Groundwater Salinity Risk Assessment in the Southern Plains of the Bakhtegan Watershed Using Statistical and Data Mining Models and Fuzzy Hierarchical Analysis Process

Abstract

Issues related to implanting national policies for reducing natural hazards in developing countries are seen as an important and worrying challenge. In this research, the factors affecting the vulnerability, hazard, and risk of groundwater salinity in the southern plains of the Bakhtegan watershed with an area of 6137 km2 were assessed using statistical index models, Classification and Regression Tree (CART), and fuzzy analytic hierarchy process (FAHP). At first, by assessing the data of groundwater quality from 124 wells in the watershed, the points of groundwater salinity occurrence (based on the electrical conductivity parameter) were recognized. After calculating and preparing basic maps related to each of the indicators, we looked for a way for selecting indicators to identify key indicators. To this end, the recursive feature removal (FRE) method was utilized for finding and selecting the main combination of indicators to prepare a map of groundwater salinity hazards. Accordingly, after determining the significance of each indicator by calculating the jackknife test, the salinity hazard map of groundwater was predicted using two models (SI) and (CART). Also, the degree of vulnerability of the region to the salinity of the groundwater was determined using the (FAHP) process. The sensitivity, specificity, accuracy and kappa coefficient statistics were used to evaluate the performance of two-variable models. Also, in order to evaluate the performance of multivariate models, fore criteria of coefficient of determination (R2), conformity coefficient (CC), Kling-Gupta efficiency (KGE), and correlation coefficient (COR) were used. The results of groundwater salinity class risk study showed that 4.13 square kilometers of irrigated areas are in a very high-risk class, which among the various land uses with the highest vulnerability is related to this type of land use. Based on the results of this study, it can be stated that the development of comprehensive management plans, monitoring of executive activities, including change of uses, or taking into account the prevailing conditions in the implementation of rehabilitation operations can help control and manage risk in these areas.

Keywords

• bivariate and multivariate models
• salinity
• risk assessment
• recursive feature elimination