Clustering applied to cloud wind determination

Reliable cloud motion wind generation from Meteosat images requires good target selection. This is usually done by examining the infrared channel and selecting target windows which have a temperature variation between an upper limit and lower limit, i.e., windows containing essentially a single cloud layer. In this paper we apply an optimised multi-spectral clustering algorithm in an attempt to extract the principal cloud targets prior to target tracking. Experimental results show an increase in the number of trackable targets compared to conventional techniques based on raw data. The paper also examines the optimal target size and compares the performance of several target tracking techniques.