Accurate snow depth (SD) estimation in forested mountain environments remains challenging owing to terrain complexity and vegetation effects. Although UAV-LiDAR provides high-resolution three-dimensional measurements, LiDAR-derived SD estimation may be insufficient to account for SD spatial variability associated with forest structure and local terrain conditions. This study developed an integrated framework combining UAV-LiDAR data, structure-from-motion photogrammetry, and machine learning (ML) algorithms to improve high-resolution SD estimation. UAV-LiDAR surveys were conducted during snow-covered and snow-free periods in 2024, and the datasets were processed using DJI Terra and Pix4Dmapper, to generate digital canopy height models (DCHMs) and digital elevation models (DEMs). Initial SD estimates were obtained by subtracting the snow-free DEM from the snow-covered DEM. Subsequently, remotely sensed variables, including initial SD estimates, DEMs, DCHMs, and distance to trees, were integrated with ML models and evaluated the predictive performance with validation of field data (n = 167). A linear mixed-effects model was used to compare the predictive accuracies of the different integrated approaches. The results demonstrated that the predictive accuracy differed significantly among the integrated approaches. Among all combinations, the estimated SD range (0.419–0.925 m) generated using the DJI Terra-derived SD map, DCHM, and distance to trees with Random Forest model achieved the highest predictive performance (R2 = 0.811, RMSE = 0.049, MBE = 0.008, and MAE = 0.037). The findings demonstrate that integrating DJI Terra-derived SD with forest structural and tree proximity information through ML can improve high-resolution SD estimation, thereby supporting operational snow monitoring in forested mountain environments.
The management and mitigation of disasters increasingly rely on remote sensing data. However, the spatial resolution of publicly available satellite imagery often limits the detection of fine-scale features, reducing the accuracy of small-scale environmental analysis. This paper proposes a deep learning-based super-resolution (SR) framework to enhance the spatial resolution of Sentinel-2 imagery. The proposed method, termed STCN-SR (Spatial Transformer–CNN Network for Super-Resolution), integrates convolutional neural networks and transformer-based attention mechanisms to jointly capture local textures and global spatial relationships. The model is trained on low and high-resolution image pairs and reconstructs super-resolved images from 10 m to 5 m resolution. The generated outputs are georeferenced using metadata from the original tiles to maintain alignment within GIS workflows. Quantitative evaluation using peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) demonstrates improved reconstruction quality, achieving a maximum PSNR of 43.91 dB on the test dataset. Qualitative comparison with UAV reference imagery indicates improved boundary sharpness and clearer small-scale structures, suggesting potential benefits for environmental monitoring tasks, particularly in flood-affected landscapes. Overall, the results demonstrate the potential of deep learning-based super-resolution to improve the utility of Sentinel-2 imagery for geospatial applications.
Increasing hydro-climatic variability is challenging the stability of meteorological baselines in tropical rice agro-ecosystems. This study analyzed 3,652 daily ERA5-Land observations from 1 January 2016 to 30 December 2025 for Malaysia’s Barat Laut Selangor (BLS) rice granary using structural-break analysis, Granger predictive association analysis, and machine-learning models, including Random Forest (RF) and Multiple Linear Regression (MLR), to compare the predictability of Layer 1 soil moisture and ERA5-Land low-vegetation Leaf Area Index (LAI) using hydro-climatic predictors. Structural-break analysis indicated a temporal discontinuity in monthly precipitation, although sensitivity analyses showed that its precise timing was method-dependent and uncertain. Three-month accumulated precipitation–potential evaporation balances showed recurrent periods during which ERA5-Land potential evaporation exceeded precipitation, indicating periods of increased atmospheric water demand. Granger predictive analysis identified a significant lagged predictive association between Layer 1 soil moisture and LAI, whereas immediate meteorological variables showed weaker predictive associations with LAI. Under chronological temporal validation, RF showed strong predictive performance for Layer 1 soil moisture (R² = 0.9311), whereas prediction of ERA5-Land low-vegetation LAI remained limited (MLR: R² = 0.2910; RF: R² = 0.2460). Across 49 expanding-window rolling-origin folds, RF maintained comparatively strong soil-moisture performance (R² = 0.7374 ± 0.1688), while LAI prediction remained weak (MLR: R² = 0.2185 ± 0.1182; RF: R² = 0.1983 ± 0.1914). The contrasting predictability of the two targets suggests that the selected hydro-climatic predictors represented the physical soil-water state more effectively than the model-derived vegetation indicator. The absence of improved LAI performance with RF further indicates that greater model complexity alone did not resolve this predictive limitation. Potential influences of crop phenology, irrigation, management practices, spatial-scale differences, and reanalysis-product uncertainty remain plausible but untested because direct agronomic observations were unavailable. Future modeling should integrate longer observational records, independent vegetation measurements, crop calendars, irrigation information, and field-level agronomic data.
Satellite images have recently been made accessible using web services in a web-based Geographic Information System (WebGIS). Web Map Service (WMS) and Web Processing Service (WPS) are the main services used by the information system for viewing and processing remotely sensed data online, respectively. While WMS is widely used to render satellite images for visual interpretation, few image processing functions are implemented using WPS. Formulating the service to classify satellite images available via WMS provides a very accessible, efficient and cost-effective satellite image processing resource for many applications. Artificial Neural Network (ANN) computing is one of the machine learning methods used for satellite image classification. Unlike the conventional per-pixel parametric statistics classifier, an ANN is a distribution-assumption-independent algorithm capable of processing complex nonlinear relationships, which is useful for identifying spectral and spatial patterns of land use types in satellite images. This study focuses on the formulation of WPS to classify satellite images made available as a color composite WMS using ANN. The results of the study show that an accurate satellite image classifier using an ANN can be implemented with WPS, using WMS URL as input. It is further demonstrated that the WPS-based ANN classifier could be easily accessed using WebGIS for easy implementation. This study provides a method to increase the usability of satellite image color composite WMSs, from data exclusively used for visual interpretation, to WPS inputs for a wide array of web-based image-processing applications.
