Streamflow Forecasting Using a Hybrid Modelling Coupled with Different Components

Accurate streamflow forecasting remains a challenging task in water resources management—such as power generation, water supply, and flood control—particularly in catchments dominated by snow and glacier melt. This paper introduces a hybrid modeling approach that incorporates rainfall, snowmelt, and glacier melt streamflow components to enhance forecasting accuracy. The hybrid models combine a Long Short-Term Memory (LSTM) network, the Snowmelt Runoff Model (SRM), and a degree-day model to predict the respective streamflow components from rainfall, snowmelt, and glacier melt. These models were tested in two watersheds: Yellow River and North Platte River. Results showed that the hybrid models significantly outperformed individual models by integrating different physical processes, thereby better capturing the complexities of hydrological dynamics. In particular, during the snowmelt season, the hybrid models accounted for the influence of snowmelt and glacier melt on streamflow, greatly improving forecasting accuracy.