Haiti is still rebuilding from the massive earthquake that struck 11 years ago, as well as dealing with the fallout of its president’s assassination in July. So the island nation was ill-prepared for the magnitude 7.2 earthquake that hit its western region on August 14. At the time of publication, the official death toll hovers around 2,000, although for the first few critical days after the quake, media reports listed a death toll in the hundreds.

But the true number of deaths is probably at least five to 50 times that number, according to a scientific model from the U.S. Geological Survey. Its tool, known as Prompt Assessment of Global Earthquakes for Response (PAGER), automatically combines information about an earthquake with demographic and other data from the affected region to model the likely scale of the disaster, including deaths and economic impact.

PAGER, which was launched in 2007, does not predict the exact number of fatalities. Rather it estimates the probability that the number lies within a certain range. For the recent earthquake in Haiti, PAGER gives a 35 percent chance that between 10,000 and 100,000 people have died and a 32 percent chance that the death toll will top 100,000.

This earthquake was probably less deadly than the one that hit Haiti in 2010, for which PAGER correctly predicted more than 100,000 deaths. The tool also estimates a 66 percent chance that the cost of damage for this year’s quake lies between $100 million and $10 billion. For comparison, the 2010 earthquake cost between $7 billion and $14 billion.

Having an accurate estimate of deaths and damage early on could greatly influence the scale of the response. News organizations tend to report the official death count, which is the number of bodies found so far. In a country such as Haiti, which has poor infrastructure and limited access to remote regions, determining the real toll this way could take months. This creates a problem for international responders, humanitarian aid organizations and other nations, all of who need to assess the extent of the damage and determine an appropriate response.

Scientific American spoke with David Wald, a seismologist at the USGS who helped develop PAGER and other earthquake modeling tools.

[An edited transcript of the interview follows.]

Why are Haiti’s official fatality numbers so much lower than PAGER’s?

There are a lot of things that you don’t see that the models predict. What you’re not seeing is all the small towns and villages and remote buildings, some of them on these hillsides that may be affected not only by shaking but by landslides. It’ll be a long time before those areas are all reached, especially because access and civil infrastructure and governance are all challenged. So those [fatality] numbers will tend to grow over time, as we always expect. We can say, very definitively, that there are going to be problems with the roads because of landslides—because the roads cross the steep terrain where there was strong shaking.

What data are used by PAGER to model fatalities?

Getting the exact number of fatalities is unachievable. There are just too many uncertainties and too many unknowns. But to create a prediction, there are three ingredients: the shaking, the population exposed to each shaking level and how vulnerable that population is, based on the buildings [people] are in.

We know that for larger earthquakes, the pattern of shaking can be extremely complicated, so we try to capture that as best we can through a tool called ShakeMap. If you had thousands of stations, you would know the shaking everywhere, and it would be a really well-constrained map.

In California, you have hundreds of stations. If you go to Haiti, where there are not very many seismic stations, there is additional uncertainty. It’s always going to be more uncertain until we help equip Haiti with more instruments. The second ingredient is the population exposed to these different shaking levels. You can just take a population grid and calculate the population exposed for all of the different shaking intensity levels. The final question is the impact of that. In China or Haiti, you have very vulnerable buildings. For the same shaking level, you can get many times more fatalities than you would in, say, California or New Zealand or Japan, where you have better building codes.

In the case of Haiti, there’s a high population, strong shaking and very vulnerable buildings. And that leads you to the conclusion that there’s going to be what we call a red alert on PAGER, where we have probably 1,000 fatalities or higher.

Why is there so much uncertainty?

You don’t have many instruments. Haiti has a seismic network where it can locate earthquakes. But we can’t use those instruments for the ShakeMap. They have to be special instruments called strong motion instruments. There are probably six or seven in Haiti, and we only have the data from two of them.

An additional source of uncertainty in our calculations is trying to figure out [the location and shape of] the fault. It also has dependencies on the data and how good the data are and how complicated the earthquake was. In this case, it’s a pretty complicated, challenging fault to determine. Sometimes those things take months to really iron out. But we always do our best to do it within the first few hours and days.

For the modeling, a big problem is: we don’t even know how many people died in 2010. The estimates range from 100,000 to 320,000. Most countries have very good reporting, and if the number of fatalities is three, four or five, you can assume it’s pretty damn accurate. But once you get into these really big losses, such as in Haiti, the truth is uncertain.

We can never be exact in this business, but we can be useful. And for Haiti, it was in the red alert, no matter what, so it was an international type of response.

How does PAGER record an earthquake when it strikes?

First of all, the National Earthquake Information Center [determines] the magnitude and epicenter. It operates 24/7, so that information gets sent out and triggers ShakeMap, which will generate a map of the shaking with the seismic data that are available. As soon as the ShakeMap is made, PAGER runs. It takes the shaking, and it overlays population. And with the model for that country, it estimates fatalities. It also sends that ShakeMap around the world to systems that estimate shaking and damage.

Most earthquakes are in the green zone because they’ll be in the ocean or a low population area. Even magnitude 8.0s are often in the subduction zones [collisions between tectonic plates in which one sinks beneath the other] offshore and don’t affect anything. One of the most important things about PAGER is to say that nothing happened as opposed to that something happened.

If it’s an orange or red alert like this was for Haiti, however, it will page us, and we can watch the model get generated. We will review the results before sending those out. We sit on it for maybe 10 or 20 minutes while we look at the other information that’s coming in, such as better magnitudes and better locations, and we’re confident that that’s a good starting place.

News reports use the official numbers, which are probably far lower than the actual numbers. Does that mean people will pay less attention to the disaster?

With the media, there have been a number of cases where an earthquake will happen at night, and nobody will be paying attention. Then it’ll be another news cycle, with Afghanistan and COVID and so many other things, so the media are paying attention to those.

We’re saying, “It’s going to be worse than the initial reports. It’s going to be much worse.” We may be overestimating the total losses, based on some of these uncertainties. It’s a challenge for the media to work with uncertain numbers. And it’s a challenge for the agencies and the financial institutions and urban search-and-rescue teams. But we’re pushing them in the direction of “it’s worse than we’ve seen so far.”

Different PAGER users have different timelines. Urban search and rescue should begin within hours, so you go with these uncertain numbers. Whether to send $1 billion worth of aid can wait a little bit longer. Over days, things tend to stabilize, and you reach the answer that’s a little bit more constrained than the initial estimates. So you can take the model as an uncertain estimate, along with what’s happened on the ground, and weigh those appropriately.