Scientists develop dynamic landslide prediction methods using hydrological and machine learning data

Scientists from Northwestern University and University of California Los Angeles (UCLA) have developed a new process-based framework that provides a more accurate and dynamic approach to landslide prediction across large-scale regions.

Traditional landslide prediction methods often rely solely on rainfall intensity, but new approaches integrate a variety of water-related processes and machine learning models. By taking into account a variety of, sometimes complex factors, the framework provides a stronger understanding of what drives these disruptive events.

With further development, this new framework will help improve early warning systems, inform hazard plans and strengthen climate resilience strategies for landslide vulnerable regions. Ultimately, these approaches can help save lives and prevent damage.

The study states that “mixed hydrogen ear processes explain the potential for local landslides.” Geophysical research book.

“Current early warning systems tend to derive information from historical precipitation events and landslides,” says Chuxuan Li, the first author of the study.

“Because it is based on historical data, we do not consider climate change. We expect stronger precipitation and more heavier precipitation events in the future. These systems often do not consider snow melting or other ground conditions. Our model considers a wider range of factors, allowing us to identify more diverse pathways that lead to Earth viewing than larger spatial scales.”

“Various landslides can be caused by different hydrological processes,” says Daniel E. Houghton, senior author of Northwestern's study.

“We are trying to identify which processes are triggered. But the scale that coincides with the storms that cause these events is trying to think about it from a much wider scale. Our ideal is to develop useful tools in a wider region, such as California.”

Horton is an associate professor of Earth, Environment and Planetary Science at Weinberg University of Arts and Sciences in Northwestern, and leads the Climate Change Research Group. Li is a PhD. He graduated from the Horton Institute in Northwestern and is a current postdoctoral researcher at UCLA.

Simulate the “parade” of Arashi

Dangerous flows of water, mud and rock can be difficult to predict, especially across a wide area with a variety of landscapes and different climates. To better understand how and why extensive landslides occur, the Northwest and the UCLA team turned to the extreme weather of a month in California.

During the winter of 2022-23, California experienced nine consecutive unprecedented “parades” of Atmospheric Rivers, causing catastrophic flooding and over 600 landslides.

To understand the pathways that caused these landslides, scientists adopted a community-developed computer model that simulates how water moves the environment, including penetrating the ground, running through the surface, evaporating, and freezing or thawing snow and ice.

To drive the model, the team used a diverse array of weather, geographical and historical data. This included information on topography, soil depth, past wildfires, precipitation, weather and climatic conditions.

Using the model output, the team developed a metric called “Water Balance Status” (WBS) to assess whether there was too much water in a particular area. Positive WBS means there is more water than the ground can treat by absorption, storage, evaporation, or drainage. This also means there is a high chance of landslides.

Identify the main route

Finally, Northwestern and the UCLA team applied machine learning techniques to group similar landslides based on specific conditions on the site. Through this technique, they identified three main routes that led to landslides in California: heavy rainfall, rain in already saturated soil, snow and ice.

The team is predicting a heavy, rapid downpour that caused about 32% of landslides. Approximately 53% of the landslide occurred after moderate rain fell on soil already saturated from the previous storm. Approximately 15% of landslides were linked to snow and ice, and rain was accelerating the melting of snow and ice.

“We found that most landslides were triggered under excessively wet conditions,” Li said. “Because it is over wet, precipitation means that it exceeds the soil's ability to retain or drain water, which is especially dangerous on steep slopes.”

When scientists compared these events with models, they found that a significant majority (89%) of California landslides occurred in areas with positive WBS. This finding examined that metrics could accurately identify ripe conditions for landslides.

“The research looks backwards to understand past events, but our ultimate goal is to look forward to this method making predictions,” Houghton said. “We will take this modeling framework we developed and use it in collaboration with the weather forecasting model.”

A better model for an uncertain future

As the global climate continues to change, forecasting systems are more important than ever. Warm air can hold more water vapor, allowing storms to throw away more water. And more water often shows more dangerous floods and landslides.

In a review published on ScienceHoughton and his collaborators have looked into how natural disasters such as the Atmospheric River can cause other disasters and cause chain reactions. In this article, the authors highlight the important need to integrate diverse datasets and build sophisticated models to improve their ability to predict and prepare natural disasters.

“Atmosphere rivers aren't necessarily more common,” Horton said. “However, when they land, their impact is becoming more severe. Recently, we have seen their rainfall intensity increase, consistent with the global trend of humans experiencing more intense precipitation events due to climate change caused by humans.”