Data analysis
As the points sampled are geographically aggregated, the autocorrelation
of the data was tested through the Moran’s Index. Spatial maps of
suitable vectors are considered the best way to control autocorrelation
and are good representatives of the spatial structure present in the
data (Mauricio Bini et al. 2009). The eigenvectors of the first
eigenvalues are related to the local structuring of communities,
species with small dispersion skills. In order to perform spatial
eigenvectors maps is necessary to know the relationship between all
pairs of points present in the analysis, known as the Weight (W) matrix
(Grffith et al. 2006). In this way we define four W: (i) Global W
(Appendix 1 - S1 - GW), defined by the linear distance between all
streams (Appendix 2); (ii) Local W (Appendix 1 - S1 - LW), defined by
the linear distance between streams present in the same hydrographic
unit. Streams in different units had no interaction and connectivity
values equal to zero (Appendix 2); (iii) Water W (Appendix 1 - S1 - WW),
defined by the hydrographic distance. The same way as in Local W,
streams in different units had no interaction and connectivity values
equal to zero (Appendix 2); and (iv) W Hydalt (Appendix 1 - S1 - HW),
defined by the hydrographic distance between the points weighted by the
slope (Appendix 2).
We measured ecological diversity using two distinct metrics: (i) species
richness and (ii) beta diversity. Species richness was defined as the
number of species present at the site of interest. The beta diversity
was calculated according to the procedure described by Baselga (2010),
which defines beta diversity as the Sorensen dissimilarity index.
Diversity indices were calculated for both the total fish community and
for different feeding guilds (detritivores, insectivores and omnivores).
The classification of the fish species by trophic guild were performed
considering the literature. We used generalized linear models (GLM) to
identify the best way to represent environmental conditions for streams
fish communities. We compared models of community diversity indices and
the average of the environmental conditions with a second model
including the average together with the standard deviations of
environmental conditions as predictor variables. This procedure was
adopted to identify the best way to describe environment conditions, the
average or variance. The GLM models was performed using all W matrices
(Table 1) to identify the best means to represent neutral processes
(linear or hydrographic distances). To avoid multicollinearity, a
Principal Component Analysis (PCA) was performed with the average
values, and a second PCA with the average values together with the
standard deviation values of the descriptors, the first two axes of
explanation of the PCA being used as predictor variables on the models.
After identifying the best way to represent environmental conditions
(mean or mean with standard deviation), GLMs were performed for each
trophic guild and all W matrices. This procedure was used to identify if
results found for the whole community were equivalent to those found for
individual feeding guilds.
Table 1. Models used to determine the best set of descriptors
of environmental conditions and spatial structure considering the
richness and beta diversity of ichthyofauna.