Predicting landslide susceptibility and risks using GIS-based machine learning simulations, case of upper Nyabarongo catchment

Abstract

Sustainable landslide mitigation requires appropriate approaches to predict susceptible zones. This study compared the performance of Logistic Model Tree (LMT), Random Forest (RF) and Naïve-Bayes Tree (NBT) in predicting landslide susceptibility for the upper Nyabarongo catchment (Rwanda). 196 past landslides were mapped using field investigations. Thus, the inventory map was split into training and testing datasets. Fifteen predisposing factors were analysed and information gain (IG) technique was used to analyse the correlation between factors and observed landslides. Therefore, the area under receiver operating characteristic (AUROC) with other statistical estimators including accuracy, precision, and root mean square error (RMSE) were employed to compare the models. The AUC values were 78.7%, 80.9% and 82.4% for RF, LMT and NBT models, respectively. Additionally, the NBT produced the highest accuracy and precision values (0.799 and 0.745, respectively). Regarding RMSE values, the NBT model achieved an optimized prediction than RF and LMT models (0.301; 0.428 and 0.364, respectively). The results of the current study may inform further studies and appropriate landslide risk reduction and mitigation measures. They can also be instrumental for policy and decision making in regards with natural risk management.

A comparison among fuzzy multi-criteria decision making, bivariate, multivariate and machine learning models in landslide susceptibility mapping

“…Regarding the machine learning models, the random forest classifier (RFC) is based on a combination of decision tree classifiers and, therefore, it is considered a powerful supervised algorithm for solving binary classification tasks (Breiman 2001). The RFC model has been used in several LS studies, such as Chen et al (2018aChen et al ( , 2018b, Sevgen et al (2019), Nsengiyumva and Valentino (2020), Kocaman et al (2020) or Zhao et al (2020). Similarly to RFC model, Naïve Bayes classifier (NBC), a supervised probabilistic algorithm built on Bayes theorem, have been applied in several studies in recent years (Tsangaratos and Ilia 2016;He et al 2019;Chen et al 2020aChen et al , 2020bLee et al 2020;Lei et al 2020aLei et al , 2020b.…”

Integrating InSAR Observables and Multiple Geological Factors for Landslide Susceptibility Assessment

“…The index calculations of these factors are described below. It should be mentioned that these spatial factors were first selected based on suggestions reported in the relevant studies in the literature [16][17][18][19][20]. A significance test was then performed to identify the most influential factors that have the high correlation with the landslides in the study areas.…”

“…Machine learning algorithms enrich the quality and accuracy of generated susceptibility maps. Researchers use and compare various machine learning models on the basis of different data [16][17][18][19], integrate different machine learning models to improve accuracy [20][21][22][23], or develop new algorithms that are based on traditional machine learning models to strengthen landslide prediction results [24][25][26]. These techniques perform better than do classical methods.…”

“…The later-observed spatial factors are inputted into the model to determine the landslide probability of each slope unit. The naive Bayes prediction model is based on the probability density function of the Bayesian classification method [33], as shown in Equation (16).…”

A comparative evaluation of machine learning algorithms and an improved optimal model for landslide susceptibility: a case study

“…Unfortunately, landslides are natural disasters that cannot be avoided and are triggered by numerous factors (e.g., abundant rainfall, complex geo-environmental settings, high-frequency earthquakes, etc.) in mountainous communities (Parker et al 2011;Gorum et al 2013;Nsengiyumva and Valentino 2020).…”