Deriving Complete Upper-Air Station Histories Using Sensitive Data Variables—An Essential Step in Homogenizing the Atmospheric Climate Record

Archived radiosonde soundings have near-global coverage back to 1957, but to compute credible atmospheric temperature and moisture trends, data must be adjusted to compensate for instrument biases. Many soundings have manufacturer-specific corrections applied, but corrections were developed from limited tests so data biases remain. As radiosonde errors have gradually decreased, the global trend is contaminated by artificial cooling and drying.

Historical metadata listing upper air stations, locations, elevations, instruments, and change dates is needed to properly identify discontinuities (breakpoints) for developing rigorous corrections. However, metadata compilations are quite fragmentary, such as catalogs updated every few years that rarely list change dates, so researchers often make data adjustments that do not coincide with instrument changes. Most efforts to identify breakpoints have many missed and false detections because natural fluctuations (especially in temperature) mask many instrument discontinuities, many stations frequently alternate between radiosonde models so discontinuities are blurred, many projects do not use available instrument information, and few projects search for humidity discontinuities that often precisely identify an actual instrument change.

This project focuses on developing serially complete station radiosonde histories by consolidating many previously-unused current and historical metadata sources, then using archived soundings to check all metadata and to fill in missing information. So far, >3000 upper air instrument types and >6000 stations are documented. An appropriate discontinuity in a station time series usually identifies the actual change date between different instruments reported in successive catalogs. Computing the surface elevation hydrostatically from reported heights has identified the time of thousands of station moves, and ~1000 launch locations have been verified to ~10 m accuracy using site photos or descriptions, and online satellite photos.

In sounding data, instrument discontinuities are best detected and most attributed to specific instrument models using “sensitive” variables focusing on extremes of sensor reporting (especially humidity near 0 or 100%). In several cases, an unknown discontinuity indicated an unreported instrument type, which was later confirmed by personal communications. All stations using the same instrument model usually have similar data characteristics, and distinct characteristics among soundings at a station can sometimes identify exact use of up to 3 instrument types.