If I didn’t know anything about the changing pattern of wildfire burns, I would’ve probably concluded “all clear”. To see this, consider the following picture I could’ve drawn a year ago:

Figure 1: Acres burned (black) and acres burned year-to-date (YTD) 4 August (red), and log linear regression fit based on 1983-2019 sample (gray). Source: NIFC1, NIFC2 , and author’s calculations.

As it turns out, a regression of log acres on log acres-YTD would have provided a respectable adj-R² of about 0.88. However, adding 2020 shows why looking at things early — and without any institutional or historical knowledge — can lead one astray, perfectly illustrated by one comment written a week ago:

Year to date, number of fires is just above ten year average; acres burned is well below ten year average.

With the clear implication that there isn’t much to worry about. Here is the picture as of today.

Figure 2: Acres burned (black) and acres burned year-to-date 4 August (red), yeat-to-date 28 September (teal), and log linear regression fit based on 1983-2020 sample (gray). Source: NIFC1, NIFC2 , and author’s calculations.

In other words, in these times, a lot can happen in August and September. Here is the August and September fire potential outlook from the NIFC:

Source: NIFC, accessed 8/4/2021.

Source: NIFC, accessed 8/4/2021.

NIFC’s Monthly Seasonal Outlook report for August is here.

Adding in 2020 into the regression drops the adj-R² to 0.43 (although the slope coefficient is still statistically significantly different from zero).  Interestingly over the 1983-2019 period, using 28 September YTD as the regressor yields an adj-R2 of 0.99, declining slightly to 0.96 if using 1983-2020.

So, with 2020’s data in hand, best to approach forecasting 2021 with humility and circumspection.