Estimating Number of Pteridophyte and Melastomataceae Species from Satellite Images in Western Amazonian Rain Forests

Abstract

The paper studies the variation of rain forests vegetation by means of satellite images and data of 30 field 9 species of understorey pteridophytes (ferns and fern allies) and Melastomataceae mainly small trees and shrubs). Pteridophytes and Melastomataceae species were seen as indicators of more general floristic patterns in the present study. The numbers of species and individuals in taxonomically and ecologically defined species groups were estimated using Landsat TM images and field measurements in primary lowland rain forests in Amazonian Ecuador. The k-nearest neighbours (k-nn) estimation method was applied. Different spectral features and weighting of features were tested. The final estimations were computed using means and standard deviations of bands TM1–TM5 and TM7 calculated in a window of 7x7 pixels and weighted using the optimal weight seeking method. The estimates were evaluated by the leave-one-out cross-validation method and by using a separate test data set. A destriping method was developed. The systematic brightness variation towards the eastern edge of the image was corrected. The destriping method worked well if the striping was continuous from one image edge to another. The results showed that the number of species could be estimated relatively accurately with the method applied. The root mean squared errors (RMSE) for the estimates of the number of individuals were mostly high. The estimates for pteridophyte species were more accurate than for Melastomataceae species. Visualisations showed that the estimates for the number of species in ecological groups varied spatially, and that observed spatial pattern could be largely explained by topography. Success in satellite based estimation of understorey species suggest that the variation of these species is related to the spatial patterns of canopy trees.