Data-Driven Modeling of Groundwater Level with Least-Square Support Vector Machine and Spatial-Temporal Analysis

Investigation of groundwater level is considered a prominent research topic for the study of underground hydrologic system. Due to the complexities of underground geological structure, the accuracy of real-time ground water level prediction is limited. In this study, a novel two-phase data-driven framework to model the time-series groundwater level with spatial-temporal analysis and least square support vector machine is proposed. Groundwater data collected from four monitoring sites in the northern region of United Kingdom is utilize in this study. In phase I, the time-series analysis is conducted to study the temporal characteristics of the groundwater. Based on the time-series analysis, least square support vector machine is performed to construct the prediction model to forecast the future groundwater level. In phase-II, the spatial correlation between the water levels in four sites are computed to construct a comprehensive model regarding the interrelation between the monitoring sites. Computational results illustrated the outperformance of least square support vector machine in predicting time-series groundwater levels compared with other state-of-arts machine learning algorithms. It has been demonstrated that the spatial-temporal model may serve as an applicable approach for the future research of groundwater resources.