stochtastic

You might be interested to know that private sector weather forecasting firms have a long history of lobbying to prevent NOAA/NWS from building its own end user services and apps [0].

I agree that NWS [text] forecasts are usually the best, though it can be time consuming to digest them. Nate Silver's book [1] makes the interesting point that commercial forecasts almost always reduce the quality of the input data they're given from NOAA, but that part of this boils down to incentives: Nobody complains if you forecast a small chance of rain and it turns out to be sunny. The problem spot is in ruining someone's picnic. Hence, forecasts tend to bias heavily towards rain.

[0]: https://www.bloomberg.com/news/features/2018-06-14/trump-s-p...

[1]: https://en.wikipedia.org/wiki/The_Signal_and_the_Noise

I wrote a chapter of my PhD explaining why the linked PNAS paper is highly misleading (see the letter we published in PNAS in response for a summary: http://www.pnas.org/content/110/7/E546.full and a somewhat extended comment including figures here: http://stochtastic.blogspot.com/2013/02/temperature-extremes... )

TL;DR: While regional normalization makes sense in a stationary climate, the presence of trends tends to exaggerate extremes by increasing the apparent variance of the distribution.

Eyeballing the data from the NYT article, it appears that the authors have continued to make the same statistical mistakes. It is unfortunate that NYT would publish this, as the field (climate science) is quite aware of the problem with the original 2012 paper.