A few days ago, reader Carl commented that the National Weather Service forecasts for overnight low temperatures in Fairbanks seem to have been biased high recently, with many occasions of low temperatures near or below freezing when much higher temperatures were predicted.
To look at this objectively, I was able to pull a history of NWS forecasts all the way back to 2009, and it turns out that the mean absolute error on the low temperature forecasts in the month of May was the highest in the data set.
The errors, as Carl observed, were largely due to a warm bias; the low temperature forecasts were generally warmer than the corresponding outcome.
(A small caveat is in order here: I'm comparing the NWS forecasts for "Fairbanks Metro Area" to the outcome at the airport, and there are microclimate differences that vary depending on the weather pattern.)
Here's a look at the daily errors so far in 2026: the warm bias for minimum temperatures has been quite persistent.
For extra credit, I also did a comparison of the recent NWS forecasts with pure model forecasts (NBM - National Blend of Models) for Fairbanks airport - see below, with the black line in the lower panel revealing the frost and freeze events that Carl mentioned. As we might expect, the NWS forecasts closely follow the model predictions, so we can basically blame the recent warm bias on the models. Ideally the forecasters might have adjusted the model numbers to improve the results, and actually it looks like they did start making a downward adjustment to the low temperatures starting on May 25.
None of this should be seen as criticism of the NWS folks in the Fairbanks office, who cover the weather over a staggering 359,000 square miles, over 4 times more area than any NWS office in the lower 48.
