Applying multivariate analyses to the study of bird-habitat relationships: A case study with montane, non-forest passerines

Full article

Published:

Volume 50(1), June 2003. Pages 47-58.

Language:

Spanish

Original title:

Aplicación de análisis multivariantes al estudio de las relaciones entre las aves y sus hábitats: un ejemplo con paseriformes montanos no forestales

Keywords:

Canonical correspondence analysis (CCA), habitat selection, models of response, montane non-forest passerines, multivariate analysis, principal components analysis (PCA), principal components regression (PCR)

Abstract:

This paper compares the results obtained after studying habitat selection of an assemblage of montane, non-forest passerines in the Gorbeia Natural Park (N of the Iberian Peninsula) by means of two multivariate approaches. The first (indirect) approach implied the ordination of the sampling sites by means of principal component analysis (PCA) of the environmental variables determined at each sampling site, followed by the regression of the specific abundances on the gradients described by the PCA by means of generalized linear models (GLM) assembled by polynomial regression. The second (direct) approach involved the simultaneous ordination of the sampling sites and the studied species by means of canonical correspondence analysis (CCA) of abundances constrained by environmental variables, followed by GLM polynomial regression of abundances on CCA axes. The first three PCA axes explained 47.5% of the total variability in environmental data. Our indirect approach allowed us to obtain 6 out of 8 statistically significant GLM models, being their corresponding goodness-of-fit between 26.9% and 47.1% of the original deviance in abundance data. Both the first and all the CCA axes were statistically significant, as verified by means of Monte Carlo permutation tests. Canonical axes represented a good solution to the global ordination of species abundances as a function of explicative environmental variables (explained inertia = 36.6%), segregating species by means of their habitat selection. After partialing out the effect of physical variables (altitude, slope and direction), the principal explicative variables (21.3% of the total inertia) were descriptors of habitat structure (relative cover by herbaceous vegetation, ferns, heath shrubs, tree shrubs, spiny gorses and rocks). By means of our direct approach a total of 6 out of 8 statistically significant GLM models were obtained, being their goodness-of-fit from 34.4% to 72.3% of the corresponding null model deviance. Goodness-of-fit of the significant models obtained with both approaches was statistically superior in the GLM models obtained following CCA than after PCA (Wilcoxon signed ranks test: Z = 2.023; n1 = n2 = 5; P = 0.043). In conclusion, when modeling the bird-habitat relationships the direct approach (canonical correspondence analysis plus appropriate GLM modeling) seems to be more effective than the indirect approach (principal components regression). The models obtained allow improving previous descriptions of bird habitat selection, adjusting them to the high resolution of our research scale, which makes them useful in the management of the Natural Park montane, non-forest habitats.