By Sid Perkins of Nature magazine

Landslides claim an order of magnitude more lives each year than has been previously recognized, a study reports. Yet this annual toll of several thousand deaths worldwide may still underestimate the long-term average.

Although landslides, rockfalls and other 'debris flows' are common, the number of lives lost to them has been poorly quantified. This is mainly because databases tend to classify relevant fatalities as caused by the earthquake or weather event that triggered the landslide, rather than the landslide itself.

But David Petley, a geographer at the International Landslide Centre at Durham University, UK, has compiled his own global database using government statistics, aid-agency reports and research papers. Writing in Geology, he reports that between 2004 and 2010, 2,620 fatal landslides killed a total of 32,322 people. That figure excludes landslides triggered by earthquakes, and comes in at a little more than half the total number of people killed by floods, which claimed more than 7,600 lives annually between 1990 and 2006. Wildfires, by comparison, slayed 47 people per year, on average, between those dates. “These results are astounding in terms of the numbers of people dying from landslides,” says Rex Baum, a geologist with the US Geological Survey in Denver, Colorado. “This hasn’t been well documented on a global scale previously.” Petley reports that landslides tend to occur during the Northern Hemisphere summer and autumn, when monsoons strike eastern and southern Asia and hurricanes and typhoons slam Central America, islands of the Caribbean and land bordering the northwestern Pacific Ocean.

Fatality factors
In general, says Petley, fatalities happen where slopes are steep, there is plentiful rainfall for at least part of the year, and populations are dense. Although this combination of factors is unsurprising, it hasn’t previously been shown statistically, says Dalia Kirschbaum, a geoscientist at NASA’s Goddard Space Flight Center in Greenbelt, Maryland.

Despite the large numbers, the results are probably underestimates of long-term averages. Several regions purportedly prone to landslides, such as North Korea and Ethiopia, report few of the events that actually happen. Furthermore, many people who die from landslide-caused injuries long after the event itself are not included in final death tallies. And finally, the interval that Petley studied did not contain an extended El Niño — a climate pattern that usually increases the number of hurricanes striking the Caribbean and Central America, and can therefore dramatically boost occurrences of fatal landslides.

The data set doesn't cover enough years to be useful for making predictions, but it does hint that the frequency of landslides is rising.

“It’s certainly true that there are opportunities for future increases in fatalities,” says Baum. Populations in affected areas are growing and encroaching on deforested land that is prone to landslides. But without long-term data, he notes, it is hard to estimate just how high the death toll could mount.

This article is reproduced with permission from the magazine Nature. The article was first published on August 8, 2012.

2 Comments

1. 1. jtdwyer 01:21 PM 8/9/12

   It seems reasonable to guess that increasing populations and the construction of new residences in dangerous locations contribute to any increase in fatalities - the hills of L.A. being a prime example. It would be interesting to record for each landslide how long residences had been located at that site.

   Reply | Report Abuse | Link to this

2. 2. evosburgh 01:41 PM 8/9/12

   Question: is it that landslides are increasing in frequency or is it that we have better reporting/recording?
   
   I come to this same question each time I address a new time series data set, whether it be in my profession (reservoir characterization) or climate data (hurricanes included) or landslide data or whatever. The problem is that the data we collect now is much more complete that what has been recorded in the past due to sampling bias.
   
   For example: examine the historical hurricane records (which I have: http://weather.unisys.com/hurricane/) and see that in the past unless a hurricane made landfall or was encountered by a ship then it was not counted (re weather satellite days). Also the maximum intensity was not accounted for when it was not measured as we do now.
   
   So now we go and start looking at the relative intensity of hurricanes over time and what do we have to compare our current record against? What we have is an incomplete data set that needs some massaging in an attempt to give the historical data some of the depth that the current data set(s) have.
   
   Therefore the conclusions that are made are based upon data sets that needed some tuning, which is based upon some assumptions, and therefore we get results with error in them. The real trick is to accurately model the range of error in you solutions from which you are basing your conclusions.
   
   It turns out that it is all viscious cycle as you want to make forward projections from your work which is, in part based upon incomplete data. However, you want to state your results with confidence but you know there is error.
   
   So what do you do now? You have an incomplete data set that has a reasonable amount of error in it that you are applying analysis methodologies, which certainly also have some error in them, and coming to conclusions. I do this all day long every day, for the past 15+ years, and I have no reasonable way to explain my answers without the appropriate uncertainty error.

   Reply | Report Abuse | Link to this