This is an evaluation of a retrieval, data assimilation, and forecast experiment on satellite sounding data from the NOAA-9 and NOAA-10 polar orbiting satellites during a 15.5 day period in January–February 1987. Retrievals were produced globally through the 3I (improved initialization inversion) scheme developed at Laboratoire de Météorologie Dynamique, Paris. In this paper we use the European Centre for Medium Range Weather Forecasts (ECMWF) global data assimilation system to evaluate the global performance of the 3I retrieval scheme. The 3I-retrieved temperature profiles have also been compared with the operational retrievals produced by NESDIS (National Environmental Satellite, Data, and Information Service), Washington.
It is shown that the large airmass-dependent biases, characteristic of the NESDIS retrievals, also occur with 3I, but with smaller amplitude. Some large local 3I retrieval errors are identified and explained, and their impact on analysis and forecasts demonstrated. The forecast scores with the 3I retrievals are worse than the scores with the NESDIS retrievals. This may be due to the much larger volume of soundings used in the 3I experiment. The use of a global assimilation system has identified aspects of the 3I system which can be improved.