Predicting Standardized Precipitation Evapotranspiration Index (Spei) Using SVM and Arima Models: A Comparative Study

This study evaluates the predictive performance of Support Vector Machines (SVM) and Autoregressive Integrated Moving Average (ARIMA) models in forecasting the Standardized Precipitation Evapotranspiration Index (SPEI) for three critical agricultural regions in Egypt: Nubariyah, Wadi Al-Natrun, and Al-Boseli. Accurate SPEI forecasting is necessary for effective water management and agricultural planning, especially in arid regions. Through comprehensive analysis involving time series decomposition and model evaluation using Mean Squared Error (MSE) and Mean Absolute Error (MAE), ARIMA models consistently outperformed SVM models across all locations. The ARIMA (1,1,1) model showed superior predictive accuracy, with MSE reductions ranging from 1.4% to 14% over the SVM models. In Nubariyah, the ARIMA model achieved an MSE of 1.7499 compared to 1.7746 for the SVM model. In Wadi Al Natrun, the ARIMA model's MSE was 2.0735, significantly lower than the SVM model's 2.4113. In contrast, in Al Boseli, the ARIMA model recorded an MSE of 1.8033 versus 2.0844 for the SVM model. The decomposition of SPEI values into trend, seasonal, and residual components revealed a long-term trend towards increasing dryness over the 30 years, alongside regular annual fluctuations. These insights are essential for understanding climatic behavior and informing water management strategies. The ARIMA model's superior performance underscores its effectiveness in anticipating drought conditions and optimizing water usage. Research should explore advanced models like Recurrent Neural Networks (RNN) to enhance forecasting accuracy further and expand the analysis to additional regions with more recent data to validate these findings, thereby improving drought prediction and water resource management.