Landslide Prediction Methods: A Comparative Overview of Traditional and AI-Based Decision-Making Approaches

Landslides pose significant global hazards, and traditional methods for risk assessment, such as expert systems and geotechnical models, are resource-intensive and limited by human subjectivity. With the advent of artificial intelligence (AI), machine learning (ML) and deep learning (DL) methods have become prominent in landslide detection, offering advanced capabilities for modelling complex geotechnical data. Despite initial reliance on visual interpretation of remote sensing data, AI-driven methods have shown potential to improve accuracy and objectivity. ML models have been used to detect landslides based on pixel data and geological features, while DL models, especially CNNs like U-net and ResNet, have gradually replaced ML. However, DL methods are still in their early stages and face challenges due to the complexity of landslide characteristics across regions.

I. Introduction

Landslides are a major hazard worldwide, creating significant risks to human society [1]. Due to the complex and variable nature of soils and rocks, effective prevention is difficult. Landslide risk assessment involves evaluating the likelihood of landslides and their potential impacts on living beings, [2], [3]. Traditional methods used in landslide management primarily depend on knowledge-based expert systems, which integrate human expertise to assess and predict landslide occurrences. Although these systems offer specialized insights, they are inherently restricted by human subjectivity, as expert judgments can vary, impacting consistency. Moreover, these methods are often tailored to region-specific characteristics, including climatic, geological, geomorphological, topographical, and seismic factors, which makes them difficult to apply effectively outside their intended regions. Additionally, anthropogenic influences, such as urban development and land-use changes, introduce further complexity, as these factors may vary significantly across regions and alter landslide susceptibility. This reliance on localized data and conditions limits the adaptability and accuracy of knowledge-based systems in diverse settings, reducing their effectiveness in areas with different environmental and human variables [4]–[6].