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

scatterplot shows that ERA5 has better skill (Fig 7). The higher accuracy of ERA5 is confirmed by its aggregate RMSE (0.6 hPa), which is less than half of that of 24d18h (Table 2). The RMSE of ERA5 remains relatively constant and less than one during the whole lifetime of the PL (Fig 8B). Like 24d18h, ERA5 shows a very strong positive correlation with the observations, but its correlation coefficient is slightly higher than that of 24d18h (Table 2, Fig 8C). 3.3.2 Temperature at 2 m. All the datasets have a positive temperature bias (Fig 9). The aggregate statistics indicate that 24d12h and 24d18h have higher mean bias and lower accu- racy, but higher correlation coefficient, than the other simulations (Table 3). These two simu- lations have a higher ME (2˚C) and a slightly higher RMSE (2.4˚C) compared to the simulations that did not simulate the PL (ME between 0.8 and 1.3˚C, and RMSE between 1.9 and 2.1˚C). The simulation that captured a small and short-lived low, 23d12h, has a ME (1.7˚C) and a RMSE (2.3˚C) lower than those of 24d12h and 24d18h, but higher than those of the other simulations. All simulations show a quite strong correlation with the observations. Although 24d12h and 24d18h show the highest Spearman correlation coefficient (0.78), the difference between the simulations is very small. In the initial stage of the PL, all simulations show a high positive ME, with 24d12h and 24d18h showing the highest bias (Fig 10A). This could indicate that the MCAO is not well sim- ulated. However, there are only two observations available at 0100 and one at 0200 UTC, so

the results should be interpreted with care. From 1000 UTC until the dissipation of the PL, the ME of 24d12h and 24d18h remains notably higher than that of the simulations that have not captured the PL. The ME of 23d12h remains lower than that of 24d12h and 24d18h, but higher than that of the other simulations virtually all the time. The important surface heat transfer from the ocean to the atmosphere that takes place in the simulations that have captured the PL (not shown) likely explains why these simulations have higher ME. The RMSE of 23d12h, 24d12h and 24d18h remains higher than that of the other simulations most of the time (not shown). The correlation coefficients of all simulations notably increase from around 1100 UTC until around 1500 UTC, and then the simulations show a strong or a quite strong correla- tion with the observations (not shown). Surprisingly, the aggregate ME of ERA5 (0.9˚C) is closer to that of the simulations that have not captured the PL than to that of the simulations that have captured it (Table 3). The ME of ERA5 remains significantly lower than that of 24d12h and 24d18h, and closer to that of the simulations that have not captured the PL, during the whole observed lifetime of the PL (Fig 10A). With an aggregate RMSE of 1.5˚C, ERA5 has slightly higher accuracy than all the simulations (Table 3). The RMSE of ERA5 remains lower than that of the simulations during virtually the whole observed lifetime of the PL, and the difference in accuracy between ERA5

Table 2. Aggregate statistics computed to verify the simulated and ERA5 SLP against the observations from sur- face stations and drifting buoys. ME [hPa] RMSE [hPa] r 23d00h 3 3.8 0.54 23d06h 3.5 4.5 0.48 23d12h 2.8 3.6 0.71 23d18h 3.2 3.9 0.67 24d00h 2.1 3.2 0.59 24d06h 2.5 3.6 0.53 24d12h -0.2 2 0.81 24d18h -0.2 1.4 0.92 ERA5 0.1 0.6 0.98 https://doi.org/10.1371/journal.pone.0292250.t002 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 13 / 28