Evaluating Global Forecast System (GFS) Model Biases Relative to Upgrade Implementations

This project presents an evaluation of changes in biases of the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS). Numerical weather forecast models are often upgraded by the entities which maintain and develop them. Determining how these upgrades to numerical weather models impact the accuracy of the model’s forecasts is central to understanding how to further improve modeling efforts, especially as models become more complex with many coupled systems. Changes to either the model physics or model resolution made in an upgrade, results in varied changes to the biases of different meteorological fields.

Focusing on the Continental U.S. (CONUS), the two meter air temperature and dew point temperature biases were analyzed for the various CONUS subregions to determine which regions had the largest fingerprint of the upgrades implemented over the last ten years. Overall, the forecasts have been improved after each upgrade particularly the cold two meter temperature bias. However, the warm two meter temperature bias became worse over a few regions. Two regions were identified to be clearly impacted by upgrades during this time period. The cold two meter temperature bias was significantly reduced in the Northwest CONUS (NWC). But evident increases to two meter air temperature warm bias in the Southern Plains(SPL), especially during the nighttime and early morning hours, were identified following two major upgrades during the last ten years.

The warm nighttime two meter bias is an outstanding issue of the GFS. Therefore, the GFS model output was compared to observational flux tower station data from stations in this region to further characterize the physics and resolution changes of the upgrades with the largest influence on the two meter air temperature bias. Specifically, differences in latent heat flux and soil moisture of the GFS model data and the station data were studied to better understand the relationship between evaporation and the observed increase of the GFS warm bias in the Southern Plains region following the upgrades studied. Comparison of GFS data with the station data revealed that bais changes were due to a number of reasons depending on the station’s location. Contributors to the observed bias increase include changes to land processes such as the latent heat flux and evaporation, large local precipitation events, and potential discrepancies in the surface radiation budget.