GROUNDWATER SYSTEM CHARACTERIZATION AND MAPPING USING WATER-QUALITY AND HYDROGEOPHYSICAL DATA AND MACHINE-LEARNING WORKFLOWS

摘要

Multi-step machine-learning (ML) workflows utilize fusion, clustering, and estimation operations on water-quality and hydrogeophysical data to derive hydrostratigraphic units (HSUs). The term HSU refers to spatially-distinct units based on integrating combinations of biologic, chemical, and physical characteristics associated with flow and transport. Data used in this process involves combinations of measured, derived, interpolated, and estimated values. Data fusion results in a hypersurface following the training of a self-organizing map (SOM) with these data. The application of Davies-Bouldin criteria to K-means clustering of SOM nodes determines the number and location of HSUs. Estimation is handled by iterative least-squares minimization of the SOM quantization and topographical errors. The application of ML workflow for characterization of HSUs is done for data collected from different hydrogeologic settings. In workflow 1, the ML sequence is applied to surface-water and groundwater system attributes (aquatic biology, aqueous chemistry, dissolved oxygen, temperature, hydraulic properties, and lithologies) measured across the Southland region, NZ. In workflow 2, the ML sequence is applied to airborne (1D conductivity profiles interpolated to boreholes) and borehole (lithoiogy, induction conductivity, natural gamma, hydraulic conductivity, water content, pH, EC, and TDS) data from locations across a portion of the Broken Hill Managed Aquifer Recharge region of New South Wales, AU. The respective 1D conductivity profiles and hydraulic conductivity are products derived following numerical inversion of AEM and nuclear magnetic resonance. Performance metrics and validation are used to test each step of both workflows.