Brazilian Legal Amazon (BLA) was created to mitigate forest loss. The system, Detection of Deforestation in Real Time (Deter), from the National Institute for Space Research (INPE), alerts to seven types of anthropogenic forest disturbances through the visual interpretation of optical satellite imagery from CBERS-4, CBERS-4A and Amazônia-1. Many near-real-time systems currently generate alerts using automated algorithms, primarily leveraging SAR sensors to compensate for the absence of cloud-free images over tropical forests. Deter uses spatial patterns to identify types of disturbances, minimising commission errors, while most algorithms prioritise the temporal dimension for early-stage detections. Discrepancies in space and time across systems and disturbance types, such as omissions, delays, and mismatches, are linked to the selection of sensor technologies, forest masks, and algorithm strategies. Forest disturbances detected between 2020 and 2024 for the entire Brazilian Amazon Biome were extracted from the systems: Deter, Prodes, MapBiomas, SAD, RADD, GLAD, LUCA and TropiSCO. Based on this dataset, we conducted an exploratory analysis revealing agreement and disagreement between detection systems regarding five classes of disturbances (clear-cut, selective logging degradation, fire scars, mining and windthrow). The results emphasise the importance of systems that consider the trade-off between spatial and temporal context to detect different disturbance types, similar to Deter, but using automated near-real-time alert approaches.

In archaeology, the long-standing problem of low efficiency and high experience-dependence in manual matching of numerous unearthed pottery shards has been a challenge. This paper presents and develops an intelligent matching and annotation tool for pottery shard images, integrating advanced computer vision technologies. Using 35,159 pottery shard images from Pit H690 at the Daxinzhuang Site in Shandong as the dataset, a comprehensive “segmentation-feature extraction-cross-verification-screening” technical process is established. The core steps are as follows: First, the natural-language-based visual segmentation model Lang SAM is employed to precisely segment individual pottery shards from the original images, obtaining clean front and back images. Second, the self-supervised visual feature model DINO v2 is used to extract deep visual feature vectors of the shards, calculate image similarities for the front and back sides respectively, and generate a Top-N candidate matching list for each shard. Finally, cross-verification is carried out by taking the intersection of the front and back candidate lists, and the final screening is conducted with archaeological metadata. This research demonstrates the great application potential of AI in archaeological fragment assembly, offering an automated, interpretable, and efficient solution for handling massive cultural relic fragments.

Effective maritime surveillance and management of small-scale fisheries remains challenging in coastal waters because small vessels are not systematically tracked and are weakly represented in medium-resolution satellite imagery. Within the AI4COPSEC Horizon Europe framework, this study investigates an object-detection workflow for small-vessel monitoring along the Adriatic coasts of Marche and Puglia, Italy. A multisource dataset was prepared in which Sentinel-2 and PlanetScope optical imagery were manually annotated to enrich an existing SAR and optical imagery training dataset and support a two-stage training strategy. The first stage used a larger, more heterogeneous dataset for robust feature learning, while the second refined the model on a smaller, higher-quality subset to improve domain adaptation and detection performance. The resulting dataset comprised 4,202 image tiles (pretraining) and 706 image tiles (fine-tuning), with 16,096 and 1,716 vessel annotations, respectively, all belonging to a single target class.

Detection experiments were conducted with several YOLOv26 configurations trained under a consistent protocol to assess the trade-off between model complexity, accuracy and computational efficiency. Among the standard variants, YOLOv26-M achieved the most balanced performance, with a Precision of 0.813, Recall of 0.846, F1-score of 0.829, Accuracy of 0.719 and mAP50-95 of 0.306. Pruned and lightweight alternatives showed competitive efficiency-oriented behaviour. Results indicate that, in small-target coastal environments, scaling up model size does not necessarily yield proportional gains, whereas task-oriented architectural design improves the balance between detection quality and computational cost. The workflow provides a practical benchmark for AI-enabled maritime monitoring and supports the advancement of Copernicus-oriented coastal surveillance applications.

The growing use of artificial intelligence (AI) and data-driven methods in architectural heritage research requires structured and reusable semantic units to support consistent modelling, annotation, and knowledge alignment. In this context, Historic Building Information Modelling (HBIM) can serve as a semantic anchor by linking surveyed geometry with object-based representations and non-geometric information. However, current HBIM workflows remain semantically fragmented: point cloud segmentation often relies on project-specific labels, object modelling adopts inconsistent decomposition and naming logics, and semantic enrichment is frequently implemented through custom parameters without a shared component-level framework. Although Industry Foundation Classes (IFC) provide the most widely adopted canonical structure for interoperability, their standard entities are often too coarse to represent heritage-specific subcomponents. To address this gap, this study proposes an IFC-compatible semantic framework for component-level representation in HBIM. The framework combines a canonical IFC-aligned layer with a heritage extension layer and introduces a mapping strategy for representing semantically meaningful subcomponents without modifying the core IFC schema. A Serliana arch on the church of SS. Paolo e Barnaba in Milan is used as a case study to illustrate the implementation of the proposed approach. The study establishes a preliminary semantic foundation for component-level heritage representation in HBIM, providing both a conceptual basis for structuring heritage subcomponents and an operational basis for their IFC-compatible implementation. This foundation may also support future developments in ontology alignment and cross-modal AI applications, where stable semantic anchors are required for data integration and annotation.

Thousand-Buddha niches on the walls of grotto temples are core carriers of China's Buddhist cultural heritage. Their high-precision digital extraction is a key prerequisite for virtual restoration of cultural relics, stylistic lineage research, and digital display. Currently, close-range photogrammetry is mostly used for digital acquisition of small and medium-sized grotto temples to obtain point clouds. This technology, through non-contact multi-view image collection and matching, can not only retain the fine morphological features of niches but also comply with the core requirement of "non-destructiveness" in cultural relic protection, making it the mainstream method for grotto temple point cloud collection. However, the segmentation of thousand-Buddha niche point clouds still faces two core challenges: first, the sample scarcity bottleneck in cultural relic scenes. Manual annotation of niches requires professional archaeological knowledge, which is time-consuming and labor-intensive, resulting in limited sample size that is difficult to support the full training of deep learning models; second, the segmentation adaptation problem of target characteristics. Niches are densely distributed with similar shapes, and point clouds from close-range photogrammetry are prone to local noise due to lighting differences. Traditional segmentation methods are prone to boundary blurring, misclassification, and missing segmentation. Pure transfer learning without combining the characteristics of cultural relic scenes leads to insufficient segmentation accuracy.

Remote sensing image super-resolution aims to recover fine structural and textural details from degraded low-resolution observations. However, conventional methods and early deep learning models often produce over-smoothed results and struggle to reconstruct realistic high-frequency content. Stable Diffusion-based (SD-based) methods offer a promising alternative by using strong generative priors to synthesize more natural, detail-rich super-resolved images. Although many SD-based super-resolution methods have been proposed in computer vision, their use in remote sensing imagery remains limited, and systematic comparative evaluation in this domain is still lacking, leaving insufficient empirical guidance for method development. Therefore, this paper compares four representative SD-based super-resolution methods, namely Stable Super-Resolution (StableSR), Semantics-Aware Super-Resolution (SeeSR), Different Blind Image Restoration (DiffBIR), and Pixel-Aware Stable Diffusion (PASD), on the WHU-Mix remote sensing dataset. The evaluation uses seven metrics: Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), Learned Perceptual Image Patch Similarity (LPIPS), Frechet Inception Distance (FID), CLIP Image Quality Assessment (CLIP-IQA), Multi-Scale Image Quality Transformer (MUSIQ), and Multi-Dimension Attention Network for No-Reference Image Quality Assessment (MANIQA). Quantitative results show that StableSR achieves the highest PSNR of 23.16 dB, PASD obtains the best SSIM of 0.81 and lowest LPIPS of 0.45, SeeSR achieves the best MUSIQ of 64.57 and MANIQA of 0.46, and DiffBIR achieves the best FID of 110.58 and CLIP-IQA of 0.68 but with weaker full-reference fidelity. These findings indicate that current SD-based methods favor different aspects, including fidelity preservation, perceptual quality, and generative realism, and should be selected according to the target remote sensing application.

Monocular depth estimation (MDE) infers depth from a single image, offering significant advantages in computational efficiency and memory consumption compared to conventional Multi-View Stereo (MVS) methods. However, most MDE methods suffer from poor multi-view geometric consistency, which limits their application to photogrammetric 3D reconstruction. To address this issue, this paper employs sparse point clouds of Structure-from-Motion (SfM) as extra geometric constraints and proposes a framework that achieves photogrammetric 3D reconstruction using off-the-shelf learning-based MDE models without the need for additional fine-tuning. Specifically, when SfM priors are available during inference, globally geometrically consistent depth maps can be directly predicted. Otherwise, the estimated monocular depths are aligned to a consistent scale using SfM results via a post-correction step. The resulting depth maps are then fused using a truncated signed distance function (TSDF) to generate dense 3D reconstructions. Experiments on photogrammetric datasets demonstrate that the proposed framework effectively improves geometric consistency across depth maps and enables high-quality scene reconstruction. In addition, we systematically analyze the impact of key parameters in depth inference and fusion, including depth map resolution, voxel size, denoising steps, and ensemble size, on reconstruction performance, and further explore the potential of MDE for photogrammetric 3D reconstruction.

Recognising individual trees has important applications in forest ecology and management. Conventional individual tree segmentation methods tend to favour dominant trees with pronounced canopy surface features but have limited capability in detecting subdominant trees that are partially occluded or have smaller crowns. To mitigate this issue, we propose a morphology-based method for individual tree segmentation. First, a treetop extraction method is developed based on morphological criteria. Candidate treetops are initially detected using local maximum filtering, followed by classification and validation through vertical profile analysis integrated with crown morphological characteristics. Subsequently, the extracted treetops serve as seed points to guide individual tree crown delineation within a Min cut/Max flow graph cut framework, leveraging the spatial relationships among points. Our method enhances the detection of subdominant trees, with detection rates climbing to 90–95%, and achieves an average F score of 0.8 for crown delineation, which outperforms the other methods by 0.24 points. By integrating treetop information with local crown features, the proposed method improves the detection and segmentation accuracy of subdominant trees in complex forest environments, supporting overstory structure analysis and individual tree inventory in intricate forests.

This paper presents an automated approach for quantifying corrosion surface areas in steel transmission towers by integrating Unmanned Aerial Vehicle (UAV) photogrammetry and deep convolutional neural networks (DCNNs). Traditional visual inspections for corrosion pose significant challenges to structural safety and maintenance planning due to their complexity, subjective nature, high costs, and safety risks associated with inspecting tall structures. The proposed methodology utilizes a DeepLabv3+ model for the semantic segmentation of corroded areas. The network was trained and validated using a robust dataset of 999 field photographs collected from on-field tower inspections. A comparative analysis of DCNN backbones identified MobileNetV2 as the optimal choice, offering a superior balance between accuracy and computational efficiency. After fine-tuning, the network achieved an acceptable validation accuracy of 90.8% and a validation loss of 0.23. A major contribution of this study is the integration of these deep learning algorithms with metrically accurate photogrammetric products. The trained network was applied to orthomosaics derived from the 3D reconstruction of the South-East tower at the Torino Eremo broadcasting center. Unlike traditional image segmentation which lacks spatial reference, the photogrammetric approach enables the quantification and localization of the corrosion extent in exact physical dimensions. The high accuracy of the orthomosaic was confirmed against ground-truth measurements, achieving a root mean square error of 0.87 mm.

This automated, deep learning-based framework streamlines the detection process, provides reliable and quantitative data for assessing structural integrity, and represents a significant advancement over manual inspections, enhancing the overall efficiency, safety, and accuracy of infrastructure maintenance

Accurate identification of urban building functions is crucial for smart city planning and sustainable development. AlphaEarth Foundations introduce a new paradigm in remote sensing by providing semantically rich, pre-trained embeddings that integrate multi-sensor, spatiotemporal, and contextual information. In this study, we propose a novel fusion of 64-dimensional AlphaEarth embeddings and OpenStreetMap (OSM) derived building spatial indicators. We use the city of Toulouse as the study area, with the French official OCS GE database providing the ground truth labels. A random forest classification model was constructed, and the classification performance of single-source versus multi-source feature fusion was systematically compared. Results demonstrate that the multi-source feature fusion model achieves optimal classification performance, with an overall accuracy of 72.1\%, significantly surpassing models relying solely on embedding features (68.7\%) or spatial features (53.3\%).

The findings demonstrate the effectiveness and superiority of integrating AlphaEarth embeddings and OSM-derived building spatial indicators for automated urban building function identification, and provide a reliable technical approach for achieving large-scale and high-precision urban functional mapping.

We present an LLM-driven framework for retrieving remote sensing data from cloud-based geospatial catalogues using natural language queries. The system converts user intent into structured API calls, enabling efficient access to satellite imagery and environmental datasets. The architecture integrates three agents: Guardrail for safety and policy enforcement, General-QA for intent interpretation, and Recommender-Analyst for schema-aware API call generation. This coordinated design ensures reliable, semantically aligned interaction with external data services. The modular framework is portable across platforms through API schema substitution and supports applications in environmental monitoring, disaster response, and climate analysis. It establishes a scalable interface between user intent and geospatial infrastructure, enabling streamlined and automated Earth observation workflows. Preliminary experiments under adversarial multi-turn settings show that prompt-level safety instructions improve robustness, although rare high-impact failures persist in API manipulation scenarios and highlight the need for adaptive, system-level defenses that balance safety, usability, and cost efficiency, which motivates the use of our intercept-level Guardrail agent.

This paper introduces the second version of an end-to-end framework, which is the EDGE-Solar Farm Observation System (EDGE-SFOS v2.0). This system was developed for real-time solar farm defect detection with Edge generative detectors using drone images.

This study addresses the critical issue of systematic biases in gridded meteorological datasets, which can lead to inaccurate agricultural predictions and flawed decision-making. The primary objective is to develop a unified, high-accuracy meteorological dataset for Manitoba to support agricultural applications. The study focuses on the 2005–2024 period and on key variables commonly used in agriculture, including minimum temperature, maximum temperature, precipitation, and solar radiation. The methodology involves two main stages. First, four widely used national and international gridded datasets, ERA5-Land, Daymet, CHIRPS, and ANUSPLIN, will be benchmarked by comparing gridded values extracted at the locations of more than 120 Manitoba weather stations with the corresponding station observations. Second, the best-performing dataset for each variable will be selected for bias adjustment. Traditional statistical methods, such as Linear Scaling and Quantile Mapping, will be compared with machine-learning and deep-learning approaches, including Linear Regression, Random Forest, XGBoost, DNN, LSTM, and 1D-CNN. The study is expected to provide a quantified assessment of dataset reliability for Manitoba and to produce an improved bias-adjusted meteorological dataset for regional applications. The resulting dataset is intended to support more accurate agro-climatic assessments, regional yield estimation, and crop modelling, while also offering a scalable framework for similar agricultural regions.

Accurate mapping of urban tree canopy is essential for quantifying ecosystem services and assessing the impact of green infrastructure on wellbeing and public health. This study evaluates and compares three Geospatial Artificial Intelligence (GeoAI) frameworks for the automated detection and segmentation of tree cover. The frameworks are YOLO, Detectree, and TreeEyed Utilizing high-resolution aerial imagery (0.2 m and 0.5 m ground sampling distance), the research tests different deep-learning paradigms, including object detection and semantic segmentation. The results indicate that while object-based models like YOLO align closely with statistical baselines (30.83% vs 30.11%), pixel-based models such as Detectree may underestimate fragmented urban vegetation. The study highlights the effectiveness of the TreeEyed QGIS plugin for urban applications and emphasizes the necessity of local LiDAR-derived data for model validation. Further studies would benefit from ad-hoc training with correct co-registration and consistent coordinate reference systems across layers.

Maritime vehicles face significant positioning challenges under adverse weather conditions where visual and laser SLAM systems suffer from severe degradation. Millimeter-wave radar offers inherent robustness to weather interference, yet single-band radar cannot simultaneously achieve accurate translation and robust attitude estimation.This paper proposes a complementary fusion framework for multi-band radar odometry.This system leverages W-band radar (CFEAR) for reliable translation estimation and combines it with X-band radar (LodeStar) to improve rotational estimation robustness. The main innovations are as follows:(1) A complementary fusion framework exploiting the complementary characteristics of W-band and X-band radar; (2) A quality-aware adaptive weighting mechanism dynamically computing fusion weights based on sensor data quality assessment; (3) A consistency gating mechanism monitoring inter-sensor agreement and activating protective measures during sensor degradation.Experiments on the MOANA maritime dataset demonstrate that the proposed method achieves stable and reliable local motion estimation, reaching an RTE RMSE of 1.67 m on the Near-Port sequence.

In recent years, Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have emerged as advanced methods for photogrammetry-based 3D reconstruction. Since its introduction in 2020, NeRF has gained significant attention due to its capability to generate high-fidelity reconstructions from multi-view imagery. More recently, 3D Gaussian Splatting (3DGS), introduced in 2023, has proposed an alternative explicit scene representation based on a collection of anisotropic Gaussian primitives optimized directly in 3D space. This representation allows efficient rendering and scalable modelling of complex scenes while maintaining high visual quality. This paper analyses the performance of different 3DGS methods when dealing with complex geometry and less-cooperative surfaces compared to standard SfM IM procedures. Included in the comparison is also the Mesh-In-the-Loop Gaussian Splatting for Detailed and Efficient Surface Reconstruction (MILo), a novel meshing method using Gaussian splats. Three Gaussian splatting methods as implemented in the Postshot commercial software were also tested. Our experiments show that MILo shows very promising results in terms of detail reconstruction, while standard Gaussian splatting excels in visualisation but is still plagued by a high rate of noise especially when converted into a geometric point cloud form.

In this paper, we present the design and pilot implementation of Slovenia’s national geospatial Digital Twin (DT), coordinated by the Surveying and Mapping Authority of the Republic of Slovenia (GURS). Geospatial digital twins are enriched digital replicas of real world environments, dynamic models capturing past, present, and projected states to support geospatial decision making, location based services, and scenario simulations.

To demonstrate how the Slovenian Geospatial DT can be applied in practice, a prototype for modelling and managing flood hazards was developed. A flood-hazard prototype demonstrates the approach using the August 2023 event. The flood-modelling framework integrates very high-resolution (VHR) geospatial datasets with in situ environmental observations to ensure detailed spatial representation and analytical consistency. It combines ALS-derived terrain models with hydrological time series, meteorological forecasts, and satellite-based water detection from sources such as Sentinel-1/-2 and PlanetScope, enabling three-dimensional simulation and visualisation of flood dynamics.

The results show how a geospatial DT can transform authoritative datasets into operational intelligence for emergency management, spatial planning and climate-risk scenarios. Beyond floods, the architecture generalises to landslides, drought monitoring, infrastructure condition assessment and biodiversity applications.

Post-construction vegetation monitoring along linear infrastructures is increasingly required to support evidence-based restoration assessment, yet conventional ground surveys remain spatially sparse and difficult to scale over narrow, heterogeneous corridors. This limitation is particularly critical in recently replanted pipeline clearings, where plant-level restoration outcomes must be inferred under operational constraints and where satellite-based monitoring cannot reliably resolve early post-restoration signals at the scale of individual saplings. This study addresses the problem by developing a UAV data-fusion workflow that integrates UAV laser scanning (ULS), UAV multispectral imagery (UAV-MS), and ultra-high-resolution UAV-RGB observations for sapling-level vitality assessment. The workflow was tested in two restored pipeline corridor sites in the central Apennines (Italy), Ponte Baffoni (4.6 ha) and Ca' Romano (1.4 ha), surveyed in May 2025. ULS data were used to detect and geolocate individual saplings, UAV-MS data were used to extract vegetation-index metrics (NDVI, GNDVI, NDRE), and UAV-RGB imagery supported plot-level expert validation. A PCA-based soft-labelling strategy generated proxy vitality labels, which were then used to train a Random Forest classifier to derive corridor-scale probabilistic maps of sapling vitality, subsequently expressed as ALIVE, DEAD, and UNCERTAIN classes. Random Forest classification achieved balanced accuracies of 0.78 and 0.83, respectively. The resulting corridor-scale maps suggested mortality rates of 48.9% in Ponte Baffoni and 40.0% in Ca' Romano. These results suggest that multi-sensor UAV fusion can provide spatially explicit, sapling-level indicators of restoration performance, complementing field surveys and supporting operational post-construction assessment in narrow restoration corridors.

Automatic building reconstruction is a strategic component for creating urban Digital Twins (DTs), enabling the generation of accurate and interoperable Level of Detail 2 (LOD2) models. These models provide an essential standard for applications such as Geographic Information Systems (GIS), energy and hydraulic simulations, and urban planning. To address these needs, the MEDUSA (MEDiterraneo: Uso Sostenibile dell’Ambiente) project, promoted by the University of Salerno and funded by the Italian Space Agency (ASI), developed an innovative pipeline. The method was optimized to model areas with complex geometries and articulated roofs, utilizing the Amalfi Coast as a test area. The developed workflow is based on the City3D algorithm, integrating LiDAR (Light Detection And Ranging) data with building footprints derived from the Regional Topographic Database (RTDB). The process involves point cloud segmentation to isolate buildings and the generation of a Triangulated Irregular Network (TIN) mesh. Roof contours are identified using edge detection operators, simplified into polylines, and regularized using geometric constraints like parallelism and orthogonality to ensure LOD2 compliance. Finally, polygons are vertically extruded and optimized through the PolyFit framework, ensuring closed and topologically correct polygonal models. To overcome computational challenges and LiDAR data variability, significant improvements were introduced, including process parallelization, alignment with the Digital Terrain Model (DTM), and batch management of GeoJSON files. These enhancements successfully increased the pipeline's robustness and efficiency. The enriched pipeline produces high-quality LOD2 models, laying a solid foundation for next-generation urban modeling capable of meeting the scalability and interoperability requirements of future smart cities.

The identification and systematic recording of industrial buildings pose significant challenges for modern monument preservation. In particular, system halls have shaped the industrial landscape since the 19th century but often elude complete documentation because of their widespread distribution. These buildings serve as vital witnesses to technical innovations and economic transformation; however, assessing their architectural value requires a comprehensive inventory to determine the rarity or preservation state of specific building types.

Deep learning (DL) approaches are commonly used for the automatic recording of these buildings in aerial photographs, where the primary obstacle is the scarcity of curated training datasets. We overcome this by employing generative AI, specifically Stable Diffusion (SD), to produce synthetic data. By fine-tuning the SD model with Low-Rank Adaptation (LoRA), we successfully replicate the appearance and textures of various hall types. To resolve the spatial incoherence and geometric inaccuracies inherent in standard text-to-image generation, we integrated ControlNet. This allows for precise structural grounding using semantic masks, where specific colors represent building types, and polygon shapes define their exact locations. The resulting model generates accurate synthetic samples that maintain both spectral authenticity and an accurate spatial layout. Their usability was assessed by training a building detection model on both the real and synthetic datasets, achieving 71.9 and 66.7 mIoU, respectively. Moreover, introducing a few real samples for validation during training increased the mIoU to 82.7. The detection results demonstrate that the synthetic dataset is a reliable source for training, yielding robust generalization.

The PRISMA hyperspectral narrow-band data covering part of the Jind district during the Kharif Season 2024 were acquired to discriminate between two rice varieties, namely High-Yielding Variety (HYV) and Aromatic Basmati. In this study, hyperspectral bands were selected from the 240 hyperspectral bands of PRISMA data using Selective Principal Component Analysis (SPCA), which is specifically useful for crop classification. The subset of 9 PRISMA hyperspectral bands corresponding to the Sentinel-2 MSI bands was selected for rice crop classification and variety discrimination. The main difference between PCA and SPCA is that SPCA chooses only a subset of bands depending on the desired objectives of the study. The first three principal components (PCs) explained over 98 % of the variance of all spectral bands. The scatter plots of PC-1 and PC-2 indicated that there is a clear distinction between HYV and Basmati rice varieties.

In the present analysis, narrow-band hyperspectral red-edge group indices, such as Ratio Vegetation Index (RVIs), Green Normalised Difference Vegetation Index (GNDVI), and Chlorophyll Green Index (Clgreen), were generated to study their effectiveness for rice variety discrimination. The Spectral Angle Mapper (SAM) algorithm was used for supervised classification, and the results were validated using the time series S1 and S2 classified data. The results of validation indicated that using single-date hyperspectral data with 30 m spatial resolution, it was possible to discriminate between Basmati and HYV rice; however, it was not possible to discriminate between traditional and evolved Basmati rice varieties.

Real-time visualization of cadastral information through augmented reality (AR) has emerged as a significant challenge for public authorities in recent years. This paper addresses the potential, usage, and challenges of AR in the public sector. The prototype developed for this study demonstrates the visualization of geospatial data from ALKIS (Amtliches Liegenschaftskatasterinformationssystem, engl. Authoritative Real Estate Cadastre Information System), visualizing the boundaries and points of parcels in AR. Field tests conducted within this study assess the accuracy and usability of the AR visualization. As part of the study, existing AR libraries and frameworks were evaluated to select the most suitable platform for the prototype. The research underlines the potential of AR for geospatial applications, although it points out current precision limitations in the absence of external GNSS (Global Navigation Satellite System) receivers. The outcomes demonstrate the capabilities of AR visualization in a geospatial context and provide concrete approaches for optimizing future applications and research initiatives.

Modelling old buildings according to BIM standards is challenging, as historical architecture often features complex geometries and subject-specific information that is difficult to classify. This applies also to historic timber roof structures. The geometric complexity of historic timber structures makes them laborious and time-consuming to model using standard 3D software. In the case of aged heritage wooden beams, a lot of additional information should be parameterised. This information is derived from optical analysis as well as timber geometry and surface features, what is usually omitted in Open BIM.

In this paper we demonstrate a pipeline of data transfer from smartphone-based interface analysing automatically wood strength factor to BIM. This prototype interface allowing wood knottiness estimation for assessment of unknown strength values by aged heritage timbers as well as information connection to BIM framework.

The exponential growth of remote sensing data in recent years has underscored the need for intelligent, fast, and user-friendly analytical tools. Despite advancements in platforms such as Google Earth Engine and ENVI, the computation of spectral indices still demands specialized expertise, considerable time, and complex parameter tuning. This study aims to reduce the complexity of spatial data analysis and enhance its accessibility for non-expert users by developing an intelligent system capable of transforming simple natural language commands into automated remote-sensing index calculations. The main innovation lies in integrating Large Language Models (LLMs) with geospatial processing to establish a lightweight, multilingual, and fully automated framework capable of identifying index types and selecting appropriate spectral bands from Landsat data. The system was implemented using the Bloomz-560m language model in combination with open-source image-processing engines and deployed as a web-based interface. Experimental results over Tehran demonstrated that the model outputs were highly consistent with those generated by Google Earth Engine and ENVI, achieving an RMSE of 0.016 and a correlation coefficient of R² = 0.957. The total processing time was under 45 seconds, with the entire workflow executed automatically without user intervention. By simplifying the analytical process and significantly reducing computation time, this framework represents a crucial step toward democratizing remote sensing and spatial analysis. It can be effectively applied to urban surface heat island (SUHI) monitoring, water resource management, and precision agriculture applications.

Global Earth Observation provides coarse LCLU maps, classifying complex urban areas as a single Built-Up class. This limits urban modeling and product validation. Conversely, local SLAM offers fine-grained semantic detail but suffers from large-scale drift and lacks a global coordinate system.

We introduce Urban-Graph, a novel AI fusion framework to bridge this gap. Our system centers on a semantic scene graph to manage multi-scale information. It fuses three data sources: satellite imagery as a global prior, vehicle-based SLAM for local semantic detail, and fixed roadside infrastructure for high-precision GNSS anchors. A factor graph optimizer integrates these local, global, and anchor constraints.

This process generates a large-scale, globally-consistent, and geospatially-anchored semantic map. This resulting graph serves a dual purpose. It provides a drift-free map for local systems and functions as a scalable, high-fidelity ground-truth product to automate the fine-grained validation and decomposition of coarse urban LCLU classes.

High-resolution 3D documentation of cultural heritage sites is essential for their preservation. While terrestrial laser scanning (TLS) remains the gold standard, it is often cost-intensive compared to photogrammetry. This study evaluates three image-based reconstruction techniques, Multi-View Stereo (MVS), Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), by applying them to a complex scene featuring a chapel and its surrounding vegetation, sensed from an uncrewed aerial vehicle (UAV). A hybrid TLS/MVS model provides a high-accuracy reference. Using identical interior and exterior camera parameters of the 105 UAV-acquired images, we generate dense point clouds with all methods and assess geometric accuracy and completeness using the M3C2 algorithm. Results show that MVS achieves superior accuracy (standard deviation of all M3C2 distances: MVS = 0.11 m, NeRF = 0.15 m), whereas NeRF attains up to 20% higher completeness, particularly in low-texture and vegetation-occluded regions. The 3DGS point cloud was deemed too sparse and was therefore not used for further analysis. The study highlights the potential of NeRFs to recover partially occluded or sparsely textured geometries that are challenging for MVS and suggests a complementary use of both approaches for cost-efficient documentation of cultural heritage.

This study presents a machine learning-driven approach to forecast forest fire risk in California’s high-risk regions, aiming to predict fire-prone areas one month in advance. By integrating static topographical features with dynamic vegetation indices—such as NDVI, NDWI, GPP, and LAI—and their derivative components like trend and Exponential Moving Average (EMA), the model captures critical indicators of vegetation health and moisture. Among several algorithms tested, Logistic Regression (LR) consistently outperformed others, achieving a validation AUC of 0.90 when combining static and dynamic features. A 12-month historical time window proved most effective, enabling the model to learn seasonal and long-term vegetation patterns. Validation on independent datasets showed promising results for 2021 (AUC 0.84), though performance dropped in 2024 (AUC 0.64), likely due to satellite data shifts and ecological changes. These findings underscore the importance of long-term vegetation monitoring and robust feature engineering for accurate fire risk prediction. The study offers a practical tool for early warning systems, while highlighting the need to address data variability and environmental dynamics for sustained performance.

This work showcases mapping a natural disaster using social media posts of users (tweets) during the ongoing event. We have finetuned an encoder-decoder model and created a model that detects toponyms from tweets very efficiently. Toponyms are then resolved to geographical coordinates and features so that temporal heatmaps can be created effectively mapping the natural disasters through social media posts.

Encoder-decoder models are generally used for machine translation or summarization tasks in NLP. We show that through finetuning with proper data, a lightweight encoder-decoder model deployed locally can generate comparable results to prompting web-deployed large language models.

High-resolution mapping of urban land cover and urban green infrastructure (IVU) is essential for medium-sized cities, where global datasets often fail to capture fine-scale patterns. This study presents a local-scale land-cover classification for Alta Gracia, Argentina. The approach integrates medium- and high-resolution imagery with object-based segmentation (SNIC) and Random Forest classification. Six vegetation indices (NDVI, EVI, SAVI, GNDVI, MSAVI, VARI) were used to enhance class separability, while PlanetScope mosaics and local orthophotos improve spatial detail. Accuracy was assessed using overall accuracy, Cohen’s Kappa, and F1 Score.

The resulting land-cover map was used to delineate and quantify urban green infrastructure. Green-cover areas were summarized across city-defined sectors. Results were compared with regional layers from IDECOR and the global Dynamic World product, showing that global datasets underestimate fine-scale vegetation and fail to capture small or fragmented patches. The high-resolution local map substantially improves spatial accuracy and IVU delineation, serving as a baseline for urban planning, green-space management, and climate-resilient strategies.

This study demonstrates the value of combining multi-source imagery, object-based methods, and machine-learning classification to refine local land-cover mapping and IVU assessment. The methodology is reproducible using open-source tools (Google Earth Engine, QGIS, and R) and transferable to other medium-sized Latin American cities with limited data availability. Future work will integrate LiDAR-derived canopy metrics and citizen science to validate and enhance local products.

This contribution links local mapping with broader land-cover/use frameworks, supporting the ISPRS ThS21 global–local dialogue and providing actionable evidence for sustainable urban development.

This work presents a hybrid camera calibration approach that combines the strengths of standard photogrammetric camera calibration with data-driven lens distortions correction. Conventional calibration methods, such as those based on Zhang’s model, estimate lens distortions by fitting polynomial functions to the calibration images coordinates. While these methods are well established, they may struggle to fully describe complex or setup-dependent distortions, particularly near image borders or under varying environmental conditions.

To address this, a learning-based model is introduced to directly predict the distortion coefficients from calibration images. The network is trained using real data, allowing it to capture lens- or condition-specific variations that conventional calibration may overlook. The predicted coefficients maintain the same format as those used in standard photogrammetric models, ensuring compatibility with existing calibration toolchains such as OpenCV or MATLAB.

The proposed approach, therefore, aims to automate the estimation of distortion parameters while preserving the interpretability and mathematical foundations of traditional models. Although the primary focus is on camera calibration, the method offers further advantages for optical metrology systems such as fringe projection, where accurate and consistent distortion compensation is essential for depth measurement reliability.

Digital twins play a crucial role in autonomous driving applications and transportation system simulations. The need for large scale and dynamic information has increased interest in generating urban digital twins from remote sensing data.

Aerial high resolution imagery of urban areas serves as the one of the most important data sources for this task. Advances in deep learning and machine learning allow more accurate and automated extraction of urban elements. In recent years, we have developed and integrated advanced deep learning models to extract various land cover types surrounding road networks, including buildings, roads, and vegetation. Furthermore, we have conducted proof of concept studies aimed at detecting and delineating linear landmarks from aerial imagery, including curbstones and road borders. These developments contribute to the creation of more accurate and detailed urban digital twins, which are essential for advanced urban analytics and intelligent transportation systems.

Results from the deep learning models are presented for the Schwarzer Berg district in Brunswick, Germany, which is a test region for the development of mobility services and technologies at the German Aerospace Center (DLR). The AI models are trained using benchmark datasets from other urban regions, indicating that the proposed approaches can be readily transferred and evaluated in other European cities.

In this work, we demonstrate our early results in geocoding oceangraphic research articles across Canada. Through the use of AI, we extracted locations out of the abstracts of research articles and then assigned a latitude and longitude to those works based off of the locations extracted. The geocoded works are then displayed. Our work allows a user to identify locates across Canada that are being actively researched and find research specialists of those locations. We intend to develop this tool further by collaborating with journalists.

Land use and land cover (LULC) classification plays a crucial role in addressing numerous real-world challenges. Hence, we proposed methodological advances in LULC classification from a data-centric artificial intelligence perspective, which prioritizes data quality as a key factor in improving machine learning performance. The main contributions include evaluations of novel approaches for: (i) constructing an accurately labeled dataset based on agreement among existing reliable maps; (ii) curating remote sensing data to improve accuracy, consistency, unbiasedness, relevance, diversity, and completeness; (iii) generating training samples that capture the spatial, temporal, and spectral dimensions of remote sensing data; and (iv) developing a deep learning model designed to leverage multidimensional features. The study evaluates a sample generation method grounded in reference map agreement and multidimensional feature extraction, along with a deep learning model that leverages these features, attaining high accuracy across all LULC classes and providing a robust basis for large-scale, data-centric LULC mapping.

Global forested areas are decreasing at a rapid rate, leading to environmental instability, altered climate patterns, and a decline in ecosystem services. In the present study, the Western Ghat (WG) region is one of the major forest resources in the Indian southern peninsula; it regulates/balances the weather conditions with the unique features of high-rise mountains and tall trees. This mountain chain is recognised as one of the world’s eight ‘hottest hotspots’ of biological diversity. These mountains cover an area of approximately 140,000 km² along a 1,600 km long stretch, traversing the states of Kerala, Tamil Nadu, Karnataka, Maharashtra, Goa, and Gujarat. This region is one of the richest biodiverse hotspots and biosphere reserves identified by UNESCO. The WG region is of immense global importance for the conservation of biological diversity and endemism. This region encompasses a number of protection regimes, ranging from Tiger Reserves, National Parks, Wildlife Sanctuaries, and Reserved Forests. The forests of the WG include some of the world's best representatives of non-equatorial tropical evergreen forests. Around 325 globally threatened species (IUCN Red List) occur in the Western Ghats, of which 129 are classified as vulnerable, 145 as endangered, and 51 as critically endangered.

This research introduces an innovative platform designed to enhance citizen engagement in urban planning and management by integrating emerging technologies such as Artificial Intelligence (AI), Large Language Models (LLMs), and chatbots. Traditional Public Participation Geographic Information Systems (PPGIS) often face challenges in effectively capturing and analyzing citizen input. This platform addresses these limitations by enabling users to articulate urban issues or ideas in natural language, which are then processed through AI-driven Natural Language Processing (NLP) techniques to identify key elements such as location, issue type, and intensity. Furthermore, the platform facilitates interactive dialogues, allowing citizens to inquire about perspectives from other community members, thereby fostering a dynamic exchange of views. In the absence of an initial user base, a dataset comprising 2,000 tweets related to Montreal's public transportation was curated. An LLM was fine-tuned using this data, equipping the model to respond to queries concerning Montreal's public transportation system. The findings demonstrate the feasibility of leveraging AI and LLMs to create a responsive and interactive platform that not only streamlines data collection but also enriches the participatory planning process. This approach has the potential to transform urban governance by making it more inclusive and data driven.

Remote Sensing Image-Text Retrieval (RSITR) aims to retrieve target textual descriptions from the gallery images, and vice versa. RSITR faces the key challenge of establishing accurate alignment between two heterogeneous modalities. Existing methods typically assume that image-text pairs are semantically aligned, where each textual description corresponds to a single image. However, this assumption does not always hold because factual errors in textual descriptions lead to incorrectly-correlated relationships. Moreover, some samples exhibit weakly-correlated relationships, i.e., an image corresponds to multiple similar texts. These incorrectly- and weakly-correlated relationships hinder effective cross-modal alignment. To address these challenges, we propose the Robust Dual Embedding Alignment (RDEA) network, which improves the robustness of cross-modal alignment by jointly learning both instance-level and feature-level correspondence between image and text modalities. Firstly, we propose an Incorrectly-Correlated Feature Rectification (ICFR) module, which employs a dynamic margin-guided mechanism to adaptively balance original and auxiliary descriptions generated by a large language model, guiding the model to learn correct image-text correspondences at the instance-level. Secondly, a Weakly-Correlated Feature Decoupling (WCFD) module constructs modality-specific intermediate features via learnable distributions, which decouple overlapping semantics across modalities. These intermediate features enable the model to distinguish semantically similar texts, thereby establishing more discriminative and accurate image-text correspondences at the feature-level. We conduct extensive experiments on benchmark datasets, demonstrating that our approach outperforms state-of-the-art methods.

This contribution presents an operational framework that advances BIM–SAR fusion into a commercial, API-based Digital Twin service for building deformation monitoring. Building on the BIMSAR research project, the system integrates multi-frequency MTInSAR results from Sentinel-1, TerraSAR-X, and PALSAR-2 with IFC-based BIM models to provide semantically structured deformation indicators for individual building components. Persistent and distributed scatterer analyses generate millimetre-scale deformation time series, which are stored in a harmonized database and exposed through a RESTful API that supports standardized queries for deformation values, risk metrics, and metadata. A pilot implementation in Ahlen, Germany, demonstrates the service’s interoperability with existing digital twin platforms and validates the workflow using previously established BIMSAR datasets. Developed jointly by EFTAS Remote Sensing and EnviroTrust, the system showcases the successful transition of research-driven BIM–SAR fusion methods into an operational, cloud-ready monitoring service supporting resilient building and infrastructure management.

Accurate tree species classification is fundamental to forest ecology, biodiversity monitoring, and sustainable resource management. However, large-scale species-level labeling in remote sensing remains challenging due to the need for expert annotation and the limited generalization of supervised models. This study introduces TreeCLIP, an unsupervised framework that adapts the CLIP vision–language model for ecological analysis through multi-view feature fusion.

TreeCLIP renders each individual tree point cloud into multiple orthogonal 2D projections that capture its geometric and morphological characteristics. CLIP’s pre-trained image encoder extracts visual embeddings from each view, which are then L2-normalized and fused into a unified multi-view representation. By applying clustering methods such as K-means and DBSCAN, TreeCLIP achieves species-level grouping without any manually defined textual prompts or labeled training data.

Experiments on multi-platform airborne laser scanning datasets from German forest stands demonstrate that TreeCLIP surpasses traditional machine learning approaches (e.g., Random Forest, SVM) and achieves accuracy comparable to supervised deep models. The results highlight CLIP’s capacity to generalize across domains and reveal the potential of foundation models for fine-grained ecological recognition.

TreeCLIP provides a scalable, annotation-efficient framework for large-scale forest inventory and vegetation monitoring, bridging the gap between general-purpose vision–language models and domain-specific ecological applications.

Construction sites are complex, dynamic environments that demand accurate, real-time monitoring for progress and safety management. Traditional on-site supervision and image-based UAV monitoring often fall short in providing detailed and timely 3D information. Recent digital twin technologies offer virtual replicas of construction sites, but existing 3D reconstruction methods—typically relying on LiDAR or depth cameras—remain limited by high hardware costs, heavy energy consumption, and extensive manual annotation requirements.

This study investigates the feasibility of applying 3D Gaussian Splatting (3DGS) for 3D scene reconstruction and segmentation in digital twin–based construction monitoring. Leveraging only visual inputs, 3DGS enables high-fidelity modeling while avoiding costly hardware. Combined with foundation models such as the Segment Anything Model (SAM), it supports unsupervised or weakly supervised segmentation adaptable to continuously evolving site conditions. Moreover, integrating 3DGS with large vision–language models allows for interactive segmentation through clicks or natural language prompts, advancing toward intelligent and adaptive digital twins.

We evaluate several Gaussian-based segmentation algorithms on construction-related datasets, assessing their effectiveness in capturing structural details and object semantics. Results show that 3DGS-based methods achieve promising segmentation quality for simple geometric objects but face challenges in complex, cluttered environments. These findings highlight both the potential and current limitations of 3DGS in realizing fully automated, adaptive digital twins for smart construction management.

EarthDaily FM is a foundation model purpose-built for high-frequency change detection and short- to medium-range forecasting across global Earth Observation (EO) time series. It is designed around the forthcoming EarthDaily Constellation (EDC)—a systematic, near-daily mission with 22 VNIR, SWIR, and LWIR bands engineered for AI-ready analytics, high geolocation and radiometric accuracy, CEOS ARD compliance, and spectral compatibility with Sentinel-2 and Landsat. This design enables a single self-supervised model to fuse years of historical S2/Landsat data with new daily EDC observations, closing the temporal gap that constrains existing EO foundation models focused on static scene understanding.

Preliminary experiments using open and proxy datasets demonstrate the model’s capability for diverse forecasting tasks, including harvest date prediction, crop yield estimation, and soil moisture retrieval. Using VENµS imagery as a proxy for EDC’s cadence and 5-m resolution, the model achieves low median errors in harvest date prediction at 50–60-day lead times, while multimodal training with meteorological and radar inputs improves soil moisture estimation. The impact of incorporating EarthDaily Constellation data on forecasting accuracy and model generalization will be demonstrated as new observations become available.

EarthDaily FM represents a practical step toward operational, time-aware EO modeling—integrating optical, radar, and weather data to support forecasting in agriculture, water resources, and environmental resilience.

We propose an innovative method to improve the reliability of Planet Fusion surface reflectance during periods of extended cloud cover. Planet Fusion offers daily, 3 m, cloud-free data (RGB-NIR) by radiometrically harmonizing all available PlanetScope imagery using the CESTEM algorithm, which employs MODIS/VIIRS and FORCE data for correction, and then uses a spatially and temporally driven gap-filling algorithm to ensure spatial completeness. A critical weakness arises during prolonged cloudiness, where the certainty of Planet Fusion's gap-filled pixels diminishes.

The proposed research directly addresses this weakness by incorporating Sentinel-1 synthetic aperture radar and AMSR-2 brightness temperature data. Both Sentinel-1 and AMSR-2 operate in the microwave spectrum, guaranteeing data acquisition regardless of weather or light conditions. By fusing these multi-sensor, multi-modal datasets into the Planet Fusion workflow, we are able to improve the accuracy of gap-filled pixels during months-long periods of persistent cloud cover. This work not only seeks to increase the reliability of the Planet Fusion product, but also advances the field of multi-modal data fusion, highlighting its necessity for uninterrupted, observation-driven monitoring of land surface change from space.

This work presents the development of a web‑based interface designed to support both on‑site and remote exploration of environmental sensor deployments.

The growing accessibility and standardisation of IoT technologies have led to their adoption across diverse fields, including environmental studies, urban planning, architecture, agriculture, archaeology, and museum studies, yet shared challenges persist around planning, deployment, interpretation, and communication of sensor data. When multiple disciplines operate within the same test environment, their activities can affect one another, highlighting the need for interfaces that reduce disciplinary barriers and rely on spatially grounded visualisation rather than domain‑specific terminology. The system builds on principles of Situated Analytics, enabling data to be interpreted directly within its spatial or contextual setting while also supporting remote interaction through proxy representations of real‑world environments.

In this contribution, three modelling techniques, dense point cloud, 3D Tiles, and Gaussian Splatting, were generated from drone images and integrated into a Babylon.js platform. A WebAR application, developed with 8th Wall, allowed sensor locations to be placed in situ, with data visualised through a shared information layer using MQTT to stream live or simulated readings.

The results indicate promising developments for cross‑disciplinary knowledge exchange through accessible, device‑agnostic web tools. Ongoing work explores the improvements to point‑cloud handling, AR localisation accuracy, and the long‑term collection of historical environmental data.

To address the common deterioration of ancient Chinese murals—including pigment loss, texture blurring, and color fading—this paper proposes a deep learning-based approach integrating multi-scale attention and texture enhancement modules for high-fidelity virtual restoration. The model employs a multi-scale attention mechanism to maintain structural continuity and a dedicated texture enhancement module to recover fine details often lost in conventional methods. The restoration process consists of three stages: multi-scale feature extraction using partial convolutions, feature reconstruction that transfers statistical properties from intact regions, and a texture refinement module for detail completion. Evaluated on the Dunhuang mural dataset, the method outperforms existing techniques in PSNR, SSIM, and FID scores, producing visually coherent and stylistically consistent results. This approach offers a scalable and adaptable solution for digital conservation, supporting customizable restoration levels tailored to various degrees of damage.

Climate change and population growth intensify the demand for precise agriculture mapping to enhance food security. Such mapping tasks require robust modeling of multi-scale spatiotemporal patterns from fine field textures to landscape context, and from short-term phenology to full growing-season dynamics. Existing methods often process spatial and temporal features separately, limiting their ability to capture essential agricultural dynamics. While transformer-based remote sensing foundation models (RSFMs) offer unified spatiotemporal modeling ability, most of them remain suboptimal: they either use fixed windows that ignore multi-scale crop characteristics or neglect temporal information entirely. To address these gaps, we propose AgriFM, a multi-source, multi-temporal foundation model for agriculture mapping. AgriFM introduces a synchronized spatiotemporal downsampling strategy within a Video Swin Transformer backbone, enabling efficient handling of long and variable-length satellite time series while preserving multi-scale spatial and phenological information. It is pre-trained on a globally representative dataset comprising over 25 million samples from MODIS, Landsat-8/9, and Sentinel-2 with land cover fractions as pre-training supervision. AgriFM further integrates a versatile decoder specifically designed to dynamically fuse multi-source features from different stages of backbone and accommodate varying temporal lengths, thereby supporting consistent and scalable agriculture mapping across diverse satellite sources and task requirements. It supports diverse tasks including agricultural land mapping, field boundary delineation, agricultural land use / land cover mapping, and specific crop mapping (e.g., winter wheat and paddy rice) with difference data sources. Comprehensive evaluations show that AgriFM consistently outperforms the general-purpose RSFMs across multiple agriculture mapping tasks.

An AI-driven framework for structural identification and stock analysis of bamboo scaffold systems to enable lifecycle management for firms, regulators, and workers. The method addresses irregular geometry, dense packing, and occlusions through three components. First, Node-guided Pole Fitting detects bamboo nodes and poles; the Bamboo of Building dataset trains a neural network to generate a Node Candidate Set. Within each node’s bounding box, Line Segment Detector (LSD) extracts linear features; representative segments are clustered, connected, and curve-fitted to model a pole. Second, multi-view 3D reconstruction maps the scaffold; cross-image matching projects poles into a unified space, refining NCS into Real Node Set and Fake Node Set for reliable topology. Third, a digital model estimates member lengths/diameters to quantify stock and potential CO2 reductions.

Agricultural drought is one of the most damaging natural hazards, causing ecological disruption, economic losses, and reduced crop yields. Recent extreme droughts in Central Europe, particularly after 2018, have underscored the need for reliable, spatially explicit drought monitoring. Traditional ground-based indices often fail to capture crop-specific physiological responses, while commonly used remote-sensing indicators, such as NDVI, are limited by soil background effects and saturation in dense vegetation. Sun-Induced Chlorophyll Fluorescence (SIF) directly reflects plant photosynthesis and responds sensitively to water and heat stress, making it a promising alternative for drought assessment.

Despite its potential, SIF-based drought monitoring remains largely unexplored in Germany. Most studies focus on specific regions or individual crops and rely on other remote-sensing indices rather than SIF. To fill this gap, this study evaluates whether multi-temporal SIF data can detect agricultural drought signals across Germany and how consistently these signals relate to crop yield anomalies. Using the Soil–Climate Regions (SCRs) of Germany as an ecologically meaningful spatial framework, we examine spatial correlations between SIF and yield across SCRs, and compare time-series SIF anomalies with average yield anomalies.

This research highlights the potential of SIF as an early and robust indicator of agricultural drought, offering insights for improved drought monitoring and crop management strategies in Germany.

Vision-based displacement measurement offers a promising pathway toward UAV-enabled structural monitoring, where contact-free, lightweight, and rapidly deployable sensing is essential. However, existing vision approaches typically estimate only 2D motion or require model training, artificial markers, or complex calibration, which hinders their applicability on real structures.

To address these limitations, this paper presents a new training-, marker-, and calibration-free vision-based framework designed with future UAV deployment in mind for structural 3D displacement measurement. Leveraging the reasoning capability of a state-of-the-art vision foundation model, the proposed method achieves millimeter-level 3D displacement accuracy without any scene-specific training, calibration, or fine-tuning.

To support rigorous evaluation, we establish a compact multi-modal dataset collected from two full-scale bridges, including synchronized stereo videos, accelerometer measurements, and an evaluation protocol. Experiments on real bridges demonstrate that the proposed framework delivers accurate, robust, and practical in-situ 3D displacement measurement under uncontrolled field conditions.

The system is inherently suited for airborne visual sensing, and integrating the framework with UAV-based data acquisition constitutes the next step of this research.

Invasive pests such as the desert locust are both detrimental to people and the environment. The desert locust is documented as one of the most destructive polyphagous plant pests. This study, about the integration of crowd-sourced field dataset and Google Earth Engine (GEE) satellite data, demonstrates how community-based initiatives and freely available cloud-based earth observation resources can be used to provide innovative, evidence-based and data-driven decision support insights that are of critical use to government agencies in desert locust crisis management. The study integrated 160,810 desert locust field survey records collected from January 2020 to December 2021, with vegetation and water indices time-series computed from Sentinel 2 bands B2, B3, B4, B8 and B11 on GEE. The results indicate that the peak of desert locust mature adult (67) and hopper (75) incidents coincided with the highest spectral index values in June 2020. However, the peak of desert locust immature adult (70) incidents in February 2021 coincided with low spectral index values. This means that spectral indices can be used to identify suitable breeding areas for desert locusts, but may not reliably identify all the areas where the pest might be present. Among the assessed indices, the Modified soil adjustment vegetation index (MSAVI) produced the best prediction with a β=0.703, t=6.983 and p=<0.001. The study concludes that, because Hotspot 1 denotes arid and semi-arid lands (ASAL), MSAVI would be the most suitable for monitoring desert locusts in this area, as the index accounts for soil brightness in the deserts.

This paper reviews the evolution, achievements, and future direction of remote sensing across the Great Lakes Basin (GLB), emphasizing the unique binational collaboration between the United States and Canada. Beginning with post–World War II aerial photography, remote sensing in the region rapidly expanded through pioneering work in forestry, water quality mapping, and early satellite-based observation. The formation of the Great Lakes Alliance for Remote Sensing (GLARS) marked a major step toward coordinated, cross-border environmental intelligence. Enabled in part by the Great Lakes Restoration Initiative (GLRI), GLARS brought together federal agencies, universities, and private partners to deliver high-resolution, multi-temporal products supporting natural resource management.

Key achievements include production of 2-meter digital surface models for the entire basin using petascale computing; integrated optical and SAR approaches for dynamic wetland mapping; multi-year RADARSAT-2 monitoring of seasonal wetland saturation; InSAR applications for water-level change detection; and successful classification of invasive species such as Phragmites australis using multi-sensor datasets.

Looking ahead, the paper identifies priorities such as harnessing new SAR missions (RCM, NISAR), expanding daily high-resolution multispectral monitoring, building fully automated analysis pipelines, and formalizing binational data-sharing systems. Continued integration of AI, cloud computing, and stakeholder-driven design is essential for climate-resilient management of the world’s largest freshwater system.

This work details the development of a comprehensive national assessment application for rooftop solar photovoltaic (PV) potential in Israel, designed to support the national target of 30% renewable electricity generation by 2030. Faced with limited land and increasing electricity demand, Israel's policy prioritizes PV installations on existing building rooftops.

The technological approach integrates solar radiation modeling, Deep Learning (AI) obstacle segmentation, GIS, and governmental data. The system utilizes advanced models incorporating DSM data, shading, and meteorological variables to calculate solar radiation. Crucially, multiple Convolutional Neural Network (CNN) models (U-net, Mask RCNN) were trained on high-resolution aerial imagery to accurately segment and deduct rooftop obstacles, such as existing PV systems, solar collectors, and vegetation, achieving over 95% IoU.

The final assessment feeds into a two-pronged system:

A Public Application allowing citizens and businesses to receive address-specific estimates of usable roof area, expected electricity production, and economic return on investment.

A National Management System and Dashboard for policymakers and local authorities, enabling spatial examination, progress monitoring, and data-driven strategy formulation (e.g., targeted encouragement campaigns).

This multi-level system, combining remote sensing, machine learning, and governmental data, provides an adaptable, data-driven framework for facilitating the renewable energy transition across all stakeholder levels.

Low-altitude remote sensing using unmanned aerial vehicles (UAVs) has become a crucial method for large-scale 3D reconstruction in various applications, including urban planning, environmental monitoring, and disaster management. However, due to issues such as proportion blurring, projection distortion, and failed loop closure, obtaining precise and dense 3D point cloud maps from monocular RGB cameras remains challenging. Recent advances in feed-forward 3D scene reconstruction, such as VGGT (Visual Geometry Grounded Transformer), which generates dense point clouds and camera poses from uncalibrated RGB images, offer potentially promising solutions. VGGT-SLAM extends this capability to large-scale scenes by aligning local submaps optimized on the SL(4) manifold, which addresses projective ambiguity that similarity transformations (Sim(3)) cannot resolve. The enhanced large-scale reconstruction capability of VGGT-SLAM is precisely what is needed for 3D reconstruction of remote sensing datasets. This study investigates the feasibility of applying VGGT-SLAM to UDD (Urban drone datasets) and highlights its limitations in real-world scenarios.

Accurate LiDAR–camera extrinsic calibration is crucial for reliable multi-sensor fusion in robotics, autonomous navigation, and UAV photogrammetry. This study presents a robust two stage LiDAR–camera calibration framework that integrates geometric and monocular depth assisted information constraints within a unified joint optimization scheme.

In the initial stage, geometric features from both LiDAR and camera views are extracted and aligned via Singular Value Decomposition (SVD) to provide stable initialization. The refined stage introduces a hybrid optimization that combines spatial distance constraints with a Normalized Mutual Information Distance (NID) term between LiDAR-measured depth and monocular depth estimation (MDE) results. The deep learning–based MDE provides dense and metrically consistent depth maps, effectively bridging the modality gap between 3D point clouds and 2D images. This dual-constraint formulation enhances calibration robustness against LiDAR sparsity and texture deficiencies. Experimental evaluations using a circular calibration target demonstrate mean rotational errors below 0.3° and translational errors under 3 cm, surpassing traditional FastCalib methods. Qualitative visualizations further confirm precise alignment between LiDAR projections and image contours. The proposed framework eliminates the need for precise calibration targets and manual initialization, achieving automatic, high-accuracy extrinsic calibration adaptable to complex outdoor environments

Extreme rainfall from Hurricane Helene (September 2024) triggered widespread landslides across the southern Appalachian region, highlighting the need for rapid landslide susceptibility assessments that capture both landslide initiation and downstream runout. Traditional susceptibility models often focus solely on initiation zones, limiting their ability to identify which slopes will generate destructive landslides or where material will travel. This study addresses that gap by (1) integrating Geographic Information System (GIS)-based machine learning susceptibility modeling using ArcGIS Pro: Maximum Entropy (MaxEnt) and Random Forest-Based and Boosted Classification and Regression (FBBC) and (2) the U.S. Geological Survey (USGS) Grfin (Growth, Flow, and Inundation) runout toolbox. The study focuses on the Nolichucky River Gorge in eastern Tennessee and western North Carolina, where intense rainfall (4-20 in;10.1-50.8 cm) triggered numerous shallow landslides. Results provide a framework for emergency response along TN-107 and US-19W corridors, infrastructure vulnerability assessments, and hazard planning in Unicoi and Carter counties.

The Surveying and Mapping Authority of the Republic of Slovenia (GURS) carried out nationwide airborne laser scanning project (CLSS) between 2023 and 2025, with a minimum spatial resolution of ten points per square metre across the entire territory of Slovenia. In 2025, the project for automated building extraction from the acquired LiDAR data was initiated, with the objective of systematically processing approximately one third of Slovenia’s territory per year. The automatically extracted building data (2.5D building footprints and 3D building models) will serve as a fundamental topographic dataset, a key source for detecting and monitoring changes in the Real Estate Cadastre, and a foundational dataset for property valuation at scale. Moreover, this initiative represents a pivotal step towards the establishment of a geospatial digital twin of Slovenia. The production workflow is based on an integrated processing method that combines a classified LiDAR point cloud (GKOT) and True Orthophoto imagery (POF) from CLSS. The quality evaluation is conducted in accordance with the international standard ISO 19157 — Geographic Information — Data Quality.

Climate change is expected to significantly alter environmental conditions in southern Sweden, increasing the risk of natural hazards such as wildfires. This study assesses wildfire susceptibility in forest areas of Scania under projected climate conditions corresponding to the Representative Concentration Pathways RCP8.5 scenario. Using Geographic Information Systems (GIS) and a fuzzy multicriteria decision analysis (MCDA), climatic variables (temperature, precipitation, wind speed) and forest type data were integrated to generate a continuous fire risk map. Forest types were reclassified based on fire susceptibility, and fuzzy membership functions were applied to climatic variables, with a fuzzy gamma overlay (γ = 0.6) used to combine criteria. Results indicate that several coastal and fragmented forest areas exhibit high wildfire risk, while northern inland regions show relatively lower susceptibility. The fuzzy approach enables a nuanced representation of risk gradients, providing valuable spatial information for climate adaptation and hazard mitigation planning. Despite limitations in input data and parameter quantification, the produced map highlights priority areas for monitoring and management under future climate scenarios.

The debut of the Constellation Optique en 3D (CO3D) in July 2025 represented a significant advancement in Earth observation. This state-of-the-art satellite mission captures the earth in breathtaking three dimensions by using four satellites in a novel out-of-phase tandem arrangement that mimics mammalian vision. CO3D produces high-resolution Digital Elevation Models (DEMs) at one-meter grid spacing with previously unheard-of accuracy—one-meter relative height precision and four-meter absolute height precision.

Synchronous stereo imaging enables tracking of moving objects even in the dark, and each CO3D satellite provides 0.50-meter resolution images in the red, green, blue, and near-infrared bands. This innovative technology, which offers cutting-edge capabilities for coastal monitoring, disaster response, urban planning, and climate research, helps the scientific, defense, and civil communities equally.

Applications for CO3D are numerous, ranging from improving post-disaster evaluations and urban resilience to tracking glaciers and coastal erosion. CO3D enables governments, businesses, and researchers to tackle important issues with unmatched accuracy by offering almost worldwide 3D data. Welcome to the era of CO3D, the future of Earth observation.

The oil slick look-alike challenge occurs when natural ocean phenomena reduce synthetic aperture radar (SAR) return in the same backscatter range as mineral oil. We revisit this challenge through the lens of geospatial foundation models (FMs), large neural networks which are a current frontier in automatic, deep learning-based mapping methods. In their benchmark evaluations, FMs promote state-of-the-art performance across a wide range of downstream tasks including segmentation. In contrast, our findings suggest that, in their current state, FMs do not outperform other neural network backbones for segmentation in an unconventional remote sensing modality such as SAR imaging of oceans. Surprisingly, backbones that were partly pretrained on SAR data do not show improved segmentation over those pretrained on natural images (here ImageNet). Rather than improving model backbones for segmentation, we argue that the breakthrough made by FMs may well lie elsewhere, such as in data management and pruning techniques. We make available the dataset used in our experiments, consisting of Sentinel-1 IW images annotated for semantic segmentation of oil slicks.

This study will evaluate whether the analysis-ready, global, cloud-free, annual, 10 m resolution embedding field layers of the Google AlphaEarth and Tessera foundation models can be used to improve estimation and prediction of biophysical variables such as live fuel moisture content, as well as contributing to an understanding of the global transferability of developed models to different regions.

Quantifying biodiversity intactness—a central indicator of ecosystem health and resilience—remains difficult across Africa due to scarce standardized baseline data and limited biodiversity monitoring. Traditional indicators based on vertebrates or vegetation provide only partial insights, as they respond more slowly to environmental change and have limited spatial coverage.

This study presents a novel, continent-wide framework that integrates multi-sensor Earth Observation (EO) data (Sentinel-2, GEDI, and TerraClimate) with extensive in situ insect occurrence records to derive an insect-based biodiversity intactness index (IBI). Insects, which dominate terrestrial biodiversity and respond rapidly to microclimatic and habitat changes, are used as sensitive ecological proxies for ecosystem condition. Their ubiquity and fine-scale environmental sensitivity make them particularly suited to detect patterns of habitat degradation and recovery that other taxa may overlook.

By coupling EO-derived indicators of vegetation structure, productivity, and climatic variability with insect diversity models, the framework provides spatially explicit, continuous estimates of ecosystem integrity across Africa. The resulting IBI fills a major information gap in biodiversity monitoring by offering a harmonized, scalable, and policy-relevant assessment tool.

The approach directly supports reporting needs under the Kunming–Montreal Global Biodiversity Framework (GBF) and African Union ecosystem restoration goals. It demonstrates how EO and biodiversity data integration can operationalize continent-wide monitoring of ecosystem condition—helping countries to track progress toward conservation and sustainable land-use targets through an ecologically grounded, insect-based lens.

Tree-species classification from multispectral remote sensing has advanced rapidly with the improved spatial and spectral capabilities of sensors such as Sentinel-2, enabling accurate discrimination of forest taxa across large areas. This paper deals with two approaches for tree species classification at the national scale using multi-temporal S2 imagery. We compare a machine learning algorithm (LightGBM) and a deep learning transformer-based model (ForestFormer) to classify dominant tree species in Slovenia based on seasonal characteristics. The resulting classifications are validated against National Forest Inventory datasets, provided by the Slovenian Forestry Institute.

This work explores the practicalities of emulating a closed-source Earth embedding AI model from a large set of its pre-computed outputs. It also demonstrates how behaviour of a multi-modal multi-temporal embedding dataset can be probed using individual observational inputs. The framework is tested using Major TOM Core datasets with Sentinel-2 and Sentinel-1 data and an existing global dataset of AlphaEarth Foundations embeddings.

In an ever-changing global context, accurate and up-to-date land use and land cover (LULC) information becomes critical to understanding the dynamics of the Earth surface and managing natural resources. Nowadays, a common workflow for LULC classification involves training a supervised machine learning model using satellite image time series (SITS) and a collection of ground truth (GT) samples. Unfortunately, GT data are not always available across years due to costly and time consuming field campaigns or restrictions on field access. For this reason, the possibility of transferring a model learned on a particular year (with GT data available) to another mapping year (without GT data) has received traction, recently. To cope with such temporal transfer scenario, unsupervised domain adaptation (UDA) has been considered in order to address possible data distribution shifts originating from different acquisition conditions affecting mapping years. In recent years, self-supervised learning has emerged as a promising paradigm to mitigate the reliance on large amounts of GT data through the learning of general purpose and robust feature representations, enabling the development of geospatial foundation models (GFM) in Earth observation. GFM, trained on large volume of multi-modal geospatial data can provide embeddings that encode rich spatio-temporal, spectral, and semantic information. A notable example is AlphaEarth satellite embedding, released lately on a global scale and annual basis for the seven past years. In this study, we propose to evaluate its potential for temporal transfer scenarios in LULC classification, using a multi-year open dataset collected in Burkina Faso, West Africa.

One of the primary objectives of Earth observation is to capture the complex dynamics of the Earth system using satellite image time series. This process encompasses tasks such as reconstructing continuous cloud-free image sequences, identifying changes in land cover types, and forecasting future surface evolution. However, existing methods typically require specialized models tailored to different tasks, lacking unified modeling of spatiotemporal features across multiple time series tasks. In this paper, we propose a Unified Time Series Generative Model (UniTS), a general framework applicable to various time series tasks, including time series reconstruction, time series cloud removal, time series semantic change detection, and time series forecasting. Based on the flow matching generative paradigm, UniTS constructs a deterministic evolution path from noise to targets under the guidance of task-specific conditions, achieving unified modeling of spatiotemporal representations for multiple tasks. Furthermore, we construct two high-quality multimodal time series datasets, TS-S12 and TS-S12CR, filling the gap of benchmark datasets for time series cloud removal and forecasting tasks. Extensive experiments demonstrate that UniTS exhibits exceptional generative and comprehension capabilities in both low-level and high-level time series tasks. More details can be found on the project page: https://yuxiangzhang-bit.github.io/UniTS-website/

Advancements in remote sensing technologies and spatial data analytics have continued to transform how we map and monitor landscapes, including urban and agroforestry systems. Land use and land cover (LULC) analysis provides useful insights for sustainable land management, especially for agricultural stakeholders. Cocoa production is an agricultural system that benefits greatly from appropriate land use management. The system provides economic stability for millions of households worldwide through job creation, livelihoods, and raw materials for confectionery industries. However, its sustainability faces growing threats from environmental and socioeconomic challenges, such as climate change, land use conflicts, and extensive deforestation. One serious threat to cocoa production, particularly in West Africa (which supplies over 70% of the world's cocoa), is the widespread occurrence of the cocoa swollen shoot virus, among other pests and diseases that substantially decrease annual yields.

Therefore, accurate and current maps of cocoa farms are required for managing deforestation, supporting disease monitoring, and guiding climate-resilient agricultural strategies in the region. Previous efforts in mapping cocoa landscapes with remote sensing have not achieved the desired results, partly due to their spectral similarity to forests and shrublands, especially where they are part of agroforestry systems. This study aims to overcome this challenge by developing a robust methodology for detecting full-sun cocoa plantations using high-resolution satellite imagery and machine learning techniques for sustainable land utilisation.

This research addresses the challenges in classifying patterns of textile cultural relics by developing a multi-modal feature fusion approach. Current methods struggle with fine-grained classification and cultural-period analysis due to fragmented data and insufficient feature integration. The proposed framework integrates high-resolution images, historical documents, and spectral data through Vision Transformers and BERT models, enhanced by a Feature Enhancement Fusion Module. Validation on Han and Tang dynasty textiles demonstrates 3-5% accuracy improvement in fine-grained classification while maintaining model size under 300MB. This research establishes a new paradigm for digital heritage preservation, enabling precise pattern recognition and cultural evolution analysis with practical applications in museums and digital curation.

This study examines the impact of various personal laser scanning (PLS) schemes on the accuracy of individual tree attribute estimation in lowland pedunculate oak (Quercus robur L.) forests in central Croatia. Using a FARO Orbis PLS system, three scanning schemes were tested on sample plots with different densities: (i) a walking scheme with a planned trajectory, (ii) a static flash-scanning scheme with multiple fixed positions, and (iii) a combined scheme integrating walking and static scans. For each plot, multi-scan terrestrial laser scanning (TLS) was first conducted and used as a reference for diameter at breast height (DBH) and tree height (H). All PLS point clouds were processed using a consistent workflow, which included filtering, normalisation, individual-tree segmentation, and attribute estimation, and then compared against TLS-derived values. Preliminary results indicate that, although the static scheme yields denser point clouds and higher measurement precision, it does not consistently improve DBH and H accuracy compared to the walking scheme and can even increase errors in denser plots. The combined scheme performs similarly to the walking scheme. These findings indicate that well-designed walking-based PLS schees can provide accurate, operationally efficient estimates of individual-tree attributes in structurally complex deciduous stands, supporting wider adoption of PLS in forest inventory practice.

Despite its popularity, IMU preintegration is often perceived as requiring a dedicated implementation that is separate from conventional IMU propagation. In practice, however, many codebases already contain a reliable propagation module, often tied to a particular state or error-state definition. This raises two practical questions. First, does adopting IMU preintegration require reimplementing the IMU model from scratch? Second, how can one validate that a preintegration implementation, especially its bias Jacobians and covariance, is correct?

This note shows that IMU preintegration and IMU propagation can be viewed as two equivalent realizations of the same underlying computation. We first describe both in a way that is not tied to a particular perturbation convention. We then show that the preintegrated measurement, its Jacobian with respect to the initial IMU bias, and its covariance can all be obtained by wrapping an existing IMU propagation routine. Conversely, a preintegration module can be used to recover state-transition matrices and propagated covariances. This view also clarifies how to adapt preintegration across different error-state definitions without re-deriving bias Jacobians and residual covariances from scratch.

We validate the analysis by converting an RK4-based IMU propagation implementation to and from the GTSAM preintegration modules. In experiments with random IMU sequences, the recovered Jacobians, covariances, and transition matrices closely match those produced by GTSAM's tangent and manifold preintegration. These results suggest that a robust propagation implementation can serve both as a simple path to preintegration and as a practical reference for validating preintegration code.

Accurate information on tree structure is fundamental for forest management, biomass estimation, and carbon accounting. Personal Laser Scanning (PLS) has recently emerged as an efficient method for capturing detailed three-dimensional representations of trees under operational field conditions. At the same time, Quantitative Structure Models (QSMs) have become an important tool for deriving structural attributes such as diameter at breast height (DBH), tree height, and total tree volume directly from point cloud data. Despite increasing use of these approaches, systematic comparisons of different QSM reconstruction methods applied to PLS data remain limited.

This study evaluates two QSM workflows, PyTLidar and AdQSM, using PLS point clouds collected for pedunculate oak and European beech trees in leaf-off conditions. Data were acquired with the FARO Orbis system using both continuous mobile scanning and stationary flash scans, enabling the creation of mobile-only, flash-only, and combined point cloud variants. After preprocessing and single-tree extraction, each tree cloud was reconstructed separately with both QSM approaches. Key structural attributes were derived from each reconstruction to assess how the methods differ in estimating tree volume.

The comparison employs statistical measures that quantify natural variability among trees relative to variability introduced by each workflow. This allows the study to identify situations in which the two QSM methods produce consistent results and where their outputs diverge. The findings will support improved understanding of QSM behaviour when applied to PLS data and contribute to ongoing efforts to strengthen digital tree modelling for forest monitoring and ecological applications.

LIDAR point clouds combined with canopy-light extinction software can provide 2D simulations of shortwave radiation to identify crucial microclimates that control the overall water balance in savanna ecosystems. However, the point size necessary to accurately depict the wide range of tree species and forms that temperate savannas contain is largely unknown. To determine the optimal point size, hemispherical canopy imagery and field measured insolation will be compared to synthetical hemispherical imagery derived from LIDAR point clouds at different point sizes. The optimal point size will be validated against FLApy predictions and Hobo MX2022 measured illumination across 20 sample plots. The index of agreement between observed and predicted values will quantify systemic biases. Accurate point size is needed to assess tree removal scenarios and equip ecologists with the tools needed to understand the long-term implications for tree removal choices and how to best restore the tree canopy for long-term savanna resilience.

As Earth’s climate changes, it is impacting disasters and extreme weather events across the planet. Record-breaking heat waves, drenching rainfalls, extreme wildfires, and widespread flooding during hurricanes are all becoming more frequent and more intense. Rapid and efficient response to disaster events is essential for urban climate resilience and sustainability. A key challenge in disaster response is to correctly and quickly identify diaster locations for timely decision-making and resources allocation. In this paper, we propose a Probabilistic Cross-view Geolocalization approach, called ProbGLC, exploring new pathways towards generative location awareness for rapid disaster response. Herein, we combine the probabilistic and deterministic geolocalization models into a unified framework to simultaneously ensure model explainability and state-of-the-art geolocalization performance. Designed for rapid diaster response, the ProbGLC is able to address cross-view geolocalization across multiple diaster events as well as to offer unique features of model explainability and uncertainty quantification.