Learning simple additive weighting parameters for subsidence vulnerability indices in Tehran plain (Iran) by artificial intelligence methods

Abstract

Rates and weightings are specified for every data layer to compensate for the data layer’s regional fluctuation and relative importance accordingly using simple additive weighting frameworks (SAWFs). To this end, the present article follows three learning steps using artificial intelligence (AI): In Level 1, SFL, ANN and SVM models are used to improve weighting factors from regional and observational data. In Level 2, the AI strategy of gene expression programming (GEP) is applied to understand further. In Level 3, the results from Level 1 are reused as inputs and observational data. Using the anticipated impact measurement and monitoring (AIMM) approach, the study paves the ground to create some frameworks for mapping indices (subsidence susceptibility). The main results of the AIMM analysis are three-fold: First, the NSE coefficient for SAWF obtained through land subsidence values is approximately 0.41. Second, the NSE value is enhanced to 0.88 by learning the weighting factors at Level 1 using SFL. Third, the NSE value is even further enhanced to 0.92 by Level 2 learning of the weighting factors via GEP. With the subsidence vulnerability index map, it is possible to manage underground water resources to prevent excessive subsidence in probable areas.