Abstract
Several recent papers have reported that temperatures from satellites lack the variance found in radiosonde temperatures. Because of these results, we performed an experiment to distinguish between limitations of the satellite processing system and characteristics of the procedures used to produce temperatures. In the experiment, we calculated radiances, the quantities observed by the satellite, for a set of radiosondes for which we also had observed radiances. We then obtained regression coefficients for both the observed and calculated radiances and applied them to an independent set. We discovered two significant features: One is that the variances produced by the current satellite retrieval system are significantly closer to variances obtained from radiosondes than was reported in the earlier papers. We attributed the change to modifications in the retrieval system which wore designed to increase the accuracy of satellite retrievals. The second is that with the present system, a large portion of the remaining variance loss is recovered through using more accurate radiances.
