Introducing a novel approach for assessment of groundwater salinity hazard, vulnerability, and risk in a semiarid region

Abstract

The application of machine learning approaches to improve groundwater salinity risk mapping is limited despite all potential advantages. Therefore, there is an ongoing need for investigations that present new techniques, like machine learning, validated against conventional methods. These advances are particularly important for arid and semiarid regions, such as the Bakhtegan Basin (southern Iran), where groundwater use may far outweigh recharge, but groundwater management improvements are essential. To address these needs, groundwater salinity hazard, vulnerability, and risk maps were investigated using an integrated application of statistical (i.e., frequency ratio and statistical index models), machine learning (i.e., random forest and classification and regression trees algorithms), and decision-making models (fuzzy analytic hierarchy process (FAHP)). Results showed that the risk of groundwater salinity was high in the central areas of the region as well as at the margins of Bakhtegan Lake and irrigated farming lands. Based on the modeling results in the testing phase, it was found that the frequency ratio and random forest models exhibited better performance, with Nash–Sutcliffe efficiency metrics of 0.73 and 0.70, respectively, compared to the classification and regression trees and statistical index models, which had Nash–Sutcliffe efficiency metrics of 0.65 and 0.63, respectively. The innovative techniques developed in the current work accurately identified variables such as normalized difference salinity index, distance from mines, and land use associated with the highest weight values (0.59, 0.14, and 0.11, respectively) compared to other variables to identify groundwater salinity areas of vulnerability. The application of the fuzzy analytic hierarchy process method in developing vulnerability maps, integrating them with hazard maps, and developing groundwater salinity risk maps was innovative in this study. The innovative approach is transferrable for groundwater salinity susceptibility, vulnerability, and risk assessments globally in the study area and similar settings.