Science, predicting landslide risk through machine learning models

Developing a more accurate and dynamic approach capable of more precisely predicting the likelihood of landslides. This is the goal of a study published in the journal Geophysical Research Letters, conducted by scientists at Northwestern University and the University of California, Los Angeles. The team, led by Chuxuan Li and Daniel E. Horton , designed a system that leverages hydrological predictors to better anticipate landslide risk. Traditional methods, the experts explain, rely on analyzing rainfall intensity. The new approach, the authors report, integrates various water-related processes with a machine learning model. The framework offers a more robust understanding of the causes of these destructive events. With further development, the new framework could help improve early warning systems, guide risk planning, and enhance climate resilience strategies in regions vulnerable to landslides. "Our model," Li says, "considers a wide range of factors, allowing us to identify diverse pathways leading to landslides. These events, in fact, are not all the same and may depend on various ideological processes."

As part of the study, researchers analyzed a month of extreme weather in California. During the winter of 2022-23, the region experienced over 600 landslides, with nine rivers triggering catastrophic floods. The researchers employed a computer model that simulates water movement in the environment, including rainfall infiltration into the ground, surface runoff, evaporation, and the freezing or melting of snow and ice. Using the model's results, the research team developed a metric called the "water balance state" (WBS) to assess the overabundance of water in a given area. Machine learning groups similar landslides based on site-specific conditions. Using this technique, they identified three main pathways that led to landslides in California: heavy rainfall, rainfall on already saturated ground, and melting snow or ice. The team estimates that heavy, rapid downpours caused approximately 32 percent of the landslides , while 53 percent of the events were associated with rainfall on saturated soil, and 15 percent were caused by melting snow or ice. "Excessive moisture played a central role in the landslide," Horton concludes, "which is particularly dangerous on steep slopes. Our system was able to accurately identify conditions favorable to landslides. Our ultimate goal is to develop a tool capable of making predictions."