Comparisons of 7-year Radiosonde Data from Two Neighboring Stations and Estimation of Random Error Variances for Four Types of Radiosondes

It has always been a challenge to assess the accuracy and precision of radiosonde instrumentation due to the lack of transfer standards for comparisons. One of ways for such assessment is to compare data collected from neighboring stations. The U.S. National Weather Service radiosonde station at Norman, OK is 25 km away from a research station located at Purcell, OK operated by the Atmospheric Radiation Measurement (ARM) program. During 1996-2002, four types of radiosonde, VIZ-B, VIZ-B2, Vaisala RS80-H and Vaisala RS90, were launched at the two stations. A total of 490 pairs of sondes were launched within a half hour and based on visual examination, sampled the same air mass. These co-incident soundings enabled four types of inter-comparisons, VIZ-B vs. RS80-H, VIZ-B2 vs. RS80-H, RS80-H vs. RS80-H, and RS80-H vs. RS90. The comparisons confirm the previous finding that the Sippican (formally VIZ) carbon hygristor fails to respond to humidity changes in the upper troposphere (UT), sometimes even in the middle troposphere. This lack of response produced significant and artificial relative humidity (RH) changes in the UT when a transition occurred and resulted in the inability of the carbon hygristor to measure RH vertical and seasonal variations. The comparisons between Vaisala RS80-H and RS90 data consistently show unexplained, significantly drier (~5% in RH) RS90 data in the UT. When RS80-H was launched at both stations, the temperatures in the middle and upper troposphere at Norman were colder than those at ARM-B6. Such a cold bias is associated with a known coding error in the post processing software in the U.S. operational radiosonde ground system. Random instrumentation error variances for four types of radiosondes, which are critical for the data assimilation, are estimated based on these paired soundings and will be presented.