SPI and SPEI Drought Assessment and Prediction Using TBATS and ARIMA Models, Jordan

Abstract

Drought is a complex threat where its propagation is not yet controllable, causing more environmental, social, and economic damage. This research assesses the effects of incessant warming and decrescent precipitation by calculating SPI and SPEI from 1985 to 2021 in the Amman –Zarqa Basin based on five grid points on time and space scales. The study applied the Pearson Correlation Coefficient (PCC) between each one-to-one index at different time scales and the Mann–Kendall test (MKT) to determine trends with different data sources to measure the inferential capturing of historical drought features. Machine learning algorithms are used to predict near-future droughts from 2022 to 2025. TBATS and ARIMA models run diverse input datasets, including observations, CSIC, and CMIP6-ssp126 datasets. The longest drought duration was 14 months. Drought severity and average intensity were found to be −24.64 and −1.76, −23.80 and −1.83, −23.57 and −1.96, and −23.44 and −2.0 where the corresponding drought categories were SPI-12 Sweileh, SPI-9 Sweileh, SPI-12 Wadi Dhullal, SPI-12 Amman Airport, respectively. The dominant drought event occurred between Oct 2020 and Dec 2021. CMIP6-ssp126 can capture the drought occurrence and severity by measuring SPI but did not capture the severity magnitude as the observations (SPI was −2.87 by observation and −1.77 by CMIP6). There are significant differences in drought dimensions between SPI and SPEI, where SPI was more sensitive to drought assessment than SPEI. Using CMIP6-ssp126, ARIMA was more accurate than TBATS, as well as using the observed historical SPEI and CSIC across all stations. The performance metrics ME, RMSE, MAE, and MASE implied significantly promising forecasting models with values of −0.0046, 0.278, 0.179, and 0.193, respectively, for ARIMA and −0.0181, 0.538, 0.416, and 0.466, respectively, for TBATS. The outcomes suggest an increased risk of drought incidents and, consequently, water deficits in the future. Hybrid modelling is suggested for more consistency and robustness of forecasting approaches.