“…Evaluation studies showed that RO temperatures are generally derived with an accuracy of better than 1 K (e.g., Hajj et al, 2002;Wickert et al, 2004;Steiner et al, 2004) though it has also been found that RO retrieval results of major processing centres show, compared to other remote sensing instruments, radiosondes, and amongst themselves, biases exceeding 1 K above 25 km (e.g., Wang et al, 2004;Wickert et al, 2004;von Engeln, 2006) which is mainly due to the methodology of integrating a priori information in the highaltitude initialisation of the retrieval process as will be discussed in this study. Biases induced by downward propagated a priori information are a particular concern for climate applications like, e.g., long-term trend studies, since a priori data (often meteorological analyses) are on one hand particularly prone to model errors due to sparse observational data at high altitudes, and on the other hand they are far from long-term stable due to frequent model, observational data, and data assimilation system changes and improvements.…”