Application of machine learning to aquifer analyses：Locating hydrogeological boundaries with water table monitoring data

Abstract

Groundwater level fluctuations in China are being monitored with unprecedented frequency and density, which drives the need for mining such types of data. In a typical aquifer analysis project, groundwater level data is generally applied after the completion of the aquifer conceptual framework. When the temporal and spatial density of groundwater level data gradually increases, the information gain needs to be effectively transformed into conceptual knowledge of the model. In this study, we propose a method to identify hydrological boundaries based on the groundwater level monitoring data. In this method we discretize space into a triangular mesh using monitoring wells as the initial nodes, and a transformation function gradF is defined to calculate the hydraulic gradient at any given location on the mesh. The hydraulic gradient field is subsequently use to drive an array of randomly scattered particles to obtain the streamline representation of the flow field, which will in turn serve as the basis for deducing and refining the recharge and discharge boundaries of a hydrogeological domain. This method is implemented into the geo-environmental scientific computation platform (EnviFusion-CGS), and a detailed work flow is developed to facilitate the development of the aquifer conceptual model. This method is applied to the hydrogeological investigation of the Dagu aquifer located in Qingdao of Shandong Province, where the spatial distributions and dynamic fluctuations of the hydrogeological boundaries are identified.

Keywords

• high density
• groundwater level monitoring
• conceptual model
• groundwater
• recharge boundary
• discharge boundary