Integration and comparison of algorithmic weight of evidence and logistic regression in landslide susceptibility mapping of the Orumba North erosion-prone region, Nigeria

In recent times, weight of evidence (WoE) and logistic regression (LR) methods in GIS-based landslide susceptibility mapping (LSM) have been remarked reliable, providing larger areal coverage. This paper aims to integrate and compare the performances of WoE and LR methods in LSM of the Orumba North Region in southeastern Nigeria. This study approach has not been implemented in Nigeria before. To perform the LSM in the region, eleven unique conditioning factors (elevation, slope degree, slope aspect, plan curvature, distance from road, topographic wetness index, rainfall, stream power index, land cover, distance from road, and geology) were considered, using 107 landslide inventories. These factors were selected based on the environmental characteristics and data availability. Five vulnerable zones were produced after reclassifying each conditioning element based on WoE into different classes: very low, low, medium, high, and very high. Furthermore, the LR mapping also reclassified the eleven factors into five landslide zones. The WoE and LR methods indicated that most parts of the area are characterized by moderate to very high landslide risks. Explicitly, the southern part of the region has higher risk of landslide occurrence whereas the northern part is dominated by low to very low susceptibility. Area under curve (AUC) values were used to validate the model performance and reliability. For training dataset, the AUC values obtained for the WoE and the LR were 0.986 and 0.992, and 0.995 and 0.998 for testing dataset, respectively. It was indicated that both models performed excellently and promise to be reliable. However, the LR slightly outperformed the WoE. This paper provides baseline information on the application of WoE and LR for landslide assessment in Nigeria and also provides insights for effective disaster management and land-use planning.