Comment by nullc

The underlying paper is the opposite of convincing.

Essentially they fit a logistic regression of climate measurements the amount of area burned by wild fires each year in the forested parts of northern California to try to express how much is burned as a function e.g. of humidity, temperature, max temp, rainfall, wind, etc. Then the took historical weather data and eliminated the trend in order to try to construct an alternative time-series without human influence then their apply their aforementioned coefficients to figure out how much fire would be had in the counterfactual climate conditions.

To their credit (or perhaps their reviewers credit) the paper does observe the most obvious flaw the wildfires don't work that way-- that fuel builds up over time then is cleared by fires and once an area is burned it can't burn again for a long time. While the structure of the model is such that that if the air gets dry enough it will tell you that will constantly be fire everywhere forever no matter how much has already burned. They constructed a number of dynamic models that attempt to account for that and the increase largely disappears, with a constant level being shown for the next decade. True that the dynamic corrections seem even more adhoc (they don't seem to have data that allows them to fit the dynamic parameters), but the model that ignores these effects is pretty obviously wrong in a meaningful sense.

Even without that correction, their model doesn't fit the last ten years of data with many times the number of acres burned than the model predicts.

Their approach also has the effect that if run on the data from the first third of the study or so, it would instead result in claiming that climate change was reducing wildfires. (because wildfire acres burned were decreasing over that period)

More fundamentally, you could instead run the same analysis using any other measurements that increased over the same period that wildfires in the region increased and the model would come back attributing significant levels of wildfire to it. E.g. plugging in metrics of internet traffic growth into it looks like it would probably work even better. (See also: https://www.tylervigen.com/spurious-correlations )