December 4, 2018
Guest post by Bob Vislocky
In light of the devastating wildfires that ravished California last month, I thought it would be interesting to critically review a frequently referenced article published in the Proceedings of the National Academy of Sciences in 2016 titled “Impact of Anthropogenic Climate Change on Wildfire Across Western US Forests” by John T. Abatzoglou and A. Park Williams.
In the paper, which was the first and most significant quantitative effort to link climate change to wildfires, the authors conclude that roughly 50% of the wild fire acreage in the Western United States from 1979 to 2015 can be attributed to Anthropogenic Global Warming (AGW) as displayed in the figures below:
Not so surprisingly the article contains several serious deficiencies which will be presented below.
1) In the article, the authors create a wildfire regression model by correlating various weather parameters (temperature, humidity, etc…) to wildfire acreage. Then the AGW component of those parameters is removed from the raw weather data (using CMIP5 climate models) and the adjusted weather parameters are input to the wildfire regression model to determine the amount of acreage that would have burned without AGW.
However, the AGW components filtered out for maximum temperature in the summer and fall (which is the height of fire season) appear to be way too high (~2*C according to figure S1 in the article). In reality, the AGW component for maximum temperature during the summer and fall is closer to 0.5*C when computed from actual observations using the same baseline period as in the article (see graph below):
The estimates of the AGW component for minimum temperature in summer and fall were also inflated by the authors, but to a lesser extent. From the same figure S1 in their article it appears that from June to November they used an AGW contribution for minimum temperature around 1.5 to 2.0*C across the region. In reality, the AGW component for the summer and fall using actual observations is closer to 0.9*C (see figure below).
As a result of the authors’ desire to use climate models to determine the AGW component for the necessary weather parameters, their estimates of the AGW component for maximum and minimum temperature during fire season was roughly 2-4 times too high compared to trends in actual observations figured over a larger scale. Moreover, that 2-4 times factor is a conservative overestimation that assumes the entire observed temperature trend is due to AGW. For example, if the observed trend in max/min temperatures was 50% natural and 50% anthropogenic, then the authors would have overestimated the AGW component by 4-8 times. In the event the authors were to revise the study with more realistic AGW-adjusted temperatures then it is likely there would be much less attribution of wildfire acreage as a result of AGW.
2) Post-mortem attribution studies like these almost always lead to biased results. Let me explain how. Consider a variable like precipitation where the global trends over the past century are very weak or non-existent. Despite the lack of a global trend there will still always be several locations or regions where the local trend is highly positive or negative due to natural regional variability. All that is required is to select one of those regions, say New England, figure out that the number of 1+ inch rain events has increased over the last 100 years and then run an attribution study which shows the increased rains are correlated with global warming. In the end, however, that amounts to nothing more than cherry picking. Over the last century, there have been NO increases in the severity of drought during fire season when measured over a nationwide scale despite all the added CO2 (see figure below). Therefore none of the recent drought that the western US is currently experiencing can be attributed to global warming over natural regional variability.
3) Post-mortem event attribution studies like these are also inherently flawed because they look at the research area in isolation without consideration to the weather or climate around the rest of the world. An example will be provided using hurricane Michael that destroyed parts of the Florida Panhandle this year. All that is required is to subtract out the +1*C temperature anomaly from the Gulf of Mexico region caused by AGW and to re-run the computer models using the adjusted weather & sea surface data. When the results come back that Michael would only have reached Category 2 strength without AGW the climate scientists can claim that global warming significantly contributed to the storm’s intensity and all the newspapers will pick up on the story!
However the fatal flaw is that similar but opposite attribution tests are not performed simultaneously over other the rest of the hurricane season and other areas of open water around the world to see if a strong hurricane would have developed that otherwise didn’t because there was excess global warming in the wrong places. Otherwise, attribution studies could be run on all the hurricanes in the past couple decades to show that they all would have been 25-50% weaker without AGW. However, since there is no apparent overall trend in hurricane frequency or strength, an explanation would then be required to explain the mechanism by why hurricane strength would have been decreasing so substantially over the last few decades without an AGW component. That’s why it’s important to look at the overall trend on a national or global scale rather than fixate on a certain storm or region in doing these attribution studies.
In the case of wildfires, if the west was getting lots of rain instead of arid conditions this year, then maybe Texas or Florida would have gotten more wildfires instead, or perhaps the Midwest would have had a major drought that affected their crops. There is no way an event-driven attribution test can property detect an AGW component, especially since climate models are unable to reproduce natural regional variability.
4. The authors did not include known natural contributions to arid conditions in the Western US that have nothing to do with CO2, such as the Pacific Decadal Oscillation, which has been shown in several publications to be correlated to wildfires in the Western US, as seen in the figure below.
5) According to the US Forest Service, roughly 85% of all wild fires are started by humans. Since the population of the US has grown by 40% over their period of study, often extending into areas where previously there was less human activity, it would not be a stretch to claim that a portion of the growth in wildfires in the past several decades could be attributed to human carelessness unrelated to CO2. Moreover, over the past 20 years there has been a systematic 30-70% reduction in the fleet of large air tankers used by the Forest Service to combat wildfires. These and other human facets were not considered.
Now, in all fairness, the authors do provide a list of assumptions toward the end of the paper. In addition, they did provide a more conservative analysis that de-trended the burn data to account for potential extraneous correlations (such as those mentioned above in bullet points 4 & 5), and that modified analysis showed only 19% of the burned acres was due to AGW. However, this is buried in the middle of the article in a two-sentence blurb that nobody is going to read along with the assumptions, plus it still doesn’t address the first three points above. Unfortunately the near 50% contribution of AGW toward wildfire acreage mentioned in the abstract is what gets quoted when the study is referenced.
Conclusion: There is little question that recent drought conditions enhanced the spread of November’s wildfires in California. However, due to the many considerations expressed above it would likely be a huge over-reach to conclude that CO2-induced climate change is responsible for anywhere close to the 50% of the wildfire acreage estimated by the authors over the last few decades in the western US. Regrettably, results from their article spread like a wildfire through the media and has appeared in publications like the New York Times, Scientific American, and most recently the Fourth National Climate Assessment.