Extreme GFS Temperature Forecasts During the Summer of 2022

The Global Forecast System (GFS) consistently generated forecasts of unrealistically hot 2-m temperatures during the summer of 2022 over the CONUS. These forecasts were generally most extreme at longer forecast ranges, causing much concern among model customers, stakeholders, and the broad user community. The most extreme examples were over the Plains region, but the issue was observed in other regions as well.

EMC’s Model Evaluation Group (MEG) investigated this issue with GFS forecasts, and findings will be shared in this presentation. During the evaluation of Version 16 of the GFS (which is the version of the model that was run operationally last summer), the MEG had found that GFSv16 consistently overmixed the planetary boundary layer (PBL) in an examination of retrospective cases, and statistical analysis hinted at a warm bias at extended forecast ranges during the summer retrospective periods. It is noted that GFSv16 uses observed precipitation to initialize the land states, a change from the previous model version which used model precipitation for the initialization.

In the summer 2022 cases, the model consistently dried out the top soil layer through the model integration, almost certainly contributing to the intense mixing of the PBL, resulting in high temperature and low moisture predictions at the surface. Since the drying of the soil had a cumulative effect during the model integration, it is not surprising that the hottest model temperatures were seen during the medium and extended ranges of the forecast. That said, even the shorter range GFS forecasts were characterized by an overmixed PBL, although this affected the low-level moisture profiles more than the low-level temperature profiles.

This presentation will first examine the background information from the GFSv16 pre-implementation testing and evaluation. It will then compare GFS forecasts of 2-m temperatures at short and extended forecast ranges, along with model soundings, with some comparisons made to other modeling systems which did not display the hot bias. Soil moisture forecasts will then be examined, and the presentation will conclude with plans for GFSv17 to mitigate this bias.