Comparing probabilistic and statistical methods in landslide susceptibility modeling in Rwanda/Centre-Eastern Africa

Jean Baptiste Nsengiyumva; Geping Luo; Amobichukwu Chukwudi Amanambu; Richard Mind'je; Gabriel Habiyaremye; Fidele Karamage; Friday Uchenna Ochege; Christophe Mupenzi

doi:10.1016/j.scitotenv.2018.12.248

Comparing probabilistic and statistical methods in landslide susceptibility modeling in Rwanda/Centre-Eastern Africa

Sci Total Environ. 2019 Apr 1:659:1457-1472. doi: 10.1016/j.scitotenv.2018.12.248. Epub 2018 Dec 18.

Authors

Jean Baptiste Nsengiyumva¹, Geping Luo², Amobichukwu Chukwudi Amanambu³, Richard Mind'je⁴, Gabriel Habiyaremye⁵, Fidele Karamage⁶, Friday Uchenna Ochege⁷, Christophe Mupenzi⁸

Affiliations

¹ State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, No. 818, South Beijing Road, Urumqi 830011, China; University of Chinese Academy of Sciences, Beijing 100049, China; Ministry in Charge of Emergency Management, P.O. Box 4386, Kigali, Rwanda; Faculty of Environmental Studies, University of Lay Adventists of Kigali (UNILAK), P.O. Box 6392, Kigali, Rwanda.
² State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, No. 818, South Beijing Road, Urumqi 830011, China; University of Chinese Academy of Sciences, Beijing 100049, China. Electronic address: luogp@ms.xjb.ac.cn.
³ State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, No. 818, South Beijing Road, Urumqi 830011, China; University of Chinese Academy of Sciences, Beijing 100049, China; Water, Engineering and Development Centre, School of Civil and Building Engineering, Loughborough University, Leicestershire, UK.
⁴ State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, No. 818, South Beijing Road, Urumqi 830011, China; University of Chinese Academy of Sciences, Beijing 100049, China; Faculty of Environmental Studies, University of Lay Adventists of Kigali (UNILAK), P.O. Box 6392, Kigali, Rwanda.
⁵ Faculty of Environmental Studies, University of Lay Adventists of Kigali (UNILAK), P.O. Box 6392, Kigali, Rwanda; Lancaster University, Lancaster Environment Centre, Faculty of Science and Technology, LA1 4YQ, UK.
⁶ University of Chinese Academy of Sciences, Beijing 100049, China; Key Laboratory of Watershed Geographic Sciences, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing 210008, China.
⁷ State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, No. 818, South Beijing Road, Urumqi 830011, China; University of Chinese Academy of Sciences, Beijing 100049, China; Department of Geography and Environmental Management, University of Port Harcourt, PMB 5323 Choba, East-West, Port Harcourt, Nigeria.
⁸ Faculty of Environmental Studies, University of Lay Adventists of Kigali (UNILAK), P.O. Box 6392, Kigali, Rwanda.

PMID: 31096356
DOI: 10.1016/j.scitotenv.2018.12.248

Abstract

Application of suitable methods to generate landslide susceptibility maps (LSM) can play a key role in risk management. Rwanda, located in centre-eastern Africa experiences frequent and intense landslides which cause substantial impacts. The main aim of the current study was to effectively generate susceptibility maps through exploring and comparing different statistical and probabilistic models. These included weights of evidence (WoE), logistic regression (LR), frequency ratio (FR) and statistical index (SI). Experiments were conducted in Rwanda as a study area. Past landslide locations have been identified through extensive field surveys and historical records. Totally, 692 landslide points were collected and prepared to produce the inventory map. This was applied to calibrate and validate the models. Fourteen maps of conditioning factors were produced for landslide susceptibility modeling, namely: elevation, slope degree, topographic wetness index (TWI), curvature, aspect, distance from rivers and streams, distance to main roads, lithology, soil texture, soil depth, topographic factor (LS), land use/land cover (LULC), precipitation and normalized difference vegetation index (NDVI). Thus, the produced susceptibility maps were validated using the receiver operating characteristic curves (ROC/AUC). The findings from this study disclosed that prediction rates were 92.7%, 86.9%, 81.2% and 79.5% respectively for WoE, FR, LR and SI models. The WoE achieved the highest AUC value (92.7%) while the SI produced a lowest AUC value (79.5%). Additionally, 20.42% of Rwanda (5048.07 km²) was modeled as highly susceptible to landslides with the western part the highly susceptible comparing to other parts of the country. Conclusively, the comparison of produced maps revealed that all applied models are promising approaches for landslide susceptibility studying in Rwanda. The results of the present study may be useful for landslide risk mitigation in the study area and in other areas with similar terrain and geomorphological conditions. More studies should be performed to include other important conditioning factors that exacerbate increases in susceptibility especially anthropogenic factors.

Keywords: Frequency ratio; Landslide; Logistic regression; Rwanda; Statistical index; Susceptibility.