A Novel Approach to Aquifer Classification Using Hysteresis Loop Analysis and Deep Learning for Sustainable Groundwater Management

Aquifer classification plays a pivotal role in understanding groundwater dynamics and informing sustainable water resource management, especially in regions under significant stress from over-extraction. This study presents a novel remote sensing-based methodology for classifying aquifers represented by monitoring wells within the study area. The approach integrates stress-strain analysis, incorporating deformation data derived from Interferometric Synthetic Aperture Radar (InSAR) and groundwater head measurements from monitoring wells, utilizing advanced deep-learning techniques. Groundwater data from piezometric wells are utilized to create image-based representations of hysteresis loops derived from stress-strain diagrams, capturing aquifer deformation under varying drawdown and recovery cycles. A convolutional neural network is applied to extract high-dimensional features characterizing aquifer response dynamics. Principal component analysis is then employed to reduce dimensionality, highlighting the most significant features driving classification. Finally, unsupervised clustering methods are used to group piezometric wells, revealing distinct aquifer types and deformation patterns. The proposed methodology is tested in three hydrologically and geologically diverse regions of Iran: Shabestar, Urmia, and Neyshabur Plains. In the Shabestar and Urmia Plains, located near the hypersaline Lake Urmia, intensive groundwater extraction has severely strained local hydrological and ecological systems, contributing to declining lake levels and increased stress on water resources. Similarly, in the Neyshabur Plain in northeastern Iran, characterized by its arid to semi-arid environment and intricate geological features, excessive groundwater use has led to significant aquifer depletion and land subsidence. The proposed approach effectively identifies different aquifer types, analyzes the balance between elastic and inelastic deformation, and determines aquifer responses to varying degrees of groundwater extraction. By integrating InSAR-based deformation monitoring of ground surface with advanced deep learning techniques, the study provides a comprehensive framework for aquifer system characterization. The findings are particularly valuable for regions with scarce geological and hydrological data, offering insights to guide sustainable groundwater management practices, mitigate environmental degradation, and support effective decision-making.