Bayesian Model Averaging Ensemble Approach for Multi-Time-Ahead Groundwater Level Prediction Combining the GRACE, GLEAM, and GLDAS Data in Arid Areas

Accurate groundwater level (GWL) prediction is essential for the sustainable management of resources. However, GWLs remains a challenge due to insufficient data and complicated hydrogeological system. In this study, we investigated ability Gravity Recovery Climate Experiment (GRACE) satellite data, Global Land Evaporation Amsterdam Model (GLEAM) Data Assimilation System (GLDAS) publicly available meteorological in 1-, 2-, 3-month-ahead GWL using three traditional machine learning models (extreme machine, ELM; support vector SVR; random forest, RF). Meanwhile, further developed Bayesian model averaging (BMA) by combining ELM, SVR, RF avoid uncertainty single improve predicting accuracy. The validity forcing BMA were assessed monitoring wells Zhangye Basin Northwest China. results indicated that applied could be treated as validated inputs predict up 3 months ahead achieved high accuracy (NS > 0.55). significantly performance models. Overall, reduced RMSE testing period about 13.75%, 24.01%, 17.69%, respectively; while it improved NS 8.32%, 16.13%, 9.67% prediction, respectively. analysis also verified reliability multi-time-ahead predicting. This highlighted efficiency satellite-based substitute machine-learning-based particularly areas with or missing data. ensemble strategy can serve powerful reliable approach when risk-based decision making needed lack relevant impedes application physical