How can statistical and artificial intelligence approaches predict piping erosion susceptibility?

Abstract It is of fundamental importance to model the relationship between geo-environmental factors and piping erosion because of the environmental degradation attributed to soil loss. Methods that identify areas prone to piping erosion at the regional scale are limited. The main objective of this research is to develop a novel modeling approach by using three machine learning algorithms—mixture discriminant analysis (MDA), flexible discriminant analysis (FDA), and support vector machine (SVM) in addition to an unmanned aerial vehicle (UAV) images to map susceptibility to piping erosion in the loess-covered hilly region of Golestan Province, Northeast Iran. In this research, we have used 22 geo-environmental indices/factors and 345 identified pipes as predictors and dependent variables. The piping susceptibility maps were assessed by the area under the ROC curve (AUC). Validation of the results showed that the AUC for the three mentioned algorithms varied from 90.32% to 92.45%. We concluded that the proposed approach could efficiently produce a piping susceptibility map. Graphical abstract Unlabelled Image Highlights • Piping susceptibility can be predicted with SVM, MDA, FDA, and LR models. • Piping erosion occurrence are mainly controlled by silt content based on ME model. • Bulk density had the most influence on the occurrence of piping based on LR model. • The SVM model shows the best result in piping prediction (SVM = 93.15%). [ABSTRACT FROM AUTHOR]