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

located in the west-north-west/north-west/north-north-west (WNW/NW/NNW) quadrants of the wind rose (not shown), which correspond to the direction of the wind responsible for the MCAO. The main direction of the wind in 24d12h and 24d18h is NW, but these simula- tions also show winds coming from a wide range of directions, like the observed winds. How- ever, the number of observations of the wind direction in the NW quadrant is much less compared to the simulations and ERA5, which is likely due to the fact that many wind observa- tions are not represented in the wind rose because the recorded wind speed is zero. The simu- lation with the lowest RMSE-VWD is 24d18h (4.9 m s-1), followed by 24d12h (5.9 m s-1) and 23d12h (6.2 m s-1) (Table 6). Overall, the RMSE-VWD of 24d12h and 24d18h increases from 1100 UTC to 1700 UTC, as the PL deepens (Fig 11A). The RMSE-VWD of 24d18h and 24d12h is lower than that of the other simulations during, respectively, almost all of the time and half of the time. The wind rose of ERA5 is quite similar to that of 24d12h and 24d18h, the main difference being that ERA5 represents more frequent winds from the north-north-east (NNE) (not shown). ERA5 has the lowest RMSE-VWD (3.7 m s-1), even when compared with the simula- tions that have captured the PL (Table 6). The time series of the RMSE-VWD of ERA5 is simi- lar to that of 24d12h and 24d18h, with the difference that the RMSE-VWD of ERA5 is lower throughout the whole period (Fig 11A). At the end of the lifetime of the PL, the RMSE-VWD of all the simulations and ERA5 tends to converge to around 5 m s-1.

Given that the simulation and ERA5 capture the observed SLP quite well, it is surprising that the skill of both at capturing the near-surface wind is not as good. This is likely due not only to model error, but also to representativeness error and observational error. In complex terrain, wind observations from weather stations may not be representative of the average wind over a larger area. In addition, measurements by anemometers are affected by topogra- phy, surface cover and surrounding obstacles [49]. The differences between the observed and simulated winds may be also due to the different period used to obtain the average wind in the different datasets. 3.3.6 Maximum wind gusts at 10 m. The ME of 24d12h and 24d18h is 1 m s-1, and that of 23d12h is 0.1 m s-1, whereas the ME of the simulations that have not captured the PL is neg- ative or equal to zero (Table 7). The simulation 24d18h shows the lowest RMSE (4 m s-1), fol- lowed by 24d12h (4.9 m s-1). The other simulations have lower accuracy, with their RMSE ranging from 5.1 to 7.5 m s-1. The simulations 24d12h and 24d18h have a quite strong correla- tion and strong correlation, respectively, with the observations, whereas all the others except for 24d00h have modest or virtually no correlation with the observations. The ME of the simu- lations that have captured the PL remains positive most of the time, whereas the ME of most of Table 6. Aggregate RMSE-WVD computed to verify the simulated and ERA5 10-m wind against the observations from surface stations. RMSE-WVD [m s-1] 23d00h 7.1 23d06h 7.8 23d12h 6.2

23d18h 6.3 24d00h 6.6 24d06h 6.8 24d12h 5.9 24d18h 4.9 ERA5 3.7 https://doi.org/10.1371/journal.pone.0292250.t006 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 18 / 28