Abstract
Abstract
Land subsidence due to groundwater recharge and withdrawal has emerged as a widespread, societally impactful consequence of human development. However, the relationship between groundwater depletion and land subsidence is incredibly complex. It depends on a suite of environmental variables and non-linear processes at and beneath the land surface. One major constraint on land surface subsidence and rebound is the complex geologic matrix and depth-varying responses of distinct geologic features to groundwater drainage and/or filling. In many regions globally, sufficient soil geology information at depth is unavailable. In this work, for a study section of California's Central Valley aquifer, we show that time-series characteristics of surface deformation detected using interferometric SAR (Synthetic Aperture Radar) observations, when combined with deep neural network learning approaches, can successfully provide screening-level insight into integrated soil-column responses to predict a map of basic geologic composition and coarse grain percentages of an aquifer. As aquifer grain sizes dictate aquifer elasticity, elucidating grain-size composition based on surface signals is valuable for predicting future aquifer behavior and groundwater depletion for many of the world's critical aquifers.