A Research on the Applicability of Spatial Regression Test in Meteorological Datasets

With the rapid growth of the AWS spatial distribution density, it is more rational to use spatial consistency checks in quality control of meteorological observations of some newly built stations and single element observing stations (i.e. the automatic rainfall stations densely covered all over our country) . For that under these situations traditional historical comparative study and different elements comparing are difficult to do for lacking of data.A new spatial regression checking method used abroad is introduced in detail and applied to spatial checking of some basic surface observing meteorological elements of China for the year of 2003 in order to evaluate the applicability of this approach in China. The method is designed for identification of suspected observing values among neighboring observations. First, some neighboring stations are selected by distance. Second, the root mean square (RMS) errors of the univariate regression equations which are established basing on the examined station observation and neighboring observations are calculated and five reference stations are determined by minimizing root mean square errors. The five reference stations are weighted differentially. Stations with smaller RMS errors get more weighting points. Then, the weighting estimate values and their weighting standard errors of the examined station are computed and used to determine the data range. Data not in this range would be flagged suspected.The spatial checking tests are conducted on 7 basic meteorological elements including daily mean temperature, maximum temperature, minimum temperature, mean vapor pressure, mean wind speed, mean surface temperature and precipitation. The data are obtained from 671 weather stations all over China, and to get more reasonable results the data are divided into 10 districts according to their station designator.Results show that this method works well in identifying errors of single meteorological element especially to the elements with larger spatial variation such as precipitation and wind speed. It should be noticed that there are some differences when applied this method to different areas and different elements. To get a fixed error rate, the values of f should be selected according to the corresponding districts. Finally, same as other spatial checking methods, geographical environment and distribution of neighboring weather stations should be concerned necessarily as influence factors. The approach performs poorer under the condition of sparse station density.