Integrating remote sensing and deep learning forecasting model: A fluid-environment interface study

This study proposes an approach that integrates remote sensing (RS) data with deep learning (DL) techniques to predict variations in lake area. The model incorporates key input variables, including historical rainfall and glacier temperature data, which are the primary contributors to lake inflow. As global warming accelerates, changes in glacier melting patterns and increased evaporation due to rising temperatures are significantly impacting water resource ecosystems. To address this, the study leverages multiyear historical data on correlated variables-glacier temperature, rainfall, and lake area-to forecast future lake area dynamics. A sequential deep learning model based on gated recurrent units (GRUs) is developed and evaluated using current RS data. The model achieves strong predictive performance, with a Root Mean Square Error (RMSE) of 0.005, a Mean Absolute Error (MAE) of 0.004, and a coefficient of determination ($R^{2}$) of $\mathbf{0. 9 9}$. These results demonstrate the effectiveness of integrating RS data with DL models for environmental forecasting, particularly when supported by wellestablished correlations among input variables.