Remote sensing spatiotemporal fusion aims to generate image sequences with both high spatial and temporal resolution. However, many deep learning approaches treat fusion as a black-box regression problem, overlooking differences in sensor imaging physics—such as the point spread function (PSF) and pixel-area integration—which often leads to boundary blurring, spectral mixing, or even physically implausible high-frequency details over heterogeneous landscapes. In addition, these methods commonly rely on paired coarse-/fine-resolution reference images at the same time, which limits data availability. To address these issues, we propose SPP-STF, a superpixel-based fusion network trained with a physics-guided degradation-consistency regularizer that requires only a reference fine-resolution image and a target coarse-resolution observation, while deriving an edge intensity map as a structural prior. The model adopts a U-Net-style, three-branch encoder–decoder to fuse texture, temporal variation, and edge cues at multiple scales. An EdgeResBlock uses high-frequency gating to sharpen boundaries and reduce cross-class leakage, and a superpixel cross-attention Transformer learns object-level correspondences to alleviate alignment errors caused by mixed pixels. During training, we introduce a learnable anisotropic Gaussian PSF-based degradation-consistency term: the predicted fine-resolution output is blurred and downsampled via area integration to match the observed coarse image, and this term is optimized jointly with the reconstruction loss. Quantitatively, SPP-STF achieves the best RMSE, SAM, PSNR, and UIQI on CIA (0.0254, 0.1522, 31.9330 dB, and 0.8451), the best results across all six metrics on LGC (including RMSE 0.0188 and PSNR 34.6307 dB), and the best RMSE, SSIM, SAM, PSNR, and ERGAS on Tianjin (0.0410, 0.7640, 0.2459, 27.8666 dB, and 30.4838). Ablation studies confirm the complementary benefits of the superpixel and edge modules. The PSF term is interpreted here as a training-time regularizer that provides consistent and physically more reasonable improvements across the three datasets. Structured pruning further indicates that the model can retain spatial structure under substantial compression, while spectral fidelity becomes notably more sensitive beyond roughly 70% pruning.
Offshore wind is an important renewable infrastructure supporting the global energy transition, making accurate offshore wind turbine (OWT) mapping essential for understanding infrastructure expansion and informing marine spatial management. Existing long-term approaches often extend single-period detections over time based on structural-stability assumptions, but accumulated identification errors and regional imaging differences limit consistent object-level representation across years and regions. To address this challenge, this study proposes a coordinated three-stage object-level SAR mapping framework, in which a stable and balanced sample representation strategy under unified annual temporal windows provides a consistent data basis across regions and years. On this basis, the lightweight YOLO-SAR model integrates frequency-aware feature modeling and hierarchical structural representation to improve OWT detection under heterogeneous conditions. Compared with YOLO11n, YOLO-SAR reduces the parameter count from 2.58 M to 2.30 M while improving mAP@0.5:0.95 from 0.3956 to 0.4553. Spatial reconstruction then transforms detection outputs into consistent annual object-level OWT maps, forming a continuous spatial database covering 2015–2025. Multi-region validation and comparisons with public datasets further support the reliability of these spatial records. These records reveal multiscale patterns of offshore wind spatial evolution: shifts in regional contributions reshape global growth, regional expansion pathways and internal spatial restructuring exhibit stage-dependent differences, and site-boundary expansion and turbine-spacing adjustments determine local spatial organization. These findings demonstrate that global offshore wind development is not merely a process of quantitative growth, but a multiscale spatial evolution driven by regional shifts, site expansion, and internal spatial reorganization. The proposed framework provides a consistent data foundation for global offshore wind evolution analysis and can be extended to other marine energy infrastructures.
Long-term deformation of high earth-rockfill dams reflects post-construction consolidation, creep, reservoir loading, and cyclic operation. Separating these superimposed components is essential for gaining process-level insights into spatial patterns and differential responses. This study investigates the Pubugou Dam, a 186 m high earth-rockfill dam in Southwest China, using ascending and descending Sentinel-1 observations from 2017–2023. The two viewing geometries were jointly inverted to recover vertical and east-west displacements. A wavelet-guided decomposition, in which dominant periods identified by the continuous wavelet transform constrain the window length and component grouping in singular spectrum analysis, was proposed to separate the displacement records into long-term trend, annual cyclic, and residual components. Synthetic experiments yielded root mean square errors of 0.69–2.16 mm, indicating effective decomposition performance. Cross wavelet transform and wavelet transform coherence quantified localized coherence and phase lags between the cyclic components and reservoir water level. The central earth-rockfill body exhibits the largest magnitudes of vertical and east-west deformation rates, whereas the spillway and abutments remained comparatively stable. The trend components and settlement indices are consistent with relatively slow long-term post-construction settlement during the observation period. Annual cyclic deformation is more pronounced in the east-west direction and exhibits spatially variable lags relative to reservoir level fluctuations. Settlement inclination and apparent horizontal strain remain low under the adopted criteria, suggesting limited crest-cracking susceptibility during the observation period. This work demonstrates how wavelet-guided decomposition of two-dimensional InSAR time series provides spatially explicit insights into long-term settlement and cyclic deformation of high earth-rockfill dams.
Mangroves are among the most important blue carbon ecosystems, contributing substantially to global carbon sequestration and biodiversity conservation. Canopy carbon (CCC), nitrogen (CNC), and potassium (CKC) contents provide critical indicators of nutrient cycling, ecosystem functioning, and ecological health. However, the structural complexity and strong spatial heterogeneity of mangroves present major obstacles to understanding the spectral response mechanisms and achieving reliable large-scale retrieval of these biochemical parameters. To address these limitations, we developed an interpretable retrieval framework that integrates causal analysis with physical constraints for reliable estimation of mangrove biochemical parameters. First, we proposed a Multi-Channel Chatterjee-Enhanced Correlation Spectrum Analysis (MC-ξ-CS) method that integrates causal discovery to systematically identify spectral bands causally associated with CCC, CNC, and CKC. For large-scale retrieval of mangrove canopy parameters, the identified causal relationships were transferred to satellite observations through a Global Angle-Divergence Spectral Transfer (GADT) framework, yielding robust estimation (R2 = 0.511–0.669). Subsequently, a Dual-Mechanism Guided Physics-Informed Neural Network (DM-PINN) was developed to synergistically integrate causal analysis and physical constraints, achieving robust retrieval performance (R2 = 0.604–0.83) for three representative mangrove species and enhancing the physical interpretability of the results. To the best of our knowledge, this study provides the first comprehensive large-scale retrieval of key mangrove biochemical parameters along the southeastern coast of China, revealing CCC values of 409.18–566.89 g/kg (C/N = 9.77–14.21), CNC values of 14.38–74.9 g/kg (N/K = 1.99–3.86), and CKC values of 4.78–19.67 g/kg (C/K = 28.22–37.67). Distinct inter-provincial spatial patterns were observed: CNC and CKC exhibited similar gradients, increasing from landward to seaward regions, whereas CCC displayed an opposite spatial distribution pattern with lower values in coastal regions. Moreover, CCC was strongly positively correlated with CNC, while CKC showed negative correlations with both parameters. Collectively, this study established a novel theoretical and methodological framework for elucidating the spectral response mechanisms of mangrove biochemical parameters, offering critical scientific support for large-scale mangrove degradation monitoring, ecological restoration assessment, and biodiversity conservation.
The southeastern Tibetan Plateau (SETP), a critical component of the “Asian Water Tower”, hosts a large concentration of maritime temperate glaciers that are highly sensitive to climate change. However, persistent cloud cover, seasonal snow, and supraglacial debris complicate consistent long-term glacier mapping, while the use of static glacier boundaries may introduce temporal mismatches in altimetry-based glacier mass-change estimates. In this study, we developed a multi-source data-fusion framework on Google Earth Engine (GEE) to reconstruct glacier-area changes in the SETP from 2000 to 2025 at five-year intervals and 30 m spatial resolution. Random Forest classifiers were used to integrate Landsat and Sentinel-2 optical imagery, Sentinel-1 synthetic-aperture-radar data, and NASADEM-derived topographic information, supporting glacier delineation under complex surface and atmospheric conditions. The reconstructed glacier area decreased from 8,083 ± 664 km2 in 2000 to 6,228 ± 572 km2 in 2025, corresponding to a cumulative loss of 22.9 ± 3.6% and a mean loss rate of 80.51 ± 10.51 km2 yr−1. The mean area-loss rates were 51.31 ± 6.51 km2 yr−1 during 2000–2010 and 92.99 ± 15.55 km2 yr−1 during 2010–2025, indicating a higher average loss rate in the later period. By combining satellite-altimetry-derived elevation changes from ICESat, CryoSat-2, and ICESat-2 with the reconstructed time-varying glacier areas, we estimated a mean glacier mass-change rate of −6.51 ± 1.22 Gt yr−1 during 2003–2022. Compared with using a fixed 2003 glacier area, the time-varying-area calculation reduced the estimated mass-loss magnitude by 0.63 Gt yr−1, equivalent to 8.8% of the fixed-area estimate. These results indicate that incorporating time-varying glacier areas can reduce temporal mismatch in altimetry-based mass-change estimation and improve the temporal consistency of regional glacier mass-change assessments.
