Land subsidence susceptibility linked to unsustainable agriculture and groundwater depletion in water-stressed regions

Abstract Land subsidence is a growing hazard with far-reaching socioeconomic, environmental, and safety implications, especially in densely populated and water-stressed regions. In areas where groundwater is the primary water source, prolonged overextraction causes declines in water tables and destabilizes the ground, leading to widespread land deformation. We developed a multi-scale deep neural network model, an artificial intelligence technique that integrates environmental and anthropogenic predictors across predictor-specific spatial scales, to identify the key drivers of land subsidence in water-stressed landscapes. We used 631 confirmed subsidence presence points paired with 631 pseudo-absence points to train our model. For each point, we extracted a set of environmental variables, including elevation (DEM), slope, plan and profile curvature, NDVI, topographic wetness index, distance to river, population density, well density, geology, land use, and soil type. Our results show that agriculture within a 4,000 m radius is the strongest predictor of subsidence susceptibility, followed by well density at 2,000 m and Qft2 geology at 3,000 m. Susceptibility increases sharply with rangeland cover up to 20%, while the presence of rock outcrops in Inceptisols leads to a distinct rise in susceptibility. Areas with the highest predicted subsidence susceptibility align with regions of intensive agriculture in the western, northwestern, and northern parts of the study area. Moderate- susceptibility zones are associated with dense well infrastructure and sensitive geological formations, whereas lower overall predicted susceptibility characterizes many rangeland-dominated landscapes, reflecting the combined influence of all predictors rather than any single variable’s local contribution. These patterns highlight the urgent need for sustainable groundwater management policies, particularly in agricultural landscapes built on subsidence-prone geology, to prevent further environmental degradation and infrastructure vulnerability.