Comparative Analysis Of Atmospheric Temperature Between Three Reanalysis Datasets And Radiosonde Dataset In China

Based on radiosonde dataset from105stations in China, NCEP/NCAR reanalysis dataset, ERA reanalysis dataset, JRA reanalysis dataset in1980-2008, a study is performed of reliability of atmospheric temperature of reanalysis datasets based on the annual and monthly mean in China by means of many kinds of statistical analysis technique. At the same time, we select the representative station in area I (northeast) and V (northwest) in China to study based on the divided method of REOF. The main conclusions are as follows:(1) In numerical aspects, three reanalysis datasets are lower than the radiosonde dataset, NCEP dataset is much closer to observed dataset in upper troposphere while ERA and JRA datasets are closer in middle and lower troposphere. So far as seasonal mean, the reliability in autumn is much more than other seasons.(2) In terms of annual and interdecadal trend of atmospheric temperature, ERA dataset has higher ability to reproduce those trend in the north region of upper troposphere while NCEP dataset is better in the south region of upper troposphere. Three reanalysis datasets have comparable ability to show the trend in middle and lower troposphere.(3) In the aspect of spatial and temporal variation characteristics, So far as annual mean, the interdecadal variation character of radiosonde dataset reveals the opposite phase change between middle and low troposphere and upper troposphere while the annual variation character displays the feature of the opposite phase change between north and south of the whole layer temperature. NCEP and ERA datasets can preferably perform the feature of interdecadal variance while ERA and JRA datasets can reveal the characteristic of annual variance well. So far as seasonal mean, there are quite differences of three reanalysis datasets in the distribution of the first mode while the second mode correlates quite the same with each other in four seasons.(4) After redivided by the method of REOF, we can use the station of Haerbin and Korla as the representative station in area I (northeast) and V (northwest), to find out the variation character of three reanalysis datasets and radiosonde dataset in these areas. In the aspect of climate mean, the root-mean-square error is much lower in middle and low troposphere then in upper troposphere; Compared to radiosonde dataset, the root-mean-square error of JRA datasets is low while NCEP datasets is high; The root-mean-square error in winter is low while it is high in summer; The root-mean-square error in area V is higher than area I. From the perspective of annual variance, as for area I, the quality of three reanalysis dataset in low levels is obviously better then in high levels, and more reliable in winter and spring than in summer. The reliability of JRA dataset is much better than other datasets. As for area V, when compared to area I, the difference between three reanalysis dataset and radiosonde dataset is much bigger. There are not apparent variation in the difference between three reanalysis dataset and radiosonde dataset, and the other character is quite the same as area I. In the aspect of long-term trend, three reanalysis dataset both show the better conditions in the low troposphere. As for area I, NCEP dataset can reveals the cooling trend of radiosonde dataset better when above300hPa. As for area V, JRA dataset can reveals the cooling trend of radiosonde dataset better when above200hPa.