How Artificial Intelligence Helps Better Predict Landslides in Minnesota

Landslides pose a major geological risk, causing significant damage to infrastructure and human losses every year. In Minnesota, a region shaped by glaciation, these phenomena remain poorly mapped at the regional scale. Recent research has used advanced artificial intelligence methods to create the first detailed map of at-risk areas in this U.S. state.

Five machine learning and deep learning models were compared to identify the most exposed areas. Among them, two approaches proved particularly effective: random forests and a specialized neural network called TabKANet. These models analyzed data such as slope, elevation, land use, and precipitation. The results show that steep slopes and low-altitude areas are the most vulnerable, but local factors, such as water concentration or human activities, can also play a decisive role.

The study also used a technique called SHAP to explain how each factor influences risk. For example, a steeper slope clearly increases the likelihood of a landslide, while lower elevation, often associated with water-saturated soils, also worsens the situation. However, in the field, other elements such as drainage or human-induced landscape changes can become determining factors.

A major innovation of this research is the use of “counterfactuals,” a method that simulates the changes needed to stabilize an unstable area. For example, reducing the slope, improving drainage, or strengthening vegetation could be enough to prevent a landslide. These tools help authorities prioritize prevention actions and better understand the mechanisms at play.

This approach combines precision and transparency, providing risk managers and urban planners with a reliable framework for making informed decisions. It could be applied in other regions of the world facing similar challenges, thereby improving the safety of populations and the resilience of infrastructure.

Sources and Credits

Source Study

DOI: https://doi.org/10.1007/s41748-026-01114-6

Title: Explainable AI (xAI) for Landslide Susceptibility Modeling: A Comparative Analysis of Machine Learning and Deep Learning Approaches

Journal: Earth Systems and Environment

Publisher: Springer Science and Business Media LLC

Authors: Ambikesh Dwivedi; Surya Sarat Chandra Congress; Raul Velasquez; Prince Kumar; Ujwalkumar Patil