Remote Sensing doi: 10.3390/rs18203567

Authors: Pengfei Zhang Leigang Sun Qinyi Wang Qingtao Hao Xiaofeng Jia Junjing Lu Lu Zuo

Mamba has recently become a popular visual representation architecture for building change detection (BCD) in the remote sensing field. Although it effectively models global spatial dependencies, it still struggles to capture fine-grained local structures, thereby limiting its representation capability. To overcome this limitation, we propose ChangeMoE, a mixture-of-experts-based Mamba network for BCD from high-resolution remote sensing imagery. The main contributions and innovations are as follows: (1) A scan-specific mixture-of-experts (ScanMoE) module is proposed, which leverages a scan gating network to enhance global feature representations using eight selective scan mechanisms defined as scan-specific experts. We further quantify their contributions to improve the interpretability of global representation learning. (2) A global-and-local mixture-of-experts (GLMoE) module is proposed. We design the local and global expert networks to capture local details and global dependencies. A gating network is used to reveal their relative importance and adaptively fuse their complementary features. Experimental results on the LEVIR-CD and WHU-CD benchmarks validate the superiority of ChangeMoE over existing state-of-the-art networks. ChangeMoE achieves F1-scores of 92.55% and 94.11%, with IoU values of 86.13% and 88.88%, respectively. Compared with ChangeMamba as the baseline, the proposed ChangeMoE achieves improvements of 1.96% in F1-score and 3.33% in IoU on the LEVIR-CD dataset. These results demonstrate the effectiveness of the proposed ChangeMoE and provide an interpretable expert-fusion paradigm for understanding feature representation learning in BCD.

Remote Sensing doi: 10.3390/rs18203566

Authors: Fangze Hu Jinling Song Jing Wang Xiaoyu Zhou

Forest resources are globally valuable and critical in many ecological processes. The forest canopy remote sensing model is an essential tool for understanding forest growth conditions. Over time, the model has evolved to consider the effects of the vertical heterogeneity of forests and understory vegetation on radiative transfer. Despite the effectiveness of the mainstream three-dimensional radiative transfer model, it still faces challenges related to parameter complexity and poor timeliness. To address these issues, we developed a two-layer model (Two-Layer GORT) based on the geometric optical-radiative transfer hybrid (GORT) model. The Two-Layer GORT extends the GORT model to include a layer of understory vegetation under the canopy. The effectiveness of the Two-Layer GORT model is evaluated by simulating the gap probability and BRDF of the North China larch (Larix principis-rupprechtii) sample plot in the Saihanba Jixie Forest Farm, Hebei Province. Additionally, the model is compared to the model in the RAMI project to evaluate its effectiveness. The results indicate that the Two-Layer GORT model has a good ability to simulate the radiative transfer process, demonstrating the potential of the GORT model to simulate increasingly complex scenarios. This study is effective for understanding the growth state of a forest and sets the future goal of constructing a generalized two-layer model that considers scenarios with increasing complexity.

Remote Sensing doi: 10.3390/rs18203565

Authors: Yilinuer Alifujiang Jiale Chen Alim Samat Ping Yang

Arid inland river basins are ecologically fragile zones where habitat status is jointly shaped by anthropogenic land use interference and water-resource constraints; however, the combined effects of land use change and multi-scale hydrological conditions on habitat quality remain insufficiently understood. Taking the Kaidu–Kongque River Basin (KKRB) in northwest China as the study area, this study adopted 30 m CLCD land-cover datasets and a localized-parameter improved InVEST model to investigate habitat-quality dynamics from 2000 to 2023, combined with the PLUS model for multi-scenario simulation. Results revealed prominent “productive-land expansion and ecological-land contraction”: cultivated land increased by 36.2% and construction land expanded 10.3-fold, mainly encroaching on grassland, forming a typical “upstream improvement-downstream degradation” spatial pattern of habitat quality. Based on correlation analysis, we identified cross-scale ecohydrological coupling characteristics: meteorological water deficit exhibited a strong linear correlation with habitat quality at the basin scale, while surface soil moisture displayed an inverted-U-shaped non-linear trade-off relationship with habitat quality at local scales, jointly mediating spatial heterogeneity of habitat quality in this arid basin. Future scenario simulations indicated that only the Ecological Protection Scenario could substantially reverse degradation, with mean habitat quality projected to increase by 21.3% by 2035 relative to the natural-development scenario. This study clarifies how land use alteration and multi-scale water constraints jointly shape habitat-quality patterns, providing scientific support for land use optimization and ecological restoration in arid inland watersheds.

Remote Sensing doi: 10.3390/rs18203564

Authors: Yujin Lai Zhigang Lu Xin Lan Qing Tang Yu Li Fang Li Rui Wang

Landslide inventory mapping supports emergency response and hazard assessment in mountainous regions. Accurate identification and delineation of individual landslides remain challenging because of scale variation, irregular morphology, and interference from visually similar backgrounds in complex mountainous terrain. To address these challenges, we modified Mask R-CNN’s feature pathway and mask supervision. ResNet50 was replaced with SwinV2-Tiny for hierarchical feature extraction, while the feature pyramid network (FPN) was retained for multi-scale fusion. Spatially adaptive feature modulation (SAFM) was applied after the FPN to adjust spatial responses using local features and broader context. Binary cross-entropy (BCE) and Dice losses were combined to supervise pixel-wise foreground–background discrimination and instance-mask overlap. A curated dataset combining imagery from Bijie, the Diverse Mountainous Landslide Dataset (DMLD), and ScienceDB was used for training, validation, and internal testing. All configurations were evaluated using the Common Objects in Context (COCO) instance-mask protocol. Across three training seeds, SwinV2–FPN–SAFM Mask R-CNN achieved an internal test average precision (AP0.50:0.95) of 0.6158±0.0082 (mean ± sample standard deviation). Its mean mask AP0.50:0.95 exceeded those of ResNet50–FPN Mask R-CNN and Mask2Former–Swin-T by 6.21 and 3.51 percentage points, respectively. It achieved the highest mean mask AP0.50:0.95 among the ten evaluated configurations. The ablations indicated small positive increments in mean mask AP0.50:0.95 with SAFM and BCE–Dice, and their incremental benefits remained uncertain. The higher mean mask AP0.50:0.95 indicates improved overall performance in identifying landslide instances and delineating their extent on the internal test set. The resulting instance masks may assist inventory compilation and post-disaster mapping in complex mountainous terrain.

Remote Sensing doi: 10.3390/rs18203562

Authors: Xingxin Song Bing Zhou Jiaju Ying Jiale Zhao Lei Deng Qi Chen

Hyperspectral camouflaged target detection is challenging because target and background spectra are highly coupled, positive pixels are extremely scarce, and scene-dependent spectral variation limits cross-dataset robustness. Existing detectors often rely on a single spectral representation or use spatial models with substantial parameters and computation, which may discard complementary evidence or increase deployment costs. This article presents a lightweight dual-branch target-aware network for land-based hyperspectral camouflage target detection. The model preserves fine raw spectral details and whitened PCA features while integrating multi-scale spatial context. A target-aware spatial gate is further employed to suppress background-like responses, and a center-aware decision head alleviates the dilution of small targets during patch pooling. Class-weighted loss and balanced sampling address severe class imbalance. On four land-based hyperspectral camouflage datasets with 1% training, 4% validation, and 95% testing pixels, the proposed method achieves F1 scores of 0.9438–0.9601, PR-AUC values of 0.9814–0.9938, and false-alarm rates of 0.0002–0.0003, outperforming eight representative detectors. Ablation studies confirm the complementarity of the raw and PCA branches and the effectiveness of the target-aware gate and center-aware head. In cross-dataset evaluation with only 0.5% target pixels for fine-tuning and another 0.5% for validation, PR-AUC remains between 0.8833 and 0.9288. The model requires only 0.0556 M parameters and 8.0699 M FLOPs per patch, indicating a favorable trade-off between accuracy, generalization, and lightweight deployment.

Remote Sensing doi: 10.3390/rs18203563

Authors: Yanqiang Wang Yechuan Zhu Tingyu Yan Shiping Guo Ke Ke Shun Zhou

Accurate temperature measurements in the mesosphere and lower thermosphere (MLT) are important for investigating atmospheric dynamics and vertical coupling. Temperature retrieval from the OI 557.7 nm airglow interferometry requires an appropriate optical path difference (OPD), while the measured fringe visibility is also affected by variations in instrumental visibility. To address the coupling between these two effects, this study develops a dual-OPD interferometric method in which the line-of-sight effective temperature and common instrumental-visibility variation are jointly considered in the retrieval. The OPD pair is selected according to both temperature-dependent coherence sensitivity and the expected instrumental-visibility uncertainty. Under a baseline uncertainty of 1%, a 12.0/13.5 cm configuration provides temperature retrieval precision comparable to that of an optimized single-OPD measurement under the same total photon budget. Under the adopted forward-model assumptions, end-to-end Monte Carlo simulations with noise introduced directly into four phase-stepped interferograms show close agreement with the analytical uncertainty model and yield a maximum simulated temperature RMSE of approximately 3 K over 140–210 K at an interferogram SNR of 150. More importantly, as instrumental visibility becomes less well-constrained, the preferred dual-OPD configuration changes from two closely spaced temperature-sensitive OPDs to a widely separated pair. The shorter OPD then provides a reference-like constraint on the common instrumental response, while the longer OPD retains the main temperature sensitivity. Consequently, when instrumental-visibility variations are predominantly common to the two channels, the re-optimized dual-OPD design degrades more slowly than the optimized single-OPD design as instrumental-visibility uncertainty increases. Additional sensitivity analyses show that channel-dependent instrumental-visibility variations progressively weaken this benefit, while residual background-subtraction errors can introduce systematic temperature bias. The results show that dual-OPD sampling can retain comparable photon-limited precision to an optimized single OPD under well-constrained instrumental visibility while providing greater robustness to predominantly common-mode instrumental-visibility uncertainty. These results provide design-level guidance for selecting fixed OPDs according to the expected atmospheric temperature range and instrumental stability.

Remote Sensing doi: 10.3390/rs18203561

Authors: Jacob S. Nadolsky John L. Schroeder Brian D. Hirth

Given its large range and spatial footprint, radar remote sensing has been met with increased interest in the wind energy community. Paired radars can yield complete wind speed and direction measurements over this large spatial footprint, an invaluable asset for wind farm operators, and also the main responsibility of the Texas Tech University X-band radars installed for the American WAKE ExperimeNt (AWAKEN). However, care must be taken in how these measurements are quality-controlled and understood. This study examines the data filtering algorithms used in AWAKEN, and how their refinement can characterize atmospheric flows around wind farms differently. The original filtering method employed in AWAKEN was found to allow some non-atmospheric measurements to remain while also expunging other, higher-quality measurements, yielding a high bias in radar wind measurements. Closer inspection of the filtering process and its output in a separate field experiment led to the discovery of biological or non-atmospheric tracers influencing radial velocity fields, and to the development of a new algorithm that more effectively removes low-quality and non-atmospheric data while preserving high-quality data. When applied to the AWAKEN dataset, comparisons between radar data processed with the new algorithm and a ground-based lidar in the radars’ overlapping scanning areas show substantial reduction in the previously noted radar bias, demonstrating the utility of the new algorithm and the sensitivity of radar measurements to processing techniques.

Remote Sensing doi: 10.3390/rs18203560

Authors: Tao Long Xiaojuan Qin Runbin Zhang Xiao Ke Yunchao Jiang Bo Wang

Seasonal snow regulates spring water supply and vegetation development, thereby shaping vegetation responses to summer compound dry–hot events. Whether snowmelt timing relative to event onset explains the contrast between peak damage and cumulative deficit remains unresolved. Regional observations from 2006 to 2025 were integrated with high-resolution satellite data across Northwest China to combine regional association analyses with local comparisons within the same events. Snowmelt completion timing, the snowmelt–event interval, antecedent root-zone moisture, and local snow–terrain conditions were evaluated. Earlier snowmelt was associated with lower peak damage but greater seasonal cumulative vegetation deficit. As the snowmelt–event interval increased from 56 to 148 days, cumulative deficit decreased by 9.30 z·observations. The mean standardized root-zone moisture residual declined from 0.12 during the initial post-snowmelt period to −0.43 in the last valid period. In local annual-baseline comparisons, late-snow candidates showed greater peak damage and cumulative deficit than matched controls. Under an event-preceding baseline, candidate pixels had 0.0489 lower cumulative EVI2 deficit than controls across 77 event-sensitive samples, and this direction persisted after sequentially excluding mountain systems. Future scenarios showed jointly high peak damage and cumulative deficit in the Altai under some climate combinations. These results provide evidence for incorporating snowmelt preconditioning into assessments of mountain vegetation risk that account for both event impacts and seasonal losses.

Remote Sensing doi: 10.3390/rs18203559

Authors: Mustapha Meftah Christophe Dufour Tsovinar Karapetyan Alain Sarkissian Ping Zhu Nahla Ben-Naghrouzi Pierre-Richard Dahoo Philippe Keckhut

Earth’s energy imbalance (EEI) is the small residual between absorbed solar radiation and outgoing thermal radiation at the top of the atmosphere (TOA). We analyse global monthly fluxes from the Clouds and the Earth’s Radiant Energy System (CERES) Energy Balanced and Filled (EBAF) Ed4.2.1 record from March 2000 to March 2026, together with complementary solar, cloud, greenhouse gas, El Niño–Southern Oscillation (ENSO), sea surface temperature (SST), sea-ice, and ocean heat content datasets. Over 2001–2025, the CERES net TOA radiative flux (NET) anomaly relative to 2001–2020 shows a trend of +0.451±0.196Wm−2decade−1 (95% confidence interval; R2=0.496). The algebraic decomposition of this trend using N=ISR−OSR−OLR, where ISR, OSR, and OLR denote incoming solar radiation, reflected shortwave radiation, and outgoing longwave radiation, respectively, yields component contributions of +0.071, +0.806, and −0.426Wm−2decade−1. The reduction in reflected shortwave radiation is therefore the largest positive contribution to the observed NET trend. Raw annual correlations with greenhouse gas indicators, SST, and sea ice are substantially weakened after detrending. When accounting for annual serial correlation and false discovery rate (FDR) control, only CERES cloud area and ocean heat uptake from the Estimating the Circulation and Climate of the Ocean (ECCO) state estimate retain significant annual associations with NET. Detrended monthly lag analyses further show significant negative relationships between CERES NET and the Niño-3.4 and MEI.v2 indices, with the strongest correlations occurring when the ENSO indices precede NET by approximately 6–7 months. Over the strictly common 222-month interval (July 2005–December 2023), CERES NET shows strong temporal covariability with complementary heat-uptake estimates. The correlation coefficients over this common interval are r=0.857 for ECCO ocean heat uptake and r=0.804 for the planetary heat-storage estimate including atmospheric energy tendency. The corresponding detrended correlations are r=0.760 and r=0.640, respectively. The Uvsq-Sat NET record overlaps CERES observations, while Uvsq-Sat NG continues this measurement sequence. Its current time series remains too short to support climatological or long-term trend analyses and is therefore not yet used for such analyses. These results motivate a hybrid observing architecture combining reference radiometers, cross-calibrated broadband nano-satellites, targeted spectral measurements, and planetary heat-storage estimates.

Remote Sensing doi: 10.3390/rs18203558

Authors: Xueru Liang Yueqian Shen Jinhu Wang Jinguo Wang Yanming Chen Dong Chen

Accurate fallen tree detection from airborne laser scanning (ALS) point clouds is important for forest structural assessment and disturbance monitoring. However, fallen trees remain difficult to extract from operational ALS data because near-ground stems are often fragmented by understory occlusion, uneven sampling, and geometric confusion with shrubs, terrain surfaces, and other low vegetation. This study proposes a physics-guided supervoxel framework for fallen tree detection from ALS point clouds. The framework represents ALS observations as spatially coherent supervoxels and describes each supervoxel using local geometric features, cylindrical morphology, and graph-based neighborhood context. Cylindrical descriptors encode the elongated stem structure of fallen trees, while graph-topological features represent spatial continuity among adjacent supervoxels. A stacking ensemble classifier is trained with SHAP-based feature selection, followed by graph-constrained spatial refinement to improve local label consistency. The framework was trained using 30 plots and evaluated on seven held-out forest plots from the Dutch Actueel Hoogtebestand Nederland 4 (AHN4) dataset, with point densities of 10–15 pts/m2. Using point-level reference labels, the stacking classifier achieved a mean fallen tree F1-score of 83.86% across the test plots, which increased to 85.39% after spatial refinement. Ablation results show that cylindrical descriptors provided the largest gain, improving the fallen tree F1 score by 11.77%. These results indicate that physics-guided supervoxel learning can produce interpretable fallen tree maps from operational ALS data and support subsequent analyses of near-ground forest structure and disturbance effects.

Remote Sensing doi: 10.3390/rs18203556

Authors: Eastia Sahanama Zhen Guo Jing Yu Zhiwei Zhang K M Azam Chowdhury Jiawei Shen Yuhang Qi

Mangrove ecosystems are increasingly threatened by anthropogenic disturbances and climate change, despite providing ecological services, including biodiversity conservation, coastal protection, and blue carbon storage. Long-term evaluations of mangrove dynamics and associated environmental conditions remain limited, particularly in the Sundarbans, the largest mangrove forest in the world. This study examined the spatiotemporal dynamics of the Sundarbans mangrove ecosystem, Bangladesh, using multi-temporal Landsat imagery (1990–2025), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and SHapley Additive exPlanations (SHAP). Model performance was evaluated using the Global Mangrove Watch (GMW) dataset and spatially blocked holdout validation. Classification accuracy was high (94.56–100%) under internal and spatially blocked validation (93.5–99.8%), whereas comparison with GMW yielded lower accuracies (66.4–95.7%). Change detection showed that mangrove area expanded from 2096.66 km2 in 1990 to 3875.95 km2 in 2025, a net gain of almost 1779.29 km2, with substantial inter-period variability and a pronounced mapped decline between 2000 and 2005. SHAP analysis indicated that the relative contributions of model predictors changed over time, with NDVI contributing more strongly to model predictions in recent years than NDWI and NDMI. Mangrove gain, loss, and net change were statistically associated with cyclone exposure and elevation, while precipitation and river discharge showed less consistent relationships. This study highlights a practical method for monitoring spectral dynamics in vulnerable coastal ecosystems and supports mangrove conservation.

Remote Sensing doi: 10.3390/rs18203555

Authors: Qiyuan Xiao Junyi Zhang Zhuangwei Jiang Liujun Li

Marine heatwaves (MHWs) in the Indonesian Seas develop within a long-term rise in sea surface temperature and strong interannual climate variability. We analyze daily NOAA Optimum Interpolation Sea Surface Temperature version 2.1 for 1982–2022 and compare five Eastern-Pacific (EP) with five Central-Pacific (CP) El Niño events. MHWs are detected against a fixed 1982–2011 baseline, and each event is represented by the common 1 July–30 June window spanning its developing and decaying phases. EP-minus-CP differences are 14.25 area-weighted MHW days and 22.34 °C·days of cumulative intensity. Their event-bootstrap intervals span zero, the exact permutation tests are nonsignificant, and no grid cell passes false-discovery-rate control for either metric. Annual MHW days increase by an area-weighted 11.37 days per decade; 90.5% of the domain passes false-discovery-rate control, and the temporal-block field test yields p=0.001. The increase remains widespread after linear Niño-3.4 and dipole-mode-index adjustment. Direct physical diagnostics, spatial-block habitat validation, and official Fisheries Management Area (FMA) summaries are used to assess the strength and limitations of the proposed process and management interpretations. The physical-diagnostic intervals include zero, and the MaxEnt output represents modeled environmental suitability. The results establish a spatially extensive long-term increase in MHW exposure and quantify the uncertainty in the event-scale physical and ecological pathways.

Remote Sensing doi: 10.3390/rs18203553

Authors: Yishan Zhang Ruijun Chen

Effective water quality monitoring is essential for protecting aquatic environments and supporting sustainable management of urban rivers. Accurate prediction and visualization of water quality parameters (WQPs) distributions across spatial, temporal, and vertical dimensions remained challenging for unmanned aerial vehicle (UAV)-based remote sensing approaches, which often neglected depth-wise variability and required extensive water sampling to maintain stable prediction performance for multiple WQPs. To address these limitations, this study proposed a feature propagation-based extended long short-term memory framework (FP-xLSTM) for unified prediction and visualization of multi-depth WQPs, including total phosphorus (TP), total nitrogen (TN), chlorophyll-a (Chl-a), chemical oxygen demand (COD), biochemical oxygen demand (BOD), and total suspended solids (TSS). By integrating spectral characteristics with spatio-temporal dependencies through feature propagation across consecutive temporal observations, FP-xLSTM effectively captured short-term fluctuations and learned vertical distribution patterns of heterogeneous WQPs from multi-source data while reducing sampling requirements. A field study with an eight-month interval between the first and last sampling campaigns was conducted to evaluate the proposed framework under cross-region experimental settings and against state-of-the-art baseline models. Results demonstrated that FP-xLSTM consistently outperformed competing methods, achieving average coefficients of determination (R2) ranging from 0.66 to 0.95 and mean absolute percentage errors (MAPE) between 9.33% and 19.06%. Furthermore, the framework has potential for deployment in practical environmental monitoring applications to support local environmental authorities in tracking river water quality dynamics and optimizing management strategies. Overall, the proposed method provided an effective and scalable intelligent framework for UAV-assisted urban water quality assessment and multi-depth environmental visualization.

Remote Sensing doi: 10.3390/rs18203557

Authors: Dingzhao Li Chuanzhi Liu Mingzhang Zhou Haixin Sun

Camera-based underwater optical monitoring enables persistent and non-invasive observation of marine communities, but its automated interpretation is challenged by two coupled uncertainties: the natural long-tailed frequency distribution of observed organisms and the appearance of categories absent from model development. Rare known categories provide insufficient observations for reliably estimating class location and dispersion, whereas dominant categories may form broad acceptance regions that absorb previously unseen organisms. We propose Frequency-Shrunk Prototype and Risk Calibration (FSPRC), a known-only post hoc framework for long-tailed underwater optical monitoring built on an ImageNet-pretrained ConvNeXt-Tiny. FSPRC regularizes each empirical feature direction toward its corresponding classifier direction and shrinks its intraclass radius toward a frequency-group reference, with both corrections controlled by the number of training observations. Local prototype deviation and global energy evidence are then robustly normalized within the predicted head, middle, or tail group and combined without a trainable fusion module. Independent known-only calibration subsets determine group-conditional rejection thresholds, while held-out categories remain excluded from backbone adaptation, checkpoint selection, geometric estimation, score normalization, and threshold fitting. On the predefined fold-0 leakage-controlled protocols for AQUA20 and the track-disjoint Fish4Knowledge-23 dataset, FSPRC achieves open-set accuracies of 83.31% and 98.19%, open-set classification rate (OSCR) values of 85.96% and 98.96%, and unknown category macro area under the receiver operating characteristic curve (AUROC) values of 92.62% and 99.55%, respectively. Across exhaustive class-held-out partitions, open-set accuracy is 83.25±0.24% over four AQUA20 folds and 98.18±0.08% over five Fish4Knowledge-23 folds. These results demonstrate that frequency-dependent geometric stabilization and risk calibration can improve the recognition-stage balance between retaining data-scarce known organisms and referring unfamiliar observations for subsequent expert inspection.

Remote Sensing doi: 10.3390/rs18203554

Authors: Eslam Muhammed Ahmed Shaker

Maintaining accurate, georeferenced dense maps requires reliable global navigation satellite systems (GNSS) observations to constrain global drift. Degraded observations must be prevented from distorting trajectories and local maps as reception transitions from open sky to urban obstruction and complete signal denial indoors. This study develops and evaluates Dense LiDAR–Inertial–GNSS simultaneous localization and mapping (DLIG-SLAM) to adapt GNSS integration and local mapping as reception conditions change. Built on Dense Multi-Scan Adjustment, the framework estimates the local trajectory and dense map separately from the transformation that connects them to a global coordinate frame. GNSS observations are assessed separately for updating this transformation and correcting the estimator. An online switcher selects among integration strategies according to measurement reliability, estimator readiness, and trajectory consistency. Outdoor–indoor state changes reconfigure local mapping, while voxel size adapts to the observed scan point population. Evaluation was performed on four KITTI driving sequences and an urban canyon outdoor–indoor mission in Toronto, recorded with standalone single-point positioning GNSS. GNSS coupling through an inertial navigation estimate achieved the lowest three-dimensional trajectory root mean square error on three of the four KITTI sequences. Online switching did not consistently outperform fixed integration strategies in the KITTI trajectory evaluation. In the Toronto dataset, inertial navigation-mediated coupling achieved the lowest horizontal georeferencing root mean square error of 1.29 m, while full-navigation factor graph smoothing achieved the lowest three-dimensional georeferencing root mean square error (RMSE) of 1.71 m. Online switching achieved the lowest cloud-to-cloud root mean square distance of 6.1 cm against a terrestrial laser scanning reference within the evaluated facade and ground floor region. Controlled experiments also evaluated the effect of separate GNSS admission paths and global transformation policies. The results support adapting GNSS integration to changing reception conditions and mapping objectives, as stronger coupling did not uniformly improve trajectory accuracy, global placement, and local geometric agreement.

Remote Sensing doi: 10.3390/rs18203552

Authors: Songtao Shao Ruilin Li Ming Yu

High-resolution remote sensing image semantic segmentation faces challenges from large-scale variation, fine boundary preservation, and class imbalance. This paper proposes CAP-DBNet, a compact framework that combines a RepViT M2.3 backbone with a global–local collaborative decoder (GLCD), a class-adaptive hybrid loss (CAHL), and a class-adaptive prototype-guided calibration module (CAPC). CAP-DBNet integrates class-center global and window-local modeling through bidirectional feature exchange, mini-batch class-frequency-aware supervision, and prototype-based residual calibration. Rather than introducing an isolated new attention or prototype mechanism, this work focuses on the coordinated design of these components within an efficient segmentation framework. In five repeated experiments with different random seeds, CAP-DBNet achieves average mIoUs of 85.16%, 88.32%, and 55.75% on Vaihingen, Potsdam, and LoveDA, respectively, improving over the same-backbone LOGCAN++ baseline by 1.43, 1.24, and 1.44 percentage points. Compared with D2LS under our evaluation protocol, CAP-DBNet uses 3.55× fewer parameters and requires 8.35× fewer floating-point operations (FLOPs). These results demonstrate an accuracy-efficiency-oriented global–local and class-adaptive framework for remote sensing semantic segmentation across three public benchmarks.

Remote Sensing doi: 10.3390/rs18203549

Authors: Jie Cui Kaiqi Lang Chao Liang Hanyu Wang Honghai Shen

Infrared and visible image fusion aims to integrate complementary thermal target information and visible structural details for multimodal perception and remote sensing applications. However, existing deep learning methods often underutilize semantic information during fusion and mainly perform cross-modal interaction in the spatial domain, limiting the exploitation of complementary representations. To address these issues, we propose SDFNet, a semantic driven frequency-domain infrared and visible image fusion network based on an encoder–decoder architecture. Guided by semantic priors extracted from a pre-trained segmentation model, SDFNet introduces a frequency-domain feature aggregation module to enhance cross-modal feature interaction and semantic information preservation. Extensive experiments on the MSRS dataset demonstrate the effectiveness of the proposed method. On the 361-pair MSRS test set, SDFNet ranks among the top two on five of the six evaluated fusion metrics, achieving EN, AG, SF, SCD, and Qabf scores of 6.73, 3.78, 11.59, 1.74, and 0.69, respectively, while obtaining a competitive VIF score of 1.02. In the downstream semantic utility evaluation, SDFNet achieves the highest mIoU of 78.76 among the compared methods, further demonstrating its ability to preserve task-relevant semantic information.

Remote Sensing doi: 10.3390/rs18203550

Authors: Jin Wang Guoping Zhang Shuai Xing Xun Geng Xinlei Zhang Jiayao Wang

Satellite-derived bathymetry (SDB) commonly treats the sea surface as a source of radiometric disturbance, although the same surface wave field can also contain water depth information through gravity wave propagation. This article presents a Wave-Aware Physics-Informed Neural Network (WA-PINN) that represents these two roles within a unified shallow water bathymetry framework. The optical geometry stream combines Sentinel-2 multispectral observations with seafloor geometry. The disturbance stream converts European Centre for Medium Range Weather Forecasts Reanalysis Version 5 (ERA5) wind forcing into Mean Square Slope (MSS) through the Cox–Munk relationship and introduces a roughness-dependent sunglint contribution into the optical physics constraint. The signal stream retrieves local wavelength, propagation direction, and wave celerity from Sentinel-2 surface texture and constrains predicted depth with the finite depth gravity wave dispersion relation. A Wave Validity Factor (WVF) controls the influence of the dispersion constraint according to the coherence and physical plausibility of the retrieved wave field. The three streams are fused in a shared depth predictor constrained by supervised depth loss, optical physics loss, and wave dispersion physics loss. The framework is demonstrated over Culebra, St. Thomas, and St. John in the northeastern Caribbean Sea. The complete Dual Role model reaches an overall Root Mean Square Error (RMSE) of 1.21 m, a Mean Absolute Error (MAE) 0.77 m, and an R2 of 0.96. Relative to the Slope Aware configuration, the Disturbance and Signal models reduce RMSE by 15.7% and 12.8%, while the complete model reduces RMSE by 29.7%. The disturbance contribution increases under stronger wind forcing, whereas the signal contribution increases with WVF. Depth-dependent RMSE increases from 0.82 m at 0 to 5 m to 1.58 m at 20 to 30 m. These results illustrate the intended complementarity between correcting wave-induced optical contamination and exploiting wave propagation as an additional depth constraint.

Remote Sensing doi: 10.3390/rs18203551

Authors: Jun Wang Junxing Cao Yajuan Xue

Fractures are important storage spaces and flow pathways in tight reservoirs, but small-scale fractures with weak seismic responses are difficult to characterize with band-limited post-stack seismic data. This study proposes a high-resolution multiscale seismic method for predicting fracture distribution based on basis pursuit frequency extension (BPFE)-enhanced multi-attribute fusion. BPFE represents each trace sparsely on a wedge dictionary built from a zero-phase Morlet wavelet, which is derived from the average dominant frequency of the data. It then re-synthesizes the basis pursuit denoising solution with a time-compressed wavelet, thereby extending the effective bandwidth without estimating the seismic wavelet. Dip-steered third-generation coherence and ant-tracking attributes are subsequently normalized and fused with explicit weights according to their attribute-sensitive scales, and objective rules are provided for selecting the BPFE parameters. Synthetic tests with known reflectivity show that BPFE reproduces the ideal high-resolution section with correlation coefficients of 0.978 (noise-free) and 0.882 (signal-to-noise ratio of 10 dB). BPFE increases the dominant frequency from 30.8 to 49.3 Hz and reduces the tuning thickness from 16 to 11 ms without systematic phase rotation. At 10 dB, it also raises the area under the receiver operating characteristic curve for detecting weak bed-confined discontinuities from 0.580 to 0.687, outperforming spectral whitening, spiking deconvolution, inverse-Q filtering and sparse-spike inversion in this respect. Applied to the tight sandstone reservoir in the northeastern Sichuan Basin, the workflow delineates main faults, fault-associated fracture zones and small-scale discontinuities more completely than coherence, ant tracking or their fusion without BPFE, which is consistent with the actual geological understanding. The workflow offers a wavelet-estimation-free approach for characterizing weak-response small-scale fractures, although its generalization requires further field validation.

Remote Sensing doi: 10.3390/rs18203548

Authors: Seongmin Lee Byounggil Kim Jae Young Chang Kwang-Jae Lee Haechul Choi

Segmentation performance is measured against pixel-wise ground truth (GT); without it, that measurement cannot be made. Remote sensing segmentation models are evaluated likewise, yet operational systems receive new unlabeled images, and the properties of those images shift with season, region, and sensor aging. A way to check performance outside the training dataset is needed. We propose quantifying segmentation output without labels, using public geospatial data: a random-forest context-aware model predicts class area composition from OpenStreetMap (OSM) and digital surface model (DSM) statistics. Quantile mapping transforms its output to match the distribution of measured compositions, yielding the expected distribution, and the Label-Free Composition Score (LFCS) is the similarity between that distribution and the composition of the segmentation output. The context-aware model is fitted only to geospatial features and GT compositions from the segmentation model’s training dataset, so the expected distribution is independent of the segmentation model’s structure, parameters, and outputs while depending on that dataset’s GT, class scheme, and composition distribution. The evaluation stage requires no GT labels. To test the method, we saved 430 segmentation model checkpoints from 75 training runs under six controlled training variations, so that the mean intersection over union (mIoU) of each, measured against GT on validation tiles held out of training, was known in advance. These GT mIoU values ranged from 0.04 to 0.69. The models were grouped by GT mIoU in steps of 0.1. Their mean LFCS rose with every step: from 0.474 to 0.937 on held-out images from the same dataset, and from 0.467 to 0.893 on 900 globally distributed unlabeled images. Existing measures that read only a model’s own output tracked GT mIoU less consistently. Rebuilding the entire method reproduced the same behavior on two public datasets with different class schemes, confirming the validity of the method under the conditions tested.

Remote Sensing doi: 10.3390/rs18203547

Authors: Piyapong Tongdeenok Naruemol Kaewjampa Surachit Waengsothorn Sitthisak Moukomla

Dry dipterocarp forest (DDF) in mainland Southeast Asia burns frequently in surface fires, yet the vertical fuel strata observable from airborne and spaceborne sensors remain uncertain. We compared handheld SLAM LiDAR, UAV LiDAR, GEDI, ICESat-2, PlanetScope, and Sentinel-1/2 at Sakaerat, Thailand, using spatial block-bootstrap validation. SLAM and UAV LiDAR converged on the 1–2 m and 2–5 m strata (r = 0.53 for both, n = 212) but not below 0.5 m, identifying the 2–5 m ladder fuel fraction as the strongest cross-platform target. Optical models explained at most 32% of understory and 37% of ladder fuel variance, with no benefit from finer pixels or Sentinel-1. GEDI relative-height metrics failed below 2 m because the ground return was broad (mean FWHM 4.2 m) and widened with slope. Subtracting a fitted ground Gaussian recovered a reproducible 1–5 m waveform signal correlated with UAV ladder fuel in both 2025 (ρ = 0.47) and 2026 (ρ = 0.42), robust across detection thresholds and fitting windows. Only 22% of regional GEDI footprints supported two-mode canopy-base detection, while ICESat-2 showed the same height-ordered sensitivity. In this open DDF stand, remote sensing was therefore best suited to ladder fuel and vertical continuity; near-surface fuel requires ground-based observation.

Remote Sensing doi: 10.3390/rs18203546

Authors: Jiawei Wan Haoyuan Gao Bin Li Yang Gao Jun Li Xingxing Du

Large catastrophic landslides in the southeastern Tibetan Plateau are characterized by huge volumes and complex chains that frequently cause river damming and casualties. Based on field surveys, multi-temporal satellite imagery, Unmanned Aerial Vehicle (UAV), Interferometric Synthetic Aperture Radar (InSAR) deformation measurements, and meteorological and seismic records, this study classifies these landslides into three types by their initiation and disaster-chain processes: (1) rock–ice avalanche → debris flow → landslide dam; (2) debris flow → landslide dam; (3) rock avalanche → landslide dam. Representative cases illustrate the material sources, initiation settings, and downstream impacts of each type. The occurrence of such landslides is influenced by the interplay of four groups of factors: fractured rock masses and abundant loose deposits provide the material basis; high mountains and narrow river valleys provide the spatial framework for landslide evolution and promote river blockage; pronounced climatic changes provide the background conditions and may affect slope stability in glaciated areas through glacier retreat; and frequent strong earthquakes can both instantaneously trigger failure and cause cumulative damage that predisposes slopes to subsequent failures. Ongoing climate warming and repeated seismic activity are expected to increase landslide susceptibility, highlighting the urgency of multi-hazard early warning and proactive mitigation. This study supports early warning prioritization and risk governance in high-mountain valleys across the Tibetan Plateau.

Remote Sensing doi: 10.3390/rs18203545

Authors: Frances C. O’Leary Desirée Marques Ramos Leonor Patricia Cerdeira Morellato Vera De Cauwer Francisco Maiato Gonçalves Casey M. Ryan Manuel Cachissapa Gabriel I. K. Uusiku John L. Godlee Bruna Alberton Matúš Seči Timothy R. Baker Kyle G. Dexter

Reliable monitoring of leaf phenology across environmental gradients is essential for understanding ecosystem functioning and climate responses, yet existing approaches in African savannas are constrained by either sparse ground-based observations or the limited resolution of satellite products. We evaluated the ability of MODIS (Moderate Resolution Imaging Spectroradiometer), VIIRS (Visible Infrared Imaging Radiometer Suite), and Sentinel-2-derived vegetation indices (Normalised-Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Green Chromatic Coordinate (GCC)) to track phenocam-derived phenology across 10 plots along a water-availability gradient in Africa. Satellite vegetation index time series were smoothed using generalized additive models and compared with phenocam-derived GCC, including estimates of transition dates. All satellite datasets estimated longer growing seasons than phenocams, associated with earlier start-of-season (SOS) and later end-of-season (EOS) estimates. MODIS NDVI and VIIRS NDVI most accurately estimated transition dates (SOS root-mean-squared error (RMSE), a standard measure of model accuracy, of 10 and 11 days, respectively; EOS RMSE of 27 days in both datasets). MODIS GCC and VIIRS EVI followed with high estimated transition date accuracy (SOS RMSE of 11 days; EOS RMSE of 32 days, in both datasets). VIIRS EVI and Sentinel-2 (S2) GCC most accurately tracked phenocam temporal fluctuations (r = 0.83 ± 0.08 and 0.83 ± 0.10, respectively), while MODIS GCC and MODIS NDVI also showed high correlations (r = 0.78 ± 0.12 and 0.78 ± 0.15). These MODIS products tracked intraseasonal fluctuations more closely than the VIIRS and S2 datasets. We consider MODIS NDVI the most accurate dataset overall, given its accuracy in estimating SOS and EOS, visible tracking of intraseasonal fluctuations, and high temporal correlation with phenocam phenology. These results support using MODIS and VIIRS for large-scale phenology monitoring and upscaling phenocam observations in African savannas.

Remote Sensing doi: 10.3390/rs18203544

Authors: Siyun Hou Xiangyi Zhang Jianlong Wang Wengen Li

Sea surface temperature (SST) forecasts support marine environmental monitoring, hazard assessment, and operational ocean services. Regional multi-step prediction remains difficult because SST-only models must represent transport, spatial spreading, and local nonlinear variation from historical fields while avoiding the recursive use of predicted states. We propose ADR-Net, a direct multi-step forecasting model that combines a flexible data-driven stream with a process-inspired advection–diffusion–reaction (ADR) stream in a shared latent bottleneck. The ADR stream organizes latent SST evolution through three operators that represent transport-like displacement, spatial spreading, and local nonlinear variation. A temporal projection module (TPM) then projects the fused latent representation and multi-scale encoder features directly to all forecast leads. We evaluated ADR-Net against eight learned baselines and two non-learned references using daily OISST v2.1 data from the East China Sea (ECS), South China Sea (SCS), Niño 1+2, and Niño 3.4 regions. Separate models were trained with output lengths L∈{1,7,13}. ADR-Net achieved the lowest MSE, MAE, and RMSE in all 12 region–output length settings across four regions and three output lengths. In the ECS, it reduced the MSE by 1.46%, 8.07%, and 5.70% relative to the best MSE baseline for L=1, L=7, and L=13, respectively. For the ECS L=13 task, ADR-Net maintained the lowest MAE and highest SSIM at every lead day and extended the mean effective forecast horizon from 3.91 to 4.49 days under grid cell thresholds of RMSE ≤ 0.7 °C and R2≥0.95. Component ablations on the ECS seven-day task showed that removing the TPM or the process-inspired ADR stream increased the mean MSE by 4.55% and 3.86%, respectively, while removing both increased it by 8.32%. ADR-Net used 7.09 million parameters and 13.07 GFLOPs per ECS seven-day forecast sample. Across four storm-associated episodes, ADR-Net achieved the lowest event-mean cooling footprint MAE at every forecast lead from day 1 to day 13. Together, these results support the combination of process-inspired latent modeling and direct temporal projection for accurate, spatially coherent regional SST forecasting.

Remote Sensing doi: 10.3390/rs18203543

Authors: Wei Song

Cropland abandonment alters agricultural land use and may enable improvements in vegetation cover. On the Chinese Loess Plateau, the extent and spatial variation of these changes remain critical questions for land management. This study used annual land-cover trajectories to identify persistent cropland withdrawal from 2000 to 2020 and Landsat-derived fractional vegetation cover (FVC) to assess vegetation change. The total first-abandonment area was 80,700.64 km2, and the annual area of first abandonment peaked in 2004 at 7176.16 km2 and subsequently declined amid interannual fluctuations, with abandoned cropland concentrated on gentle slopes. Across five representative years, the mean FVC of land classified as abandoned each year rose from 0.273 in 2000 to 0.381 in 2020. These values reflect changes in pixel populations rather than the trajectory of a fixed set of fields. Among pixels mapped as abandoned in 2020, 86.07% showed increasing FVC over the entire study period, including 60.12% with significant increases. Extreme gradient boosting (XGBoost) analysis with Shapley additive explanations (SHAP) identified precipitation as the strongest predictor of spatial variation in the FVC change rate, accounting for 39.6% of the model’s relative importance. SHAP dependence patterns showed model-based transition points at approximately 395 mm of precipitation and 10 °C of temperature. These findings characterize vegetation-cover change associated with abandoned land and its environmental variation, whereas the full-period trends cannot isolate vegetation change caused by abandonment.

Remote Sensing doi: 10.3390/rs18203542

Authors: Pelagia Gawronek Maria Makuch Krzysztof Dorula Maksymilian Gawronek

Terrestrial laser scanning (TLS) and unmanned aerial vehicle (UAV) photogrammetry provide complementary observations for infrastructure documentation, but their suitability depends on the required map features. This study compares TLS- and UAV-derived as-built products for a wastewater treatment plant with a conventional survey reference and illustrates a preliminary, retrospective, feature-level data source selection approach. Positional agreement was assessed at 30 independent check points selected purposively from details identifiable in all three products. RMSE3D was 7.51 cm for TLS and 7.66 cm for UAV; these values characterize the common observable sample rather than the accuracy of the complete infrastructure. TLS and UAV mapped 128 and 102 of 135 reference objects, corresponding to object detection completeness rates of 94.8% and 75.6%, respectively. UAV required less recorded acquisition effort, while TLS recovered more reference objects using the tested configuration. A transparent score combines planimetric accuracy and object detection completeness where both are available, with source eligibility assessed separately. The application illustrates source comparison using existing products; it does not establish improved map quality or lower production cost. A completed selected source map and a merge-all alternative were not independently evaluated.

Remote Sensing doi: 10.3390/rs18203541

Authors: Baozhan Liu Yikai Huang Kaihua Huang Bingxin Liu Cuiling Yin Zhendong Li

Litter stranded on beaches can severely degrade local ecosystems and diminish the tourism value of coastal areas. The integration of deep learning into unmanned aerial vehicles (UAVs) provides an effective means for the automatic segmentation of coastal litter. However, large variations in litter scale and complex coastal backgrounds lead to relatively low detection accuracy. In addition, most existing detection models require substantial computational resources, which limits their deployment. To address these issues, we constructed a coastal litter segmentation dataset and developed a lightweight segmentation model, CoastalLitterNet. The dataset contains litter objects of different sizes, colors, and shapes, with various backgrounds. In CoastalLitterNet, a Lightweight Asymmetric U-shaped Backbone Network (L-Net) was designed to reduce model complexity, and Atrous Spatial Pyramid Pooling (ASPP) was introduced to segment litter objects at different scales. A CoastalLitterGate (CLG) was proposed to enhance litter-related features and improve adaptability to complex backgrounds. Comparative experiments showed that the mIoU, IoU1 and F1 of CoastalLitterNet reached 93.95%, 88.42% and 93.85%, respectively. The Params and FLOPs of the model are only 12.90 M and 189.30 G, respectively. The quantitative and qualitative results show that CoastalLitterNet effectively segments litter objects at varying scales and adapts well to complex coastal environments while maintaining a lightweight architecture.

Remote Sensing doi: 10.3390/rs18203540

Authors: Pan Chen Yu Han Xiaoli Li Yongqing Bai Hongzhong Li Longlong Zhao Jinsong Chen Zhengchao Chen

Litchi is an important cash crop in China. Accurate litchi orchard extraction is crucial for agricultural management and economic development. However, litchi orchards are spectrally very similar to evergreen broad-leaved forests and are poorly resolved in medium- and low-resolution satellite imagery, so large-scale extraction remains challenging. A cross-resolution fusion network via semantic anchor-graph enhancement (SAGE-Net) is therefore proposed, using a dual-branch encoder–decoder to aggregate complementary features of AlphaEarth Foundations (AEF) embeddings and GF-1 imagery. To enable semantic interaction across resolutions, the SAGE module is designed to gather the AEF embeddings into a few semantic anchors, reason over the anchor graph, and scatter them back under GF-1 guidance, so that semantic content and spatial extent are supplied by the source better suited to each. To refine the coarse prediction to full resolution without discarding what is already reliable, the reliability-adaptive correction (RAC) head is designed to gate a detail-guided correction by the confidence and local consistency of that prediction, concentrating refinement where it is uncertain. In Shenzhen, SAGE-Net attains the highest IoU (0.5660) and F1-score (0.7229) among eight comparison methods, with a Precision of 0.7162 and a Recall of 0.7297. Under stricter protocols that separate the training and the evaluation patches by a 256 m buffer or by whole spatial blocks, the IoU falls to 0.5224 and 0.5000 and SAGE-Net remains the most accurate of the networks retested, by a smaller margin. The results demonstrate the value of role-aware enhancement and reliability-adaptive correction for fine-grained orchard mapping, and confirm AEF embeddings as an effective supplementary data source for large-scale mapping.

Remote Sensing doi: 10.3390/rs18203539

Authors: Xi Kan Xianwu Wang Jing Wang Linglong Zhu

High-resolution snow depth (SD) information supports hydrological and cryospheric studies in complex mountain regions. Coarse SD products limit representation of terrain-related snow variability in these environments. This study proposes SDFNet, a Snow Depth Fusion Network combining Sentinel-1 SAR, Landsat optical data, digital elevation model (DEM) variables, and fractional snow cover (FSC). Multi-scale feature extraction, channel enhancement, and feature aggregation modules represent relationships among these inputs. Observations from 95 meteorological stations across the Qinghai-Tibet Plateau supplied supervision and reference measurements for evaluation. Station SD was the regression target rather than a prediction input. In the station-based comparison, SDFNet achieved an RMSE of 5.246 cm and an R2 of 0.702. It had the lowest RMSE and highest R2 among the compared models. However, SD above 40 cm produced an RMSE of 21.927 cm and a bias of −16.061 cm, indicating substantial deep-snow underestimation. Sparse station coverage limits inference beyond the sampled environments, and independent spatial validation remains necessary.

Remote Sensing doi: 10.3390/rs18203538

Authors: Eunna Jang Sujung Bae Jae-Hyun Ahn Jong-Kuk Choi

Establishing data continuity across satellite missions is essential for multi-decadal ocean color time-series. This study presents an integration framework to achieve radiometric harmonization and spectral continuity between the First and Second Geostationary Ocean Color Imagers (GOCI and GOCI-II) over the East China Sea. Using collocated observations during their mission overlap period, a two-step transformation scheme was established. First, systematic calibration offsets across six overlapping bands (412–680 nm) were eliminated by harmonizing GOCI to the GOCI-II via ordinary least-squares regression, bringing mean bias values near zeros (R2 > 0.95 and relative RMSD ≈ 12–13% in mid-visible bands). Second, GOCI remote sensing reflectance (Rrs) was expanded to reconstruct three GOCI-II-specific missing spectral bands (510, 620, and 709 nm) using two independent approaches: the physics-based Quasi-Analytical Algorithm (QAA) and Trust Region Reflective (TRR) optimization. Comparative validation demonstrated that the TRR model consistently outperformed QAA across dynamic coastal gradients. TRR achieved high spectral fidelity while suppressing coastal inversion errors. Furthermore, regional spectral curves verified that the harmonized and TRR-derived GOCI Rrs closely converged with GOCI-II Rrs across moderately turbid, highly turbid, and clear offshore waters. This framework verifies the feasibility of constructing a seamless geostationary ocean color database spanning GOCI and GOCI-II, establishing a solid foundation for long-term marine monitoring and future missions like GOCI-III.

Remote Sensing doi: 10.3390/rs18203537

Authors: Xuhan Huang Yulin Zhan Junwei Liu Jian Yang

Solar-induced chlorophyll fluorescence (SIF) provides a new pathway for the accurate estimation of gross primary productivity (GPP) by directly probing vegetation photosynthetic functioning, but mechanistic estimation is still constrained by the difficulty of obtaining the seasonal dynamics of key physiological parameters. Using maize, a representative C4 crop, as an experimental case, this study integrated long-term tower-based SIF observations with the SCOPE model to propose a cross-variable constraint strategy, in which key vegetation parameters were optimized against in situ SIF observations and independently validated against eddy covariance-derived GPP. Specifically, key sensitive parameters were first identified via Sobol global sensitivity analysis. A double-baseline asymmetric Gaussian phenology model was then constructed to characterize the seasonal dynamics of Vcmax, while Cab was derived from Vcmax using a smoothed piecewise-linear empirical relationship. A two-stage Bayesian optimization scheme was applied to optimize the parameters, which were subsequently used to run the SCOPE model and reconstruct continuous SIF and GPP time series. The results showed that the estimated daily SIF and GPP agreed well with ground observations, yielding R2 values of 0.86 and 0.74, respectively. Furthermore, the upscaled results correlated strongly with satellite products (MODIS: R = 0.94, R2 = −2.69; PML: R = 0.95, R2 = 0.81). This study demonstrates the feasibility of using SIF observations to constrain SCOPE vegetation parameters for improved GPP estimation, providing a parameterization basis and methodological support for mechanistic remote sensing monitoring of vegetation productivity.

Remote Sensing doi: 10.3390/rs18203536

Authors: Jilin Men Hongxing Liu Ekaterina Miliutina Yuehan Lu Lei Wang Haibin Su

The USGS Landsat 8 Provisional Aquatic Reflectance (PAR) product offers a standardized basis for inland water monitoring, but its reliability is constrained by spectral uncertainty and valid-data availability. This study evaluated PAR using GLORIA in situ observations, Lake Lanier field measurements, and cross-sensor comparisons, with ACOLITE and OC-SMART as alternative atmospheric correction methods. In the full GLORIA comparison, PAR showed median absolute percentage differences (MAPDs) of 115.9% and 75.2% at 443 and 483 nm, respectively, compared with 40.0% at 561 nm and 66.5% at 655 nm. Moreover, 310 of 579 PAR matchups contained negative reflectance in at least one band, substantially limiting usable observations. Water quality retrievals evaluated using common matchups yielded PAR MAPD values of 71.6% for chlorophyll-a, 71.0% for total suspended solids, and 117.7% for colored dissolved organic matter absorption, indicating substantial retrieval uncertainty even after sample availability was controlled. Over Lake Lanier, narrow water branches exhibited substantial PAR data loss and less reliable retrievals, highlighting limitations associated with shoreline proximity and mixed pixels. Differences in valid-data coverage also constrained direct comparisons among methods at this site. Cross-sensor comparisons over Lake Victoria and Lake Michigan, nevertheless, showed that PAR captured broad spatial patterns, while time-series comparisons suggested potential for resolving seasonal variability, particularly in chlorophyll-a. Overall, PAR may support monitoring of larger inland waters when sufficient valid observations are available. Applications in small or narrow waters, and retrievals sensitive to blue-band errors, require particular caution. Broader operational and long-term use requires further validation across sites and years.

Remote Sensing doi: 10.3390/rs18203535

Authors: Chang Woo Song Yeong Jun Koh

Referring expression grounding in overhead imagery requires oriented bounding boxes rather than horizontal ones, since aerial objects appear at arbitrary orientations. Most grounding models emit box coordinates sequentially; Parallel Box Decoding (PBD) instead predicts all geometry slots in one forward pass, but only for four-coordinate horizontal boxes. We extend this block-aligned interface to a six-slot oriented representation using a double-angle parameterization. Rather than assuming correctness from a decreasing loss curve, we verify the extension with image/prompt-shuffle controls and visibility/leakage probes, which show that the corrected model depends on both the image and the prompt rather than on prompt statistics alone. On the aerial vehicle visual grounding (AVVG) benchmark, whose targets are genuinely oriented, the extended model improves rotated intersection over union (rotIoU) from 0.1253 for the zero-shot baseline to 0.303±0.012 (mean ± standard deviation over three training seeds) across the complete 6139-record test split spanning all 120 images. Learning oriented boxes rather than their axis-aligned envelopes adds 0.055±0.008 rotIoU beyond a matched control (per-seed 95% confidence intervals all excluding zero), whereas block expansion alone yields no rotIoU gain distinguishable from zero.

Remote Sensing doi: 10.3390/rs18203533

Authors: Misganu Debella-Gilo Diress Tsegaye Richard Adade Deo D. Shirima Belachew Gizachew

Remote sensing is essential for mapping mangrove forest ecosystems, yet accuracy and spatial resolution remain limited, particularly in Africa, where mangroves are fragmented, reference data and cloud-free images are scarce. The state-of-the-art of mangrove forest mapping relies on native-resolution imagery using machine learning. This study evaluates deep learning methods using aggregated multi-date satellite imagery to improve mangrove mapping and national-scale area estimation. Super-resolution and semantic segmentation techniques were applied to Sentinel-2 (S2) and Sentinel-1 (S1) images, with training data derived from field observations and high-resolution drone and satellite imagery. Coastal regions of Tanzania and Ghana were used as representative study areas for eastern and western Africa, respectively. Results show strong classification performance, with S2 imagery alone achieving an Intersection over Union (IoU) of 85.8%. Adding S1 provided modest gains (IoU 86.3%), while super-resolution significantly improved the accuracy (IoU 88.5%). The combined super-resolved S2 and S1 dataset yielded the highest performance (IoU 89.2%). Wall-to-wall mapping and statistically rigorous validation and area estimation produced accurate results with quantified uncertainty. Tanzania showed high consistency (F1 = 96%; 1207 km2, 95% CI [1151.7, 1263]), while Ghana exhibited slightly higher uncertainty (F1 = 92%; 92.6 km2, 95% CI [87.5, 97.7]). Integrating deep learning with multi-date, multi-sensor fusion and super-resolution enables accurate mangrove mapping and area estimation for improved coastal management and restoration.

Remote Sensing doi: 10.3390/rs18203534

Authors: Yongliang Wei Yajun Yue Shouhua Liu Zhenzhe Yang

Shallow-water bathymetry is essential for coastal dynamics, navigation, engineering, and marine-resource management, especially where in situ observations are sparse. This study proposes a wave-stripe-based method for shallow-water bathymetric retrieval from Sentinel-1A VV-polarized SAR imagery. After texture-preserving denoising, multidirectional Gabor enhancement, K-means segmentation, and skeleton tracking, wave-crest centerlines are extracted and local wavelengths are measured as the normal spacing between adjacent crest lines. The dominant wave period is estimated from deep-water wavelengths, and water depth is retrieved using the linear wave dispersion relation. The method was evaluated using five Sentinel-1 acquisitions, with SAR-derived representative wave periods ranging from 12.45 to 14.96 s. Based on the mean retrieval performance across the five acquisitions, the 0–20 m depth range yielded an RMSE of 3.87 m, an MAE of 3.56 m, and r = 0.95, while the corresponding values for 0–50 m were 6.06 m, 5.02 m, and 0.97, respectively. The most stable performance occurred within the 5–20 m depth range. The method reduces fixed-window spectral averaging and improves the representation of local wavelength variability, supporting rapid bathymetric mapping along open-coast environments with clear swell-induced wave stripes, relatively gentle bathymetric slopes, and limited field observations.

Remote Sensing doi: 10.3390/rs18203532

Authors: Xinyi Du Wenbin Chen Xiong Hu Junfeng Yang Zhijin Li Shuai Lu Zewei Wang Cunying Xiao

The near-space region is of strategic importance for aerospace activities. However, accurate temperature analyses remain challenging due to sparse observations and multiscale spatial structure. In this study, a multiscale variational analysis (MSVA) configuration is applied and evaluated for near-space temperature analyses using the Mass Spectrometer and Incoherent Scatter (MSIS) empirical model as the background field and TIMED/SABER temperature observations as observational constraints. The multiscale strategy is implemented via a prescribed sum of three-dimensional Gaussian background error covariance components with scale-dependent horizontal length scales and a common vertical coupling structure. A set of observing system simulation experiments (OSSEs) and experiments using real observations are conducted to assess MSVA in comparison with a single-scale variational analysis (SSVA). Results from the multistage OSSEs show that the multiscale covariance design improves the recovery of observation-supported structures. In the three-dimensional OSSE, MSVA reduces the global RMSE from 3.93 K for 1200 km SSVA to 3.78 K, and reduces the RMSE over withheld profiles from 3.56 K to 3.16 K. Using real SABER temperature observations, both three-dimensional analyses reduce MSIS–SABER observational space departures; the mean all-profile RMSE decreases from 11.43 K for MSIS to 5.97 K for 3D SSVA and 5.35 K for 3D MSVA. Comparison with unassimilated Aura/MLS observations gives corresponding RMSEs of 8.74, 7.60, and 7.39 K, providing an independent external evaluation of the analyses using real observations. These results demonstrate that representing multiple horizontal scales improves near-space temperature analysis relative to a single-scale formulation, as evidenced by controlled OSSEs and independent MLS evaluation.

Remote Sensing doi: 10.3390/rs18203530

Authors: Benjamin J. Fogg David G. Long

Tropical cyclones pose a significant hazard to coastal populations and shipping infrastructure. By helping provide timely warnings, ocean scatterometers can mitigate this threat with ocean wind estimates. In this study, the Soil Moisture Active/Passive (SMAP) mission is used to evaluate the effectiveness of wind estimates using L-band scatterometry for tropical cyclones. Standard-resolution SMAP scatterometer normalized radar backscatter (σ∘) measurements are enhanced to a much finer spatial resolution using the Ultra-High Resolution (UHR) algorithm, from which high-to-severe winds are estimated. The influence of σ∘ polarization (Horizontal HH-, Vertical VV-, and Cross X-pol) on wind estimates at high-to-severe winds is assessed. In this study, a new wind model is created with tropical cyclone Best-Track data, SMAP σ∘ measurements, and Numerical Weather Prediction winds. The Best-Track-derived wind model and SMAP UHR σ∘ measurements are used to create a Geophysical Model Function (GMF), and wind retrieval is performed. Analysis of wind retrieval estimates demonstrates that the HH- and VV-pol σ∘ tend to saturate at high-to-severe winds while X-pol measurements do not. Retrieved UHR winds generally align with the Best-Track derived wind model, although the errors can exceed desirable tolerances for speed and direction (i.e., less than 10% m/s and 20° Root Mean Square Error). Ultimately, these results suggest that L-band scatterometer wind estimation has potential for tropical cyclone monitoring by increasing the SNR of X-pol σ∘ measurements.

Remote Sensing doi: 10.3390/rs18203531

Authors: Jiazhen Bao Xin Zhao Panpan Ma Lijian Xie Aihua Shen Xu Han Junfeng Xu Yun He Aixia Liu Jinsong Deng

Automatic building extraction from high-resolution remote sensing imagery is vital for urban applications. Although the Segment Anything Model (SAM) offers powerful semantic capabilities, its direct application faces severe challenges. The inherent low-pass filtering of Vision Transformers typically causes blurred building edges, distorted right angles, and severe adhesion among dense clusters. To address these bottlenecks, we propose a Boundary-Aware Segment Anything Model via High-Frequency Enhanced and Residual Injection (HFER-SAM). First, a Multi-scale Semantic Adaptation Module (MSAM) uses domain adapters and coordinate attention to recover spatial priors for precise domain alignment. Second, a High-Frequency Residual Injection Module (HFRIM) employs a lightweight CNN to extract roof textures and edge gradients. These are injected as residual terms into the semantic stream to explicitly sharpen physical boundaries. Finally, a Boundary-Constrained Dual-Task Head (BC-DTH) utilizes morphological supervision to sever erroneous connections between adjacent buildings and resolve target adhesion. Experiments on the WHU and Inria datasets reveal that HFER-SAM achieves an Intersection over Union of 91.53% and 82.06%, respectively. Our method significantly outperforms existing state-of-the-art models in both accuracy and boundary fidelity, providing a robust avenue for high-precision building mapping.

Remote Sensing doi: 10.3390/rs18203529

Authors: Tao Li Honghai Ma Xun Geng Yinuo Zhang Pengying Liu Zhen Peng Jin Wang Jiyuan Hu

The dense control networks constructed for regional and global image blocks often contain a substantial number of redundant tie points that increase the computational burden of bundle adjustment (BA). Reducing this redundancy without weakening the geometric support of image pairs is essential for subsequent photogrammetric processing. Existing image-centered network-thinning methods reduce point counts but do not explicitly preserve direct links between individual image pairs. This paper presents an efficient control-network thinning method for planetary photogrammetry based on image-pair convex-hull computation and grid-based interval sampling. The proposed method first organizes the input control network into an image-pair measure map. It then separates each pair’s measures into hull and interior classes using Andrew’s monotone-chain algorithm and samples the two classes at different intervals. When a tie point is selected for an image pair, its measures in both images are retained together. An optional rule can also retain measures of the same surface feature in other overlapping images to help maintain connections across the image block. The method is designed to thin an existing control network after upstream quality filtering. It does not re-evaluate tie-point quality or add new control points. We implemented the method as the cnetoptimize application within the Integrated Software for Imagers and Spectrometers (ISIS), a widely used open-source planetary photogrammetric software system developed by the United States Geological Survey (USGS). Experiments used a large-scale Lunar Reconnaissance Orbiter (LRO) Narrow Angle Camera (NAC) network and a medium-scale Mars Express (MEX) High Resolution Stereo Camera (HRSC) network. At comparable compression ratios, cnetoptimize retained strict non-degenerate support for more image pairs than the image-centered ISIS cnetthinner. Across the tested datasets and settings, cnetoptimize preserved the direct connections of all originally non-degenerate image pairs at compression ratios up to 8.4×. In a downstream test on one weak HRSC pair, bundle adjustment with the thinned cnetoptimize network yielded subpixel image residuals and a local 25 m digital orthophoto map (DOM) seam root mean square (RMS) of 36.0 m. The method provides an ISIS-compatible approach to control-network thinning for planetary mapping.

Remote Sensing doi: 10.3390/rs18203528

Authors: Qi Sun Quanjun Jiao Shiyan Chen Lulu Pan Shun Zhang Wenjiang Huang Xinyu Zhu

How retrieval performance varies with simulation settings, neural architecture, and measured-data adaptation remains unclear for hyperspectral crop trait retrieval. Here, we evaluated winter wheat leaf area index (LAI) and canopy chlorophyll content (CCC) retrieval using two simulated datasets and four dual-branch neural architectures. Models were pretrained on two PROSAIL-D datasets with default and previously optimized leaf-angle settings and then adapted separately to ground and unmanned aerial vehicle (UAV) observations. Evaluation used five repetitions with physical-plot grouping, with approximately 30% of plot groups allocated to measured-data adaptation. Simulation-trained random forest (RF) and partial least squares regression served as baselines without measured-data adaptation. Adaptation reduced mean test root mean square error (RMSE) in all 32 neural model–dataset–target–platform combinations relative to direct simulation-based prediction. Dataset II, which used the previously optimized leaf-angle setting, reduced mean RMSE in 13 of 16 neural configurations before adaptation and 11 of 16 afterwards. No neural architecture was uniformly best. RF achieved the highest mean R2 in seven of eight simulation-only tasks and outperformed the adapted Dataset II CNN–Transformer on both UAV targets. Despite adaptation gains, all adapted neural configurations retained negative mean R2 for UAV CCC. These findings support measured-data adaptation within the tested protocol, while showing that the effects of simulation design and architecture are task dependent and that relative improvements do not necessarily imply adequate absolute predictive accuracy.

Remote Sensing doi: 10.3390/rs18203527

Authors: Javier Quille-Mamani German Huayna-Felipe Jorge Muchica-Huamantuma David Ascencios Edwin Pino-Vargas

Water stress in olive orchards is often mapped from the Normalized Difference Vegetation Index (NDVI), although in sparse orchards NDVI also reflects fractional cover and bare soil. We compared an NDVI-based classification with a METRIC-derived evaporative Crop Water Stress Index (CWSIET=1−ETa/ETo) over hyper-arid, groundwater-irrigated olive orchards of La Yarada–Los Palos (Peru), using Landsat 5/8/9 imagery from eighteen austral summers (2006–2026). Both classifications rely on operational thresholds that were not physiologically calibrated for olive: the NDVI ranges were defined from field-observed canopy condition, and the CWSIET classes are equal-width bins of the reference-normalized evaporative deficit. NDVI placed most of the cropland in severe-to-extreme classes and CWSIET mainly in mild-to-moderate classes (medians 0.26–0.41); NDVI was more severe over 89% of the cropland. The divergence grew as cover decreased, persisted at native thermal resolution and was reproduced by surface temperature, although both continuous indices correlated (r=−0.61). Without in situ physiological or flux measurements, these results establish a disagreement between indicators, not which of them matches plant water status. The cropland expanded fourfold (98 to 356 km2) with no trend in mean evapotranspiration or CWSIET. NDVI-based stress maps should be interpreted with caution in sparse orchards and combined with energy-balance, meteorological or field information.

Remote Sensing doi: 10.3390/rs18203525

Authors: Louis Snyders Joris Blommaert Jonathan Leon-Tavares Mary Langsdale Lorenzo Genesio

Directional anisotropy (DA) in thermal infrared (TIR) remote sensing introduces significant variability in land surface temperature (LST) retrievals, particularly in high-resolution observations. This study presents the empirical pixel-level modelling of DA, by using multi-angular airborne data from the NETSense campaign over an agricultural area near Grosseto, Italy. Six kernel-driven models (KDMs) are evaluated for their ability to fit directional effects, with the Vinnikov–Roujean–Lagouarde (Vinnikov–RL) model achieving the highest accuracy (RMSE = 1.07 K). At 40 m resolution, modelled directional effects reach median deviations of +1.80 K at the hotspot position, −2.03 K opposite to the Sun, and −1.37 K perpendicular to the solar principal plane, with extremes up to ±6 K, highlighting their operational significance for missions such as LSTM, TRISHNA, and SBG. To improve operational feasibility, two simplified models—adapted RL (RLa) and Vinnikov–RLa—are introduced, leveraging a consistent inverse relationship between hotspot magnitude and width to reduce parameter complexity via a power-law formulation. The analysis reveals that directional effects vary with spatial resolution and are influenced by canopy structure and solar geometry, while biophysical parameters alone are insufficient for coefficient estimation. Validation using an independent flight day confirms the robustness of the simplified models under different acquisition geometries, surface types, and geophysical conditions. A tiered model strategy is proposed for future satellite missions, recommending model selection based on day or night observations and data availability. These findings support the integration of directional correction into operational pipelines, enabling consistent and accurate LST products across platforms, resolutions, and observation conditions.

Remote Sensing doi: 10.3390/rs18203526

Authors: Iacopo Mancusi Roberto Colonna Carolina Filizzola Nicola Genzano Mariano Lisi Valerio Tramutoli

In recent decades, the sharp rise in global temperatures has accelerated the melting of the polar ice caps, causing icebergs to break off and subsequently drift across the sea surface. At the same time, the increase in navigable areas in the Arctic region has led to a continued growth in maritime traffic at higher latitudes, resulting in a consequent increase in the risk of iceberg–ship collisions. In this context, it is particularly important to provide vessels at sea with timely and up-to-date information on the presence of icebergs along routes in mid- and high-latitude waters. Given this challenge, this paper presents, for the first time, the Robust Satellite Techniques (RST) methodology employed at the global level to perform iceberg detection using imagery acquired only in the B3 band (centred at 560 nm) by the Sentinel-2 (S2) constellation. A test performed near Greenland (in September 2016) and in the Labrador Sea (in August 2021), showed that, in comparison with traditional fixed-threshold methods, this approach (easily implementable in the Google Earth Engine cloud platform) demonstrated: (a) very high sensitivity by doubling the number of detected icebergs (including the smaller ones); (b) high reliability with an overall false positive below 20% that becomes zero when other objects that may pose a hazard to navigation (e.g., dark ships, shallow reefs, etc.) are also considered.

Remote Sensing doi: 10.3390/rs18203524

Authors: Xiao Chen Shihao Hu Guangming Zhao Jucai Chang Yuhan Huang

Mining-associated water bodies are typically mapped as independent annual patches, obscuring their formation, transformation and continuity over time. Changes in visible water therefore need to be distinguished from changes in water body identity and surface state. Here, we develop an object-based, interval-aware framework to reconstruct water body lifecycles across seven domains in the Huainan and Huaibei coalfields, China, separating persistent identity from geometry and surface state. The framework integrates Landsat, Sentinel-2, PlanetScope and RapidEye imagery with contextual evidence to interpret changes and constrain event timing. Across these domains, the screened visible water area increased by 853.0 ha between the 2015–2017 and 2022–2024 epochs, comprising 1758.4 ha of gain and 905.4 ha of loss. The case histories documented continuity under photovoltaic cover and engineered subdivision, as well as terrestrial surface conversion with an unresolved termination status. In the examined cases, the changes in area and component overlap did not uniquely identify lifecycle events without temporal and spatial context evidence. Dense image sequences documented the early development and expansion of individual water bodies and bounded coalescence events. An independent blinded review reproduced 94.4% (17/18) of identity-relation decisions and all endpoint decisions in the original review sample. By distinguishing object continuity from visible water change, the framework complements area-based monitoring by providing evidence of water body lifecycle processes.

Remote Sensing doi: 10.3390/rs18203522

Authors: Taewook Kang

Terrestrial laser scanning (TLS) of bridges supports as-built reconstruction and digital-twin maintenance, but semantic segmentation of such scans is limited by annotation cost, and synthetic point clouds generated from building information models (BIM/IFC) are the established remedy. The virtual scanner behind such data is usually a single fixed design decision validated on synthetic test splits, leaving open whether such benchmarks can be trusted and which acquisition-realism effects justify their cost. This research decomposes virtual laser scanning of four real IFC bridge models into four switchable realism factors—occlusion, incidence-angle dropout, range-dependent noise, and per-station registration error—and measures each factor’s effect, in a leave-one-factor-out screening design, on transfer to twenty real TLS bridge scans. Three findings result. (1) Synthetic-test accuracy overstates real performance by up to +5.2 mean intersection-over-union (mIoU) points: the least realistic generator scores highest on its own test split while transferring no better, and realistic acquisition drives the overstatement towards zero. (2) Boundary-aware metrics, introduced here to infrastructure point clouds, separate component-interface failures that mIoU conflates. (3) After fine-tuning on fifteen real bridges, only realism pre-training improves significantly over training from scratch (+3.6 mIoU points, Welch t(4.0) = 3.48, p = 0.025, Hedges’ g = 2.27), with a same-direction gain at component interfaces. IFC-derived and real-scan supervision are quantifiably complementary.

Remote Sensing doi: 10.3390/rs18203523

Authors: Jia Liu Wan-Li Yang Yi-Ming Luo Wei-Shi Chen Dong-Li Wu Shi-Hao Zhang

Bird activity levels present an increasing trend in China. Bird strike risks for the aviation industry are positively correlated with bird activity. Migratory birds pose greater threats to airplanes due to their broader distribution and larger mass. A reliable broad-area surveillance solution for migratory birds is critical for bird strike warning and prevention. This paper introduces a broad-area migratory bird identification model based on weather radar systems. A customized feature space and a sample-separation optimization strategy are proposed for more accurate identification of bird targets in weather radar resolution volumes. A new bird density calculation model is proposed to accommodate nonuniform bird distributions and provide a more reasonable bird quantity interpretation. The relevance between bird density and weather parameters is modeled by specific data sampling and a numerical regression model. Weather radar data files from two stations in China are used to verify the effectiveness of the proposed method. Results indicate that the bird identification model could provide reasonable bird surveillance information, and the density prediction models present good consistency with existing datasets. Differences among radar stations and seasons prove the necessity of building individual bird density prediction models. Identification and prediction models are expected to be further optimized with more data support and cross-validation information. Quantitative and qualitative prediction products provide valuable references for bird strike prediction and prevention.

Remote Sensing doi: 10.3390/rs18203521

Authors: Fanghao Wei Xiaoli Huang Meng Cao Yakui Shao

Forest-fire occurrence susceptibility prediction remains challenging in mountainous regions because fire occurrence is controlled by nonlinear interactions among climatic, vegetation, topographic, and anthropogenic factors. Developing interpretable prediction models that integrate multi-source environmental information is essential for improving susceptibility assessment and understanding predictor contributions. Using 2003–2022 MODIS active-fire records and multi-source environmental data from the Dabie Mountains, this study developed an explainable XGBoost-based forest-fire occurrence susceptibility prediction framework. The predictive performance of XGBoost was compared with Random Forest and radial basis function support vector machine (RBF-SVM) using five-fold cross-validation, and TreeSHAP was applied to interpret predictor contributions, nonlinear responses, and interactions. KDE, standard deviational ellipses (SDEs), and GeoDetector were further used to characterize historical fire patterns and complementary spatial associations, while an independent climate-scale XGBoost model was applied to evaluate future susceptibility sensitivity under SSP126 and SSP585 scenarios. XGBoost achieved an ROC-AUC of 0.850, and log-transformed population showed the largest global SHAP contribution. GeoDetector identified precipitation and relative humidity as the strongest spatial associations with historical fire-density patterns across KDE bandwidths. The climate-scale model achieved a pooled five-fold ROC-AUC of 0.830 and indicated stronger late-century susceptibility-score redistribution under SSP585 than under SSP126. Together, these results demonstrate an explainable susceptibility modeling approach that integrates predictive modeling, model interpretation, spatial analysis, and future climate assessment. The outputs represent relative susceptibility and scenario-specific sensitivity responses rather than unconditional fire probabilities or deterministic forecasts.

Remote Sensing doi: 10.3390/rs18203520

Authors: Tingting Pan Anyu Du Shuli Cheng Yunpeng Chen

With the rapid development of aerospace technology, remote sensing images contain increasingly rich feature information, while large variations in object scales and imbalanced category distributions pose great challenges to remote sensing semantic segmentation. Current mainstream methods often rely on attention mechanisms for global modeling, yet they tend to over-average features, making it difficult to capture minority categories and preserve fine-grained details. To address these limitations, we propose a Dual Category-Guided Spatial Mamba Network (DGSMNet) that combines dual category guidance with a spatial Mamba mechanism. Specifically, a Dual-Stage Category Guidance (DSCG) module is designed to extract class information via wavelet transforms and guide feature learning, ensuring proportional representation of all categories across channels and enabling the model to learn underrepresented classes. In addition, a Spatial Mamba (SPM) module is introduced to enhance long-range modeling while retaining sensitivity to local details. By integrating DSCG and SPM, the proposed DGSMNet improves category-aware representation and spatial modeling on the evaluated benchmarks. Extensive experiments on the ISPRS Vaihingen and Potsdam datasets demonstrate that DGSMNet achieves competitive performance among the compared methods.

Remote Sensing doi: 10.3390/rs18203519

Authors: Linning Zhou Chengzhi Yu Yichen Sun Chaohui Zheng

Mangrove mapping increasingly integrates multisource remote sensing, yet the geographically transferable value of additional sensors remains uncertain. This study evaluated Sentinel-2 multispectral imagery, Sentinel-1 and GF-3 SAR, GF-5 AHSI hyperspectral observations, terrain variables, and sensor-independent coastal-context information across five coastal regions of southern China. Five nested feature configurations (D1–D5) were tested with Random Forest, XGBoost, DeepLabV3, and SegFormer under true leave-one-study-area-out validation on 349,898 common 10 m coordinates; deep models were replicated across three random seeds. Model-averaged F1-scores were 0.9025, 0.9031, 0.9123, 0.9134, and 0.9220 for D1–D5, respectively, but differences were not statistically supported across held-out regions (Friedman χ2 = 0.48, p = 0.9754). Sentinel-1, GF-3, and their interaction showed small, geographically inconsistent contrasts. GF-5 effects were model- and preprocessing-dependent, and cross-scene normalization altered performance in contrasting directions. Gain and grouped-permutation analyses indicated model-specific predictor use and sensitivity rather than independent sensor contribution. Overall, multisource benefits depended on sensor configuration, model representation, geographic domain, and preprocessing conditions, supporting context-specific sensor selection and geographically independent validation.

Remote Sensing doi: 10.3390/rs18203518

Authors: Jie Liu Gongjian Wen Weiwei Wen Junlong Chen Weiqi Fu Shunlong Zhou

Urban planning and land-use monitoring require not only semantic land-cover maps, but also image-like overhead observations that preserve the spatial organization of urban scenes. Conventional remote-sensing segmentation and image-generation methods usually operate by obtaining available overhead imagery or in reliable overhead conditions. Street-level panoramas are easier to collect and update, but they observe the target scene from a limited terrestrial viewpoint. Recovering overhead information from a single panorama, therefore, requires inferring from an unobserved land-cover footprint before synthesizing a nadir-view remote-sensing image. We propose SDU-G2A, a two-stage framework for aerial image synthesis from a single ground panorama. Stage I predicts a probabilistic BEV land-cover footprint from the input panorama. The proposed Azimuthal-Radial Mapper propagates azimuthal evidence and applies distance-aware supervision to improve semantic recovery in occluded and far-field regions. Stage II converts the inferred footprint into a remote-sensing-like aerial image through diffusion-based synthesis, making the reconstructed urban structure available in an overhead-image form for visual interpretation and geospatial-correspondence evaluation. The main difficulty is that the inferred BEV footprint exhibits spatially varying reliability. Treating it as a deterministic condition can force cross-view segmentation errors into the generated image. SDU-G2A, therefore, derives a pixel-wise uncertainty map from Stage I predictions and uses it as a spatial reliability layer. This design preserves structures with high prediction confidence while allowing uncertain regions to be completed under text-guided and prior-driven generation. Experiments on VIGORv2 show that SDU-G2A improves BEV land-cover estimation, reduces radial prediction errors, and achieves stronger geospatial correspondence and aerial-feature distributional realism than representative G2A baselines. These results indicate that self-derived land-cover uncertainty can bridge cross-view semantic footprint inference and reliable remote-sensing image synthesis from accessible street-level observations.

Remote Sensing doi: 10.3390/rs18203517

Authors: Bocheng Yang Lujie Zheng Dingyang Lyu Jian Liu Bo Chen

Resource-coupled satellite-ground scheduling requires decisions whose computational, energy, thermal, and communication effects persist across task arrivals. We tested whether the deep Q-network (DQN) family provides a reproducible advantage over short-horizon planning in a synthetic benchmark with clipped, soft-constrained resources. Six DQN objectives used the same state representation, network architecture, replay procedure, optimizer, interaction budget, 20 model seeds, and held-out traces. Across five fully coupled regimes, every one of the 30 objective–regime comparisons had lower mean episode cost for DQN than for four-step model-predictive control (MPC-4). Every paired 95% bootstrap interval excluded zero, and all Holm-adjusted sign tests supported this ordering. The least-favorable mean difference was −5.92, with a family-wide simultaneous 95% upper bound of −5.03. Differences among DQN objectives were smaller and scenario-dependent. Additional nominal tests assessed discount-factor, training-length, greedy-rollout, and CPU decision-time sensitivity. These findings apply only to this synthetic, soft-constraint benchmark. Hard energy and thermal limits can change the feasible action set and therefore the algorithm ordering.

Remote Sensing doi: 10.3390/rs18203516

Authors: Jingjing Liu Jinliang Gao Pengcheng Wang Lingbo Qiao Yayun Cheng

The detection and spatial assessment of buried water pipeline leaks are critical for preventing resource loss and geohazards such as soil erosion and subsidence. This paper presents an initial experimental demonstration of a microwave imaging system designed for mapping subsurface anomalies induced by buried water pipeline leaks. The proposed system integrates a custom-designed broadband open-sleeve dipole antenna with a vector network analyzer, power amplifier, and a precision linear motion stage that emulates synthetic aperture radar scanning for controlled measurement validation. The antenna achieves a 66.7% impedance bandwidth (1∼2 GHz) and a stable gain of approximately 8 dBi. A phase-compensated back-projection algorithm is employed to generate two-dimensional ground-projected scattering maps. Controlled experiments involving wet-sand regions, surface potholes, and subsurface cavity surrogates yield averaged within-image response differences of 2.1∼5.2 dB between anomaly-associated regions and ground background regions. Under the specified acquisition conditions, the theoretical azimuth and range resolutions are 0.28 m and 0.15 m, respectively. These results provide preliminary evidence of imaging feasibility under the tested surrogate-target conditions. Further validation with actual leakage and representative soil and surface-cover conditions is required to assess applicability to infrastructure monitoring.

Remote Sensing doi: 10.3390/rs18203515

Authors: Zeping Wang Honglei Yang Pan Wang Jiawang Wang Tiantian Wang

Groundwater recovery can reduce land subsidence without immediately reversing aquifer-system compaction. This review synthesizes 101 original studies of the Beijing Plain to examine deformation during groundwater depletion and subsequent recovery. The evidence includes interferometric synthetic aperture radar (InSAR), leveling, groundwater records, borehole deformation measurements and laboratory experiments. Historical surveys and satellite observations document expanding subsidence centers followed by widespread deceleration and localized rebound. Groundwater pressure changes are the principal regional forcing. Sedimentary architecture, hydraulic connectivity and loading history govern the magnitude and timing of deformation. Delayed pressure equilibration and time-dependent strain can sustain compaction after aquifer heads rise, while irreversible strain limits recovery of lost elevation. Surface uplift can therefore coexist with compaction at depth and does not establish restoration of aquifer-system storage capacity. Observational coverage remains uneven: 75 studies use InSAR, whereas 17 report borehole deformation measurements. Paired pressure and interval-deformation measurements are needed to identify the sources of continuing compaction. Differential ground movement remains relevant to infrastructure assessment even where regional subsidence slows.

Remote Sensing doi: 10.3390/rs18203514

Authors: Jin Li Hanqi Xie Haipeng Yi Riwen Lai

Chinese fir (Cunninghamia lanceolata) is a major plantation species in southern China, and diameter at breast height (DBH) is an essential variable for forest inventory, biomass estimation, and carbon accounting. However, direct DBH extraction from UAV-LiDAR is limited in near-mature stands because lower-stem points are often occluded by dense canopies. This study developed a UAV-LiDAR-based workflow for individual tree segmentation, structural parameter extraction, and multivariate DBH estimation in near-mature Chinese fir plantations. Field measurements were conducted in 16 sample plots. Segmentation was evaluated against a common reference set of 428 field trees across the three CHM configurations, and 412 trees with unambiguous one-to-one field-to-crown matches were retained for DBH modeling. Canopy height models (CHMs) with spatial resolutions of 0.5, 0.25, and 0.1 m were compared using watershed segmentation. For the selected 0.25 m CHM, maximum crown-polygon height (CHM_H) and equivalent-area crown diameter (CD) were extracted. CD was calculated from crown area (CA), and CA was therefore excluded from all final multivariate models to avoid deterministic redundancy. Among the tested resolution-smoothing configurations, the 0.25 m CHM yielded the highest recall, precision, and F-score values of 94.86%, 92.69%, and 93.76%, respectively. In the height-metric comparison, CHM_H achieved R2 = 0.802 and RMSE = 0.928 m, slightly outperforming H90 under the prespecified RMSE criterion. On the same holdout, classical height–crown allometry yielded R2 = 0.605, RMSE = 1.701 cm, and MAE = 1.297 cm. BPNN retained the highest reported validation R2 (0.662) and the lowest reported RMSE (1.581 cm) and was selected as the best-performing model under the study design. The workflow provides a practical basis for non-destructive DBH estimation and precision inventory in near-mature Chinese fir plantations.

Remote Sensing doi: 10.3390/rs18203513

Authors: Mingli Xie Congjian Li Haoran Zhou Shixiong Liu Yunfeng Wan Shuli Zhu Wensheng Wang

Aerial–ground visual tracking facilitates dynamic remote sensing and scene monitoring by integrating complementary observations from unmanned aerial vehicle (UAV) and ground cameras, allowing one view to compensate when the target is poorly observed or temporarily invisible in the other. However, substantial cross-view variations in target appearance, scale, and geometry make it difficult to establish reliable target correspondences and effectively fuse complementary cues. To address these challenges, we propose CVAC-Net, a Cross-View Alignment and Adaptive Collaboration Network for aerial–ground visual tracking. Specifically, CVAC-Net uses cross-view contrastive learning to reduce semantic differences and a learned spatial offset to adjust the relative positions of aerial- and ground-view target features within their normalized search crops. Next, it uses the similarity between the main-view template and the aligned auxiliary-view search features to generate an importance map that reweights redundant and target-irrelevant regions, thereby improving the reliability of feature aggregation. Finally, its view-specific branches select and weight the top-K encoder outputs to integrate complementary cross-view features. Extensive experiments on the UAV–ground visual tracking (UGVT) benchmark show that CVAC-Net improves the baseline by 3.2, 3.1, and 1.7 percentage points in success rate (SR), normalized precision rate (NPR), and precision rate (PR), respectively. We further evaluate CVAC-Net using a two-dimensional (2D) single-object tracking protocol derived from Griffin using its predefined training and validation splits, providing supplementary validation on a simulated aerial–ground distribution.

Remote Sensing doi: 10.3390/rs18203512

Authors: Luísa Lamas Teotónio Barroqueiro

Long-term high-frequency radar (HFR) observations provide a unique opportunity to examine how coastal surface currents organise into recurrent current-field configurations. This study uses the 2013–2024 HFR-Lisbon surface-current record to identify sub-inertial regimes in the Lisbon coastal region, a dynamically complex sector of the western Iberian margin. Hourly HFR currents were quality controlled, filtered with a centred 33 h rolling mean and restricted to a sub-inertial analysis domain defined from long-term coverage and residual semidiurnal-energy criteria. A physics-guided K-means clustering approach was then applied to seven physically interpretable descriptors of current intensity, directional coherence, spatial contrast, temporal change and bathymetry-based current-speed distribution. Four recurrent regimes were identified: a low-energy regime, a shelf-intensified regime, a moderate slope/offshore regime and an energetic offshore-intensified regime. The regimes show coherent spatial structures, marked seasonal organisation, persistence over several days and preferential temporal pathways. ERA5 winds, analysed a posteriori, show the strongest contrast between the low-energy regime, associated primarily with weaker-than-average winds, and the energetic offshore-intensified regime, associated with stronger and more northerly wind conditions. The results demonstrate the value of long-term HFR observations and physics-guided clustering for identifying and interpreting recurrent coastal surface-current configurations.

Remote Sensing doi: 10.3390/rs18203511

Authors: Muhammad Nasar Ahmad Lichuan Gu Shuai Yang Xiaohui Yuan George D. Bathrellos

Accurate identification of paddy rice from satellite imagery is challenging in heterogeneous agricultural landscapes because water, wetland, and vigorous summer vegetation can exhibit spectral and radar responses similar to those of rice at individual observation dates. This study develops an interpretable, training-independent rice-mapping approach, meaning that the final classification does not rely on supervised classifier training, although manually interpreted locations were used for exploratory threshold selection. The approach identifies paddy fields from their characteristic transition from early-season inundation to subsequent canopy development. These phenological signals were integrated using explicit decision rules within a cropland-constrained domain, followed by persistent-water exclusion and spatial refinement. The framework was applied to the 2025 rice-growing season in the agricultural regions analyzed in Heilongjiang Province, China, using 30 m output mapping. Validation with 695 independent reference observations yielded an overall accuracy of 92.09%, Cohen’s kappa of 0.791, rice producer’s accuracy of 94.87%, user’s accuracy of 75.90%, and an F1-score of 84.33%. The high producer’s accuracy indicates effective detection of reference rice locations, whereas the lower user’s accuracy shows that commission error remains the principal limitation, particularly in areas with moisture and vegetation conditions resembling paddy fields. Sequential screening further showed that combining phenological evidence with cropland and water constraints substantially reduced the candidate mapping domain. The proposed approach provides an auditable alternative to training-dependent classification workflows and offers a transparent basis for repeatable rice mapping and agricultural monitoring, while requiring further evaluation across multiple years and regions before broader operational transfer.

Remote Sensing doi: 10.3390/rs18203510

Authors: Giovanni Salvatore Di Bella Arianna Beatrice Malaguti Alessandro La Spina Eleonora Amato Claudia Corradino

Effusive volcanic eruptions pose significant hazards to populated areas, making the accurate dating of lava flows essential for reconstructing eruptive histories and assessing volcanic risk. Traditional dating techniques, such as radiometric, paleomagnetic and lichenometric methods, are highly reliable but require direct sampling, complex analytical procedures, and substantial financial resources. Satellite remote sensing provides a non-invasive and spatially extensive approach that can complement conventional methods in the regional investigation of exposed volcanic surfaces. This study presents a two-stage hybrid classification–regression framework for empirical age estimation and chronological mapping of historical lava surfaces on Mount Etna, spanning the 17th to the 21st centuries. In the first stage, a Random Forest Classifier distinguishes background, pre-1600 lava surfaces, and post-1600 historical lava surfaces, thereby defining the spatial applicability domain of the chronological model. In the second stage, a Random Forest Regressor estimates age only for pixels classified as post-1600 lava by integrating ten Sentinel-2 surface-reflectance bands, elevation derived from the Digital Elevation Model (DEM), and a binary north-flank spatial indicator. On the balanced pixel-level test dataset, the RFR achieved an R2 of 0.549, a Mean Absolute Error of 66.2 years, and a Root Mean Squared Error of 89.0 years. Using the median prediction for each test eruption, the corresponding eruption-level values were R2=0.701, MAE =51.6 years, and RMSE =67.2 years. Prediction errors were larger for the oldest evaluated surfaces, indicating reduced empirical discrimination near the lower boundary of the calibrated chronological domain. This pattern may reflect calibration-boundary effects, sparse chronological coverage, eruption-specific variability, and possible convergence of surface spectral properties. Comparison with an RBF-SVR showed comparable but metric-dependent performance, without establishing the overall superiority of either regression architecture. Permutation analysis indicated that most predictive information within this dataset was associated with the Sentinel-2 spectral component, particularly the Red Edge bands. The proposed method should therefore be interpreted as a complementary fast empirical lava age-estimation and regional screening framework for exposed historical lava surfaces, rather than as a substitute for radiometric, paleomagnetic, or field-based dating.

Remote Sensing doi: 10.3390/rs18203508

Authors: Quanxiang Wang Zhaofa Zhou Zhili Zhang

Small objects in UAV imagery often exhibit weak visual representations, suffer from background clutter, and receive insufficient cross-scale feature interaction. To address these challenges, we develop MCI-DETR, a multi-level contextual interaction detector built upon RT-DETR. First, a Local-Context Modulated Interaction Module (LCMI) is designed to selectively modulate contextual features using local structural responses, thereby suppressing irrelevant background responses while expanding the effective receptive field. Second, CATM is incorporated into the high-level feature interaction stage to form a Context Interaction Module (CCIM), strengthening target-related information through spatial- and channel-domain context mapping and additive Query–Key interaction and reducing the interference of complex backgrounds with the high-level semantic representation of weak small objects. Finally, a Multi-Scale Adaptive Gated Fusion Module (MSAGF) is designed to dynamically adjust the fusion contributions of information at different scales according to the input features, while improving the quality of cross-scale feature fusion through gated coordination between local details and global context. On VisDrone2019, MCI-DETR surpasses RT-DETR by 5.2%, 4.8%, and 3.2% in mAP50, mAP50-95, and APS, respectively, demonstrating enhanced robustness for small-object recognition in cluttered aerial imagery. MCI-DETR also achieves competitive performance on UAVDT and HIT-UAV against representative detectors, demonstrating the effectiveness of the proposed design under different UAV scenes and imaging conditions. Overall, MCI-DETR strengthens the representation of small objects in aerial scenes and maintains reliable detection performance in the presence of cluttered backgrounds.

Remote Sensing doi: 10.3390/rs18203509

Authors: Tianyi Wen Huanyu Liu Yuanyuan Sheng Kefan Qu Junbao Li Shen Wang Shengbin Luo Wang Guoqing Wu

Range–Doppler (RD) spectrogram-based maritime target recognition requires effective feature learning when labeled training samples are limited and the range and Doppler axes convey different physical information. We propose DAMD-Net, which combines Asymmetric Dimension-Aware Fusion (ADF) with a Dual-Directional Attention Module (DDAM). ADF integrates range-wise scattering features, Doppler-wise motion features, and local two-dimensional context, while DDAM recalibrates the fused representation through axis-specific descriptors and learnable fusion weights. With 5% of the simulated dataset used for training, DAMD-Net achieves 86.15±0.43% accuracy and 83.57±0.41% macro-F1. With 10% used for training, the corresponding results are 89.55±0.39% accuracy and 87.48±0.51% macro-F1. At the smaller 1% and 2% training budgets, DAMD-Net improves macro-F1 over VGG16 by 1.64 and 1.82 percentage points, respectively, with positive unadjusted 95% paired confidence intervals. Each training budget uses an additional 4049 labeled validation samples. Component ablations, kernel and pooling sensitivity studies, and condition-holdout evaluations examine the contribution of directional modeling; DAMD-Net has higher mean accuracy than VGG16 in all three elevation-holdout intervals. Controlled perturbation tests show advantages under partial Doppler masking and structured synthetic clutter: at a baseline-RD-to-added-clutter ratio of −10 dB, accuracy reaches 62.74%, compared with 36.68% for VGG16. On the external MTDSP dataset, DAMD-Net achieves 97.13% accuracy; on the measured LSS-DAUR dataset, it achieves 75.00±14.03% accuracy under five-fold cross-validation. DAMD-Net also reduces parameters by 89.9% and registered FLOPs by 52.6% relative to VGG16. Together, these findings support explicit RD anisotropy modeling for recognition with restricted training data, with complementary benefits in the evaluated disturbance conditions and model storage.

Remote Sensing doi: 10.3390/rs18203507

Authors: Rubén Echeverría Luis Santamaría Fernando Parra Adrián Delgado Adrián García-Gutiérrez

The vertical wind profile within the Atmospheric Boundary Layer (ABL) is a critical variable for a wide range of applications, including wind energy, air traffic and Unmanned Aerial System (UAS) management, pollutant dispersion, and urban air mobility. However, its accurate characterization typically requires expensive vertical-profiling LiDAR instruments, which limits the number of locations that can be simultaneously monitored. Building upon previous work in which machine-learning algorithms were trained on LiDAR data from the University of León (Spain) to reconstruct full ABL wind profiles from a single near-surface measurement, this study replaces the traditional data-driven neural network approach with Physics-Informed Neural Networks (PINNs), embedding a physical constraint directly into the loss function as evaluated by automatic differentiation, and tests two such constraints under architectural control: the Ekman boundary-layer momentum balance, closed with a Monin–Obukhov eddy viscosity, and a shear-constrained formulation based on a power-law profile with an inferred exponent. With dense supervision at the target heights, neither constraint improves on an architecture-matched network without a physical term (RMSE 1.35–1.36 m/s, direction MAE 20.1–20.4∘ across all three, Kruskal–Wallis p=0.0125). In a vertical-extrapolation protocol, training with gates up to 111 m, selecting models by their skill at 140 m, and testing blind on 170–300 m, the Ekman constraint shows no detectable advantage over its own architectural control. In contrast, the shear-constrained model does, reaching 1.80±0.01 m/s speed RMSE and 27.0∘±0.3∘ direction MAE against 2.66±0.10 m/s and 34.2∘±1.8∘ for the Ekman-constrained arm and 2.61±0.12 m/s and 33.9∘±2.0∘ for the physics-free arm, which are statistically indistinguishable from one another; the shear-constrained gain is confirmed against its own architectural control (2.64±0.08 m/s, 33.0∘±1.8∘, matching the physics-free arm), ruling out the additional output as the explanation, and converts an erratic extrapolator into a stable one. The results are consistent with the benefit of embedding physics into the loss being therefore specific to the extrapolation regime and to the formulation tested, not a generic property of physics-informed learning. These findings are based on data from a single suburban site, and multi-site validation remains necessary before they can be generalized. By eliminating the need for continuous LiDAR operation after the training phase, the proposed methodology aims to enable cost-effective monitoring of multiple locations with a single instrument.

Remote Sensing doi: 10.3390/rs18203506

Authors: Zijing Song Panlong Tan Xiaoyu Zhang

Synthetic aperture radar (SAR) ship detection is essential for maritime surveillance, yet coherent speckle can obscure tiny offshore vessels, while high-response coastal structures can produce target-like responses. Repeated downsampling further weakens sparse vessel signatures, whereas a separate high-resolution prediction head increases dense prediction costs. We propose a dual-domain detector that embeds spatial–frequency encoding, multiscale refinement, and high-resolution feature transfer within the backbone–neck–pyramid representation. Scharr-derived boundary cues are coupled with learned frequency context and remain discriminative as features are fused across scales, while sample-preserving space-to-depth transformation and global–local receptive-field modeling integrate shallow detail into semantically stronger P3 features without introducing a separate P2 prediction head. Coordinated FSEM, MSFE, and SOEP modules preserve fine vessel structure and reduce clutter-induced ambiguity throughout the detection hierarchy. Experiments on SSDD, HRSID, and RSDD-SAR evaluate detection accuracy, cross-dataset transfer, limited-data behavior, sensitivity to controlled perturbations, and edge deployment. Across 30 seed-matched SSDD runs, the method improves mAP50:95 over the baseline by 2.70±0.34 percentage points. It achieves 69.3% and 71.5% mAP50:95 on HRSID and RSDD. On the Atlas 200I platform, the compiled FP16 model retains 69.3% mAP50:95 while running at 101.8 frames/s.

Remote Sensing doi: 10.3390/rs18203505

Authors: Muhammad Ali Yuxiao Qin Zeeshan Afzal Danish Raza Alisher Skabylov Dauren Zhexebay Margulan Ibraimov Alexei Belotserkovsky Alexander Nedzved

Surface deformation monitoring plays a crucial role in various fields, including geology, geodesy, and environmental studies. This study provides insights into the application of non-linear PSInSAR techniques for studying surface deformation associated with urban growth and groundwater pumping. A Persistent Scatterer InSAR (PSInSAR) analysis employing a time-segmented (non-linear) velocity model was applied to 119 Sentinel-1A ascending acquisitions spanning 10 February 2017 to 27 December 2020, yielding 38,594 persistent scatterers. The 33,238 scatterers were retained at a temporal coherence threshold of 0.70 and a mean density of 18.6 PS/km2 across the district, rising to 57 PS/km2 within the urban core, with annual mean line-of-sight (LOS) velocities measured for each calendar year. Temporal validation identifies a modest, origin-dependent forecasting benefit from segmentation in the urban core but no consistent advantage under contiguous block testing. Land use/land cover (LULC) was mapped from Landsat imagery for 2000, 2007, 2014, and 2020 using a random forest classifier in Google Earth Engine. LOS velocities span approximately +40 to −60 mm/yr over the mapped range, with the most negative values concentrated in the central urban core. Paired annual comparisons, accounting for spatial dependence and Holm adjustment across 15 tests, did not establish significant changes in the 0–2 km core. Significant contrasts occurred in two comparison annuli, but their physical origin remains unresolved. Land use classification maps for the years 2000, 2007, 2014, and 2020 show a significant expansion in built-up areas, coupled with a reduction in vegetation and, after adjustment for classification accuracy, in water bodies, which indicates the increasing pressure of urban growth on the environment. Lahore has witnessed a substantial increase in population due to urban migration, natural growth, and economic opportunities. Subsidence can lead to structural damage to buildings, infrastructure, and underground utilities. It is a serious concern in Lahore, particularly in areas where groundwater pumping is excessive.

Remote Sensing doi: 10.3390/rs18203504

Authors: Xiaolin Lan Jinkai Tan Sheng Chen Zhi Li Jianyu Zhao Chenshuai Liu Dongdong Wang Yanping Li

Accurate radar-based quantitative precipitation estimation (QPE) plays a critical role in disaster prevention and reduction. In mountainous terrain, blockage of low-elevation radar beams can lead to missing low-level precipitation observations, while errors in specific differential phase KDP retrieval further limit QPE accuracy. To address these issues, this study uses data from two S-band dual-polarization weather radars located in Fangchenggang and Beihai, together with ground rain gauges, local raindrop size distribution (DSD) data, and digital elevation model (DEM) data. Hybrid-scan elevation construction and dual-radar mosaicking are adopted to improve the effective coverage of low-elevation observations, and a threshold-fusion strategy for KDP, denoted as KDP_TF, is proposed. The strategy selects the larger of the conventionally quality-controlled KDP_QC and variationally reconstructed KDP_VR at each valid gate. Based on the CSU-HIDRO algorithm, three QPE products, namely the quality-control QPE (QC_QPE), the variational-reconstruction QPE (VR_QPE), and the threshold-fusion QPE (TF_QPE), are generated. The precipitation events associated with Typhoon Yagi in 2024 and Typhoon Matmo in 2025 are selected for evaluation. The results indicate that TF_QPE most effectively mitigates the underestimation of 24 h accumulated precipitation. Compared with QC_QPE, the magnitude of the negative relative bias (RB) of TF_QPE is reduced, from −27.30% to −20.07% and from −16.56% to −9.78%, respectively, for the Yagi and Matmo events; the root-mean-square error (RMSE) decreases from 46.76 to 43.79 mm and from 43.27 to 43.13 mm, respectively. The hourly precipitation results further show that TF_QPE better captures the peak intensity of heavy precipitation and exhibits no evident temporal lag relative to rain gauge observations. For the heavy precipitation category, TF_QPE also yields the highest probability of detection (POD) and critical success index (CSI). These results suggest that the QPE framework developed in this study, which integrates hybrid-scan elevation construction, dual-radar mosaicking, and threshold-fusion-based KDP optimization, can improve the effective low-elevation observational coverage in mountainous terrain and mitigate the underestimation of typhoon-induced heavy precipitation in radar QPE. These results provide technical references for typhoon-related heavy-rainfall monitoring and flood early-warning as well as radar-QPE applications over coastal Guangxi.

Remote Sensing doi: 10.3390/rs18203501

Authors: Don Hong Katelyn Reeps Maya Nobriga Xiaomei Lu

Karenia brevis, a harmful algal bloom (HAB) organism behind Florida’s red tide, persists in the Gulf of Mexico year-round at low concentrations. It can rapidly multiply and cause toxic outbreaks that harm marine life and human respiratory health. Red tide has been observed to suppress or intensify following tropical cyclones. Hurricanes Helene and Milton made landfall on Florida’s Gulf Coast 13 days apart in fall 2024, whereafter K. brevis proliferated along southwest Florida, with cell concentrations reaching millions of cells per liter. Remote sensing (MODIS Aqua ocean-color and sea surface temperature, 2024–2025 hurricane seasons), in situ, and hurricane best-track/reanalysis data were examined with time series analysis, statistical analysis, and spatial analysis. We found a hurricane-associated cascade within 18 days of Milton’s landfall: wind forcing, then river discharge, then chlorophyll a and K. brevis peaks, followed ~6 weeks later by another K. brevis peak within ~2% of the near-term peak in magnitude. These findings indicate that elevated wind speed and river discharge, used as a proxy for potential nutrient delivery, during and after hurricane passage are associated with a long-term K. brevis response.

Remote Sensing doi: 10.3390/rs18203503

Authors: Nana Zhang Lijun Wang Jiayao Wang Wenru Jia Zhigang Han Mengqi Wang Weiguo Fu

Characterizing the spatiotemporal differences between meteorological and agricultural drought is important for drought monitoring and risk management in major grain-producing regions. However, coordinated assessment of meteorological anomalies and crop moisture responses using multi-source data remains challenging. In this study, an integrated multi-source framework was developed to jointly characterize meteorological and agricultural drought in winter wheat across six selected growing seasons in Henan Province, China. The standardized precipitation index (SPI) and vegetation water index (VWI) were used to characterize meteorological and agricultural drought, respectively. Furthermore, standard deviational ellipse analysis and center-based spatial statistics were used to characterize drought distribution, orientation, and spatial migration, while Pearson correlation analysis was used to examine statistical associations among SPI-defined drought, VWI-defined drought, and effective irrigated area. Across the selected study periods, meteorological and agricultural drought showed distinct temporal and spatial patterns. In addition, the SPI–VWI association varied with drought severity: light drought showed the strongest positive association (r = 0.64), moderate drought showed a weak negative association (r = −0.20), and severe drought showed a moderate positive association (r = 0.57). The association between total SPI- and VWI-defined drought areas was weak (r = −0.07). At the common 1-km analysis scale, VWI-defined agricultural drought maintained a more stable spatial center and major-axis orientation than SPI-defined meteorological drought, while the difference in ellipse extent was more pronounced in the earlier study periods and narrowed markedly in the later periods. Effective irrigated area showed an exploratory negative association with total VWI-defined agricultural drought area (r = −0.72, p = 0.11; n = 6); although the association did not reach conventional statistical significance, its direction identifies irrigation conditions as relevant contextual information when interpreting variations in agricultural drought. The integrated framework provides a complementary basis for jointly assessing meteorological stress and crop moisture conditions and may support drought-risk assessment and irrigation management in major winter wheat-producing regions.

Remote Sensing doi: 10.3390/rs18203502

Authors: Wenhong Bai Zejun Xia Derong Zhou Gang Wei Cunye Wang Yan Huang Yujuan Huan Jiashuang Jiao

The Golmud River Basin is a typical inland arid river basin in the Qaidam Basin, and the dynamics of its groundwater storage are of great importance for regional ecological security and water resources management. To address the limited spatial representativeness of traditional hydrological station networks and the relatively low spatial resolution of satellite gravimetry such as the Gravity Recovery and Climate Experiment (GRACE), this study develops a multi-source geodetic inversion framework for groundwater estimation by using Interferometric Synthetic Aperture Radar (InSAR), Global Navigation Satellite System (GNSS), and GRACE satellite gravimetry. Surface deformation with high spatial resolution over the Golmud River Basin is derived from Sentinel-1 SAR images from 2017 to 2021. After correcting for modeled non-hydrological loading effects and screening observations to reduce the influence of potential non-hydrological signals, groundwater spatiotemporal variations are inverted based on elastic loading theory. The results show that from 2017 to 2021, surface deformation in the Golmud River Basin is jointly controlled by river dynamics, mineral extraction, and urbanization, forming a spatial pattern of river corridor zones and mining-related patches. Groundwater storage shows an overall increasing trend during the study period, with a particularly significant increase in the southwestern region, reaching more than 20 mm/a. A synthetic experiment demonstrates that the proposed method can recover prescribed groundwater signals at a spatial resolution of approximately 40 km under the tested conditions. Compared with the InSAR-derived estimates, GRACE-derived groundwater variations exhibit a smaller seasonal amplitude, potentially reflecting the effects of limited spatial resolution and spatial signal smoothing. The WaterGAP Global Hydrology Model (WGHM) estimates are substantially smaller in magnitude than those derived from InSAR and GRACE, highlighting the uncertainties associated with model-based groundwater estimates in this high-altitude basin. Groundwater in the Golmud River Basin experienced a rapid increase during 2017 to 2019, which is likely driven by increased precipitation and lower temperatures associated with a La Niña event, indicating a teleconnection between global climate variability and regional water resource changes. This study provides a technical reference for refined groundwater monitoring at a tens-of-kilometers scale in small and medium-scale arid basin systems.

Remote Sensing doi: 10.3390/rs18203500

Authors: Yanting Li Jianjian Kong Nan Cong Jian Yang Yuanzheng Yang Jiali Wang Wenhua Cai Jiaxing Zu

Vegetation phenology derived from satellite vegetation indices is sensitive to spatial scale, yet how spatial aggregation affects phenological timing and temporal dynamics across broad geographical extents remains insufficiently understood. In this study, the 250 m MOD13Q1 Enhanced Vegetation Index (EVI) time series from 2000 to 2024 was used as a common data source, and 1 km and 5 km datasets were generated directly through spatial aggregation. By maintaining the same sensor, vegetation index, temporal sampling, and phenology retrieval method across scales, this controlled within-source framework minimizes confounding effects associated with cross-sensor differences. The start of the growing season (SOS) and end of the growing season (EOS) were retrieved using asymmetric Gaussian fitting combined with a dynamic threshold method. Their spatiotemporal patterns, aggregation-induced cross-scale consistency, and bias were systematically analyzed. The broad spatial patterns of multi-year mean SOS and EOS remained generally consistent across the three scales. SOS showed an overall advancing tendency and EOS an overall delaying tendency, while the proportion of pixels with relatively large trend magnitudes decreased at coarser spatial scales. Cross-scale correlations remained high across all scale pairs and generally increased with latitude, with SOS exhibiting higher cross-scale consistency than EOS. Annual SOS correlations declined during 2000–2024, whereas EOS correlations increased during the early part of the record and declined thereafter, with a breakpoint around 2010. The bias showed clear differences in direction and magnitude between SOS and EOS and among vegetation types, and its absolute magnitude generally increased with increasing scale disparity. These results indicate that aggregation-induced phenological responses are not spatially uniform but vary systematically with geographical context, vegetation type, and phenological phase. These findings provide a scientific reference for the cross-scale comparison and scale transformation of multi-source remotely sensed phenology products, as well as for studies of regional ecological processes.

Remote Sensing doi: 10.3390/rs18203499

Authors: Suharti Hajar Chang Ki Kim Hyun-Goo Kim Myeongchan Oh Chang-Yeol Yun Boyoung Kim

Accurate solar irradiance data are essential for photovoltaic (PV) resource assessment in Indonesia, yet no hourly-resolution multi-database validation exists across the archipelago scale. This study evaluates four gridded solar irradiance databases, CAMS, JAXA Himawari-9, NASA POWER, and ERA5, against quality-controlled ground measurements from 26 BMKG stations spanning all three major Köppen climate zones (Af, Am, Aw) during 2023–2024. CAMS achieved the highest hourly correspondence (median R2 = 0.855, rRMSE = 28.1%), and JAXA was close behind (R2 = 0.848), while NASA POWER showed the lowest systematic bias (rMBE = +3.6%). ERA5 showed the largest hourly errors (R2 = 0.617, rRMSE = 41.2%) consistent with its model-generated cloud fields that are unable to reproduce the localized convective-cloud variability. Because CAMS and JAXA infer cloud from the same Himawari-9 AHI observations on a common 0.05° grid, their comparison reduced sensor and resolution-related differences and focused on the retrieval design. Both showed closely comparable performance (clear-sky R2 = 0.925 vs. 0.913), and this gap widens as cloud frequency increases across the three climate zones. All four databases showed substantially larger relative errors under overcast conditions (Kt < 0.35); the temporal aggregation reduced relative error substantially, with CAMS showing the lowest error at hourly and daily scales and NASA POWER the lowest monthly rRMSE. These results provide an archipelago-scale hourly-resolution validation of four structurally distinct databases across Indonesia’s three tropical climate zones, supporting product selection and bias-correction strategies for solar resource application in Indonesia.

Remote Sensing doi: 10.3390/rs18203498

Authors: Yihong Chen He Wang Haonan Wang Jianhua Zhu Yuxin Fang Gang Zheng Jingsong Yang

Significant wave height (SWH) is a fundamental parameter for ocean engineering, maritime safety, and climate studies. Although spaceborne synthetic aperture radar (SAR) wave mode observations provide global, all-weather coverage, accurate SWH retrieval remains challenging due to the complex ocean-to-SAR imaging mechanism. This paper presents an explainable machine learning algorithm for retrieving SWH from Sentinel-1 SAR wave mode data. To address the scarcity of high-quality collocations and the black-box nature of existing deep-learning approaches, we first correct ERA5 SWH against high-quality Spotter drifting in situ using quantile mapping (QM). The bias-corrected ERA5 data are then collocated with Sentinel-1 SAR to generate a large-scale, globally distributed, and Spotter-calibrated training dataset. Here, QM maps each quantile of the ERA5 data distribution onto the corresponding quantile of the collocated Spotter distribution, thereby adjusting the statistical distribution of the reanalysis. An eXtreme gradient boosting regression model is then trained on 32 physically interpretable features extracted from Sentinel-1 Level-2 products, including image statistics, spectral parameters, azimuth cutoff wavelength, normalized radar cross section, and wind-related variables. Comprehensive temporal-independent validations against Spotter buoys and multi-mission altimeter data throughout one year of 2021 (excluded both from the QM correction and from model training) demonstrate high accuracy, with root-mean-square errors of 0.36 m (WV1) and 0.45 (WV2) for SAR–Spotter matchups over the range 0.5–10.0 m. In addition, SHapley Additive exPlanation (SHAP) analysis is employed to quantify feature contributions and enhance model transparency. Furthermore, a compact 12-feature model selected via SHAP-based hierarchical clustering retains a nearly equivalent performance while improving interpretability. The proposed approach offers a reliable and transparent solution for global SAR-based SWH monitoring.

Remote Sensing doi: 10.3390/rs18203497

Authors: Xue Zhang Yanxia Wu Dan Lin Guoyin Zhang Changting Shi

Oriented object detection is essential for interpreting high-resolution remote-sensing images, but detectors trained on a single source domain often suffer from severe performance degradation under unseen imaging conditions. Existing domain generalization methods mainly address horizontal detection or treat style variation as an undifferentiated perturbation, while the geometry-sensitive nature of oriented objects, especially elongated instances, is insufficiently considered. To address these limitations, a shape-adaptive learning and style-aware decomposition framework, termed SA-DGOD, is proposed for single-source domain-generalized oriented object detection. A frozen CLIP image encoder is employed only during training to provide multi-level style statistics, from which several interpretable branch-style pseudo-domains are constructed by modulating feature statistics that characterize photometric and contrast variations. These explicit pseudo-domain identities enable structured style diversification without accessing target-domain data. To improve orientation-sensitive representation learning, a shape-adaptive soft consistency loss is introduced for rotated RoI features. Instead of using a fixed-threshold binary weighting scheme, the proposed loss adopts an aspect-ratio-aware threshold and a continuous rotated-IoU-based weight, allowing elongated objects and geometrically reliable proposals to be handled more flexibly. Furthermore, a style-aware common-specific classifier decomposition is introduced into the proposed framework to separate domain-invariant category discrimination from pseudo-domain-specific style bias. Extensive experiments on multiple remote sensing benchmarks demonstrate the effectiveness and generalization capability of SA-DGOD.

Remote Sensing doi: 10.3390/rs18203496

Authors: Bhanwar Vishvendra Raj Singh Vivek Agarwal Stuart Marsh Gajendra Singh Sisodiya Ajay Sharma

Diverse climatic regions exhibit unique agro-climatic vulnerabilities that are significantly shaped by climate stress. Understanding these distinct vulnerabilities is essential for developing effective policies to mitigate climate change impacts and ensuring sustainable environmental management. This study examines agro-climatic vulnerabilities in Nottinghamshire County (United Kingdom) and Udaipur District (India) from 2000 to 2024, comparing climate change impacts across temperate and semi-arid agricultural systems that represent fundamentally different system behaviours under contrasting climatic conditions. Using multi-temporal Landsat imagery and WorldClim climate data, nine environmental indices were analysed: the Normalized Difference Vegetation Index (NDVI), Vegetation Condition Index (VCI), Normalized Difference Water Index (NDWI), Moisture Stress Index (MSI), Land Surface Temperature (LST), Temperature Condition Index (TCI), Vegetation Health Index (VHI), Drought Risk Index (DRI), and Crop Health Index (CHI). The results reveal contrasting regional trajectories that challenge conventional assumptions about climate vulnerability. Udaipur exhibited improved vegetation and moisture conditions across the observation period, whereas Nottinghamshire showed intensifying moisture stress and declining precipitation. Warming was more pronounced in Nottinghamshire, while strong inter-index correlations validated the monitoring approach. These findings indicate that Udaipur benefits from intensified monsoon patterns, requiring flood mitigation and water-storage infrastructure, whereas Nottinghamshire’s rising moisture stress underscores the need for sustained monitoring and adaptive management. This comparative framework provides evidence-based insights for region-specific adaptation strategies essential to sustainable agricultural development across diverse agro-climatic zones.

Remote Sensing doi: 10.3390/rs18203495

Authors: Jacob L. Bongers Wayne Brennan Erin Wilkins Lauren Roach Emily Nutman Joseph W. Lehner Amy Mosig Way

Rock shelters are important cultural and archaeological sites in Australia, often containing evidence of Aboriginal occupation and symbolic expression. While rock shelters have been investigated across Australia, locating them in rugged environments through pedestrian surveys remains challenging. This research contributes a methodology which integrates remote sensing, GIS analyses, and archaeological survey to locate rock shelters in the Blue Mountains. We produced models in ArcGIS Pro that analyse high-resolution, UAV-based LiDAR datasets and publicly available, low-resolution, aircraft-based LiDAR datasets from the New South Wales government. Our models yield locations of potential rock shelters, which can be visually evaluated in 3D LiDAR point clouds and ground-truthed through archaeological survey. Ground-truthing confirmed the successful detection of six known rock shelters on UAV and NSW LiDAR datasets, producing more accurate coordinates than the ones listed in online databases. Significantly, we identified three previously unregistered rock shelters that preserve rock art and lithics. These promising results indicate capacity to improve the spatial accuracy of legacy site records and locate unregistered archaeological rock shelters within, and potentially beyond, Australia. Our methodology promotes targeted and efficient survey of rock shelters that can produce regional datasets benefitting local communities, archaeology, and cultural heritage management.

Remote Sensing doi: 10.3390/rs18203494

Authors: Roberto Colonna Carolina Filizzola Vita Elena Di Leo Iacopo Mancusi Mariano Lisi Valerio Tramutoli

High accuracy in burned area (BA) mapping is desirable for many fire-related applications. Most operational products are based on coarse-resolution satellite data with a final spatial resolution ranging from 250 m to 1000 m. The use of higher-resolution satellite data, such as that from the Multispectral Imager (MSI) on board Sentinel-2, is usually reserved for generating BA products only for specific countries and time periods. Moreover, BA algorithms are typically driven by fire detection products to minimize false identifications. The resulting accuracy of such BA products depends strongly on land cover, fire regime, and the reliability of the considered fire detection products. In this paper, we exploit the multitemporal Robust Satellite Technique (RST) to identify burned areas and implement a workflow based on Sentinel-2 images within Google Earth Engine. Test results from two complete region-wide experiments conducted under contrasting seasonal, phenological, and agricultural conditions showed that 97.85% of the mapped area was confirmed as burned within the adopted multi-source validation framework, with an area-based commission error of 1.38% and omission error of 0.60%. Missed detections documented by the Regional Spatial Data Infrastructure (RSDI) amounted to 10.60 ha, corresponding to approximately 0.9% of the burned area reported by RSDI over the tested periods. Polygon-level intercomparison with RSDI and the European Forest Fire Information System (EFFIS) yielded Intersection over Union (IoU) values of 0.69 and 0.83, respectively. The proposed approach mapped larger burned areas than those reported by both RSDI (+48%) and EFFIS (+143%) as well.

Remote Sensing doi: 10.3390/rs18203493

Authors: Mohammad Yazdanpanah Chaoshui Xu Peter Dowd

Accurate and thorough detection of rock mass discontinuities is essential in geotechnical engineering as they have a significant impact on the stability of underground and surface excavations, yet they are difficult to automate. In the work reported here, an automatic algorithm, termed PhaseVote, is proposed for the extraction of discontinuity traces from the colour (RGB) attributes of 3D meshes which, combined with the region growing method, can help derive more accurate discontinuity modelling of a rock mass. Conventional edge detection methods degrade under low contrast and variable illumination, and geometry-based point-cloud methods overlook spectral features. The proposed approach uses contrast-invariant phase-based feature detection. The detected edges are filtered and then extended or pruned to form major discontinuity traces using the tensor voting method. The first algorithm, phase analysis, detects edges independently of local contrast, while the latter, tensor voting, links and denoises the resultant edge fragments into continuous traces using perceptual grouping. The detection capability of the proposed method is assessed by applying it to two case studies and comparing the results with an existing method. PhaseVote achieved detection rates of 94.85% and 84.72% and IoUs of 70.27% and 69.80%, compared with 76.12%/45.95% and 69.81%/43.81% for a conventional existing method. The results demonstrate that the proposed method is resilient to image contrast, robust to changes in the orientation of discontinuity planes, is less subjective and has a higher degree of automation.

Remote Sensing doi: 10.3390/rs18203492

Authors: Haoze Shi Fuling Wen Guiliang Zhong Yang Jiang Liangkun Deng Shihan Shan Xin Yang Hong Tang

Accurate and timely short-term forecasting of gridded PM2.5 concentrations is essential for air-quality management and public health protection, especially for the growing elderly population who are particularly vulnerable to air pollution. Advances in multi-source environmental data and deep learning have substantially improved the spatiotemporal prediction of PM2.5 concentrations. However, existing studies have mainly focused on developing increasingly sophisticated model architectures and optimization strategies while insufficiently considering the temporal nonstationarity of PM2.5 pollution and the recurrent characteristics associated with similar seasonal environmental regimes. Changes in natural environmental conditions and anthropogenic emissions can continuously alter the relationships governing PM2.5 concentrations, challenging the stability of continuous short-term forecasting; meanwhile, historically recurring seasonal pollution characteristics provide potentially valuable prior information for model adaptation. To address this problem, we propose a Seasonal Prior-guided Elastic Weight Consolidation (Seasonal EWC) framework for dynamic forecasting of nonstationary gridded PM2.5 concentrations in China. The framework derives season-specific parameter anchors and Fisher-based parameter importance from historical seasonal conditions and uses this prior knowledge to guide parameter updates during continuous short-term forecasting. Experiments with four representative spatiotemporal forecasting models showed that Seasonal EWC produced architecture-dependent changes in forecasting performance, reducing the annual mean MAE and RMSE by 30.3% and 31.7% for U-Net and by 10.9% and 7.6% for Earthformer, with only small improvements for IAM4VP and mixed changes for ConvLSTM. Parameter analysis revealed that Seasonal EWC reduced parameter-update magnitudes by approximately 27–32% and substantially limited long-term parameter deviation, indicating more stable adaptation during continuous forecasting. These findings demonstrate that seasonal prior knowledge can effectively guide dynamic model adaptation under evolving PM2.5 conditions, providing a promising approach for continuous short-term air-quality forecasting.

Remote Sensing doi: 10.3390/rs18203491

Authors: Bingxiu Yao Liang Huang Tianwen Zhang Longyang Han

Synthetic aperture radar (SAR) can acquire maritime imagery independently of daylight and under diverse weather conditions, which makes it well suited to persistent ship monitoring and maritime awareness. Nevertheless, recognizing SAR ships from only a few labeled examples remains difficult for several reasons. First, sea clutter and speckle noise may weaken ship-related structural and scattering-response information. Second, features from different categories may become difficult to separate because between-class differences are small while within-class variations remain substantial. Third, when each category contains only a few support examples, an abnormal or noisy sample can noticeably distort the estimated prototype. To address these issues, a few-shot SAR ship recognition method based on spatial-frequency dual-domain feature enhancement and hypersphere mapping constraint (SFHMC) is proposed. First, the spatial-frequency feature enhancement and extraction module (SFEE) combines the Haar discrete wavelet transform with multi-scale spatial modeling. It enhances structural, textural, and scattering-response features under complex backgrounds. Second, the hypersphere mapping constraint module (HMCM) normalizes support and query embeddings onto a unit hypersphere. By removing differences in feature magnitude, this mapping makes similarity estimation depend mainly on feature direction and produces a more structured class distribution. Third, the sample-weighted prototype generation module (SWPGM) adaptively adjusts the contribution of each support sample according to its consistency with the initial class prototype. This reduces prototype bias caused by noisy or atypical samples. The three modules jointly optimize feature representation, metric discrimination, and prototype calibration. Experiments are conducted on FUSAR-Ship and OpenSARShip, with additional validation on MSTAR under 3-way 1-shot and 3-way 5-shot settings. Compared with MBEN-BC, SFHMC improves classification accuracy by 1.41 and 1.96 percentage points on FUSAR-Ship, 2.68 and 4.46 percentage points on OpenSARShip, and 6.40 and 0.48 percentage points on MSTAR, respectively. These results demonstrate the effectiveness of SFHMC for few-shot SAR target recognition under limited data and complex imaging conditions.

Remote Sensing doi: 10.3390/rs18203490

Authors: Xiaoxin Liao Jianting Liu Huaitao Fan Heng Zhang Jubo Zhu

Autofocus plays a pivotal role in synthetic aperture radar (SAR) imaging, which is particularly crucial for high-resolution applications. Traditional autofocus methods face inherent limitations, such as reliance on strong isolated scatterers within the scene or prohibitive computational complexity, making them inadequate for the practical demands of the rapid and batch processing of massive SAR imagery. Although recent deep learning (DL)-based methods have improved focusing efficiency, they predominantly rely on predefined high-order polynomial models for phase error estimation, which inevitably introduces parameter redundancy, restricts the model’s adaptability to complex phase errors, and is highly prone to inducing severe oscillations at the aperture boundaries. To overcome these limitations, this article proposes DeepAutofocus, a non-parametric deep learning framework for SAR autofocus. First, we design a physics-inspired subaperture correlation Transformer architecture. A multiscale convolutional neural network (CNN) is utilized to extract local spatial features, which are then transformed into serialized subaperture image patches. Subsequently, the self-attention mechanism of the Transformer is employed to capture the long-range dependencies across subapertures. Second, we introduce a data-driven cubic B-spline reconstruction strategy that directly estimates phase errors without depending on predefined high-order polynomial models, thereby eliminating boundary oscillations while maintaining high flexibility. Extensive experiments on both simulated and real measured satellite datasets demonstrate that, compared to traditional autofocus algorithms and existing DL-based methods, the proposed DeepAutofocus achieves superior focusing performance under various complex phase error conditions.

Remote Sensing doi: 10.3390/rs18203488

Authors: Tamás Várnai Alexander Marshak Alexander Kostinski

This paper examines the way sun glints from horizontally oriented ice crystals that occur in clouds impact atmospheric data products provided by the Moderate-Resolution Imaging Spectroradiometer (MODIS). For this, the paper presents a statistical analysis of two years of MODIS data from all tropical land regions. The results indicate only weak glint effects on cloud amount and height, as these values are obtained using mainly observations of (inherently glint-free) thermal infrared radiation. In contrast, glints are found to cause a roughly 10% increase in mean retrieved cloud optical thickness, and 10% decreases in mean particle size and in the prevalence of ice cloud phase. Moreover, glints from undetected ice clouds are found to cause roughly 5% increases in mean aerosol optical depth and coarse mode fraction, and positive and negative biases comparable to the magnitude of the seasonal cycle in total column ozone and precipitable water vapor amounts, respectively. Given the modest size of glints, the effects are most important to consider in studies that use data from a limited range of viewing zenith angles (e.g., around nadir). The results are relevant to MODIS and to other satellite imagers that use similar retrieval algorithms, and thus likely experience similar glint effects.

Remote Sensing doi: 10.3390/rs18203489

Authors: Zelin Wang Zhirong Men Kangxin Xiong Pengbo Wang Bing Sun Jie Chen

High-resolution squinted spaceborne sliding spotlight synthetic aperture radar (SAR) imaging requires accurate Doppler parameters. In conventional SAR processing, Doppler parameters are generally derived from prior ancillary information or estimated from the received echoes. However, Doppler parameters derived from prior information are sensitive to ancillary-measurement errors, while existing echo-based methods remain insufficient for ambiguity-resolved Doppler-centroid estimation and Doppler frequency modulation (FM)-rate refinement in high-resolution squinted sliding spotlight SAR. To reduce the reliance on prior ancillary information for Doppler-parameter determination, this paper proposes an echo-driven Doppler-parameter estimation (ED-DPE) method. The proposed method first estimates the baseband Doppler centroid using adjacent-echo cross-correlation and then resolves the Doppler ambiguity number based on the range-migration slope, thereby obtaining the absolute Doppler centroid fd. The estimated absolute Doppler centroid is subsequently used as a physical constraint on an attitude-equivalent observation geometry, which is refined through contrast-based full-aperture image formation to obtain the Doppler FM rate fr and the imaging parameters. Point-target simulations and real spaceborne SAR experiments demonstrate that the proposed ED-DPE method supports accurate focusing of 0.25-m-resolution squinted sliding spotlight SAR.

Remote Sensing doi: 10.3390/rs18203487

Authors: Yangshen He Wenzhe Zhao Guangyao Zhou

Accurate multi-scale object detection remains challenging in remote sensing (RS) imagery due to severe scale variation, limited pixel occupancy of small targets, and complex backgrounds. Existing multi-scale fusion methods propagate features along predefined cross-scale paths and apply receptive fields fixed once trained, so fusion weights and the spatial range of contextual evidence remain largely input-independent. The Adaptive Perception Fusion Network (APFNet) targets these two aspects through four operations: (1) Split-ELAN (SP-ELAN), a channel-compressed serial refinement backbone whose design is motivated by small-object feature preservation; (2) Adaptive Sampling (AS), which produces sample-dependent channel weights from a global descriptor during cross-scale alignment; (3) Adaptive Gating (AG), which is coupled with AS and aggregates the recalibrated features at complementary receptive-field scales; and (4) Adaptive Context Aggregation (ACA), which aligns three feature levels and applies parallel large-kernel depthwise branches to supply broader multi-scale context. On DIOR, APFNet-L achieves three-run mean values of 86.39% AP50, 63.09% AP, and 26.10% APS. A complete 23 component-group study under one scratch/no-augmentation protocol yields positive conditional contrasts for SP-ELAN, ACA, and the coupled AS–AG pathway. LEVIR and RSOD provide additional cross-dataset and configuration-selection results. These operations use established primitives; the contribution lies in their routing and aggregation.

Remote Sensing doi: 10.3390/rs18203486

Authors: Qing Liu Tingting Wu Ling Han

Landslides are among the most destructive natural hazards. High-resolution remote sensing imagery provides an effective means for landslide detection and mapping, supporting geological hazard assessment and early warning. However, existing methods still face challenges in complex mountainous scenes with heterogeneous land cover and geological conditions. These challenges include irregular segmentation boundaries caused by diverse landslide shapes, small-scale detection difficulties due to background interference, and foreground–background class imbalance that hinders effective model optimization. To address these issues, we propose LD-DETR, a network for landslide detection and segmentation. An Atrous Spatial Pyramid Pooling (ASPP) module is integrated into the segmentation head to improve multi-scale feature representation, enabling better segmentation of landslides with varying spatial sizes. To alleviate class imbalance caused by the dominance of non-landslide pixels, a weighted Tversky Loss is introduced to improve the learning of hard samples. Low-Rank Adaptation (LoRA) is further combined with orientation-aware data augmentation to enable parameter-efficient adaptation, reducing the proportion of trainable parameters of LD-DETR to 41.07% while keeping the pre-trained backbone parameters frozen. Experiments on the Bijie and Luding Landslide Dataset (BLLD) demonstrate that LD-DETR is competitive with the baseline model and several YOLO-based approaches. The proposed method achieves an F1-score of 95.18% and a Recall of 95%, while the segmentation task reaches an mAP of 59.28%, slightly higher than the RF-DETR baseline. Furthermore, a limited-sample adaptation experiment on the Jiuzhai landslide dataset demonstrates the transferability of the proposed method across different geographic regions. These results indicate that the proposed framework provides competitive landslide detection and segmentation performance and offers an approach for fine-grained landslide mapping using remote sensing imagery.

Remote Sensing doi: 10.3390/rs18203485

Authors: Pengyu Wang Zhanfeng Zhao Zhida Yang Youcun Qi

Biases in high-spatiotemporal-resolution radar quantitative precipitation estimation (QPE) can substantially compromise the reliability of meteorological and hydrological disaster warning. Existing QPE correction methods commonly assume predefined spatial smoothness, whereas localized biases can vary sharply over short spatial distances. We therefore propose an implicit neural representation (INR) framework for continuous radar QPE bias-field reconstruction (QPE-BiasINR). QPE-BiasINR introduces QPE-background modulation, which conditions bias estimation on the local precipitation context to account for precipitation-dependent spatial variations in radar QPE errors. Moreover, a multiscale spatial consistency (MSC) constraint is incorporated into the loss function to explicitly strengthen spatial consistency. QPE-BiasINR was evaluated using hourly multi-radar mosaic QPE and gauge observations within the coverage of four radars in the Daqing River Basin,China, during June–September 2023. Test set performance was benchmarked against the uncorrected radar QPE, mean-field bias correction, and Kriging. QPE-BiasINR achieved the highest correlation coefficient (0.84) and the lowest root mean square error (3.29 mm) and mean absolute error (1.68 mm). Ablation experiments on QPE-BiasINR demonstrated that QPE-background modulation improved accuracy, while the MSC constraint reduced localized overcorrection. These results indicate that QPE-BiasINR can improve high-resolution radar QPE, with potential value for hydrological modeling and operational flood-risk monitoring.

Remote Sensing doi: 10.3390/rs18203484

Authors: Jingjing Wang Yanhao Liu Yuncheng Xu Wentao Wang Dongyu Cai Haijun Yan

Unmanned aerial vehicle (UAV) remote sensing is widely used for plot-scale crop yield prediction, but model performance may decline when models are transferred across fields with different crop, environmental, management, and sensing conditions. This study evaluated the cross-field transferability of UAV-based winter wheat yield models using multispectral (MS), RGB, and thermal infrared (TIR) information. Categorical Boosting (CatBoost) and multilayer perceptron (MLP) models were evaluated under five strategies: target-only training (T1), direct transfer (T2), nitrogen-based adaptation (T3), cultivar-based adaptation (T4), and mixed adaptation (T5). Direct transfer increased the normalized root mean square error (NRMSE) from 12.7% to 14.0% for CatBoost and from 12.9% to 15.9% for MLP. Mixed adaptation achieved the best overall performance, with coefficient of determination (R2) values of 0.84 and 0.87 and NRMSE values of 11.7% and 10.9% for CatBoost and MLP, respectively. After controlling for calibration sample size, the difference between T3 and T4 became smaller, indicating that adaptation performance was influenced by both calibration-data composition and sample availability. CatBoost showed diminishing returns near the 40% calibration level in this experiment, whereas MLP continued to benefit from additional target-domain samples. Prediction errors were higher for samples beyond the source-domain yield range, while mixed adaptation reduced prediction errors in both yield-range groups. SHapley Additive exPlanations (SHAP) and sensor-level ablation showed that MS features provided important predictive information, while RGB and TIR provided complementary information. These results indicate that target-domain adaptation can improve cross-field UAV yield prediction and highlight the importance of calibration design, yield-range mismatch, and multi-sensor information when transferring models to new fields.

Remote Sensing doi: 10.3390/rs18203482

Authors: Shenshen Chi Bingqian Yang Lei Wang Jiajia Yuan Qingbiao Guo Xuexiang Yu

To address the difficulty in determining underground working face geometry when mining records are incomplete or mining activities are concealed, this study proposes an InSAR-constrained method for inverting the geometric parameters of concealed working faces. The method uses the line-of-sight (LOS) deformation field derived from LuTan-1 InSAR observations as the primary input. The search ranges of mining depth, mining thickness, coal seam dip angle, strike azimuth, working face center coordinates, strike length, and dip length are constrained using the morphology of the subsidence basin, available empirical parameters, and the Probability Integral Method (PIM). These constraints reduce the uncertainty caused by a large search space and parameter coupling in high-dimensional inversion. An adaptive multi-strategy differential evolution algorithm is then introduced. Latin hypercube sampling is used for population initialization. Dynamic parameter adjustment, multiple mutation strategies, reflective boundary handling, and local refinement are incorporated to improve optimization stability in complex parameter spaces. The simulation results show that the proposed method generally outperforms particle swarm optimization, genetic algorithm, evolutionary strategy, and standard differential evolution in inversion accuracy, convergence stability, and robustness to errors. It can reliably recover the main geometric parameters of a predefined working face. In the engineering case study of the 97,315 working face, the inverted active-mining geometry during the LuTan-1 observation period is generally consistent with the corresponding mine records, with a deformation-fitting RMSE of 32.53 mm and relative errors of the main geometric parameters below 7.9%. Parameter sensitivity analysis further shows that the geometric dimensions and center location of the working face have pronounced effects on the LOS deformation field. These parameters should therefore be tightly constrained during inversion. The results demonstrate that the proposed method provides a feasible remote-sensing approach for identifying concealed working faces and estimating goaf geometric parameters in mining areas with incomplete records.

Remote Sensing doi: 10.3390/rs18203483

Authors: Zhanjie Qi Xiuping Xie Jianrong Cao Ke Zheng Wenhui Xie

Downward shortwave radiation (DSR) varies strongly across mountain slopes, and products designed for horizontal surfaces may not resolve this variability. We estimated slope-parallel DSR at eight Saihanba sites using Sentinel-2 imagery, topography, atmospheric inputs, and BRDF-informed surface albedo. Both branches used XGBoost, with the clear-sky branch further incorporating a parameterized radiation anchor, bounded residual learning, and support-aware output correction. On the 2024 test set (n = 1451), the full framework achieved an RMSE of 92.60 W/m2, an NRMSE of 15.79% and an R2 of 0.897. Pooled leave-one-site-out validation across all 6096 samples yielded an RMSE of 113.68 W/m2. For horizontal-surface observations at 10 Heihe River Basin stations, NRMSE was 17.60%. Clear-sky estimates depended most on solar-incidence geometry, whereas top-of-atmosphere spectra dominated the cloudy-sky XGBoost model.

Remote Sensing doi: 10.3390/rs18203481

Authors: Aida Galfione Elsa Cattani Giulio Monte Federico Vermi Blanka Piskala Gvoždíková Pavlos Kollias Alessandro Battaglia

Convection plays a key role in the Earth’s energy balance and is responsible for severe weather phenomena such as hailstorms, which cause significant damage to infrastructure and agriculture. This study investigates convective events over the Mediterranean Basin between September 2024 and September 2025 using a synergistic combination of active and passive satellite observations from geostationary (GEO) and low Earth orbit (LEO) platforms. Convective evolution is analyzed using EUMETSAT Satellite Application Facility on Support to Nowcasting and Very Short Range Forecasting products derived from Meteosat Rapid Scan Service imagery and the Multi-sensor Approach for Satellite Hail Advection (MASHA), which combines passive microwave hail probability retrievals from the Global Precipitation Measurement constellation with high-temporal-resolution Meteosat infrared observations to monitor hail-producing storms in near real time. The analysis is further enhanced with EarthCARE Cloud Profiling Radar reflectivity and Doppler velocity measurements, whose spatial variability provides a proxy for identifying convective cores and storm intensity. For selected events, observations from the Meteosat Third Generation Lightning Imager are also examined, highlighting the close relationship between intense lightning activity and strong convective updrafts. The results demonstrate the potential of combining complementary satellite observations to improve the detection and characterization of severe convection and hail-producing storms and identify the key roadblocks and pathways towards the effective integration of LEO and GEO observations.

Remote Sensing doi: 10.3390/rs18203480

Authors: Xiao Feng Mingcheng Feng Luigi Lombardo Jing Zhang Juan Du

Interferometric synthetic aperture radar (InSAR) provides spatially extensive deformation measurements that support dynamic landslide susceptibility assessment. However, determining when landslide susceptibility maps require updating remains insufficiently studied. Here, we introduce the persistent deformation-change area ratio (PDAR), the percentage of monitored slope-unit area exhibiting persistent velocity changes relative to each unit’s accepted reference. Three-observation confirmation and reference resetting allow update intervals to respond to both increasing and decreasing deformation. A common asymmetric matrix combines the current susceptibility class with the relative velocity change, allowing at most two upward steps or one downward step. We compared PDAR with no updating and fixed one-, three-, six- and twelve-month schedules across 6956 slope units in Yunyang, Kaizhou and Wushan. Baseline construction used 2013–2016, parameter calibration used 2017–2018, and evaluation covered 2019–2022. PDAR achieved the highest area under the receiver operating characteristic curve (ROC-AUC; 0.7913 versus 0.6848 without updating), highest high/very high event capture (63.99%) and lowest false-positive rate (FPR; 13.67%). Relative to monthly updating, AUC increased by 0.1162 and FPR decreased by 12.92 percentage points. PDAR also exceeded the strongest fixed schedule, six-month updating, by 0.0688 in AUC. The fixed-window comparison showed that more frequent updating was not always beneficial. PDAR provides a state-dependent basis for selecting update windows rather than updating whenever newer observations become available.

Remote Sensing doi: 10.3390/rs18203479

Authors: Ahmet Tarık Torun Aslı Özdarıcı-Ok

Image classification is a fundamental technique that has been broadly adopted in remote sensing, computer vision and related areas. High-quality annotated datasets are necessary for the development and reliable evaluation of classification and segmentation algorithms for accurate thematic mapping. Reference data are critical in this sense, as their quality has a direct impact on the performance, accuracy and generalization of the model. However, generating reliable reference data remains challenging due to methodological, technical, and annotation-related uncertainties. This study reviews the principal challenges in reference data generation, and the solutions proposed in scientific literature. A total of 100 papers indexed in the Web of Science were examined. The review covers traditional reference data generation methods as well as emerging approaches based on semi-supervised, self-supervised, and active learning, segmentation- and post-processing-assisted labeling, benchmark dataset construction, sampling and accuracy assessment, and deep learning and automated label generation. Major challenges and emerging research directions are also discussed. The results indicate a clear shift from traditional manual techniques to more automated, learning based and accuracy driven methods. The study provides a framework of existing techniques and future research areas in image classification.

Remote Sensing doi: 10.3390/rs18203478

Authors: Cheng Chen Letian Wang Xiaowen Liu Liang Shen Haipeng Li Bo Peng

Ground Penetrating Radar (GPR) is an important non-destructive sensing technology for subsurface detection. However, strong clutter from surface reflection and heterogeneous media often masks weak target responses in B-scan images. Existing low-rank and sparse decomposition (LRSD) methods, such as Robust Principal Component Analysis (RPCA), can separate clutter and targets but suffer from uniform singular value shrinkage and high computational cost; more critically, they operate only at the original resolution, refining only the spatial sampling grid of the migrated image under the inherent physical resolution limit. In this paper, we propose a Joint Super-Resolution and Weighted Low-Rank Sparse Decomposition (JSR-LRSD) framework that synergistically suppresses clutter and improves the spatial detail and target localization of the migrated image without extending its physical resolution. Unlike cascade approaches, which perform clutter removal first and super-resolution afterward and may amplify residual clutter, our method integrates a differentiable super-resolution network directly into the alternating direction method of multipliers (ADMM) iterations. This allows for the sparse target matrix to be regularized by high-resolution image priors, while the high-resolution estimate adaptively adjusts the weighted nuclear norm via total variation feedback. An adaptive weighted nuclear norm is designed using singular values of the observed data and the total variation feedback from the high-resolution target estimate. A randomized singular value decomposition (SVD) is introduced to accelerate computation. Extensive experiments on simulated and laboratory GPR data demonstrate that the proposed JSR-LRSD achieves a higher signal-to-clutter ratio (SCR) compared to conventional and LRSD-based methods, with clear improvement in spatial detail recovery, while maintaining computational efficiency.

Remote Sensing doi: 10.3390/rs18203477

Authors: Fuqing Cui Jianing Lv Jianbing Chen Xiaona Liu Xinyue Wang Zhiyun Liu

Permafrost subgrades along the Qinghai–Tibet Highway exhibit pronounced spatial variability in seasonal freeze–thaw deformation under the combined effects of complex permafrost conditions, climate change, and engineering disturbances. Focusing on the Xidatan–Tuojula Mountains section, this study integrated field monitoring, borehole investigations, subgrade geometric parameters, and multisource remote sensing data to establish a coupled ground–atmosphere model. The maximum refreezing depth (Hf) was selected as the susceptibility indicator, and an extreme gradient boosting (XGBoost) model based on 172 numerical simulations was developed to predict the spatial distribution of seasonal freeze–thaw deformation susceptibility at a 30 m resolution. The predicted susceptibility index was classified into five levels using the Jenks natural breaks method. The results show that susceptibility along the 532.20 km investigated section exhibits significant spatial heterogeneity. Level I sections account for the largest proportion (30.64%), followed by Level II (18.52%), Level IV (15.40%), Level V (9.54%), and Level III (7.92%). High-susceptibility sections are mainly concentrated in Hoh Xil, Wudaoliang, the Fenghuo Mountains, and the Tanggula Mountains, whereas low-susceptibility sections are primarily distributed in the Tuotuo River Basin, Tongtian River Basin, Wenquan Basin, and the Tuojula Mountains section. Furthermore, SHapley Additive exPlanations (SHAP) analysis identified mean annual ground temperature (MAGT), subgrade height (SH), and ice content type (ICT) as the dominant controlling factors, contributing 51.59%, 26.56%, and 18.16%, respectively. The interactions between MAGT × SH and MAGT × ICT were identified as the dominant two-way interactions. These findings provide insights for identifying critical sections, optimizing monitoring strategies, and supporting zonation-based maintenance of permafrost highways.

Remote Sensing doi: 10.3390/rs18203475

Authors: Depeng Gao Changping Li Yuqi Ding Lipeng Gao Bingshu Wang

Object detection in remote sensing scenarios faces multiple challenges, primarily manifested in multi-scale objects, class imbalance, densely packed small targets, and occlusion. To address these issues, we propose DCA-Net, an object detection method based on deep context aggregation. Built upon the YOLO architecture, the proposed method introduces parallel multi-branch pooling paths and an adaptive feature reweighting mechanism to construct a deep context aggregation backbone (DCA-Backbone). This enables efficient fusion and selective enhancement of multi-scale contextual semantic information. Furthermore, a detection head based on efficient channel interaction (ECI-Head) is designed to enhance the representational capability for multi-scale objects and complex morphological features. Experimental results demonstrate that DCA-Net achieves mAP50:95 of 50.5% on DIOR dataset and 61.5% on NWPU VHR-10 dataset, achieving competitive performance compared with other methods. Extensive visualization and ablation experiments further validate its detection accuracy and robustness in complex remote sensing scenarios.

Remote Sensing doi: 10.3390/rs18203476

Authors: Zhengwei Shen Jianwei Yin Xiaoyu Zhang Yongheng Shang

High-resolution remote-sensing images contain dense object layouts and fine spatial details, but their spatial extent makes direct vision–language interpretation difficult. Patch retrieval offers a practical alternative by returning local regions as candidate evidence for downstream interpretation. However, independent semantic scores can leave relevant patches at isolated ranks or return lists with substantial irrelevant content. We propose Prompt-Stable Reciprocal Inter-scale Visual Evidence Reranking (PRIVER), a post-retrieval method for ranking patches within one image using controlled object-category queries. It aggregates four predefined prompt templates and combines semantic scores with geometric support, principally from a degree-normalized inter-scale graph. No encoder weights are updated, and retrieval and reranking do not access object annotations. Parameters are selected on annotated DOTA-v1.5 development data and held fixed. With relevance defined by annotated object coverage, Precision@10 on 458 DOTA-v1.5 internal-test images increases from 0.567 to 0.625 (absolute gain 0.058; 95% source-image cluster bootstrap interval [0.052, 0.065]). Normalized discounted cumulative gain at rank 10 (nDCG@10) increases from 0.647 to 0.730 and distinct-object coverage at rank 10 from 0.773 to 0.873. On a disjoint 526-image SODA-A subset, the corresponding metrics increase from 0.414 to 0.502, 0.445 to 0.550, and 0.467 to 0.681. Coverage-based gains persist with GeoRSCLIP and RemoteCLIP and under altered tiling policies on fixed image subsets. Inter-scale support is the principal source of improvement. These results support training-free, coverage-defined patch prioritization for within-image visual evidence retrieval.

Remote Sensing doi: 10.3390/rs18203473

Authors: Weizhen Wang Leilei Dong Xingze Li Feinan Xu

High-precision inversion of soil moisture content using Synthetic Aperture Radar (SAR) data still faces significant technical challenges, as the radar backscatter coefficient is influenced by complex interactions among surface roughness, vegetation cover, and soil permittivity. Therefore, systematically evaluating the mechanisms by which SAR system parameters and surface environmental variables influence soil moisture inversion is of critical scientific importance. This study utilized an improved alpha approximation algorithm in conjunction with Sentinel-1 SAR data to estimate soil moisture content for different land cover types in middle reaches of the Heihe River Basin and analyzed the sources of error in the soil moisture estimation process. The results indicate that the improved alpha approximation algorithm can achieve relatively satisfactory soil moisture inversion results. However, significant differences in the inversion results were still observed under different orbital and incidence angle conditions, with an average RMSE of 0.034 for ascending orbits and 0.037 for descending orbits. In desert regions, data from Ascending Orbit 128 with an incidence angle of 33° yielded the lowest Root Mean Square Error (RMSE, 0.030 m3/m3) and the highest determination coefficient (R2, 0.848). In agricultural areas, the minimum RMSE for 2017 and 2018 was 0.030 m3/m3 and 0.041 m3/m3, respectively, while the maximum R2 values were 0.815 and 0.728, respectively; Compared with the desert region, the accuracy of soil moisture inversion in the agricultural region was reduced due to the influence of vegetation phenology and irrigation activities. Overall, the soil moisture inversion results for both study areas agreed well with field observations but still exhibited significant seasonal and interannual variations. The angle of incidence, surface roughness, vegetation dynamics, precipitation conditions, and irrigation activities were the primary factors affecting the accuracy of soil moisture inversion. Especially, the impact of the static roughness assumption on the accuracy of the alpha approximation algorithm is more pronounced.

Remote Sensing doi: 10.3390/rs18203474

Authors: Bo Jiang Hanfei Yang Bensheng Huang Chenjuan Li

Coastal waters exhibit pronounced spatial and temporal variability in optical properties and water quality, which cannot be fully characterized by observations from discrete monitoring stations. Satellite Optical Water Type (OWT) classification provides spatially continuous information on optical conditions. However, comparisons with water quality observations have generally relied on the assigned OWT, and it remains unclear whether membership provides additional information within the same class. This study examined the relationships of assigned OWTs and membership with monthly water quality conditions in Guangdong coastal waters. Sentinel-3 OLCI observations acquired from 2017 to 2024 were matched with station observations of dissolved inorganic nitrogen (DIN), dissolved inorganic phosphorus (DIP), chemical oxygen demand (COD), dissolved oxygen (DO), and eutrophication classes. The analysis first compared water quality among assigned OWTs and then evaluated the secondary OWT and membership values within OWT 4a, the class with the largest sample size. Assigned OWTs distinguished broad differences in DIN, DIP, COD, and eutrophication classes, whereas DO did not differ significantly among OWTs. Within OWT 4a, the class associated with the second highest membership separated records with different nutrient, COD, and eutrophication conditions. Continuous membership values provided further information, but their relationships with water quality depended on the secondary OWT and were most consistent for DIN. OWT occurrence and station eutrophication patterns showed partial spatial correspondence, particularly around the Pearl River Estuary and Huangmao Bay. These findings show that assigned OWTs identify broad water quality differences, while secondary OWTs and membership values distinguish additional variation within OWT 4a. The mapped optical patterns provide complementary spatial context for interpreting station-based water quality observations.

Remote Sensing doi: 10.3390/rs18203472

Authors: Valentina Raimondi Roberto Carlà Stefano Baronti

Soil moisture (SM) assessment underpins applications ranging from climatological studies to land management and emergency response. Microwave techniques already support operational SM products, but at a spatial resolution limited to 1 km. Optical multispectral indices have been proposed as a complementary source of information at finer resolutions (10–100 m), yet vegetation heterogeneity has generally been handled by masking or stratifying the land cover rather than by modeling its effect. Here, we compare three indices (the Normalized Difference Infrared Index, NDII; the Normalized Multi-band Drought Index, NMDI; and the Visible and Shortwave Infrared Drought Index, VSDI) over a heterogeneous agricultural area near Grosseto, Tuscany, using 31 Landsat-8 OLI images acquired over four summers. Evaluated against the Copernicus Soil Water Index (SWI), VSDI performed best both when averaged over the test area and when analyzed at the pixel level (1 km) and was therefore selected for further analysis. A linear mixed-effects model with pixel-dependent intercepts resolves a within-pixel VSDI–SWI repeated measures correlation of 0.714, against 0.411 for the pooled regression. Allowing the slope to vary as well shows that vegetation density governs both the baseline level of VSDI and its sensitivity to soil moisture, with the within-pixel correlation falling from 0.84 on bare soil to 0.30 under a dense canopy.

Remote Sensing doi: 10.3390/rs18203471

Authors: Zheng Zhou Jintao Lai Xiaofei Yang Zhong Cao Wenbiao Yang Wenli Shang

Coded aperture snapshot spectral compressive imaging (CASSI) enables the recovery of hyperspectral images (HSIs) from a single two-dimensional compressive measurement, serving as a core paradigm in modern computational imaging. However, HSI reconstruction from under-sampled measurements is a severely ill-posed inverse problem. Existing deep unfolding methods typically treat each iterative stage independently and model spatial–spectral features in a decoupled manner, which neglects intrinsic spatial–spectral coupling and lacks persistent cross-stage information propagation. To tackle these issues, we propose a deep dynamic 3D memory-augmented unfolding network (3DMAUNet) for HSI reconstruction. Based on the momentum-accelerated generalized alternating projection framework, 3DMAUNet retains the strict mathematical interpretability of iterative optimization while integrating data-driven learnable components. Specifically, we design two key modules: (1) a 3D memory-augmented learning module equipped with a persistent memory bank, which explicitly encodes cross-layer spatial–spectral correlations to enable efficient inter-stage information transfer and long-range dependency modeling; and (2) a learnable dynamic fast momentum updating module, which fuses cross-stage spatial–spectral attention features to estimate task-adaptive momentum coefficients, accelerating convergence and enhancing the robustness of feature updating. Comprehensive experiments on synthetic and real-world HSI datasets demonstrate that 3DMAUNet outperforms state-of-the-art methods in both quantitative and qualitative reconstruction performance.

Remote Sensing doi: 10.3390/rs18203470

Authors: Zigeng Song Shuting Wu Wang Man Shujie Yu Yingxue Zeng Qin Nie Zongmei Li Xiaofeng Du Xinchang Zhang

Atmospheric nitrogen dioxide (NO2) is one of the major pollutants affecting environmental quality and human health, yet accurate surface NO2 (NO2Surface) estimation remains challenging due to complex nonlinear relationships between satellite-derived NO2 column density (NO2column) and near-surface concentrations. Here, we proposed an explainable transformer-based framework for daily NO2Surface estimation across Fujian Province, integrating Sentinel-5P TROPOspheric Monitoring Instrument (TROPOMI) NO2 observations with meteorological data, digital elevation model (DEM), normalized difference vegetation index (NDVI), nighttime light (NTL), and population density (PD). Using ground-based observations, Tabular Prior-data Fitted Network (TabPFN) was trained and benchmarked against five conventional machine learning (ML) algorithms. Under random five-fold cross-validation, TabPFN achieved the highest accuracy, yielding a coefficient of determination (R2) of 0.93, root mean square error (RMSE) of 2.56 μg m−3, mean absolute error (MAE) of 1.05 μg m−3, and mean absolute percentage error (MAPE) of 5.82%, with R2 improvements of 0.17–0.26 and RMSE reductions of 45.3–53.5% relative to CatBoost and RF, respectively. Additional station-grouped spatial validation and independent temporal validation yielded R2 values of 0.86 and 0.79, respectively. The reconstructed NO2Surface distributions revealed a clear coastal-to-inland gradient, with concentrations frequently exceeding 50 μg m−3 in urbanized coastal areas but generally below 15 μg m−3 in northwestern mountainous regions. Strong seasonality was evident, with the peak NO2Surface and widespread pollution in winter, contrasting with lower and more localized levels in summer. Temporal analyses revealed persistent declines in NO2Surface, with annual mean concentrations decreasing by 11.7% from 2019 to 2020 and a further decline of 1.6% in 2021, while descriptive daily slopes were approximately −0.003 μg m−3 day−1. SHapley Additive exPlanations (SHAP) analysis highlighted TROPOMI NO2column and BLH-corrected NO2 (NO2BLH) as the most influential predictors, accounting for 15.07% and 11.36% of the contribution, respectively, with temperature- and humidity-related variables exhibiting strong nonlinear responses. This study provides an accurate and interpretable framework for 1 km gridded NO2Surface estimation and characterizes the predictive relationships associated with regional NO2 variability.

Remote Sensing doi: 10.3390/rs18203469

Authors: Shiyao Guo Cengiz Ozel Yang Qu

The rapid proliferation of high-resolution remote sensing video (RSV) has made object segmentation an increasingly critical yet challenging task. Existing methods face two fundamental limitations in this domain: (1) the intrinsic complexity of RSVs arising from diminutive object sizes, dense distributions, and drastic scale variations; (2) prohibitive computational overhead and latency caused by the massive spatial resolution of remote sensing imagery. To address these challenges, we propose Trajectory-SAM, a novel object trajectory-guided segmentation framework tailored for rapid and accurate object segmentation in RSVs. Specifically, we first employ a Graph Neural Network to extract representative keyframes, improving data utilization efficiency while suppressing attention drift and interpretation bias induced by redundant frames. We then introduce Temporal-LoRA, a parameter-efficient fine-tuning strategy for the Segment Anything Model (SAM), substantially enhancing its spatiotemporal contextual understanding in remote sensing domains. Finally, we devise a sample-augmentation-based knowledge distillation strategy to accelerate inference without sacrificing segmentation quality. Extensive experiments on the SAT-MTB, iSAID, and VHR-10 datasets demonstrate that our proposed Trajectory-SAM achieves superior performance, establishing a compelling trade-off between accuracy and speed.

Remote Sensing doi: 10.3390/rs18203465

Authors: Liyang Yin Xinran Xu Xinlei Huang Fengyan Wang Mingchang Wang Qing Ding

Oilfield subsidence threatens infrastructure and sustainable development. Although facility detection, factor-association analysis, and deformation forecasting have each advanced considerably, their systematic integration remains insufficiently explored for oilfields in agricultural plains, where C-band SAR decorrelation over cropland and wetland yields sparse and uneven deformation observations. This constraint complicates facility extraction, quantitative analysis of associations among interacting factors, and robust temporal prediction. To address these challenges, this study develops and applies an integrated modular workflow for the monitoring, analysis and prediction of surface subsidence in a typical oil production area of Songyuan City, China. We first used PS-InSAR for high-precision deformation monitoring and validated the results with leveling survey data. Subsequently, the YOLOv12 model was utilized for automated oil well identification, and the Geodetector model was applied to quantify the spatial associations between subsidence and the examined natural and anthropogenic factors. Finally, a hybrid LSTM–Transformer model was developed for time-series prediction and applied to forecast deformation through November 2025, twelve months beyond the final SAR acquisition. Results indicate significant spatial differentiation in deformation, with a maximum subsidence rate of −26.0 mm/y. Analysis identifies oil well density and surface elevation as the leading spatial factors among those examined, and the buffer zone analysis indicates a distance-related association of approximately 700 m between oil wells and subsidence rates. The hybrid model demonstrated superior overall accuracy compared to standalone LSTM and Transformer models. The findings provide concrete scientific support for risk mitigation in oil cities.