A new cloud recognition algorithm for optical sensors

The Optical Cloud Recognition Algorithm (OCRA) presented in this paper is a novel approach for optical cloud recognition on a global basis, which overcomes the problems of the traditional threshold technique, and runs completely automatically. The key of the OCRA algorithm is the determination of a so called "cloud-free composite" using historical data. The cloud coverage is obtained by comparing the actual measurement with the pre-determined cloud-free composite. The cloud-free composite is generated taking the measured data set with minimum brightness (less cloudy) for a given ground pixel during a large period of time (e.g. one year). Before the comparison among multitemporal optical data sets with different viewing geometry and different atmospheric conditions can be performed, a correction is needed. This correction is done applying a physical model of image formation invariant to topography, position of the Sun, atmosphere, and position of the viewer. The cloud coverage is determined through a linear comparison of the actual optical measurement and the cloud-free composite. The novel technique developed for this algorithm is applicable to any optical sensor and works complete automatically. The results show a good correlation between the cloud coverage derived by OCRA and that produced by other algorithms. Furthermore, a good agreement is achieved compared to METEOSAT data.