Archived radiosonde data is extensive enough that global and regional trends of atmospheric temperature and moisture (total precipitable water) can potentially be calculated back to about 1958. However, all trends computed from radiosonde data are uncertain because instrument changes cause artificial discontinuities and because station histories are incomplete and inaccurate.

The first goal of the ongoing Validated Atmospheric Profiles for Operations and Research (VAPOR) project is to develop complete historical station and instrument metadata. Three major steps are involved: (1) Combine and add all available metadata updates to the most extensive historical radiosonde metadata sources. (2) At stations which are well-documented, search for consistent characteristics of specific instrument types, and discontinuities which indicate instrument changes, in time series of variables which are very sensitive to different instrument types. (3) At stations and in time periods where metadata is absent or incorrect, infer the use of specific instrument types by similar characteristics, and the dates of changes by discontinuities, in time series of the same variables.

The second goal is to develop and apply adjustments for each distinct instrument type to allow computation of unbiased atmospheric temperature and moisture trends. While the metadata is not complete, it is most complete and best checked in Japan, China, the Russian Federation, and India in the period starting 1973. This reports on initial efforts to develop adjustments to compensate for instrument changes and their biases where the documentation is nearly complete and accurate (Japan), moderately detailed (China), mostly complete and involving a large number of instrument types (Russian Federation), or highly uncertain (India). In general, because technological improvements have caused radiosondes to gradually become more sensitive and better protected from radiation errors, the instrument-caused trend is generally an erroneous cooling and drying, with the largest errors (and improvements) at the highest levels.

The method of data adjustment is called equiprobability transformation. Basically, readings from all instruments are adjusted so the probability distribution of each variable of climate interest (temperature and dew point depression) in the same climate environment (stratified by pressure interval and sun angle, and by temperature interval in the case of dew point depression) matches the probability distribution of the same variable using a target "reference" instrument type. The best comparisons of instruments are obtained from data either before and after a change to (or from) a reference instrument at the same station, or with simultaneous use of the two instrument types at different closely-located stations.