Prediction of Spatial Likelihood of Shallow Landslide Using GIS-Based Machine Learning in Awgu, Southeast/Nigeria

• Published in: Sustainability Vol. 14; no. 19; p. 12000
• Main Authors: Nnanwuba, Uzodigwe Emmanuel, Qin, Shengwu, Adeyeye, Oluwafemi Adewole, Cosmas, Ndichie Chinemelu, Yao, Jingyu, Qiao, Shuangshuang, Jingbo, Sun, Egwuonwu, Ekene Mathew
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.10.2022
• Subjects: Algorithms; Analysis; Bayesian analysis; China; Climate; Comparative analysis; Correlation coefficient; Correlation coefficients; Data mining; Distribution; Field investigations; Geographic information systems; Geology; Geomorphology; Geospatial data; Landslides; Landslides & mudslides; Learning algorithms; Literature reviews; Local government; Machine learning; Mapping; Nigeria; Rainfall; Remote sensing; Sea level; Seasons; Statistical analysis; Susceptibility; Sustainability; Nigeria; China
• ISSN: 2071-1050; 2071-1050
• DOI: 10.3390/su141912000

A landslide is a typical geomorphological phenomenon associated with the regular cycles of erosion in tropical climates occurring in hilly and mountainous terrain. Awgu, Southeast Nigeria, has suffered a severe landslide disaster, and no one has studied the landslide susceptibility in the study area using an advanced model. This study evaluated and compared the application of three machine learning algorithms, namely, extreme gradient boosting (Xgboost), Random Forest (RF), and Naïve Bayes (NB), for a landslide susceptibility assessment in Awgu, Southeast Nigeria. A hazard assessment was conducted through a field investigation, remote sensing, and a consultation of past literature reviews, and 56 previous landslide locations were prepared from various data sources. A total of 10 conditioning factors were extracted from various databases and converted into a raster. Before modeling the landslide susceptibility, the information gain ratio (IGR) was used to select and quantitatively describe the predictive ability of the conditioning factors. The Pearson correlation coefficient was used to judge the correlation between 10 conditioning factors. In this study, rainfall is the most significant factor with respect to landslide distribution and occurrence. The confusion matrix, the area under the receiver operating characteristic curve (AUROC), was used to validate and compare the models. According to the AUROC results, the prediction accuracy for the RF, NB, and XGBOOST models are 0.918, 0.916, and 0.902, respectively. This current study can support the landslide susceptibility assessment of Awgu, Southeast Nigeria, and can provide a reference for other areas with the same conditions.