Simulation of the rainfall-runoff relationshıp in the AL-Kabir AL-Shamali river basin (Latakia Governorate) usıng artifıcial neural networks
Keywords:
Rainfall-Runoff Modeling, Artificial Neural Networks (ANN), Multilayer Perceptron (MLP), Al-Kabir Al-Shamali River BasinAbstract
This study aimed to simulate the dynamic and non-linear rainfall-runoff relationship in the Al-Kabir Al-Shamali River Basin (Latakia Governorate) for the period (2001-2011) using Multilayer Perceptron (MLP) artificial neural networks. The methodology relied on monthly hydroclimatic data (precipitation, potential evapotranspiration, runoff) to develop five predictive models. These models were trained using the Levenberg-Marquardt backpropagation algorithm as an optimization tool, and their performance was evaluated using the correlation coefficient (r) and the mean squared error (MSE). The results demonstrated the superiority of the ANN-4 model with a (5-15-1) architectural structure over the other models, achieving the highest predictive accuracy. The model recorded correlation coefficients of 0.94, 0.97, and 0.91 during the training, validation, and testing phases, respectively, while attaining the lowest mean squared error (MSE = 0.10) in both the validation and testi000000 ng phases. However, the results highlighted the model's limitation in accurately simulating peak flows. Despite its constraint in modeling peak events, the developed ANN-4 model proves to be an efficient, robust, and applicable predictive tool to support sustainable water resources planning and management processes in the studied basin.