The Okavango inland Delta in Botswana is characterized by a high spatial and temporal variation in vegetation patches and flooding. Predicting the effects of escalating development projects in this pristine wildlife area is hampered by a lack of accurate maps. Efforts using traditional statistical methods have been futile. The processes forming this highly dynamic environment, however, give rise to a well‐documented consistency in the land cover pattern at scales ranging from single island architecture to an overall gradient in wetland, flood plain and island occurrence. We conducted a classification in a two‐step process starting with statistical methods, and then refining using indices and flooding data. The indices and flooding data were created and selected to make possible the inferring of knowledge about the patterns at different scales through declarative IF … THEN … statements. The initial statistical classification achieved a best result of 46% accuracy for 10 classes, whereas the rule‐based classification achieved an accuracy of 63%. Application of the derived classification for mapping islands and topography shows a surprisingly high accuracy.