Global fatal landslide occurrence from 2004 to 2016

Picture a steep hillside after days of rain. The ground looks still, but it isn't. Water seeps into cracks, pressure builds, and then—quiet becomes motion. When slopes meet people, those motions can be deadly. That's why, as Dave Petley and colleagues have argued for years, the most powerful tool we have isn't a fancy model; it's a clean, coherent record of what actually happened, where, and when. A global inventory lets us see patterns, add up losses, and—most importantly—learn how to keep people out of harm's way. You'd think we already had that. We do have global disaster databases, from the Emergency Events Database, or EM-DAT, to NASA's Global Landslide Catalog. They're useful, but they miss a lot when it comes to landslides. Petley showed back in 2012 that EM-DAT had undercounted fatal landslide events by roughly twenty times and fatalities by more than four times over part of the two-thousands. Kirschbaum and colleagues found a similar gap a few years later. Why? Because landslides are often filed as a footnote to something else—an earthquake or a storm—and because reporting varies wildly by language and location. People also die days or weeks after these events, so final tallies can lag. Even in a dedicated landslide inventory, Petley and Sepúlveda estimated the undercount might be on the order of fifteen percent. So Froude and Petley built a database designed to do one thing very well: track every fatal landslide they could find consistently, year after year, since 2004. It's called the Global Fatal Landslide Database, or GFLD, and its backbone is methodical. They start with news reports in English to catch the first signal that lives were lost. Then they triangulate: government and aid agency reports, peer-reviewed studies, and sometimes direct communications. They classify each event using the modern landslide taxonomy that Oldrich Hungr's group laid out—so not just "landslide," but debris flows, rockfalls, and so on—and tag the trigger: rainfall, seismic shaking, construction activity, mining, hill cutting, leaking pipes, and even the collapse of garbage mounds. The team re-read thousands of entries to standardize those trigger tags with keyword searches. And they didn't just slap a pin on a map. They reconciled place descriptions with administrative boundaries, road networks, and satellite imagery. For road-related events, they drew a line along the road and buffered it by 500 meters to estimate where the slope likely failed. The median spatial precision works out to an area about the size of a medium county—roughly 681 square kilometers—with a very wide range. One more crucial choice: they excluded earthquakes from the analysis here, not because quakes don't matter, but because their global record of quake-triggered landslides isn't complete yet. Those excluded events still exist in the database—168 earthquakes linked to 3,978 deaths—but leaving them out keeps the non-seismic picture clean. What does that picture look like from 2004 to 2016? It's big, it's global, and it's lopsided. The database counts 4,862 non-seismic fatal landslide events and 55,997 deaths. Asia dominates with three out of every four events, reflecting both where people live—densely—and the kind of terrain they live in: steep, wet, and rapidly changing. You also see clusters in Central America, extending through southern Mexico, the Caribbean, along the Andes in South America, in East Africa around the Tanzania–Rwanda–Kenya belt, and in pieces of Turkey, Iran, and the European Alps. And the losses are spiky. A single catastrophe can tilt a year's totals—the Kedarnath disaster in India in 2013 killed more than 5,000 people—while most landslides kill far fewer. Only a tenth of one percent of events crossed the thousand-fatality mark. The vast majority, ninety-five percent, involved one discrete slope failure rather than a chain of failures across a large area. The top-line story about triggers is both simple and sobering. Rainfall rules. Of all non-seismic landslide deaths in the record, seventy-nine percent were triggered by rain. Rainfall-triggered landslides also dominate the counts: they make up seventy-nine percent of non-seismic events. When you look at the time series, rainfall events drive almost all the wiggle you see—about ninety-three percent of the variance in the whole non-seismic record, as Froude and Petley report—with a correlation that's essentially locked in. That rhythm is seasonal. If you stack the data by five-day blocks, the autocorrelation peaks right around one year, which is a complicated way of saying there's a strong annual cycle. Non-rainfall, non-seismic events, the shorthand in the paper is NSNR, don't show that; their timing follows people, not clouds. Zoom in by region and that seasonal picture sharpens. In South Asia, landslides rise and fall with the monsoon like clockwork. The correlation between mean monthly rainfall and mean monthly landslides there is almost perfect, at zero point nine nine six, which is as tight as it gets in geoscience. East Asia isn't far behind at about zero point ninety-seven. In the Americas, where seasonal rains also have a strong cadence, Central America comes in around zero point eighty-two and South America about zero point eighty-one. Southeast Asia is the outlier; at roughly zero point seventeen, the link between rainfall and landslides is weak at this monthly scale, hinting that local storms and typhoons can scramble a simple rain-total story. You can feel the monsoon's fingerprint in the country breakdowns. India accounts for about sixteen percent of all rainfall-triggered events worldwide and Nepal about ten percent, and the overwhelming majority of those happen during their summer monsoon—seventy-seven percent for India and ninety-three percent for Nepal. Put together, roughly one in five rain-triggered landslides in the entire global record happened during those two countries' monsoons. East Asia's pattern is different but just as clear: typhoons. Between April and October, 109 fatal landslides were directly tied to typhoons—about sixteen percent of East Asia's rainfall-triggered events and three percent of the global total. In the Philippines, nearly half of rainfall-triggered landslides were typhoon-driven. Indonesia's pattern follows its own monsoon, with about seventy-two percent of rainfall-triggered events between November and April, and a smaller August-to-October bump that reflects typhoon rains sneaking in from the north. Cross the Pacific and the signal shifts with the South American monsoon system and the migration of the Intertropical Convergence Zone. Brazil and Colombia together host most of that continent's rain-triggered landslides, with Brazil at thirty-seven percent and Colombia at thirty-two percent. If you're wondering how these regional pulses add up to the global beat, the team did that math too. They fitted a hierarchical regression, adding regions in turn to see how much of the year-to-year seasonal swing each one explains. South Asia alone accounts for about half of the global seasonal variance. Adding Southeast Asia and East Asia brings the total up past eighty-six percent. Once South America and Central America are in, you've explained roughly ninety-six percent of that global seasonal cycle. It's a neat result with a simple intuition: line up the seasonal rains across a few key regions, and you've essentially built the world's fatal landslide calendar. Now, what about El Niño and La Niña, those Pacific swings we love to blame for everything? Here the authors are careful. Over just thirteen years, there's no clean global link between those climate modes and landslide incidence. Regional rainfall often dances to the tune of the El Niño–Southern Oscillation, or ENSO, but landslides add layers—local topography, land use, exposure—that smear the signal. To tease out a robust ENSO–landslide relationship, they argue you need local records on the order of thirty years or more. So far, this is a climate story. But the second act is human. In the database, non-rainfall, non-seismic triggers make up sixteen percent of events—770 in total—and 3,725 deaths. What stands out isn't their sheer number but their trajectory and texture. Construction, mining, and hill cutting are driving a growing share of multi-fatality disasters, especially in Asia's rapidly urbanizing mountains. Workers take the brunt: about ninety percent of deaths in mining-related landslides are workers, seventy-six percent in construction, and eighty-four percent in hill cutting. Take construction. In China, more than half of construction-triggered landslides in the record happened at urban sites, with only a small fraction on roads. In India and Nepal, the pattern flips; roads account for about thirty percent of construction-triggered events in India and forty-three percent in Nepal. That makes intuitive sense when you look at the pace of road building—Nepal's road network roughly quadrupled in length in about eighteen years, and India's nearly tripled in around twenty-four—often cut into steep, fragile slopes with minimal engineering. When things go wrong, they can go very wrong. The Shenzhen construction-waste collapse in December 2015 killed seventy-seven workers, not because a hillside gave way in a storm, but because a mountain of spoil was piled without proper containment until it flowed. Mining tells a parallel story. The highest shares of mining-triggered fatal landslides in the record come from India at about twelve percent, Indonesia at roughly eleven point seven percent, then China, Pakistan, and the Philippines. Illegal or unregulated extraction shows up in thirty-two countries, with Indonesia logging twenty-four illegal mining landslide events and India fifteen. The consequence can be mass-casualty disasters. Myanmar's jade fields in Kachin state recorded 403 deaths from landslides in the period studied, tied to a boom in opportunistic scavenging on unstable spoil heaps where oversight is weak and accountability weaker. Demand for rare earths and gemstones doesn't just shape markets; it reshapes hillsides. Hill cutting—literally carving into slopes for building material or space—rounds out the human drivers, and it has a different human face. In parts of Bangladesh, India, and Nepal, it's often households and informal builders at risk. The data hint at that social pattern: in Nepal, at least forty percent of hill-cutting fatalities were children. Nine of eleven recorded deaths there involved harvesting colored clay for house decoration. In Bangladesh, most victims were adults—seventy-eight percent—and seventy-nine percent were male. Regulations exist on paper, especially in Bangladesh, but enforcement is patchy, expertise is thin on the ground, and the practical slope-stability guidance that would help a family decide where not to dig is hard to come by. As Holcombe and colleagues have argued, planning alone doesn't cut it in fast-growing, low-capacity settings; partnerships with non-governmental groups that can deliver community-based slope engineering matter. If you're keeping a mental list of caveats, keep going. This is a media-seeded database in English, after all. When Sepúlveda and Petley compared it to a Spanish and Portuguese subset for Latin America, the difference was only about five percent of records, and the missing events were mostly small. The team's best estimate is that global undercounting might be up to roughly fifteen percent, driven by language gaps and events that never make the news. But the workflow—multiple sources per entry, explicit trigger classifications, careful georeferencing, and a transparent record of what's missing, like earthquakes—makes the trends we've just walked through robust. Where does this leave us? With a map and a metronome. The map shows where people and slopes and human projects collide—Asia first, then a belt of landslide-prone regions across the tropics and mid-latitudes. The metronome ticks with rain: monsoons, typhoons, seasonal shifts that raise risk on a schedule. And layered over both is the human signature—construction missteps, unregulated mining, and everyday hill cutting that's pulling more people into the path of failure. The policy levers aren't mysterious. Tighten and enforce construction standards that reflect local geology. Manage spoil heaps like the hazards they are. Bring accountability to mining, especially the illegal corners where incentives point downhill. And where state capacity is thin, invest in community engineering that gives people practical tools to build safely. One last thought about the future. It's tempting to jump straight to climate change here, but the authors resist that. Over thirteen years, they can't cleanly tie global landslide patterns to El Niño or La Niña, let alone long-term warming. That's not a dodge; it's a call for better data. Longer, multilingual records; more local rain and landslide monitoring; and smarter, open inventories—like the GFLD now posted on ArcGIS and moving toward full public release—will let us separate weather from climate, and natural triggers from human ones. When we do, the science will be sharper. The stakes, sadly, will be the same: lives on steep ground, depending on whether we choose to learn from the last collapse before the next storm arrives.