Groundwater Level Monitoring Networks Design with Machine Learning Methods

Abstract Collecting and analyzing groundwater data are essential to evaluate the regional groundwater flow patterns and quality under existing and future temporal/spatial hydraulic stressors. These data also provide researchers with the information required to calibrate the groundwater models and study the interaction between surface water and groundwater resources. In general, frequent and well-organized groundwater monitoring improves the comprehension of groundwater systems and enhances groundwater resource management. Both surface water and groundwater quality monitoring networks have been extensively studied and presented in literature reviews. In contrast, there is much less literature focused on groundwater level monitoring networks. In many regions, groundwater level monitoring networks are limited and do not provide sufficient data for decision-making purposes. This gap in data availability is due to multiple reasons but includes financial constraints and the limited interest in those areas that rely more heavily on surface water resources for human consumption and industrial purposes. Here, we introduce methods employing K-means clustering and/or Relevance Vector Machine to design an optimal groundwater level monitoring network. The machine learning algorithms utilize the hydrogeological datasets obtained from the initial groundwater MODFLOW models and consider the uncertainties in the aquifer characterization through stochastic simulations. The result of this research is three groundwater level monitoring networks which the optimal one is selected based on the minimum modeling error. The network configurations are demonstrated in terms of the number and location of the observation wells. The proposed monitoring network improves the procedure of groundwater modeling and significantly reduces modeling errors.