Soil erosion and landslide susceptibility insights based on hierarchical clustering and multilayer perceptron networks: a Nigerian case study

• Published in: International journal of environmental science and technology (Tehran) Vol. 20; no. 10; pp. 10763 - 10786
• Main Author: Egbueri, J. C.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.10.2023
• Subjects: Aquatic Pollution; case studies; cohesion; Earth and Environmental Science; Ecotoxicology; Environment; Environmental Chemistry; Environmental Science and Engineering; landslides; neural networks; Nigeria; Original Paper; prediction; risk; soil erosion; Soil Science & Conservation; Ursidae; Waste Water Technology; Water Management; Water Pollution Control; Nigeria; Multilayer perceptron networks; Machine learning techniques; Hierarchical clustering; Soil erosion and landslides; Southeastern Nigeria
• ISSN: 1735-1472; 1735-2630
• DOI: 10.1007/s13762-022-04714-7

The usefulness of machine learning algorithms (MLAs) in environmental sciences, geosciences, and hazard assessments cannot be overemphasized. MLAs provide insights for reliable prediction and understanding of the processes and risk of hazards at various scales. This paper integrated multiple MLAs, including hierarchical clusters (HCs) and multilayer perceptron networks (MPN1, MPN2 and MPN3), in investigating soil erosion and landslide susceptibility in Southeast Nigeria. The developed HCs efficiently ranked the susceptibility and risks of the studied gullies. MPN1 predicted the cohesion and ϕ of the gully soils, with its area under curve ranging from 0.857 to 1.000 for cohesion and R 2 for ϕ at 0.643. MPN2 predicted the factor of safety (FS) for dry and wet climatic conditions (based on geotechnical variables), having R 2 in the range of 0.880–0.902. MPN3 predicted the FS for both conditions (based on estimated geometrical variables and porewater pressures) and expressed R 2 ranging from 0.655 to 0.685. The findings of this study bear positive significances for erosion and landslide risk analysts and managers. Moreover, the MPN models support the motion that high accuracy thresholds would be realized when the findings are utilized for risk planning and management. The use of these MLAs has proven to be economical and much more usable and manageable than tedious and expensive conventional experimental works. However, continued research is encouraged in this subject area, in order to develop newer models that would consider variety of other erosion and landslide causative factors and possibly validate or update the findings of the present paper.