Integrating SCS-CN Method with Machine Learning Models for Rainfall Runoff Modeling: A Case Study in the Upper Geba Catchment

Abstract

Accurate rainfall runoff modelling is essential for effective water resource management, yet it remains challenging due to the complex, nonlinear interaction between meteorological inputs and catchment processes. This study investigates the application of advanced Recurrent Neural Network (RNN) architectures, Long Short Term Memory (LSTM), Gated Recurrent Units (GRU), and Bidirectional LSTM (Bi-LSTM) for daily stream flow simulation. Evaluating both the data driven models and hybrid framework integrates physical hydrological variables, potential evapotranspiration and effective rainfall derived from the soil conservation service curve number (SCS-CN) methods. To effectively capture the catchment storage effect, optimal model input lags were identified using Partial Autocorrelation Function (PACF) analysis. The model was calibrated and validated on daily hydro-meteorological dataset, calibration (1992-2008) and validation (2009-2015). Performance was assessed using Nash-Sutcliffe Efficiency (NSE), Kling-Gupta Efficiency (KGE), Root Mean Square Error (RMSE), and the Coefficient of Determination (R2). The results indicate that the GRU Outperform motherst and alone architectures, achieving the highest Validation performance (RMSE = 1.56m3/s, NSE = 0.891, R2 = 0.897, KGE = 0.944). The I-GRU further improved higher performance during calibration (RMSE =1.16 m3/s, NSE = 0.95) and maintain good performance during validation (RMSE = 1.44 m3/s, NSE = 0.89), demonstrates the strongest performance among t he integrated models. According to flow regimes the I -GRU model perform best achieving the highest NSE for low (0.815) and high (0.872) flows with the lowest RMSE values (1.092 for low and 1.833 for high flows) This finding highlights the benefit of integration and PACF based lag selection for enhancing hydrological understanding and hybridization for process consistency, and this hybrid strengthens the role of advanced recurrent neuralnetworkordeeplearningmodelsinrainfallrunoffmodelingandoperationalwaterresources Movement.