Avaliação de métodos para estimativas de dissimilaridade em gradientes ecológicos com alta diversidade beta

Abstract

There are several problems on the analysis of biological communities with sparse data, resulting from gradients with high beta diversity. I used four strategies to solve this problem (Beals smoothing, Swan, Shortest Path and Extended Dissimilarity). I randomly removed from 1% to 50% of the individuals in empirical and simulated matrices. I then performed PCoA and nMDS ordinations and used Procrustes correlation of the original two dimensional ordination with the ordination obtained using the degraded matrices. For the simulated data set, I also correlated the ordenation in two dimensions with the coordinates of the samples in the two-dimensional simulated gradients. Finally, I analyzed how robustness to degradation, quantified as Procrustean correlation, was related to the matrix properties. Different from the expected, in the comparison of the degraded and original ordinations, the uncorrected data with a traditional dissimilarity index (Bray-Curtis) produced higher fit than the four methods evaluated. In relation to the coordinates of the simulated two-dimensional gradients, the evaluated methods were slightly better than the raw data. Overall, the simulated data were more robust to the degradation than the empirical ones and the data of abundance were more robust than matrices of presence and absence. Matrices with small proportion of zeros were more robust to degradation. I conclude that the correction methods evaluated distorted the pattern on the original data. Also, data with low beta diversity (few zeros) are robust to degradation and sufficient to reconstruct the original gradient.