Conference Agenda

Beef cattle farming is a significant activity in Brazil, and forage quality has a direct impact on animal performance. However, traditional methods for estimating dry mass, which involve cutting, drying and weighing plant material, are slow and labor-intensive. UAVs equipped with multispectral sensors, such as the DJI Mavic 3M, offer a faster and more scalable alternative for monitoring mixed-forage pastures. This study estimates the dry mass of forage mixtures using multispectral UAV data in two scenarios: (i) using only spectral information and (ii) combining spectral data with canopy height measured in the field. Model performance was evaluated using R², RMSE, and percentage error. The multispectral-only model explained 55% of dry mass variability (720.56 kg/ha; 23.67%), while adding canopy height improved performance to 80% and reduced the error to 589.41 kg/ha (19.36%). Results show that canopy height enhances the accuracy and operational potential of UAV-based methods for estimating dry mass in mixed-forage areas.

Mining companies are continuously looking for cost efficient methods to monitor the success of their rehabilitation efforts. Although open access satellite imagery is available at regular temporal intervals, its usefulness for grassland biodiversity monitoring has been questioned due to its coarse spatial resolution with respect to the species size. To compensate for the low spatial resolution, previous studies have successfully explored the benefits of using a multitemporal set of Sentinel-2 (S2) images. However, unless the temporal patterns are studied as a whole, some of the phenological information such as growth rates are lost, and delayed snow cover may spread events like growth onset over multiple dates between plots. This study aims to explore the added value of temporal fitting of Sentinel-2 time series (ts) over existing baseline models applied using the full time series as such. Our set of temporal features included functional components, harmonic decomposition, frequency decomposition, and phenological metrics. Out of the compared models, the Random Forest regression model using a set of fitted temporal features achieved the highest holdout prediction accuracy (R2 = 0.36, RMSE = 3.87, relative RMSE = 0.20) and cross-validation accuracy similar to the baseline models. However, all the compared regression models underestimated extreme plant diversity to some extent. Future studies should account for varying vegetation cover and terrain features by incorporating auxiliary data.

Woody plant encroachment (WPE) threatens grassland ecosystems across the Canadian Prairies, causing grassland biodiversity loss with substantial economic impacts due to reduced forage production. While remote sensing offers scalable monitoring capabilities, existing approaches lack frameworks for distinguishing shrub and tree encroachment and often require extensive ground truth data. This study developed an ensemble machine learning framework integrating Sentinel-1 SAR and Sentinel-2 optical imagery with UAV-derived training data to map fractional shrub and tree cover across Saskatchewan's Aspen Parkland and Moist Mixed Grassland ecoregions, SK. A stacking ensemble combining Random Forest, Support Vector Machine, XGBoost, and Artificial Neural Network models with Ridge regression meta-learning outperformed individual algorithms, achieving mean R² values of 0.65 for shrub and 0.68 for tree cover prediction. Multi-scale training incorporating features at 10, 30, 50, and 70 m resolution improved performance by 15% for shrub and 24% for tree compared to single-scale approaches. Feature importance analysis revealed that shrub detection relied primarily on red-edge bands and moisture indices, while tree detection depended heavily on SAR backscatters. Quantile histogram matching enabled successful model transfer from Foam Lake Community pasture to Aberdeen Community Pasture, with resulting maps indicating that total WPE exceeded 50% in both study areas, with shrubs occupying 23.7% (Foam Lake) and 18.5% (Aberdeen) of both regions at rates higher than 5% shrub cover. The present framework provides a scalable, cost-effective approach for operational woody encroachment monitoring, enabling early detection and targeted functional management interventions to preserve grassland ecosystems.

Fractional vegetation cover (FVC) is a key land surface parameter describing vegetation abundance and structure, defined as the fraction of the ground area occupied by vegetation when viewed from nadir. FVC provides essential insights into ecosystem condition, productivity, and disturbance, making it a critical variable for biodiversity monitoring and habitat assessment. However, generating accurate and repeatable FVC estimates remains challenging due to scale effects, spatial resolution constraints, and inconsistencies in available validation data across time and space. This research develops a machine learning (ML) framework for large-scale FVC estimation that addresses these challenges by combining multi-sensor Earth observation data and Active Learning (AL) model refinement techniques. The ML framework is applied within key wood bison habitat in northern Alberta, focusing on mapping six vegetation components: spruce, pine, deciduous, shrub, herbaceous, and moss. The approach integrates Sentinel-1, Sentinel-2, Landsat-9, and GLO-30 data, optimized through feature selection and ensemble-based Random Forest modeling. The resulting FVC maps achieved strong predictive performance (R² = 0.50–0.88) and capture fine-scale spatial variability in vegetation composition. The ML pipeline provides a scalable and adaptive framework for FVC estimation that supports provincial landcover updates, improves understanding of wood bison habitat features, and contributes to ongoing ecosystem monitoring and conservation planning across boreal Alberta.

Wildlife monitoring from aerial imagery often requires precise animal localization under practical constraints where only object counts are needed. Traditional detection methods rely on bounding-box annotations, introducing unnecessary cognitive load for small objects spanning only a few dozen pixels. This work introduces DINOKey, a modified DINO transformer-based detector adapted to operate natively on point annotations rather than bounding boxes. Key contributions include: (1) architectural modifications to the DINO decoder, detection head, and denoising queries to directly predict 2D keypoints; (2) a combined loss function integrating L1 regression, focal loss, and average Hausdorff distance, with ablations validating each component; (3) open-source implementation within an existing detection framework; and (4) demonstration of improved small-object localization and reduced false positives on an aerial elephant dataset compared to box-supervised baselines. Ablation studies show that the Hausdorff distance term provides the largest accuracy gain by effectively reducing false positives, while focal loss improves stability in densely clustered regions. The proposed method achieves 0.786 mAP and accurately localizes animal centers across diverse environmental conditions, offering a practical solution for conservation practitioners working under tight logistical constraints.

Fog-dependent ecosystems in the Atacama Desert host highly specialized vegetation, yet monitoring their functional traits remains challenging due to remoteness and limited spectral detectability. The bromeliad Tillandsia landbeckii exhibits extremely low reflectance in the VIS/NIR range, rendering conventional multispectral approaches ineffective.

This study evaluates the potential of a novel UAV-based VNIR/SWIR multi-camera system (camSWIR) for estimating canopy water content (CWC) in Tillandsia landbeckii. A UAV survey conducted in northern Chile acquired high-resolution (≈3 cm GSD) SWIR imagery across four operational bands (1100–1650 nm). Field-based destructive sampling (n = 20) provided reference CWC measurements, and a statistically rigorous workflow was applied to mitigate overfitting in a high-dimensional predictor space.

Results show that the spectral slope between 1200 and 1510 nm is the most informative predictor of CWC, with cross-validated performance indicating moderate predictive skill (LOOCV R² ≈ 0.52), but reduced stability under nested validation. The repeated selection of predictors within this wavelength region confirms a physically meaningful relationship with liquid water absorption.

Despite limitations due to a small sample size and species-specific optical properties, particularly the dense trichome layer that affects light interactions, the study demonstrates the feasibility of SWIR-based, non-destructive CWC estimation in hyper-arid ecosystems. These findings provide a proof of concept for future upscaling, highlighting the need for larger calibration datasets and improved modelling to enable reliable spatial mapping of plant water status.

Aerial imagery provides a powerful avenue for monitoring wildlife populations, yet automated detection remains challenging. Animals typically occupy only a tiny fraction of large-scale aerial imagery, may be partially obscured, and appear against highly diverse Arctic and sub-Arctic backgrounds. Suppervised deep-learning detectors also depend on large, fully annotated datasets, making broad ecological surveys labor-intensive and slow to scale.

This study explores an alternative perspective: viewing wildlife as rare events within mostly background imagery. Instead of training on annotated animal samples, an anomaly-detection framework learns the visual patterns of normal landscapes and identifies deviations from these patterns as potential animal locations. To guide the model without costly labels, simple animal-like shapes are inserted into background patches during training, encouraging the network to recognise features associated with real targets while avoiding the need for detailed masks or bounding boxes.

The approach generates two outputs: patch-level predictions distinguishing empty from potentially occupied areas, and pixel-level anomaly maps highlighting likely target locations. When evaluated on a highly varied Arctic dataset, the method remains reliable despite major shifts in terrain, surface texture, animal distributions and postures, and pronounced class imbalance that often degrade supervised models. Unlike distribution-based anomaly approaches that rely on stable normal-feature statistics and frequently misinterpret natural texture variability as anomalies, this method handles heterogeneous environments more effectively.

Overall, the study shows that anomaly-oriented frameworks, typically used in industrial and medical settings, have strong potential to ease annotation demands and support scalable, automated wildlife detection in complex remote-sensing environments.

Rapid and reliable assessment of building damage after major earthquakes is essential for effective emergency response and recovery planning. This study formulates post-disaster building damage detection (BDD) as a binary image classification task (damaged vs. undamaged buildings) using multimodal satellite data and a unified ResNet-18 backbone to enable a controlled comparison of fusion strategies. The analysis focuses on the Mw 7.7 Myanmar earthquake of 28 March 2025 and integrates post-event COSMO-SkyMed Second Generation (CSG) dual-polarization (HH, HV) SAR imagery, Maxar optical data, OpenStreetMap (OSM) building footprints, and UNOSAT damage annotations. Three fusion paradigms are evaluated: Early Fusion (EF), Late Fusion (LF), and a novel Middle Fusion (MF) approach. The proposed MF framework introduces a Footprint-Guided Cross-Attention (FGCA) mechanism that uses building geometry as a spatial prior to guide feature-level interaction between SAR and optical representations. Five-fold cross-validation results show that MF consistently outperforms EF and LF, achieving higher precision, F1-score, and robustness across modality configurations. By jointly exploiting SAR structural sensitivity, optical detail, and footprint-based spatial context, the proposed Footprint-Guided Middle Fusion (FGMF) framework enables accurate and scalable building damage mapping from heterogeneous Earth Observation (EO) data.

Accurate and timely building damage detection (BDD) is crucial for disaster emergency response. Although deep learning-based change detection methods have made significant progress in remote sensing, their practical application in disasters still faces two major challenges: (1) Existing high‑accuracy models are typically computationally complex and difficult to deploy for real‑time inference on edge devices.. (2) Model performance heavily relies on large amounts of annotated data, but disaster data are extremely scarce. To address these challenges, this paper proposes a novel lightweight Local‑Global Interaction Network (LGINet) for efficient BDD. The core of LGINet is the proposed Local‑Global Interaction Unit (LGIU), which achieves efficient fusion of detailed and contextual features through a dual‑path architecture and channel‑wise cross‑attention mechanism. Furthermore, a Frequency Difference Enhancement Unit (FDEU) is proposed to generate more accurate damage features, and contrastive learning is employed to reduce the model’s sensitivity to weather conditions and its reliance on annotated data. Experimental results on the xBD and WBD datasets show that LGINet achieves F1-scores of 81.76% and 80.91%, respectively, with an inference speed of 47.83 FPS. It achieves the best balance between accuracy and efficiency, outperforming existing methods.

Recent armed conflicts have increased the need for reliable, spatially explicit damage mapping to support situational awareness, humanitarian assessment, and reconstruction planning. This contribution presents a hybrid change-detection framework for conflict-related urban damage mapping that combines AlphaEarth Foundations embedding change with Sentinel-1 SAR change indices. AlphaEarth provides semantically informed annual embeddings, while Sentinel-1 time series contribute all-weather sensitivity to structural change.

The study compares several embedding-based change metrics and combines the selected AlphaEarth indicator with SAR-derived change measures through simple scalar fusion rules. The proposed framework is designed to preserve the sharp sensitivity of SAR to abrupt structural changes while reducing part of the diffuse background response that often complicates single-source interpretation.

Experiments are conducted over war-affected urban areas in Ukraine, with illustrative examples from Bakhmut and Avdiivka. The results show that AlphaEarth and Sentinel-1 provide complementary information and that their fusion improves the spatial specificity of detected damage patterns. The contribution highlights the potential of combining foundation-model representations with radar time series for operational damage mapping in conflict settings.

Rapid and precise identification of disaster locations is essential for efficient emergency response and management. However, during the immediate post-disaster phase, the lack of timely and reliable information often impedes relief operations. Although satellite imagery and ground-based sensing systems provide valuable data, their effectiveness is constrained by factors such as time delays, high costs, and limited spatial resolution. At the same time, social media platforms such as X (formerly Twitter), Instagram, and Facebook have become valuable channels for real-time, crowd-sourced information. Users function as "human sensors," contributing extensive on-the-ground insights. Much of this content is visual—images that capture the effects of disasters with finer street-level detail and immediacy than textual posts. In this study, we propose a novel, deep learning-based image-matching framework designed to pinpoint the geographic coordinates of disaster events from social media images with street-level accuracy. The core of our approach is to match a query disaster image against a database of georeferenced Google Street View (GSV) imagery. The methodology consists of image pre-processing and feature enhancement; deep feature extraction and matching, and location inference and verification. The preliminary results on an external validation dataset are highly promising, demonstrating a high detection rate of ~90% with confidence scores above 0.9. The model proves resilient to key challenges such as partial occlusion and varied lighting, accurately segmenting multiple objects against complex backgrounds of damaged structures and flooded areas.

The paper presents a novel deep learning framework for automated disaster damage assessment using remote sensing imagery. It addresses the challenge of timely and accurate damage classification in the aftermath of disasters, aiming to improve emergency response and resource allocation. The proposed system leverages both pre- and post-disaster satellite images to assess building damage across four categories: no damage, minor damage, major damage, and destroyed. The central innovation lies in the development of a multi-modal attention mechanism, which integrates features from both pre- and post-event images to enhance damage detection. A lightweight ConvNeXT-Tiny architecture serves as the backbone, ensuring efficient processing while maintaining high performance.

Key contributions of this work include: (1) a cross-attention module that fuses multi-modal data, (2) an optimized preprocessing pipeline designed for large-scale datasets, and (3) novel data augmentation techniques that improve the model’s robustness. Experiments on a large-scale disaster damage dataset show the model achieves an impressive 94.90% classification accuracy, with strong performance in discriminating damage levels and resilience to incomplete or corrupted data. This framework represents a significant step forward in disaster response, offering a scalable solution for real-time damage detection. The research demonstrates the potential of combining remote sensing, multi-temporal imagery, and deep learning to expedite and improve disaster damage assessment, ultimately supporting more efficient emergency management.

Urban areas evolve rapidly, often increasing exposure to natural hazards, especially in seismically active regions such as southern Turkey. This contribution presents an AI-based workflow for multi-temporal analysis of urban expansion in the city of Osmaniye between 2015 and 2025. The methodology integrates Sentinel-2 multispectral imagery with a U-Net convolutional neural network trained on World Settlement Footprint (WSF) masks for binary segmentation of built-up versus non-built-up areas. After training on 2015 and 2019 data, the model was applied to the full temporal series to assess its generalisation capability and to quantify long-term urban growth.

Results show a substantial increase in built-up surfaces over the decade, with a temporary decline linked to the 2023 earthquake and a marked acceleration during the reconstruction phase. Beyond the quantitative trends, the spatial patterns identified by the model highlight how urban expansion has progressively shifted from the central districts toward peripheral zones, revealing both densification processes and outward sprawl. These observations provide valuable indications on how development pressures interact with seismic vulnerability.

The approach demonstrates the potential of AI and open satellite data for large-scale, reproducible monitoring of urban dynamics and for supporting risk-informed urban planning. Because it relies entirely on open-source datasets and tools, the workflow can be easily transferred to other hazard-prone regions, offering a scalable and transparent framework for assessing urban change, post-disaster reconstruction, and long-term exposure.

Urban surface structure and component temperatures induce significant thermal anisotropy (TA), resulting in substantial differences in observed surface temperatures across varying viewing angles. Although previous studies have investigated the temporal dynamics of TA through observations and modeling, its spatial differentiation over heterogeneous surfaces remains poorly constrained. Resolving how surface heterogeneity influences TA is hindered by the coarse spatial resolution and limited angular sampling of current multi-angle satellite observations. Consequently, most mainstream thermal-anisotropy models were developed for simplified scenes and lack systematic evaluation of their applicability to complex urban environments. To address these challenges, we coupled the microscale 3D urban energy balance model (TUF-3D) with the state-of-the-art Discrete Anisotropic Radiative Transfer (DART) model. This approach allows for rapid and accurate TA modeling of hypothetical urban scenes with varying geometric structures and component temperatures, thereby quantifying the impact of surface heterogeneity on TA. Building height variability was used to represent geometric heterogeneity, while differences in building material properties were used to characterize component temperature heterogeneity. To evaluate , The results of a series of sensitivity experiments have validated the individual effects of geometric and component temperature heterogeneity on TA. From the perspective of component temperature, changes in average component temperatures result in a maximum TA difference of 7.29 K, while temperature variability alone contributes only 0.54 K. These findings suggest that assuming simplified scenes with uniform building heights or homogeneous component temperatures can introduce biases in TA simulations, potentially compromising the accuracy of models correcting for the angular effects of land surface temperature.

Understanding the thermal behavior of soil and vegetation separately is essential for interpreting land–atmosphere energy exchange, diagnosing ecosystem stress, and improving land surface modelling. However, conventional satellite LST products only provide a mixed radiometric signal, masking the distinct thermal responses of soil and canopy. This study introduces GloSVeT, the first global dataset that provides monthly surface soil and vegetation component temperatures at 0.05° resolution for 2003–2023. The dataset is generated using an enhanced multisource fusion framework that integrates multi-temporal MODIS observations with ERA5-Land skin temperature and vegetation structural information to retrieve physically consistent component temperatures. We summarize the data sources, modelling framework, and global implementation strategy, and present an independent evaluation using flux-tower networks with screened spatial representativeness. Validation results show strong agreement with in-situ measurements, with correlations typically above 0.9 and RMSE around 2 K for both soil and vegetation temperatures. Seasonal variations in performance reflect expected hydrothermal conditions, and a small cool bias is attributable to the temporal sampling of satellite observations. GloSVeT provides a new basis for studying surface energy partitioning, monitoring hydrothermal dynamics, and supporting ecosystem and climate model applications.

This study presents the development of a ground-based observation system designed for vicarious calibration of satellite sensors operating in the mid-wave infrared (MWIR) region. Conventional natural targets used in LWIR calibration lack spectrally stable emissivity in MWIR, motivating the need for dedicated reference targets and high-sensitivity measurement instruments. We introduce a thermally controlled ground reference target whose effective emissivity can be tuned by adjusting the ratio of water and metal surfaces using perforated plates of varying hole diameters. In parallel, an MWIR radiation thermometer employing lock-in detection was developed to enable accurate measurement of low-signal MWIR radiance from room-temperature targets. The system achieved measurement uncertainties down to 20–70 mK. A field campaign was conducted at the Goheung Aerospace Center using the integrated reference targets and radiation thermometer to validate performance under real environmental conditions. The results demonstrate the feasibility of applying controlled emissivity targets and lock-in-based MWIR radiometry to improve the accuracy of MWIR vicarious calibration frameworks.

A Method for Identifying Industrial Heat Sources

This study addresses the limitations of conventional urban shortwave radiation simulations in representing complex three-dimensional morphology. A parameterization approach for large-scale urban sky view factor was proposed, significantly improving computational efficiency and spatial adaptability. Based on this, an urban solar shortwave radiation transfer model was developed to quantitatively characterize the shading and reflection effects of building clusters. Furthermore, a novel remote sensing inversion method for urban surface reflectance and solar radiation parameters was introduced, enabling high-accuracy estimation of surface radiative properties and offering a new technical pathway for urban thermal environment and energy balance research.

Urban Heat Islands (UHIs) emerge from reduced vegetation, impervious surfaces, and anthropogenic heat emissions, leading to elevated surface temperatures in urban areas. Monitoring UHIs at fine spatial and temporal scales requires thermal data capable of capturing both urban heterogeneity and daily variability—conditions not satisfied by the native 1 km resolution of MODIS Land Surface Temperature (LST).

This study presents a regime-aware machine learning workflow to downscale daily MODIS LST to the native spatial scale of MODIS NDVI (231 m) over Bologna (Italy), using only MODIS-internal predictors and meteorological forcing. The approach adopts a two-stage architecture: a Ridge regression model estimates a day-level atmospheric bias, while a Random Forest reconstructs pixel-level residuals to recover fine-scale thermal variability from vegetation, land-cover, topographic, and atmospheric predictors. To account for atmospheric control, the dataset is partitioned into three thermal regimes (COLD, MILD, HOT), with independent models trained for each regime.

Pre-processing and data integration were performed in Google Earth Engine using MODIS LST (MOD11A1/MYD11A1), NDVI, SRTM-derived terrain variables, and built-up fraction from ESA WorldCover. Experiments show strong predictive performance (RMSE < 1 K; R² ≈ 0.90) and spatial patterns consistent with Local Climate Zones.

The MILD and HOT regimes provide the largest enhancement in spatial detail compared to the original MODIS product, while the COLD regime shows reduced performance, likely due to weaker surface–atmosphere coupling. Results highlight that atmospheric conditions play a dominant role in downscaling accuracy, exceeding the impact of model architecture. The framework enables scalable, daily UHI monitoring and supports heatwave analysis and climate-resilient urban planning.

Nighttime optical remote sensing provides valuable insights into natural and, in particular, human activities. This study evaluates the nighttime imaging capabilities of the Sentinel-2 mission using the only available nighttime acquisition not limited to ocean observations for dark signal calibration, covering the United Arab Emirates with Dubai in 2015. We checked the detection limit using granules over the Persian Gulf, extracted radiance spectra for different regions of interest, and analysed lighting types and temperatures. Results suggest a conservative nighttime detection limit of approx. 0.37 W/m²/um/sr for visible/near infrared bands, and 0.08 W/m²/um/sr for short-wave infrared bands. Sentinel-2’s high spatial resolution and multispectral bands, although designed for daytime observations, were capable of detecting and classifying bright visible/near and short-wave infrared emitters. Comparisons with hyperspectral EnMAP imagery acquired in 2025 validated the classifications and revealed changes in urban lighting over a decade. While limitations apply, this study highlights S2’s potential for nighttime remote sensing and supports considerations of nighttime capabilities for future satellite missions.