Soil erosion potential hotspot zone identification using machine learning and statistical approaches in eastern India

• Published in: Natural hazards (Dordrecht) 2020-11, Vol.104 (2), p.1259-1294
• Main Authors: Chakrabortty, Rabin, Pal, Subodh Chandra, Sahana, Mehebub, Mondal, Ayan, Dou, Jie, Pham, Binh Thai, Yunus, Ali P.
• Format: Article
• Language: English
• Subjects: Civil Engineering; Earth and Environmental Science; Earth Sciences; Environmental Management; Geophysics/Geodesy; Geotechnical Engineering & Applied Earth Sciences; Hydrogeology; Natural Hazards; Original Paper
• ISSN: 0921-030X; 1573-0840
• DOI: 10.1007/s11069-020-04213-3

Land degradation is very severe in the subtropical monsoon-dominated region due to the uncertainty of rainfall in the long term, and most of the rainfall occurs with high intensity and kinetic energy over short time periods. So, keeping this scenario in view, the main objective of this work is to identify areas vulnerable to soil erosion and propose the most suitable model for soil erosion susceptibility in subtropical environment. The implementation of machine learning and artificial intelligence techniques with a GIS environment for determining erosion susceptibility is highly acceptable in terms of optimal accuracy. The point-specific values of different elements from random sampling were considered for this study. Sensitivity analysis of the predicted models (i.e., analytical neural network, geographically weighted regression and GWR–ANN ensemble) was performed using the maximum causative factors and related primary field observations. The area under curve of receiver operating system reveals precision with 87.13, 89.57 and 91.64 for GWR, ANN and ensemble GWR–ANN, respectively. The ensemble GWR–ANN is more optimal than the GWR, ANN for determining water-induced soil erosion susceptibility. The process of soil erosion is not a unidirectional process, so the multidimensional impacts from the conditioning factors have to be determined precisely by considering the maximum possible factors as well as selecting optimal models for specific regions.