Abstract
In this paper two geomorphologic maps (landform level and landscape level) are presented covering the French Guianan rainforest (84,000 km2) using full-resolution Shuttle Radar Topography Mission (SRTM) data. The entire country was segmented into 224,000 landform units on the basis of an original object-oriented approach using a modified counting box algorithm. A Principal Components Analysis (PCA) followed by k-means clustering (Ward's method) identified 12 different landform types corresponding to theoretical elementary landforms. The landscape map was generated by analyzing the spatial distribution of the different landform types. The different maps and models were compared with topographic field data collected on 92 transects totaling 260 km in length. The object-focused approach is a very efficient method that preserves geomorphologic consistency and discriminates between landforms using simple descriptors that are easily understood by non-geomorphologists. Despite major noise in the data, the landform map proved to be reliable and provided a strong spatial structure for the definition of landscape units. We recommend using the landform map at scales 1: 100,000?1: 250,000. Landscape map, used on a 1:1,000,000?1:2,000,000 scale, enabled us to draw bio-geographical limits in this region and provides exhaustive relief information that usefully supplements the geological map.