87th AMS Annual Meeting

Thursday, 18 January 2007: 8:30 AM
Validation of radiosondes using statistical reconstructions: data prior to 1958
207A (Henry B. Gonzalez Convention Center)
Andrea N. Grant, ETH, Zurich, Switzerland; and S. Brönnimann and T. Ewen
Historical radiosonde data suffer from numerous homogeneity issues due to changes in instrumentation, launch practices, station locations, and launch times. This is especially the case in early years when the global radiosonde network was still under development and due to a worldwide change in launch times which occurred in 1957-1958. We have developed a validation procedure based on comparing the monthly mean radiosonde data to a suitable reference series. As reference series we statistically reconstruct a series for each station based on a multiple linear regression model relating surface temperature and pressure to NCEP-NCAR reanalysis. This validation process was applied to pre-1958 radiosonde data at stations from several international archives including the following datasets: TD54 (US Air Force, archived at National Center for Atmospheric Research), TD-6201 (National Climatic Data Unit (NCDC)), CARDS data from 1946 and 1947 (NCDC), Integrated Global Radiosonde Archive (NCDC), comprising over 800 unique stations.

The validation technique is described as well as the results of the validation process for selected stations, including typical diagnostic plots and types of errors. We detected large errors in some series while others proved to be of excellent quality. Typical problems included station relocation (especially during war years), uncorrected (or insufficiently corrected) radiation and lag errors, temperature biases, incorrect timestamp, and inconsistent unit conversion. Additionally, during the preliminary processing, an adjustment is made for launch time based on a diurnal cycle interpolated from NCEP-NCAR Reanalysis climatology. The robustness of this diurnal cycle adjustment across the 1957-1958 launch time inhomogeneity is investigated. A summary of statistics is presented including error types and magnitudes.

Supplementary URL: