Removal of Ambiguities in Meteorological Fields Through Stochastic Minimization of the Spatial Variability

Sometimes there are a finite number of options for the value of a meteorological variable at a grid point. When no physical selection criterion is available, values are likely to be chosen in such a manner that the spatial variability of the resulting field is as small as possible. An algorithm based on simulated annealing is constructed to accomplish the optimization of smoothness in case of a large number of ambiguities. Two concrete applications are discussed, where the first one is the construction of an equivalentpotential temperature surface. There exists a rule of thumb to position the surface at the greatest possible altitude, and a couple of arguments are given why the surface resulting therefrom should be close to the smoothest possible. The simulated annealing procedure by and large confirms this presumption, but does in special situations show a distinct capability to yield considerably better results. This potential benefit of simulated annealing compared to less sophisticated approaches becomes even more apparent for the second application which deals with the selection of Meteosat infra-red cloud motion vectors from several vector fields achieved through varying the size of the targets to be tracked. Despite a certain slowness of convergence towards a final solution, simulated annealing proves to be a promising approach to the considered type of meteorological optimization problems. The cloud motion vector experiments represent the first step of an unconventional attempt of cloud/no-cloud discrimination which is based on the hypothesis that tracking of features in infra-red imagery should yield results of different character for cloudy resp. cloud free regions.