Data analysis
After evaluating the correlation matrix of all variables (Figure
A1 in the appendix ) we omitted from future analysis two variables that
were highly correlated (> 0.8); the ratio between
intercellular CO2/ambient CO2 that was
highly correlated with WUE and the delta of pre-dawn and midday water
potential that was highly correlated with midday water potential.
To show how paramo species are arranged, relative to one another, in the
trait space, to evaluate which trait explains most of the observed
variation in the data, and to select those traits that could be useful
in identifying different PFTs, we used the algorithm of the HCPC
(Hierarchical Clustering on Principal Components) method on the
10 first dimensions retained from a Factor analysis of mixed data (FAMD)
(metric, Euclidean; linkage, Ward), implemented in the FactoMineR
package (Version 1.41) (Kassambara, 2017). We run the FAMD to reduce the
dimension of the data into 10 orthogonal axes containing 81% of the
variability of the data. We applied FAMD because it is a principal
component method relevant to examine a data set containing both
quantitative and qualitative variables (Pagès, 2014). Quantitative and
qualitative variables are scaled during the analysis to balance the
influence of both types of variables. After that, we ran the
hierarchical clustering using the Ward’s criterion and letting the
analysis pick the optimal number of clusters between 3 to 10 clusters.
Finally, we plotted all figures using the Factoextra package (Version
1.0.7) and ggplot2 package (Version 3.1.0).
We explored how the different growth forms contrasted in the multiple
trait space using the growth form classification by Ramsay and Oxley
(1997) (Figure 1 ) and then used a simpler classification
scheme, where prostrate and upright shrubs were classified as shrubs,
erect and prostrate herbs as forbs and basal rosettes, stem rosettes and
acaulescent rosettes all as rosettes (Figure 2 ).
To examine the variation in all traits among the identified PFTs, we ran
ANOVAs and we used Tukey’s mean separations tests for variables with
normal distribution and ran a Kruskal-Wallis tests and Wilcox tests for
variables with a non-normal distribution. All analyses were done using R
Studio (Version 1.0.143 - R Development Core Team, 2009-2016).
RESULTS
In the factor analysis of mixed data (FAMD), we found that 42.7 % of
the variability could be explained by the three first principal
dimensions (Table 3 ). In the first axis of variation, we found
traits correlated to the allocation of carbon and light capture (leaf
area, SLA, C:N ratio), traits related to defense, such as leaf thickness
and toughness, and traits related to plant nutritional status (leaf N
and P content). On the right side of this axis are grouped together all
the stem and basal rosette species and few shrubs species due to their
similarity in having thicker and tougher leaves with low SLA and low N
and P content, traits which are generally associated with slower growth
rates and longer foliar life span (Figure 1 ). On the left side
of this axis are clustered all the forbs (erect and prostrate herbs),
one acaulescent rosette and some upright shrubs. These species share
having thin and soft leaves with high SLA that are rich in nitrogen and
phosphorus, traits associated with high growth rate and an acquisitive
strategy (Figure 1 and 2 ).
On the second axis of variation or dimension 2 we found traits
correlated with plant defense and water deficit strategies (LDMC, WUE,
water potential at noon, transpiration rate), seed dispersal syndrome,
the investment in support and in photosynthetic tissue (LDMC, LCC, SLA
and Amax) (Table 3 ). Clustered at the lower end
of Dim. 2 are all upright and prostrate shrub species with small, tough
expensive leaves and with high carbon content whose seeds are dispersed
by endo-zoochory with high water use efficiency, low leaf water
potential at noon but low photosynthetic rate. Clustered at the higher
end of Dim. 2 are all the erect and prostrate herbs and all rosette
species. They all have in common high SLA, lower LDMC, seeds dispersed
by wind, water or gravity, low WUE, that is, they use more water by each
unit of CO2 assimilated, high transpiration rate and
photosynthetic rate, and they reach lower leaf water potential at noon
(Figure 1 ).
Based on our HCPC analysis, we identified three PFTs (Figure 1
and 2 ) that broadly correspond with a coarser classification scheme of
the growth forms in the paramo; shrubs, forbs and rosettes. All shrubs,
either prostrate and upright shrubs, cluster together; all large
rosettes, whether stem or basal rosettes, cluster together, and all
erect, and prostrate herbs cluster together (Figure 1 and 2 ).
Considering these results, it is reasonable to suggest that we could use
a simplified classification of growth forms in the future to feed paramo
vegetation models.
The first and larger functional group PFT-1 (circle symbols inFigure 2 ) comprised all forbs and shrubs with relatively tender
leaves, belonging mainly to Asteraceae and Melastomataceae families
(Table 1 ). This group has thinner leaves, with high SLA, and
high nutrient content (nitrogen and phosphorus), probably indicating a
leaf rich in lipids, DNA and photosynthetically active enzymes, which
would account their slightly higher photosynthetic rate. This set of
species is also somewhat tolerant of water deficit with intermediate
values of leaf water potential at noon (Figure 3 ).
The second functional group PFT-2 (triangles symbols in Figure
2 ) contains all the rosettes except for one acaulescent rosette speciesAcaena cylindristachya (Table 1 ). Plants in this group
have the largest, thickest and toughest leaves of all groups
(Figure 3 ), with low SLA generally associated with slow growth
rates. Their leaves have high C:N ratios and are low in N and P content,
making them probably slow to decompose. Their leaves are also densely
pubescent, a strategy to reduce water loss and regulate temperature
transfer (avoid freezing). Species in this group have low WUE and the
highest leaf water potential (less negative) suggesting they are
avoiders of water deficit.
The third functional group PFT-3 (square symbols in Figure 2 )
is formed by only shrubs, both prostrate and upright shrubs, including
all the species from the Ericaceae family, one Melastomataceae, one
Asteraceae among others (Table 1 ). In this PFT-3, we found the
tallest shrubs with tough leaves, rich in carbon, but poor in
phosphorus, and with a high leaf dry matter content. These shrubs have
fruits with heavier seeds dispersed by endo-zoochory, low transpiration
rate and low rates of photosynthesis and the most negative values of
leaf water potential at noon, an indication that they could be more
tolerant of water deficit than the other groups (Figure 3 ).