Development of intelligent machine learning and Jenkins box models based on the full ensemble mode decomposition method for meteorological drought modeling (Case study: Khuzestan Province)

Drought is one of the natural hazards, especially in arid and semi-arid regions. Khuzestan Province is highly vulnerable to drought due to its strategic geographical location and strong dependence on water resources. Therefore, in the present study, the analysis and prediction of meteorological drought in Khuzestan Province was investigated with the intelligent machine and Jenkins Box GPR models CEEMD- and CEEMD-SARIMA during the 30-year statistical period (1989-2020). To assess drought conditions, the Standard Precipitation Index (SPI) obtained from data from eight synoptic stations in Khuzestan Province was used. In the next step, the modeling results were compared with each other using the aforementioned models and goodness of fit indices. The results indicated that the CEEMD-GPR model is very efficient in estimating the SPI index in Khuzestan province. Also, the long-term SPI time windows had higher accuracy than the short-term time windows. For example, at Omidiyeh station, using the 12-month SPI instead of the 1-month SPI reduced the RMSE and MAE values ​​from 0.178 and 0.097 to 0.167 and 0.087, respectively. In addition, the R and NS values ​​also increased from 0.954 and 0.969 to 0.963 and 0.974. In general, it can be stated that the CEEMD-GPR model is able to learn a more complex dynamic structure of the data by using the components extracted from the CEEMD decomposition. Accordingly, the results of this study show that hybrid models based on CEEMD decomposition, along with machine learning, are powerful and highly efficient tools for analyzing and predicting meteorological drought in arid and semi-arid climates.