This research systematically examines time series of over 200 variables from 1973 to mid-2001 at over 1200 stations, including stations with sparse data and fixed and moving ships. The goals are to validate existing metadata and infer serially complete elevation and instrument metadata. The main output of this research is a list of all stations with valid radiosonde data, which lists starting dates of each elevation, location, and instrument type, along with indicators of data quality and other comments.
Instrument types can be inferred because each radiosonde type has distinct physical characteristics, and usually specific data reporting practices. Time series show coherent signals at different levels and stations, with discontinuities coinciding with instrument changes. The most useful distinguishing variable is the 500-mb dew point depression. Signals found at well-documented stations are sought at stations with no (or suspect) metadata to infer instrument and elevation changes. Most changes are identifiable within one month, and exact observations of some changes can be identified.
Preliminary adjustments to dew points are developed to make each instrument type statistically similar to the mean of recent VIZ (Sippican) and Vaisala models, as a chosen "reference." Adjustments are made primarily to validate inferred instruments, since a wrong instrument type adjustment is immediately obvious.
The adjusted dew points also give initial indications of global trends in precipitable water. From 1973 to 1995, three distinct periods are 1973-76 (dry), 1979-87 (wet), and 1989-95 (dry), with transitions coinciding with documented climate shifts. After adjustments, 1979-87 averages 3.4% wetter than 1973-76, and 1989-95 averages 0.8% drier than 1979-87 (or 2.6% wetter than 1973-76). This analysis is being revalidated and will extend to mid-2001.