A novel implementation of pre-processing approaches and hybrid kernel-based model for short- and long-term groundwater drought forecasting

Abstract

Groundwater drought, as a form of hydrological drought, embodies the distinctive characteristics of the aquifer and human-induced disruptions within the hydrological system. The intricate nature of groundwater flow systems, coupled with challenges in acquiring field observations related to aquifers, poses significant challenges in quantitatively characterizing groundwater drought. The present paper presents a novel contribution to the time series forecasting of groundwater drought through state-of-the-art integrated GWO-SVM models. The Standardized Groundwater Level Index (SGI) was employed to monitor groundwater drought in one of the critical aquifers in Iran, and forecasts were conducted for various horizons, including short-term (3 months: t + 3), mid-term (9 months: t + 9), and long-term (12 months: t + 12) periods. Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Variation Mode Decomposition (VMD), Empirical Wavelet Transform (EWT), Empirical Fourier Decomposition (EFD), and Discrete Wavelet Transform (DWT) were further incorporated as pre-processing techniques to enhance forecasting accuracy. The trend analysis findings indicated that out of the 20 observation wells assessed, 15 observation wells (P1–P15) located in the western part of the aquifer showed a negative trend. The SOM method clustered the aquifer into five clusters, with well P8, representing cluster 1, demonstrating the highest accuracy in forecasting groundwater drought. The overall results demonstrated the significant impact of pre-processing on enhancing the forecasting accuracy of groundwater drought. The VMD-GWO-SVM model provided superior performance compared to all employed models in short to long-term horizons, achieving NSE values of 0.955, 0.915, and 0.838 for short-term, mid-term, and long-term periods, respectively.