Research On Applicability Of Climate Dynamical Downscaling Skill Using WRF Model

In order to test the downscaling ability of WRF model runs in China, national-wide and regional climate simulation tests were made, driven by the NCEP reanalysis data. Differences of simulation results caused by various re-initialization frequency and their influences on system errors due to long integration time has been examined by designing a variety of comparison tests (sensitive test). The tests were compared with one by adding an analysis nudging method in WRF based on traditional climate simulation ways(the control test). The Micaps data(temperature,wind speed,sea level pressure and geopotential height datasets are chosen) and the gradient data (temperature and relative humidity data at2m above surface,the surface sensible heat flux and latent heat flux dada and also the wind speed data at10m(by interpolation) are chosen) from Semi-Arid Climate and Environment Observatory of Lanzhou University(SACOL) were used for the validation.(1) In the simulations over China, the conclusions are as below:a. By analyzing all of4simulation results, the results show that no matter "re-initialization"or "analysis nudging" skill can result some kind of reducing of system errors caused by long time integration, there are different performances for different variables as to these4simulation cases, b. By comparing the RMSEs for "fnl"(interpolated NCEP reanalysis data)、"year"(controlling experiment)、"10days"(representing "re-initialization"skill) and "nudging"(representing "analysis nudging" skill) their downscaling skills were verified, and the results show that "fnl" has the larger errors compared with WRF simulations and "10days" is better than "nudging" for surface temperature and wind speed;"nudging" has special advantages on mass field simulation;"fnl" shows minimum RMSE on geopotential height field of500hpa.(2) The "fnl" data has better result in upper air and "nudging" takes second place, but "year" is the worst case; WRF simulations has smaller RMSEs in low level than "fnl".(3) For the single grid/station simulation tests on semi-arid area in the northwest,the two-way nested case show advantages on all examined variables compared to the one-way nested case. And in the comparison tests of two-way nested run of single station, the results show as below:a. These four kind of tests have shown consistent variation trend with observations of temperature at2m above ground but all results have been underestimated; They have very close and good simulation results in all months except July, and "nudging"shows advantage in July; it is clear that analysis with nudging can decrease the system errors caused by improper parameterization scheme selection.b. Different from temperature, the simulations by adding the updating of initial conditions has been obviously improved on relative humidity simulations at2m; The annual RMSEs compared with observation data are7.2%,7.6%and8.5%respectively for "10days","month"and "year".c. It is not as good as simulations on temperature and humidity. The simulations for surface wind speed have shown good results from January to June than those from July to September(summer); Re-initializations has shown no obvious difference compared to control experiment;"nudging" shows worst simulations for more flat annual varying trend and great underestimating compared with observations.d. Obvious improvement has obtained by re-initialization for surface sensible heat flux and surface latent heat flux simulations except "nudging";"Nudging" method resulted in overestimation for the surface sensible heat flux but underestimation for the surface latent heat flux.