SYDNEY—After a landslide hit Papua New Guinea’s highlands in the middle of the night last month, locals used spades, sticks and their bare hands to clear debris that authorities say buried people while they slept.
Half a world away, Dan Shugar, an associate professor at the University of Calgary who studies landslides, wondered whether scientific methods could have detected the impending disaster and helped to prevent the loss of life.
Looking at historical satellite imagery, Shugar found evidence of what appeared to be a small landslide in the same location in January 2022, and then an even smaller one in the days before the catastrophic collapse.
Minor landslides are common in mountainous areas and don’t always lead to disaster. But two small landslides over the previous couple of years, Shugar said, is “a warning that maybe something bigger was going to happen.”
Papua New Guinea, with high rainfall, hilly terrain and frequent earthquakes, is prone to landslides. But the May 24 event made headlines around the world after authorities estimated that hundreds of people, perhaps even 2,000, were killed.
Some experts have questioned the accuracy of the government’s death-toll estimate based on the size of the landslide—which covered about 22 acres—and the fact that it struck a rural area rather than a densely populated city. Papua New Guinea hasn’t had a reliable census for years, so it is difficult to determine the size of the local population with accuracy.
Officials with Papua New Guinea’s National Disaster Centre couldn’t be reached for comment on the death toll estimates and landslide monitoring efforts.
The disaster has put a renewed focus on efforts to better forecast when and where landslides might happen. At open-pit mines, where minerals are extracted from giant sloped pits in the ground, various tools such as ground-based radars or a laser-emitting sensor known as lidar are used to monitor the stability of slopes.
But such a detailed approach would be too costly to be deployed over wide areas, particularly in developing countries where landslides often hit. Researchers are now investigating other tools, such as satellite-based radars and machine learning, to make it easier to forecast future landslides across entire regions.
Adding to the urgency: Many researchers expect climate change will lead to more-frequent landslides. They predict intense rainfall, a common trigger of landslides, will become more common. Warmer temperatures are thawing frozen soil called permafrost and melting glaciers, destabilizing the land. And more severe wildfires could contribute to increased deforestation, which takes away the vegetation that supports the soil.
One study from Chinese researchers concluded the average annual frequency of landslides triggered by extreme precipitation in an intermediate climate-change scenario is projected to increase by 7% in the three-decade period after 2030, compared with the 1971-to-2000 period.
Countries most at risk for high casualties included China, Afghanistan and India—all of which have mountainous areas where there have been deadly landslides in the past.
Rain is one possible culprit for last month’s landslide. Papua New Guinea’s Prime Minister James Marape told his country’s parliament on Wednesday that the nation has received extraordinary rainfall in the past year, leading to river flooding, sea-level rise in coastal areas and landslides.
Complicating the relief effort is that the landslide struck an area that has been struggling with violence between tribal groups. Aid workers say people had sought refuge in the area where the landslide hit, swelling the population and adding to the uncertainty over the death toll.
“Our people in that village went to sleep for the last time, not knowing that they would breathe their last breath as they were sleeping peacefully,” Marape said.
The United Nations children’s agency, Unicef, estimated that about 7,850 people were affected by the disaster, and up to 150 structures were damaged. As of Saturday, nine bodies had been recovered, a U.N. official said. A separate U.N. analysis of satellite imagery counted 18 structures damaged or destroyed, and estimated that 2,000 people were living within 1.2 miles of the landslide.
The area was buried under as much as 26 feet of soil and debris, U.N. officials say. With more bodies potentially under the rubble, some feared that underground water flowing down the mountain would contaminate local drinking water. The land was still shifting, threatening rescue workers. And the landslide blocked a road to a town further inland, cutting the area off and raising fears of a food shortage.
Justine McMahon, the Papua New Guinea country director for aid group CARE International, said there is almost no sign that a village ever existed.
“We believe that there are some survivors who were scared and have run into the bush for safety,” she said. “The population is traumatized, particularly the children. And many of the children have stopped talking because of what they saw and what they experienced.”
Cressida Kuala, a women’s rights and environmental activist from the same province as the disaster, said people affected by the landslide need medical assistance, food and help to dig out the remains of their loved ones. Kuala, who said she passed through the village shortly before the landslide, said locals would sell goods along the road and there was a guesthouse nearby.
A multitude of variables influence whether landslides occur, making them difficult to predict. Events that trigger landslides, such as rainfall and earthquakes, aren’t easy to forecast. How a particular slope will behave is tied to material underneath the surface, which can be several layers of different rocks and soil.
“We can’t see everywhere,” said Pierre Rognon, an associate professor at the University of Sydney who studies soil mechanics. “There will always be some uncertainty about what the slope is made of.”
That means scientists need all the help they can get. Using deep learning, a type of machine learning that mimics the human brain, Italian scientists have shown that landslides can be anticipated accurately using only rainfall data. And at the University of Melbourne, researchers say they have developed an artificial-intelligence platform that can sift through massive amounts of data collected from satellite-based radars that measure ground deformation, and then predict the location of an impending landslide.
Their 2022 study looked at data tied to a landslide in China and volcano rockfalls in Italy and determined their tool would have shown where the events would occur. For the China landslide, which hit a village in Sichuan province in 2017 and left dozens dead or missing, the tool predicted the location of the disaster almost one year in advance, the study said.
But satellite-based radars have a drawback—they can have difficulty detecting land changes in areas of dense vegetation. The exact timing of a landslide can also be hard to determine, though once a location is known, scientists can use other tools for a more detailed analysis into when a landslide will occur.
“You can have two slopes of equal levels of instability, but if one gets hit by a heavy rainfall tonight, that one will be more predisposed to collapse,” said Antoinette Tordesillas, a professor at the University of Melbourne’s School of Mathematics and Statistics, and an author of the 2022 study.
There are already localized warning systems in some places where it is known there is a high landfall risk. In Hong Kong, which also has hilly terrain and intense rainfall, authorities use a combination of real-time rainfall data, weather forecasts and radars that track clouds to determine whether to issue landslide warnings.
Each rain gauge consists of a tipping bucket rainfall measuring unit and a data logger powered by solar energy, and information is beamed to cloud servers via mobile phone networks at one-minute intervals, according to information published by Hong Kong’s civil-engineering department. Though the city averages about 300 landslides a year, the last landslide fatality was in 2008.
For scientists, deciding whether locals should be evacuated isn’t always clear-cut. In one situation in Norway, nearby residents were evacuated more than a dozen times over several years before a rock slope gave way in 2019.
“The ground is very, very complicated, so it isn’t unusual to see the slope accelerating and looking really dangerous, and then stopping or slowing down,” said Dave Petley, a landslide expert and the vice chancellor at the University of Hull in England. “As a scientist in that situation, you’re really having a big impact on people’s lives.”