On the homogeneity of radiosonde wind time series

Abstract

Radiosonde observations of temperature and wind have been an important source of upper air data during the past few decades. To be useful for climate studies, the quality and temporal homogeneity of this dataset is essential. Large efforts have been put into quality control and homogenization of radiosonde temperature data. Much less emphasis has been put on the bias correction of wind data derived from radiosonde ascents. This paper takes a first look into the temporal homogeneity of the global radiosonde wind data set as it has been compiled in the 40-year ECMWF reanalysis (ERA-40) and operational ECMWF analysis databases as well as in the Integrated Global Radiosonde Archive (IGRA). More than 1000 breaks have been detected with a preliminary detection and adjustment method based on the analysis of time series of background departures from the ERA-40 and the operational ECMWF data assimilation systems. Most breaks can be attributed to significant inhomogeneities (median size 2 m/s) in the observed radiosonde wind data. Especially errors in the north alignment could be detected quite well. During some periods, especially in the mid-1970s, the homogeneity of the ERA-40 background wind is questionable, however. An additional estimation of the break sizes using neighboring stations as reference is therefore recommended. As far as could be seen, the homogenization has little systematic impact on global mean observed wind trends.