High-resolution snow depth monitoring is a crucial foundation for precise disaster early warning and optimal water resource management. Traditional snow depth estimation methods mainly rely on passive microwave remote sensing data, but due to their low spatial resolution, they have difficulties capturing the subtle changes in snow depth in complex terrain. Existing deep learning methods mostly adopt single-modal or simple band fusion, failing to fully utilize the complementarity among multi-source data and not considering that terrain factors can lead to misjudgment of the true snow signal. Therefore, this paper proposes a dual-branch intermediate fusion network (TACMF-Net) for high-latitude regions in Asia. By introducing terrain factors (DEM, slope, aspect) and conducting cross-modal feature interaction, it achieves efficient collaboration of multi-source remote sensing data. Research shows that our method has extremely high accuracy and robustness on the self-made multi-source snow depth terrain dataset.

In alignment with carbon neutrality goals, high-precision long-term wind power forecasting is vital for power system scheduling. However, non-stationarity, noise, and underexploited physical relationships limit existing models, especially during critical conditions. This paper addresses these challenges by proposing a Physics-informed Operating Condition Attention Model, Phy-OCA. (1) Phy-OCA employs a dual-branch architecture to decouple input variables into an Energy branch (wind speed and historical power) and an Efficiency branch (wind direction) based on the aerodynamic power conversion mechanism. (2) An Operating Condition Attention module is introduced to evaluate temporal patches via physical indicators, guiding the encoder (centered around scalar Long ShortTerm Memory units) to prioritize critical intervals, such as power ramps and stochastic fluctuations. (3) Wind Speed-Direction Coupling mechanism is developed to characterize how wind directional shifts affect the wind speed capture efficiency of turbines. (4) The study constructs a spatiotemporal dataset comprising 132 real-world wind turbines in Northwest China. Experimental results demonstrate that Phy-OCA outperforms state-of-the-art baselines, achieving significant accuracy improvements under critical conditions. This research offers a high-precision, interpretable solution for practical wind power forecasting.