IMPLICATIONS OF COMPLEX UPPER AIR HISTORIES ON APPROACHES TO ADJUST ARCHIVED RADIOSONDE DATA FOR INSTRUMENT DISCONTINUITIES

摘要

Radiosondes provide the longest time series of meteorological measurements above the surface, but discontinuities from frequent instrument changes at all stations contaminate atmospheric climate trends. As sensors have become more sensitive and better protected from radiative errors over time, the erroneous trend is hypothesized to be in the direction of artificial cooling and drying (although individual discontinuities can differ). Metadata describing upper air stations, instruments, and dates of changes, is incomplete and sometimes inaccurate, Trends adjusted for instrument discontinuities are questioned because researchers make adjustments while many instrument transitions and their dates are not known. This study is part of a longer-term project to develop complete radiosonde metadata and instrument adjustments based on the completed station instrument histories (Schroeder 2009). The main finding from histories developed so far is that many station instrument histories are very complex, with dozens of transitions including many alternations between 2 to 8 instrument types, and few long periods with the same instrument type. The complex station histories have implications for data adjustment methods. First, frequent alternations of instruments blur the discontinuity from a transition, sometimes for up to several years. Second, some adjustment methods are not designed to handle transitions less than about 2 years apart. Third, many of the instrument discontinuities are real but small and are not likely to be detected by automated discontinuity detection methods, but still have a noticeable cumulative effect on trends. Finally, the usual adjustment method, which equalizes the mean value of a variable for a specified averaging period before and after a discontinuity (so it can be called the "segment mean matching" method), inevitably removes a portion of the real trend, so when sta-tions have many discontinuities, the adjustments can remove a large percentage of the real trend. This study focuses on the last issue because the removal of some of the actual trend is not well publicized in accounts of upper air data adjustment projects. Any time series with artificial discontinuities for which overlapping data is not available is subject to the same type of unintentional trend removal. This paper estimates the theoretical proportion of a long-term trend that is likely to be removed by a "segment mean matching" adjustment method and shows some simple illustrations.