Temperature Trends over Germany from Homogenized Radiosonde Data

Radiosonde temperature measurements started operationally in the mid 1950’s and thus provide a long-term time series for the upper atmosphere. However the radiosonde instruments have undergone repeated developments since their introduction. This resulted in multiple abrupt changes (breakpoints) in the data. For instance the radiation errors of the large instruments lead to too high temperature readings during daytime in the early years, introducing an artificial bias in the temperature data. Therefore this raw dataset cannot be used for climate studies without any correction.

Temperature trend profiles for German stations after the homogenization. Error bars show typical ±2σ uncertainty

In this work a manual homogenization procedure was developed in order to homogenize German radiosonde data. The radiosonde soundings were investigated at 11 mandatory pressure levels, up to 50hPa. This work was a part of the MOSQUITO Module E project, in order to provide a long-time free atmosphere dataset for the validation of the MiKlip MPI-ESM decadal climate hindcasts.
The homogenization process can be split into three main parts: determination of the date of breakpoints, approximation of the biases, and the test of the results. For the first issue the available metadata were collected, and verified for instance with day-time minus night-time differences, or tropopause humidity records. It turned out that only 34% of the determined breakpoints were correctly described by the metadata, and more than 52% of them were not mentioned in the metadata at all. The determinations of the artificial biases occurred for day-time and night-time data separately. The determined biases were later cross-checked for consistency.
The biases were in the early years usually higher. Typical biases for daytime soundings are up to 3K in the stratosphere, and 0.5K in the troposphere before the 70’s, and reduce to 0.1-0.3K in the 2000’s.
Compared to other, automatically homogenized, radiosonde datasets (HadAT2, RICH), our homogenized data provide a more reliable dataset for Germany. The HadAT2 dataset shows 1K higher temperature below the tropopause starting from around 2004, during the period in which the most accurate radiosondes operated. Additionally it contains suspicious peaks in the data. These discrepancies can be an effect of the automatic evaluation procedure. Our data coincide with the RICH data better; however the RICH may underestimate the temperature bias in the stratosphere in the early years.
The amendment of the long-time temperature trends was investigated. After the homogenization process, the upper-air temperature trend profiles of the German station became more uniform. There is a positive trend in the troposphere of around +0.25K per decade. The height of the no-trend-level is around 250hPa. There is an increasing negative trend above this level reaching around -0.45K per decade at 50hPa. The trends for the same period are approximately 0.1K higher per decade, both in the stratosphere and in the troposphere, for the homogenized data than for the raw time series.



The difference between the trends after and before homogenization.
Back to List