Remote Sensing doi: 10.3390/rs18213809

Authors: Lucas Vituri Santarosa Andrés Hirigoyen

Drought affects meteorological forcing, root-zone water availability and land-surface functioning at different spatial and temporal supports. We evaluated Uruguay’s 2020–2023 drought using an observational 2016–2024 panel of 1423 SMAP parent cells. Monthly SPEI-3, SPEI-6 and SPEI-12 represented short, intermediate and persistent climatic water-balance deficits. SMAP root-zone soil moisture (RZSM) was retained at 9 km support, whereas actual evapotranspiration (ETa) and the moisture stress index (MSI) were summarized from approximately 500 m pixels using annual MapBiomas fractions for grassland, cropland and silviculture within each parent cell. For each cell and 8-day composite position, responses were expressed relative to the corresponding 2016–2019 mean. An interval was classified as locally dry when its overlap-day-weighted SPEI-12 was below −1. During 2020–2023, RZSM responses ranged from −13.09% to −16.63% among dominant landscape domains. Fraction-weighted ETa responses were −8.69% for grassland, −6.75% for cropland and −7.39% for silviculture, while oriented MSI responses were −6.62%, −3.32% and −11.20%, respectively. At lag zero, monthly RZSM correlated the most strongly with SPEI-3 (r = 0.81), followed by SPEI-6 (r = 0.64) and SPEI-12 (r = 0.51). At native 8-day support, RZSM was positively aligned with ETa (r = 0.42–0.57) and the oriented MSI (r = 0.31–0.59), although prewhitening weakened some associations. Overall, the 2020–2023 drought was consistently expressed across meteorological, root-zone and land-surface indicators. The strong association between SPEI-3 and RZSM, together with consistent land-surface departures and product-dependent ETa variation among components, highlights the value of multiscale monitoring for characterizing drought intensity and ecosystem responses across Uruguay.

Remote Sensing doi: 10.3390/rs18213808

Authors: Xu He Shaohua Gong Gaopeng Lu Xuewu Cheng Guotao Yang Hailiang Huang Faquan Li Jing Jiao Zhenquan Yin

The characteristics and formation mechanisms of lightning activity during the lifetime of Super Typhoon Doksuri (2023) were investigated using ERA5 reanalysis data and observation data from the World-Wide Lightning Location Network (WWLLN) and Himawari-8/9 satellite. The lightning activity exhibited pronounced intermittent outbreak behavior, and intense lightning flashes mainly occurred in the outer rainband region approximately 200–600 km from the typhoon center. The occurrence of intense lightning was mainly related to the development of deep convections which were characterized by a high CAPE index and low cloud-top brightness temperature. In the landfall stage, intense lightning outbreaks may have been further modulated by the orographic lifting from the underlying surface and island thermal effects. Strong updrafts within the typhoon provided favorable conditions for intracloud microphysical electrification processes. The formation of intense lightning was interpreted via the non-inductive charging mechanism and validated using E-WRF model simulations. During the first rapid-intensification episode, intense inner-core lightning preceded the typhoon intensity peak by 14–41 h; during the re-intensification episode, it occurred ~9 h after the typhoon peak. These insights into the intense lightning activity in Super Typhoon Doksuri (2023) contribute to improving typhoon forecasting and disaster warning in the Western Pacific.

Remote Sensing doi: 10.3390/rs18213806

Authors: Xin Wang Bing Feng Peng Zeng Ting Sun Tianbin Li

Landslides are among the most serious geohazards worldwide, and accurately identifying their locations is a prerequisite for effective risk prevention and mitigation. However, regional-scale screening of candidate active-slope units from InSAR observations is still hindered by subjective manual interpretation, inconsistent deformation-rate thresholds, and insufficient topographic constraints. To address these limitations, this study proposes a largely automated terrain-constrained framework for candidate active-slope screening by integrating InSAR observations, slope units, and screening based on slope-unit bootstrap intervals and a regional two-standard-deviation background envelope. First, surface deformation rates are derived from InSAR measurements and post-processed to reduce the influence of identifiable artifacts before slope-unit analysis. Then, a clustering-based method is introduced to delineate topographically constrained slope units in alpine canyon regions, whose visual terrain correspondence is examined qualitatively using terrain information and optical imagery. Next, rather than applying a fixed site-specific deformation-rate cutoff in mm/yr, data-dependent bootstrap confidence-interval comparisons are used to determine whether the deformation interval of each slope unit extends beyond the regional deformation-background interval, thereby enabling rule-based screening of candidate active-slope units. This confidence-interval screening rule provides a quantitative criterion for regional screening but requires further calibration across additional geomorphic and deformation settings. Finally, the proposed method is applied to two landslide-prone regions in western China, located in sections of the Upper Yellow River and Upper Jinsha River, where final inventories containing 92 and 25 candidate active-slope units are produced, respectively. These polygons are geomorphologically constrained candidate active-slope units produced for regional screening and should not be interpreted as the exact boundaries of individual landslides. Inventory-based spatial comparison and targeted optical inspection were reported as complementary evidence streams rather than as independent validation of contemporaneous slope activity. These results document implementation in two related alpine river-valley settings in western China and indicate the framework’s potential as a quantitatively defined, topographically constrained approach for largely automated regional screening of candidate active-slope units in comparable settings. These two applications do not establish broad applicability across substantially different environments.

Remote Sensing doi: 10.3390/rs18213807

Authors: Zhiyu Zhang Bo Wu Qi Huang

High-resolution bathymetry is critical for nearshore reefs, navigation, and island engineering. This paper proposes an Enhanced Mixture Density Network (EMDN) using dual-stream feature disentanglement and mixed total variance regularization to improve both accuracy and reliability of satellite-derived bathymetry (SDB). Validated with ICESat-2 and Sentinel-2 data over island reefs in the South China Sea and Hawaii, EMDN achieves RMSEs of 1.20 m, 1.24 m, and 3.71 m at Ganquan, Guangjin, and West Molokai, with the lowest mean uncertainties (0.76 m, 0.75 m, 2.46 m). Comparisons with the evaluated baselines demonstrate improved bathymetric accuracy and lower CRPS values, together with sharper prediction intervals and competitive calibration, highlighting the potential of EMDN for nearshore reef bathymetric mapping.

Remote Sensing doi: 10.3390/rs18213805

Authors: Tong Wang Yang Nan Bofeng Guo Xiao Yan

Spaceborne Global Navigation Satellite System reflectometry (GNSS-R) sea surface wind speed retrieval commonly relies on the normalized bistatic radar cross section (NBRCS), whose response depends on observation geometry, link conditions, signal quality, and sea state. Under low-wind, high-swell conditions, local wind speed alone cannot fully explain variations in reflected signals. We use SMAP-R hybrid compact polarimetric observations and ERA5 wind and wave data to assess whether polarimetric features improve sea surface wind speed retrieval. A dataset of 2,067,971 observations is divided by complete months into training, validation, calibration, and test sets. With total significant wave height included in every baseline, paired non-polarimetric and polarimetric configurations are evaluated on identical samples using a geophysical model function with minimum-variance fusion (GMF/MV), extreme gradient boosting (XGBoost), and a fully connected network trained by back-propagation (BP-FCN). Adding the degree of polarization and normalized third Stokes component reduces the overall root mean square error (RMSE) by 7.83%, 5.49%, and 5.31%, respectively, with the polarimetric BP-FCN configuration achieving the lowest RMSE of 1.666 m s−1. Wind speed-stratified analysis shows that the relative error reduction from polarimetric information varies non-monotonically across wind speed intervals. Across all signal-to-noise ratio (SNR) intervals represented in this study, all three methods achieve positive polarimetric gains, with paired month-block bootstrap 95% intervals lying above zero, indicating benefits across different SNR conditions. Under low-wind, high-swell conditions, RMSE decreases by 10.58%, 11.26%, and 9.82%, respectively, exceeding the overall relative gains. Error decomposition indicates that these improvements mainly reflect reduced systematic overestimation. Comparisons against ASCAT scatterometer and NDBC buoy winds using all collocated samples also show positive polarimetric gains of 4.83–5.35% and 4.80–5.78%, respectively. These results demonstrate that polarimetric observations can further improve GNSS-R wind speed retrieval accuracy beyond the information already included in the baselines, with more pronounced gains under the low-wind, high-swell conditions sampled in this study, providing useful complementary information for sea surface wind speed retrieval under complex sea states.

Remote Sensing doi: 10.3390/rs18213804

Authors: Xiaoyong Du Shudao Zhou Yudi Liu Zheng Sheng Min Wang Yanzhen Kang Wei He Qixiang Liao Qiusheng Zheng

Understanding and estimating the impact of atmospheric turbulence on the growth of precipitation particles in freezing weather can help improve the prediction of freezing rain precipitation. In this study, the turbulent kinetic energy dissipation rate (ε) during a freezing weather event was retrieved using the power spectral analysis method, based on observational data from a Ka-band millimeter-wave cloud radar, radiosondes, and ground automatic weather stations, and the influence of atmospheric turbulence on the growth of freezing rain precipitation particles was investigated using nonlinear causal methods such as transfer entropy (TE), convergent cross-mapping (CCM), the phase-randomized surrogate test, and partial CCM. Causality tests confirmed a significant bidirectional causal relationship between ε and Ze, with the information transfer from ε to Ze systematically dominating throughout the entire height range of precipitation particles. The bidirectional original ρ was far higher than the 99% quantile of the surrogate distribution, and the coupling was enhanced after removing the influence of temperature and humidity, indicating that it was independent of the thermodynamic covariant background. The causal intensity in the ice crystal growth zone above 6 km was approximately 3 times the average in the liquid zone, indicating that atmospheric turbulence is one of the key factors driving the growth of precipitation particles in freezing weather and that the effect is concentrated in the dominant layer of ice crystals. This finding provides a practical basis for improving the parameterization scheme and numerical prediction of the precipitation particle microphysical processes in freezing weather.

Remote Sensing doi: 10.3390/rs18213803

Authors: Jie Xin Yufei Yang Yun Xu Bing Wang Dongxia Wang Li Peng

In response to the application limitations of single-source navigation systems for their susceptibility to interference, insufficient positioning accuracy and limited scenario adaptability in complex environments, as well as the increasingly complex demands for high-precision, high-robustness, all-weather and all-scenario navigation services in fields such as the low-altitude economy, intelligent driving and specialized operations, multi-source fusion navigation technology is accelerating its evolution. This manuscript systematically reviews the development trajectory from reliance on a single navigation system to multi-source intelligent integrated navigation, dividing it into the following four evolutionary stages: single-system dominance, loose-tight coupling, deep integration and intelligent ubiquitous navigation stage. It focuses on a comparative analysis of the core characteristics, fusion modes, key technological breakthroughs, and practical limitations and bottlenecks of multi-source fusion technologies at each stage, with the aim of providing comprehensive reference for technological research and development, solution design and industrial implementation in related fields.

Remote Sensing doi: 10.3390/rs18213798

Authors: Yuan Xue Guojin He Ranyu Yin Guizhou Wang Zhaoming Zhang Yan Peng Tengfei Long Huichan Liu

Differences in cropland definitions, spatial resolution, and mapping strategies complicate product selection. We harmonized and evaluated eight 2020 cropland and land-cover products across mainland China using hierarchically labeled probability samples. Fixed map predictions enabled comparisons across common task-specific and product-definition-matched reference labels. Evaluation covered spatial agreement, design-weighted classification accuracy, design-based area bias, cropland-type responses, and cross-scale performance. Only 26.8% of pixels in the cropland union were classified as cropland by all products. Within the consensus-cropland evaluation domain, CropLayer achieved the highest overall accuracy (0.961) and F1 score (0.844). CCropLand30 led 30 m products in F1, whereas the Esri land-cover product (ESRI-LC) had national area bias closest to zero (−2.26%). Most 30 m products showed commission-dominated errors and positive area bias. Product responses varied across cropland types, and reference labels altered performance estimates. Switching from unified-agriculture to product-definition-matched labels reduced ESRI-LC’s F1 from 0.765 to 0.682 and shifted area bias from −8.08% to +14.88%. Spatial aggregation reduced F1 for some fine-resolution products without consistently changing local cropland-proportion error. These results show that classification accuracy and area bias yield different rankings, supporting product selection according to application, cropland definition, landscape context, evaluation metric, and spatial support.

Remote Sensing doi: 10.3390/rs18213801

Authors: Beibei Song Yunfei Chen Yuedong Tan Wenfang Sun

Hyperspectral imaging (HSI) is an important imaging modality in the field of optical remote sensing, capable of simultaneously capturing spatial structures and continuous spectral responses related to material properties. Therefore, compared with conventional red-green-blue (RGB) images, it can provide richer representations for understanding dynamic scenes. However, most existing video object segmentation methods have been developed for RGB videos, while high-quality hyperspectral video datasets with frame-by-frame pixel-level annotations remain scarce. To address this gap, this study constructed a hyperspectral video object segmentation dataset, comprising hyperspectral videos and their corresponding three-channel false-color RGB videos derived from the same HSI data, with pixel-level annotations provided for each frame. The dataset contains three spectral subsets: visible (VIS), near-infrared (NIR), and red-edge near-infrared (RedNIR). Using this dataset, we developed SSAM-SAM2, a hyperspectral video object segmentation network built on Segment Anything Model 2 (SAM 2). The network includes a spectral–spatial interaction enhancement (SSIE) module designed to identify informative spectral bands, suppress interference from redundant and noisy spectral channels, and integrate hyperspectral cues with the spatial-structure features of the corresponding false-color RGB images. In addition, the network introduces an Adaptive Correlation Memory Selection (ACMS) module that regulates memory sampling and selection according to variations in target state and inter-frame feature correlation. This design reduces the influence of redundant or unreliable historical information and improves temporal memory reliability under target deformation, motion variations, and background interference. Extensive experiments show that SSAM-SAM2 outperforms the selected comparison methods on the VIS, NIR, and RedNIR subsets under the evaluated settings. These results demonstrate the value of the proposed dataset and indicate that spectral–spatial enhancement and temporal adaptive memory mechanisms contribute to improving hyperspectral video object segmentation performance.

Remote Sensing doi: 10.3390/rs18213802

Authors: Emma Imen Turki Md Saiful Islam Carlos Lopez Solano Mayowa Basit Abdulsalam Amelie Arias

Coastal water levels during storms result from interacting tides, storm surges, waves, currents, bathymetry, and coastal geometry, producing spatially heterogeneous responses that remain difficult to characterize from point observations alone. The Surface Water and Ocean Topography (SWOT) mission provides a new opportunity to investigate this complexity through wide-swath High-Resolution (HR) observations of sea surface height (SSH). Here, SWOT HR water levels are first evaluated against five tide gauges in the English Channel and subsequently analyzed at three contrasting sites, Cherbourg, Villers-sur-Mer, and Étretat, using complementary spatial and scale-resolved approaches. SWOT HR shows strong agreement with tide gauges at the best-performing sites, with RMSE values of approximately 0.11 m and correlations exceeding 0.99, while accuracy remains site-dependent. Beyond point-based validation, the two-dimensional HR fields reveal pronounced spatial heterogeneity, including localized and elongated SSH anomalies with distinct geometries and orientations. EOF decomposition shows that more than half of the variance remains beyond the first six modes at all three sites, highlighting the complexity of the HR spatial signal. The detrended SSH anomalies show organized spatial patterns, while their point-by-point linear correlations with radar backscatter (σ0) remain close to zero. Across the three investigated coastal domains and two storm events, the two-dimensional wavelet analysis identifies enhanced SSH variability within an energetic scale range of approximately 0.5–1.0 km, although its intensity, spatial distribution, and organization differ substantially among sites. Comparison with independent wave, surface-current, and bathymetric fields further places these observations within contrasting hydrodynamic and morphological environments, suggesting that the observed sub-kilometer SSH variability may be associated with local coastal processes, although their individual contributions remain unresolved. . Together, these analyses demonstrate the added value of SWOT HR for resolving the two-dimensional, multiscale complexity of storm-time coastal SSH, providing new spatial information for investigating coastal dynamics and supporting high-resolution model validation and data assimilation.

Remote Sensing doi: 10.3390/rs18213800

Authors: Hasan Aksoy Joel Kostensalo

Aboveground biomass (AGB) is considered a key variable in surveilling the sustainability of forest management and the carbon cycle. This study examined the effects of image acquisition season and topographic variables on the performance of AGB estimates using Sentinel-2 imagery from three seasons. The study was conducted in 224 sample plots in pure Calabrian pine (Pinus brutia) stands in southern Türkiye. Random forest models were trained on 26 datasets derived from rainy, arid, and semi-arid season imagery, including single-season, two-season, and three-season combinations with and without topographic variables. Models trained using satellite images from rainy and semi-arid seasons performed better than models trained using data from the more arid season. All two-season combinations outperformed the corresponding single-season models, but none reached the accuracy of the three-season model. Including topographic variables reduced RMSE for every season and season combination. The best results were obtained using all three seasons together with topographic variables (R2 = 0.74, RMSE = 13.97 Mg ha−1), and removing topography from this model increased RMSE to 15.69 Mg ha−1. According to the SHAP analysis, elevation was the second-most influential predictor after Aug_SAVI (mean SHAP value = 1.65 Mg ha−1) and the only topographic variable that notably contributed to predictive performance. These findings indicate that, for the pure Calabrian pine stands and conditions examined in this study, combining multi-seasonal Sentinel-2 imagery with topographic variables improves AGB estimation accuracy compared with single-season models and spectral data alone.

Remote Sensing doi: 10.3390/rs18213799

Authors: Yuanjie Bai Haiquan Lu Xupei Zhang Lulu Liu Sheng Zhong

Small ship detection in synthetic aperture radar (SAR) imagery is limited by feature loss at coarse strides, sea-clutter false alarms, and the sensitivity of small boxes to annotation offsets. We propose CaSA-DETR, a detector built on RT-DETRv2 with three components. CFAR-Attention is a CFAR-inspired local contrast normalisation that reweights the stride-8 and stride-16 features bypassing the Transformer encoder, using sliding-window background statistics. Scatter-Aware Context Aggregation (SACA) collects context at four dilation rates with parallel dilated depthwise-separable convolutions and branch-adaptive gating. Inner-GIoU computes the regression loss on the inner 49% of each box, intended to reduce the influence of boundary offsets on targets smaller than 32 by 32 pixels. On a benchmark of 44,217 training images (73,407 instances, all under 32 by 32 pixels), the full model with a ResNet-101-vd backbone at 1024 by 1024 input reaches an average precision (AP) of 60.75% and an F1 score of 90.10%, 3.96 AP above D-FINE at the same input size. A cross of backbone, input size and modules shows that most of the gain over the original RT-DETRv2 baseline comes from the higher input, and that the three modules add 1.0 to 1.4 AP at every backbone and input size tested. On 2244 ship-free images, the modules cut false alarms by about 30% at a confidence of 0.5, at a cost of 0.66% more parameters and 2.3% more floating-point operations (FLOPs). The three modules improve AP at low cost.

Remote Sensing doi: 10.3390/rs18213797

Authors: Mustafa Al Marzooq Alina Almukhametova Konstantin Tertyshnikov Olivia Collet Boris Gurevich Pavel Shashkin Roman Isaenkov Nikita Beloborodov Mikhail Vorobev Roman Pevzner

Safe and reliable carbon capture and storage (CCS) requires long-term monitoring to verify secure CO2 containment. Distributed acoustic sensing vertical seismic profiling (DAS-VSP) provides a practical approach for subsurface monitoring due to its permanent installation, dense spatial sampling, and reduced operational costs. However, monitoring programs spanning decades may require hardware upgrades, including interrogator replacements and changes to fibre-optic cable configurations, affecting time-lapse repeatability and interpretability. This field case study evaluates whether the CO2-induced 4D DAS-VSP response remains interpretable across a combined change in interrogator, fibre type and optical path, whose individual contributions cannot be isolated. The assessment uses data from the CO2CRC Otway Project Stage 4, where approximately 10 kilotons of CO2-rich reservoir gas were injected into a saline aquifer at approximately 1500 m depth. Baseline and monitor surveys using different DAS configurations are compared at two monitoring wells, with a same-configuration case as a repeatability benchmark. An established 4D DAS-VSP processing workflow, supplemented by system-specific noise attenuation, is applied to mitigate configuration-dependent differences. The results show that, despite higher non-repeatability, the principal CO2-related time-lapse anomaly remains spatially localised and distinguishable, indicating that the DAS-VSP response can remain interpretable across a combined configuration change under favourable acquisition and signal-quality conditions.

Remote Sensing doi: 10.3390/rs18213796

Authors: Mengqi Jiang Liangliang Tao

Methane is a major greenhouse gas that plays a critical role in climate change. Accurate monitoring of methane at the regional scale is essential for understanding its spatiotemporal dynamics and underlying processes. Here, a machine learning-based framework is developed to retrieve column-averaged methane (XCH4) by integrating shortwave infrared spectral information from TROPOMI. Spectral channels are first analyzed through machine learning-based feature importance analysis, and normalized spectral features are subsequently constructed. These features are then combined with auxiliary variables, including meteorological conditions, observation geometry, and surface environmental factors, to develop an XCH4 retrieval model. The GOSAT L2 XCH4 product is introduced as a high-accuracy reference observation for correcting the retrieved XCH4 results, and independent validation is conducted using observations from the TCCON site. Among the evaluated models, XGBoost demonstrates the best performance, achieving an R2 of 0.91 and an RMSE of 9.28 ppb under five-fold cross-validation. The model also maintains relatively better predictive performance compared with other models under both temporal and spatial extrapolation scenarios. Compared with the TROPOMI L2 product, the retrieved XCH4 shows improved agreement with TCCON observations, with a higher correlation (R = 0.84) and reduced errors. Based on this approach, a daily, spatially continuous XCH4 dataset over China for 2021 is generated. Analysis of the retrieved dataset reveals pronounced spatiotemporal patterns, characterized by higher XCH4 concentrations in eastern China and lower values in the western regions, as well as a distinct seasonal cycle.

Remote Sensing doi: 10.3390/rs18213795

Authors: Ying Liu Dianfan Guo Pengfei Li Yuqi Sun Yuhong Zhang

Long-term monitoring of wetland degradation, recovery, and spatial resilience requires methods that can jointly exploit multi-temporal remote sensing observations and spatially heterogeneous environmental information. This study develops an intelligent remote sensing framework for monitoring the disturbance–recovery dynamics and spatial resilience of typical wetland landscapes in Northeast China by coupling Landsat-based Detection of Trends in Disturbance and Recovery (LandTrendr) with geographically weighted random forest (GWRF). Annual wetland landscape state sequences for 2000–2022 are constructed from multi-temporal Landsat observations and associated geospatial variables. LandTrendr time-series segmentation is used to identify disturbance timing, disturbance magnitude, recovery rate, and recovery duration, thereby converting long-term wetland change trajectories into quantitative indicators of degradation and recovery. At the GIS grid scale, remotely sensed vegetation and wetness indicators are integrated with wetland patch structure, hydrological conditions, topography, land-cover change, and anthropogenic pressure. GWRF then models local nonlinear relationships under spatial-distance weighting to generate a spatially explicit wetland resilience index and reveal geographically varying dominant controls. The LandTrendr–GWRF model achieved an R2 of 0.872, an RMSE of 0.057, and an MAE of 0.039, outperforming RF, GWR, XGBoost, and standalone GWRF benchmarks. The mean wetland resilience index was 0.641, with low-, medium-, and high-resilience areas accounting for 27.86%, 44.37%, and 27.77% of the study area, respectively. Significant spatial clustering was detected (Moran’s I = 0.421, Z = 6.92, p = 0.001). Hydrological regulation, landscape connectivity, fragmentation constraint, and anthropogenic disturbance pressure emerged as the four dominant spatial response types. High-resilience cells formed contiguous clusters where hydrological recharge and landscape connectivity were strong, while low-resilience cells clustered where fragmentation and construction pressure were concentrated. Disturbance magnitude and recovery duration differed markedly among the four response types, with connectivity-dominated cells returning to the pre-disturbance state within 4.4 years and anthropogenic pressure-dominated cells requiring 11.7 years. The proposed framework extends wetland remote sensing from static condition mapping toward process-oriented monitoring of degradation, recovery, connectivity, and resilience, providing a practical approach for long-term wetland restoration assessment and spatial management.

Remote Sensing doi: 10.3390/rs18213794

Authors: Mingyue Gao Shengbo Chen Xitong Xu

Recent blind hyperspectral image unmixing methods have achieved effective spectral–spatial representation, yet the reliability of estimated abundances remains difficult to assess because errors in jointly estimated endmembers and abundances can compensate for each other. To address this issue, we propose Residual Diagnosis-Guided Abundance Correction (RDAC), a post-correction framework that preserves the initial abundance estimate as an anchor and characterizes its mismatch using structured reconstruction residuals. Abundance candidates are then combined with structured residual information, while only 1% reference abundances are used to learn the correction. RDAC was evaluated on the Jasper Ridge, Samson, and Urban datasets. With UNMamba as the primary blind-unmixing backbone, RDAC achieved A-RMSEs of 0.0229, 0.0175, and 0.0359, reducing the initial errors by 55.05%, 70.80%, and 59.63%, respectively. It also achieved the lowest mean A-RMSE among the compared methods on Jasper Ridge and Samson while remaining competitive on Urban. Further experiments showed that reconstruction error alone does not reliably indicate abundance accuracy. When applied to NMF, RDAC reduced A-RMSE by 80.45%, 89.25%, and 77.61% on the three datasets. The results indicate that structured residual diagnosis can support abundance correction under limited supervision, and the NMF experiment shows that the same correction framework can also be used with an alternative blind-unmixing backbone.

Remote Sensing doi: 10.3390/rs18213793

Authors: Xu Yang Zhenyuan He Jiajia Yuan Mingfei Zhu Xuexiang Yu Lei Wang Yulong Ge Shuai Nie

Ground surface subsidence resulting from underground mining is characterized by significant spatial deformation gradients and nonlinear temporal deformation. A single monitoring technique is insufficient to simultaneously achieve extensive spatial coverage and high temporal continuity. Interferometric Synthetic Aperture Radar (InSAR) provides high spatial resolution but is limited by sparse temporal sampling, whereas the Global Navigation Satellite System (GNSS) offers dense temporal sampling but with restricted spatial coverage. To address these limitations, this study introduces a spatio-temporal fusion method combining InSAR and GNSS, based on baseline deformation field–residual decoupling (BR-STKF), specifically applied to the Banji Coal Mine in the Huainan Mining Area, Anhui Province. Sentinel-1 SBAS-InSAR is employed to derive Line-Of-Sight (LOS) deformation time series, while three-dimensional GNSS displacements are projected onto the LOS direction. A daily scale baseline deformation field is constructed using pixel-wise Kalman filtering and Rauch–Tung–Striebel (RTS) smoothing. Subsequently, residual signals between the InSAR/GNSS observations and the baseline field are estimated utilizing a correlation-anchored multiscale spatial basis and an information-form Kalman filter. Helmert variance component estimation is introduced to adaptively adjust the weights of InSAR and GNSS observations. To demonstrate the reliability of the proposed framework, a quantitative validation was conducted. The experimental results yielded root mean square error (RMSE) values of 6.37 mm for strict spatial leave-one-out cross-validation (LOOCV) and 2.84 mm for retrospective reconstruction with missing InSAR phases. These results are competitive and comparable to leading benchmarks such as the Spatio-Temporal Kalman Filter (STKF), while achieving lower mean RMSEs than baseline methods such as Spatio-Temporal Random Effects (STREs), Enhanced Spatio-Temporal Random Effects (eSTREs), Improved Spatio-Temporal Random Effects (ISTREs), Multi-resolution Orthogonal Polynomial Basis (MOP), and ST-Kriging. However, the statistical stability of differences among models requires further verification with additional independent data. The findings suggest that the baseline deformation field–residual decoupling strategy effectively integrates the spatial coverage of InSAR with daily GNSS observations, offering a viable framework for daily scale line-of-sight (LOS) deformation reconstruction in mining areas.

Remote Sensing doi: 10.3390/rs18213792

Authors: Dongdong Bu Lei Liu Zhuli Xie Gang Wan

Remote sensing image captioning enables automated interpretation of remote sensing data for applications such as environmental monitoring and urban planning. However, general-purpose multimodal models exhibit suboptimal adaptability on remote sensing imagery due to domain discrepancies in data distribution, resolution, and rotation invariance. We propose a domain-specific memory bank-enhanced framework for remote sensing image caption, built upon the lightweight EVCap architecture with three core innovations: (1) a remote sensing-specific visual memory bank providing domain-representative visual–semantic alignment; (2) a staged transfer training strategy that leverages pretrained knowledge while mitigating overfitting from limited remote sensing data; and (3) a K-means clustering-based small-sample memory bank that preserves performance while reducing memory footprint for resource-constrained deployment. Extensive experiments on the NWPU, RSICD, and Sydney datasets demonstrate substantial improvements over baselines. On the test sets, the domain-specific memory bank (R1) improves CIDEr by 1.87% on RSICD (0.3255 to 0.3316), and the clustering-based variant (R4) by 11.7% on Sydney (0.1422 to 0.1589) using only 39% of the full bank, despite decreases on NWPU and RSICD, with about 3.97 million trainable parameters.

Remote Sensing doi: 10.3390/rs18213791

Authors: Zhikun Chen Binfeng Wu Yi Zhang Yongping Ji Xiangyu Wu Shuting Sun

The HY-2B/C/D constellation delivers routine altimeter significant wave height (SWH) and sea-surface wind speed over the Chinese coastal seas through four product streams (GDR, IDR, ODR, and SDR), yet whether the three satellites are mutually consistent, and whether that consistency is the same for both parameters, has not been established. We assess three years (2023–2025) of 1 Hz Ku-band retrievals against 30 quality-controlled coastal buoys, using a deduplicated matchup database, a common-buoy and common-period protocol with bootstrap confidence intervals, and triple collocation against ERA5. Within each satellite, the product streams are consistent for both parameters: the reprocessed GDR shifts the operational SWH by a sea-state-independent −0.157 m and the wind streams coincide to rounding. Across satellites, consistency is parameter-dependent. For SWH, the biases on common buoys span 0.18 m; HY-2B has the lowest random error, significantly lower than HY-2D, and HY-2C cannot be separated from either in random error. For wind speed, HY-2C reads about 1 m s−1 low in all four Chinese seas and in every wind band once the bands are defined by an independent reference, so that most of the offset is removable by a constant. The near-shore degradation of SWH inside 50 km largely reflects the sub-0.5 m sea states that dominate the near-shore buoys, and wind speed shows no comparable coastal signal. We translate these findings into applicability bounds and a parameter-aware recipe for merging the constellation.

Remote Sensing doi: 10.3390/rs18213790

Authors: Hao Zhang Yusen Liu Chao Wen Xianhong Zhao Fei Han Tingting He Enlong Wang Yicheng Wang Shouzheng Zhu Genghua Huang Fashuai Li Yuwei Chen

Hyperspectral image (HSI) classification is a fundamental remote sensing task, but the high-dimensional nature of HSI data poses challenges for acquisition, storage, transmission, and processing. To mitigate these issues, researchers have developed compressive classification methods based on the coded aperture snapshot spectral imaging (CASSI) system, enabling direct classification in the compressed domain without reconstructing the full hyperspectral cube. However, existing methods often fail to fully leverage spatial-spectral correlations, suffer from limited multi-scale feature extraction, and are constrained by fixed receptive fields and degradation caused by compression. To overcome these limitations, we propose ECSA-UNet, an end-to-end framework that directly classifies compressive spectral measurements from CASSI. The model introduces a spatial-spectral decoupling module and refines the U-Net architecture with global–local feature extraction, Spatial-Spectral Attention, a Multi-Scale Enhancement Pyramid and Cross-Stage Connections, enabling dynamic receptive fields, improved long-range dependency modeling, and robust local–global feature fusion. Moreover, the CASSI coded aperture is parameterized as a trainable component, allowing for joint optimization of coding and classification within an end-to-end learning paradigm. Extensive experiments on four benchmarks show ECSA-UNet achieves state-of-the-art accuracy, and ablations confirm each module’s contribution. In addition, it maintains an effective trade-off between parameter overhead and computational efficiency, offering a viable solution for latency-sensitive hyperspectral remote sensing tasks.

Remote Sensing doi: 10.3390/rs18213789

Authors: Feng Xing Yuchuan Yang Deng Pan Caiyan Qin

Unmanned aerial vehicle (UAV) thermal infrared remote sensing can rapidly acquire thermal-response information for large-scale photovoltaic (PV) arrays. Nevertheless, local thermal anomalies occupy small regions within aerial images; their features tend to degrade during successive down-sampling. Moreover, model deployment on edge devices is constrained by limited computational and memory resources. To address these challenges, this paper proposes TP-YOLO, a photovoltaic anomaly-detection approach, and implements it on the Zynq-7020 hardware platform. First, a three-channel thermal-prior input consisting of luminance distribution, local luminance difference, and a luminance gradient is constructed from pseudo-color infrared images, and explicitly characterizes thermal anomalies without requiring absolute-temperature measurements. Next, the detection scales of YOLOv11n are adjusted to P2/P3/P4. Modules including A-C3k2, Conv + Add, and TLCB are adopted to optimize feature extraction and multi-scale fusion, preserving local anomaly features while restraining model complexity. On the hardware side, a PS/PL co-processing architecture is employed. A dedicated convolution IP, ADD IP, and Window IP are designed for convolution, element-wise fusion, and local-window operations, respectively. INT8 quantization and ping-pong buffering are further utilized to realize network deployment. On the PV HSD test set, TP-YOLO achieves an mAP@0.5 of 77.5% and an mAP@0.5:0.95 of 33.7%, with 1.91 M parameters and 6.2 GFLOPs computational cost. In cross-dataset evaluation on the Drone Dataset without fine-tuning, the mAP@0.5 reaches 64.4%. After deploying the INT8 quantized model onto Zynq-7020, the implemented system yields an mAP@0.5 of 76.8%, achieving single-frame inference latency of 162 ms and throughput of 6.17 FPS at 150 MHz, with a power consumption of 3.4 W.

Remote Sensing doi: 10.3390/rs18213788

Authors: Pengfei Li Shuzhe Huang Ran Liu Wang Gao Fan Zhang Chao Wang

Compound temperature and soil moisture extremes threaten terrestrial carbon uptake, but their contrasting ecological impacts remain poorly understood. We assessed dry–hot (DH), wet–hot (WH), dry–cold (DC), and wet–cold (WC) events across China from 2001 to 2025. The analyses combined daily air temperature, root-zone soil moisture, and MODIS-derived gross primary productivity (GPP). We quantified event occurrence, negative-response probability, cumulative GPP loss, and recovery time, and examined environmental predictors using explainable machine learning. WC events were most frequent, whereas DH events lasted longest. WH frequency increased over the study period, while DC frequency declined. DH and DC produced negative GPP responses in 63.5% and 67.1% of classifiable events, respectively. Among negative-response events, DH had the largest national mean cumulative event-period GPP loss below the −0.5 SD threshold (2.85 g C m−2) and the longest conditional recovery time (3.17 days). WC losses exceeded DC losses despite a lower negative-response probability. Losses varied with elevation, season, and event severity. Vapor pressure deficit dominated predictions of DH negative responses and losses, whereas other event types showed greater contributions from radiation, precipitation, and soil texture. These findings highlight the need to distinguish event occurrence, response probability, productivity loss, and recovery when assessing ecological risks from compound extremes.

Remote Sensing doi: 10.3390/rs18213787

Authors: Kuiran Wang Xumeng Han Xuehui Yu Zhipei Huang Zhenjun Han

Single-point annotation in oriented object detection of remote sensing scenarios is gaining increasing attention due to its cost-effectiveness. However, due to the granularity ambiguity of points, there is a significant performance gap between previous methods and those with full supervision. In this study, we introduce P2RBox, which employs point prompt to generate rotated box (RBox) annotation for oriented object detection. P2RBox employs the SAM model to generate high-quality mask proposals. These proposals are then refined using the semantic and spatial information from annotation points. The best masks are converted into oriented boxes based on the feature directions suggested by the model. P2RBox incorporates two advanced guidance cues: Boundary Sensitive Mask guidance, which leverages semantic information, and Centrality guidance, which utilizes spatial information to reduce granularity ambiguity. This combination enhances detection capabilities significantly. To demonstrate the effectiveness of this method, enhancements based on the baseline were observed by integrating three different detectors. Furthermore, compared to the state-of-the-art point-annotated generative method PointOBB, P2RBox outperforms by about 29% mAP (62.43% vs. 33.31%) on DOTA-v1.0 dataset and 7.1 mean IoU (68.9 vs. 76.0) on WHU building dataset, which provides new possibilities for the practical application of point annotation techniques in multi-class object detection (e.g., building) from remote sensing imagery.

Remote Sensing doi: 10.3390/rs18213785

Authors: Jiayong Yu Susu Zhang Qi Chen Deliang Lv Jiachun Guo Peng Lin

To address scale-like errors in local point clouds caused by pitch angle errors in unmanned aerial vehicle-borne LiDAR bathymetry (UAV-ALB) systems, this study proposes a correction method based on scanline keypoint constraints. First, the point cloud is traversed using scanline information to extract geometric keypoint features from each scanline. Next, a scale-like error correction model is constructed based on the distance differences between keypoints at the start and end of overlapping scanlines, and the pitch angle correction value is estimated using a forward–backward fusion approach. The estimated correction is then applied to the attitude data, and the point cloud is reprocessed. This procedure is iterated until the convergence criterion is satisfied. Experimental results show that, after correction, the horizontal root mean square errors (RMSE) of the two datasets decrease from 0.217 and 0.283 m to 0.025 and 0.030 m, respectively, while the elevation RMSE values decrease from 0.227 and 0.241 m to 0.006 and 0.004 m, respectively. These results demonstrate that the proposed method effectively mitigates scale-like errors in UAV-ALB point clouds and substantially improves their geometric consistency and elevation accuracy.

Remote Sensing doi: 10.3390/rs18213786

Authors: Xinyu Li Jiaxing Mo Zhiyuan Wang Qian Tang Meng Zhang Hui Lin

Forest canopy height is a key parameter for estimating forest biomass and carbon stocks, as well as a fundamental indicator for assessing forest ecosystem health. Although existing 10 m resolution global canopy height products are free and open-source, they suffer from substantial systematic errors due to the inherent biases of spaceborne LiDAR and the nonlinear relationship between optical imagery and canopy height, making them insufficiently accurate as direct training labels for deep learning. This study proposes a complete framework of “adaptive stratified screening, followed by weakly supervised UNet training and minimal ALS post-processing refinement.” First, using the 12 spectral bands of Sentinel-2 multispectral imagery, the OOB distance is calculated via weighted random forest as an anomaly metric. Through systematic comparison of three anomaly metrics (Cook’s distance, OOB distance, and standardized residuals) and five retention ratios (30–50%), the OOB distance at 40% retention is identified as optimal. An adaptive retention strategy based on tree height stratification is then implemented: for low-tree areas (<20 m), 40% of low-anomaly samples are uniformly retained; for tall-tree areas (≥20 m), the retention ratio is dynamically adjusted according to the proportion of tall-tree pixels (95% when tall-tree proportion < 10%, 70% when 10–30%, and 50% when ≥30%). The screened high-confidence Lang data are then used as supervisory signals to train a UNet model for predicting canopy height across the entire region, producing an initial canopy height map (MAE = 6.97 m, Bias = +4.21 m). Finally, using only a minimal amount of airborne LiDAR data (down to 0.1% area proportion), with a 14-dimensional input comprising the initial predicted height, screening mask values, and the 12 spectral bands of Sentinel-2, a multilayer perceptron is independently trained for each region to perform post-processing refinement of the initial predictions. Experiments based on 68 independent regions from the Open-Canopy dataset, with a strict regional split (34 validation/34 test), show that post-processing refinement reduces the test set MAE from 6.97 m (ASS) to 3.29 m (0.1% ALS, a 52.8% improvement, 58.1% improvement over the original Lang product), Bias from +4.21 m to −0.13 m, and RMSE from 8.95 m to 4.72 m. Even when the ALS proportion is reduced to 0.01%, the MAE remains 3.57 m (a 48.8% improvement). Statistical significance testing (paired Wilcoxon signed-rank test, n = 34) indicates that ASS+MLP(0.1%) significantly outperforms ASS (p < 0.00001). This study provides an effective technical pathway for regional refinement of global canopy height products.

Remote Sensing doi: 10.3390/rs18213784

Authors: Jie Tang Fugen Jiang Hua Sun Kaisen Ma

Diameter at breast height (DBH) and tree height are core parameters for forest monitoring. LiDAR enables rapid high-precision point cloud acquisition, yet single-platform data have limitations: airborne laser scanning (ALS) cannot capture sub-canopy structure, while terrestrial laser scanning (TLS) cannot fully acquire canopy information. Based on 92 plots in Guangxi, this study proposes a weighted point-to-normal iterative closest point (WPN-ICP) registration algorithm for high-precision fusion of ALS and TLS point clouds, with comparisons to trimmed iterative closest point (TrICP) and iterative closest point (ICP). An adaptive density-based spatial clustering of applications with noise (DBSCAN) algorithm was developed for diameter at breast height (DBH) estimation, and a DBH cluster-based vertical layering method (DCVL-THE) for tree height estimation. Three segmentation-based methods—canopy height model (CHM), region growing (RG), and comparative shortest path (CSP)—were selected for comparison. WPN-ICP achieved a mean MAE of 0.0359 m, with reductions of 85.87%, 69.03%, and 73.00% compared with coarse registration, TrICP, and ICP. Using the fused point cloud, DBH estimation achieved an RMSE of 0.6326 cm with an RRMSE of 5.18%, and tree height estimation achieved an RMSE of 0.7160 m with an RRMSE of 6.58%. Compared with the suboptimal CSP method, the RMSE and RRMSE for DBH were reduced by 41.79% and 41.68%, and for tree height by 25.66% and 25.68%. This method does not require full individual-tree or crown segmentation, thereby avoiding the error propagation associated with such segmentation and enabling accurate stand-scale estimation.

Remote Sensing doi: 10.3390/rs18213783

Authors: Caterina Balletti Enrico Breggion Andrea Martino Luisa Berto Irene Rocca Diego Alejandro Talledo

Historic hill towns require geometric information capable of relating urban morphology, construction features, and structural details within a common spatial and semantic framework. This paper presents a multi-sensor, multi-scale geomatics workflow for establishing an integrated geospatial information framework for the historic centre of Fiumefreddo Bruzio in Calabria, southern Italy. A control network of 38 points provided the spatial reference for UAV and terrestrial photogrammetry, six SLAM trajectories, and 270 static laser scans covering five buildings. Point clouds generated by Structure from Motion and range-based sensors were co-registered in the RDN2008/UTM zone 33N coordinate reference system and used to produce orthophotos, digital terrain and surface models, contour lines, textured meshes, and building-scale drawings. These products were organised within an interoperable geospatial information system based on persistent links between settlement-, aggregate-, building-, façade-, space-, and macroelement-level entities. The framework also supported structural characterisation and seismic vulnerability assessment at building and urban scales, linking building screening and potential local mechanisms to their spatial interpretation. The resulting survey-based digital model supports conservation documentation, seismic screening, and street-network analysis, while providing the foundation for future digital twin development rather than an operational implementation. The study offers a transferable acquisition-to-information workflow, preserving data provenance and metric quality for risk-informed management of historic settlements.

Remote Sensing doi: 10.3390/rs18213782

Authors: Mauro Morata Junior Lucas Prado Osco Lúcio André de Castro Jorge Ana Paula Marques Ramos

Huanglongbing (HLB) is the most destructive disease affecting citrus production worldwide, causing severe yield losses and tree decline. Hyperspectral proximal sensing provides detailed information on leaf reflectance; however, its effectiveness for reliable HLB detection depends on the identification of consistent and discriminative spectral patterns. In this study, hyperspectral reflectance data (400–2450 nm) were combined with supervised machine learning algorithms to distinguish HLB-positive and HLB-negative citrus samples at two complementary scales: (i) leaf-level classification using individual spectral signatures and (ii) plant-level classification based on aggregated spectral responses. Model performance was evaluated under a leave-one-field-out validation framework designed to minimize spatial dependence between training and testing data. Results demonstrated consistently high classification performance, with Logistic Regression achieving accuracies above 0.94 at the leaf level and up to 1.00 at the plant level in several configurations. Dimensionality reduction using Principal Component Analysis (PCA) preserved most of the discriminative information, although with a slight reduction in recall under certain FIELD combinations. In contrast, Class balancing produced no observed improvement, whereas hyperparameter optimization showed limited and configuration-dependent effects in classification performance. Spectral coefficient analysis revealed that the most relevant wavelengths are concentrated in the visible (400–700 nm), red-edge (700–750 nm), near-infrared (750–1300 nm), and shortwave infrared (1500–2500 nm) regions, reflecting physiological alterations associated with pigment degradation, mesophyll disruption, and changes in water and biochemical composition. Overall, hyperspectral reflectance combined with parsimonious machine learning models enabled consistent discrimination of HLB-positive and HLB-negative samples within the evaluated orchard under validation designed to reduce spatial dependence, particularly at the plant scale.

Remote Sensing doi: 10.3390/rs18213781

Authors: Chenyu Shao Zhenyu Fang Heping Song Gen Li Jiangbin Zheng

Hyperspectral anomaly detection (HAD) requires effective background modeling while limiting the reconstruction of rare anomalous spectra. Conventional non-negative matrix factorization provides an interpretable endmember–abundance representation but offers limited flexibility for estimating background representations under complex spectral variability, whereas reconstruction-oriented autoencoders provide flexible nonlinear inference but often lack an explicit mixing-model structure. To bridge these two paradigms, we propose a Non-negative Latent Factorization Autoencoder (NLFA) for anomaly-oriented background modeling. NLFA employs a nonlinear encoder to estimate simplex-constrained abundance-like latent coefficients and a linear mixing model (LMM)-consistent linear decoder to represent the dominant background through endmember-like spectral bases. A column-sparse residual is jointly optimized with the latent factorization so that spectral components insufficiently explained by the learned background are retained for anomaly detection, while local spatial smoothness and endmember-diversity constraints regularize the background representation. Experiments on eight benchmark hyperspectral datasets show that NLFA achieves competitive detection performance.

Remote Sensing doi: 10.3390/rs18213780

Authors: Kyeong-Hun Cho Dochul Yang Seung-Kuk Lee

Accurate ground elevation beneath forest canopies is important for terrain and forest-structure analysis, yet satellite-based elevation data can be affected by canopy and terrain conditions, while spaceborne light detection and ranging (LiDAR) provides ground elevations only at discrete locations. Here, the elevation offset between the Copernicus GLO-30 DEM (CopDEM) and Global Ecosystem Dynamics Investigation (GEDI) L2A ground elevation was estimated using an Explainable Boosting Machine (EBM) with topographic, canopy-structure, and CopDEM-uncertainty predictors to generate continuous forest digital terrain models (DTMs) for two regions in the Republic of Korea. Validation against independent airborne LiDAR data showed root mean square errors (RMSEs) of the proposed DTM of 4.10 m in the central-western region and 3.67 m in the eastern region. In the EBM, the CopDEM height error mask (HEM), canopy cover, and aspect contributed most to the estimated offset, and the magnitude and direction of their contributions varied across canopy–terrain conditions and between regions. Residual bias increased with canopy height and became more negative on steep slopes, where RMSE also increased. These results show that the proposed approach can generate continuous forest DTMs while quantifying the contributions of the main predictors to the offset, their variation across environmental conditions, and the conditions under which residual errors increase.

Remote Sensing doi: 10.3390/rs18213779

Authors: Flora Stanley Laiba Amarouche Bruno Picard Laura Hermozo Xavier Boulanger Marie-Laure Frery Nicolas Picot Patrice Gonzalez Gerald Dibarboure

A key correction in altimetry measurements is the wet tropospheric correction (WTC), which accounts for the delay on the measured range predominantly caused by water vapor in the troposphere. The WTC is currently one of the main sources of error affecting the altimeter range over ocean surfaces, with direct impact on estimates of sea surface height and the global mean sea level. Most current altimetry missions rely on onboard passive microwave radiometers that measure the brightness temperature at key frequencies for the spatio-temporal collocated estimation of the WTC. For such missions, the current operational algorithms for the estimation of the WTC can be placed in two categories: those using a neural network and those using a log-linear equation dependent on classes of WTC and windspeed. These two methodologies both follow a semi-empirical approach, but differ significantly in the inputs used, the training databases on which they are developed, and the mathematical approach itself. The objective of this study is to conduct a comprehensive comparison between these two algorithms, focusing solely on the mathematical aspect, in order to determine if the differences in the approaches of the two algorithms have a significant effect on the resulting WTC products, their performance and their limitations. To ensure the identical conditions required to achieve this, we develop the two algorithms so that they take the same inputs, have the same training database, and then we apply them to identical simulated and observed datasets to determine their performance. We use the three-frequency configuration (18.7 GHz−23.8 GHz−34 GHz) of the Advanced Microwave Radiometer for Climate (AMR-C) on Copernicus Sentinel-6 Michael Freilich (Sentinel-6MF), and radiative transfer modelling using the Radiative Transfer for TOVS (RTTOV), to create a simulated database that will serve for both development of the algorithms and performance evaluation. Both algorithms are then applied to the same observations from Copernicus Sentinel-6MF to characterise their observational performance in comparison to both model and operational products, as well as to each other. We find that, despite the significant differences in the two methodologies, when the algorithms are developed and applied to the same databases, they deliver very similar performances both on simulated and observed datasets and face similar limitations. In addition, for both algorithms we find performances similar to or slightly better than that of the operational WTC product for Copernicus Sentinel-6MF.

Remote Sensing doi: 10.3390/rs18213778

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

The Mamba network has recently emerged as a powerful architecture for remote sensing change detection (RSCD). Notably, the fusion of multi-scale features and heterogeneous attention mechanisms has proven effective in enhancing the performance of Mamba network. However, existing methods fail to quantitatively evaluate the actual contribution weights of different components during feature fusion, thereby limiting the interpretability of their improvements. To address this issue, this paper proposes a novel mixture-of-scale-and-attention-experts Mamba Network (MoEMamba) for the RSCD task, which has the following innovations and contributions: (1) a Siamese Mamba architecture is employed to extract features from bi-temporal images; (2) a mixture-of-scale-experts module is designed to enhance the generalization ability across multi-scale objects, which quantitatively evaluates the contribution of multiple scale-specific experts; and (3) a mixture-of-attention-experts module is proposed to improve the accuracy, which adaptively fuses complementary responses of heterogeneous attention mechanisms based on the learned weights. Extensive experiments on the WHU-CD and LEVIR-CD datasets demonstrate that MoEMamba outperforms existing state-of-the-art methods. Specifically, it achieves F1 scores of 94.20% and 92.62%, and IoU values of 89.04% and 86.25%, respectively. Importantly, this study provides valuable insights into the interpretability of multi-scale feature fusion and heterogeneous attention fusion for RSCD research.

Remote Sensing doi: 10.3390/rs18213777

Authors: Changjiang Shi Cui Zhou Chuanqi Liu Dewei Yin Zhixin Liu Yaning Yi

The NASA–ISRO Synthetic Aperture Radar (NISAR) mission provides L-band soil moisture at 200 m, but validating its early products depends on retrieval flags, overpass collocation and sensor depth. We evaluated the beta Level-3 product against seven U.S. in situ networks (October 2025–January 2026) and the provisional product at 21 National Ecological Observatory Network sites (June–July 2026). Beta retrievals with flag 0, which averages all three candidate algorithms, were matched to in situ data within 60 min, merged by orbit and averaged over 0–10 cm sensors, yielding 261 site overpasses from 117 sites. Beta NISAR had a preliminary bias of 0.003 m3 m−3, a root mean square error (RMSE) of 0.081 m3 m−3 (95% interval 0.066–0.098) and R=0.578. Accepting all flag classes increased coverage ninefold and RMSE to 0.096 m3 m−3. On common site overpasses, deeper sensors did not have significantly higher RMSE than 5 cm sensors. The exploratory provisional assessment found a wet bias of 0.110 m3 m−3, concentrated at very dry reference sites, with a pooled R of −0.038 but an anomaly R within sites of 0.259. These estimates rest on few sites and short records; validation of early NISAR products should report flag classes, overpass construction and matched depths and separate pooled from temporal agreement.

Remote Sensing doi: 10.3390/rs18213776

Authors: Nicolae Ajtai Alexandra-Maria Câțu Alexandru Mereuță Camelia Gînscă Horațiu-Ioan Ștefănie Andrei-Titus Radovici Camelia Talianu Livio Belegante

Smoke from the Canadian wildfires of August 2025 reached Cluj-Napoca, Romania, on 6–9 August as a thin layer at 5–8 km. We followed the plume from its source with the multi-angle polarimeters HARP2 and SPEXone and the OCI radiometer on NASA’s PACE and characterised it at the RADO-Cluj ACTRIS National Facility with a multiwavelength depolarisation lidar, a ceilometer, AERONET photometry, GRASP inversions, and HYSPLIT trajectories. Along the air mass trajectory, the HARP2 retrievals resolved the ageing of the smoke: an optical depth decaying from 0.9 to 0.2–0.3, a layer rising to 10 km and descending to 5 km, growth of the fine-mode radius, and a single-scattering albedo consistent with brown carbon loss. At the site, the lidar ratio and particle depolarisation ratio retrieved by HARP2 (52 sr, 0.042) agreed with the lidar’s derived lidar ratio and measured depolarisation (50 sr, 0.042), whereas the three satellite optical depths differed by 0.2 around an AERONET value of 0.1 and the NATALI typing networks returned three different classes. For this faint, aged plume, intensive polarimetric and vertical quantities could no longer be identified for the aerosol by its optical depth and Ångström exponent, which motivates testing an intensive-first pairing of satellite polarimetry with ground-based lidar on further episodes.

Remote Sensing doi: 10.3390/rs18213775

Authors: Yingying Kong Dongyu Yuan

Limited labeled data and class imbalance complicate polarimetric synthetic aperture radar (PolSAR) image classification. We propose a complex-valued multiscale feature interaction attention vision transformer (CV-MfIaViT) that combines spatial analytic-envelope preprocessing, complex-valued local feature extraction and global modeling. A two-dimensional Hilbert transform (2D HT) provides envelope features to the shallow pathway. The complex-valued amplitude–phase interaction attention (CV-APIA) module generates positive real gains through cross-modulation of pooled features; the final gating preserves the element-wise phase of non-zero input features. A complex-valued feature complementary mapper (CV-FCM), a complex-valued multiscale extractor (CV-MSE) and a complex-valued vision transformer (CV-ViT) complete the architecture. Under random-pixel sampling, the reported average accuracy (AA) exceeds CV-MsAtViT by 1.83, 3.42 and 2.48 percentage points on Flevoland, San Francisco and Nanjing, respectively. The advantage is also retained under spatial-block partitioning with shared label-free scene-level normalization. Component ablations and comparisons with two simple learned front ends characterize the combined design. The results support the evaluated architecture without isolating phase preservation as the cause of its accuracy gains.

Remote Sensing doi: 10.3390/rs18213774

Authors: Mohamed Chikh Essbiti Morad Dalal Simon Jones Hasna Eloudi Abdenbi Elaloui Mohamed Elhaou Maryem Ismaili Abdullah B. Saadon Mohammed S. Fnais Kamal Abdelrahman Samira Krimissa Mustapha Namous

Accurate forest-cover mapping in fragmented Mediterranean landscapes remains challenging because forests, croplands, grasslands, and sparsely vegetated surfaces often exhibit overlapping spectral responses. This study benchmarks ten supervised machine-learning classifiers for regional forest-cover mapping in the Beni Mellal-Khenifra region of Morocco, evaluates the contribution of spectral, textural, and topographic predictors, and assesses the consistency of the resulting forest map with the Hansen Global Forest Change product. A reference dataset of 2800 samples representing forest, crops/grass, bare ground, urban areas, and water was combined with an annual 2025 Sentinel-2 median composite, spectral indices, NDVI-derived gray-level co-occurrence matrix textures, and SRTM-derived elevation and slope. The samples were divided once into a stratified 70% training set and a 30% test set common to all classifiers. Redundancy, correlation (|r| > 0.80), and variance inflation factor (VIF < 10) screening, fitted on the training set only, reduced 56 candidate predictors to 16, and hyperparameters were tuned by five-fold grid-search cross-validation within the training set. On the common test set (n = 840), CatBoost, XGBoost, and the multilayer perceptron each achieved an overall accuracy of 95.95% (Cohen’s κ = 0.947), followed by Gradient Tree Boosting (95.83%); McNemar tests with Holm correction detected no significant differences among these four models. Spatial block cross-validation reduced overall accuracy by 0.3–2.7 percentage points, and CatBoost retained the highest spatially blocked accuracy (95.51% with 10 km blocks). SHapley Additive exPlanations (SHAP) identified slope, SATVI, NDWI, and the blue band as the dominant contributors to CatBoost forest predictions. The leading maps estimated 859,057–870,663 ha of forest. CatBoost captured 90.9% of the Hansen forest extent (≥30% canopy cover) but mapped a much larger forest area; blind interpretation of stratified samples showed that 90.5% of the CatBoost-only forest was woody vegetation, mainly open woodland and matorral. These findings show that boosting ensembles combined with a compact, non-redundant predictor set provide accurate and spatially robust forest-cover maps and that differences from global products largely reflect forest definitions.

Remote Sensing doi: 10.3390/rs18213773

Authors: Feifei Pan

Accurate monitoring of river water surface elevation (WSE) is important for hydrologic and flood applications, particularly where in situ observations are limited. This study evaluates Sentinel-1 Synthetic Aperture Radar (SAR) for estimating river WSE using a river inundation area–water surface elevation (RIA–WSE) relationship derived from high-resolution topobathymetric elevation data and investigates environmental factors affecting estimation accuracy. A total of 265 Sentinel-1 SAR images acquired during 2015–2026 covering the USGS Thebes gauging station along the Mississippi River were analyzed using six VV- and VH-based backscatter metrics, yielding 1590 WSE estimates. Comparison with a conventional 1-m digital elevation model (DEM) revealed that its representation of river water-surface elevations at the time of the elevation survey, rather than submerged channel-bed elevations, limits its applicability for estimating WSE at low river stages. Among the six metrics, log10(VV) produced the best overall performance, with an RMSE of 1.61 m, mean error of −0.96 m, R2 of 0.81, and percentage of lowest absolute error (PLE) of 44.5%. A sensitivity analysis using SAR image grid spacings of 10, 5, 2, and 1 m showed that log10(VV) maintained relatively stable estimation accuracy, whereas finer resampling did not consistently improve WSE estimates. Errors were smallest under low-flow conditions (WSE < 97.0 m), largest under medium- to high-flow conditions (97.0–102.5 m) and decreased again during flooding (WSE > 102.5 m). Analysis of 10 National Agricultural Imagery Program (NAIP) images revealed increasing vegetation along the inundation boundary above approximately 100 m WSE. Under medium- to high-flow conditions, calm conditions generally produced smaller errors than non-calm conditions, although errors did not increase monotonically with wind speed. E–W and SE–NW wind orientations generally produced larger errors than N–S and NE–SW orientations. These results indicate that WSE estimation accuracy reflects combined effects of vegetation, wind-generated water surface roughness, wind orientation, effective fetch, water depth, and river morphology. Although log10(VV) performed best overall, it provided the lowest absolute error for only 44.5% of acquisitions, highlighting the potential for condition-dependent backscatter metric selection to improve SAR-based river WSE retrieval.

Remote Sensing doi: 10.3390/rs18213770

Authors: Yifan Deng Xiang Chen Zhigao Chen Jiaxing Zhou

Image registration is fundamental to UAV remote sensing data processing, but repetitive textures, viewpoint and scale variations, terrain relief, and local deformation can introduce numerous outliers into tentative correspondences. The existing random-sampling and spatial-consistency methods commonly rely on pointwise confidence or isolated pairwise relations and therefore cannot reliably determine whether a locally compatible relation receives stable group-level structural support. To address this limitation, we propose SPGSC++, a second-order spatial guided sampling method for outlier removal. SPGSC++ first uses PCA to reduce the dimensionality of tentative correspondences, thereby reducing the computational burden of subsequent spatial relation construction. It then derives second-order structural quality from the common-neighbor support of directly compatible edges and uses this quality for initial data subset selection and residual-informed data subset refinement to generate and optimize a homography. Comparative experiments show that, on the five datasets CIAP, SUIRD, UAV, ZHANGZHOU, and PAN, the proposed method achieves mean Precision, Recall, F-Score, Runtime, and mean reprojection error values of 0.9627, 0.9797, 0.9702, 0.3724 s, and 1.6454, respectively. Relative to the original SPGSC, SPGSC++ increases the mean F-Score by 0.0391 and reduces Runtime and mean reprojection error by 32.39% and 58.12%, respectively. Overall, the innovation of SPGSC++ lies in using second-order common-neighbor structural quality to guide initial subset selection and residual-informed refinement for robust estimation in UAV remote sensing image registration.

Remote Sensing doi: 10.3390/rs18213771

Authors: Sagar Rawal Guoquan Wang Rajesh Silwal Jack Thomson Muhammad Asif

Interferometric Synthetic Aperture Radar (InSAR) is a satellite–based remote sensing technique capable of measuring surface displacement. However, InSAR measurements can be affected by temporal decorrelation, baseline geometry, and surface characteristics. Persistent scatterer (PS) and small baseline subset (SBAS) are two widely used multi-temporal InSAR techniques for addressing these limitations. Processing SAR acquisitions to derive deformation maps and displacement time series requires a series of complex steps. Many existing open-source and commercial software packages support PS or SBAS, while some support both methods, each with its own challenges and limitations. In this study, we presnet SNAPxtra, a Python-based InSAR automatic workflow to generate PS and SBAS interferometric stacks. These can be exported in the format required for time-series analysis in LiCSBAS and StaMPS. SNAPxtra has automated master selection, multi-swath and multi-frame processing, polygon-based masking, trial processing, and network generation features within an integrated workflow. The reproducibility and adaptability of SNAPxtra are demonstrated through case studies in Quetta (Pakistan), Surat (India), and Fort Worth (USA) as case studies using both PS and SBAS methods.

Remote Sensing doi: 10.3390/rs18213772

Authors: Mingzhong Xiao Dongdong Kong

The consequences of drought for rivers depend not only on the atmospheric water deficit but also on how catchments store and release water. It remains unclear how well catchment characteristics distinguish streamflow drought occurrence under similar meteorological conditions. We combined AlphaEarth satellite descriptors with principal component analysis to examine this question in 361 relatively undisturbed catchments across the conterminous United States. Using the three-month Standardized Precipitation Evapotranspiration Index (SPEI) and Standardized Streamflow Index (SSI), we linked meteorological droughts to subsequent streamflow droughts and assessed whether catchment landscapes helped explain differences in their occurrence, intensity and duration under comparable meteorological and initial streamflow conditions. Occurrence was most clearly associated with PC1, the dominant component of the AlphaEarth descriptors, whose landscape differences were partly reflected in drainage density and forest cover. At comparable meteorological drought durations and the same meteorological intensity and initial streamflow conditions, catchments in the upper PC1 quartile had estimated occurrence probabilities 8.1–20.9 percentage points below those in the lower quartile. The separation was greater for moderate-or-stronger drought, at 13.8–22.3 points. Higher PC1 also tended to accompany lower drought intensity, but the evidence for intensity and duration differences was less conclusive than for occurrence. These findings support the use of a common satellite-derived landscape description to identify catchments more susceptible to streamflow drought under comparable meteorological conditions, while showing that occurrence probability alone cannot indicate how severe or persistent the resulting drought will be.

Remote Sensing doi: 10.3390/rs18213769

Authors: Abu Ubaida Khurram Zeeshan Haider Syed Ali Asad Naqvi Adnanul Rehman Farhan Aadil Shoaib Khalid

Deep-learning research on flood inundation mapping is driven largely by architectural innovation and single-run reporting. We ask whether the segmentation architecture or the input representation contributes more to accuracy in flood-event surface-water mapping. We introduce GeoHybridNet, a hybrid CNN–Transformer network, and compare it on the hand-labeled Sen1Floods11 benchmark with a U-Net, an Attention U-Net, a parameter-matched wide U-Net, and SegFormer under matched data and training. The four convolutional and hybrid models differ by at most 1.4 IoU points, and none are significantly different under event-level tests; a compact SegFormer trained from scratch is about 3.5 points lower. In factorial designs crossing three architectures with up to five input representations (five seeds each), input representation explains about 95% of the variance in water IoU and architecture about 1%. The dominant input factor is radiometric normalization: raw bands standardized per chip reach only 60% IoU and adding NDWI restores 76%, whereas the same bands standardized with training-set statistics reach 77.5% without any index. The pattern holds when permanent water is excluded. On three held-out flood events, accuracy drops by 24–26 IoU points for every model. On this benchmark, input preparation and honest evaluation matter more than architecture.

Remote Sensing doi: 10.3390/rs18213768

Authors: Joel Younger Anthony Finn Steven Andriolo Stefan Peters Jim O’Hehir

Data from terrestrial GNSS receivers have been combined with point clouds from aerial lidar to estimate the average attenuation coefficients of foliage in planted pine forests at GNSS frequencies. The absorption and scattering of radio signals due to the presence of vegetation can degrade signals from communications and satellite-based navigation systems, making accurate estimates of attenuation essential for predicting equipment performance. Given the potentially limited ability of lidar to penetrate dense foliage, a method was developed to estimate the average attenuation of signals based on distance travelled beneath the maximum height envelope of tree canopies. This was accomplished by using aerial lidar data to determine the outer surface of the forest canopy based on the maximum height in each horizontal bin. Each GNSS satellite detection vector was then analysed to determine the total length traversed inside of the canopy envelope. Application to regions of planted radiata pine in a managed forest outside Adelaide, South Australia, yielded average attenuation coefficient estimates for satellites with a 30–60° elevation angle of 0.36 dB m−1 for 10-year-old medium-height and 0.35 dB m−1 for 19-year-old tall trees. The demonstrated method and results provide site and vegetation-type-specific data to inform predictive models of GNSS receiver performance at specific locations.

Remote Sensing doi: 10.3390/rs18213767

Authors: Ke Ye Jinghui Fan Joaquim Joao Sousa Hongli Zhao Hua Guo Shibiao Bai Guang Liu

Tropospheric phase delay is a major error source that limits the accuracy of Interferometric Synthetic Aperture Radar (InSAR) deformation measurements, including those from the C- and X-band systems commonly examined in the studies reviewed here. Traditional correction methods, including external atmospheric models, spatiotemporal filtering, and phase–elevation empirical approaches, can partially mitigate atmospheric delays but often suffer from limited resolution, strong assumptions, or difficulties in separating atmospheric signals from complex surface deformation. Recent advances in deep learning provide new opportunities for InSAR tropospheric correction by learning nonlinear spatial and temporal relationships from interferometric observations and auxiliary information. This review provides a focused overview of deep learning-based approaches for correcting tropospheric phase delays in InSAR. Existing studies are organized according to their dominant modeling strategies, including convolutional neural network (CNN)-based spatial modeling, recurrent neural network (RNN)-based temporal modeling, generative adversarial network (GAN)-based generative learning, and other feedforward regression approaches. Key network-design strategies, training-label and dataset construction methods, and representative applications are discussed. Finally, current limitations and future research directions are discussed, emphasizing physical consistency, deformation preservation, cross-region generalization, and independent geodetic validation.

Remote Sensing doi: 10.3390/rs18213766

Authors: Corey T. White Pratikshya Regmi William Reckling Helena Mitasova

Civil Air Patrol (CAP) imagery, often the first post-disaster observation, is designed for 2D interpretation rather than multi-temporal volumetric measurement. We build post-storm digital surface and terrain models from uncontrolled CAP imagery with structure-from-motion in western North Carolina’s open Swannanoa Valley and forested Hickory Nut Gorge. Surfaces are co-registered to pre-event LiDAR using roads as pseudo ground control, corrected for terrain-correlated bias, and thresholded with a spatially variable, calibrated Level of Detection. Before bias correction, bare-earth surfaces match withheld post-event LiDAR on stable terrain to 0.68 and 0.82 m normalized median absolute deviation. Volumes are 2017-to-2024 change, lower bounds in coverage, and conditional on the bias model, residual screen, and detection window. Corrected valley surfaces record 125,000 m3 of erosion and 123,000 m3 of deposition, and a sampled image check dates about 13% of that change to the sortie. Under the same limit, the uncorrected surface reports five times the deposition, and the alternative bias model gives gross terms 3.6 to 4.2 times larger. The gorge at Chimney Rock recorded 76,000 m3 of erosion and 84,000 m3 of positive change, just under half dating to the sortie, including land deposition partly cleared before the post-event LiDAR flew. CAP-derived surfaces can capture transient storage that later surveys cannot.

Remote Sensing doi: 10.3390/rs18213765

Authors: Lisa L. Sinclair Su Zhang Christopher D. Lippitt Esteban H. Muldavin

Fine-scale vegetation mapping supports habitat management, landscape change monitoring, and local planning. However, fine-scale maps that distinguish a large number of vegetation classes across broad extents remain scarce. Previous classifications using imagery with spatial resolutions of 1 m or finer have generally distinguished only six to ten classes at broad physiognomic levels and relied on training data generated exclusively through image interpretation. This study develops an object-based classification workflow to map vegetation at the Group Level of the United States National Vegetation Classification across two pilot study areas in New Mexico totaling 705,154 hectares, including the montane Jemez Mountains and the semi-arid Sevilleta region. Segmentation was performed using Geographic Object-Based Image Analysis (GEOBIA) at two hierarchical levels with 0.6 m National Agriculture Imagery Program (NAIP) imagery and a LiDAR-derived digital height model. Classification was conducted using a random forest model with 65 predictor variables derived from NAIP imagery, PlanetScope imagery, Sentinel-2 imagery, LiDAR-derived terrain surfaces, and Soil Moisture Active Passive (SMAP) soil moisture. Training and validation data were obtained from a legacy ecological plot database originally collected for purposes other than classification. Of the 34 classes initially targeted for mapping, 27 were ultimately mapped after spectrally inseparable classes were collapsed into coarser hierarchical levels. Mean overall accuracy across 1000 training and validation partitions was 54.8%, with a minimum of 51.2% to a maximum of 65.6%. Phenological variables derived from PlanetScope imagery improved overall accuracy by an average of 6.5 percentage points. Overall classification accuracy varied by 12.7 percentage points across partitions, indicating that overall accuracy estimates based on a single training and validation split may provide a poor representation of expected model performance.

Remote Sensing doi: 10.3390/rs18213764

Authors: Shentao Wang Yue Hong Ka Po Wong Jin Yeu Tsou Yuanzhi Zhang

Cross-sensor landslide mapping is limited by differences in image resolution and unreliable target pseudo-labels. We present SCALAR (Scale Anchoring and Label-Free Recalibration), an unsupervised adaptation workflow built on an exponential-moving-average teacher and ClassMix. GSD Anchoring (GSA) uses ground sampling distance metadata for bounded resampling. Logit-Otsu Calibration (LOC) jointly adjusts the pseudo-label boundary and relative retention margins. Ratio-Signed Regularization (RSR) makes these margins asymmetric according to the target-to-anchor GSD ratio. It is exploratory; GSA + LOC with symmetric margins is the recommended configuration. These modules introduce no trainable parameters. In the primary CAS region-held-out evaluation across five seeds, GSA + LOC increased the landslide intersection over union (IoU) by 17.9±4.5 points beyond scale anchoring when transferring satellite/aerial imagery (SAT) to unmanned aerial vehicle imagery (UAV), with no resolved additional gain in UAV-to-SAT. At seed 0, SAT-to-UAV threshold-independent average precision (AP) increased from 0.3484 for the anchored source to 0.4809 with GSA + LOC on held-out regions, compared with an increase from 0.6371 to 0.6495 under the original transductive protocol. Under that protocol, fixed-checkpoint region-cluster bootstrap supported GSA gains in both directions and LOC gains only in SAT-to-UAV; RSR’s 95% intervals included zero in both directions. UAV-to-SAT benefits varied across backbones, and negative transfer occurred in some tested settings. The findings support coordinating scale and pseudo-label decisions within CAS while identifying teacher reliability and target resolution as conditions to examine before broader deployment.

Remote Sensing doi: 10.3390/rs18213763

Authors: Chaoli Tang Liupeng Qin Pengfei Wu Yuanyuan Wei

Stratospheric ozone exhibits quasi-biennial-range variability, whose spatial organization can differ among pressure levels and longitudes. We analyzed Aqua Atmospheric Infrared Sounder (AIRS) Version 7 monthly ozone observations from September 2002 to December 2025 at 50, 10, and 3 hPa. A Morlet wavelet analysis of standardized anomalies characterized the 20–36-month variability, with AR(1) Monte Carlo significance tests and complementary multiple-testing and uncertainty assessments. The tropical 10 hPa band showed relatively enhanced normalized power and widespread longitudinal support: 326 of 359 eligible longitudes remained significant after a within-level Benjamini–Hochberg adjustment of band-level tests. In the global 10 hPa grid-cell analysis, 6792 of 62,191 eligible cells passed the corresponding global Benjamini–Hochberg adjustment. Conditional block-bootstrap comparisons indicated differences in some predefined longitude-sector contrasts, whereas the confidence intervals for the 10-minus-50 hPa longitude-power coefficient-of-variation difference included zero. Regressions using NOAA CPC and KIT/FU Berlin QBO wind records indicated longitude-sector heterogeneity in the 10 hPa ozone–QBO regression associations using the 50 hPa wind index. The ERA5 temperature and low-wavenumber geopotential height patterns provided meteorological context, although the assessed ozone–ERA5 spatial alignments did not pass the adopted spatial-null tests. These findings offer a pressure-dependent, longitude-resolved description of quasi-biennial-range ozone variability and selected QBO-associated statistical relationships, without establishing a unique dynamical or chemical attribution.

Remote Sensing doi: 10.3390/rs18213762

Authors: Xingyuan Zhu Yang Cai Gaoxu Wang Dehua Zhu Kuan Liu Xuan Zhang Liyu Quan Ziwen Yin

Radar precipitation information with high spatial and temporal resolution is essential for flood forecasting, urban pluvial flood warning, and digital twin basins, yet radar quantitative precipitation estimation (QPE) is affected by observational errors, retrieval relationships, terrain effects, and multi-source merging. Based on 1328 Web of Science Core Collection publications (2000–2025), this study integrates bibliometric analysis, Transformer-based semantic representation, BERTopic modeling, semantic association analysis, and bibliographic coupling to map the topic structure, thematic evolution, and methodological lineages of research on radar precipitation uncertainty and hydrological applications. Nineteen micro-level topics were identified and grouped into four categories: measurement error identification and correction, uncertainty characterization and modeling, error propagation and scale effects, and operational applications and risk decision-making. The semantic structure shows that observation and correction and hydrological propagation constitute the main branches, while AI-driven forecasting is emerging. Methodological lineage analysis indicates that Statistical Fusion dominates the field, Physics provides a foundation, and AI and Hybrid methods have grown markedly in recent years, with Hybrid methods bridging physical mechanisms, statistical fusion, and data-driven approaches. Future research should integrate explainable AI, physical constraints, probabilistic modeling, and hydrological-response consistency testing to advance chain-wide uncertainty representation from observation and forecasting to hydrological propagation and risk-informed decision-making.

Remote Sensing doi: 10.3390/rs18213760

Authors: Wei Deng Changying Xue Haibo Chen Aifang Su Hongwei Tian Xiefei Zhi Yiling Qi Bingchen Li Jie Wu Mingjie Zhang Haijiang Kong Yanling Zhang

Remote-sensing-derived static terrain and land-cover datasets influence regional extreme heat simulations, but their effects on internal model responses and temperature performance require joint assessment. Five numerical experiments were conducted for the June 2022 Henan heatwave using five configurations drawn from three terrain datasets and four land-cover datasets. Atmospheric forcing and model physics were identical, and the inner grid spacing remained fixed at 3 km. Simulated skin temperature (TSK) was evaluated against Moderate Resolution Imaging Spectroradiometer (MODIS) land-surface temperature after pixel-level spatiotemporal matching. Simulated 2 m air temperature (T2) was evaluated against observations from 111 national meteorological stations. Two diagnostic pathways examined terrain-related and land-cover-related responses. Replacing MODIS15s-lake with MCD12Q1 changed the dominant land-use category in 22.2% of Henan grid cells. Daytime sensible heat flux difference fields had spatial root-mean-square (RMS) values exceeding 20 W/m2. On the WRF Preprocessing System (WPS)-mapped 3 km grid, terrain fields derived from ASTER30 and GMTED5m had mean absolute differences of 7.5 and 22.8 m, respectively, relative to GMTED30s. Terrain-height and surface-pressure differences were strongly inversely correlated (R = −0.9990 to −0.9993), supporting hydrostatic consistency. Direct interaction residuals indicated non-additive TSK and T2 responses, and their full-period mean fields had spatial RMS values of 0.421 and 0.225 K, respectively. All configurations exhibited pooled TSK warm biases of 5.52–5.74 K, compared with an inter-configuration bias spread of approximately 0.22 K. No configuration consistently minimized temperature errors across variables and periods, and larger internal response magnitudes did not consistently correspond to smaller temperature errors. For this event and model configuration, products with higher nominal native resolution did not consistently improve temperature simulation; the independent resolution effect was not isolated. The framework combines internal diagnostics with station and satellite evaluation to assess remote-sensing-derived static inputs.

Remote Sensing doi: 10.3390/rs18213761

Authors: Aris Kyparissis

Accurate assessment of crop phenology under changing climate conditions is essential for agricultural planning and food security, particularly in regions where wheat production is sensitive to temperature and water availability. This study examines the long-term phenological responses of wheat across six sites in Greece and five in Cyprus over a 24-year period (2001–2024), using MODIS-derived NDVI time series to extract start of season (SOS), end of season (EOS) and length of season (LOS). SOS advanced consistently in both countries, while EOS was delayed in Greece but advanced in Cyprus. As a result, LOS increased across Greek sites but showed mixed responses in Cyprus, with no significant average change. Spatial differences were associated with latitude, with higher-latitude Greek sites exhibiting later SOS and EOS, whereas temporal trends were linked to climatic variability. Linear mixed-model analyses identified temperature as the dominant climatic driver of SOS and EOS in both regions, with precipitation exerting additional, but region-specific influences. A key finding is that atmospheric CO2 emerged as an additional statistical predictor of EOS in Greece, where it was associated with later EOS, offsetting the climate-driven advance in the model, and was marginally associated with LOS extension. In contrast, no CO2 association was detected for Cyprus, possibly reflecting more constrained environmental conditions. These findings highlight the importance of integrating geographic, climatic and atmospheric variables to capture crop-specific responses and support region-specific adaptation strategies in Mediterranean agroecosystems vulnerable to climate change.

Remote Sensing doi: 10.3390/rs18213759

Authors: Piyapong Tongdeenok Naruemol Kaewjampa Surachit Waengsothorn Sitthisak Moukomla

Individual-tree segmentation of dense point clouds increasingly supplies the crown dimensions from which crown–stem allometry is fitted, yet no study has measured how much the fitted crown–stem exponent changes with what is done to material a segment holds but may not own. In two 1 ha plots spanning the open-to-closed canopy gradient of a Southeast Asian tropical dry forest, scanned with handheld SLAM LiDAR against a field stem map of 1320 trees, we voxelized each of 340 segments at 5 cm, split it into 26-connected components and computed crown projected area three times with identical code: on the delivered segment, after removing every component not physically continuous with the tree’s own bole (strict) and after removing only those components in which a field-mapped stem other than this tree falls inside the component’s own footprint (conservative). In the closed evergreen stand the strict treatment moved the exponent of crown area on stem diameter from 1.54 to 1.04 (paired-bootstrap difference −0.49, 95% CI −0.65 to −0.34). The conservative treatment moved it by −0.003 (−0.005 to −0.001), and its median removal was 0.000 of a segment’s points in both stands. Removal based on stem-footprint overlap therefore moves the exponent by two orders of magnitude less than removal of every disconnected component, a contrast that bounds the sensitivity without establishing the ownership of either group. The result holds with the crown base fixed at its delivered value, with the largest segments removed and on the segments whose crown base came from the primary detection rule (−0.372, −0.587 to −0.147). Its direction also holds across bole gates, crown-base thresholds, an alpha-shape crown area and connectivity voxels of 3 to 10 cm, although its magnitude scales with the voxel (−0.89 to −0.24). The crown–stem exponent obtained from segmented closed-canopy point clouds is therefore not determined by the point cloud alone, and we quantify how much of it the analyst decides.

Remote Sensing doi: 10.3390/rs18213758

Authors: Yang Zheng Zhikai Wang Ruimin Li Junling Fan Wenbo Jia

Enhancing the environmental generalization of Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) requires diverse target–clutter combinations. Despite abundant target-free clutter, exhaustively capturing multi-background target samples is prohibitive. However, existing generative methods either distort authentic clutter statistics through full-scene reconstruction or lack autonomous generation capabilities by strictly relying on pre-existing target instances. To overcome these limitations, we propose the Masked Inpainting and Residual Blending Network (MIRB-Net) for instance-free, background-preserving, multi-class coupled generation directly onto authentic measured backgrounds. Treating the background as a measured clutter carrier, MIRB-Net synthesizes class-conditional targets in situ using three synergistic mechanisms. First, a Background-Preserving Residual Editing (BPRE) mechanism explicitly decouples spatial blending from residual prediction, effectively confining generative perturbations to the target vicinity. Second, a self-supervised Masked Inpainting Pretraining (MIP) strategy specifically designed for BPRE equips the network with crucial priors of target scattering and background statistics. Third, joint global–local adversarial supervision alongside tri-region spatial constraints ensures seamless radiometric coupling across target–clutter boundaries. Evaluated under a dedicated multidimensional assessment framework on the MSTAR dataset, MIRB-Net demonstrates superior local structural fidelity, faithful background preservation, and natural target–clutter coupling. Furthermore, hybrid data-augmentation experiments confirm that the autonomously generated samples consistently improve downstream ATR performance across diverse deep learning architectures.

Remote Sensing doi: 10.3390/rs18213757

Authors: Xinyu Chen Yongfeng Zhu Jianxiong Zhou Yibin Liu Zezhou Wu Pengcheng Guo

Accurate target prior information is an important basis for improving the detection and recognition performance of high-resolution range profiles (HRRPs). To address the insufficient reliability of prior templates in scattering-center-assisted HRRP target recognition based on SAR images, where scattering centers extracted from a single image are susceptible to noise, sidelobes, local scattering fluctuations, and variations in observation azimuth angle, this paper proposes an HRRP target recognition method based on the fusion of scattering-center priors derived from SAR images acquired at adjacent azimuth angles. Two-dimensional scattering centers at different azimuth angles are extracted by combining an adaptive point-spread function (PSF) with the two-dimensional CLEAN algorithm. Amplitude-weighted clustering is then employed to retain stable scattering structures with spatial consistency, thereby constructing a fused scattering center model. Subsequently, the fused model is projected onto the range dimension under different azimuth offsets to establish a one-dimensional scattering center prior template library. Target classification is finally performed using sliding matching based on the maximum correlation coefficient. Same-band, cross-band, and comparative experiments conducted with ground-target data sets demonstrate that the proposed method effectively improves the stability of the prior templates and target recognition performance, while exhibiting reasonable adaptability to variations in observation azimuth angle, noise level, and radar operating frequency band.

Remote Sensing doi: 10.3390/rs18213756

Authors: Wilian H. Vargas-Farfan Abner S. Rivera-Fernandez Jhon A. Zabaleta-Santisteban Angel J. Medina-Medina Martha R. Pacheco-Huamani Katerin M. Tuesta-Trauco Alexander Cotrina-Sanchez Swati Suman Teodoro B. Silva-Melendez Elgar Barboza

Flood monitoring in Andean–Amazonian valleys is constrained by sparse hydrological records and persistent cloud cover. This study mapped observed flood recurrence in the Utcubamba River Valley, Peru, during 2019–2025 using Sentinel-1 imagery processed in Google Earth Engine. PRE–POST backscatter differences were classified using Otsu thresholding and spatial constraints, including a maximum slope of 5° and a 1000 m hydrographic buffer. Eight of 13 cataloged events were retained. VH thresholds had a mean ± standard deviation of −2.83 ± 0.05 dB, compared with −1.73 ± 1.11 dB for VV and −3.33 ± 1.44 dB for (VH+VV)2; the low VH dispersion is a property of the retained subset, which was selected partly on VH threshold behavior. The eight-event map covered 9842.02 ha detected as flooded at least once, with 6.29% classified as high or very high observed recurrence, concentrated near the Utcubamba–Marañón confluence. For E13, assessment against a balanced sample of 1360 reference points yielded a sample-based overall accuracy of 85.15%, flooded-class user’s accuracy of 91.64%, producer’s accuracy of 77.35%, and F1 score of 0.839. These metrics characterize the sampled sectors of E13. Agricultural classes accounted for 73.70% of the area in the 2024 land-cover overlay. Irrigation-related changes could not be fully distinguished from river flooding. The resulting count of detections across cataloged events provides complementary information for land-use planning, with interpretation bounded by temporal coverage, event selection, and reference-data availability.

Remote Sensing doi: 10.3390/rs18213755

Authors: Cristina Soares Sofia Cerasoli João M. N. Silva

Satellite-based characterization of vegetation phenology in structurally complex Mediterranean ecosystems remains challenging because of the contrasting spatial and temporal properties of remote sensing platforms. This study evaluates how differences between sensors, including spatial and temporal resolution, influence phenological parameter extraction from NDVI time series in a cork oak woodland in central Portugal. NDVI time series from Landsat 8/9, Sentinel-2, and a ground-based spectrometer (FloX) were processed in TIMESAT to extract six phenological metrics across three vegetation types (cork oak, herbaceous vegetation, and open woodland) over three growing seasons (2022–2025). Landsat 8/9 and Sentinel-2 agreed closely for herbaceous vegetation and open woodland, whereas differences concentrated in cork oak, where Sentinel-2 estimated a start of season 161 days earlier in Year 1 and an end of season 33 to 61 days later in Years 2 and 3. Sentinel-2 retrieved higher amplitude and seasonal integral values in all vegetation types and years. Compared with FloX, satellite end of season and peak timing for cork oak were later in all years (by 72–198 and 87–203 days, respectively), whereas deviations for herbaceous vegetation were 15 days or less. These findings support multi-scale phenological characterization of Mediterranean montados and offer a baseline for future multisensor assessments of dry-land ecosystems.

Remote Sensing doi: 10.3390/rs18213754

Authors: Xinwei Sun Kang Xu Huafu Pei Bin Pan Yongheng Tang

Unmanned aerial vehicle (UAV) optical remote sensing provides a flexible means of mapping and inspecting power-transmission corridors. However, the automatic extraction of power lines from aerial imagery remains challenging because conductors have slender geometry, weak texture, low contrast, and discontinuous visibility and are easily affected by imaging noise, occlusion, and complex background interference. Consequently, segmentation-based methods often produce fragmented masks that limit the reliability of subsequent structural reconstruction. This study presents a power line detection and reconstruction framework based on global keypoint regression for UAV optical imagery. Rather than recursively associating adjacent keypoints, the framework regresses every keypoint belonging to the same power line toward a common start point. This shared reference is designed to avoid recursive propagation through a chain of local associations. Three complementary components are incorporated: a multi-scale region-aware feature extractor (MSRAFE) for combining local and global semantics across scales, a line feature aggregation (LFA) module for adaptive multi-level feature fusion, and a linear structure-aware feature strategy (LSAFS) for coupling local geometric cues with global structural relationships. The method was evaluated on two public aerial-image benchmarks and a self-constructed dataset. It achieved intersection-over-union (IoU) and F1 scores of 85.76% and 91.76% on PLDU and 75.23% and 83.95% on TTPLA, respectively. On the self-constructed CPLD dataset, it obtained the highest IoU (70.44%) and precision (83.87%) among the compared methods. These pixel-level results support the performance of the complete framework for detecting thin linear objects in airborne optical remote sensing imagery.

Remote Sensing doi: 10.3390/rs18213753

Authors: Wenrui Qi Chen Chen Yang Zhou Yuhang Bai Qingfeng Dong Shuqiang Wang

Ground clutter persists as residual stationary or recurrent echo in gridded reflectivity-only weather radar products, then is converted into spurious light rainfall when reflectivity is mapped to rain rate. This study presents an image-level quality control filter for Météo-France MeteoNet composites from northwestern France. Trained annotators labeled 346,629 pixels from clear-air and weather windows; complete events were withheld from training during validation. Intensity, spatial, temporal, and object features were combined in a random forest (RF) that separated clutter from weather echo on unseen events, reaching F1 = 0.833 with weather echo preservation of 0.952 and clutter suppression of 0.763. Choosing a 95% preservation operating point from training events alone maintained 0.953 preservation on outer test events; a gradient-boosted decision tree model achieved comparable discrimination. Rain gauges supported period screening, station-level recurrence review, and paired 30-min quantitative precipitation estimation (QPE). At nine reviewed recurrent-echo stations, targeted RF gating reduced false alarm ratio from 0.277 to 0.211 and mean absolute error from 0.0549 to 0.0526 mm by filtering the light rain echoes that produced false precipitation, while retaining most genuine rainfall. A mixed-echo analysis preserved 0.961 of overlapping labeled pixels, and a model trained only on 2016 data transferred to a 2017 test day with reduced error at the same stations. The resulting framework filters false precipitation effectively at locations with recurrent clutter, providing an auditable event-grouped workflow for quality control of reflectivity-only radar imagery.

Remote Sensing doi: 10.3390/rs18213752

Authors: Lina Wang Yanan Gu Xin Chen Huaidan Liang Guangfeng Shi

Optical and synthetic aperture radar (SAR) image matching is fundamental to numerous remote sensing applications. Although phase congruency-based methods have been widely adopted for this task, the structural features they extract remain inherently sensitive to speckle noise, which significantly degrades matching performance. To address this limitation, we propose a novel noise-robust multiscale structural feature (NMSF) method. Unlike existing relative total variation (RTV)-based approaches that modify the RTV model or rely on phase congruency, our framework employs the standard RTV model with a principled exponential scaling scheme to construct multiscale structural maps (MSM), effectively capturing salient structural information across multiple scales while suppressing noise and mitigating modality discrepancies. Building upon the MSM, we develop an MSM-ORB detector with non-maximal suppression to extract reliable and well-distributed keypoints. We further introduce a multilayer features of oriented gradient (MFOG) descriptor with a square grid nested architecture, which fundamentally differs from the log-polar structures used in prior methods and achieves multiscale integration without dimensional inflation. Extensive experiments on publicly available datasets demonstrate that NMSF consistently outperforms state-of-the-art approaches, delivering superior robustness and accuracy under nonlinear intensity variations, geometric distortions, and severe speckle noise.

Remote Sensing doi: 10.3390/rs18213751

Authors: Bo Wang Xina Cheng Zhiqi Hu Yuan Li Xu Tang Xiangrong Zhang

Accurate aircraft-region segmentation in aerial infrared imagery can provide a useful scene-understanding component for image-assisted aviation inspection. However, target scale variation and ambiguous boundaries remain challenging in simulated aerial infrared scenes. To address these issues, this paper develops a bidirectional collaborative framework that couples global semantic fusion with edge optimization. The Semantics-guided Multi-scale Context Fusion branch uses adaptive multi-scale feature extraction and cross-stage calibration to produce a reliable initial segmentation, while the Edge-Sensitive Superpixel Collaborative Refinement branch improves boundary localization through semantic-guided superpixel consistency. Their interaction forms a bidirectional closed loop in which semantic context guides edge refinement and refined boundaries feed back to semantic learning. Experiments on AISD, a dedicated physics-based simulated aerial infrared dataset, achieve an 86.29% mPA, 83.35% mF1, and 76.20% mIoU. Relative to the strongest baseline, P-Mamba, the gains are 0.89 percentage points in mPA, 1.52 points in mF1, and 0.88 points in mIoU, and a sourcewise holdout comparison indicates consistent performance under unseen source content within the AISD simulation pipeline. These results concern aircraft-region segmentation rather than direct diagnosis of thermal defects.

Remote Sensing doi: 10.3390/rs18213750

Authors: Jiayi Guo Shufeng Li Haoxing Zhang Tao He Wei Yang Hongcheng Zeng Yamin Wang

Squinted multi-channel spaceborne synthetic aperture radar (SAR) offers wide-swath coverage and flexible observation geometry. However, imaging algorithms based on chirp scaling (CS) may broaden the range spectrum beyond the available sampling bandwidth, causing range spectrum aliasing and degrading the imaging quality of targets at the range edges of wide-swath scenes. To address this problem, this paper proposes a range-subband imaging method. First, a range spectrum broadening model is established to identify the main factors affecting spectral broadening. A subband division criterion is then derived to determine the minimum number of subbands required to avoid aliasing. Finally, a processing procedure comprising range-subband division, individual subband focusing, and coherent fusion is developed to recover the full-band range-focused signal. Point-target and distributed-target simulations verify that the proposed method suppresses range-spectrum aliasing and preserves full-band range resolution, while a broadside airborne experiment confirms that it introduces no appreciable image-quality degradation.

Remote Sensing doi: 10.3390/rs18213749

Authors: Davaajargal Myagmarsuren Haibin Wu Aili Wang Minhui Wang

This study presents a reliability-aware feedback-controlled fusion framework for hyperspectral image (HSI) and LiDAR classification under spatially disjoint conditions. Instead of committing to a fixed fusion configuration, the framework couples a library of fusion operators (geometry-aware fusion, adaptive gating, cross-modal transformer fusion and physics-informed fusion), integrated within a unified GhostConv-based architecture, with a closed-loop controller that evaluates candidate operators on validation data, accepts them only when they improve validation accuracy, and applies feedback-guided residual refinement. Experiments are conducted on Houston2013, Muufl, and Augsburg under a sample-center spatially disjoint protocol, in which training and validation samples are drawn from development blocks that are spatially disjoint from the test blocks, with seven labeled samples per class and ten random partitions. The framework achieves mean overall accuracies of 70.00%, 73.20%, and 66.34%, improving the Baseline-Ghost model on Houston2013, Muufl, and Augsburg; the improvement achieved by closed-loop training is preserved under a patch-disjoint protocol. Budget-matched analyses show that, with seven labels per class, the gain stems mainly from validation-gated closed-loop training and feedback-guided refinement, whereas the benefit of adaptive operator switching emerges as the label budget increases.

Remote Sensing doi: 10.3390/rs18213748

Authors: Can Su Wei Yang Hongcheng Zeng Xinwei An Danni Jiang Zhengqi Wen Yikui Zhai Xiaolin Tian Jie Chen

One-bit quantization technology effectively simplifies the hardware architecture of synthetic aperture radar (SAR) and significantly reduces the volume of echo data, making it a pivotal research direction in spaceborne SAR systems. However, the high computational cost and latency associated with one-bit SAR signal reconstruction have limited the efficiency of downstream target detection applications. To address the demand for efficient ship target detection in one-bit SAR imagery, a label self-supervised ship detection training method on one-bit non-reconstructed SAR images is proposed in this paper, which enables generic deep learning models to perform detection directly in one-bit SAR images based on the chirp scaling algorithm (CSA) without signal recovery during inference. First, a training framework based on curriculum learning is designed, aiming to discriminate between real ship targets and false targets. Second, a one-bit reconstruction and training label self-supervised correction branch is incorporated into the training framework to autonomously generate false target labels and progressively guide the model to learn false target features. Third, a one-bit non-reconstructed image dataset is proposed to validate the trained model, which maps only reliable ship target annotations from full-precision echo imaging results. Finally, experiments conducted on real SAR data demonstrate that the detection model derived from the proposed training method can directly detect ship targets in the non-reconstructed images. Compared with the classic “reconstruction-and-detection” workflow, the trained model runs inference, bypassing the reconstruction process, and achieves a 20-fold improvement in efficiency without compromising detection accuracy. This method provides a practical solution for the rapid acquisition of target information in large-scale ship monitoring using one-bit SAR.

Remote Sensing doi: 10.3390/rs18213747

Authors: Ya Li Yiqi Zhu Hanwei Liang Sophia Shuang Chen Liang Dong

Existing environmentally adjusted development indices can assign similar scores to regions with contrasting environmental–social configurations, while provincial averages conceal subprovincial variation. We developed a Coupling-Embedded Sustainable Development Index (CESDI), which retains the Sustainable Development Index (SDI) as baseline performance while incorporating an adjustment for the joint balance and overall development level of the environmental and social subsystems. CESDI was applied to 30 Chinese provincial-level units from 2002 to 2022. Exploratory predictions for 337 Chinese cities were generated using MOSAIKS-inspired Landsat features, nighttime lights, and dual-penalty ridge regression. CESDI preserved the interprovincial SDI gradient (Pearson r = 0.870 to 0.995) while differentiating provinces with similar SDI values but contrasting joint environmental–social conditions. National mean CESDI increased from 0.248 in 2002 to 0.465 in 2017 and declined to 0.371 in 2022. Nested province-grouped cross-validation yielded R2 = 0.4471 and RMSE = 0.1388. Feature-block ablation showed that the dual-penalty model outperformed daytime-only, nighttime-light-only, and equal-penalty specifications. The predicted share of high and very high classes increased from 12.5% to 37.1%, while within-province heterogeneity rose in 21 of 26 multi-unit provinces. The framework integrates baseline sustainability performance with a CCD-defined balance-and-level adjustment and provides an exploratory spatial-screening layer for identifying within-province variation.

Remote Sensing doi: 10.3390/rs18213746

Authors: Ran Zhang Kaihong Guo

High-resolution remote sensing images exhibit large variations in object scale, ambiguous boundary transitions, and complex interactions among adjacent land-cover regions. Under sparse pixel supervision, these factors may cause object fragmentation, erroneous merging, and boundary displacement. To address these issues, a Dual-Domain Coupled 2-Additive Fuzzy Measure Network (DC2CNet) is proposed for sparse pixel-supervised remote sensing image segmentation. The proposed method extends pixel-level class prediction by jointly modeling regional consistency and boundary discrimination. A region-domain branch and a boundary-domain branch are constructed to characterize intra-region coherence and class-transition information, respectively. A dual-domain coupling mechanism is further introduced to integrate regional and boundary evidence. In addition, a 2-additive fuzzy measure decoding strategy is developed to model individual evidence contributions and pairwise evidence interactions through Möbius representation. During training, sparse pixel annotations are used to generate boundary discrimination probabilities, and neighborhood consistency constraints are introduced to optimize the boundary branch. Experimental results demonstrate that DC2CNet effectively reduces boundary ambiguity and region confusion, while maintaining accurate segmentation of small objects, elongated structures, and complex interface regions.

Remote Sensing doi: 10.3390/rs18213745

Authors: Marián Marčiš Károly Bence Póka Adam Young Marek Fraštia

The accuracy of 3D digitisation methods is strongly influenced by the surface properties of the observed object, particularly its texture, reflectance, and material structure. This is especially relevant to the systematic metrological analysis of ancient stone vessels undertaken by the Artifact Research Foundation, where high-resolution 3D models are increasingly used for geometric analysis, including through dedicated tools such as Petrie Scope. This study assesses the potential influence of crystalline granite texture on residual dispersion in Structure-from-Motion (SfM) photogrammetry and handheld laser-triangulation scanning using granite metrological reference objects measured under controlled conditions before and after application of a self-sublimating scanning spray. For the unsprayed cube, the pooled empirical 99% residual interval was −33.8 to +33.1 µm for photogrammetry and −49.9 to +54.5 µm for handheld scanning; for the unsprayed cylinder, the corresponding intervals were −51.6 to +30.7 µm and −76.3 to +55.3 µm. Sprayed reconstructions in the principal comparison exhibited lower residual dispersion and circularity RMSD. Repeated handheld-scanner acquisitions supported this trend, while supplementary SfM analyses demonstrated the influence of camera-network configuration and calibration treatment on surface-residual dispersion. By linking surface residuals to circularity and centre-offset descriptors, this study provides an empirical basis for interpreting geometric regularity in ancient stone vessels. The results for the tested granite surfaces and workflows highlight the need to distinguish measurement-induced deviations from genuine manufacturing irregularities when analysing polished stone artefacts and ancient vessels.

Remote Sensing doi: 10.3390/rs18213744

Authors: Valerio Pampanoni Emilio Chuvieco Alvise Ferrari Giovanni Laneve Simone Saquella

Live fuel moisture content (LFMC) influences vegetation flammability, but field measurements are too sparse to describe its variation over large areas. We present, to our knowledge, the first LFMC framework that combines observations from multiple optical satellite sensors within the same trained models. More than two decades of GlobeLFMC 2.0 measurements are matched to Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS), Visible Infrared Imaging Radiometer Suite (VIIRS), Landsat 8/9, Sentinel-2 and Sentinel-3. We screen the vegetation surrounding sampling locations and adjust reflectance for differences between sensors before calculating vegetation indices. Weather, topography and time of year provide additional information. Separate models are trained for grasses, shrubs and trees. We test prediction on dates not used for training at sites where other dates contribute to training. At each site, records using the same satellite image are kept in either the training or the test set. The final Random Forest models achieve R2 values of 0.715, 0.693 and 0.691, respectively. Models trained on the combined archive give higher R2 and lower root mean square error than models trained on one satellite product in all 18 product–vegetation comparisons on identical test observations. Paired 95% confidence intervals support the improvement in 16 comparisons. Testing at sites excluded from training gives lower accuracy. The method produces LFMC maps at the selected satellite product’s resolution, from 20–30 m to 300–500 m.

Remote Sensing doi: 10.3390/rs18213743

Authors: Guoqing Zhang Xiaosan Ge Lin Zhou Xiudong Qi Qinrui Zhu

Unmanned aerial vehicle (UAV) remote sensing provides a flexible and efficient approach to early forest fire monitoring and warning. However, small fire spots and smoke plumes in UAV imagery vary substantially in texture, morphology, and spatial distribution across scales. Interactions between complex forest backgrounds and target features cause information loss during successive feature compression and misaligned spatial and semantic responses during cross-layer fusion. Limited onboard computing and storage further constrain model complexity. To address these challenges, we propose the lightweight feature alignment and multi-scale enhancement network (FAME-Net), which alleviates feature degradation and response mismatches during encoding and fusion while reducing computational redundancy in multi-scale prediction. At the feature encoding stage, the lightweight multi-scale enhanced backbone (LMEB) combines lightweight downsampling with multi-scale context modeling to reduce computational overhead and preserve fine details of small fire spots and weakly textured smoke. Its hierarchical features are fed into the alignment-aware multi-scale feature fusion neck (AMFF-Neck), which coordinates spatial and semantic responses across scales through cross-layer interactions, reducing fusion bias caused by scale discrepancies and background-induced coupling. The resulting features are processed by the lightweight shared convolution detection head (LSCD), which performs scale alignment, shared convolution, and learnable scale compensation for efficient multi-scale prediction. FAME-Net achieves mAP50 values of 71.5% and 65.3% on the integrated forest fire dataset and the Yuxi Wildfire dataset, respectively, with 1.7 M parameters and 3.9 GFLOPs, indicating a favorable accuracy–efficiency trade-off. On an NVIDIA Jetson Orin Nano Super (8 GB), it achieves inference speeds of 36.2 and 31.6 FPS on the two datasets, respectively, further supporting its potential for resource-constrained UAV deployment.

Remote Sensing doi: 10.3390/rs18213742

Authors: Mengyan Ren Dongyou Zhang Qi Shen Zhaopeng Wang Xiangyou Li Shuting Yang Jianfei Cao Xibo Xu

Monitoring soil total nitrogen (TN) status and spatial variation is essential for preventing land degradation and supporting precision agriculture. Remote sensing (RS) bands and their derived spectral indices are important covariates for soil TN mapping. However, optical RS-based soil mapping may be ineffective in fog-prone regions because water vapor interferes with spectral information. Here, multi-source earth observation data were acquired, including synthetic aperture radar (SAR), multispectral, and topographic data. Spectral indices, SAR-derived indices, and topographic indices related to soil TN were systematically developed. Subsequently, the PCC-VIF method was used to assess the importance of RS-based indices and select the optimal covariates, referred to as SST covariates. Finally, SST covariates and measured soil TN values were used as the independent and dependent variables, respectively, to construct an XGBoost-based soil TN mapping model for regions with persistent foggy cover. The results showed that: (1) SST covariates included Ela, TWI, SOS, VHplusVV, VHdiffVV, B2, and NDGI; (2) the proposed RS-based soil TN mapping model achieved satisfactory accuracy, with an R2 of 0.82 and an RPIQ of 3.03. Compared with models developed using traditional approaches, the R2 increased by approximately 0.16–0.52; and (3) areas with high soil TN content were mainly located in the southern part of the study area, whereas lower TN values occurred in the northern part. Soil TN content and its spatial distribution were mainly influenced by agricultural planting activities and microtopographic conditions. This study provides new insights into RS-based soil TN mapping in fog-prone regions. The robust soil TN mapping framework based on SST covariates and the XGBoost algorithm can support food security and help prevent environmental degradation.

Remote Sensing doi: 10.3390/rs18213741

Authors: Yu Shi Fei Yang Asen Grytsai Diana Zazubyk Gennadi Milinevsky

Sudden Stratospheric Warmings (SSWs) are rare polar stratospheric extremes far less documented over Antarctica than the Arctic. A pair of successive midwinter Antarctic SSWs occurred in July–August 2024, featuring severe polar vortex distortion and sharp stratospheric temperature rises. Using MERRA-2/ERA5 reanalysis and MSR-2 total ozone column data, we diagnose wave 1 and wave 2 planetary wave forcing and associated ozone variations. Abnormally strong upward-propagating Southern Hemisphere tropospheric planetary waves drove the 2024 SSW. Wave 1 triggered the first July warming peak, while amplified wave 2 dominated the second intense vortex disruption in early August. These waves delivered westward momentum that weakened the polar night jet and destabilized the vortex. The ozone redistribution is consistent with an inferred enhancement of the Brewer–Dobson circulation (BDC) driven by the amplified wave activity, raising polar ozone levels and supplying extra stratospheric heat. Wave–ozone coupling shows clear selectivity: wave 1 is associated with negative total ozone column anomalies across 0–50°S (low-to-midlatitudes), whereas wave 2 is positively correlated with total ozone column anomalies over the 60–70°S polar collar region. Fading planetary wave activity after the warming peaks gradually restored the polar vortex and stratospheric thermal structure. This study provides key observational evidence for rare Antarctic SSWs and advances understanding of stratosphere–troposphere coupling and polar ozone dynamics under global warming.

Remote Sensing doi: 10.3390/rs18213739

Authors: Carlos Guerrero-Cubero Anabel Calzado Carretero Cristina Pérez-Carral Reyes Alejano-Monge Javier Vázquez-Piqué

Old-growth forests (OGFs) are conservation priorities under the EU Biodiversity Strategy for 2030, but identifying relict stands in Mediterranean Pinus nigra (black pine) forests is challenging due to structural complexity. We evaluated a multisensor remote sensing framework to estimate an Old-Growth Index (OGI) in Cazorla Natural Park (Spain) using high-density Airborne Laser Scanning (ALS) data from the Spanish National Plan for Aerial Orthophotography (PNOA-LiDAR, 2024), RGB–NIR orthophotographs, and multispectral imagery (Sentinel-2 and WorldView-3). Four predictor sets (ALS metrics only; ALS plus canopy cover fraction (CCF); ALS + CCF + Sentinel-2; and ALS + CCF + WorldView-3) were evaluated across Multiple Linear Regression, Random Forest, and Support Vector Regression (SVR) using repeated nested cross-validation. SVR combined with the predictor set ALS + CCF + WorldView-3 achieved the highest nested R2 (0.79), outperforming Random Forest and Multiple Linear Regression. However, models relying solely on ALS (R2 = 0.71) or ALS with CCF (R2 = 0.77) yielded statistically comparable performance. Furthermore, integrating Sentinel-2 (R2 = 0.61–0.68) provided no clear gain over ALS-based models. These findings demonstrate that open-access national LiDAR programs provide sufficient 3D structural information to map stand maturity without requiring commercial satellite imagery. Continuous spatial predictions of OGI offer valuable baseline maps to support conservation planning and OGF monitoring in Mediterranean pine ecosystems.

Remote Sensing doi: 10.3390/rs18213737

Authors: Muhammad Azeem Sadiq Philip Kibet Langat Arjun Neupane

Emerging remote sensing technologies increasingly support coastal observation through Earth observation (EO) derivatives such as satellite-derived shorelines, unmanned aerial vehicle (UAV)- and light detection and ranging (LiDAR)-based elevation models, coastal land-use/land-cover products, bathymetric grids, turbidity and chlorophyll-a indicators, habitat maps, flood extents, and artificial intelligence (AI)-generated segmentation outputs. These EO-derived products are central to shoreline monitoring, coastal vulnerability assessment, flood-response planning, marine ecosystem management, and land–sea administration. Yet their evidential value depends on transparent provenance: users must know which sensor, acquisition condition, preprocessing chain, algorithm, model version, validation dataset, and uncertainty estimate produced each coastal product. The review synthesises 156 eligible publications identified through a systematic review process. Of these, 89 studies containing sufficient implementation and methodological detail were selected for detailed analytical coding. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol, the review analyses application domains, provenance capture modes, representation strategies, semantic web technologies, and interoperability practices. A coastal EO thematic refinement translates the review findings into provenance requirements for satellite, UAV, LiDAR, Internet of Things (IoT), open-water, and AI-enabled coastal observation workflows. The review identifies persistent gaps in feature-level lineage, alignment between the World Wide Web Consortium Provenance Ontology (PROV-O) and International Organization for Standardization (ISO) geospatial metadata standards, spatial semantics, AI workflow transparency, uncertainty traceability, and scalable querying. To address these limitations, the paper introduces GeoPROV, a lightweight semantic profile of PROV-O designed for spatially explicit, feature-level provenance. GeoPROV aligns PROV-O with ISO 19115/19157, GeoSPARQL, Shapes Constraint Language (SHACL), and SpatioTemporal Asset Catalogue (STAC)-compatible catalogue practices. The framework enables EO-derived coastal products to be traced from raw observation through preprocessing, AI/geoprocessing, validation, catalogue publication, and coastal-administrative decision-making. The resulting GeoPROV profile integrates PROV-O, GeoSPARQL, SHACL, ISO 19115/19157 lineage concepts, and STAC metadata within a lightweight semantic framework demonstrated through a coastal EO product workflow and machine-actionable provenance implementation.

Remote Sensing doi: 10.3390/rs18213740

Authors: Peixin Cai Bingxin Liu Xiaoyang Li Xinhao Li Siqi Wang Peng Liu Peng Chen Ying Li

The Journal retracts the article “Detecting Ship-to-Ship Transfer by MOSA: Multi-Source Observation Framework with SAR and AIS” [...]

Remote Sensing doi: 10.3390/rs18213738

Authors: Yiqi Wang Yang Du

This study evaluates whether the Whittle–Matérn (WM) covariance representation improves pixel-wise SAR inversion over bare agricultural soil which can exhibit scale-dependent roughness. WM includes a smoothness parameter that controls covariance and roughness spectrum shape and has a Gaussian–Gamma scale-mixture interpretation. It was compared with exponential, Gaussian, and powered exponential (PE) covariance models within the same modified extended advanced integral equation model. Co-registered RADARSAT-2 and Sentinel-1 observations provided five channels for inversion. All experiments used identical observations, ancillary inputs, parameter bounds, objective function, and multi-start optimization settings. The analysis focused on the three co-polarized channels, for which covariance structure enters the scattering model directly. At a co-polarized residual tolerance of 2dB, WM produced model-consistent solutions for 425 of 439 retained pixels, compared with 384, 387, and 423 pixels for the exponential, Gaussian, and PE models, respectively. Its blind point ratio was 3.19%, versus 12.53%, 11.85%, and 3.64%, respectively. On the 342 pixels retrievable by all four models, WM gave the lowest mean absolute error for all channels except Sentinel-1 VH. The largest improvement was obtained for Sentinel-1 VV, where WM reduced RMSE to 0.501dB, compared with 0.820dB, 0.786dB, and 0.606dB for the exponential, Gaussian, and PE models. WM also yielded the most complete retrieval maps and scene mean soil moisture estimates closest to the available SMAP and station reference means. These results show that covariance representation is a material source of structural difference in unregularized multi-sensor SAR inversion.

Remote Sensing doi: 10.3390/rs18213736

Authors: Ting Li Chuping Fu Chaokui Li Hanwu Yu Ping Zhang Ling Jiang Jia Fan

Ecological changes in mining areas are characterized by long-term dynamics, temporal stages, and spatial heterogeneity. Existing monitoring methods often face challenges in balancing multidimensional ecological condition characterization with spatially consistent analysis and have limited capability to identify complex disturbance–recovery trajectories. To address these issues, an Object-based Ecological Index-driven Framework (OEI-GTCD) is proposed to monitor ecological disturbance and recovery trajectories in mining areas. The framework uses the Mining-area Ecological Index (MEI) to characterize multidimensional ecological conditions and conducts analysis based on spatially consistent object units. On this basis, a Globally Optimized Time-series Change Detection method (GTCD) is proposed to improve the identification of complex ecological disturbance–recovery trajectories by integrating trend-change detection, abrupt-change detection, and segmentation optimization. Landsat imagery from 2000 to 2024 was used to characterize the spatiotemporal patterns of ecological disturbance and recovery across the Daye Mining Area, China. The findings suggest that MEI captures multiple dimensions of ecological conditions, while object-scale analysis characterizes ecological changes through overall object-level trajectories, supporting their spatial interpretation. OEI-GTCD achieved disturbance and recovery detection accuracies of 88.4% and 84.8%, respectively, representing improvements of 6.0 and 5.2 percentage points over OEI-LandTrendr. Ablation analyses further highlighted the contributions of abrupt-change detection and the improved segmentation criterion to event identification, as well as the role of local breakpoint refinement in temporal localization. From 2000 to 2024, stable and continuously improving ecological states together accounted for 82.07% of the area represented by the non-water objects used in the ecological analysis, while disturbance and recovery processes exhibited pronounced spatiotemporal heterogeneity. Overall, OEI-GTCD provides an integrated approach for monitoring ecological disturbance and recovery and characterizing their spatiotemporal patterns in mining areas.

Remote Sensing doi: 10.3390/rs18213735

Authors: Jiaqi Li Qinghua Zeng Zhi Xiong Songnan Duan Junjie Wu Yudi Sun

Small-object detection in UAV imagery remains challenging because repeated downsampling weakens fine structural cues, conventional feature pyramids are poorly aligned with small-object-dominated scale distributions, and fixed detection heads provide limited adaptation. To address these issues, a Hierarchical–Bidirectional–Dynamic Network (HBD-Net) is developed as a parameter-compact detector based on YOLOv11s. HEEM reinforces edge, contour, and texture information at multiple backbone stages using orthogonal depthwise strip convolutions, residual detail extraction, and zero-initialized fusion. PLSS replaces the conventional P3–P5 hierarchy with P2–P4 to emphasize high-resolution features, while P2-BiFPN introduces bidirectional propagation and learnable weighted fusion. An integrated Dynamic Head further refines features across scale, spatial, and task dimensions. On VisDrone, HBD-Net achieves 39.0% mAP50 and 22.6% mAP50–95, improving the YOLOv11s baseline by 6.9 and 4.3 percentage points, respectively, while reducing parameters from 9.4 M to 3.7 M. Ablation studies confirm the complementary effects of the components. On HIT-UAV, HBD-Net reaches 84.1% mAP50 and 57.0% mAP50–95, indicating effectiveness under infrared aerial imaging. These results demonstrate that coordinated structural enhancement, high-resolution multiscale fusion, and adaptive prediction can improve UAV small-object detection with a compact parameter footprint.

Remote Sensing doi: 10.3390/rs18213734

Authors: Zekun Ding Xiangzhi Kong Yanhu Mu Minghao Liu Zhuang Wang Lingjie Li Lijie Zhang Wudong Wang

Mountain vegetation change is commonly summarized as a mean greening or browning trend, or as the displacement of a discrete boundary; both can miss changes in the shape of the elevation profile of vegetation greenness. Here we tested whether the greenness profile on the southern flank of the Tianshan Mountains reorganized between 1989 and 2024, using a Landsat transect along the Dushanzi–Kuqa (G217) highway. NDVI and EVI2 profiles were constructed over a fixed 1000–4100 m elevation domain using only sampling locations valid in all comparison periods; profiles were mean-centered, and the early-to-recent contrast was tested with spatial-block permutations across 36 combinations of temporal window, corridor width, elevation-bin width, and index. All 36 combinations showed significant shape change (p ≤ 0.005) with cross-scenario concordance, and the contrast persisted after adjustment for cross-sensor differences. The contrast followed a consistent three-band structure, with relative decreases at 1100–1950 m, increases at 2550–3300 m, and renewed decreases at 3550–4100 m. Absolute greenness increased almost everywhere (more than 97% of 25,334 samples), but the increase was four to five times larger in the mid-elevation band than in the lower and upper bands, varied eightfold along the transect, and was organized primarily by elevation and along-transect position (spatially cross-validated random-forest R2 ≈ 0.66). The vegetation contrast was not directionally consistent with decadal temperature, soil-moisture, or snow contrasts. Greenness change along this transect was therefore an elevation-dependent redistribution of the whole profile rather than uniform greening or a single boundary displacement, and no robust directional correspondence with decadal hydroclimate contrasts was detected under the present decadal analysis.

Remote Sensing doi: 10.3390/rs18213733

Authors: David J. Durden Tristan Goulden Michael SanClements Cedric J. Hagen Michael Wussow John Adler Joshua Roberti

This study evaluates the accuracy and usability of two low-cost uncrewed aerial system (UAS) multispectral sensors—MicaSense RedEdge-M and SlantRange 3p—using a dual-scale comparison framework. Calibrated 3% and 48% reflectance tarps and Analytical Spectral Device (ASD) field spectrometry were used for point-scale absolute validation. The NEON imaging spectrometer (NIS), a well-calibrated airborne hyperspectral sensor, was utilized for a field-scale comparison by spectrally resampling the hyperspectral data to align with the multispectral bands and performing spatial co-registration and aggregation of the multispectral data to a common 1 m grid. Over the 48% tarp, MicaSense Near-infrared (NIR) reflectance was 7.2% below the ASD reference while SlantRange NIR was 7.9% lower, validating previous studies that have observed the underestimation of the NIR band with UAS multispectral sensors. Both sensors overestimated the 3% tarp across all bands when compared to the ASD measurements. Field-scale regressions yielded R2 values above 0.90 for most bands, while NIR slopes were 0.74 for MicaSense and 0.72 for SlantRange. Consistent with these trends, the largest radiometric errors occurred in the NIR region, where both sensors increasingly underestimated high-reflectance surfaces relative to NIS. Median Normalized Difference Vegetation Index (NDVI) differences relative to NIS were −0.027 and −0.064, and these became more negative with an increasing NIS NDVI. Edge buffering reduced MicaSense bias, whereas the SlantRange showed limited edge effects. These results demonstrate that low-cost UAS multispectral imagery can support ecological applications when calibration, reflectance range, view geometry, and ecosystem-specific validation requirements are explicitly considered.

Remote Sensing doi: 10.3390/rs18213731

Authors: Xu Wang Jiahao Zhang Tongxu Hu Jialu Liu Jin Li Rui Min Zongjie Cao Yiming Pi

THz-SAR provides high-resolution imaging but is highly sensitive to trajectory perturbations, while video SAR further requires efficient reconstruction of consecutive sub-aperture frames from incomplete measurements. Under variable curvature motion, forward-model errors introduce geometric and phase inconsistencies that weaken the interframe correlation exploited by tensor recovery. This article develops an integrated radius-of-curvature change rate (RCRC)-guided low-tubal-rank recovery framework for sparse aperture THz video SAR. An RCRC-aware variable curvature trajectory operator models deterministic motion-dependent effects, a tensor nuclear-norm prior exploits interframe redundancy, and pulsewise unit modulus phase factors compensate for residual perturbations. The resulting joint problem is solved by alternating proximal-gradient, tensor singular value thresholding, and closed-form phase updates. Controlled point target simulations and measured data experiments are conducted at pulse retention ratios of 60%, 40%, and 30%. A single global slow-time mask is applied to the shared pulse stream before sub-aperture slicing, ensuring consistent pulse omissions across overlapping frames. Comparisons with three literature baselines evaluate the integrated method, while two internal ablations examine the effects of RCRC modeling and residual phase estimation within the adopted TNN-regularized formulation. The 40% case preserves the principal scattering structure and is treated as a candidate sparse observation operating point, whereas the degradation at 30% identifies a more severe robustness boundary. Because the sparse observations are generated retrospectively from full-rate measurements, the experiments validate reconstruction under missing pulses but do not demonstrate physical reductions in PRF, transmitted pulses, or collected raw data volume.

Remote Sensing doi: 10.3390/rs18213732

Authors: Qingming Chen Lvhao Chen Yaoliang Chen Zhijian Xie Keyan Fang Kunping Yang

Remote-sensing image change captioning (RSICC) generally assumes correctly paired bitemporal images and focuses on isolated sentence descriptions, limiting applicability to archives where the correct temporal counterpart may be unknown. We introduce the Noisy-Reference Paragraph-Level Change-Captioning (NRP-CC) task, which requires retrieving the post-change observation from a noisy candidate pool before generating a coherent multi-statement paragraph. We construct the SECOND-ParCap dataset with image-level semantic annotations and paragraph-level change descriptions, and propose a Weakly Supervised Detection-Guided Historical-Cascaded Network (WSDHC-Net). WSDHC-Net leverages Dual Similarity Matching (DS-Matching) for retrieval from noisy candidate pools and employs weakly supervised detection to mine semantic cues for the Historical-Cascaded Paragraph Refinement Change-Captioning (HCPR-CC) module, which generates multi-statement paragraphs describing scene changes. We further introduce SA-ROUGE to evaluate sentence-level correspondence across the generated statements. On SECOND-ParCap, WSDHC-Net achieves Acc@1 scores of 99.53%, 99.29%, and 98.52% under appearance-hard retrieval with 5, 10, and 20 candidates, respectively. In a separate captioning benchmark using identical frozen retrieved inputs across all compared methods, WSDHC-Net attains ROUGE-L and SA-ROUGE scores of 0.5017 and 0.5016, respectively, under reference-assisted sentence-level evaluation; the corresponding paragraph-level scores are 0.2749 and 1.9305. These results support the feasibility of integrating temporal-correspondence retrieval with paragraph-level change description in NRP-CC.

Remote Sensing doi: 10.3390/rs18213730

Authors: Diming Wu Rui Yang Ying Cao Qing Cheng

Reliable drought assessment requires spatially continuous drought information, whereas station-based drought indices are spatially discrete and many existing gridded products are available at relatively coarse spatial scales. To address this issue, this study generates a monthly Geographically Weighted Composite Drought Index, termed GCDI-1, on a 1-km output grid over China. A geographically weighted SPEI-related auxiliary variable, termed GeoSPEI-1, is constructed from neighboring station-based SPEI-1 values to incorporate neighboring-station drought information into drought index generation. GeoSPEI-1 is jointly used with multi-source remote sensing and reanalysis variables as input to a geographically weighted feature-learning model, termed GeoCSL, which combines convolutional neural network, self-attention, and long short-term memory components to process the multi-source inputs. Input-combination experiments showed that GeoSPEI-1 was the primary contributor among the evaluated input variables within GeoCSL, while the multi-source environmental variables provided additional complementary information. Ten-fold cross-validation using the 2001–2010 training samples yielded an RMSE of 0.546 and an MAE of 0.407 for GeoCSL against station-based SPEI-1. Station-referenced evaluation using the temporally held-out 2011–2012 data further showed that GCDI-1 achieved an RMSE of 0.556 and an MAE of 0.410, compared with 0.562 and 0.421 for leave-one-station-out Kriging and 0.754 and 0.577 for SPEIbase-1, respectively. These results indicate that incorporating geographically weighted neighboring-station information together with multi-source environmental variables can support the generation of spatially continuous drought information and the use of GCDI-1 as a supplementary gridded product for regional drought monitoring and assessment in China.

Remote Sensing doi: 10.3390/rs18213729

Authors: Xiaodong Yin Bei Liu Peng Chen Yuchao Lin Zezhao Li Dongyang Fu

This study examines ocean conditions and rainfall-related flood candidates during Severe Tropical Storm Maysak using multisource observations in the northwestern South China Sea and Guangxi, China. Satellite products, atmospheric data, operational ocean products, published gauge summaries and official impact reports were analyzed with explicit observation windows and sensor provenance. Regional daily mean wind speed, significant wave height and mixed-layer depth peaked around 4 July 2026. Daily mean sea surface temperature (SST) on 5 July was 1.12 °C below the maximum during the study period. A daily first frontal analysis south of Hainan showed reductions of 15.3% in mean SST gradient and 25.6% in the mean daily strong front area between paired 15-day windows. Gradient decreases persisted across the tested windows and smoothing settings. Photosynthetically available radiation decreased by 24.1%, and diffuse attenuation at 490 nm increased by 7.9% across their respective windows. The merged chlorophyll-a retrievals increased descriptively by approximately 5.0%; sensor differences and limited common observations prevent interpreting this estimate as a bio-optical response independently corroborated across sensors. Integrated Multisatellite Retrievals for the Global Precipitation Measurement mission (IMERG) indicated a maximum accumulation of approximately 702 mm over the prolonged 2–16 July rainfall episode, including subsequent weather systems. Water-change masks identified 126.5 km2 of potential new inundation candidates, including 1.5 km2 of same-pixel multisensor overlap. The analysis documents marine changes, prolonged rainfall and regional water-change candidates within their stated observation windows.

Remote Sensing doi: 10.3390/rs18213726

Authors: Yuguo Wang Wensong Wang Xuanmei Fan Xin Wang Li Chen

Conventional landslide susceptibility assessments rely primarily on static environmental factors, which provide limited information on actual slope movement. The additional predictive value of interferometric synthetic aperture radar (InSAR) deformation features, and the consistency of this value, remain insufficiently established. This study assessed their contribution to landslide susceptibility mapping in Luding County, Sichuan Province, China. Sixteen static environmental factors were combined with three features derived from small baseline subset InSAR (SBAS-InSAR): long-term mean deformation velocity (Vmean), velocity variability (Vstd), and velocity trend (Kv). Random forest (RF), XGBoost (XGB), LightGBM (LGBM), and CatBoost (CAT) models were evaluated using 30 repeated spatial holdout splits and paired ablation experiments. Adding Vmean yielded modest median paired AUC gains of 0.003–0.005 across the four models under the primary 10 km spatial validation scheme. XGB achieved the highest mean AUC (0.914) with the M1 feature set, comprising the static factors and Vmean. Adding Kv or Vstd did not yield consistent further gains. Out-of-sample Shapley additive explanations (OOS-SHAP) identified rainfall, the coefficient of variation in elevation (CV), distance to faults, and lithology as the leading predictors; Vmean also made an appreciable contribution. The main predictors exhibited nonlinear responses. Spatial SHAP mapping and Moran’s I analysis further revealed pronounced spatial variation and significant clustering in feature contributions. These findings indicate that long-term mean InSAR deformation velocity provides a consistent but modest complement to conventional static factors, primarily through local adjustments to predicted susceptibility and the identification of regional differences.

Remote Sensing doi: 10.3390/rs18213727

Authors: Şenol Hakan Kutoğlu Mustafa Softa Elif Akgün

Deformation of glacier–bedrock systems provides evidence of slope adjustment before catastrophic failure. In a post-event assessment, we investigate the 26 August 2026 Langtang rock–ice failure in Nepal using NASA-ISRO Synthetic Aperture Radar (NISAR) observations to relate pre-event displacement to rock–ice interaction and valley confinement. Seven ascending line-of-sight (LOS) intervals characterize temporal change, while the final pre-event ascending–descending pair (P4) provides east–west and vertical displacement estimates. P4 spans 24-day windows ending on 19 and 16 August 2026. Relative to the surrounding terrain, median eastward and vertical displacements are +24.85 and −19.03 mm within the sampled possible detachment mapped by the United Nations Satellite Centre (UNOSAT), compared with +41.05 and +33.12 mm in the neighboring LOS deformation area. Opposing vertical tendencies and local east–west contrasts indicate differential movement across the rock–ice system. A positive LOS cluster lies 1.393 km from the possible detachment boundary and retains 78.8% of its connected area at coherence ≥ 0.70. The ascending record shows a positive pre-event contrast, a negative event-spanning contrast and a mixed post-event field. Together, these observations underpin a geological model in which oblique sliding increases loading at neighboring ice–bedrock contacts, promotes uplift and westward deflection, and transfers loading toward a valley constriction. We interpret this sequence as south to north, counter-clockwise adjustment accompanying differential sliding. The model connects the measured deformation pattern to structural weakness, changing glacial support and topographic confinement, providing a geological explanation for pre-event rock–ice adjustment within the source to valley hazard cascade.

Remote Sensing doi: 10.3390/rs18213728

Authors: Yuhang Zhou Zishuo Gu Yuanjin Xu

Under dense vegetation cover, deep networks for lithological classification must resolve small spectral differences between rock–vegetation mixtures. To address this issue, we propose the Dual-Branch Multi-Source Lithological Classification Network (DBMSL-Net), which uses Gaofen-5 Advanced Hyperspectral Imager (GF-5 AHSI) hyperspectral imagery, Phased Array type L-band Synthetic Aperture Radar-2 (PALSAR-2) data, and terrain data. Specifically, one branch encodes pixel spectra, and the other encodes spatial features from radar and terrain observations in the surrounding neighborhood. The spectral branch retains the overall spectral shape and local responses within intervals derived from absorption features in laboratory rock spectra. Laboratory pretraining initializes this branch, and source-domain class prototypes provide additional spectral references. A reliability gate adjusts the prototype contribution using target-pixel features and their similarity to the prototypes. The two branches are fused using weights that depend on the normalized difference vegetation index and pixel features. The core classifier, the differentiable binning residual module (DBRM), is trained jointly with both branches. It learns piecewise feature–class relationships through soft binning and refines class scores through successive residual updates. During inference, full-spectrum and feature-interval predictions are combined using class-specific weights. Neighborhood adjustment accounts for target extent. Across runs, boundary-aligned minimum-margin consensus is used for the spatially restricted target, whereas probability averaging is used for the extensive target. We evaluate DBMSL-Net in Lannitang, Yunnan, China, where quartz monzonite porphyry and quartz diorite porphyrite have similar mixed-pixel spectra and markedly different labeled sample sizes. On the within-area test subset, DBMSL-Net achieves a Target Macro F1 of 0.8301. Average accuracy, macro-averaged F1, and mean intersection over union reach 0.9388, 0.8780, and 0.7897, respectively. All four metrics exceed those of the nine comparison methods under the shared evaluation framework. Ablation experiments show that neither branch alone approaches the complete system (0.5727 and 0.7135, compared with 0.8301), and that the residual stage adds a measurable gain in both the initial and the complete configuration. The classification maps preserve the distributions of both the local and the extensive target and support lithological discrimination and field verification in this vegetated study area.

Remote Sensing doi: 10.3390/rs18213724

Authors: Lei Xia Yinnian Liu

Hyperspectral anomaly detection (HAD) aims to identify Earth’s surface anomalies without prior knowledge. However, most existing unsupervised HAD methods rely on training data roughly partitioned into background and anomalies, suffering from severe class imbalance and yielding feature representations heavily tied to scene-specific distributions. In contrast, supervised hyperspectral classification networks can extract well-organized spectral–spatial representations from diverse land-cover types; yet, this valuable knowledge remains underutilized in HAD due to the absence of labeled anomaly samples. To circumvent this limitation, this paper proposes a novel knowledge transfer paradigm from classification tasks to anomaly detection. The proposed framework employs a pretrained convolutional enhanced Transformer (CET) classification network as the teacher model, transferring its rich spectral–spatial context encodings via feature-level distillation. Concurrently, by synergizing latent feature alignment, Gaussian mixture model (GMM) priors, and reconstruction objectives, the student network is equipped with compact background modeling and highly discriminative representation abilities, which provide robust support for feature-space anomaly detection. Extensive experiments on spaceborne and airborne hyperspectral datasets demonstrate that our method substantially improves detection accuracy against complex terrestrial backgrounds with fast inference, while exhibiting superior cross-scene generalization.

Remote Sensing doi: 10.3390/rs18213725

Authors: Han Li Jiaqi Lei Pengnan Zheng Liang Chen Chao Wang

Complex rotational motion of a ship induces severe range migration and time-varying Doppler modulation in radar echoes. The time-domain generalized Radon–Fourier transform (TGRFT) can jointly compensate for these effects, but its multidimensional search over the initial range and all motion coefficients leads to a high computational burden. This paper proposes a two-dimensional frequency-domain GRFT (2D-FGRFT) for fast motion-parameter estimation and inverse synthetic aperture radar (ISAR) imaging of ships. The two-dimensional frequency-domain expression of a radar echo with a polynomial slant-range history is first derived, based on which a corresponding frequency-domain matched filter is constructed. The matched filter is parameterized only by the second- and higher-order frequency-domain motion coefficients. By exploiting the structure of the derived expression, a two-dimensional inverse fast Fourier transform (2D-IFFT) is performed after matched filtering, replacing the explicit searches over the initial range and first-order motion coefficient while preserving coherent integration of echoes from multiple scatterers on the ship. Finally, the motion parameters are determined from the matched-filter parameters and the peak locations obtained through constant false alarm rate (CFAR) detection, enabling fast ISAR imaging of the ship. Computer simulations and experimental results validate the computational efficiency, noise robustness, and overall performance of the proposed algorithm.

Remote Sensing doi: 10.3390/rs18213723

Authors: Haklim Choi Mi Eun Park Ji Eun Choi

Atmospheric methane is the second-largest contributor to radiative forcing among long-lived greenhouse gases, and characterizing its spatiotemporal distribution is essential for identifying sources and evaluating mitigation policies. Methane emissions are most directly reflected in near-surface concentrations, whereas satellite observations retrieve only the column-averaged methane concentrations. This study presents a physics-informed stacking ensemble framework that estimates surface CH4 directly from TROPOMI SWIR observations. SWIR radiance channels within the 2305–2385 nm band were optimally selected on the basis of their CH4 information content, including the contribution of the lower atmosphere, as quantified by radiative transfer analysis, and these L1B radiances were combined with L2 products and auxiliary variables governing surface concentration formation. Predictions from machine learning models with different learning schemes were fused through a stacking ensemble that exploited their complementary error structures and outperformed all individual base learners against ground-based in situ observations. For years and stations that were not used in training, the ensemble achieved R2 = 0.75 and 0.66, respectively. The selected channels improved estimation performance, and the L1B-only configuration performed comparably without a dedicated CH4 retrieval product, indicating that the observed spectra carry information relevant to surface CH4 before retrieval. The reconstructed global fields reproduced large-scale spatial and seasonal variability with stronger regional contrasts than the input column observations, demonstrating that physically informed channel selection combined with machine learning enables satellite-based surface CH4 estimation.

Remote Sensing doi: 10.3390/rs18213722

Authors: Abdelkader Riche Carla Arellano Elena Nafieva Jachin Jonathan van Ek Yann Rebois Pierluigi Confuorto Silvia Bianchini Daniel Hölbling

Rapid post-event landslide mapping is essential for emergency response and disaster management; however, accurately delineating rainfall-induced landslides remains challenging because of their fragmented morphology, heterogeneous landscapes, and the complementary nature of Earth Observation data. Although recent deep learning approaches have demonstrated promising performance using optical imagery, the contribution of different geo-auxiliary information to landslide segmentation and its transferability across regions remain insufficiently understood. This study proposes a Residual Dual-Branch SegFormer that integrates heterogeneous geospatial information through a lightweight residual Convolutional Neural Network (CNN) encoder and a novel Multi-Scale Geo-Auxiliary Gated Fusion Module (MGFM), enabling adaptive feature fusion while preserving the original SegFormer architecture. Using a multimodal dataset constructed from the March 2023 Tropical Cyclone Freddy event in southern Malawi, we systematically evaluate RGB imagery combined with optical, terrain, and Sentinel-1 SAR-derived auxiliary layers. The framework is assessed on a geographically independent test area and further validated through zero-shot transfer on the public Landslide4Scene and Lushan datasets. Experimental results demonstrate the following: (1) Integrating pre- and post-event SAR coherence with RGB imagery achieves the best overall performance when applied to areas that match the training data, reaching a mean Intersection over Union (mIoU) of 80.56% and an F1-score of 78.03% in the independent Malawi test area. (2) RGB combined with slope represents the most effective among the evaluated static topographic configurations in zero-shot transfer to the two external datasets, consistently outperforming the RGB baseline. (3) The proposed framework exhibits strong cross-region generalization, achieving the highest transfer performance on both external datasets while demonstrating that SAR coherence provides the most discriminative auxiliary information for post-event rainfall-induced landslide segmentation. These findings provide practical guidance for designing robust multimodal deep learning frameworks for operational landslide mapping.

Remote Sensing doi: 10.3390/rs18213721

Authors: Siyu Wu Zhiguo Pang Daling Cao Jun’e Fu Wei Jiang Yao Zhao

The difficulty in accurately determining channel Manning’s roughness has long constrained the accuracy of river flood simulation. At present, Manning’s roughness is mainly assigned using empirical approaches such as lookup tables and visual comparison, which are insufficient to represent the spatial heterogeneity of roughness in natural rivers. Remote sensing provides valuable data support for flood simulation in areas with sparse hydrological gauges. Remote-sensing observations are objective and can be acquired over large areas, but they are temporally discontinuous, whereas hydrodynamic models provide physically meaningful and temporally continuous simulations. These two information sources are therefore naturally complementary, providing a new avenue for Manning’s roughness inversion through remote sensing and hydrodynamic model coupling. To address these challenges, this study proposes a new remote sensing and hydrodynamic model coupling framework centered on roughness inversion. SAR-derived inundation extent is used as a spatial constraint, and an objective function together with particle swarm optimization (PSO) drives an iterative closed loop between the one-dimensional HEC-RAS model and the remote-sensing constraint to autonomously approach the optimal roughness parameters and improve the reliability of flood simulation. To bridge the scale mismatch between one-dimensional hydrodynamic outputs and two-dimensional SAR observations, the HEC-RAS cross-section geometry TIN structure is used to spatialize the simulated water surface, enabling direct comparison between simulated and SAR-observed inundation extents. The Le’an River in Wuyuan, Jiangxi Province, China, was selected as the test reach to evaluate the applicability and reliability of the framework under sparse-gauge conditions. The model was calibrated using the SAR-derived inundation extent, and simulated water levels were subsequently compared with rating-curve-derived reference levels. Under the SAR inundation constraint, the agreement between simulated and observed inundation extents improved, with IoU increasing from 0.763 under the initial parameter set to 0.789. Comparison with rating-curve-derived reference water levels further showed that the optimized Manning’s roughness parameters improved water-level simulation, reducing RMSE from 0.811 m to 0.594 m and increasing NSE from 0.950 to 0.973. These results demonstrate the effectiveness of the proposed method when continuous hydrological observations are unavailable. The framework provides an automated remote-sensing-constrained approach for Manning’s roughness inversion in hydrodynamic models of small and medium-sized rivers with limited in situ hydrological data.

Remote Sensing doi: 10.3390/rs18213720

Authors: Leli Zong Zi Chen Xiaonan Niu Shengjun Xiao Yanwei Sun Shangxiao Wang Yue Chang

Accurate landslide susceptibility assessment (LSA) is essential for regional disaster prevention and mitigation planning. Although machine learning and deep learning models have been widely adopted for LSA, their transferability across regions is often compromised by distributional discrepancies between source and target datasets. To address this challenge, we adapted the conditional domain adversarial network (CDAN) framework for cross-regional landslide susceptibility assessment to mitigate domain shift and improve cross-regional robustness. Two cross-regional transfer experiments were conducted to compare CDAN with conventional machine learning, direct-transfer deep learning, and representative domain adaptation methods. The results showed that CDAN achieved the highest mean Accuracy and F1 score among the benchmark models in both target domains, with Accuracy values of 0.77 and 0.80 and F1 scores of 0.78 and 0.80, respectively. The corresponding values of the area under the receiver operating characteristic curve (AUC) were 0.837 and 0.836, respectively. Neighborhood-based spatial information also improved the direct-transfer performance of conventional machine learning models compared with single-pixel representations. These results indicate that the CDAN-based framework can reduce the reliance on labeled target-domain samples and support county-level cross-regional landslide susceptibility assessment under domain shift.

Remote Sensing doi: 10.3390/rs18213719

Authors: Huilin Shu Hanqing Song Li Lei Chunmei Su Xiaomin Li Xianwei Li Lianghua Qi

Ma bamboo (Dendrocalamus latiflorus Munro) is an ecologically and economically important sympodial bamboo species widely cultivated in southern China. Mapping its spatial distribution with satellite imagery remains challenging due to spectral response overlaps between Ma bamboo and other forest types. This study assessed the potential of integrating seasonal spectral variation and a targeted SWIR2–RE1 spectral contrast from Sentinel-2 to improve Ma bamboo mapping. Four normalized seasonal metrics were derived from winter and summer observations. A Ma Bamboo Spectral Index (MBSI) was developed based on the normalized contrast between SWIR2 and RE1. Four feature configurations were evaluated using XGBoost and nested spatial cross-validation. SHAP and accumulated local effects were further employed to interpret the model. The results showed that the model integrating annual spectral variables, seasonal metrics, and MBSI performed best, with an overall accuracy of 0.833, an F1 score of 0.819, and an AUC of 0.875. Seasonal metrics primarily improved producer’s accuracy, whereas MBSI contributed more to user’s accuracy, indicating complementary effects on reducing omission and commission errors. MBSI was the most important predictor, while RE1c was the most important seasonal metric and showed the strongest interaction with MBSI. These results show that combining seasonal spectral variation with the SWIR2–RE1 contrast can improve the completeness and reliability of Ma bamboo mapping and provide an interpretable framework for mapping in southern China.

Remote Sensing doi: 10.3390/rs18213718

Authors: Nikola Zorić Anita Simic Milas Jochem Verrelst Dinka Matošević Ante Seletković Ivan Balenović

Hybrid radiative transfer model–machine learning (RTM–ML) approaches can incorporate physical constraints into vegetation-trait retrieval, but their advantage over empirical learning remains uncertain for very-high-resolution, limited-band observations of structurally heterogeneous forest canopies. We compared six empirical machine-learning regression algorithms with a hybrid PROSAIL–Gaussian Process Regression (GPR) approach for retrieving derived canopy chlorophyll content (CCC) from four-band UAV multispectral imagery of pedunculate oak (Quercus robur L.) stands under contrasting oak lace bug (Corythucha arcuata) pressure. CCC was derived from field-derived leaf chlorophyll content and UAV-estimated leaf area index. Data were collected across three sampling periods from 2021 to 2023, and models were evaluated using tree-grouped three-fold cross-validation to maintain independence between training and held-out trees. Empirical performance varied among algorithms, sites, and sampling periods. At the infested site, GPR provided the strongest performance in May (R2 = 0.61, RMSE = 27.04 µg cm−2, NRMSE = 10.87%), whereas Random Forest performed best in September. At the reference site, Random Forest, least-squares linear regression, and kernel ridge regression provided the strongest results depending on sampling period and performance metric, with R2 reaching 0.60. According to the primary cross-validated predictive metrics (R2, RMSE, and NRMSE), the tested PROSAIL–GPR approach generally showed lower predictive performance than the empirical models under the same tree-grouped evaluation framework. Empirical GPR maps resolved substantial within-stand spatial variation in derived CCC. Overall, incorporating PROSAIL-simulated information did not improve predictive performance under the tested four-band, crown-scale configuration. Rather than identifying a single cause, the results indicate that the benefit of physically constrained retrieval depends on compatibility among model representation and parameterization, canopy structure, spectral information, and available training data.

Remote Sensing doi: 10.3390/rs18213717

Authors: Linru Xu Tong Liu Jie He Weixing Zhang Pinzhuo Wang

To address visual misclassification caused by subtle surface signatures, variable preservation, and similar patterns from natural landforms and modern features, we propose a knowledge-enhanced multimodal reasoning framework for the northwestern margin of the Tarim Basin. Site Objects organize multi-source data, while expert knowledge from the Morphology, Shadow, Context, and Temporal dimensions is represented as structured evidence and hierarchical decision rules. Knowledge-guided supervision and domain adaptation using low-rank adaptation enable candidate assessments to consider archaeological support, non-archaeological alternative explanations, and insufficient evidence. Independent testing showed that knowledge-informed prompts alone did not consistently improve performance. After domain adaptation, strict precision increased from 22.31% to 58.70% and the F1 score from 33.54% to 62.79%, while strict recall remained essentially unchanged. Strict false-positive rates for modern anthropogenic features and natural geomorphic look-alikes decreased from 53.75% and 31.25% to 10.00% and 7.50%, respectively, while candidate recall improved. Regional application demonstrated candidate screening across contiguous remote sensing scenes. The framework retains potential sites while constraining visually similar confounders, thus providing more reliable and traceable screening support for large-scale surveys in complex arid regions.

Remote Sensing doi: 10.3390/rs18213716

Authors: Yang Liu Yibin Yao Bao Zhang Qingzhi Zhao Mingxian Hu Junyu Li Zhaohui Xiong Jing Zhang

Accurate hourly regional precipitation forecasting is essential for flood warning and weather-risk management, but station-scale global navigation satellite system (GNSS) precipitation models cannot fully describe the spatial evolution of rainfall between stations. This study proposes an hourly regional precipitation forecasting (HRPF) model that combines gridded GNSS-derived precipitable water vapor (PWV) and gridded precipitation with a convolutional long short-term memory (ConvLSTM) encoder–decoder network. Missing values in the hourly PWV and precipitation records from 69 GNSS-collocated meteorological stations in Taiwan Province during 2008–2012 were reconstructed using singular spectrum analysis and then resampled onto a 0.06° latitude–longitude grid using nearest-neighbor interpolation. The model uses PWV and precipitation fields from the preceding 1–6 h as two input channels and predicts the precipitation field one hour ahead. Lagged autocorrelation analysis showed strong spatiotemporal dependence in both precipitation and PWV fields, with coefficients above 0.86 and 0.85, respectively. In annual-scale external validation, the HRPF model maintained probability of detection (POD) values above 90%, false alarm rate (FAR) values below 55%, and threat score (TS) values of 46.22–46.58%. Seasonal external validation showed POD values of 88.8–97.8%, with the highest TS in summer (52.8–53.3%) owing to lower FAR values (45.8–46.4%). These results indicate that GNSS-derived PWV can provide useful spatial water-vapor information for one-hour-ahead regional precipitation forecasting, while model skill remains season-dependent over Taiwan’s complex terrain.

Remote Sensing doi: 10.3390/rs18213715

Authors: Xiaokun Sun Shuang Li Canbin Hu Fei Ma

Infrared imagery supports low-light target perception, but acquiring, registering, and annotating paired visible–infrared imagery is costly. Visible-to-infrared image translation can alleviate this demand, yet cross-modal differences often degrade object boundaries and spatial layouts, especially for small targets. We propose DFSMamba to preserve geometric structure throughout visible-to-infrared translation by connecting frequency-guided feature formation, input-dependent directional aggregation, and infrared-domain reconstruction supervision. A discrete wavelet transform (DWT)-based detail-guided module supplements bottleneck features with learned structural cues, direction-adaptive 2D selective scan adjusts the fusion of aligned directional responses, and state-space similarity matching constrains the output in a fixed multiscale feature space learned through real-infrared reconstruction. Experiments on FMB, VEDAI, and AVIID-1 include image-quality comparisons, target-level evaluation, ablation studies, efficiency analysis, and downstream vehicle detection. Among evaluated methods, DFSMamba achieves the best SSIM, MS-SSIM, PSNR, and L1 values on all three datasets. On AVIID-1, it obtains an SSIM of 0.985 and a PSNR of 41.799 dB. A YOLOv9-C detector trained on generated images reaches an AP50:95 of 0.636 on real infrared test images, compared with 0.640 when trained on real infrared images. These results show that DFSMamba preserves detection-relevant structure and supports the use of generated images as synthetic training data in the evaluated aerial vehicle-detection setting.

Remote Sensing doi: 10.3390/rs18213714

Authors: Zhonghua Li Bin Xiao Zhanghai Wu

In general object detection tasks, accurate responses to the boundaries and classifications of potential objects are required of detectors. Given the diverse shapes of objects, these bounding boxes are typically represented as horizontal rectangles. However, in aerial imagery scenarios, horizontal bounding boxes can easily miss valid detections when objects are densely distributed. In recent years, oriented object detection algorithms have achieved considerable success. However, they focus on the boundaries of the object without taking into account its head–tail attributes and true orientation. Getting the true orientation of objects serves as the long-term aim of remote sensing object detection task. To resolve these issues, we decompose the single Gaussian distribution probability loss into a sequential dual-probability distribution, thereby establishing a one-to-one collaborative loss mechanism. This enables the model to learn the real orientation relationship from the dual distribution loss, achieving the acquisition of objects’ true orientation information. To more accurately reflect the preference of detectors, we design a new evaluation metric based on the cosine function and average precision. We conduct a series of experiments on the DOTA dataset, HRSC dataset, and FAIR1M dataset, achieving satisfactory results that validate the effectiveness of the proposed method.

Remote Sensing doi: 10.3390/rs18213712

Authors: Anbin Yu Wensheng Mei Dapeng Zhu

Fine-crack segmentation in high-resolution UAV images of bridges is hindered by weak crack signals, complex backgrounds, and substantial computational demands. This study proposes a dual-stage framework combining lightweight grid-wise screening with graph-enhanced segmentation. CrackClsNet identifies candidate regions, and CrackSegNet processes contextual patches comprising each candidate and its eight neighbors. Multiscale and deformable convolutions, attention, and cross-patch graph reasoning support feature extraction, while overlapping predictions are fused during reconstruction. In experiments using 1000 pixel-annotated crack-containing images collected from 13 highway bridges and 500 crack-free images, neighborhood retention increased candidate region crack-pixel coverage from 91.97% to 95.65% on the classification test set. The end-to-end framework achieved an F1-score of 0.5826, Crack IoU of 0.4110, and mIoU of 0.7052, exceeding six baseline methods in these metrics. The results of ablation experiments further support the benefits of neighboring context and cross-patch feature exchange, with 3 × 3 neighborhoods balancing segmentation quality and computational cost. With approximately 4.41 million parameters, the framework achieved 4.10 FPS with a reported memory consumption of 6.94 GB at 4096 × 5632 pixels, providing the highest throughput and lowest reported memory consumption among the evaluated methods at this resolution. These results indicate a promising balance between segmentation performance and computational cost on the evaluated bridge image dataset.

Remote Sensing doi: 10.3390/rs18213713

Authors: Zhaonan Xue Lin Liu Zhicai Luo

High-altitude lakes on the Tibetan Plateau (TP) are important components of regional water-cycle regulation and climate responses, yet their long-term monitoring remains challenging due to limited in situ observations and sparse spatial sampling of conventional satellite altimetry. The Surface Water and Ocean Topography (SWOT) mission provides two-dimensional observations of inland water-surface elevation (WSE), offering new opportunities for lake monitoring in remote regions. This study evaluates the applicability of SWOT L2 Pixel Cloud (PIXC) observations for TP lake monitoring through cross-sensor comparison with ICESat-2 ATL13 observations. A total of 50 representative TP lakes were analyzed using quality-controlled SWOT–ICESat-2 matched observations. The results demonstrate that SWOT PIXC and ICESat-2 ATL13 generally capture comparable lake-scale WSE variations across the TP. The mean MAE and RMSE were 0.164 m and 0.231 m, respectively, indicating decimeter-scale cross-sensor consistency for most investigated lakes. Lake area showed no significant association with the RMSE, suggesting that lake size alone does not explain the variability of cross-sensor residuals. Further analysis reveals that residual differences are strongly scene-dependent and reflect combined influences from satellite retrieval characteristics, spatial sampling differences, environmental conditions, and temporal offsets rather than SWOT-only uncertainty. These findings provide an independent assessment of SWOT observations over high-altitude lakes and demonstrate their potential for spatially continuous monitoring of TP lake dynamics. The results also establish a foundation for integrating SWOT with multi-source satellite observations to reconstruct long-term lake water-level and storage variations.

Remote Sensing doi: 10.3390/rs18213711

Authors: Yan Lv Guanghao Kang Yan Cao Shiyu Wang Liang Guo Huaiwu Peng Wei Feng

The inherent class imbalance and scarcity of labeled samples remain major challenges in hyperspectral image (HSI) classification. SMOTE-based methods alleviate class imbalance by generating synthetic minority-class samples. However, their interpolation mechanisms often fail to capture spatial–spectral features adequately, resulting in noisy or boundary-ambiguous samples. To address this issue, this paper proposes Spatial–Spectral Prototype-guided Deep SMOTE (SSPD) for imbalanced HSI classification. Firstly, SSPD constructs a spatial–spectral representation guided by feature importance to suppress redundant spectral bands and isolated spatial noise. Secondly, SSPD employs a prototype-guided oversampling strategy to generate synthetic minority-class samples. Class prototypes determine the generation direction, while feature-importance weights constrain the interpolation process to preserve discriminative spectral responses. Finally, a stepwise ensemble with soft voting is employed to integrate classifiers trained under different imbalance ratios. Extensive experiments were conducted on Indian Pines, Pavia University, Salinas, and WHU-Hi benchmark datasets. Compared with the individual machine-learning classifiers, SSPD achieves more stable and competitive performance in terms of Overall Accuracy (OA), Average Accuracy (AA), and Kappa Coefficient (KC).

Remote Sensing doi: 10.3390/rs18213710

Authors: Shicheng Li Zhilong Zhao Binjie Liu Qin Ye Huimin Zhu Zhihua Liu Meijiao Li Mingjie Song

Protected Areas (PAs) are essential for global biodiversity conservation, yet assessing their management and conservation effectiveness remains a formidable challenge. While remote sensing (RS) enables synoptic, large-scale monitoring, existing reviews are often fragmented—either decoupled from PA governance contexts or confined to isolated threat mappings. To bridge this gap, this review proposes a dual-dimensional analytical framework integrating management effectiveness and conservation effectiveness. Based on the literature retrieved via a systematic Boolean search in the Web of Science, we first establish a multi-source RS technology stack covering space-borne optical, radar, nighttime-light, LiDAR and cloud-computing platforms. We then review RS applications within the two dimensions. For management effectiveness, we trace the progression from single-disturbance tracking to integrated pressure assessments using counterfactual matching designs. For conservation effectiveness, we synthesize advances in biodiversity dynamics, ecosystem pattern integrity, quality evaluation, and ecosystem service quantification. Our review demonstrates that RS has become a major component of large-scale PA evaluation, revealing measurable successes while highlighting the risk of overestimating conservation outcomes when relying solely on greenness-based proxies. Persistent challenges include the lack of standardized, biome-specific assessment frameworks and the disconnect between biophysical monitoring and socio-ecological drivers. Addressing these gaps requires unified evaluation systems aligned with the Kunming-Montreal Global Biodiversity Framework, AI-driven multi-sensor fusion, and the operationalization of RS into routine PA governance to support the “30 × 30” target and ecological product value realization.