Ecological Archives E096-255-A1

Geneviève Lajoie and Mark Vellend. 2015. Understanding context dependence in the contribution of intraspecific variation to community trait–environment matching. Ecology 96:2912–2922.

Appendix A. Additional methodological details on environmental data collection, reducing the dimensionality of environmental axes, and trait measurement.

1.   Environmental data collection

Elevation was determined using a digital elevation model. Daily air temperature 0.5 m above ground was recorded with Thermocron IButtons (Maxim Integrated) along two of the transects. For each plot, we calculated the average temperature across one year, from November 2012 to October 2013. We estimated slope using a clinometer, by averaging measurements taken 4 m upslope and 4 m downslope at each end of the rectangular plot. We characterized percent canopy openness by taking canopy photos (with Nikon Coolpix 5000 camera mounted by a Nikon Fisheye Converter FC-E8 lens) above the herbaceous cover every 4 m across the length of the plot at full canopy closure, which we then analyzed using Gap Light Analyzer (Frazer et al. 1999). Soil depth was measured and averaged from 14 sampling points on the periphery of each plot using a soil probe. At each of these points, we also took 15 cm-deep soil cores, which were pooled per site for further characterization of nine soil physico-chemical characteristics, seven of which were measured (extractable NH4+-N concentration, extractable NO3--N concentration, extractable phosphate-P concentration, percent organic matter, pH, total soil carbon and total soil nitrogen) and two derived from the first seven (C:N ratio and total inorganic nitrogen concentration). Analyses were performed in the Université de Sherbrooke Soils Analysis Laboratory.

Upon collection, aliquots of fresh soil were extracted in 50 mL 1 N aqueous KCl on a reciprocating shaker for one hour. The suspension was then gravity-filtered through Whatman no.5 filter papers. Extractable NH4+-N and NO3--N concentrations of the resulting extracts were determined via colorimetry with a Technicon Autoanalyser II system, using the indophenol blue (alkaline salicylate-nitroprusside-hypochlorite) and Griess-Ilosvay (Cd reduction-NED-sulphanilamide) reactions, respectively. Sample absorbances were compared against those of standard curves that had been prepared for NH4SO4 and KNO3. Extractable inorganic-N concentrations (mg/L) were then corrected to an oven-dry soil mass-basis (60°C for 48 h).

Remaining soil was then air-dried and sieved with a 5 mm sieve. Aliquots were extracted in 30 mL Mehlich 3 reagent, after which the suspensions were agitated on a reciprocating shaker for 5 minutes and filtered through Whatman no.5 filter paper. The phosphate-P concentrations of the extracts were then determined by manual colorimetry using Murphy-Riley reagent. Sample absorbances (882 nm) were read after 10 minutes of color development and compared against those of standard solutions (0, 2, 4, 6, 8, 10 ppm phosphorus as KH2PO4). Organic matter content of air-dried soil was estimated via loss on ignition, (400°C for 12 h). Samples were weighed before and after combustion, and organic matter mass determined by difference of the two weights, standardized by dividing by original pre-combustion mass. Soil subsamples were then sieved with a 2-mm sieve and pH determined using a ThermoScientific Orion Star A211 pH meter. For estimating total soil carbon and nitrogen, aliquots of 100–150 mg of dry soil were grounded and weighed to the nearest 0.1 mg. These samples were then flash-combusted at 960°C in a Vario Macro multi-element analyser (Elementar analysensysteme GMbH, Haan, Germany), after which the dried and reduced (830°C) combustion gases were passed through a thermoconductivity detector for sequential determination of total N (mg N/g) and total C (mg C/g).

The two additional variables derived from the above were calculated as follows: C:N ratio as the ratio between total soil carbon and nitrogen; and total inorganic nitrogen concentration as the sum of extractable NH4+-N and NO3-N concentrations.

2.   Reducing the dimensionality of the environmental data

In order to reduce multicollinearity among the environmental predictor variables, we calculated variance inflation factors for each variable and eliminated those with the highest scores until none was higher than 4. To further simplify the environmental dataset, we then performed a PCA of plots based on the remaining environmental characteristics, and we extracted site scores along the axis that was the most strongly correlated with variables characterizing site acidity (pH, phosphorus concentration and percent organic matter). Environmental variables included in the final analyses were elevation, slope, canopy openness, total inorganic nitrogen concentration (extractable NH4+ + NO3-) and the acidity axis.

3.   Trait measurement

In sampling plants for SLA and height measurement, reproductive individuals were excluded to limit differences in physiological status of the sampled individuals across species and across sites within species, since for many species reproductive individuals were often not present at a given site. To reduce the impact of ontological variation due to differences in the timing of vegetative development across elevations, individuals were sampled at each site shortly after the peak flowering date of the species at that site. Nonreproductive fronds or shoots were selected from spore-bearing pteridophytes, entirely nonreproductive individuals being rare. Individuals of summer-flowering species and pteridophytes were sampled in 2012 while ephemerals were sampled in 2013.

For a small proportion of species-plot combinations, we could not estimate trait values either due to the small size of the population or to the poor condition of the individuals present. In those instances, we interpolated missing data via multiple regressions using the environmental variables as predictors. We predicted missing plot-level trait values for the very few species-plot combinations where less than 3 observations were locally recorded for a given trait using multiple regressions (Legendre and Legendre 2012), as described in the next paragraph. For SLA and height, this was done for 9/567 and 14/567 species-plot combinations, respectively. There was no missing value in the peak flowering date data set.

For each species missing data for a given trait in a given plot, and for which we had trait values for at least 5 other plots, we built a multiple regression model with the trait as the dependent variable, and a reduced set of environmental variables as predictors. The reduced set of variables was selected upon visual examination of individual trait-environment scatterplots, such that only those variables that were most related with trait variation would be included in the starting model. We thereby ensured that the number of predictors did not exceed the number of plots available for estimating the trait-environment relationship. We did not use the reduced set of non-collinear variables presented in the methods above because our objective was to maximise model fit, not to assess independent relationships with different variables (Legendre and Legendre 2012). From this starting model, we then performed a stepwise selection procedure using function step in {stats} package in R to further reduce the number of predictors. We used the final model parameter estimates to predict missing trait values, using the environmental characteristics measured at each plot for which the data was missing. If the predicted value was far outside the range of the observed data, we used the species trait mean as the trait value for this plot. We discarded two rare species from the data set because observations were missing for more than 20% of the sites where they occurred.

Literature cited

Legendre, P., and L. F. J. Legendre. 2012. Numerical Ecology. Elsevier Science.

[Back to E096-255]