Evaluation of three numerical weather prediction models for the Weddell Sea region for the Austral winter 2013

Abstract

It is widely recognized that numerical weather prediction (NWP) results for the Antarctic are relatively poor compared to the mid-latitudes. In this study, we evaluate output from three operational NWP systems: the ECMWF, Global Forecast System (GFS) and Antarctic Mesoscale Prediction System (AMPS), for the Austral winter (June-August) of 2013 for the Weddell Sea region, paying special attention to regional patterns of error statistics. This is the first evaluation of NWP systems over the Southern Ocean that also addresses the accuracy of forecasted vertical profiles. In the evaluation, we use data from land- and ship-based automatic weather stations (AWS) and radiosoundings. While the ECMWF and AMPS forecasts are on average biased cold and dry near the surface, the GFS forecasts are on average biased warm and moist. The near-surface wind speed is on average overestimated by the AMPS forecasts, whereas it is slightly underestimated by the forecasts of the other two NWP systems. Among the variables investigated, all three NWP systems forecast the near-surface specific humidity most accurately, followed by the temperature and then the wind speed. The forecast quality for the near-surface and upper-air wind speed degrades the most rapidly with increasing lead time, compared to the other variables. ECMWF is the overall best NWP system when compared against both the near-surface and upper-air observations, followed by AMPS and then GFS. The generally poorest model performance is found in locations with complex terrain along the coast of the Antarctic continent, and the best over the ocean.