Assessing landslide susceptibility using a machine learning-based approach to achieving land degradation neutrality

Achour Y., Saidani Z., Touati R., Pham Q. B., Pal S. C., Mustafa F., ...More

ENVIRONMENTAL EARTH SCIENCES, vol.80, no.17, 2021 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 80 Issue: 17
  • Publication Date: 2021
  • Doi Number: 10.1007/s12665-021-09889-9
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, IBZ Online, PASCAL, Aerospace Database, Applied Science & Technology Source, Aqualine, Aquatic Science & Fisheries Abstracts (ASFA), BIOSIS, CAB Abstracts, Compendex, Computer & Applied Sciences, Environment Index, Geobase, INSPEC, Pollution Abstracts, Veterinary Science Database, Civil Engineering Abstracts
  • Keywords: Landslide vulnerability, Machine learning, Land degradation neutrality, 2030 UN agenda for sustainable development, Algeria, ANALYTICAL HIERARCHY PROCESS, SUPPORT VECTOR MACHINE, LOGISTIC-REGRESSION, NEURAL-NETWORKS, GIS, MODELS, REGION, PREDICTION, BIVARIATE, HAZARD
  • Yıldız Technical University Affiliated: Yes


The aim of this study is to develop landslide susceptibility models for the northern part of the Bordj Bou Arreridj (BBA) region, Northeast Algeria, to reduce the physical degradation caused by landslides and, to inspect what is required to properly control it. A comprehensive landslide inventory and susceptibility assessment of this region are not available, even though this region is prone to frequent disruption by geological hazards, mainly landslides. To achieve this objective, an inventory map and 12 variables (including geomorphic, geological, hydrological and environmental factors) are created. The inventory dataset is divided to training dataset with 148 landslides (70%) and validation dataset with 64 landslides (30%). Then, 2 machine learning (ML) techniques are applied to learn the internal relationship between the target set (212 landslide locations) and the 12 variables as inputs. The used methods are Random Forest (RF) and eXtreme Gradient Boosting (XGBoost). Their performances are assessed through the receiver operating characteristic (ROC) curve, the standard error (Std. error), and the confidence interval (CI) at 95%. As the main results, RF and XGBoost models give identical predictive accuracy (AUC) of approximate to 90%. This indicates that the proposed procedure can be useful for handling and monitoring present and future landslides. In addition, the models proposed in this study will be useful for the continuous assessment of land degradation trends for this region. Therefore, presenting these models in the best possible way allows stakeholders to benefit from them to identify key areas that may be targeted for protection and restoration procedures to achieve Land Degradation Neutrality (LDN) goals by 2030.