Abstract
We used Landsat Thematic Mapper (TM) data, field inventories and the k nearest neighbours (k‐nn) estimation method to estimate local species richness of pteridophytes and Melastomataceae in different ecological and taxonomical groups in an area of ca 670 km2 of primary rainforest in Amazonian Ecuador. We tested different spectral and spatial features of a satellite image and weighting of spectral features. To validate our estimates, we used a leave‐one‐out cross‐validation, an independent test dataset, and estimates based on randomly selected neighbours. The numbers of species in taxonomic groups were estimated less accurately than the numbers of species in ecological groups. The numbers of species in ecological categories were also estimated more accurately than the estimates based on randomly selected neighbours, except for the species on floodplain forest. Most of the estimated variables produced spatial patterns which followed the edaphic and topographical variation in the study area. The study showed that values for field variables can be predicted with Landsat TM data and k‐nn method in Amazonian rainforests.
Acknowledgements
We thank Robbin C. Moran, Axel D. Poulsen, Catalina Quintana, Gustavo Cañas, Jorge Celi, Rommel Montúfar and Sami Markkanen for collaboration in the field, Kai Mäkisara for doing the atmospheric correction of the satellite image, and Sandra Luque and an anonymous reviewer for comments on the manuscript. The study was mainly funded by the Academy of Finland, the Danish Center for Tropical Biodiversity, and the European Union.