Figure 10. Verification of T-SHiELD 2017 and 2018 during the 2017 Atlantic hurricane season against the Best Track Dataset: intensity (a) error and (b) bias; (c) track error; (d) 64-kt (33 m s-1, hurricane-force) radii. Units shown (kt, nautical miles) are standard for US operational prediction. In a–c the number of cases (individual storms) available at each lead time is shown in parentheses; in (d) the number in parentheses is the number of storm quadrants available for validation.
The multiple changes in the 2018 version of T-SHiELD combined to create tropical cyclones which are stronger overall (Figure 10a,b), with little to no bias towards more intense storms at all lead times. There is a minor degradation in track error in the 2018 version at longer lead times (Figure 10c). The adoption of the PD scheme and YSU PBL scheme likely created forecasts of more intense storms mitigated by the introduction of the interactive mixed-layer ocean. While the weak bias of the 2017 version was alleviated, intensity predictions were not appreciably improved except at 120-h lead time, and in fact were degraded between 36 and 72 hours after initialization. These results show once again the great challenge of improving intensity prediction. The reduction in RMWs in simulations using the PD scheme will be discussed in more detail in a forthcoming manuscript.
3.3 C-SHiELD Nest for Continental US Convection
C-SHiELD was designed to efficiently reach convective-scale resolutions in a global domain, in this case to replicate the capability of regional convective-scale models for continental convection such as the 3-km NAM Nest and the members of the High-Resolution Ensemble Forecast (HREF). C-SHiELD also is designed to extend convective-scale forecasts beyond the 18-to 60-hour ranges of existing US operational CONUS models into the medium-range timescales and beyond. The nested domain of C-SHiELD serves as a prototype for the Regional Forecast System (RFS; Carley et al. 2020) and the Rapid-Refresh Forecast System (RRFS; Alexander et al. 2020), both using the regional domain capability being developed within FV3.
The 2017 version of C-SHiELD is described in Harris et al. (2019). Modified versions of C-SHiELD with different microphysics and PBL schemes are described in Zhang et al. (2018) and Snook et al. (2019). C-SHiELD 2018 saw considerable updates as shown in Table 2; C-SHiELD 2019 added incremental updates, including re-configuration of the numerical diffusion and GFDL microphysics. We will limit our discussion to the evolution of broad forecast characteristics, but we will perform year-round validation instead of restricting the analysis to a single season. The time periods evaluated are given in Table 1. The exception is for the Surrogate Severe verification below, which is only verified for peak severe weather season of April to August of each year.