Kathmandu - Every morning, Nepali primary school teacher Bina Tamang steps outside her home and checks the rain gauge, part of an early warning system in one of the world’s most landslide-prone regions.

She contributes to an AI-powered early warning system that uses rainfall and ground movement data, local observations and satellite imagery to predict landslides up to weeks in advance, according to its developers at the University of Melbourne.

From her home in Kimtang village in the hills of northwest Nepal, 29-year-old Tamang sends photos of the water level to experts in the capital Kathmandu, a five-hour drive to the south.

“Our village is located in difficult terrain, and landslides are frequent here, like many villages in Nepal,” Ms Tamang told AFP.

Every year during the monsoon season, floods and landslides wreak havoc across South Asia, killing hundreds of people.

Nepal is especially vulnerable due to unstable geology, shifting rainfall patterns and poorly planned development.

As a mountainous country, it is already “highly prone” to landslides, said Mr Rajendra Sharma, an early warning expert at the National Disaster Risk Reduction and Management Authority.

“And climate change is fuelling them further. Shifting rainfall patterns, rain instead of snowfall in high altitudes and even increase in wildfires are triggering soil erosion,” he told AFP.

Landslides killed more than 300 people in 2024 and were responsible for 70 per cent of monsoon-linked deaths, government data shows.

Ms Tamang knows the risks first hand.

When she was just five years old, her family and dozens of others relocated after soil erosion threatened their village homes.

They moved about 1km uphill, but a strong 2015 earthquake left the area even more unstable, prompting many families to flee again.