As climate change fuels heavier rainfall and increases the frequency of extreme weather events, landslides have become more common, more widespread, and harder to predict. But a groundbreaking new study from Northwestern University and UCLA offers a game-changing tool in the battle to forecast and potentially prevent these destructive natural disasters.
Published in Geophysical Research Letters, the study introduces a new process-based framework for landslide prediction that moves beyond traditional, rainfall-only models. The innovation? A dynamic system that simulates the full water cycle from rain and runoff to soil saturation and snowmelt and uses artificial intelligence to identify the different pathways that lead to landslides.
“Current early warning systems often fail to account for how changing climate conditions affect the land,” said lead author Chuxuan Li, now a postdoctoral researcher at UCLA. “By incorporating a wider range of hydrological processes, our model is better equipped to identify when and where landslides are likely to occur even across large, diverse landscapes.”
To test the system, researchers analyzed a particularly devastating month in California’s recent history: the winter of 2022–23, when a rare “parade” of nine consecutive atmospheric rivers unleashed catastrophic flooding and triggered more than 600 landslides. Using a sophisticated, community-built computer model, the team simulated how water moved through terrain absorbing into soil, running off surfaces, or melting snow and ice.
They then created a new metric, “water balance status” (WBS), to measure how much water the ground could handle. When the WBS was positive indicating more water than the soil could absorb or drain the risk of landslides skyrocketed.
Their results were striking: 89% of all landslides during that month occurred in areas where the WBS was positive.
Further analysis revealed three dominant pathways to landslides:
•Intense rainfall (32% of cases),
•Rainfall on already saturated soils (53%),
•Rainfall-induced snow or ice melt (15%).
“This shows that it’s not just about how hard it rains but also what came before,” said Northwestern’s Daniel E. Horton, the study’s senior author and head of the university’s Climate Change Research Group. “Many areas were already soaked from earlier storms, or had melting snowpack. Our model accounts for all these factors in real time.”
The machine-learning component of the model grouped landslides based on shared environmental features, helping researchers pinpoint what exact conditions trigger different types of slides. This information could vastly improve how early warning systems operate, moving them from reactive to predictive and making them far more reliable, especially as climate extremes increase.
Atmospheric rivers, while not more frequent, are now more intense. Horton points out that warmer air holds more moisture, meaning these storms carry and drop more water. This intensifies flooding and makes landslides more probable and more destructive.
And that’s where this new framework becomes critical. By coupling hydrological simulations with AI-powered pattern recognition, the model not only explains past disasters but holds immense promise for forecasting future ones.
Looking ahead, the researchers plan to integrate their framework with live weather forecasting systems to build real-time prediction tools. These tools could help governments and communities better prepare for and respond to landslides, saving lives and reducing damage in vulnerable regions from California to the Himalayas.
As global climate change alters the natural rhythms of precipitation and snowmelt, scientists say this kind of modeling is essential. “We’re facing a future of more extreme weather, and we need early warning systems that can keep up,” Horton said. “Our goal is to make these models operational so we can not just understand landslides, but predict them before they happen.”