ABSTRACT This study proposes a geospatial procedure that leverages Landsat 8/9, TERRACLIMATE, and SRTM-DEM data, to spatialize drought patterns and dynamics, from a confrontation–confirmation–completion perspective of a reference (empirical) and observed (records) warmest period. First, Landsat 8/9 bi-month median image is used to develop the Improved Land Surface General Drought Index – version 3 (LSGDI3) with the Euclidean function. Next, a multi-regression computation supports the downscaling-adjustment of climate (climatic drought) and topography (topographic drought) to the previous step output. Further, the Weighted Residuals Aggregation Polynomial (WRAP) model is proposed, to produce the Land Surface-Climate-Topography Combined Drought Model (LSCTDMComb). As overall finding, the dual data reference/observed alternative is highly efficient to model drought spatial patterns at a regional scale and reflecting the global warming dynamics. For the 2014–2023 studied decade, the average decade cross-receiver operating characteristic/area under curve (cross-ROC/AUC) results are satisfying up to 0.818, sign of a balanced and robust model. Further support is provided by lower values of the root mean square error (RMSE), mean absolute error (MAE), and relative absolute error (RAE) metrics, respectively [0.0368–0.246], [0.0031–0.0233], and [0.005–0.0393]. Whereas the coefficient of determination values obtained between reference and observed warmest bi-months are, R2: [0.64–0.96], regardless of the temporal difference. Spatially, all LSCTDMComb agree on increasing risk of drought hazards with the latitudes at the Sahel–Sahara interface. These trends are similar to the patterns depicted by compared popular models, despite some caveats raised, especially the resolution discrepancies.

ABSTRACT Building extraction using deep learning is an important research direction in the field of remote sensing. HRNet is able to maintain high-resolution representations throughout the whole process, providing a powerful tool for dense prediction tasks. However, HRNet is not specifically designed for building extraction, leading to poor performance when applied directly. In addition, it suffers from excessive parameter counts and high computational complexity. To address these problems, this paper proposes LBEHRNet, an innovative lightweight network for building extraction. LBEHRNet is achieved by reconstructing the framework of HRNet and developing three novel core functional modules. First, the network framework is reconstructed to boost the overall computational efficiency and preserve more spatial details. Second, a lightweight star convolutional module is proposed by reconstructing the classic star convolution unit and further improving it with asymmetric convolutions and attention mechanisms, which significantly reduces computational overhead during the encoding stage while enhancing building feature representation. Next, an efficient cross-resolution assisted hybrid attention module is constructed to strengthen the collaborative enhancement across multi-resolution features through explicit interaction between spatial and channel attention mechanisms. Finally, a star operation-based multi-resolution fusion prediction head is designed to reinforce the model’s capability for local detail reconstruction of building features and global decoding. Experimental results on three public building extraction datasets, Inria, WHU-S and WHU-A, confirmed the effectiveness of LBEHRNet in terms of both accuracy and efficiency.

ABSTRACT Multi-LiDAR calibration is essential for ensuring accurate and reliable spatial measurements in applications such as 3D mapping, environmental monitoring, urban modelling, autonomous driving, and robotics. Conventional calibration approaches often rely on additional artificial targets and structured environments, limiting their applicability in dynamic and complex settings such as forests. This paper presents a robust extrinsic multi-LiDAR calibration framework, termed ExCalibR, designed to enable reliable sensor fusion in diverse environments without requiring external targets. The proposed framework employs a dual correspondence search strategy to establish point correspondences and integrates plane-to-plane matching with an adaptive pseudo-Huber loss kernel, an accelerated momentum update scheme, and a reduced Branch-and-Bound (BnB) algorithm to ensure global optimality and improved robustness. The performance of ExCalibR is evaluated against commonly used registration algorithms across multiple real-world datasets representing both natural and built-up environments. Comprehensive experiments and evaluation on collected and publicly available datasets demonstrate that the proposed framework consistently estimates stable extrinsic parameters, achieving the least variation across different environmental conditions. The results demonstrate the robustness and generalizability of the proposed framework, highlighting its applicability across diverse sensor configurations and complex environmental conditions without reliance on additional artificial targets.

ABSTRACT Hyperspectral images (HSIs) are inevitably degraded by sensor noise and atmospheric interference, which severely limits their reliability in subsequent analysis tasks. Recent denoising approaches have leveraged the intrinsic spectral low-rank subspace structure of HSIs, typically represented by low-dimensional spectral bases and corresponding spatial coefficients. Despite their effectiveness, most existing methods neglect the distributional discrepancy between denoised results and clean HSIs, thereby constraining the expressive power of deep models. In this paper, we propose SLR-GAN, a subspace low-rank generative adversarial network for hyperspectral image denoising. Specifically, the noisy HSI is first decomposed into a compact spectral basis and spatial coefficient images. A Wasserstein-guided adversarial learning framework is then introduced to refine the coefficient images by explicitly aligning their distribution with that of clean counterparts, leading to more realistic and faithful reconstructions. Extensive experiments on both simulated and real-world HSIs demonstrate that SLR-GAN consistently outperforms state-of-the-art methods in terms of denoising accuracy and robustness.

ABSTRACT Historical aerial imagery acts as a crucial bridge to the pre-satellite era, becoming increasingly vital for quantifying long-term land cover changes and supporting retrospective environmental studies. This study presents a comprehensive comparison of CNN-based (DeepLabV3+, PSPNet, HRNet, ConvNeXt) and transformer-based (ViT, Swin Transformer, BEiT) architectures for semantic segmentation of historical aerial imagery. In particular, the potential of Transformer-based architectures for semantic segmentation of panchromatic historical aerial imagery in the context of increasing segmentation complexity remains underexplored. We assess how effectively each architecture handles increasing class granularity and captures finer-grained land-cover semantics. Experiments are conducted at three levels of detail: binary, three-class, and five-class segmentation. The dataset comprises panchromatic images from 1960, captured in Turkey. Precision, recall, F1-score, and intersection-over-union (IoU) metrics were employed to evaluate class-wise performance. In the binary ground and non-ground segmentation task, all architectures achieved mean F1-scores of 0.945–0.959 and mean IoU values of 0.905–0.921. In the three-class and five-class tasks, mean F1-scores ranged from 0.861–0.902 and 0.754–0.803, with mean IoU values of 0.777–0.831 and 0.623–0.682, respectively. In the three-class setting, ConvNeXt, Swin Transformer, and BEiT clustered towards the upper end of the mean metrics with only small differences among them. Increasing class granularity from three to five reduced mean performance across architectures, with decreases of up to 0.123 in mean F1-score and up to 0.176 in mean IoU. Across all classification levels, ConvNeXt consistently achieved higher performance, distinguishing itself specifically in the five-class task with the highest F1-scores and IoU values. ConvNeXt outperformed both the CNN-based baselines and Transformer-based models in the challenging building and road categories.

ABSTRACT Crop classification serves as a fundamental component for agricultural resource monitoring and refined cropland management. However, in regions characterized by complex crop types and diverse planting patterns, traditional remote sensing methods frequently struggle to maintain classification stability across multiple years and varying geographic scales. This study focuses on a typical irrigated agricultural area in Changji Prefecture, Xinjiang, where a multi-source time-series feature set was constructed by integrating Sentinel-2 optical imagery and Sentinel-1 radar data. To address existing limitations, we propose a dual-branch crop classification model, termed Multi-source Spatial-Temporal-Phenological Integration (MSTPI), which synergizes the spatial feature extraction capabilities of SegFormer with the temporal phenological recognition strengths of the Time-Weighted Dynamic Time Warping (TWDTW) algorithm. Additionally, refined farmland boundaries were delineated using the CLCFormer model to establish unified spatial analysis units, thereby mitigating classification uncertainties caused by mixed pixels. In a 2022 benchmark experiment, the TWDTW branch integrating the kernel Normalized Difference Vegetation Index (kNDVI) and VH features achieved an overall accuracy (OA) of 87.91% under a decision-level fusion strategy. With the introduction of the SegFormer branch, the OA increased to 93.56%, significantly outperforming models based on single features or feature-level fusion. The average OA for temporal transfer testing from 2020 to 2024 was 88.17%, while spatial transfer validation across seven counties yielded an average OA of 86.24%, highlighting the model’s robust generalization potential in both temporal and spatial dimensions. Parcel-level maps revealed distinct spatial patterns, and the identified crop rotation patterns demonstrated strong correspondence with local agricultural practices. Ultimately, the MSTPI model exhibited exceptional performance in terms of classification accuracy and regional adaptability, offering a highly effective framework for crop type identification in complex agricultural landscapes

ABSTRACT The Intraclass Correlation Coefficient (ICC) is commonly used to assess the consistency and reliability of different measurement methods for the same quantitative variable. The accuracy and reliability of multi-method measurement results, including the consistency in evaluation, are critical in meteorological research. In this study, vertical profile data from wind profiler radar and L-band radar at the Beihai National Meteorological Observatory in Guangxi, collected in 2022, were evaluated with ICC methodology to assess the consistency between the wind measurements. The results indicate that: (1) the overall ICC of the u component in the wind profile radar and the L-band radiosonde radar’s wind measurement data is 0.946, while the ICC of the v component is 0.908, demonstrating a very high degree of consistency in the wind component measurements; (2) values for different seasons were all greater than 0.88, with the consistency being higher in spring and summer compared to autumn and winter; (3) the ICC values at 08:00 and 20:00 were both greater than 0.91, showing nearly identical consistency between the two radars; (4) the ICC for wind speeds greater than 7.1 m·s−1 and below 7.1 m·s−1 were both above 0.85, with a higher consistency observed at higher wind speeds; (5) the ICC values both under precipitation and non-precipitation conditions were greater than 0.9, with no significant difference in the consistency degree between the two radar datasets; (6) the ICC values under varying mean relative humidity conditions were all greater than 0.83, indicating a high degree of consistency, and the consistency was significantly lower when the mean relative humidity was below 60% compared to when it was above 60%; (7) the ICC values under different detection altitudes were all greater than 0.88 with a high degree of consistency. Notably, the agreement above 2,000 metres is significantly stronger than that below this altitude.

ABSTRACT Air pollution is responsible for various health issues, including respiratory and cardiovascular diseases, among individuals. However, previous studies have not successfully identified the sources of air pollution that contribute to the acceleration of climate change. To address this gap, a novel approach known as the Adaptive Exponential Sigmoid Fuzzy Tsallis Entropy Interference System (AES-FTEIS) is proposed for identifying air pollution sources. Datasets from remote sensing and ground-level air pollution measurements are collected, temporally aligned using the Prior Distribution Regularized Kalman Filter (PDRKF), and imputed using Cross-Entropy Minimization Spline Interpolation (CEMSI). Additionally, aerosol particles such as PM2.5 and PM10 are extracted from the dataset and incorporated into the analysis. Subsequently, the data are organized by location and time using Transfer Entropy Spectral Clustering (TESC), and their correlation are analysed using Spearman Rank Correlation (SRC). An exploratory data analysis is conducted on the time-based grouped results through Box plots, leading to feature extraction. Finally, the AES-FTEIS is utilized to identify the pollution sources based on the levels of aerosol particle concentrations. The experimental results show that the proposed WOLSTM-ASLRCNN classifier achieves 97.56% of accuracy and 98.5% of precision, outperforming existing models such as CNN (95.3%), LSTM (93.84%), GRU (92.57%), and RNN (89.66%). The proposed TESC clustering method obtained a silhouette score of 0.9721, higher than SC (0.9687), AC (0.9428), HC (0.9271) and KMC (0.9087). Moreover, the AES-FTEIS source identification approach reduced the rule generation time to 1483 ms, demonstrating the effectiveness of the proposed framework.

ABSTRACT Understanding the optical properties of different sea ice types is crucial for improving knowledge of energy balance, heat and mass budgets, and for enhancing the accuracy of remote sensing retrievals. The in situ-based spectral characteristics of coastal first-year sea ice in eastern Liaodong Bay, Bohai Sea, were systematically analysed. A regional reflectance database was established in January 2025, collecting 49 ice samples with diverse types, thicknesses, densities, sampling locations, and observation angles. The results show that scale-like ice exhibited the highest visible-band reflectance (>0.7), followed by young ice with considerable variability, while frazil ice, nilas, and needle ice exhibited lower reflectance due to their thin structure, loose texture, and uneven surfaces, and sediment-laden ice had the lowest reflectance. Sea ice reflectance was strongly related to ice density, with lower-density ice showing higher values, and displayed a significant exponential relationship with ice thickness in both ultraviolet and visible bands (p < 0.01), with the ultraviolet band yielding higher coefficients of determination. The Hemispherical Directional Reflectance Factor (HDRF) increased with azimuth angle, reaching a minimum at nadir (0° zenith), and reflectance grew with zenith angle, with forward scattering being dominant, followed by side and back scattering. Additionally, comparisons between Geostationary Ocean Colour Imager II (GOCI-II) satellite data and in situ measurements showed similar spectral patterns, including primary peaks in the ultraviolet–blue bands and a secondary maximum near 600–700 nm, although GOCI-II reflectance was generally 10–20% lower. Moreover, preliminary results based on four machine learning algorithms indicate that GOCI-II satellite data have clear discriminatory potential for classifying sea ice by both region and thickness.