Figure 8 . Global 2-m temperature (deg K) bias (top) and RMSE (bottom) for 13-km SHiELD (orange) compared to contemporary GFS (blue), both validated against ERA5 Reanalysis (Hersbach et al. 2020).
3.2 T-SHiELD North Atlantic Nest for Tropical Cyclone Prediction
T-SHiELD uses the variable-resolution capabilities of FV3 to replicate the tropical cyclone track skill of global models and the intensity skill of convective-scale regional hurricane models. This configuration uses the 13-km SHiELD grid and then places a large factor-of-four two-way nest over the tropical North Atlantic (Figure 1). The resulting nested domain has grid cells of about 3-km width and interacts with its parent global domain. Earlier experiments and a comprehensive evaluation of T-SHiELD 2017 were described in Hazelton et al. (2018a, 2018b). T-SHiELD has been used as the initial prototype for the Hurricane Analysis and Forecast System (HAFS; Hazelton et al., 2020). Here we will describe further evolution of T-SHiELD, including progress towards rectifying two forecast issues in T-SHiELD 2017: an under-intensification bias for rapidly intensifying storms, and storms with a radius of maximum winds (RMW) that is too large. Note that there is no 2019 version of T-SHiELD.
Hazelton et al. (2018b) found that the RMW in T-SHiELD 2017 was often larger than observed and in particular larger than that in HWRF simulations from the same set of cases. Zhang et al. (2015) found that reducing the parameterized mixing in the PBL scheme reduced the size of the RMW in HWRF. While reducing the parameterized mixing in the hybrid EDMF scheme gave modest improvement to hurricane structure in T-SHiELD, there was no appreciable reduction in the size of the eyewall. A dramatic and immediate impact was instead found by using the positive-definite (PD) advection scheme for water vapor and microphysical tracers. Results from T-SHiELD 2018 simulations of Major Hurricane Irma, initialized prior to its rapid intensification, show that a simulation using the older monotonic advection scheme (Figure 9) produces a gradually expanding vortex that does not intensify. Meanwhile, the simulation with the new PD scheme and no other changes to the physics or dynamics, including advection of dynamical quantities, produces an intensifying storm with a contracting eyewall. Notably, the vertical velocity within the eyewall is much more coherent with the PD scheme and is continually displaced within the eyewall, which we suspect may be driving both the intensification of Irma and a continued contraction of the eye, as well as contributing to enhanced precipitation within the eyewall. For this reason, the positive-definite advection scheme was selected for T-SHiELD 2018.