Evaluation Of Polar WRF Simulations Of Antarctic Atmospheric Circulation

In recent years there has been considerable growth in requirement for accurate numericalweather forecasts for the Antarctic because of our increases in complex scientific researchactivities and the rapid growth of tourism to the continent. Therefore, we evaluate a completeannual cycle of nonhydrostatic mesoscale model simulations of the Antarctic atmosphericcirculation. The year-long time series are compiled from a series of overlapping short-duration(48h) simulations of the atmospheric state with the first24h being discarded for modelspin-up of the physical and boundary layer processes, and the24-48h periods used for modelevaluation. The simulations are generated with the Polar WRF version3.2.1, with a horizontalresolution of50km. The model has been run for the period of1st January-31st September2008, initialed by ECMWF ERA-Interim reanalysis and NCEP GFS-FNL analysis,respectively. The model output is compared with near surface meteorological observations (12sites) and the upper-air GPS sounding observations (5sites) over timescales from diurnal toannual. The impact of uncertainty in the initial and boundary conditions, large-scalecirculation seasonal variability, sea ice characteristics, horizontal resolution and differentparameterization schemes for cumulus, microphysics, planetary boundary layer (PBL) andradiation are discussed. The results show that:1. The model skill varies seasonally reflecting the different roles played by local andsynoptic scale systems at different times of the year, but it has high forecast skill for capturingthe whole year circulation features of the near surface pressure, air temperature, specifichumidity and wind speed, as well as the upper-air geopotential heights, temperatures, windspeeds and wind directions, especially for capturing the synoptic scale variability of nearsurface and upper-air meteorological elements during austral summer and winter, e.g. thevariations of high and low pressure processes, the tropospheric suddenly warmer or cooling ofwinter temperature in a short period of time and strong winds. For the near surface, themonthly differences between the simulations and observations are usually less than2hPa forsurface pressure,<2°C for air temperature over the interior regions, and <2m/s for the nearsurface wind speed. The model exhibits a cold bias (<2°C) at the coastal zones throughout theyear and a slight warmer winter (<0.7°C) at the interior areas. For the upper-air, the biasesbetween simulation and observations are usually±20gpm for geopotentail height,<1°C fortemperature,<1m/s for wind speed and±5°for wind direction throughout the entire atmospheric column. The temperature (wind speed) absolutely biases are c.2°C (c.2m/s)only at the near surface and the tropopause (c.300hPa). The related coefficients betweensimulations and observations are usually>0.98for surface pressure,>0.85for airtemperature,>0.99for geopotential height profiles,>0.9for temperature profiles,>0.8forwind speed profiles, and0.6~0.9for wind direction profiles. In addition, Polar WRF cancapture the coastal summer diurnal cycles of each near surface meteorological elements(including katabatic winds), and even can accurately reproduce the bimodal type wind diurnalvariation at Vostok, with the extreme wind speed biases just less than0.5m/s. The highforecast skill for temperature profiles determines the model have the ability to accuratelysimulate the tropopause height and temperature and the vertical distribution of atmosphericstatic stability (Brunt-V is l frequence). The minimum monthly tropopause height biasoccurs in January with a vaule about-0.1km and the maximum bias appears in August with avalue of0.6km. The monthly mean tropopause temperautre biases are usually less than2°C.2. The model forecasting skill of near surface meteorological elements over coastalregions of East Antarctica can be improved either by reducing the sea ice thickness, or sea icealbedo in summer, or reducing them both. In addition, the effect of sea ice albedo issignificantly larger than that of its thickness, making the summer mean absolute bias of coastalair temperature decrease0.22°C. Greater influence has been found over the coastal regions aschanging the sea ice characteristics, especially in summer when the sea ice albedo decreases,its impact on the surface pressure and air temperature over the coastal zones was nearly as5times large as that on the interior regions. The albedo reduction could make the unstablestratifications increasing and its influence was more significant than that from the sea icethickness reduction. The response height in summer of local atmosphere to the change of seaice albedo can be up to about880hPa, and the horizontal extent of response can reach theinterior regions about600km far away from the coast.3. The near surface meteorological conditions are very sensitive to the horizontalresolution, especially for air temperature and wind speed over the coastal regions, even on therelative flat interior Plateau areas like Dome A and South Pole are no exception. The modelshows different seasonal forecasting skills for different meteorological elements in thedifferent sites under the high horizontal resolution. For example, the summer and winter meanabsolute biases of air temperature decrease1.20°C and0.84°C under16.7-km grid than thatunder50-km coarse resolution, respectively, but the surface pressure (wind speed) biasesrespectively increase1.10hPa (1.42m/s) and0.65hPa (1.99m/s). It shows that it is not thehigher horizontal resolution the more accurate simulation.4. For the cumulus parameterization schemes, Betts-Miller-Janjic scheme shows the bestresults both over the coastal and interior regions. Kain-Fritsch scheme presents the best forecasting skill for the summer synoptic processes in coastal regions and New Grell schemegives the best results in interior regions during austral summer. For the microphysicsparameterization schemes, Thompson scheme shows the best results in coastal and interiorregions both in summer and winter. The performance of Morrison scheme in coastal area isnearly same as Thompson scheme, but it is not suitable for the simulation in interior regions,especially for the winter air temperature. The PBL parameterization sensitivity tests show thatACM2scheme is best for the simulation in coastal and interior regions both in summer andwinter. YSU scheme is not suitable for the simulation of Antarctic winter temperatures. For theradiation parameterization schemes, RRTM long-wave and CAM (or RRTMG) short wavecombination radiation scheme shows the best performance in the coastal regions duringsummer and winter; RRTMG long-wave and RRTMG short wave combination radiationscheme has a high forecasting skill for the coastal atmospheric processes in summer, andRRTM long-wave radiation scheme shows a slightly better forecasting skill for the interiorregions than RRTMG long-wave radiation scheme. In addition, the model’s sensitivity to thePBL and radiation parameterization schemes is obviously higher than that to the microphysicaland cumulus parameterization schemes.5. The most skillful forecasts for the near surface air temperature, specific humidity andupper-air temperature and wind speed are using the ERA-Interim reanalysis for initial andlateral boundary conditions and the highest forecasting skill for the surface pressure andupper-air geopotential height are using the GFS-FNL analysis for initial and lateral boundaryconditions. The performance of the simulated near surface air temperature, specific humidityand surface pressure in coastal regions are similar as the performance of the initial fieldsERA-Interim and GFS-FNL. In addition, ERA-Interim and GFS-FNL respectively as theinitial field the forecasting skill for other near surface parameters and upper-air elements showthe seasonal varies in different regions, but the biases between them are small and obviouslysmaller than the differences between ERA-Interim and GFS-FNL. This shows that the errorsof initial and lateral boundary conditions would in certain extent directly influence theinaccuracy of the simulations. On the while, Polar WRF model has an ability to reduce thedifferences between different initial fields through adjusting itself (i.e. dynamical core andphysical processes). ERA-Interim reanalysis and GFS-FNL analysis both are good choice forthe model’s initial and lateral boundary conditions over Antarctic region.The results of this study could provide some references for the improvement anddevelopment of Polar WRF of Antarctic simulation and lay a good foundation for developingour own Antarctic numerical weather prediction system and for further research.