GIS-based landslide susceptibility mapping of Western Rwanda: an integrated artificial neural network, frequency ratio, and Shannon entropy approach

Vincent E. Nwazelibe, Johnbosco C. Egbueri*, Chinanu O. Unigwe, Johnson C. Agbasi, Daniel A. Ayejoto, Sani I. Abba

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

14 Scopus citations

Abstract

The May 2nd and 3rd, 2023 landslide in Rwanda’s Western Province caused a devastating natural disaster, resulting in the tragic loss of 95 lives. Ngororero, Rubavu, Nyabihu, and Karongi were the worst-hit areas, as reported by Rwanda Broadcasting Agency (RBA). Such recurring disasters have posed significant challenges to the affected communities, requiring strong measures like susceptibility mapping to address their impact in the future. The literature review indicates that statistic and machine-learning susceptibility mapping efforts have been applied in the study region. However, these studies have not focused explicitly on localized scale studies of the western province; instead, they have mainly concentrated on examining the entire country. Using artificial neural networks (ANN), Shannon entropy (SE), and frequency ratio (FR), this paper aims to fill some gaps in the Rwandan landslide literature by integrating localized studies of the landslide susceptibility mapping (LSM) of the western province of Rwanda using the available higher data resolution. The LSM studies took 1157 landslide inventory locations and considered a broader range of landslide-conditioning factors compared to the previous studies on the region (distance from the road, aspect, elevation, slope degree, stream power index, normalized differential vegetation index, plan curvature, distance from the river, topographic wetness index, geology, and rainfall). In model training, 70% (810 points) of the landslide points underwent utilization, while the remaining 30% (347 points) served the purpose of model testing. The obtained area under the curve (AUC) values from model validation and testing provided reliable accuracy measures for the three LSM methods: ANN (AUC = 0.929 and 0.924), FR (AUC = 0.895 and 0.889), and SE (AUC = 0.768 and 0.750). Despite varying data handling, the models show that Rutsiro, Ngororero, and Karongi in Rwanda's Western Province have the highest landslide concentration. The relative importance of conditioning factors indicates that geology, rainfall, distance to the road, slope, and NDVI factors played a crucial role in landslides in the studied area. The slope can be stabilized by enhancing drainage, modifying slope angles, and implementing structural fortifications. It is hoped that the findings of this study will aid Rwandan policymakers and global researchers mitigate landslides and their dynamics.

Original languageEnglish
Article number439
JournalEnvironmental Earth Sciences
Volume82
Issue number19
DOIs
StatePublished - Oct 2023

Bibliographical note

Publisher Copyright:
© 2023, The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.

Keywords

  • Artificial neural networks (ANN)
  • Frequency ratio (FR)
  • Landslide susceptibility mapping (LSM)
  • Landslides
  • Shannon entropy (SE)

ASJC Scopus subject areas

  • Global and Planetary Change
  • Environmental Chemistry
  • Water Science and Technology
  • Soil Science
  • Pollution
  • Geology
  • Earth-Surface Processes

Fingerprint

Dive into the research topics of 'GIS-based landslide susceptibility mapping of Western Rwanda: an integrated artificial neural network, frequency ratio, and Shannon entropy approach'. Together they form a unique fingerprint.

Cite this