Mapping gradual landscape‐scale floristic changes in Amazonian primary rain forests by combining ordination and remote sensing

Abstract

ABSTRACTAim We present a new method to economically map gradual changes in plant species composition in lowland rain forests using field data and satellite images. Such a method will be a useful tool in planning the sustainable use and conservation of Amazonian rain forests.Location The study covered an area of c. 700 km2 of primary rain forest in Amazonian Ecuador.Methods We field inventoried the species composition of pteridophytes and Melastomataceae in 340 inventory plots (5 m × 50 m), described the prevailing topography and analysed soil cation concentration and texture. We used non‐metric multidimensional scaling (NMDS) to summarize the floristic variation among the inventory plots in three ordination dimensions. The scores of the three ordination axes were predicted to non‐visited places using a Landsat TM (thematic mapper) satellite image and the k nearest neighbours (knn) estimation method. To avoid extrapolation, we excluded from the analysis those pixel windows whose spectral values were not represented in the areas covered by field sampling. The accuracy of the predictions was evaluated by cross‐validation and by comparing the predictions based on spectrally nearest neighbours to the predictions based on random neighbours.Results The floristic gradients presented by NMDS ordination were interpretable in terms of topography, drainage and soil cation content. Thirteen percent of the cloud‐free pixels were excluded from the knn analysis to avoid extrapolation. The estimates of the floristic ordination scores based on spectrally nearest neighbours were always more accurate than estimates based on random neighbours.Main conclusions The presented method needs a relatively small input of work and resources, is mechanistic and produces maps that give relevant information on floristic variation over forest areas that are traditionally considered essentially homogeneous. Therefore, the method appears to have a great potential for use in mapping large areas of Amazonian rain forests.