Investigation of scene identification algorithms for radiation budget measurements

Computation of the Earth radiation budget from satellite measurements requires identification of the scene in order to select spectral factors and bidirectional models. Evaluation of the effects of scene identification errors on the resulting radiant exitances requires scene identification probability matrices, which quote the probability that a scene which is, in truth, type j is identified as type i . These matrices are empirically evaluated by intercomparing the results from the Earth Radiation Budget Experiment (ERBE) scene identification algorithm with results using Advanced Very High Resolution Radiometer (AVHRR) data. A scene identification procedure is developed for AVHRR data by using radiative transfer models. Scene types for these AVHRR pixels are attached to matching ERBE pixels, and the results are sorted into scene identification probability matrices. These scene intercomparisons explain some of the strengths and problems of the ERBE scene identification algorithm. A tendency for underestimation of cloudiness over ocean at high cloud amounts for the ERBE relative to the AVHRR results is found and explained. Estimates of the errors of exitances due to scene misidentification demonstrate the high dependency on the probability matrices. While the longwave error can generally be neglected, the shortwave deviations can reach maximum values of more than 13% of the respective exitances.