New high resolution satellites for commercial purposes became available in the few years. This increases the need of new automatic knowledge based image interpretation methods. Such methods partially emulate the reasoning of an image analyst during the visual image interpretation. The present work falls into this context and proposes an automatic classification model for high resolution remotely sensed images. The model consists of three stages. In the first stage only spectral and textural information are used for classification. In the second stage GIS (geographic information system) data are combined with the result of the spectral analysis by means of fuzzy rules. In the third stage the multitemporal knowledge is introduced by estimating class transition possibilities within a given time span. To validate the proposed model experiments were performed based on IKONOS images from 2001 and 2002 as well as aerial photos from 1999 of the Pedra Branca Park, which is an important Atlantic Forest fragment in the State Rio de Janeiro in Southeast Brazil.