Cloud top inference for hyperspectral infrared radiance assimilation

Numerical weather prediction centers are now refining techniques to infer cloud height in order to assimilate hyperspectral (e.g. AIRS) radiances which are not affected by clouds. This is the object of this study. A first estimate of cloud top is obtained from radiative transfer calculations from 6-hour forecasts and assuming that the field of view is overcast: Povc. A second estimate uses the well known “CO2 slicing” (CS) technique to infer both cloud top height Pcs and effective cloud fraction (F) using a pair or multiple pairs of channels. In principle, Pcs is higher than Povc if F < 1. There are situations where the CS technique does not work well, especially for low clouds. Imperfect registration and observational biases are additional problems. From tests using GOES and AIRS data, the study proposes a robust algorithm involving a minum number of channels. As expected, some channels are more suitable to infer ice clouds (cirrus) while others are more suitable to infer water clouds. The influence of the selection on the radiance monitoring statistics and on the data volume is assessed.