In numerical weather prediction one responds to the chaotic nature of weather by performing many simulations with slightly modified starting conditions. This ensemble of simulations helps to judge the reliability of the forecast and to estimate the probability of the occurrence of certain weather events. When using this technique on a regular basis it is desirable to detect interesting weather phenomena automatically in an ensemble and to present the results in a concise graphical way. This requires that the essential structures of the weather phenomena can be adequately characterized and that they can be identified in large data sets. In this interdisciplinary project the meteorologists will enhance their methods for detecting and analyzing weather phenomena in such a way that the computer scientists can develop efficient algorithms for searching through the simulation data in order to find the considered weather situation. In this context the minimization of external memory access as well as the parallelization of the processes will play an important role. The object oriented analysis will be followed by the question how to visualize the results in a meaningful and clear manner.