Assessment of landslide hazard risk in Kenya based on different statistical models

Abstract

Kenya is an important fulcrum of China's Belt and Road initiative in east Africa. However, due to its plateau rift terrain and aboriginal rain and drought season, geological disasters occur frequently in Kenya. The study used historical landslide data in Kenya as samples and selected several evaluation indexes, including elevation, slope, aspect, landform, plane curvature, soil type, annual average rainfall, stream power index, terrain witness index, and land use type. The landslide risk in Kenya was evaluated based on the information value model (IV), logistic regression model (LR), and extreme learning machine model (ELM), with the ELM model considering SIG, SIN, and HARDLIM functions as activation functions for discussion. The main findings are as follows: (1) The high-risk and above-grade areas of landslide disasters in Kenya are mainly concentrated in the plateau and plateau-rift transition zone in the southwest. (2) The ROC curve was used to evaluate the accuracy of the models, and the AUC values of the 0.977(IV), 0.965(LR), 0.859(ELM-SIG), 0.900(ELM-SIN), and 0.941(ELM-HARDLIM) models illustrate their validity. (3) Considering the PR curve results comprehensively, the recall rate and precision rate of the LR model are at a high level, marking it better than other models. (4) Nairobi, Central, Nyanza and Western provinces in Kenya account for a significant proportion of the high-risk and above-grade areas of landslide disasters.

Keywords

• kenya
• risk
• information value
• logistic regression
• machine learning