Evolutionary polynomial modeling for interpretable drought prediction and resilient resource management

Abstract

Droughts are natural hazards that exist in nature and can have a serious impact on the environment and society, which includes water shortages, crop failures, fires and, in some cases, soil manipulation. To assess and predict droughts, various methods, such as the Standardized Precipitation Index (SPI), were designed to segregate drought trends and excess rainfall over a period ranging from 3 to 48 months. This study proposes an innovative approach to predicting drought use, the Evolutionary Polynomial Expansion with Feature Selection (EPEFS) model, a hybrid method that integrates polynomial regression with feature selection to increase accuracy and interpretability. The methodology was applied to historical precipitation data from six meteorological stations in Türkiye, covering the period from 1971 to 2016. The drought index Standardized Precipitation Index (SPI) was used as the primary indicator, with predictions made for three different time scales: SPI-3, SPI-6 and SPI-12. Furthermore, a time series cross-validation strategy was employed to ensure performance assessment. The EPEFS model obtained R${}^2$ coefficients of 0.880, 0.903 and 0.929 for SPI-3, SPI-6 and SPI-12, respectively, surpassing the other models analyzed. Furthermore, the model presented less complexity in the generated expressions. The results suggest that the EPEFS model holds promise as a robust and interpretable tool for drought forecasting, with potential applications in early warning systems and mitigation strategies.