Climate change and global population growth have intensified extreme weather events such as floods and droughts. These kinds of events make forecasting process difficult and cause significant damages. To overcome these challenges, hydroelectric reservoirs have been constructed to store water for production, flood control, and electricity generation. In this research, we present an integrated forecasting framework that combines the modern Visual Geometry Group Network-19 and Seasonal AutoRegressive Integrated Moving Average with eXogenous factors to process multi-source data, including numerical data and satellite imagery. Experimental results from the An Khe hydroelectric reservoir in Vietnam demonstrate that this forecasting framework achieves higher accuracy compared to traditional models, highlighting that the integration of multiple data sources significantly enhances the forecasting performance. The proposed framework achieves a smaller prediction error compared to traditional models of 11.7% for MAE, 26.8% for MSE, and 14.5% for RMSE.