Landslide susceptibility mapping core-base factors and models' performance variability: a systematic review

Landslides cause significant economic, social, and environmental impacts worldwide. However, selecting the most suitable model and factors for landslide susceptibility mapping (LSM) remains challenging due to the diverse factors influencing landslides and the unique environmental settings in which they occur. Here, we conducted a systematic literature review from 2001 to 2021 to identify the main core-base factors and models used in LSM and highlight areas for future research. We found that there is a need for increased research collaboration with leading knowledge-producing countries and research efforts in underrepresented regions such as Africa, Central America, and South America. Of the 31 most used landslide susceptibility factors, we identified the core-base factors slope, elevation, lithology, land use/land cover, and distance from road, which were the most used, top-ranked predictors and commonly used together when mapping landslide susceptibility. Although aspect was the third most used factor, it ranked among the eight least effective predictors of LSM. Among the core-base factors of LSM, road density, elevation, and slope exhibited the least ranking variability as LSM predictors. The most used methods in LSM were random forest, logistic regression, support vector machine, and artificial neural network, with hybrid, ensemble, and deep learning methods currently trending. Random forest was the most accurate of the four most commonly used models, followed by artificial neural networks. However, artificial neural networks exhibited the least performance variability, followed by support vector machines. This comprehensive review provides valuable insights for researchers in selecting appropriate factors and models for LSM and identifies potential areas for future collaboration and research.