The upward trend in acres burned is shown below.
Figure 1: Acres burned (black) and acres burned year-to-date (YTD) 28 July (red), and log linear regression fit (gray), forecast based on log first differences regression 2012-18 (purple x) and plus/minus one standard error (light small purple x). Source: NIFC1, NIFC2 , and author’s calculations.
If one were to use year-to-date numbers to infer the severity of the season to come (as has at least one commentator – the same one who also had questions about confidence intervals), one could make a serious error. All one has to do is to absorb the oft-remarked point that fire season is expanding in time, so that it runs later into the year. The trend in divergence between full-year and year-to-data figures — i.e. the gap between black and red line — is consistent with this point.
How to use this fact in order to get a handle on 2021 acres burned? Examination of the first difference of logged variables is illuminating.
The correlation breaks down for 2019 and 2020, but overall seems otherwise strong. I use that correlation between 2012-18.
The regression equation is:
d(log(ACRES)) = 0.046 + 0.894×d(log(ACRES_YTD))
bold denotes significance at 1% msl using HAC robust standard errors. Adj.-R2 = 0.86. DW = 2.45, n=7.
Hence, even without believing in any regression results, one should be circumspect in using YTD numbers this early in the fire season to infer final annual numbers, particularly in the new climate normal I believe we are in.