Models are only useful if they address specific questions. They must simplify the representation of what can be complex systems.

We presented here the development of a forecasting method from the observation of the regions area based on maps of geographical data over time. The method has been applied over a region located in the Ariège mountains called ``Soulane'' to describe the evolution of its landscape from three provided maps (1942, 1962 et 1993). These models allow a quick generation of a set of information in a GIS database.

This approach of evolution modeling takes into account historical spatial data about the regions of progression and regression. Such mechanism allows the integration of several important landscape features by means of out-image data.

It uses raster maps processing and manages the inherent uncertainties and imprecision that surround the evolution modeling of a landscape providing a flexible interpretation of the results.

The results from the experimental tests presented here encourages further work to enhance the method of spatio-temporal forecasting described in this approach.