Roger Pielke Contributor
I research and write about science, policy and politics.
Earlier this week a paper published by the Proceedings of the National Academy of Sciences (PNAS) by a team of authors led by Aslak Grinsted, a scientist who studies ice sheets at the University of Copenhagen, claimed that “the frequency of the very most damaging hurricanes has increased at a rate of 330% per century.”
The press release accompanying the paper announced that United States mainland “hurricanes are becoming bigger, stronger and more dangerous” and with the new study, “doubt has been eradicated.”
If true, the paper (which I’ll call G19, using its lead author’s initial and year of publication) would overturn decades of research and observations that have indicated over the past century or more, there are no upwards trends in U.S. hurricane landfalls and no upwards trends in the strongest storms at landfall. These conclusions have been reinforced by the assessments of the Intergovernmental Panel on Climate Change (IPCC), U.S. National Climate Assessment, and most recently of the World Meteorological Organization.
In fact, however, the new PNAS paper is fatally flawed. The conclusions of major scientific assessments remain solid. As I’ll show below, G19 contains several major errors and as a result it should be retracted.
The first big problem with G19 is that it purports to say something about climatological trends in hurricanes, but it uses no actual climate data on hurricanes. That’s right, it instead uses data on economic losses from hurricanes to arrive at conclusions about climate trends. The economic data that it uses are based on research that I and colleagues have conducted over more than two decades, which makes me uniquely situated to tell you about the mistakes in G19.Compare the counts of hurricanes reported in G19 with those that can be found in climate data from the National Oceanic and Atmospheric Administration.
From 1900 to 1958, the first half of the period under study, NOAA reports that there were 117 total hurricanes that struck the mainland U.S.. But in contrast, G19 has only 92. They are missing 25 hurricanes. In the second half of the dataset, from 1959 to 2017, NOAA has 91 hurricanes that struck the U.S., and G19 has 155, that is 64 extra hurricanes.
The AP passed along the incorrect information when it reported that the new study looks at “247 hurricanes that hit the U.S. since 1900.” According to NOAA, from 1900 to 2017 there were in fact only 197 hurricanes that made 208 unique landfalls (9 storms had multiple landfalls).
Part of this difference can be explained by the fact that G19 focus on economic damage, not hurricanes. If a hurricane from early in the 20th century resulted in no reported damage, then according to G19 it did not exist. That’s one reason why we don’t use economic data to make conclusions about climate. A second reason for the mismatched counts is that G19 counts many non-hurricanes as hurricanes, and disproportionately so in the second half of the dataset.
The mismatch between hurricane counts in G19 versus those of NOAA by itself calls into question the entire paper. But it gets much worse.
The dataset on losses from hurricanes used by G19 to generate its top-line conclusions is based on my research. That dataset has been maintained by a company called ICAT located in Colorado. The ICAT dataset was initially created about a decade ago by a former student and collaborator of mine, Joel Gratz, based entirely on our 2008 hurricane loss dataset (which I’ll call P08).
In the years since, ICAT has made some significant changes to its dataset, most notably, by replacing P08 loss estimates with loss estimates from the “billion dollar disasters” tabulation kept by the NOAA National Centers for Environmental Information (NCEI). The replacement data begins in 1980, at the start of the NCEI dataset.
This process created a new hybrid dataset, from 1900 to 1980 the ICAT dataset is based on P08 and for 1980 to 2018 it is based on NCEI. This is hugely problematic for G19, which was apparently unaware that of the details of the dataset that they had found online.
In our comprehensive update of P08 published last year (Weinkle et al. 2018, or W18) we explained that the NCEI methodology for calculating losses included many factors that had historically not been included in tabulations of the U.S. National Hurricane Center, “for instance, to include federal disaster aid, federal flood insurance payouts, national and local agricultural commodity effects and other macro-economic impacts.”
That meant that one cannot, as ICAT has done, simply append the NCEI dataset from 1980 to the end of the P08 dataset starting in 1900. They are not apples to apples. Indeed, a big part of our work in the W18 update of P08 was to ensure that the data was apples to apples across the entire dataset, and we performed several statistical consistency checks to ensure that was the case.
The new PNAS paper, G19 unwittingly uses the ICAT dataset that staples together P08 and NCEI. I have shown with several graphs on Twitter why this matters: Before 1940, G19 and W18 loss estimates for individual are just about the same. After 1980, however, G19 loss estimates for individual storms are on average about 33% higher than those of W18. The result is a data incontinuity that introduces spurious trends to the dataset.