Cherry picked and misrepresented climate science undermines FiveThirtyEight brand
Posted on 25 March 2014 by dana1981, Albatross, thingsbreak
Nate Silver has launched a new FiveThirtyEight blog with the intent of applying his data-driven approach to a wide variety of subjects. The problem is that Nate Silver is himself only one man, so FiveThirtyEight has hired a variety of contributors to write about the subjects that are outside his expertise and comfort zone. For the topic of climate change, Silver decided to hire the renowned obfuscator Roger Pielke, Jr.
This was immediately disappointing for those familiar with Pielke's work, because FiveThirtyEight is a statistics site, and frankly Pielke is not good at statistics. Instead, Pielke is known for taking a selective view of the peer-reviewed scientific literature in order to downplay the connection between human-caused global warming and extreme weather. Predictably, Pielke's first two posts at FiveThirtyEight did exactly that, and included a litany of errors:
- The headline and main point of his post are wrong.
- He misrepresents his own research.
- The references he provides don't say what he claims and don't support his argument.
- Research he neglects contradicts his conclusions.
- He doesn't include all available data.
- He incorrectly claims that weather-related disasters aren't becoming more frequent.
- He fails to account for the costs of improved technology and the damages they prevent.
- He considers only land-falling hurricanes whose damages are highly variable.
- His conclusions are contradicted by the increased intensity of North Atlantic hurricanes, and global warming's contribution to their storm surges and flooding.
Pielke vs. Munich Re
Pielke's first FiveThirtyEight post begins by plotting global disaster economic losses, from data provided by reinsurance company Munich Re. Except for some reason Pielke only plots the data from 1990 to 2013, when Munich Re provides data beginning in 1980.
The Center for Research on the Epidemiology of Disasters also has natural disaster cost estimates ranging back to the year 1900. And ironically, when Munich Re published its report, Pielke criticized them for only including data since 1980, saying,
"Thirty years is not an appropriate length of time for a climate analysis, much less finding causal factors like climate change"
Yet Pielke has dismissed climate change as a causal factor using data from just 1990.
As it turns out, the positive trend in global disaster losses in the Munich Re data is almost 30 percent larger for 1980–2013 than it is for 1990–2013. What's more, the trend in disaster losses for 1980–2013 is statistically significant at the 99 percent confidence level, whereas the trend for the 1990–2013 window cherry picked by Pielke is not statistically significant.
Pielke then simply took these disaster costs as a proportion of global Gross Domestic Product (GDP), finding that the trend then flattens out since 1990, because while disaster loses have grown, so has GDP. Based on this less than robust analysis, Pielke asked three important questions, and wrongly claimed the answer to each is "no."
"In the last two decades, natural disaster costs worldwide went from about $100 billion per year to almost twice that amount. That’s a huge problem, right? ... no"
Whether or not the rising costs of extreme weather is "a huge problem" is subjective; many would disagree with Pielke's answer to this first question (for example, many residents of New Jersey).
"Indicative of more frequent disasters punishing communities worldwide? ... no"
Pielke's answer to this second question is unambiguously wrong. For starters, he's looking at the wrong data. To determine if weather disasters are becoming more frequent, you don't look at economic loss data. Munich Re also provides data about the number of annual disasters, and the frequency of these events is indeed rising. Pielke claims,
"When you read that the cost of disasters is increasing, it’s tempting to think that it must be because more storms are happening. They’re not."
They are unequivocally increasing as illustrated in the figures above and below.
"Perhaps [rising disaster losses are due to] the effects of climate change? ... no"
Munich Re disagrees with Pielke on the answer to this third question, concluding following a study of "Severe Weather in North America" that (emphasis added),
"Among many other risk insights the study now provides new evidence for the emerging impact of climate change. For thunderstorm-related losses the analysis reveals increasing volatility and a significant long-term upward trend in the normalized figures over the last 40 years. These figures have been adjusted to account for factors such as increasing values, population growth and inflation ... In all likelihood, we have to regard this finding as an initial climate-change footprint in our US loss data from the last four decades."
As we'll see below, several peer-reviewed papers neglected by Pielke are consistent with the conclusions of Munich Re.
Perhaps Pielke is simply confusing storm frequency with 'normalized' storm costs. After all, his main argument is that disaster costs are only rising because we've become wealthier.
"In reality, the numbers reflect more damage from catastrophes because the world is getting wealthier. We’re seeing ever-larger losses simply because we have more to lose — when an earthquake or flood occurs, more stuff gets damaged."
It's true that some of the rising costs due to these disasters can be explained by increased wealth. As the first figure above shows, geophysical events like earthquakes aren't becoming more frequent, but they are costing more than they used to. However, the frequency of storms and floods is increasing rapidly, and their costs are rising faster than the costs of earthquakes. This suggests that something (i.e. climate change) is adding to the frequency and costs of these disaster events.
Pielke's research focuses on the 'normalized' costs of land-falling hurricanes in the US. In those papers, his approach is similar to that in his FiveThirtyEight post, but he accounts for rising wealth and population density in regions prone to hurricanes. When accounting for a few such factors, Pielke finds no long-term trend in 'normalized' damages, and thus concludes that climate change isn't making these storms any more expensive. He argues we've just got more stuff in areas hit by hurricanes, and that increase in stuff is what's causing storm costs to rise.
However, Pielke ignores some important factors in his normalization procedure. For example, our engineering has improved significantly over the past century to make buildings more resistant to hurricane damage. We now have instruments orbiting the Earth on satellites tracking storms, weather models predicting their paths, and communications technology to warn people well before they make landfall. This technology costs money, as evidenced by the fact that funding cuts may soon cripple our ability to predict hurricane tracks, for example. Yet Pielke fails to account for the costs of these improved technologies. This point was made by Judith Curry in 2007.
"The second problem with the analysis is that the paper does not account for major engineering improvements that rendered these regions in Florida less susceptible to damage."
Kevin Trenberth also made this point in Science in 2010.
"He completely ignores the benefits from improvements in hurricane warning times, changes in building codes, and other factors that have been important in reducing losses."
Munich Re also disagrees with Pielke (emphasis added),
"Several of the events of 2013 illustrated how well warnings and loss minimisation measures can restrict the impact of natural catastrophes. In the case of the most recent winter storms in Europe, for example, the losses remained comparatively low”, said Torsten Jeworrek, Munich Re Board member responsible for global reinsurance business."
The other issue is that focusing only on land-falling hurricanes significantly shrinks the available data and introduces a lot of variability. As Trenberth notes, the damage done by a hurricane depends on many factors like precisely where it makes landfall, the angle of the track with respect to the coast, and what time of day it strikes.
Curry also points out that the lack of a long-term trend in hurricane damage losses in Pielke's analysis depends heavily on two large, expensive hurricane strikes in the 1920s (the Great Miami Hurricane of 1926 and the Lake Okeechobee Hurricane of 1928). However, property values were badly inflated leading up to the US stock market crash of 1929, and Pielke fails to account for this inflation factor. Curry concludes,
"If you omit the data prior to the 1930’s, and look for the decade early in the period with the largest total damage, it turns out to be 1936-1945 ... The period post 1929 with the greatest amount damage is 1996-2005, which is 84% greater than the period 1936-1945. Such a conclusion is counter to Pielke’s conclusion that found no trend in damage."
Pielke Misrepresents His Own Research
In a follow-up post on FiveThirtyEight to try and justify his first entry, Pielke dug himself even deeper into a hole by claiming that efforts and technologies to mitigate disaster damages don't make a difference in damage trends "for floods, U.S. hurricanes or tornadoes." The problem is that those referenced papers he links don't support his claims.
The US hurricanes reference goes to one of Pielke's own papers, published in 2008. That paper specifically states,
"The normalization methodologies do not explicitly reflect two important factors driving losses: demand surge and loss mitigation. Adjustments for these factors are beyond the scope of this paper, but it is important for those using this study to consider their potential effect."
The paper includes a brief discussion of improved building codes in Florida potentially reducing disaster losses by up to 40%, but then dismisses their importance by claiming "As strong codes have only been implemented in recent years and in some cases vary significantly on a county-by-county basis, their effect on overall losses is unlikely to be large." However, this speculation is not substantiated with any sort of analysis, does not include areas outside of Florida, and does not include mitigation measures other than improved building codes.
In fact, a study of Florida building performance found that homes built after 2002 sustained less average hurricane damage than those built between 1994 and 2001, which in turn sustained less damage than those built before 1994. This is strong evidence that improvements in building resistance to hurricane damage is making a marked difference in damage losses – a difference Pielke does not account for.
Elsewhere, Pielke referenced a 2011 paper by Barthel & Neumeyer to try and support his argument that mitigation makes no difference in cost trends. However, the same authors published a paper in 2012 that concluded (emphasis added),
"A trend analysis of normalized insured damage from natural disasters is not only of interest to the insurance industry, but can potentially be useful for attempts at detecting whether there has been an increase in the frequency and/or intensity of natural hazards, whether caused by natural climate variability or anthropogenic climate change...We find no significant trends at the global level, but we detect statistically significant upward trends in normalized insured losses from all non-geophysical disasters as well as from certain specific disaster types in the United States and West Germany."
Another paper, Schmidt et al. (2009) also looks at US hurricane losses and concludes,
"In the period 1971–2005, since the beginning of a trend towards increased intense cyclone activity, losses excluding socio-economic effects show an annual increase of 4% per annum. This increase must therefore be at least due to the impact of natural climate variability but, more likely than not, also due to anthropogenic forcings."
In short, Pielke's own paper does not remotely support Pielke's claim that mitigation doesn't make a difference in US hurricane disaster damages. Pielke has misrepresented his own research. The other referenced papers don't seem to support Pielke's claims regarding flood or tornado mitigation either, and research Pielke neglects does find rising trends in normalized disaster losses in some areas, like from US hurricanes and severe thunderstorms.
Climate Change is Strengthening North Atlantic Hurricanes and Damages
It's also not surprising that hurricanes would now be doing more damage, because research has shown that the most intense hurricanes are already occurring more often as a result of human-caused global warming. This, combined with rising sea levels, has also led to larger storm surges and the costs of the damage that goes with them. As Grinsted et al. (2013) concluded,
"we have probably crossed the threshold where Katrina magnitude hurricane surges are more likely caused by global warming than not."
Global warming also adds moisture to the atmosphere, with the increase in precipitation also adding to the flooding associated with these storms, and the damages they cause.
Pielke's Contradictory Conclusion
Oddly, Pielke seems to contradict himself in his concluding statement, saying,
"As countries become richer, they are better able to deal with disasters — meaning more people are protected and fewer lose their lives. Increased property losses, it turns out, are a price worth paying."
This implies that Pielke understands that the costs of the technologies that he neglects, because those costs "are a price worth paying." Apparently they're just not worth including in his calculations.
It's also worth noting that another of Pielke's claims, "In fact, today’s climate models suggest that future changes in extremes that cause the most damage won’t be detectable in the statistics of weather (or damage) for many decades," is not remotely true, as illustrated in this post. To support this incorrect claim, Pielke references his own work, which is limited to the statistics of economic loss data for tropical cyclones only (and is contradicted by other research on that subject, as discussed above). However, trends in many extremes are already detectable, including in North Atlantic hurricane intensity (e.g. see Emanuel 2005, Elsner et al. 2008, Knutson et al. 2010, Emanuel 2012, Kunkel et al. 2013, Grinsted et al. 2013, Holland and Bruyère 2013). Pielke's own paper even acknowledges this point, so once again he's misrepresented his own research:
"This result confirms the general agreement that it is far more efficient to seek to detect anthropogenic signals in geophysical data directly rather than in loss data"
Note that while Pielke's post has received extensive criticism (e.g. Climate Progress, Things Break, The Huffington Post, Daily Kos, and Columbia Journalism Review), some have claimed that criticism is because he "flunked the green purity test" (i.e. National Journal). The latter argument falls prey to the false "honest broker" narrative. Pielke isn't criticized because of a lack of "purity," he's criticized because he consistently provides a skewed representation of the body of peer-reviewed science, downplaying links between climate change and extreme weather and only focusing on areas where those links are uncertain. On FiveThirtyEight he's done even worse, making a number of false claims, including about his own research. That is why those of us who care about accurately representing the scientific literature criticize Pielke.
The bottom line is that many types of extreme weather are being intensified by human-caused global warming, and that will continue in the future. And there is evidence that climate change is adding to the costs of extreme weather damage. There's an important lesson for FiveThirtyEight to learn here – sometimes conclusions are counter-intuitive because they're wrong.