Modeling of Rainfall-Runoff Relationship Using Artificial Neural Network Model: A Case of Mille Watershed, Awash Basin, Ethiopia

This study effectively utilizes an Artificial Neural Network (ANN) model to simulate the rainfall-runoff relationship for the Mille watershed in the Awash River basin. The ANN model was trained and cross-validated using MATLAB, supported by the NN toolbox package. Further, the ANN model was developed using feed feed-forward backpropagation algorithm. Hydrometeorological data for the Mille River watershed, which was collected from the Ministry of Water, Irrigation and Energy and the Ethiopian Meteorological Agency was used to train, validate and test the model. Statistical Packages for Social Science (SPSS) software were employed to determine rainfall-runoff correlation and to select the main input for the ANN model. Three ANN models, one with one input variable (rainfall only), two input variables (rainfall and previous runoff) and another with three input variables (rainfall, previous runoff and rainfall), were selected to model the rainfall-runoff relationship of the Mille watershed. The three ANN models were trained, and tested by considering 8 years of data (2005-2012) for model training and 2 years of data (2013 and 2014) for model testing and their performance was evaluated using the Correlation coefficient (R2), Root Mean Square Error (RMSE) and Nash and Sutcliffe Simulation Efficiency (NSE) during training and testing phases. Among the three ANN models tested, the model with rainfall and previous runoff as inputs (M2) achieved the highest performance, with notable results in both the correlation coefficient (R2) and Root Mean Square Error (RMSE). The study's findings highlight the ANN model's capability to accurately model daily rainfall-runoff dynamics and offer valuable insights for managing water resources in the Mille watershed and similar river basins.