Performance Evaluation of a Gated Recurrent Unit (GRU) Model for Rainfall–Runoff Modeling in the Sher Basin Using CAMELS-IND and India-WRIS Data

Rainfall–runoff modeling is essential for understanding watershed dynamics and supporting effective water resource management under changing climatic conditions. This study evaluates the performance of a Gated Recurrent Unit (GRU) model for rainfall–runoff simulation in the Sher River Basin, a sub-basin of the Upper Narmada River system in Madhya Pradesh, India. The basin covers an area of 1479.13 km² and exhibits diverse hydrological and physiographic characteristics.

A 21-year daily dataset (2000–2020) obtained from CAMELS-IND and India-WRIS databases was used, incorporating key hydro-meteorological variables such as antecedent rainfall, potential evapotranspiration, actual evapotranspiration, soil moisture (up to 100 cm depth), and water level observations. Data preprocessing and model development were conducted using Python-based libraries, including NumPy, Pandas, and TensorFlow/Keras. The dataset was divided into training and testing sets using an 80:20 ratio.

The GRU model effectively captured temporal dependencies in the time-series data and demonstrated strong predictive performance. Model evaluation based on the coefficient of determination (R²) indicated that the inclusion of one-day and two-day antecedent rainfall, evapotranspiration, and water level significantly improved model accuracy. The India-WRIS dataset yielded superior results with an R² value of 0.97 compared to CAMELS-IND, likely due to its higher spatial and temporal resolution.

The findings suggest that the GRU model provides a robust and computationally efficient framework for rainfall–runoff modeling. This approach has significant potential for improving flood forecasting, water resource planning, and sustainable watershed management in data-driven hydrological studies.