Incorporating dominant species as proxies for biotic interactions strengthens plant
community models
Peter C. le Roux, Loïc Pellissier, Mary S. Wisz and Miska Luoto
Appendix S1. Supplementary materials
Supplementary materials and methods
Vascular plant species cover and environmental characteristics were quantified at two sites on
the Saana massif (69° N 20° E) in north-western Finland. The study sites were located on a
north- and south-facing slope, separated by 2.5 km. Both sites were above the birch (Betula
pubescence ssp. czerepanovii) treeline, at c. 700 m a.s.l. in the Saana Nature Reserve. At the
nearby Kilpisjärvi research station (< 2 km away; 480 m a.s.l.) January and July temperatures
average -13.4 and 11.0 °C, with mean annual precipitation of 458 mm (1961 – 2011; Finnish
Meteorological Institute, Finland).
The mesotopography of each quadrat was classified on a ten-point scale following Bruun et
al. (2006), with convex ridge tops assigned the maximum value and the bottom of
depressions the minimum. Soil moisture and temperature were measured in each quadrat
during the peak growing season (16 & 17 July 2012 for the north and south sites
respectively; > 24 hours after previous rainfall) using a hand-held time-domain reflectometry
sensor (FieldScout TDR 300, Spectrum Technologies, Plainfield, IL, USA; using 7.5 cm
sensor rods) and a digital temperature probe (TFX 392 SKW-T thermometer, Ebro
Electronic; Ingolstadt, Germany; 10 cm depth; see Aalto, le Roux & Luoto 2013 for details).
Maximum potential solar radiation (i.e. assuming clear sky conditions) was calculated for
each quadrat (McCune & Keon 2002), using slope and aspect values recorded in the field.
Soil samples were collected at 4 m intervals across each grid, where after soil pH was
determined in the Laboratory of Geoscience and Geography (University of Helsinki) from
air-dried soil samples following the standardized ISO 10390:1994(E) procedure. Bilinear
interpolation was subsequently used to estimate the pH within each quadrat. Rock cover (i.e.
percentage areal cover) was visually estimated in each quadrat from exposed rock.
All analyses were conducted in R statistical software (R Development Core Team 2011),
using the mgcv (Wood 2011) and gbm (Elith et al. 2008) packages to implement generalized
additive models and boosted regression trees and the QuantReg package (Koenker 2009) to
run quantile regression.
References for supplementary materials
Aalto, J., le Roux, P. C. & Luoto, M. (2013) Vegetation mediates soil temperature and
moisture in arctic-alpine environments. Arctic, Antarctic, and Alpine Research, 45, 111.
Bruun, H. H., Moen, J., Virtanen, R., Grytnes, J. A., Oksanen, L. & Angerbjörn, A. (2006)
Effects of altitude and topography on species richness of vascular plants, bryophytes
and lichens in alpine communities. Journal of Vegetation Science, 17, 37-46.
Elith, J., Leathwick, J. R. & Hastie, T. (2008) A working guide to boosted regression trees.
Journal of Animal Ecology, 77, 802-813.
Koenker, R. (2009) quantreg: Quantile Regression. Retrievable from http://CRAN.Rproject.org/package=quantreg.
McCune, B. & Keon, D. (2002) Equations for potential annual direct incident radiation and
heat load. Journal of Vegetation Science, 13, 603-606.
R Development Core Team (2011) R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna, Austria.
Wood, S. N. (2011) Fast stable restricted maximum likelihood and marginal likelihood
estimation of semiparametric generalized linear models. Journal of the Royal
Statistical Society: Series B (Statistical methodology), 73, 3 - 36.
Appendix S2. Modelling the cover of dominant species (Supplementary materials and
results)
Analyses in this study use the observed cover of three dominant plant species as predictor
variables, assuming plant cover to be a reasonable a proxy for the frequency (and therefore
also total intensity) with which these species interaction with, and impact upon, co-occurring
sub-dominant plant species. Since Betula nana, Empetrum nigrum ssp. hermaphroditum and
Junipersus communis contributed significantly to explaining patterns of community richness,
composition and functional structure at our two study sites, accurate predictions of their cover
outside of our study location would likely benefit models of arctic-alpine tundra vegetation
elsewhere in this habitat type. To determine the accuracy with which the cover of the three
species could be predicted we used a six-fold non-random cross-validation method
(implemented using generalized linear models, generalized additive models and boosted
regression trees), based on the same six abiotic predictor variables described in the main text:
mesotopography, soil temperature, soil moisture, maximum potential solar radiation, soil pH
and rock cover. Boosted regression trees provide the most accurate predictions of the cover of
these species, explaining on average 36 and 30% of the deviance in their cover in the north
and south site respectively. Generalized linear models and generalized additive models
performed much worse, explaining 11 and 10% of deviance in cover at the south site, and 2
and 2% at the north site respectively. Variable importance varied strongly between sites, but
was more consistent between methods (Fig. S7). Thus, at fine-scales the six abiotic variables
have relatively low predictive power for the cover of the dominant species. Therefore, in the
absence of additional abiotic predictor variables with which to improve models of dominant
species cover, field-quantified measurements (and not modelled estimates) of dominant
species cover appear necessary to improve community models of arctic-alpine vegetation.
Table S1. Characteristics of the two study sites, including total vascular plant cover, the
cover of the three dominant plant species, and vascular plant biomass. All values are mean ±
SE, except for soil pH where the median value is presented.
North
South
Betula nana
5.76 ± 0.25
4.00 ± 0.24
Empetrum nigrum
13.87 ± 0.66
20.57 ± 0.72
Juniperus communis
0.65 ± 0.13
5.52 ± 0.43
All vascular species
27.23 ± 0.73
43.98 ± 0.74
Biomass (grams per 0.04 m2)
9.86 ± 0.34
15.63 ± 0.52
Vascular plant species richness (1 m2)
11.19 ± 0.22
14.21 ± 0.20
Median vegetation height (cm)
4.19 ± 0.07
6.54 ± 0.13
Leaf dry matter content (mg.g-1)
300.7 ± 13.0
279.4 ± 12.8
Moisture (%)
31.29 ± 0.48
28.71 ± 0.29
Temperature (°C)
8.50 ± 0.05
8.93 ± 0.06
pH (median)
4.63 ± 0.01
5.37 ± 0.02
Rock cover (%)
9.20 ± 0.43
20.62 ± 0.78
0.26 ± 0.01
0.72 ± 0.01
Vegetation cover (%)
Soil characteristics
Potential solar radiation (MJ.cm–2.yr–1)
Figure S1. The mean (± SE) fit of simple (six abiotic predictor variables) and full (six
abiotic and three biotic predictors) models of species occurrence, measured by the area under
the curve of a receiver operating characteristic plot (AUC) and true skill statistic (TSS), for
both the northern (N; n = 43 species) and southern (S; n = 54 species) study sites. Three
species distribution modelling techniques were implemented: generalized linear models
(GLM), generalized additive models (GAM), and boosted regression trees (BRT).
Significance of improvement assessed using one-tailed paired t-tests.
Figure S2. Variable importance (%; mean ± SE) for the six abiotic predictor and the three
biotic predictor variable when modelling the occurrence of every sub-ordinated species, as
determined from the full models in both the north (n = 43 species) and south (n = 54) study
sites. Three statistical techniques were implemented: generalized linear models (GLM),
generalized additive models (GAM), and boosted regression trees (BRT). Variable
importance was calculated for GLMs and GAMs from each variable’s drop contribution (i.e.
change in deviance associated with exclusion of that variable from a model containing all the
other predictors), while the method of Friedman (2001) was used for BRTs. Variables’
contributions were scaled to sum to 100, with higher values indicating stronger influence on
the response variable.
Figure S3. Loess smooth (± 95 % CI) fitted to the relationship between community-weighted
mean leaf dry matter content (observed and predicted) and the combined cover of the three
dominant species (Betula nana + Empetrum nigrum ssp. hermaphroditum + Juniperus
communis) in the north and south site. Three statistical methods were used; GLM =
generalized linear models, GAM = generalized additive models, BRT = boosted regression
trees. The “simple” predictions are models using only abiotic predictor variables, while the
“full” model comprised both abiotic and biotic predictor variables. Observed and predicted
species richness exclude occurrences of the three dominant plant species.
Figure S4. Relationship between community-weighted mean leaf dry matter content and the
cover of the three dominant plant species (assumed here to represent a gradient of increasing
competitive pressure). Dashed lines represent results from quantile regression performed on
the upper and lower 20th percentiles of the data.
Figure S5. Variable importance when directly modelling sub-ordinate species richness and
community-weighted mean leaf dry matter content (based on full models comprising six
abiotic and three biotic predictor variables) for both the northern and southern sites.
Figure S6. Histogram of leaf dry matter content (LDMC) of all modelled species for which
data were available. The LDMC values for the three dominant species (Empetrum nigrum
ssp. hermaphroditum, Betula nana and Juniperus communis) are indicated with arrows, and
rank 22nd, sixth and second respectively when compared to the values of the modelled subdominant species.
Figure S7. Mean (and maximum) relative importance of six abiotic predictor variables when
modelling the cover of the three dominant species (Betula nana, Empetrum nigrum ssp.
hermaphroditum and Juniperus communis). GLM = generalized linear model, GAM =
generalized additive model, BRT = boosted regression trees, Mesotop. = mesotopography,
Moisture = soil moisture, Temperature = soil temperature, Radiation = maximum potential
solar radiation, pH = soil pH, Rock = rock cover.