Rainfall Forecasting in Muscat Governorate Using Artificial Neural Networks and Hybrid Modeling Approaches

Abstract

This study presents an advanced method for forecasting seasonal and annual rainfall in the Muscat Governorate of Oman, using artificial neural network (ANN) models and a hybrid approach combining wavelet decomposition with neural learning. Historical rainfall data from 1872 to 2017, sourced from Oman’s meteorological records, were analyzed to uncover long-term patterns, seasonal variability, and drought trends using the Standardized Precipitation Index (SPI). Initial statistical evaluations revealed high interannual variability and a slight declining trend in total annual rainfall, with the majority of precipitation concentrated in winter months. Artificial neural networks were developed to predict both annual and monthly rainfall based on autoregressive inputs and prior-month rainfall values. While the ANN models demonstrated moderate skill, limitations were observed in capturing extremely wet or dry years. To address this, a hybrid Wavelet-ANN model was constructed, enabling decomposition of rainfall signals into low- and high-frequency components for more targeted forecasting. The hybrid model showed improved performance, offering a more nuanced understanding of rainfall dynamics. Despite promising results, the models underscore the need for incorporating global climate predictors such as ENSO and IOD to improve forecast accuracy. The study concludes that ANN and hybrid methods provide a practical and scalable framework for enhancing regional rainfall forecasting capabilities, with significant implications for water resource planning and climate resilience in arid regions like Muscat.