Operational weather forecasting in Antarctica with the WRF model in sub-kilometer resolution

The coastal regions of Antarctica are subject to frequent passages of low-pressure systems and transient pressure ridges. This synoptic-scale setup favors abrupt weather changes and occurrence of dangerous outdoor conditions such as snowstorms, extremely low wind-chill values due to low air temperature and strong winds, or intense solar radiation potentially leading to snow blindness. In this coastal zone most Antarctic research stations are located, and numerous scientific and logistical field activities are conducted. Under these circumstances, operational weather forecasts are of utmost importance. In this contribution, a WRF-based high-resolution weather prediction system is presented and evaluated. This experimental system was run to support the 2023 and 2024 summer expeditions to the J. G. Mendel Station on James Ross Island, Antarctic Peninsula. Initial and boundary conditions were provided by the GFS model, model topography by the Reference Elevation Model for Antarctica. The model configuration included the 3DTKE boundary layer scheme suitable for sub-kilometer resolutions, Thompson microphysics, RRTMG longwave and shortwave radiation schemes and the NoahMP land surface model. The model was run once a day in 500-m horizontal resolution (132-h lead time) and 1.5-km horizontal resolution (96-h lead time, more recent initial conditions). Forecasted time series of air temperature, wind speed and direction, precipitation amount, snow height, global radiation and sea-level pressure for multiple field locations were sent to the Mendel Station via a satellite internet service. The WRF model forecasts were validated with in-situ observations at the coastal Mendel Station (10 m a.s.l.) and the top of Davies Dome glacier (514 m a.s.l.). Furthermore, the model accuracy was compared with the output of publicly available Antarctic Mesoscale Prediction System (AMPS). Compared to AMPS, the WRF model in 500-m resolution massively improved air temperature prediction at Mendel Station, reducing mean bias from -4.2 °C to -0.8 °C in 2023. In late 2023, multiple AMPS physical parameterizations were updated, possibly contributing to reduced bias of -2.9 °C in the 2024 season. However, the WRF model still performed significantly better with bias of ‑0.5 °C. On Davies Dome, the WRF model performed slightly better in 2023 (by 0.4°C) while in 2024 the models performed similarly well.  Regarding wind speed, both the WRF model and AMPS provided comparable results with mean bias 1.5 - 1.9 m·s^‑1 at Mendel Station (more favourable for WRF) and 0.1 - 0.7 m·s^‑1 on Davies Dome (more favourable for AMPS). Prediction of two significant snowfall events in 2023 was done with a very good accuracy.