From Watts Up With That?

By Charles Rotter

In a compelling study published in the Journal of Applied Meteorology and Climatology (April 2025), Roger Pielke Jr., previously a professor at the University of Colorado Boulder and now a Senior Fellow at the American Enterprise Institute and professor emeritus, exposes critical flaws in a widely used dataset of U.S. hurricane losses, known as the ICAT dataset. This dataset, originally derived from Pielke’s own peer-reviewed work, was modified without documentation by an insurance company, leading to biased results in peer-reviewed studies and major climate assessments. Pielke’s paper, titled “Do Not Use the ICAT Hurricane Loss ‘Dataset’: An Opportunity for Course Correction in Climate Science,” is a clarion call for the climate science community to uphold rigorous standards and correct these errors.

A fatally flawed time series of U.S. hurricane losses assembled by an insurance company almost a decade ago has found its way into analyses published in the peer-reviewed literature. The flawed time series is based on undocumented modifications to a research-quality dataset that I and my colleagues published almost two decades ago.

Pielke’s study meticulously documents how the ICAT dataset, initially based on his team’s carefully curated hurricane loss data (Pielke et al. 2008; Weinkle et al. 2018), was altered by International Catastrophe Insurance Managers, LLC (ICAT) after a corporate acquisition. These changes, made without transparency or scientific rigor, introduced significant biases, particularly inflating post-1980 loss estimates. The result is a “Frankenstein dataset” that combines incompatible methodologies, rendering it unsuitable for research.

The ICAT dataset’s flaws are not trivial. Pielke demonstrates that it includes 61 additional loss events compared to the Weinkle et al. (2018) dataset, with 53 of these occurring in the latter half of the time series, creating an artificial upward trend in losses. Furthermore, post-1980 data were replaced with estimates from NOAA’s “Billion Dollar Disaster” (BDD) database, which uses a different methodology that inflates losses by including factors like government flood insurance and commodity effects.

The problem with replacing base damages originally from Pielke et al. (2008) with those from NOAA NCEI (and extending the dataset forward using NCEI data to 2017) is that the methodologies used to develop the hurricane loss estimates in each time series are very different. The loss estimates are simply apples and oranges.

This methodological mismatch has led to erroneous conclusions in studies like Willoughby et al. (2024) and Grinsted et al. (2019), which reported increasing trends in normalized hurricane losses and attributed them to climate change. Pielke shows that these trends disappear when the Weinkle et al. (2018) dataset is used instead, underscoring the ICAT dataset’s role in driving misleading results.

The findings of Willoughby et al. (2024) and Grinsted et al. (2019)—of an upward trend in normalized hurricane losses—are due entirely to the use of the flawed ICAT base damage time series and the Frankenstein extensions.

Pielke’s critique extends beyond data quality to the broader implications for climate science. The ICAT dataset’s use in high-profile reports, including the IPCC’s Sixth Assessment Report and the U.S. National Climate Assessment, risks undermining public trust in climate research. He argues that economic loss data, like hurricane damages, should not be used to detect climate trends, as direct meteorological data (e.g., hurricane landfall frequency) are more appropriate. Notably, studies like Klotzbach et al. (2018) and Vecchi et al. (2021) find no upward trend in U.S. hurricane landfalls since 1900, consistent with NOAA and IPCC assessments that refrain from attributing hurricane changes to greenhouse gas emissions.

There is no upward trend in landfall U.S. hurricanes or major hurricanes since 1900, so we should not expect to detect a change in normalized losses resulting from more or more intense landfalls.

Pielke’s paper is a model of scientific self-correction, emphasizing the importance of transparency and accountability. He calls for the retraction of papers relying on the ICAT dataset, arguing that their errors are “so obvious and significant” that they demand action to prevent further misuse. This stance aligns with the National Academy of Sciences’ guidance on ensuring research data integrity.

Science advances knowledge and sustains public trust in part because of the commitment of scientists to self-correction. … The errors are so obvious and significant that editorial boards from JAMC and PNAS should retract both of these papers to prevent the further misuse of a fatally flawed dataset.

This study is a timely reminder of the need for rigorous data provenance in science. By exposing the ICAT dataset’s flaws, Pielke not only protects the integrity of hurricane loss research but also offers a path forward for the scientific community to correct course. His work underscores the importance of grounding science, including “climate” science, in robust, transparent data to ensure that policy and public understanding are based on sound evidence.