Accurate monitoring of snow and ice cover is essential for climate research and disaster management, but conventional remote sensing methods often struggle in complex terrain and fog-contaminated conditions. To address the challenges of high-resolution UAV-based snow and ice segmentation—including visual similarity, fragmented spatial distributions, and terrain shadow interference—we introduce IceSnow-Net, a U-Net-based architecture enhanced with three key components: (1) a ResNet50 backbone with atrous convolutions to expand the receptive field, (2) an Atrous Spatial Pyramid Pooling (ASPP) module for multi-scale context aggregation, and (3) an auxiliary path loss for deep supervision to enhance boundary delineation and training stability. The model was trained and validated on UAV-captured orthoimagery from Ganzi Prefecture, Sichuan, China. The experimental results demonstrate that IceSnow-Net achieved excellent performance compared to other models, attaining a mean Intersection over Union (mIoU) of 98.74%, while delivering 27% higher computational efficiency than U-Mamba. Ablation studies further validated the individual contributions of each module. Overall, IceSnow-Net provides an effective and accurate solution for cryosphere monitoring in topographically complex environments using UAV imagery.

Sea surface salinity (SSS) observations play a crucial role in the study of ocean circulation, climate variability, and marine ecosystems. However, current satellite SSS products suffer from systematic biases due to factors such as radio frequency interference (RFI) and land contamination, resulting in fundamental limitations to their application for SSS monitoring. To address this issue, we propose a physics-informed neural network (PINN) approach that directly integrates radiative transfer physical processes into the neural network architecture for SMAP L2 SSS bias correction. This method ensures oceanographically consistent corrections by embedding physical constraints into the forward propagation model. The results demonstrate that PINN achieved a root mean square error (RMSE) of 0.249 PSU, representing a 5.3% to 8.5% relative performance improvement compared to conventional methods—GBRT, ANN, and XGBoost. Further temporal stability analysis reveals that PINN exhibits significantly reduced RMSE variations over multi-year periods, demonstrating exceptional long-term correction stability. Meanwhile, this method achieves more uniform bias improvement in contaminated nearshore regions, showing distinct advantages over the inconsistent correction patterns of conventional methods. This study establishes a physics-constrained machine learning framework for satellite SSS data correction by integrating oceanographic domain knowledge, providing a novel technical pathway for reliable enhancement of Earth observation data.