Assessment of simulations of a polar low with the Canadian Regional Climate Model

has shown that 24d18h has more skill than 24d12h at reproducing most of the near-surface variables analysed. These results indicate that the initialisation time has an important impact on whether the model captures or not this PL, and on how well it is represented. The two latest initialized simulations show northerly cold air advection and winds that are more intense than in the other simulations, leading to baroclinic instability and, subsequently, to the genesis of the PL. Nevertheless, since the environmental conditions– strong low-level temperature gradi- ent and an upper-level through –on the 24 and 25 March are favourable for PL development, a PL can form at a later time if the low-level conditions are favourable for baroclinic instability to grow, which is what happens in 23d12h. In view of these results, it is suggested that future studies should investigate the potential to improve PL forecasts by using spectral nudging to maintain the low-level atmospheric fields and small scale features close to the driving data. Sensitivity tests should be conducted with different spectral nudging parameters and nudging horizontal wind, temperature, or both. Another key finding is that the processes involved in the development of the PL need to be improved in the model in order to decrease the mean bias of the simulations that have cap- tured it. Although all the statistics clearly show the better performance of 24d12h and 24d18h at reproducing SLP compared to the other simulations, it is notable that, for the other vari-

ables, these two simulations show similar or higher aggregate absolute mean bias. In particular, the parameterization of the surface heat fluxes in the CRCM6/GEM4 needs to be improved. In effect, the fact that 24d12h and 24d18h represent a PL deeper than the observed one, and show higher temperature mean bias compared to the other simulations and ERA5, seems to indicate that the ocean surface fluxes may be too strong. Finally, the results have shown that ERA5 has more skill than the simulations, including those that have captured the PL, at reproducing the observed PL during its mature and dissi- pation stages. Table 9 shows the added value of the CRCM6/GEM4 compared to ERA5 when considering the best simulation (24d18h). For all the near-surface variables analysed here, the model does not provide added value in terms of accuracy (based on the RMSE val- ues shown in Tables 2–8). It is surprising that the CRMC6/GEM4, a high-resolution model, does not provide added value compared to ERA5, the coarser reanalysis that drives it. There are two main reasons that could explain the fact that ERA5 has better skill than CRMC6/ GEM4. First, conventional observations are assimilated into ERA5. Second, the verification of high-resolution simulations using standard statistics has some limitations. For instance, when verifying the simulation of a PL using dropsonde observations, Stoll et al. [17] found that a fuzzy verification method showed that the regional model AROME-Arctic had higher skill at capturing extreme values at small scales than the global model ECMWF HRES,

whereas standard verification statistics were similar for both models. Finally, note that for this work we used GEM4, but a new version with improved physics parameterizations, GEM5, was recently released [51]. Therefore, an interesting course of research would be to analyse if this new version of GEM provides added value compared to ERA5 and to the cur- rent CRCM6/GEM4 simulations. Table 9. Added value of the CRCM6/GEM4 (simulation 24d18h) compared to ERA5 for the following variables: Sea level pressure (SLP), 2-m temperature (T2m), 2-m dew point temperature (Td,2m), 10-m wind (V10m), wind gusts (WG) and 1-h accumulated precipitation (PR). The added value has been computed using the aggregate RMSE (RMSE-WVD for the 10-m wind), based on values presented in Tables 2–8. The added value computation is based on the study of Di Luca et al. [52; Equa- tion 1)]. SLP T2m Td,2m RH V10m WG PR -0.8 hPa -0.9˚C -0.9˚C -2% -1.2 m s-1 -0.1 m s-1 -0.3 mm https://doi.org/10.1371/journal.pone.0292250.t009 PLOS ONE Assessment of simulations of a polar low with the canadian regional climate model PLOS ONE | https://doi.org/10.1371/journal.pone.0292250 October 5, 2023 25 / 28