If I look at the year-to-date data from a year ago, what would I predict is the burned area of ​​acres in 2020?

If I knew nothing about the changing patterns of wildfire burning, I might come to the conclusion that “everything is clear”. To see this, consider the following picture I drew a year ago:

figure 1: The number of acres burned (black) and the number of acres burned (red) from August 4 to present (YTD), and the log-linear regression fitting based on the sample from 1983 to 2019 (gray). source: NIFC1, NIFC2 , And the author’s calculation.

It turns out that the return of log acres to log acres-YTD will provide considerable adj-R² About 0.88. However, coupled with the fact that 2020 illustrates why looking at things prematurely without any institutional or historical knowledge can lead to misleading, a comment written a week ago perfectly illustrates this point:

Year-to-date, the number of fires is slightly higher than the 10-year average; the burned area is far below the 10-year average.

A clear hint has nothing to worry about. This is today’s photo.

figure 2: Year-to-date August 4 (red), September 28 (blue-green) and acres burned (black) and acres burned based on the log-linear regression fitting of the 1983-2020 sample (grey). source: NIFC1, NIFC2 , And the author’s calculation.

In other words, at these times, many things may happen in August.This is the potential outlook for the fire in August National Financial Center:

source: National Financial Center, Visited on April 8, 2021.

NIFC Monthly seasonal outlook The August report is here.

Adding 2020 to the regression will lower adj-R² To 0.43 (although the slope coefficient is still statistically significantly different from zero). Interestingly, during the period 1983-2019, using the year-to-date September 28th as the regression variable adj-R2 is 0.99, and if 1983-2020 is used, it drops slightly to 0.96.

Therefore, with 2020 data, it is best to forecast 2021 with humility and caution.