Simulation of Extended Satellite Scenes for Validation Purposes

Abstract

Remote sensing of clouds is a challenging task, due to the inherent inhomogeneity of clouds on all spatial and temporal scales. Passive cloud remote sensing is generally based on two assumptions: (1) that clouds and reflected or emitted radiance are homogeneous over the sensor's field-of-view; and (2), that individual pixels may be treated independently of each other. Both assumptions have been shown to cause systematic bias and random noise in the retrieved optical and micro-physical properties. For this reason, validation of the satellite-derived properties with independently measured data is urgently needed. On the other hand, independent observations of cloud optical properties, e.g. the optical thickness, are very hard to obtain for the very same reason: the high variability of clouds in space and time. To validate passive remote sensing methods we started a self-consistent model-based validation approach. Extended satellite images the size of central Europe, based on the output of a weather forecast model (the German Lokalmodell, LM), are used as input for 1D and 3D radiative transfer simulations to calculate satellite radiances. The thus generated artificial images provide an ideal basis for the validation of remote sensing methods because satellite radiances are known as well as the cloud properties. The generated scenes are large enough for a real operational retrieval test which we illustrate by an example application of a cloud detection algorithm and a cloud property retrieval for the SEVIRI imager on Meteosat Second Generation (MSG) to such a scene. The cloud scenes could also complement the already existing EarthCARE simulator.