An automatic clustering technique applied to the study of vegetation fire patterns distribution in the African continent

The aim is to develop and evaluate the capability of an automatic clustering technique, In recognizing and quantitatively describing vegetation fires patterns as derived from NOAA-AVHRR GAC data. A refinement of the moment preserving clustering technique is proposed: the algorithm is validated on synthetic test data, and applied to the analysis of vegetation fire patterns distribution at regional and continental scale in Africa. The effect of scaling process on clustering is also discussed, and parametrization concerning the relative disposition of fire clusters and the size and shape of the clusters is proposed