Ecological Archives E096-047-A1
Katie E. Marshall and Jennifer L. Baltzer. 2015. Decreased competitive interaction drive a reverse species richness latitudinal gradient in subarctic forests. Ecology 96:461–470. http://dx.doi.org/10.1890/14-0717.1
Appendix A. Additional statistical methods used to generate forest type classification, PC scores for climate and tree stand parameters, and geographic and community distances.
A forest classification was generated from Ward hierarchical clustering of the Hellinger-transformed percent cover of all tree species in each of these plots using R (R Core Team 2013). Five forest types were identified by the dominant species in each type (summarized in Table C1) including: white spruce (Picea glauca), black spruce (Picea mariana), black spruce/larch (Picea mariana/Larix laricina), white spruce/black spruce (Picea glauca/Picea mariana), and mixed (Picea glauca, Pinus banksiana, Betula neoalaskiana). Two additional forest types for sites with no living trees present were added (either sites with no trees > 9 cm dbh or sites with all standing dead trees), for a total of seven forest types.
Two sets of climate data were examined. The first was a data set describing the mean yearly climate conditions for each PSP for the 20 year period between 1982 and 2002 (when the first vegetation plots were catalogued). The second was a data set that included climate data for each PSP for the year immediately prior to the year the vegetation for that plot was catalogued. A principal components analysis using the prcomp() function with scaled and centered data in R (R Core Team, 2013) was used to extract components that described the variation in the climate data set. Parameters that did not load with an absolute value > 0.25 on the first four principle components were removed, and the PCA was re-run. All principal components with eigenvalues > 1.5 (two in the mean climate dataset, three in the year before vegetation catalogue dataset) were used to generate principal component scores for each site (PC loadings in Tables D1-2). Tree stand characteristics were also simplified using principal components analysis after square-root transforming all stand characteristics, except for site age, to improve normality (PC loadings in Table E1). Climate PC scores were transformed by multiplying by -1 to ensure that increasing positive values corresponded to increasing temperature and seasonality, while tree stand characteristics principal component scores were transformed to ensure that increasing positive values corresponded to increased quantity of standing wood and increased number of living trees.
Geographic and community distance
Geographic distance between points was examined by first splitting the species data set into a "northern" and "southern" set at 64 degrees latitude. Geographic Euclidean distance between each site within the northern and southern sets was calculated by the dist function in the base package of R from UTM coordinates for each site. Euclidean distance between sites based on Hellinger-transformed species percent cover was also calculated for northern and southern sites. To examine distance-decay in community composition, a Mantel test with 999 permutations for both the northern and southern sites was conducted using the community-composition distances and geographical distances using the mantel function in the vegan package of R (Oksanen et al. 2013).
Community structure
Community structure was investigated using scores generated from nonmetric multidimensional scaling (Methods). Differences in MDS score between northern and southern sites were compared using ANOVA, while differences in variation in MDS score were investigated using Fisher tests in R (var.test function). Distances between MDS scores were also investigated using the distance function as described above.
Literature Cited
Oksanen, J., Blanchet, F. G., Kindt, R., Legendre, P., Minchin, P. R., O'Hara, R. B., … Wagner, H. 2013. vegan: Community Ecology Package. Retrieved from http://cran.r-project.org/package=vegan
R Core Team. (2013). R: A language and environment for statistical computing. Vienna, Austria. Retrieved from http://www.r-project.org/