Cold air coming this weekend, MOS can’t predict

Unless all the major computer models are wrong (the GFS, NOGAPS, European, and Canadian), a strong cold front will blast through Alabama late this week, bringing the coldest air of the season by far. We could see highs on Friday and Saturday struggle to reach 55 degrees, with strong north winds making it feel colder. And, by Sunday morning, temperatures in the 30s are likely in north Alabama, maybe as far south as Birmingham.

So far, the computer-generated MOS (model output statistics) are not picking up on this, due to their bias towards normal temperatures at days 3-7 in the forecast timeframe. This is a problem that needs to be addressed, in my opinion. Consistently, due to statistical averaging and the attempt to limit occasional large errors, MOS forecasts for weather events are too conservative. MOS numbers are based on a mixture of the actual model output and normal temperatures, but the influence of “normal” temperatures gets higher as the forecast gets farther out.

An example is this weekend’s cold outbreak. Here are the GFS MOS high/low temperatures for Birmingham for Monday-Sunday, on the 00 UTC model run:

Mon 71/62
Tue 77/66
Wed 74/62
Thu 73/56
Fri 65/47
Sat 63/43
Sun 69/47

This is despite all models showing cold air pouring south. I understand the need for some component of “normal” temperatures in the MOS, since models sometimes go haywire and predict snow, etc. and change from one run to the next. However, there should be some level of model consistency (one run to the next and between different models) considered here, since all models show the cold air. I’d guess Friday and Saturday’s high temperatures will be 10 degrees colder (or more) than the MOS numbers.

A drastic example of this failure of MOS was during the heat wave in 2007. Below are MOS predicted high temperatures for Birmingham on August 8 for the following week, and the actual high temperature at Birmingham.

8/09 99/103
8/10 98/102
8/11 97/102
8/12 99/102
8/13 94/103
8/14 95/104
8/15 93/105

You can see how the MOS predicted temperatures returning closer to normal as time went on (back to 93 by the 15th), when it actually got hotter. Sometimes, we have to look past the MOS output and look at the real weather, and what the computer models, not climatology-biased derivatives of them, are saying.

Category: Uncategorized