Comparing probabilistic and statistical methods in landslide susceptibility modeling in Rwanda/Centre-Eastern Africa

Area of Science:

• Geosciences
• Environmental Science
• Risk Management

Background:

• Rwanda faces frequent and intense landslides, necessitating effective risk management strategies.
• Landslide susceptibility maps (LSM) are crucial tools for assessing and mitigating landslide risks.
• Various statistical and probabilistic models exist for LSM generation, each with varying performance.

Purpose of the Study:

• To generate and compare landslide susceptibility maps in Rwanda using four distinct models.
• To evaluate the predictive accuracy of weights of evidence (WoE), logistic regression (LR), frequency ratio (FR), and statistical index (SI) models.
• To identify areas with high landslide susceptibility in Rwanda for targeted risk reduction efforts.

Main Methods:

• Utilized 692 past landslide locations for model calibration and validation.
• Incorporated fourteen conditioning factors: elevation, slope, TWI, curvature, aspect, distance to rivers/roads, lithology, soil properties, LS factor, LULC, precipitation, and NDVI.
• Validated models using receiver operating characteristic curves (ROC/AUC).

Main Results:

• The weights of evidence (WoE) model demonstrated the highest prediction rate (92.7% AUC).
• Frequency ratio (FR), logistic regression (LR), and statistical index (SI) models showed prediction rates of 86.9%, 81.2%, and 79.5% AUC, respectively.
• Approximately 20.42% of Rwanda's land area, particularly the western region, was identified as highly susceptible to landslides.

Conclusions:

• All applied models are promising for landslide susceptibility assessment in Rwanda.
• The weights of evidence (WoE) model is recommended for its superior accuracy in the study area.
• Findings provide valuable data for landslide risk mitigation in Rwanda and similar regions, emphasizing the need for further research on anthropogenic factors.