Landslide susceptibility maps (LSMs) are very crucial for planningpolicies in hazardous areas. However, the accuracy and reliability ofLSMs depend on available data and the selection of suitable methods.This study is ...Landslide susceptibility maps (LSMs) are very crucial for planningpolicies in hazardous areas. However, the accuracy and reliability ofLSMs depend on available data and the selection of suitable methods.This study is conducted to produce LSMs by combinations of machinelearning methods and weighting techniques for Ha Giang province,Vietnam, where has limited data. In study area, we gather 11 landslideconditioning factors and establish a landslide inventory map.Computing the weights of classes (or factors) is very important toprepare data for machine learning methods to generate LSMs. Wefrst use frequency ratio (FR) and analytic hierarchy process (AHP)techniques to generate the weights. Then, random forest (RF), supportvector machine (SVM), logistic regression (LR), and AHP methods arecombined with FR and AHP weights to yield accurate and reliableLSMs. Finally, the performance of these methods is evaluated by fivestatistical metrics, ROC and R-index. The empirical results have shownthat RF is the best method in terms of R-index and the five metrics, i.e.TP rate (0.9661), FP rate (0.0), ACC (0.9835), MAE (0.0046), and RMSE(0.0350) for this study area. This study opens the perspective of weightbasedmachine learning methods for landslide susceptibility mapping.展开更多
基金supported by the Research on scientific basis to develop a set of criteria and to identify the areas highly-susceptible to landslides,debris flows,flash floods in the mountainous and hilly regions of Vietnam.Funded by Ministry of Natural Resources and Environment(MONRE)[TNMT.2021.02.08].
文摘Landslide susceptibility maps (LSMs) are very crucial for planningpolicies in hazardous areas. However, the accuracy and reliability ofLSMs depend on available data and the selection of suitable methods.This study is conducted to produce LSMs by combinations of machinelearning methods and weighting techniques for Ha Giang province,Vietnam, where has limited data. In study area, we gather 11 landslideconditioning factors and establish a landslide inventory map.Computing the weights of classes (or factors) is very important toprepare data for machine learning methods to generate LSMs. Wefrst use frequency ratio (FR) and analytic hierarchy process (AHP)techniques to generate the weights. Then, random forest (RF), supportvector machine (SVM), logistic regression (LR), and AHP methods arecombined with FR and AHP weights to yield accurate and reliableLSMs. Finally, the performance of these methods is evaluated by fivestatistical metrics, ROC and R-index. The empirical results have shownthat RF is the best method in terms of R-index and the five metrics, i.e.TP rate (0.9661), FP rate (0.0), ACC (0.9835), MAE (0.0046), and RMSE(0.0350) for this study area. This study opens the perspective of weightbasedmachine learning methods for landslide susceptibility mapping.
