Fast Forward Modelling of Cloudy Atmospheric States

Cloud detection always relies on some knowledge of how clear and cloudy observations will differ. In a full Bayesian determination of the probability that an infrared image pixel contains cloud, an estimate of the brightness temperature distribution for clear and cloudy cases is required. A method for estimating this distribution for cloudy atmospheric states through exploitation of the knowledge already held about an imaged scene is presented here. Relationships are found between cloud properties and the brightnesstemperature predictions of a fast radiative transfer model, run with atmospheric information specific to the imaged scene. This means that the number of model runs can be limited, without limiting the number of clouds represented in the distribution. The technique is demonstrated here in a case study, the results of which suggest that clear areas of an image can be identified with more certainty.