Predicting susceptibility to landslides under climate change impacts in metropolitan areas of South Korea using machine learning

Abstract

To cite this article: Sang-Jin Park & Dong-kun Lee (2021) Predicting susceptibility to landslides under climate change impacts in metropolitan areas of South Korea using machine learning,

Multi-hazard exposure mapping under climate crisis using random forest algorithm for the Kalimantan Islands, Indonesia

“…According to the literature review, RF models perform better than other models in a variety of risk vulnerabilities than SVM and multivariate statistical analysis 25 , 47 , 48 . Most of this assessment suggested that creating several risk maps using optimal algorithm that deliver high accuracy and prediction performance of a single risk type can help to gain highly accurate information regarding relevance and interaction 38 .…”

“…Furthermore, this study aligned with previous research where Random Forest (RF) demonstrated superior performance, employing supervised classification using spatial and geographic data, akin to findings by earlier researchers 49 – 51 . One of the distinguishing factors from prior studies lies in conducting spatial risk analysis of individual hazards and compound disasters through future climate scenarios 25 , 52 . Despite the inherent uncertainties of future climate scenarios, various scenarios from AR6 were employed, and the impacts based on policy scenarios were analyzed 53 , 54 .…”

“…In the previous literatures, many researches have previously conducted hazard risk spatial analysis using ensemble models such as Support Vector Machine (SVM), Naive Bayes Classifier (NB), k-Nearest Neighbor (kNN), and Decision Tree (DT). Among them, the results from RF were found to be the highest 22 , 25 , 26 . Additionally, high AUC values have been observed in studies that evaluated hazards using RF alone 23 , 24 , 27 .…”

“…AUC is a numerical measure and a technique without regard to thresholds 40 . A model’s categorization of a location is ideal when the area is 1, while a model’s classification of a location is poor when the area is 0.5 or less 25 . In the current research, Python 3.9 was used to create MH probability maps of woodland fires and droughts.…”

“…Future climate scenarios (RCP-SSP2-4.5 and RCP-SSP5-8.5) were used to predict future risk potentials. In the previous studies, we found that the two scenarios were compared and reviewed 25 , 41 – 44 . In general, 2–4.5 are known realistic scenarios in which climate change is mitigated in the future with carbon dioxide reduction policies.…”