Development of a data-driven process to cost-efficiently estimate hydraulic property of aquifer

The pattern of groundwater level fluctuation is a representation of the hydraulic characteristics of the aquifer (e.g., hydraulic conductivity, recharge capacity, etc.). In this study, the data-driven-based process for estimating the hydraulic properties of the aquifer was proposed. The process consists of three main steps: 1) classifying the aquifer showing the similar pattern of the groundwater level fluctuation based on denoising autoencoder (DAE), 2) extracting the coordinate of each groundwater level fluctuation in the dimensionality-reduced subspace (i.e., coding layer in DAE network), and 3) constructing the regression model with the subspace coordinate and the known hydrological properties as input and output variables, respectively. In this study, the regression model to predict the hydraulic conductivity is constructed where the property of the applied groundwater level data is measured by fitting PP-model (Park and Parker, 2008). For validation, the proposed process was applied to the actual groundwater level data acquired from 65 monitoring stations in South Korea. From the results, it is found that DAE is effective to classify the monitoring stations showing the similar fluctuation patterns in the dimensionality-reduced subspace. In addition, the regression model based on the result of DAE shows reasonable performance on the prediction of hydrological conductivity. Therefore, it can be confirmed that the hydraulic property of the unknown aquifer can be evaluated without hydraulic test or optimization using the analytical solution by the proposed process. The proposed process can be cost-efficiently applied to evaluate the aquifer characteristics in terms of time and money, which can also contribute to managing the sustainability of groundwater resources in response to drought, vulnerability on the contaminant sources, etc.