The AI-enhanced SAFE-RICCS system for predicting landslides has been rolled out in Nepal. The system analyses satellite images taken by NASA and other space agencies. Climate change is expected to trigger more devastating landslides.
A landslide forecasting system driven by artificial intelligence (AI) is being rolled out in Nepal, one of the most landslide prone countries in the world, as the monsoon season approaches.
Devastating landslides in Papua New Guinea last month show the need for better forecasting and early warning systems to protect lives and properties, especially in mountainous developing countries in Asia Pacific which lack dedicated disaster monitoring systems and the means to communicate risks to the population.
In Nepal, more than 80 per cent of land is on a slope and much of it was destabilised during the 2015 earthquake in Gorkha, which killed around 9,000 people, according to project lead Antoinette Tordesillas from the University of Melbourne.
“With the monsoons due anytime now, we are helping policymakers and risk managers better prepare for future monsoons when increasingly frequent and heavy rains can trigger more devastating landslides,” she said.
Between June 2023 and June 2024, Nepal had 513 landslides, which killed 48 people and caused an estimated loss of more than US$ 440,000, national data shows.
Weeks in advance
A new AI tool has been added to the SAFE-RISCCS (Spatiotemporal Analytics, Forecasting and Estimation of Risks from Climate Change Systems) platform to improve forecasting, allowing it to predict landslides days or even weeks in advance. This gives people time to vacate areas where there are impending landslide threats.
The platform, developed by the University of Melbourne in 2018, continuously analyses satellite images of the earth taken by NASA, the European Space Agency and the Japan Aerospace Exploration Agency. Together with rain measurements and ever-changing ground motion data, it can monitor and achieve near real-time forecasting on landslide risks at a given location.
Tordesillas says different data types – depending on the location, topography and climate – can be integrated by the system to produce a forecast.
“The forecasting feature will then find the most relevant pattern or clue – for example ground motion holds clues when a slope is about to slide – in the input data and use it to quickly and accurately predict the where and when of the impending landslide,” she told SciDev.Net.
“The quality of the input data in the forecasting tool will be critical for accuracy in landslide prediction,” she added.
Such information can help planners and policymakers make risk assessments for land use planning, noted Basanta Adhikari, director of the Centre for Disaster Studies at Tribhuvan University in Nepal, a project partner.
Most natural landslides are triggered by earthquakes or rainfall, or a combination of both. Climate change-driven hazards, including frequent and intense rainfall, are expected to trigger more devastating landslides.
UN Secretary-General Antonio Guterres has called for “every person on earth to be protected by early warning systems by 2027”.
“Besides the effects of seismotectonic activities and climate change, unsustainable development – the non-engineering road construction, slope modification, deforestation and inappropriate drainage system – are triggering landslides in the Nepal Himalayas,” Adhikari told SciDev.Net.
Seeking partners
The tool is currently not available for open access distribution and deployment.
But Tordesillas said: “We would welcome the opportunity to partner with other institutions, industry and governments to use it in their landslide early warning systems. The Papua New Guinea landslide is a timely reminder of the huge costs to human life and livelihoods that are at stake.”
As weather events become more extreme, early warnings are a must for reducing disaster deaths and losses.
However, the researchers say there are challenges in deploying the tool in lower-income countries in the Asia Pacific region, such as lack of comprehensive historical data on landslides, limited financial resources and technical expertise, as well as bureaucratic hurdles.
David Petley, an earth scientist and vice-chancellor at the University of Hull in the United Kingdom, told SciDev.Net: “There is no doubt that satellite-based tools present a major opportunity for identifying areas of landslide risk and in providing advanced warning to those living within them.”
However, he says the tendency for landslides to be triggered by localised, short duration rainfall could make it challenging to predict them in real-time.
He added: “All early warning systems also suffer problems with communicating the information to the right people at the right time and to get them to respond appropriately. It will be interesting to see how this challenge is addressed.”
This piece has been sourced from SciDev.Net
Image: Wikimedia / Sridhar Rao
