25 Degree MOS Temperature Errors in the US Chinook Belt

This study examined extremely large temperature forecast busts by NCEP Model Output Statistics (MOS) in the part of the United States most affected by chinook winds and cold surges: the lee of the Rockies or its downstream island ranges. From a baseline of AVN/GFS MOS at Great Falls, MT, it was found that a threshold of 25 degrees Fahrenheit resulted in 33 events in the past 16 years (2002-2017). This same threshold was then used to identify cases in ETA/NAM MOS and for both models in Rapid City, SD and Denver, CO. AVN/GFS events decreased sharply (to 16) at Rapid City and slightly more (to 10) at Denver. At all stations, ETA/NAM cases exceeded AVN/GFS by 14-15.

There is also a major decrease with time early in the study period. 35-50% of the busts occur in the three years ending in 2004, which is when a mountain blocking parameterization was introduced in the global model. Subsequent to this frequencies have varied with multi-year periods of 0-2 events total separated by consecutive years with 1-4 events each. This appears to be due to a combination of changes in model configurations and seasonal variations in winter temperatures.

Initial examinations of each case show that the simulation of a late cold surge passage is a common cause of extreme busts. While the frequencies has decreased since 2004, they still average about 25% for all stations and models. Denver has the highest proportions including three of the six recent GFS busts.

Another important cause is overprediction of temperatures due to the mishandling of chinook winds. This can be due to early onset of a sustained event, spreading an isolated event of a few hours to earlier and later times, and missing brief foehn pauses at night. Collectively, this "overchinooking" represents 40-50% of the model busts at Great Falls and still accounts for 15-25% of events at Denver and Rapid City. The proportion is consistently higher in the AVN/GFS MOS compared with the ETA/NAM.

Other factors which generally appear three or fewer times in each dataset are lower observed temperatures due to thunderstorm activity, discrepancies in cloud/precipitation amounts, unique cases requiring further study, and errors with cold surge and chinooks resulting in temperature underprediction. The latter represent 10-25% of busts at Rapid City, but are much rarer in Great Falls and Denver. Furthermore, out of only six events, four occur in the NAM in 2007 and 2008.

Further work will provide more statistical analysis (such as the length of time that large errors remain in the forecasts) and increased areal coverage by adding Billings, MT, Cheyenne, WY, and Colorado Springs, CO.