Evaluation on the Random Error of Second Level Sounding Data

With the development of science and technology, the performance of sounding system, including the data acquisition rate, accuracy, reliability and automation are improved significantly. Comparison and statistical methods to estimate various errors are also needed to be improved. Relative system error and random error are concern variables of the sounding information users, errors evaluated by the reasonable method can reflect typical characteristics of error to same extent. So far, there have not a satisfactory standard radiosonde developed as a reference, relative system error and random error are obtained only through direct intercomparison simultaneously. The random error, it is not determined by dual-launching the same type of radiosondes because of the heavy workload. It is mainly used the indirect estimation method, that is the random error of the specify instrument used as a reference, and then the random error of unknown radiosonde is isolated from the variance between reference and unknown radiosondes. But whether the indirect calculation method of random error is suitable for the second level sounding data or not, the further discussion should be adopted. An overview of the random error is explained including the definition and determination method. And then two datasets are used to analyze the effect on the random error by different degree of data smoothing. One is the data of domestic GPS radiosondes comparison experiments in June 2007 and June-July 2008, the other dataset is the 8th WMO radiosonde comparison at Yangjiang China in 2010. The intercomparison analysis shows that the indirect calculation method of random error could not fully be applicable to the second level sounding data, especially for the estimation of random error of wind, temperature in stratospheric and relative humidity in tropospheric. The second level sounding data can detect the more detail caused by the swing of rising balloon, the raw data should be smoothed to reduce the impact of the above. If smoothing degrees of the original data compared are consistent, the indirect calculation method of random errors could be used suitably. The deviation is small, conversely, it might be problematical, which will produce large bias if it exists the difference in smoothing degree of the original data. In the scheme of direct intercomparision, in order to obtain the relative system error and random error of the different types of radiosonde systems, it is best to hang more than one of the same types of radiosonde in the same balloon to contrast synchronously, which can reduce the influence on evaluating the unknown radiosonde random error because of the own error of reference instrument. The more radiosondes of the same type are used, the more valid data could be obtained, the more accurate evaluation of random errors could be obtained.