Assessment of the WRF model performance in air temperature and wind speed simulation over a complex Antarctic topography

The Antarctic Peninsula experiences a strong climate variability which is well reflected in glacier mass balance and state of the other cryospheric components. A better insight into the interactions of the atmosphere and the cryosphere can be obtained using numerical atmospheric models. In the presented work, the Weather Research and Forecasting (WRF) model at 700 m horizontal resolution was validated in the northern part of James Ross Island, Antarctic Peninsula. The model topography was based on the Reference Elevation Model of Antarctica. Sea ice cover was updated daily using high–resolution satellite observations. The WRF output was compared with air temperature, wind speed and wind direction observations from multiple automatic weather stations located in a complex topography and a mosaic of land cover types of Ulu Peninsula. Two periods in 2019/2020 representing contrasting seasons (summer and winter) were selected to identify possible seasonal effects on model accuracy. The three WRF boundary layer schemes (MYJ, MYNN, QNSE) were tested and the spatial and seasonal variability in their performance was evaluated. Simulated air temperatures were in very good agreement with measurements (mean bias –1.7 °C to 1.4 °C). The model was within 2 °C of observations in 47–72 % of the winter period and in 66–79 % of the summer period. An exception was a strong air temperature inversion at two winter days when the model accuracy decreased at low–altitude sites. Additional analysis of the WRF output revealed a good skill in simulating near–surface wind speed with higher correlation coefficients in winter (0.81–0.93) than in summer (0.41–0.59). Wind speed mean bias was mostly lower than 2.5 m·s^–1, but higher wind speed overestimation was found at a coastal site during the winter validation period. The model successfully captured wind direction, showing only small differences to the observed values. Finally, the model accuracy at coastal and low–altitude sites was found to be more sensitive to the strength of synoptic–scale wind than at higher–altitude sites.