Mapping potential groundwater-dependent ecosystems in Central Mexico: Expert knowledge and machine learning approaches

Groundwater sustainability requires meeting current and future human needs while maintaining groundwater discharge and interactions with Groundwater-Dependent Ecosystems (GDE). The first step in including GDEs in water management policies is identifying their location and extent in the landscape. Approaches to mapping GDE include those based on expert knowledge and machine learning methods.  

Meanwhile, Mexico is one of the countries currently facing major groundwater challenges due to intensive groundwater abstraction, land use change, and climate change, putting to risk the structure and function of GDEs. Therefore, GDE mapping is needed in Mexico to facilitate their inclusion in water management.

For this purpose, this study evaluated the performance of the Analytic Hierarchy Process (AHP) method and the Logistic Regression (LR) method to map GDEs using topographic, hydrogeological, structural, and vegetation variables obtained from remote sensing products and geospatial data in a study area located in Central Mexico. The two methods were compared by the AUC and ROC curve based on ground-truth data obtained from springs and groundwater-dependent wetland inventories.

The results show insights into each method's predictive power in identifying areas associated with GDEs, with AHP emphasizing the prioritization of criteria based on expert knowledge and LR revealing statistical relationships within the dataset.

The use of different explanatory variables and methods enables the development of distinct frameworks for GDE mapping, each with distinct strengths. Nevertheless, this study shows different approaches that can be successfully applied by decision-makers to map GDEs at local and regional scales and ease their inclusion into water management policies.