In all-weather Earth observation and complex unstructured environments, traditional single-modal remote sensing object detection often fails due to low illumination and strong background interference. While RGB–infrared fusion provides complementary information, existing methods are typically computationally intensive and struggle with dense small objects and modality discrepancies, limiting their deployment on resource-constrained platforms. To address these challenges, we propose ISDG-Net, a lightweight and efficient visible-infrared dual-modal object detection framework specifically tailored for edge deployment. ISDG-Net integrates four core components: (1) a channel-separated inverted bottleneck backbone (IBC-Conv) that reduces parameter redundancy while preserving modality-specific semantics; (2) a dynamic sparse attention module (DySparse) based on Bi-Level Routing Attention, enabling long-range dependency modeling with low computational cost; (3) an adaptive spatial fusion detection head (Detect-SASD) that aligns visible and infrared features at the pixel level to resolve semantic inconsistency and scale mismatch; and (4) a geometry-aware IoU selector (GIS) that mitigates over-suppression in crowded scenes by incorporating multi-dimensional geometric constraints into post-processing. Extensive experiments on the VEDAI, M3FD, and LLVIP datasets demonstrate the effectiveness and efficiency of ISDG-Net. It achieves 55.1% and 77.1% mAP@0.5 on VEDAI and M3FD, respectively, and 93.7% mAP@0.5 with 89.7% recall on LLVIP, while maintaining a compact model size of 4.2 M parameters and 11.3 GFLOPs. These results validate that accurate RGB–infrared detection is achievable under strict resource constraints, making ISDG-Net well-suited for deployment in edge-based remote sensing systems.

Effective monitoring of surface damage in mining areas is vital for ecological restoration. Unmanned aerial vehicles (UAVs) have been widely used to obtain ground subsidence data owing to their low cost and ease of operation. The images captured by UAVs can generate dense image matching (DIM) point clouds, which, after screening, can be used to create a digital elevation model (DEM) required for deformation analysis. Existing filtering algorithms mainly rely on the spatial geometric features of point clouds and rarely utilize color information, which limits their accuracy in areas with vegetation coverage. To address this issue, this study proposes a H-PTD method that combines visible light with progressive triangulated irregular network densification (PTD). First, initial ground seeds are selected based on the H value in the HSV space. Subsequently, a triangulated irregular network (TIN) is constructed, and iterative densification is performed by evaluating the relationship between the target point and adjacent triangular faces, thereby achieving an accurate distinction between ground and non-ground. Evaluated on three terrain datasets and against five classical methods, the results indicate that the Total error in the H-PTD cross-matrix is controlled between 2.9% and 7.8%, and remains below 8% overall. The standard deviation of the DEM difference is around 0.02 m. Compared to other methods, H-PTD shows higher filtering accuracy and better terrain adaptability, making it more promising for monitoring mining areas and providing a more reliable tool for subsidence detection based on UAVs.

Fine-grained object detection (FGOD) is crucial for identifying visually similar sub-categories in remote sensing imagery. However, existing detectors suffer from severe supervision imbalance because static label assignment strategies assign a fixed number of positive samples to all sub-categories and targets. To address this challenge, this paper presents Cumulative Quality-based Dynamic Assignment (CQDA), a fine-grained aware label assignment algorithm that dynamically calculates the optimal positive budget for each instance based on its cumulative alignment quality. Moreover, to further resolve feature-space confusion, this paper introduces two modules: a frequency-decoupled enhancement algorithm to sharpen discriminative features, and an orthogonal classification head to maximize inter-class separability. Integrated into the KFIoU framework, extensive experiments demonstrate that the proposed method consistently achieves performance improvements of 4.2, 15.8, and 35.3 in mAP@0.5 on the fine-grained oriented object detection datasets FAIR1M-v2, MAR20, and ShipRSImageNet, respectively.

Accurate nowcasting of thunderstorm gales, a highly destructive form of severe convective weather, is critical for mitigating wind-related disasters and ensuring the safety of life and property. Existing deep learning approaches face challenges such as performance decay at high wind speed thresholds, limited capability in capturing extreme events, and difficulties in processing high-resolution data. To address these issues, this paper proposes a novel physics-driven multi-source fusion recurrent neural network named MSF-PhyDRNN. The model incorporates a multi-source fusion module that integrates radar composite reflectivity and surface wind field data through feature decoupling and hierarchical fusion. Additionally, we improved the recurrent unit in PhyDNet to enhance short-term wind capture and reduce redundancy, leveraging its cascaded memory and spatiotemporal propagation mechanisms. Experimental results indicate that, compared to the advanced MFWPN model, MSF-PhyDRNN achieves an average increase of 14.3% in the Critical Success Index (CSI), 27.2% in the Probability of Detection (POD), and 19.7% in the Heidke Skill Score (HSS) across the Jiangsu and South China datasets.

Forest canopy water storage capacity is a critical component of ecohydrological research. However, because most current studies focus on the plot or stand scale, upscaling these fine-scale measurements to regional spatial scales remains a major challenge. Taking the forest in southern Jiangxi province as a case study, we integrated water immersion experiments, Handheld Laser Scanning (HLS), Unmanned Aerial Vehicle LiDAR (UAV-LiDAR), and optical remote sensing data to construct a spatial upscaling model. This model aims to quantify regional canopy water storage capacity and delineate its spatial patterns. The results indicate that: (1) the water storage capacity of branches and leaves per unit surface area of coniferous trees was significantly higher than that of broad-leaved trees, and the water storage capacity of branches was 6.0–10.7 times that of leaves. The mean canopy water storage capacities of coniferous forests, mixed coniferous and broad-leaved forests, and broad-leaved forests were 1.41 ± 0.27 mm, 1.30 ± 0.45 mm, and 1.26 ± 0.36 mm, respectively. (2) The canopy water storage capacity was significantly positively correlated with canopy volume (VC) and average canopy area (AC) extracted from UAV-LiDAR data, and vegetation structure factors such as normalized difference vegetation index (NDVI) and vegetation cover (FVC) extracted from optical remote sensing, and significantly negatively correlated with altitude and slope. Among them, canopy closure (C), average canopy area (AC), and altitude were key factors affecting canopy water storage capacity. (3) The upscaling prediction models based on UAV-LiDAR data and optical remote sensing factors, respectively, show reliable prediction performance, with R2 values of 0.884 and 0.815, RMSE of 0.951 and 0.116 mm, respectively. (4) The canopy water storage in the study area ranged from 0 to 1.76 mm, with a prediction uncertainty ranging from 0.12 to 0.49 mm. Canopy water storage is higher in the continuous middle and low mountain and hill areas within the region, while it is relatively lower in the high elevation ridge areas along the western, eastern, and southern margins. The results provide baseline structural information for understanding the spatial patterns of regional forest canopy interception potential.

Atmospheric heatwaves (AHWs) and marine heatwaves (MHWs) are intensifying under climate change, yet their coupled behavior in the Red Sea remains insufficiently quantified. This study investigates the spatial and temporal characteristics of AHWs, MHWs, and their concurrent occurrence across the Red Sea from 1990 to 2024 using ERA5 surface air temperature (SAT) and NOAA OISST v2.1 satellite-derived sea surface temperature (SST). Remote-sensing daily satellite-derived Level-4 (L4) OISST products were used in this study to enable spatially complete and temporally consistent detection of MHWs in this narrow, semi-enclosed basin despite contamination and coastal sampling constraints. Both SAT and SST exhibit statistically significant warming trends (p < 0.05), with basin mean increases of 0.40 ± 0.07 °C/decade and 0.31 ± 0.05 °C/decade, respectively. The strongest warming was observed in the central and northern Red Sea. This warming is accompanied by significant increases in the frequency and duration of AHWs, MHWs, and their concurrent AHW-MHW events, particularly after 2010, indicating a shift toward more frequent heatwave conditions. AHWs occur more frequently than MHWs across the Red Sea, whereas MHWs exhibit long duration, particularly in the northern Red Sea, where annual durations exceed 45–50 days/year. Concurrent AHW-MHW events account for about 66% of MHWs in the Red Sea, and their characteristics show a significant increasing trend across the entire basin. These findings identify the Red Sea as a regional hotspot of increasing concurrent heatwave events and highlight the importance of satellite-based monitoring for assessing evolving climate risks in semi-enclosed basins.

Milltailings deposition areas in abandoned mines are inherently unstable and spatially extensive and heterogeneous, making regional-scale field investigations challenging under intense rainfall. With the advancement of space–airborne remote sensing technologies, large-scale surface deformation monitoring has become feasible. In this study, a 22.02 km² abandoned mine in Lingqiu County, Shanxi Province, was selected as a case site; during the late-July 2023 extreme rainfall event, the site experienced large-scale surface displacements. Surface deformation was interpreted using Sentinel-1 SBAS-InSAR data, combined with differential digital elevation models (DEMs) derived from UAV surveys before and after heavy rainfall. A bivariate spatial autocorrelation analysis was conducted to evaluate the spatial relationship between differential DEMs and InSAR-derived deformation. The results indicate that: (1) SBAS-InSAR revealed significant spatial heterogeneity of ground deformation, with pronounced subsidence observed in the milltailings deposits; (2) the bivariate spatial autocorrelation analysis yielded a Moran’s I value of 0.2, suggesting a weak but positive spatial correlation between the DEM differences and InSAR results, with dispersed correlation patterns; (3) hotspot analysis highlighted notable clustering of deformation, with approximately 27.84% of the study area showing strong deformation responses, while 25.81% represented low–low clusters with limited deformation. Beyond tailings-deposit settings, this workflow is also applicable to the regional investigation of rainfall-responsive deformation and debris-flow-related terrain change on natural slopes under global change, providing technical support for surface investigations and offering insights for disaster early warning and ecological restoration in similar regions.

Accurately monitoring the surface stabilization of waste dumps in open-pit coal mines is critical for hazard prevention and ecological reclamation. In arid and semi-arid regions, traditional optical remote sensing vegetation indices suffer from a systematic “response lag” in assessing physical stability due to the slow establishment of pioneer vegetation. To overcome this biological limitation, this study proposes a quantitative spatiotemporal monitoring framework based on time-series Interferometric Synthetic Aperture Radar (InSAR) coherence to detect early-stage geotechnical stabilization. Using Sentinel-1 imagery of the Balongtu coal mine, a sliding-window detection algorithm was developed to capture the physical transition of surface electromagnetic scattering mechanisms from active disturbance to stable consolidation. The main findings are as follows: (1) Statistical analysis identified a critical geophysical coherence threshold of 0.15, which effectively and objectively distinguishes active dumping disturbance zones from structurally stable areas. (2) The spatiotemporal evolution dynamics of the completed dump areas from 2017 to 2023 were successfully characterized, revealing that 87.6% of the open-pit areas achieved physical stabilization within three years post-mining, with a spatial distribution highly consistent with the objective operational rule of “mining first, dumping later”. (3) Accuracy assessment using 700 spatiotemporally balanced validation points—derived through strict visual interpretation of high-resolution optical imagery—demonstrated high algorithm reliability, achieving overall accuracies (OA) of 87.57% and 90.43% at half-yearly and annual monitoring intervals, respectively. By decoupling physical surface stabilization from optical greenness, this study provides a timely abiotic precursor indicator, offering scientific, quantitative decision support for precision ecological zoning and accelerated land turnover approval in mining areas.

As fundamental units of forest ecosystems, individual trees provide essential structural characteristics for forest resource assessment. However, existing LiDAR-based individual tree segmentation methods are often limited by a trade-off between information preservation and computational efficiency. This study proposes a novel framework for individual tree segmentation from LiDAR data based on canopy surface points (CSP), aiming to balance this trade-off. The framework introduces a Snow-Covered Filter (SCF) that simulates snow deposition to extract surface points from the point cloud. After removing ground points from these surface points, the resulting CSP retains the core 3D structure of the canopy while significantly reducing data volume. We validate the proposed framework on four multi-platform datasets using four algorithms that represent the evolution of individual tree segmentation methods: Dalponte2016, K-means, Li2012, and SegmentAnyTree. The results demonstrate that: (a) the SCF effectively extracts surface points, with an average F1-score of 0.703; (b) segmentation using CSP achieves accuracy comparable to that obtained using all points or raster data (mean ΔF = 0.027), with the primary gap observed for SegmentAnyTree (maximum F-score reduction of 0.259); (c) the framework offers substantial efficiency gains: >40% point reduction, ~38.4% average runtime reduction (maximum saving ~4660 s), and lower memory consumption. By providing a lightweight yet structurally rich data representation, this work presents an innovative and efficient approach to individual tree segmentation, with promising potential for large-scale forest resource management.

Hyperspectral unmixing aims to estimate endmember spectra and their corresponding abundance fractions at the subpixel scale, which is a critical preprocessing step for quantitative analysis of hyperspectral remote sensing imagery. While deep learning-based methods have achieved remarkable progress, three fundamental challenges remain: (i) reliance on a single shared spatial prior that cannot decouple the heterogeneous spatial patterns of different land covers; (ii) the lack of synergy in jointly optimizing endmember extraction and abundance estimation; (iii) the poor robustness of unsupervised training to complex mixtures, noise, and class imbalance. To address these issues, we propose a novel unsupervised unmixing framework that integrates adaptive orthogonal multi-faceted graph representation with curriculum learning. Specifically, we design an Adaptive Orthogonal Multi-Faceted Graph Generator (AOMFG) to learn a set of independent orthogonal graph structures, achieving spatially informed decoupling of land cover patterns. Then, a dual-branch collaborative optimization network is constructed: a Graph Convolutional Network (GCN) branch that incorporates the learned spatial topological priors for abundance estimation, and a 1D Convolutional Neural Network (1DCNN) branch that employs a query-attention mechanism to adaptively aggregate pure spectral features for endmember extraction. Finally, we introduce a three-stage curriculum learning strategy that progressively fine-tunes the model, which significantly enhances its performance. Extensive experiments on three widely used real-world benchmark datasets demonstrate that our proposed framework consistently outperforms state-of-the-art methods in both endmember extraction and abundance estimation accuracy. Comprehensive ablation studies, parameter sensitivity analysis, and noise robustness tests further validate the effectiveness of each core component.