Landslide Susceptibility Mapping in Tegucigalpa, Honduras, Using Data Mining Methods

Abstract

Being located in a mountainous area in the heart of Central America, with a young and partially very soft volcanic geology under the influence of a humid tropical climate with high rainfall during the rainy season, the area of the Honduran capital city, Tegucigalpa, is highly prone to landslides. Due to rapid and uncontrolled urbanization, especially on the slopes that surround the city, there is a strong interaction between human activities and landslides, further increasing landslide occurrences and causing a high vulnerability of the poorest part of the population. We here employed a landslide inventory, a geological map, and a dataset of landslide related factors, generated from data that is freely available for the analysis of landslide susceptibility with the help of data mining techniques. An input dataset of 21 variables, such as lithology, landform and drainage characteristics, and road density, was pre-processed, explored, coded in the IBM SPSS Modeler software, and implemented for the prediction of landslide occurrences with Artificial Neural Networks (ANN), Bayesian Networks (BN), and Decision Trees (DT). Different techniques were applied to enhance the performance of the model predictions by preparing the dataset to make it mathematically more accessible. The models with the balanced dataset yielded promising overall correct predictions of landslide and non-landslide cases of 85% (ANN) to 90% (DT) and correct predictions of landslides of 35% (BN) to 63% (ANN).