Spatial prediction of loose aquifer water abundance mapping based on a hybrid statistical learning approach
Earth Science Informatics, 14(3):1349 - 1365, 2021.
Summary
This paper develops a method for mapping the water abundance of loose aquifers where borehole observations are sparse. Five indicators describe the aquifer and its investigation: thickness, the relative thickness of sand and clay, drilling-fluid consumption, core recovery, and hydraulic conductivity. Kriging is used to estimate how these quantities vary between observations. The assessment then proceeds in two stages. Fisher discriminant analysis first converts the normalized measurements into a smaller set of informative coordinates suitable for visualization. Three classifiers subsequently predict water-abundance categories from those coordinates, with leave-one-out validation used to make efficient use of the limited data. A case study at the Baodian coal mine illustrates the complete procedure and identifies the areas with the greatest water abundance. The resulting zoning map is intended to support mining decisions and the management of mine-water hazards.