SCM key findings
Key finding #1
• Given identical forcing, the GFS-GF suite produces weaker convective tendencies and convective transport than GFS-SAS. This alters the relationship among the physics schemes within the suite, leading to the explicit microphysics scheme in GFS-GF to have a greater relative response to the forcing.
Key finding #2
• Use of the GFS-GF suite reduces the dry bias in the boundary layer and generally produces a higher cloud fraction during the deep convective period compared to GFS-SAS for this case.
Key finding #3
• For the suppressed convection phase of the case, the GFS-GF suite produces an elevated temperature inversion and associated steep gradients in water vapor, leading to spurious cloud generation near the boundary layer top.
Key finding #4
• The GFS-GF suite produces a much lower convective precipitation ratio compared to the GFS-SAS suite.
Key finding #5
• During the deep convective period, the forcing ensemble elicits greater variability from the GFS-GF suite than the GFS-SAS suite.
Global model key findings
Key finding #1
• The results of RMSE and bias comparisons vary by forecast lead time, level, and region, with GFS-SAS displaying superior forecasts in more instances than GFS-GF. In upper levels, there are more differences between GFS-SAS and GFS-GF in temperature and relative humidity RMSE than in wind RMSE. When statistically significant pairwise differences were noted for wind speed RMSE, they nearly always favored GFS-SAS, regardless of level or region.
Key finding #2
• Upper-air temperature and relative humidity RMSE values are generally larger for GFS-GF than GFS-SAS but the favored configuration depends on forecast lead time and vertical level. The advantage of GFS-SAS over GFS-GF is larger and more frequent earlier in the forecast; as forecast lead time progresses, the gap in performance narrows and GFS-GF is superior to GFS-SAS for some levels, lead times, and regions. This suggests that the GF scheme may not be in balance with the initial conditions used in this test (operational GFS analyses), and that the GF might perform better in a cycled experiment.
Key finding #3
• A pronounced diurnal cycle in 2-m temperature bias is clear for both the GFS-GF and the GFS-SAS configurations. The GFS-SAS gets progressively warmer through the forecast period over CONUS throughout the troposphere and at the surface, and gets colder in the tropics. The diurnal GFS-GF bias amplitude grows with forecast lead time.
Key finding #4
• In extratropical regions, precipitation frequency biases are overall similar between the model configurations, with over-precipitation for low thresholds and under-precipitation for high thresholds. However, the diurnal cycle of errors over the continental US are distinct between the configurations.
Key finding #5
• Overall, GFS-SAS is more skillful at predicting precipitation.
Key finding #6
• The partition of precipitation (convective and explicit) is different between the configurations, with SAS producing more total convective precipitation than GF.
Key finding #7
• Tropical Cyclone track errors averaged over the Atlantic, Eastern North Pacific, and Western North Pacific basins are similar for both model configurations. While accuracy in TC intensity forecasts is not expected of a model run at this coarse resolution, it is of interest to note that storms in GFS-SAS are more intense and have less absolute intensity error than those in GFS-GF.
Key finding #8
• While verification of cyclogenesis is beyond the scope of this report, it is noticeable that the models have different behaviors, with GFS-GF producing more storms.
