Johnny Vega, Fabio Humberto Sepúlveda-Murillo, Melissa Parra
Air, Soil and Water Research 16 (1), (7 September 2023) https://doi.org/10.1177/11786221231195824
KEYWORDS: Colombian Andes, landslides, machine learning, SHAP, STATISTICAL METHODS, susceptibility
Landslides are a geological hazard commonly induced by rainfall, earthquakes, deforestation, or human activity causing loss of human life every year specially on highlands or mountain slopes with serious impacts that threaten communities and its infrastructure. The incidence and recurrence of landslides are conditioned by several aspects related to soil properties, geological structure, climatic conditions, soil cover, and water flow. Precisely, Colombia is one of the most affected by this type of natural hazard, as well as by floods, since they are the natural phenomena that bring with them the most severe risks for communities. In this work, we articulated the statistical approach of the landslide conditioning factors, Machine Learning Algorithms (MLA), and Geographic Information System (GIS), evaluating a flexible and agile methodology to estimate the landslide susceptibility defining areas prone to the landslide occurrence. The MLA were validated in a case study in the “La Liboriana” River basin, located in the Municipality of Salgar in the Colombian mountains Andes where Landslide Susceptibility Maps (LSMs) were obtained. The obtained MLA results hold immense potential in the field of regional landslide mapping, facilitating the development of effective strategies aimed at minimizing the devastating impacts on human lives, infrastructure, and the natural environment. By leveraging these findings, proactive measures can be devised to safeguard vulnerable areas, mitigate risks, and ensure the safety and well-being of communities. Seven supervised MLA were employed, two regression algorithms (Logistic) and five decision tree algorithms (Recursive Partitioning and Regression Trees [RPART], Conditional Inference Trees [CTREE], Random Forest [RF], Ranger, and Extreme Gradient Boosting Algorithm [XGBoost]). The LSMs were produced for each MLA. Considering different performance metrics, the RF model yields the best classification accuracy with an area under receiver operating characteristic (ROC) curve of 95% and 90% of accuracy, providing the most representative results. Finally, the contribution of each landslide conditioning factor on predictions with RF model is explained using the SHAP method.