Figure 7. Precipitation PDF for 13-km SHiELD (orange) compared
to contemporary GFS (blue). Top: Tropical (30S–30N) precipitation vs.
TRMM (black). Bottom: CONUS precipitation vs. StageIV (black).
Another sensible weather metric is 2-m temperature, which has an
interesting development history (Figure 8). The initial 2016 version of
SHiELD had a very small warm bias, significantly less than the small
(0.3 K) warm bias of the operational GFS. The 2018 version of SHiELD,
which otherwise had significant improvements in other skill metrics,
developed a cool bias which increased to 0.6 K by day 10. Investigation
traced the cool bias to two sources: the switch from the hybrid EDMF PBL
to YSU, which by default has significantly less near-surface mixing and
thereby allows the surface to cool too much, and the change in how cloud
droplets absorb radiation when the Inline GFDL Microphysics was
introduced. In 2019 the cloud-radiation interactions were significantly
revised, and the background diffusion in the YSU PBL was increased,
which significantly reduced both the cold bias and the error in 2-m
temperature. The cold bias in SHiELD 2019 ranges from 0.1 K on the first
day to 0.35 K on day 10, which is approximately equal to the positive
bias of the operational GFS.