Mapping Landslide Hazard Risk Using Random Forest Algorithm in Guixi, Jiangxi, China

Yang Zhang,Xiaoting Zhou,Lifeng Xie,Yaozu Qin,Penghui Ou,Xiaolan Huang,Weicheng Wu,Chongjian Shao,Renxiang Chen,Shanling Peng,Yong Song,Ziyu Lin,Wenchao Huangfu,Guiliang Zhang,Tao Lang

doi:10.3390/ijgi9110695

Yang Zhang, Xiaoting Zhou + Show 13 more

Open Access

https://doi.org/10.3390/ijgi9110695

Copy DOI

Abstract

Landslide hazards affect the security of human life and property. Mapping the spatial distribution of landslide hazard risk is critical for decision-makers to implement disaster prevention measures. This study aimed to predict and zone landslide hazard risk, using Guixi County in eastern Jiangxi, China, as an example. An integrated dataset composed of 21 geo-information layers, including lithology, rainfall, altitude, slope, distances to faults, roads and rivers, and thickness of the weathering crust, was used to achieve the aim. Non-digital layers were digitized and assigned weights based on their landslide propensity. Landslide locations and non-risk zones (flat areas) were both vectorized as polygons and randomly divided into two groups to create a training set (70%) and a validation set (30%). Using this training set, the Random Forests (RF) algorithm, which is known for its accurate prediction, was applied to the integrated dataset for risk modeling. The results were assessed against the validation set. Overall accuracy of 91.23% and Kappa Coefficient of 0.82 were obtained. The calculated probability for each pixel was consequently graded into different zones for risk mapping. Hence, we conclude that landslide risk zoning using the RF algorithm can serve as a pertinent reference for local government in their disaster prevention and early warning measures.

Highlights

Landslides are a major natural hazard and can be defined as phenomena in which a rock and soil body on a slope slides down a certain interface under the action of gravity, rainfall, and groundwater
In spite of its capacity to deliver rather accurate prediction when number of decision trees (NT) was set to 100, the prediction was more robust with higher overall accuracy (OA) and Kappa Coefficient (KC) when it was set to 300 and 500
We demonstrated the procedure on how to digitize non-digital geo-environmental factors and assign a weight value in terms of their proximity or propensity so that quantitative analysis and modeling were made possible

Summary

Introduction

Landslides are a major natural hazard and can be defined as phenomena in which a rock and soil body on a slope slides down a certain interface under the action of gravity, rainfall, and groundwater. Landslides are one of the most frequent geological disasters in China. In 2019, 6181 geohazards were recorded in China, including 4220 landslides, accounting for 68.27% of the total hazards. These geohazards resulted in 211 deaths, 13 missing persons, and 75 injuries, and a direct economic loss of 2.77 billion yuan (China Geological Survey, 2020) [1]. The Ministry of Natural Resources of the People’s Republic of China announced that 1747 geological hazards occurred in the first half of 2020, with a direct economic loss of 1.01 billion yuan. It was predicted that the situation will remain severe in the second half of the year (http://www.mnr.gov.cn/)

Objectives

Methods

Results

Conclusion