The archived radiosonde record is less useful than desired for identifying the causes of ongoing climate changes in the last few decades, or in some cases even for determining the temperature or moisture trend, because of frequent instrument changes at all stations. In general, newer radiosonde models are more sensitive and are assumed to be better protected from radiative errors than early models, so on the average an artificial cooling and drying trend is hypothesized to be superimposed on the true climate trend. Researchers attempt to adjust radiosonde data to compensate for instrument changes, but the adjustments are questionable due to incomplete and inaccurate knowledge of the instrument types used and the times when instruments changed at each station.

Previous work in this project has developed nearly-complete station histories of instrument changes by systematically searching time series of variables that are very sensitive to different instrument types. These variables show common data characteristics of the same instrument type used at individual stations, and discontinuities indicate instrument changes. Station histories for the Russian Federation, developed back to 1973, appear to be very reliable with consistent signals at over 200 stations. At least 12 radiosonde models are currently in use. Most station histories are complicated because of frequent alternations of instrument types.

It is a much more feasible task to determine differences between instrument types and to develop instrument corrections when reliable station histories are available. This project will test various methods to determine differences between instrument types and develop adjustments, first for reported temperatures and then for dew points. The primary method is "histogram matching," where cumulative probability distributions of reported data values are developed for common circumstances involving each distinct instrument type. To make one instrument statistically equivalent to another "reference" instrument type, the data value with the observed percentile is replaced by the data value with the same percentile in observations obtained with the "reference" instrument.

For a proper transformation, the data circumstances for the two instrument types involved in each comparison must be equivalent. Probability distributions will be developed in categories stratified by pressure level and sun angle, and also temperature in the case of dew point comparisons. However, either the long-term trend or short-term climate fluctuations such as ENSO or volcanic eruptions can affect the probability distributions and contaminate the adjustments. For example, if adjustments are based on the differences in probability distributions before and after an instrument change at a station, the adjustments remove part of the long-term trend or project short-term climate anomalies into other time periods. Averaging the adjustments over a large number of stations making the same transition at various times only partially relieves this problem. To determine if more satisfactory adjustments can be developed, this project will examine other comparisons such as alternating use of the instruments at a station, simultaneous use of the instruments at nearby stations, or transitions in both directions between instruments.