Artificial Intelligence-Based Geospatial Framework to Simulate Landslide Susceptibility

Abstract

In the context of natural disasters, landslides take more significance with the threat to life and property. The uncertainty of the occurrence of landslides, and the scarcity of accurate models for precise predictions have led to huge losses. Studies have been conducted in this paradigm worldwide to provide both quantitative and qualitative analysis. Moreover, many approaches have been taken to develop ML-based quantitative models. Yet, there is a timely need to develop an explainable AI model to predict and interpret landslides with logical arguments comprehensively. This study is focused on developing an AI model incorporating both training and logic to predict landslides which can be used to map landslide susceptibility. Landslide data, including topographical, climatic, and geological factors affecting landslides, related to Sri Lanka, retrieved from National Building Research Organization (NBRO) of Sri Lanka, which includes 3,000 data points 1,500 representing landslide occurrences and 1,500 representing non-occurrence. This research incorporates 25 features in modelling an artificial neural network (ANN) which learns by training to give predictions. Comparatively higher accuracy was obtained achieving a test accuracy of 94.45%, a precision of 88.46%, and a recall of 98.17%, demonstrating superior accuracy compared to existing models. In addition, a logical/rule-based model based on Expert Systems and Fuzzy Reasoning will be incorporated into the ANN model to predict the landslide occurrence giving probabilities and reasoning. Finally, a Geospatial framework will be developed to simulate landslide susceptibility. The approach will lead to mitigating the drawbacks of existing early warning systems and present the general public with a logical and more accurate mapping of landslide susceptibility and minimize losses to life and property. The findings of this research indicate the effectiveness of AI in predicting landslide-prone areas, potentially reducing the risk and impact of landslides in the study area