Community resilience: storms, earthquakes, climate change, oh my!


Roger Pielke wrote an article at FiveThirtyEight about whether the rising costs of disasters are linked to climate change. The article got a lot of comments and generated some critical press As a result, Roger posted a followup to quell the critics.

What’s all the hubbub?

Roger’s conclusion is that rising disaster costs are not the result of climate change. Or at least it is not a major driving force. Instead, he argues that the cost of disasters isn’t really rising if you consider the increase in global GDP in the same timeframe. We put more toys in the sandbox and so more toys get damaged. GDP is basically a measure of how many toys the world has (and how many games we play, technically…or almost technically).

In other words, the rising cost of disasters can’t be the result of climate change because the cost of disasters really isn’t rising. Other studies have come to a similar conclusion.

I have my guesses as to why Roger’s article and studies are so controversial. I’ve had quite the reaction from friends on Facebook posting his and similar studies. There’s always confusion that the point being made is about whether climate change is occurring or is anthropogenic. But that’s not the point at all. The point is what Roger write’s at the end of his first article, “When you next hear someone tell you that worthy and useful efforts to mitigate climate change will lead to fewer natural disasters, remember these numbers and instead focus on what we can control.”

Admittedly, that sentence ends a bit ambiguously. But my interpretation (and self-serving extrapolation) of it boils down to this: community disaster resilience and climate change resilience are not the same thing. Hazard mitigation is needed, as well as (“worthy and useful”) climate change mitigation. Perfect climate change mitigation will not completely or, arguably, significantly mitigate against extreme hazard events. There is no doubt that there are commonalities between resilience or mitigation related to climate change and natural hazards. However, it is really important we don’t confuse the two.

I am not shy about saying that folks focused more on climate change than disasters tend to do this.

Community disaster resilience is about reducing the loss due to some distinct hazard event and then recovering from that loss. Loss and recovery can be represented in many different ways . Climate change isn’t a single event or it can only be seen as distinct at a very large temporal scale. In other words, the speed of onset is much, much(, much) slower than natural hazard events. The difference in temporal scale means that climate change isn’t a major consideration for community disaster resilience. Yes, I can succeed at my work without ever thinking about climate change. It doesn’t mean I shouldn’t or don’t, it means that its not necessary.

Roger and others have made the point empirically. I make the point theoretically in this post .

In Roger’s follow up article, he brings up the issue that some disasters are related to natural hazards that are not influenced by climate change — geophysical hazards. This, by the way, is absolutely not true. (It could only be true if water was not a factor in geophysical hazards. …water, water everywhere!) But, okay, climate change is a bigger factor for storms than earthquakes. Turns out that, as Roger reports, losses associated to weather events specifically are not rising much or maybe are even on the decline. So from the loss side of the community disaster resilience equation, maybe we’re doing okay—based on GDP as a metric of disaster resilience, anyway.


That result, plus a more forceful point in the comments to one of Roger’s articles, made me curious about comparing the losses from storm disasters to earthquake disasters. Are we doing better or worse with respect to earthquakes? GDP can’t represent the complexities of human development that are embedded in disasters. No single metric can, as Butan knows . But if we normalize storm data by earthquake data we can perhaps cancel out this variable. (One can argue that the exposed and impacted development between the two hazards are different. This is true; however, the same argument applies to GDP normalization.)

I grabbed data from EMDAT to give this a quick go (and I mean quick). Specifically, I downloaded data on occurrence, loss (in US$), deaths, injuries, and homelessness for storms and earthquakes. EMDAT defines occurrences as any event with more than 10 dead or 100 affected or international assistance was requested or an emergency was declared. They define storms as tropical storms, winter storms, and local/convective storms. Earthquakes are things that make the ground shake that are not caused by living creatures. (Okay, I made that one up.)

Since 1900, here’s the occurrence and loss trends for both storms and earthquakes. The rate of occurrence looks like it increases considerably about 1950. About 1990 for losses. (I’m going to eventually do an analysis to statistically nail down those dates.) I’ll let you guess as to why.


Here’s the same trends but just after 1980 since that is as far back as Roger shows in his two FiveThirtyEight articles.


Here’s that chart again, but you’ll notice I smoothed the data (rolling mean with an n=3 window). The randomness of hazard events justifies this, I think. I’ll be referring to just the smoothed data from now on. I did everything on the raw data, as well; the general insights from the data are the same but less emphatic in some cases.


I bet you’re curious how correlated all of those squiggles are.

For earthquakes the correlation between occurrence and losses is 0.85 between 1900 and 2013. This is based on the Spearman correlation, which is just checking to see if both trends monotonically increase at the same rate. It doesn’t check for linearity. For storms, the correlation is 0.94. (Dang.) If we look at just the data between 1980 and 2013, we see a change. For earthquakes, 0.28. For storms, 0.74. In the past few decades there’s been much less of a relationship between the occurrence of disasters and the amount of loss from them—more so for earthquakes.

What about the correlation between the same metrics between earthquakes and storms?

From 1900 onwards, the occurrence of storm and earthquake disasters have a correlation coefficient (monotonic, not linear) of 0.92. The correlation between the two hazard types with respect to losses is less: 0.84. Between 1980 and 2013, correlation goes down a bit: 0.82 for occurrence and 0.60 for losses.

Recapping: As we’d expect there’s a relationship between whether disasters occur and the cost associated with them (remember, EMDAT does not define disaster occurrence based on cost). Perhaps more surprisingly, there is also a strong relationship between the occurrence of earthquake disasters and the occurrence of storm disasters. The relationship isn’t as strong with respect monetary loss but its still strong. So maybe there is room for some other driving force to explain the variance? Climate change? I personally don’t think so. (It’s probably more that human development has more of a role in loss from storms than earthquakes—development puts more toys at risk, but it also worsens the hazard with impervious parking lots and the like.)

And what’s up with the past few decades? The relationships start to fall apart for all of the comparison above. Something is definitely up.

Another way we can compare earthquake and storm disasters is by looking at the differences between the various metrics: storm metrics minus earthquake metrics.

Simpler math!

Here is a plot of the difference between both occurrence and losses for storms and earthquakes. This is from 1900. Quickly you see that there have been an increasing occurrence of storm disasters with respect to earthquakes. Though, don’t get tricked by the scale of the y-axis. It only goes up to 80ish. There isn’t an obvious relationship with respect to the difference in losses, other than that craziness from 1980ish on. Really, the trend is rather erratic ( ?). It also looks like the rate of difference in occurrences goes up considerably about that time.


Clearly the reason is Reagonomics .

Here’s the same graph, but just showing the crazy years. Apparently storm disasters took a bit of a break after Hurricane Katrina in 2005. Awfully considerate, don’t you think?

earthquake_storm_diff1980 You might be wondering what the trends look like for other disaster metrics. Money isn’t everything, you know.

Below, is a figure that adds deaths, injuries and the number of homeless. Again, its the difference between storms and earthquakes. The most obvious thing here I think is how much more volatility is in the data for these additional metrics. Overall, the trend for losses and deaths has stayed constant. For injuries and homelessness, you don’t see much difference until the 1960s. I’d like to say this is because we learned to save lives and so there were more people alive to be injured and homeless. Maybe. But I’d also guess its a data quality thing—deaths were easier to track back then.


You’ll notice I computed a trend line for each metric so you don’t have to do linear regression visually. (You’re welcome.)

Here are the trend lines for just the crazy period after 1980.



If you put any stock in the data and the utility of the trend lines, with respect to earthquakes, we can make the following generalizations between 1980 and 2012:

  • The occurrence of storm disasters has gone up (by about 40 per year)
  • The monetary loss from storm disasters has gone up a little (by about $8.5billion per year)
  • The number of deaths from storm disasters has gone down (about 70,000)
  • The number of injuries from storm disasters has gone down (about 40,000)
  • The number of homeless from storm disasters has gone down (about 1.5 million)

Can we make any conclusions about whether, as a global community, we have become any more resilient to storms than to earthquakes over time?

First, any conclusion to this question can only be in terms of the loss side of resilience. Clearly, if there is zero loss, there doesn’t have to be any recovery and a community is perfectly resilient. But this never happens, so our conclusion is only looking at a very partial representation of resilience. Second, it depends on what variables you think should be used to represent resilience. Broadly, I think those variables are well-being , identity , services , and capitals . The metrics that EMDAT has data for are very imperfect ways of representing the variable of well-being, but not the other three variables.

Regardless, we can say something about what this data shows. Given an increasing number of storm disasters, it seems okay to say that the world is becoming more resilient to storm disasters than to earthquake disasters with respect to the metrics of deaths, injuries, and homelessness. It seems like its draw with respect to losses, since both storm occurrence and storm losses are going up relative to earthquakes. Maybe someone can see this as a chance to say “because climate change.”  But that ignores the fact that the actual losses have been extremely erratic since the 1980s–sometimes more for storms and sometimes less. And remember that Roger Pielke pointed to work suggesting that costs from storms could be going down.

The take away for me at the moment is that we need to spend time thinking about what’s been going on the last few decades with respect to vulnerability around the world. (There are certainly folks who do this.) And, well, I think we need some more progress with respect to seismic safety and improving community resilience to earthquakes.

If you want to play with the data, my is . (Non-programmers should check this out to see the analysis in action.) And you can grab the data .

Comments are closed.