Sudden and unpredictable, landslides and avalanches kill thousands of people each year and cause billions of dollars in damage. But what if we could anticipate them before they happen?

In the village of Kimtang in central Nepal, there are clear warning signs that something is wrong. Cracks run through concrete steps, and trees grow at unusual angles—evidence that the ground beneath the village is slowly shifting. The key question is: how much movement is actually taking place?

Antoinette Tordesillas, a mathematician at the University of Melbourne, studies this using advanced technology. Through a Zoom call, she displays an overhead image of Kimtang—not a standard satellite photo, but a color-coded map created by an artificial intelligence (AI) system.

In the image, a bright red area stands out against the dark blue hillside. This red zone marks a highly unstable section of land directly beneath the village, indicating a serious risk of a major landslide. According to Tordesillas, the villagers live and farm on a dangerous slope.

Although landslides often seem sudden and impossible to predict, radar-based satellite images can reveal subtle early warning signs. These include tiny movements in the ground, where particles of soil slowly begin to separate—sometimes days, weeks, or even years before a collapse.

For the people of Kimtang, this AI-generated map is both a warning and an opportunity.

Landslides are becoming more frequent due to climate change and human activities such as construction and mining. In the United States alone, they kill between 25 and 50 people annually. Globally, the death toll reaches into the thousands. Predicting where and when they will occur remains extremely challenging, but AI is beginning to change that by helping scientists identify high-risk areas around the world.

Nepal, home to some of the tallest peaks in the Himalayas, is especially vulnerable. In October 2025, a series of landslides killed around 60 people there.

Monitoring such vast and rugged terrain produces enormous amounts of data. Analyzing it manually would be nearly impossible, but machine learning—a type of AI—can process it efficiently.

Tordesillas explains that their AI system is guided by established knowledge of how slopes fail. While not perfect, this approach ensures the system reflects scientific understanding of landslide behavior.

In fact, Kimtang’s residents were already relocated in 2019 after a nearby landslide struck their farmland. Ironically, the area they were moved to is now believed to be the most unstable in the region.

The data that revealed this came from Sentinel-1, a European satellite that uses radar to map the Earth’s surface in fine detail. It scans the ground at around 2,000 pulses per second. Fortunately, the satellite passed over this part of Nepal at an ideal angle, allowing researchers to gather high-quality measurements.

The most recent images, from January 2025, show that no landslide has yet occurred. However, the analysis has enabled scientists to warn villagers, develop ground-level monitoring strategies, and plan evacuation routes.

The AI map also highlights safer areas. For example, the local high school appears to be located on one of the most stable parts of the terrain—valuable information that has been shared with local authorities.

Similar work is being carried out elsewhere. In the UK, researchers are using AI to analyze radar data from Sentinel-1 across 300,000 slopes in Great Britain. They identified around 3,000 slopes that are actively moving.

These movements are extremely small—just millimeters per year—and invisible to the naked eye. However, they may signal the potential for future landslides. Even minor shifts can affect thousands of miles of roads and railways over time.

According to Alessandro Novellino of the British Geological Survey, tasks that would take humans months or years can now be completed by AI in hours. This capability is enabling entirely new types of scientific research.

Although real-time satellite data is expensive and not always available, even data that is one or two years old can provide valuable insights into long-term ground movement and risk.

In other cases, AI is used after disasters occur. Following major landslides on the Indonesian island of Sumatra, researchers used AI to map more than 4,000 landslides across half the island. This information helped authorities assess damage and coordinate emergency responses.

Organizations like Network Rail in the UK are also using AI to monitor land stability around railway tracks. By combining data from sensors, drones, and inspection trains, they can detect early signs of ground movement and take preventive action.

Speed is especially critical when dealing with avalanches. These sudden flows of snow and ice can be deadly. This winter alone, more than 100 people died in avalanches in the European Alps. In February, nine skiers were killed in an avalanche near Lake Tahoe in California.

Researchers have found an unexpected source of avalanche data: webcams across the Alps. James Fox, a student at the Swiss Federal Institute of Technology in Lausanne, helped develop an AI system that can detect avalanches in these images.

To train the system, around 4,000 avalanche images were manually labeled. This allowed a deep learning model to learn what avalanches look like and eventually recognize them automatically.

While this technology shows promise, human oversight remains essential. AI systems can make mistakes—for example, confusing exposed rocks with avalanches. Still, such tools could help alert authorities more quickly and improve emergency response.

As climate change continues to destabilize mountain environments, technologies like these may become increasingly important. By detecting subtle changes in the Earth before disaster strikes, AI offers a powerful new way to protect lives and infrastructure.