<p>Effective water management in snow-dominated mountainous regions presents significant challenges due to the complex and non-linear interactions between meteorological variables, snowmelt processes, and river discharge. The Gilgit River Basin, a vital tributary of the Upper Indus Basin in northern Pakistan, is particularly sensitive to climatic fluctuations, with snowmelt contributing substantially to river flow. Traditional physically based hydrological models often struggle to capture these dynamics, especially in data-scarce environments. This study explores the application of Artificial Neural Networks (ANN), specifically Recurrent Neural Networks (RNN), for modeling and predicting daily river discharge in the Gilgit River Basin. The research integrates a 16-year dataset (2005–2020) including precipitation, temperature, snow cover area (SCA), and lagged discharge values, obtained from national agencies and MODIS satellite products. Five ANN architectures of increasing complexity were developed and tested using Python in a Jupyter Notebook environment. The models were evaluated using four statistical performance metrics: Nash–Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Mean Squared Error (MSE), and the Coefficient of Determination (R²). Architect 3, which incorporated SCA and previous-day discharge data, outperformed all other architectures with NSE = 0.9663 and RMSE = 0.0272. The model successfully captured seasonal discharge trends and high-flow events, indicating its robustness for snowmelt-driven hydrological systems. These findings demonstrate the effectiveness of RNN-based models for hydrological forecasting in high-altitude basins. The proposed approach offers a valuable decision-support tool for water resource management, flood prediction, and climate adaptation planning in Pakistan’s mountainous regions and similar snow-fed river basins globally.</p>