Environmental and Earth Sciences

Abstract: High-resolution land cover classification is critical for monitoring environmental change and managing natural resources. This study presents a fully unsupervised framework that integrates Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 optical imagery at 10-meter spatial resolution. A cloud-native export protocol in Google Earth Engine (GEE) enables the generation of consistent, cloud-free, and snow-free seasonal composites across Ontario, Canada. A comprehensive feature engineering pipeline combines spectral indices, radar backscatter metrics, terrain derivatives from digital elevation models (DEMs), and temporal statistics to create a rich multi-sensor input space. Dimensionality reduction is performed using Sparse Principal Component Analysis (SparsePCA) and mutual-information based feature selection. Clustering is conducted using three complementary algorithms: centroid-based K-means, density-based Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN), and reachability-based Ordering Points To Identify the Clustering Structure (OPTICS). Final land cover labels are assigned via a majority-voting ensemble, with prediction ties resolved deterministically using OPTICS. OPTICS is particularly effective for modeling heterogeneous landscapes due to its ability to detect clusters of varying density without requiring a global threshold. The resulting classification maps are validated against reference land cover data, demonstrating the scalability and effectiveness of the proposed label-free mapping approach.

Abstract: Sentinel-1 Synthetic Aperture Radar (SAR) is a multi-purpose monitoring satellite suite that, among many applications, provides sea surface wind speeds at high spatial resolution. The overall aim of the study is to quantify the accuracy of the SAR wind products from Copernicus Ocean Wind, called OCN OWI, and from the Technical University of Denmark (DTU) Department of Wind and Energy Systems’ product called DTU SAR. Both products serve as a basis for offshore wind resource mapping for offshore wind energy planning. With the growth in offshore wind farms, offshore wind resource information is highly relevant. However, a comparison between the two products is lacking. This study fills this gap by presenting a comprehensive validation of the two Sentinel-1 wind speed products using wind speed measurements from 18 weather buoys and 10 floating wind lidars in the European Seas. It is the first time a comprehensive wind lidar data set has been used for SAR wind validation. Key findings: OCN OWI vs. lidar (buoy) shows R2 = 0.93 (0.84), root mean square error (RMSE) = 1.18 m/s (1.61 m/s), mean absolute error (MAE) = 0.86 m/s (1.24 m/s), and bias = -0.5 m/s (-0.6 m/s). DTU SAR vs. lidar (buoy) shows R2 = 0.88 (0.84), RMSE = 1.3 m/s (1.6 m/s), MAE = 0.92 m/s (1.22 m/s), and bias = 0.02 m/s (-0.04 m/s). OCN OWI provides a filtered data set, and validation vs. lidar shows R2 = 0.95 and RMSE = 0.88 m/s; however, at the expense of discarding more than 50% of all data. The lidar vs. SAR wind speed statistics outperformed the buoy comparison statistics for all metrics studied. The 3 km Norwegian reanalysis (NORA3) wind speeds vs. lidar (buoy) show RMSE = 1.27 m/s (1.64 m/s) and bias = -0.01 m/s (-0.43 m/s). Lidar wind speed data are more accurate than buoy data and give a more trustworthy validation of SAR wind speed and model wind speeds than buoy data. Lidar data are recommended for validation studies on Geophysical Model Functions on SAR winds. Satellite Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals (IMERG) are collected at the buoy and lidar sites for comparison of SAR-based wind speed accuracy during precipitation. SAR and NORA3 show consistently higher RMSE values vs. buoy and lidar data with increasing precipitation and higher mean wind speeds at higher precipitation rates, but no systematic bias. Creating a precipitation flag for Sentinel-1 SAR-based winds would reduce the number of available samples and potentially lead to biased estimates of the wind resource. Vertical wind profiles at lidar locations are compared to SAR-based wind profile extrapolation, including stability correction.

Abstract: Landslide susceptibility mapping (LSM) is fundamental to disaster prevention and spatial risk management in mountainous regions. However, traditional LSM methods typically rely on static geo-environmental indicators and subjective classification thresholds, resulting in limited temporal adaptability and reduced interpretability. To address these limitations, this study proposed a surface deformation-constrained LSM framework that integrated SBAS-InSAR-derived ground deformation data with machine learning models to enhance the accuracy and temporal relevance of susceptibility results. Using the Wangmo County area in Guizhou Province, China as a case study, multi-source data, including remote sensing imagery, geo-environmental factors, and field investigations, were used to construct and validate four machine learning models. The Random Forest (RF) and Back Propagation Neural Network (BPNN) models exhibited superior performance and were further enhanced through Shapley value analysis to improve interpretability. Surface deformation was extracted from 31 Sentinel-1A images using SBAS-InSAR technology, and a Pearson correlation-based optimization approach was applied to align susceptibility classifications with ground deformation patterns. The improved LSM showed significantly improved temporal relevance compared to traditional LSM methods and better performance in identifying landslide activity during specific periods, as verified by two representative cases. The proposed framework significantly improved the temporal sensitivity and scientific robustness of LSM, providing a promising tool for dynamic landslide risk monitoring and decision-making in complex terrains.

Abstract: Land subsidence driven by excessive groundwater extraction in the Choushui River alluvial fan of central Taiwan poses a significant threat to the structural integrity of the Taiwan High-Speed Rail (THSR). This study presents an integrated approach combining multi-temporal Interferometric Synthetic Aperture Radar (MT-InSAR), continuous and campaign Global Navigation Satellite System (GNSS) measurements, and precise leveling surveys to characterize both vertical and horizontal surface displacements along the THSR corridor. Sentinel-1 C-band SAR data from ascending (A69) and descending (D105) tracks were processed using the Small Baseline Subset (SBAS) technique over the period 2015–2021, and decomposed into east-west (EW) and vertical components via 2.5D decomposition. The InSAR-derived EW velocity field was calibrated using GNSS Ordinary Kriging interpolation, improving R2 from 0.147 (RMSE = 4.24 mm/yr) to 0.992 (RMSE = 0.24 mm/yr). The vertical velocity field was corrected using a polynomial trend surface fitted to 922 leveling benchmarks and 38 CGPS stations, reducing RMSE from 6.03 to 4.85 mm/yr (R2 from 0.889 to 0.898) and virtually eliminating the systematic bias from +3.47 to −0.47 mm/yr. Maximum subsidence exceeding 60 mm/yr was identified in the Yunlin Tuku area, while three secondary subsidence centers exist in Changhua. The horizontal velocity field reveals a convergent pattern directed toward subsidence centers, with magnitudes of 2–10 mm/yr, confirming that aquifer compaction induces significant lateral deformation. Along the THSR corridor, differential EW velocities across the Xizhou and Tuku subsidence zones highlight potential risks to rail alignment and structural safety, with horizontal strain rates reaching approximately 10⁻⁶/yr. GNSS observations additionally provide the north-south velocity component that InSAR cannot detect, enabling a more complete three-dimensional deformation characterization.

Abstract: Vignetting introduces spatial radiometric nonuniformity into remote sensing images and degrades subsequent radiometric analysis, image interpretation, and calibration-related applications. To address this problem, this paper proposes a vignetting correction method based on low-rank modeling and polynomial fitting. The method constructs a data matrix in the logarithmic domain, extracts the common vignette component through low-rank decomposition, and further recovers a smooth vignette field by polynomial fitting. Experiments were conducted using real remote sensing images, simulated vignetted images, and star images. On simulated vignetted datasets, the proposed full method achieved the best overall performance, with mean absolute error (MAE), mean absolute deviation (MAD), center-region MAE, and edge-region MAE of 0.482%, 3.646%, 0.138%, and 0.519%, respectively. Compared with the low-rank-only method, these four metrics were reduced by 22.9%, 32.9%, 71.7%, and 19.9%, respectively. For star images, the method reduced image-plane nonuniformity from 1.39-1.92 to 0.59-0.80 while preserving the stability of background-subtracted stellar DN values. These results demonstrate that the proposed method effectively suppresses image-plane nonuniformity while maintaining radiometric consistency, thereby providing an effective solution for remote sensing image vignetting correction.

Abstract: Remote sensing-based change detection for infrastructure monitoring demands methods that are simultaneously accurate, robust to severe class imbalance, and transparent in their decision logic. This study proposes MS-HySAN, a hybrid change-detection framework that addresses these requirements through three coordinated design decisions: (i) a truncated, attention-augmented Siamese encoder that serves as a frozen feature extractor rather than an end-to-end pixel classifier, (ii) a latent–physical fusion strategy that concatenates multi-scale CNN difference features with physically interpretable spectral-index differences, and (iii) a LightGBM classifier that performs internal sparse feature selection and exposes gradient-based SHAP attributions for post-hoc analysis. The framework is evaluated on high-resolution PlanetScope imagery (4-band and 8-band) over a national highway construction corridor in Indore, India, using 21 acquisitions from 2022–2025 with geographic k-fold cross-validation to enforce spatial independence. Experimental results show that the proposed hybrid model consistently outperforms conventional deep learning baselines including U-Net and Siamese U-Net across bi-temporal multi-class change-detection tasks, and competes with bi-temporal architectures (ChangeFormer, SNUNet, BIT) under the same training conditions. A SHAP interpretability analysis reveals complementary and physically meaningful contributions from the learned deep features and the handcrafted spectral indices, validating the fusion strategy. In the best-case setting, MS-HySAN (bi-temporal, indices + reflectance) achieves an overall mean F1-score of 0.95 (Kappa: 0.90), outperforming the corresponding deep baseline by +6 F1 points while maintaining stable cross-fold performance.

Abstract: Ship detection in synthetic aperture radar (SAR) imagery, an indispensable all-weather technology for marine engineering and coastal safety, remains challenging in complex nearshore scenes due to coupled speckle noise, sea-land clutter, large scale variation, and extreme class imbalance. Existing decoupled pipelines fail to jointly mitigate these degradations, leading to high false alarm rates and poor generalization. We propose DN-AnchorNet, an end-to-end unified framework integrating a detection-oriented structure-preserving enhancement branch, a scale-adaptive anchor mechanism, and an adaptive weighted Smooth L1 loss. The detection-guided enhancement branch operates without paired clean data to preserve critical ship structures. The scale-adaptive anchor design enhances matching for small, elongated, and arbitrarily oriented ships, while the tailored loss improves robustness against hard samples and scale errors under class imbalance. Extensive experiments on challenging nearshore subsets of RSDD-SAR and SSDD+ show that DN-AnchorNet achieves the best overall performance among all compared representative oriented object detectors, with AP₅₀ values of 0.699 and 0.610, and F1-scores of 0.757 and 0.689, respectively. A strict zero-shot cross-dataset evaluation on HRSID further demonstrates strong generalization to unseen marine SAR conditions. These results confirm that joint optimization achieves a favorable accuracy-false alarm balance for practical coastal monitoring applications. Code is available at: https://github.com/yongqi011210/Dn-anchornet.

Abstract: Satellite-derived environmental features can predict county-level life expectancy (LE) across the contiguous United States with a mean absolute error of 1.08 years over two decades, without using any census or sociodemographic inputs. We assembled 61,680 county-year observations across 3,084 counties from 2000–2019, integrating features from 11 satellite and gridded data streams. The data streams include the Moderate Resolution Imaging Spectroradiometer (MODIS) land surface temperature and vegetation indices, Sentinel-1 synthetic aperture radar, Sentinel-2 and Landsat optical imagery, the United States Department of Agriculture (USDA) Cropland Data Layer, the European Commission Joint Research Centre (JRC) Global Surface Water layer, the Copernicus Digital Elevation Model, the European Space Agency Climate Change Initiative (ESA CCI) soil moisture record, and the Food and Agriculture Organization (FAO) gridded livestock densities. After a supervised pruning step that removed low-importance variables, a Random Forest regressor was trained and evaluated using 5-fold cross-validation grouped by county. The grouping places all 20 years of each county exclusively in either the training set or the test set, which prevents spatial information leakage between folds. Coefficient of determination, mean absolute error, and root mean squared error are reported as R² = 0.631 ± 0.013, MAE = 1.08 ± 0.02 years, and RMSE = 1.48 ± 0.04 years. Moran's I, a measure of residual spatial autocorrelation, is 0.0988 (p = 0.001), which supports geographic generalisation. Multimodal fusion reduces unexplained variance by approximately one-third relative to the strongest single-modality baseline (MODIS land surface temperature alone, R² = 0.442). TreeSHAP attribution analysis reveals a feature hierarchy in which nighttime land surface temperature features carry roughly 6.16× the cumulative attribution weight of all daytime channels combined. The model response shows a protective inflection near a minimum overnight temperature of about 7.5°C. Because all input streams are globally available, the framework is extensible to regions where civil registration and vital statistics systems are incomplete, and supports satellite-based monitoring of United Nations Sustainable Development Goal (UN SDG) Target 3.9.

Abstract: Hyperspectral remote sensing provides rich spectral information and has been widely used in fine-grained land-cover classification and forest monitoring. However, accurate tree species classification remains challenging due to subtle interspecific spectral differences, similar spatial structures among related species, redundant spectral bands, and the limited ability of existing methods to model long-range spatial–spectral dependencies efficiently. In addition, many existing hyperspectral image classification methods rely on patch-based inputs and sliding-window inference, which often lead to redundant computation and insufficient utilization of global image context. To address these issues, this paper proposes MambaHSINet, a dual-branch bidirectional state space network for full-image pixel-wise hyperspectral tree species classification. Specifically, the proposed network employs a spectral branch and a spatial branch to explicitly extract complementary spectral responses and spatial structural features. A lightweight channel attention mechanism is introduced to emphasize informative spectral–spatial representations while suppressing redundant information. Subsequently, a bidirectional Mamba global modeling module based on selective state space modeling is adopted to capture long-range contextual dependencies in both forward and backward directions with linear computational complexity. Unlike conventional patch-based methods, MambaHSINet takes the entire hyperspectral image as input and produces full-resolution pixel-wise classification maps, thereby avoiding repeated cropping and redundant sliding-window inference. Experimental results on public and self-collected hyperspectral datasets demonstrate that the proposed method achieves a favorable balance among classification accuracy, inference efficiency, and model complexity, showing strong potential for practical tree species classification applications.The source code of this paper will be made available at https://github.com/YFENG-123/mambaHSI after the publication of the paper.

Abstract: Medicanes are mesoscale cyclones that develop over the Mediterranean Sea and display tropical-like cyclone characteristics, including a warm core, spiral cloud organization, and deep convection over warm sea surfaces. Since their structure and position can change rapidly on short lead times before coastal impact, robust near-real-time tracking algorithms are essential for timely warning and operational decision support. To advance this research direction, this work introduces the Deep learning Medicane Tracking (DeMeTrA) algorithm, an end-to-end deep-learning framework for medicane detection and rotation-center localization from SEVIRI Rapid Scan Airmass RGB imagery. The proposed methodology consists of three-stage VideoMAE v2 architecture encompassing: (i) self-supervised domain specialization on unlabeled satellite image sequences, (ii) supervised binary classification of cyclone versus non-cyclone events, and (iii) supervised coordinate regression for rotation-center tracking. The training corpus spans several time windows of Meteosat Second Generation observations across the Mediterranean basin, with ground-truth annotations derived from a consensus cyclone-track reference. On event-based splits, cyclones detection reaches 91% balanced accuracy on a balanced validation set, and 89% on an unbalanced test set representative of operational conditions. Tracking results show generally low localization errors (mostly below 20 km), with limited outliers in the most complex cases. These findings support the use of Transformer-based video models for operational medicane monitoring and establish a baseline for future developments.

Abstract: Preventing large-scale illegal migration is one of the EU's highest priorities. In this study, we analyse the potential for integrating and fusing remote sensor data with a wider range of data streams to enhance border security situational awareness, specifical-ly targeting illegal migration. The aim was to develop a dynamic predictive risk analysis model to identify high-risk zones for illegal border crossings at Croatia's external EU borders. The model’s methodological framework is based on the integration of Geo-graphic Information Systems (GIS), Multi-Criteria Analysis (MCA), and the Analytic Hi-erarchy Process (AHP). The model utilizes various environmental and infrastructure var-iables derived from the open-source databases ESA WorldCover and OpenStreetMap to generate a categorised risk map showing areas of lowest, moderate, and highest risk for illegal border crossing. High-resolution historical satellite imagery showing activities re-lated to illegal migration is used for model verification and generation of labelled da-tasets for AI training. Features such as suspicious vans, river boats, tyre tracks, tents, il-legal campsites, and clusters of individuals were observed in high-resolution Airbus and Maxar historical satellite images. The model can be used for various practical applica-tions, including the strategic allocation of surveillance resources and the enhancement of frontier and pre-frontier intelligence, enabling more informed actions and optimised op-erations.

Abstract: MODIS satellite images were used together weather data for large-scale water balance assessments in rainfed sugarcane crops, with a long-term data set from 2007 to 2024 in Northeast Brazil. Precipitation (P) was spatially quantified from interpolated pluviometer data and actual evapotranspiration (ET) estimated by applying the SAFER (Simple Algorithm for Evapotranspiration Retrieving). Considering the sugarcane cropped areas for each year, the mean annual P ranged from 3,172 million cubic meters in 2021 to 8,555 million cubic meters in 2009, while for ET this range was from 3,375 million cubic meters in 2017 to 5,869 million cubic meters in 2007. The assessments showed gaps between rainfall water supplies and the root-zone moisture levels, highlighting the evaporative fraction (Ef), i.e. the ratio of ET to reference evapotranspiration (ET0), as the best root-zone moisture indicator, which showed benefits of supplementary irrigation during the transition from grand growth to maturation stages. It was demonstrated that the assessments of large-scale sugar cane water balance using long-term series of data has potential to subsidize public policies which aim the sustainable crop management as well as for its rational expansion in areas with environmental aptitude under the actual climate and land-use changes scenarios.

Abstract: Accurate daily precipitation prediction in data-scarce tropical regions remains a critical challenge for climate monitoring, agriculture, and public health. Satellite products such as CHIRPS offer broad spatial coverage but exhibit systematic biases relative to ground-based observations—particularly in complex terrain under bimodal tropical regimes influenced by ENSO. We propose a Functional Generalized Additive Mixed Model (FGAMM) that corrects CHIRPS-derived precipitation estimates by treating the annual accumulated precipitation curve as a functional response and the satellite accumulation curve as a functional covariate, while incorporating station-level random effects and the Southern Oscillation Index. Applied to 62 IDEAM stations in the Valle del Cauca department of Colombia (2012–2020), the FGAMM achieves a mean cross-validation RMSE of 0.68 mm/day (95% bootstrap CI: 0.61–0.75), outperforming linear regression, SVM, and Random Forest by a factor of more than four; the performance gap is statistically significant across all competing methods. Because CHIRPS provides near-global daily coverage from 1981 to the present, the methodology is directly transferable to any tropical or subtropical region with a sparse reference station network, including areas of Latin America, sub-Saharan Africa, and South Asia where station density is similarly limited.

Abstract: Accurate characterization of riverbed substrate from remote sensing imagery is essential for applications in fluvial geomorphology, habitat modeling, and river management. While recent advances in computer vision, particularly deep learning, have improved sediment mapping capabilities, their reliance on large annotated datasets and computational resources limits their broader applicability. This study presents a scalable workflow for categorical substrate classification using ultra-high-resolution UAV-derived RGB orthoimagery in clear-water river environments. The approach integrates spectral information with statistical and structural texture descriptors derived from Gray-Level Co-occurrence Matrices (GLCM) and Local Binary Patterns (LBP), combined within a Random Forest classification framework. The methodology consists of two main steps: (i) manual annotation of homogeneous substrate patches within a standard GIS environment and (ii) automated feature extraction, model optimization, and full-domain classification. Model performance is evaluated using spatially aware cross-validation and design-based probability sampling to account for spatial autocorrelation and provide unbiased accuracy estimates. The method was applied in four geomorphologically distinct alpine river reaches, achieving design-based overall accuracy ranging from 69.77% to 95.21%. These results demonstrate that RGB-based approaches can achieve reliable reach-scale categorical substrate classification when combined with appropriate feature representation and rigorous validation strategies. However, limitations remain for visually similar or transitional substrate classes, particularly fine sediments such as sand and clay, which are difficult to distinguish consistently even during manual annotation. The workflow is implemented using open-source tools and is applicable to clear-water conditions where the riverbed remains optically visible.

Abstract: Coastal biodiversity conservation is challenged by fragmented datasets and the limited integration of environmental conditions into marine spatial planning (MSP). This study develops an operationalized adaptive Marine Spatial Planning (MSP) to support biodiversity conservation by linking remote sensing data, IoT-based water quality measurements, and spatial optimization within Spatial Decision Support System (SDSS). The Tidung Islands are used as a case study, where benthic habitats are mapped from 3 m PlanetScope imagery. Water quality observations are processed into the Nemerow Pollution Index (NPI) and subsequently interpolated through an ensemble approach that combines inverse distance weighting, random forest, and gradient boosting. A key innovation of this study is the incorporation of the Nemerow Pollution Index (NPI) as a dynamic environmental cost layer within Marxan-based conservation prioritization. These data were incorporated alongside anthropogenic pressures to evaluate multiple conservation scenarios The ensemble interpolation demonstrated strong predictive performance (R²=0.76;MAE=0.0306), enabling reliable spatial representation of environmental conditions. The results show that integrating environmental quality into MSP significantly improves spatial efficiency, reduces fragmentation, and enhances ecological representation compared to conventional approaches based on static variables. Moderate conservation targets (≈30%) produced the most optimal solutions (~2,200 cost; ~11 km boundary), while more ambitious targets resulted in fragmented and inefficient spatial configurations. The proposed framework offers a transferable approach for data-limited coastal regions, contributing to the advancement of adaptive biodiversity conservation strategies.

Abstract: Accurate rainfall estimates are essential for managing water resources and planning for climate risks in semi‑arid regions, yet long‑term gauge networks in these environments are often extremely limited. In this study, we evaluate three widely used multi‑source precipitation datasets; CHIRPS, IMERG, and ERA5‑Land, against long‑term observations from Ed Dueim and Kosti, the two main reference stations in White Nile State, central Sudan. The assessment covers monthly and annual scales across each product’s available record (1952–2022) and uses a broad set of metrics, including Pearson and Spearman correlations, NSE, KGE, RMSE, MAE, percent bias, and categorical detection scores (POD, FAR, CSI). All three datasets capture the region’s single‑peak June–October monsoon pattern, but their accuracy differs sharply when it comes to rainfall amounts and year‑to‑year variability. CHIRPS performs best overall, with monthly NSE values around 0.77 and KGE between 0.79 and 0.88, along with a consistent dry bias of 5–13%—a predictable error that can be corrected operationally. IMERG shows strong monthly correlations but consistently overestimates rainfall by 25–42%, which leads to unreliable annual totals (NSE = −1.93 to −2.21). ERA5‑Land performs worst across nearly all metrics, with monthly NSE near or below zero, annual NSE dropping to −15.34, and frequent false alarms during the dry season. Taken together, the evidence points to CHIRPS as the most reliable dataset for routine hydro‑climatic monitoring in White Nile State, while IMERG and ERA5‑Land may still be useful in more specialized or time‑specific applications.

Abstract: The "spin-up" problem—where convection-permitting models require hours to develop realistic clouds from large-scale initial fields—critically limits short-term severe weather forecasting. Cloud analysis offers a potential solution by directly incorporating hydrome-teor information from remote sensing observations. In this study, we leverage multi-source remote sensing data, including three-dimensional mosaic radar reflectivity, hourly aver-aged FY-2G satellite black-body temperature (TBB), and FY-2G total cloud water products, within a stepwise cloud-analysis initialization scheme. The scheme is implemented in a convective-scale ensemble forecasting system (CMA-Meso, 3 km resolution) for a heavy rainfall event. For each ensemble member, three-dimensional hydrometeor increments are independently generated from these remote sensing retrievals and gradually introduced over the first ten time steps, ensuring smooth coordination with the model's dynam-ic-thermal framework. Results demonstrate that the remote sensing-driven cloud analysis substantially enhances ensemble system performance across multiple dimensions: (i) spin-up time is significant-ly reduced, with precipitation forecasts exhibiting reasonable structure from the initial forecast hour; (ii) deterministic forecast accuracy improves systematically, with reduced RMSE for geopotential height, temperature, and wind fields across all levels; (iii) proba-bilistic forecasting skill is enhanced, evidenced by improved CRPS and AROC for surface elements and precipitation thresholds; (iv) ensemble reliability is optimized, with spread better matching forecast errors. Mechanistic analysis reveals that these improvements stem from physically coordinated hydrometeor-latent heat initial perturbations and sub-sequent cloud-radiation feedbacks that continuously regulate thermal-dynamic structures. This study establishes that assimilating diverse remote sensing data via cloud analysis is an effective approach for addressing spin-up challenges in convective-scale ensemble prediction.

Abstract: Floodplain wetlands are dynamic and biodiverse environments that provide important ecosystem services. This study analyzes the temporal and spatial dynamics of hydrogeomorphological attributes, vegetation, and water in floodplain wetlands. The methodology consisted of applying PCA in temporal (T) and spatial (S) modes, decomposing spectral indices (NDVI, NDMI, MNDWI) to identify variability patterns associated with ENSO events. The results revealed that C2 was the main descriptor of hydrological anomalies, with strong temporal synchrony between vegetation vigor (NDVI) and the expansion of the water surface (MNDWI), contrasting with the water stress response captured by NDMI. PCA highlighted environmental heterogeneity within the floodplain, with peatland areas standing out as zones of high spatial complexity and greater water retention capacity. Temporal variability responded primarily to climatic extremes, whereas spatial variability was modulated by hydrogeomorphology.

Abstract: AlphaEarth Foundations (AEF) unifies multi-modal and multi-temporal observations into analysis-ready low-dimensional representations, introducing a representation-driven paradigm that addresses key limitations of traditional task-centric approaches, including high engineering complexity, limited cross-regional transferability, and strong dependence on labeled data. This paper presents the first systematic review of AEF, synthesizing 23 research articles published up to March 2026 from three perspectives: conceptual positioning, technical framework, and application practices. It first clarifies the distinctions between AEF and related paradigms, including foundation models, remote sensing large models, and existing unified embedding approaches. It then summarizes the organization of its data system and key technical components. Based on a systematic literature survey, the paper further provides a structured synthesis of current studies in terms of thematic and regional distribution, scientific questions, usage patterns, and evaluation methods. The analysis indicates that AEF represents an important step toward a paradigm shift from task-driven to representation-driven Earth observation, offering clear advantages in reducing engineering barriers, enabling cross-regional transfer, and establishing a unified environmental semantic foundation. However, limitations remain in temporal resolution, semantic interpretability, and performance in specific task scenarios. Future work should focus on enhancing dynamic representation capabilities, cross-domain adaptation mechanisms, multi-source integration frameworks, systematic evaluation and annotation systems, and asset-oriented applications. By delineating the capability boundaries and applicable contexts of AEF from both methodological and empirical perspectives, this study provides a systematic reference for the standardized development and rational application of unified surface embeddings as an emerging geospatial information infrastructure.

Abstract: Precise assessment of landslide potential in tectonically active mountain areas like Darjeeling Sikkim Himalaya (DSH) is a scientific challenge due to the complexity of different landslide conditioning factors that control the slope stability. Despite several studies for landslide susceptibility mapping, most of the conventional methods struggle to capture the nonlinear relationships and spatial heterogeneity that characterize landslides. Besides, the current use of pixel-based methods is insufficient to depict geomorphological units and slope-scale processes, thus limiting their effectiveness in boundary demarcation of landslide-prone areas. These limitations highlight the need for more robust machine learning frameworks that integrate geomorphology-based terrain segmentation with advanced machine learning models, which would not only facilitate modeling the multifaceted interactions among environmental components but also improve the understanding of the landslide driving forces. In this study, we have used slope unit based landslide susceptibility mapping with 4380 slope units integrated with 17 conditioning factors, and 8373 total updated inventories using six models Random Forest (RF), Generalized Additive Model (GAM), Categorical Boosting (CatBoost), Tabular Neural Network (TabNet), Bayesian Additive Regression Trees (BART), and Convolutional Neural Network (CNN). The model hyperparameters were optimized using Bayesian optimization, except for the BART model. Among the six models, RF (AUC = 0.848) and CatBoost (AUC = 0.846) were the best two performing models. Furthermore, SHAP analysis reveals that elevation, aspect, slope, distance to faults, NDVI, and proximity to roads and drainage networks are the main landslide controlling factors in DSH. The interaction analysis using SHAP indicates that the occurrence of landslides is controlled by nonlinear and threshold-dependent relations, especially among slope-rainfall, rainfall-soil moisture, and slope-distance to roads and faults, which represents a complex interaction between the hydrological triggering factor, geomorphic processes, tectonic activity, and human interventions.