Implications of complex upper air station histories on approaches to adjust archived radiosonde data for instrument discontinuities

The archived record of radiosonde soundings has nearly-global coverage since 1957 so it is potentially very useful for analyzing the effects of increasing greenhouse gases, but computed atmospheric temperature and moisture trends are not accurate due to frequent instrument changes. Documentation of station instrument histories is incomplete and some of the available information is erroneous, so the process of developing adjustments for instrument changes is hindered because the nature and timing of instrument changes is uncertain.

An ongoing project is developing complete station instrument histories by examining time series of sensitive variables that tend to amplify instrument differences. These variables are mostly moisture-related and show a very high degree of consistency for the same instrument type, with smooth variations with pressure level, season, and climatic environment of different regions. Signatures of specific instrument types are first developed using well-documented stations. Discontinuities involving interrelated variables indicate instrument changes and are so consistent that in many cases an instrument change can be identified to the exact observation. Similar signals are seen at stations with no (or erroneous) documentation, allowing the development of complete instrument histories with only minor uncertainties.

Histories constructed so far show that no station is homogeneous for a long period. In many countries, it is typical for the same station to use multiple instrument types for long periods, with dozens or hundreds of alternations. For example, in Russia, many stations used 4 to 7 instrument models from 2007 to 2009, and past transitions have typically involved varying mixes of 2 instrument types, often for several years.

This project proposes to develop instrument type adjustments, rather than separate adjustments for each station, using "histogram matching." To adjust each instrument type to be statistically equivalent to a chosen "reference" instrument type, the cumulative probability distributions of a variable are developed using carefully matched circumstances (such as stratification by pressure layer and sun angle) for each instrument type. To adjust an instrument to be equivalent to the reference, when the percentile (cumulative probability) of a data value (such as temperature at a pressure level) is determined, the adjustment is simply to substitute the data value at the same percentile for the reference instrument for the reported value.

The main implications of complex station histories for adjustment methods developed by other researchers are that (1) With frequent instrument changes, adjustments based on before and after each transition will probably remove a large portion of the true trend, (2) Change points are often blurred because many transitions are not a change from instrument type exclusively to another, and (3) Automated methods to detect undocumented change points probably have many false detections or omissions of change points or detections at the wrong time, resulting in incorrect estimates of the magnitudes of discontinuities. However, any method should benefit from being applied with better knowledge of station histories. If a method identifies a change point that this study has not detected, the change point needs to be further examined to determine if it is a station radiosonde change, but if the change points identified by that method agree with the instrument changes developed here, confidence in the instrument histories is increased.