Rainfall Prediction in Khorasan Razavi Stations Using a Hybrid Neural Network and Genetic Algorithm Approach

Accurate rainfall prediction is crucial for effective water resource management, especially in arid and semi-arid regions. This study proposes a novel hybrid approach, combining the Non-linear Auto Regressive with eXogenous inputs (NARX) neural network with a Genetic Algorithm (GA) for parameter optimization, aiming to improve daily rainfall prediction in Khorasan Razavi province, Iran. The performance of the proposed NARXGA model was compared with several benchmark models, including traditional time series models ARIMA, Holt-Winters Exponential Smoothing (HWES), and machine learning models, such as LSTM, CNN1D and the standalone NARX network. The models were trained and tested using five years of daily meteorological data from Mashhad. The results showed that the NARXGA model achieved the lowest Mean Squared Error (MSE) on both the training and test datasets, with values of 9.7453 and 11.5565, respectively, thus showing that the method can more effectively capture the non-linear patterns in rainfall data. A convergence analysis of the GA was also provided, as well as histograms of the error distributions, which further validated the superior performance of the proposed NARXGA model. This research highlights the potential of hybrid AI models for enhancing rainfall prediction accuracy and providing valuable insights for water management and drought mitigation in arid and semi-arid regions.