SUPPORTING INFORMATION
Kärnä, O.-M., Grönroos, M., Antikainen, H., Hjort, J., Ilmonen, J., Paasivirta, L. & Heino, J.
(2015) Inferring the effects of potential dispersal routes on the metacommunity structure of
stream insects: as the crow flies, as the fish swims or as the fox runs? Journal of Animal
Ecology.
Appendix S1. Results of the BIO-ENV analysis.
The number of environmental variables selected varied greatly in the BIO-ENV analysis,
varying between body size or dispersal mode classes from one to seven (Table S1a, b). Moss
cover was arguably the most important variable in our study because it was also selected in
the final environmental distance matrices for all body size or dispersal mode classes. Other
habitat and water chemistry variables were selected less often in the best environmental
distance matrices. However, the results changed to some degree when presence-absence data
were used. Although moss cover was still the variable selected in all final environmental
distance matrices, the best matrices incorporated more variables than those in the analysis of
abundance data. The best environmental distance matrices correlating with biological
dissimilarities based on presence-absence data hence included various habitat and water
chemistry variables.
1
Table S1. Environmental variables selected by BIO-ENV for the final environmental
distance matrices for each response dissimilarity matrix (i.e., entire data, active dispersers,
passive dispersers, and each size class). Shown are results based on dissimilarity matrices of
(a) abundance data with Bray-Curtis coefficient and (b) presence-absence data with Sørensen
coefficient. Mantel correlation coefficients (r) are shown.
(a)
Selected variables for abundance data
r
Entire data
Moss cover
0.459
0 - 0.25 cm
Iron, manganese, moss cover, channel width
0.253
0.25 - 0.50 cm Total nitrogen, moss cover
0.251
0.50 - 1 cm
Total nitrogen, manganese, cobble, moss cover, depth, channel width
0.398
1 - 2 cm
pH, gravel, moss cover
0.373
2 - 4 cm
Gravel, cobble, moss cover, velocity
0.287
Active
pH, gravel, moss cover
0.390
Passive
Total nitrogen, moss cover
0.409
(b)
Selected variables for presence-absence data
Entire data
Total nitrogen, boulder, moss cover, depth
0.453
0 - 0.25 cm
Iron, channel width
0.283
r
0.25 - 0.50 cm Total nitrogen, colour, pH, boulder, moss cover, velocity
0.290
0.50 - 1 cm
Total nitrogen, manganese, boulder, moss cover, depth
0.425
1 - 2 cm
Boulder, moss cover, depth, channel width
0.329
2 - 4 cm
Sand, moss cover, velocity
0.217
Active
Total nitrogen, sand, boulder, moss cover, velocity, depth, channel width
0.353
Passive
Total nitrogen, manganese, boulder, moss cover
0.470
2
Appendix S2. Modelling variation in raw data using environmental and spatial variables.
To validate our Mantel test results, we also used a commonly-used modelling approach to
examine variation in community structure among sites (Legendre, Borcard & Peres-Neto
2005). This consisted of using multiple spatial and environmental variables in constrained
ordination. Constrained ordination analyses were run for all species as well as for active,
passive, size class 0.25 – 0.50 cm, size class 0.50 - 1 cm, size class 1-2 cm and size class 2-4
cm. Species belonging to size class 0.25 – 0.50 cm occurred at very few sites and were not
considered here.
We used Moran’s eigenvector maps to model spatial structures among the provinces
and to provide spatial variables for our modelling endeavours (Griffith & Peres-Neto 2006;
Legendre & Legendre 2012). Of the methods belonging to the family of Moran’s eigenvector
maps, we used the traditional principal coordinates of neighbour matrix analyses (PCNM)
based on Euclidean distances among the sites (Borcard, Gillet & Legendre 2011). We used
the PCNM eigenvectors showing positive spatial autocorrelation as explanatory variables in
analyses aimed to explain variation in assemblage composition. The first PCNM eigenvectors
with large eigenvalues describe broad-scale spatial structures, whereas the PCNM
eigenvectors with small eigenvalues describe fine-scale spatial variation (Borcard &
Legendre 2002; Legendre & Legendre 2012). The PCNM eigenvectors are mutually
orthogonal, linearly unrelated spatial variables and can thus be used to account for spatial
autocorrelation in assemblage composition (Borcard & Legendre 2002; Legendre &
Legendre, 2012). Significant spatial variation in assemblage composition that is related to
such spatial variables may result from environmental autocorrelation, dispersal limitation or
historical effects on assemblage composition (Dray et al. 2012). PCNM analysis was
conducted using the R package PCNM in the R version 2.15.3 (Legendre et al. 2013).
3
We used redundancy analysis (RDA; Rao 1964) to analyse variation in raw species
abundance data. RDA examines variation in species composition (Y) in relation to sets of
predictor variables that were, in the present study, environmental variables (E) and spatial
variables (S) derived from principal coordinates of neighbour matrix analysis (see above).
Prior to the RDA, site-by-species abundance data were Hellinger-transformed to make the
data analysable using linear methods (Legendre & Gallagher 2001). We selected significant
variables in the final RDA models of each set of variables (E or S) following the forward
selection method with two stopping rules (Blanchet et al. (2008) and using the function
“ordiR2step” in the R package vegan (Oksanen et al. 2013). We used redundancy analysis
(RDA) to partition variation in species composition (Y) between E and S following the
widely-used variation partitioning approach (Borcard et al. 1992; Legendre & Legendre
2012). Variation partitioning of species composition (Y) between two sets of predictor
variables results in pure environment (E│S) and pure spatial (S│E) fractions, as well as their
shared effect (E∩S) and unexplained variance (U). We used adjusted R2 values in all
analyses because they are unbiased estimates of variation (Peres-Neto et al. 2006).
PCNM produced 17 spatial variables showing positive autocorrelation. None to seven
spatial variables were selected in the spatial models of different groupings of insects (Table
S1). The most important spatial variable was V1 that was selected in all but one spatial
model. Correspondingly, one to five environmental variables were selected in the
environmental models. The most important environmental variable was moss cover that
occurred in all environmental models (Table S1). Environmental variables were more
important than spatial variables, except for size class 0.5-1 cm and passive dispersers for
which spatial variables were more important than environmental variables.
4
Table S2. Results of variation partitioning for the entire insect data, different size classes,
active dispersers and passive dispersers. Abbreviations: env = environmental model, spa =
spatial model. “-” = no variable was significant.
Df
5
4
9
Adj. R2
0.225
0.104
0.301
5
0
4
0.197
0.028
0.076
0.699
3
3
6
0.103
0.071
0.145
3
0
3
0.074
0.028
0.043
0.854
Size 0.5-1 cm
[a+b] = env
[b+c] = spa
[a+b+c]
Individual fractions
[a] = env|spa
[b] = shared
[c] = spa|env
[d] = Residuals
Df
3
6
9
Adj. R2
0.159
0.166
0.302
3
0
6
0.136
0.023
0.143
0.698
Size 1-2 cm
[a+b] = env
[b+c] = spa
[a+b+c]
Individual fractions
[a] = env|spa
[b] = shared
[c] = spa|env
[d] = Residuals
Df
4
1
5
Adj. R2
0.313
0.060
0.355
4
0
1
0.296
0.018
0.042
0.645
Size 2-4 cm
[a+b] = env
[b+c] = spa
[a+b+c]
Individual fractions
[a] = env|spa
[b] = shared
[c] = spa|env
[d] = Residuals
Df
1
-
Adj. R2
0.070
-
-
0.070
0.930
All species
[a+b] = env
[b+c] = spa
[a+b+c]
Individual fractions
[a] = env|spa
[b] = shared
[c] = spa|env
[d] = Residuals
Size 0.25-0.50 cm
[a+b] = env
[b+c] = spa
[a+b+c]
Individual fractions
[a] = env|spa
[b] = shared
[c] = spa|env
[d] = Residuals
-
Variables selected
Moss, width, manganese, boulder, conductivity
V1, V3, V2, V9
Boulder, moss, manganese
V6, V3, V16
Moss, manganese, width
V3, V7, V9, V1, V2, V12
Moss, iron, width, boulder
V1
Moss
5
Table S1. Cont.
Active
[a+b] = env
[b+c] = spa
[a+b+c]
Individual fractions
[a] = env|spa
[b] = shared
[c] = spa|env
[d] = Residuals
Df
5
1
6
Adj. R2
0.309
0.044
0.343
5
0
1
0.299
0.010
0.034
0.657
Passive
[a+b] = env
[b+c] = spa
[a+b+c]
Individual fractions
[a] = env|spa
[b] = shared
[c] = spa|env
[d] = Residuals
Df
4
7
11
Adj. R2
0.141
0.172
0.286
4
0
7
0.114
0.027
0.145
0.714
Moss, width, boulder, manganese, iron
V1
Moss, manganese, boulder, width
V3,V7, V9, V1, V2, V12, V6
References
Blanchet, F.G., Legendre, P. & Borcard, D. (2008) Forward selection of explanatory
variables. Ecology, 89, 2623–2632.
Borcard, D., Gillet, F. & Legendre, P. (2011) Numerical Ecology with R. Springer, New
York.
Borcard, D. & Legendre, P. (2002) All-scale spatial analysis of ecological data by means of
principal coordinates of neighbour matrices. Ecological Modelling, 153, 51–68.
Borcard, D., Legendre, P., & Drapeau, P. (1992) Partialling out the spatial component of
ecological variation. Ecology, 73, 1045–1055.
Diniz-Filho, J. A. F. & Bini, L. M. (2005) Modelling geographical patterns in species
richness using eigenvector-based spatial filters. Global Ecology and
Biogeography, 14, 177–185.
Dray, S., Pélissier, R., Couteron, P., Fortin, M. J., Legendre, P., Peres-Neto, P. R., Bellier, E.,
Bivand, R., Blanchet, F. G., De Cáceres, M., Dufour, A. B., Heegaard, E.,
Jombart, T., Munoz, F., Oksanen, J., Thioulouse, J. & Wagner, H. H. (2012)
6
Community ecology in the age of multivariate multiscale spatial analysis.
Ecological Monographs, 82, 257-275.
Griffith, D.A. & Peres-Neto, P.R. (2006) Spatial modeling in ecology: the flexibility of
eigenfunction spatial analyses. Ecology, 87, 2603–2613.
Legendre, P., Borcard, D., Blanchet, F.G. & Dray, S. (2013) PCNM: MEM spatial
eigenfunction and principal coordinate analyses. R package version 2.1-2/r109.
http://R-Forge.R-project.org/projects/sedar/
Legendre, P., Borcard, D. & Peres-Neto, P.R. (2005) Analyzing beta diversity: partitioning
the spatial variation of community composition data. Ecological Monographs,
75, 435–450.
Legendre, P. & Gallagher, E.D. (2001) Ecologically meaningful transformations for
ordination of species data. Oecologia, 129, 271-280.
Legendre, P. & Legendre, L. (2012) Numerical Ecology. Third Edition. Elsevier, Amsterdam.
Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., Minchin, P.R., O'Hara, R.B., Simpson,
G.L., Solymos, P., Stevens, M.H.H. & Wagner, H. 2013. vegan: Community
Ecology Package. R package version 2.0-9. http://CRAN.Rproject.org/package=vegan.
Peres-Neto, P.R., Legendre, P., Dray, S. & Borcard, D. (2006) Variation partitioning of
species data matrices: estimation and comparison of fractions. Ecology, 87,
2614–2625.
Rao, C.R. (1964) The use and interpretation of principal component analysis in applied
research. Sankhyā: The Indian Journal of Statistics, Series A, 26, 329–358.
7