ELECTRONIC SUPPLEMENTARY MATERIAL (ESM)

OECOLOGIA

Trait assembly in grasslands depends on habitat history and spatial scale

Liina Saar1*, Francesco de Bello2, Meelis Pärtel1, Aveliina Helm1

1. Institute of Ecology and Earth Sciences, University of Tartu, Lai 40, Tartu, 51005, Estonia

2. Institute of Botany, Czech Academy of Sciences, Dukelská 135, 379 82 Trebon, Department

of Botany University of South Bohemia, Na Zlate stoce 1, 370 05 Ceske Budejovice, Czech

Republic

* Correspondence: Liina Saar, Institute of Ecology and Earth Sciences, University of Tartu, Lai

40, Tartu, 51005, Estonia

E-mail: [email protected]

Phone: +372 533 00787

1 Figure S1 Location of study sites (dots) in Saaremaa (larger) and Muhu (smaller) islands (n = 35)

2 Figure S2 a) Design and b) depiction of scales used in the current study: a) local community (~ 50 m radius habitat area) with a 2×2 meter sample plot located in the centre of local community.

Numbers inside the habitat area or in the square indicate average species number recorded across the 35 local communities or 35 sample plots of a given habitat respectively. b) The habitat species 3 pool (cumulative set of all species for a given habitat across 35 grassland complexes). Numbers indicate shared species between species pools. Numbers in parentheses indicate total species number recorded in 35 sample plots or 35 habitat species pools for given habitats.

4 Table S1 Correlations between analysed traits. Continuous variables were related by using Pearson

product–moment correlation (r, with n = number or pairs), categorical variables were related by

Chi2 test (with d.f. degrees of freedom), and continuous variables were related to categorical

variables by one-way ANOVA (F-statistic with d.f. Statistical significance is given as ***P < 0.001,

**P = 0.001–0.01, *P = 0.01–0.05. Statistically significant tests are marked in bold

Insect Mean plant Mean seed Short life Specific Terminal Vegetative Name of trait pollinatio height^ weight^ span leaf area velocity growth n

Mean plant height^ F1,189= 1.78 Mean seed weight^ F1,210= r=0.13 5.46* n=174 2 Short life span Χ =2.0 F1,197=0.37 F1,216=0.31 3 d.f.=1 *** Specific leaf area F1,160= r=0.02 r=0.01 F1,166=12.5 1.16 n=135 n=156 *** Terminal velocity F1,125= r=-0.06 n=104 r=0.58 F1,126=1.66 r=0.04 0.01 n=123 n=105 2 2 Vegetative growth Χ =0.0 F1,197=0.32 F1,216=1.44 Χ =0.18 F1,166=0.27 F1,126=1.49 5 d.f.=1 d.f.=1 2 *** 2 *** 2 Wind dispersal Χ =5.5 F1,135=0.45 F1,155=16.2 Χ =1.40 F1,113=0.41 F1,102=28.6 Χ =0.48 1* d.f.=1 d.f.=1 d.f.=1 ^Log-transformed variables

5 Table S2 Trait values availability minimally and on average per site

Trait name Trait values available per Trait values available per site site minimally (%) on average (%) Dispersal mechanism 61.2 77.9 Life span 100.0 100.0 Main pollen vector 82.6 96.9 Mean plant height 57.7 77.3 Mean seed weight 86.0 96.9 Mode of reproduction 100.0 100.0 Specific leaf area 56.5 76.2 Terminal velocity 44.0 68.0

6 Table S3 Divergence (diver.) and convergence (conv.) patterns in grasslands with different land- use history at two nested spatial scales analysed over individual traits. Degrees of freedom for all models are 34. Statistically significant patterns are marked in bold

Historical grassland Former grassland Developing grassland Pattern t Pa t Pattern t Trait name Scale tte rn Dispersal Local community –> -0.5 co -3.2 -3.6 mechanism plot-scale n v. random P conv. P = 0.32 = P = 0.0005 0. 00 14 Habitat species pool –> -1.2 co -2.1 -9.9 local community n v. random P conv. P = 0.12 = P < 0.0001 0. 01 83 Life span Local community –> -1.2 co -3.1 -2.7 plot-scale n v. random P conv. P = 0.13 = P = 0.0047 0. 00 19 Habitat species pool –> -3.4 co -7.7 -7.3 local community n v. conv. P conv. P = 0.0008 < P < 0.0001 0. 00 01 Main pollen Local community –> diver. 5.6 di 3.3 random 1.4 vector plot-scale P < 0.0001 ve P = 0.0803 r. P = 0.

7 00 12 Habitat species pool –> 1.0 co -2.2 4.6 local community n v. random P diver. P = 0.16 = P < 0.0001 0. 02 03 Mean plant Local community –> 0.9 di 3.7 -0.2 height plot-scale ve r. random P random P = 0.81 = P = 0.88 0. 00 04 Habitat species pool –> 5.7 di 3.1 –6.5 local community ve r. diver. P conv. P < 0.0001 = P < 0.0001 0. 00 20 Mean seed Local community –> -4.8 co -2.8 -1.4 weight plot-scale n v. conv. P random P < 0.0001 = P = 0.0813 0. 00 41 Habitat species pool –> -7.9 co -2.5 -4.7 local community n v. conv. P conv. P < 0.0001 = P < 0.0001 0. 00 99 Mode of Local community –> diver. 2.9 ra -0.1 conv. -1.7 reproduction plot-scale P = 0.0035 nd P = 0.0449 o m

8 P = 0. 49 Habitat species pool –> -6.3 co -3.6 -1.8 local community n v. conv. P conv. P < 0.0001 = P = 0.0422 0. 00 04 Specific leaf Local community –> -0.7 co -3.7 -1.7 area plot-scale n v. random P random P = 0.25 = P = 0.0507 0. 00 03 Habitat species pool –> -12.8 co -11.1 -9.3 local community n v. conv. P conv. P < 0.0001 < P < 0.0001 0. 00 01 Terminal Local community –> -3.6 co -3.9 -3.9 velocity plot-scale n v. conv. P conv. P = 0.0005 = P = 0.0002 0. 00 02 Habitat species pool –> 4.8 co -6.3 -9.2 local community n v. conv. P conv. P < 0.0001 < P < 0.0001 0. 00 01 All traits Local community –> diver. 2.0 co -3.2 conv. -4.7 plot-scale P = 0.0268 n P < 0.0001

9 v. P = 0. 00 13 Habitat species pool –> -5.4 co -9.9 -10.6 local community n v. conv. P conv. P < 0.0001 < P < 0.0001 0. 00 01

10 S4 Testing the null-model approach as alternative statistical method

To assess whether the use of different methods would lead to different results, we also applied null-model approach, a method that has a long tradition in testing patterns of species assembly, but at the same time has shown to have a lot of methodological constraints (de Bello 2012). We compiled simulated null communities via randomizations across each scale by maintaining the species richness. Functional diversity was calculated by compiling eight traits together: dispersal mechanism, life span, main pollen vector, mean plant height, mean seed weight, mode of reproduction, specific leaf area, terminal velocity of diaspores. By using simulated null communities for each grassland type and for each scale, we calculated the standardized effect sizes (SES = (observed functional diversity – mean of simulated functional diversity)/standard deviation of simulated functional diversity)). t-test was used to test whether SES values are significantly different from zero. SES > 0 suggests trait divergence and SES < 0 suggests trait convergence. We found that SES was bigger than zero for historical grasslands and lower than zero for former and developing grasslands, indicating that stable grasslands are divergent and the former and developing grasslands convergent (Fig. S4b, see Fig. 1a in the main text for comparison). There was also strong relationship between SES resulting from null models and effect size resulting from the use of mean pairwise trait dissimilarity (Effect size (mpdplot- mpdcommunity), Fig. S4a). To conclude, use of null-models leads to same results as the use of functional species pool method.

11 ) y t i n u

m a b m 2 o 0 c 1 .

0 d s t

R-square=0.94 i p a 0 r m t -

t l l o l 2 a

1 p 0 - . S d 0 - E p S 2 - m (

e 6 z 3 i - 0 . s

t 0 - c

e -3 -2 -1 0 1 Historical Former Developing f f E SES all traits

Figure S4 (a) Comparison of the effect size obtained from mean pairwise trait dissimilarity

(MPD) differences in finer scale (MPD at the 2×2 m plot-scale minus MPD at the local community scale) with the standardized effect size (SES) resulting from null-model approach. (b)

Standardized effect size (SES) for 2×2 m plots of grasslands with different development histories.

For each grassland type, effect sizes significantly differ from zero (P < 0.05), SES > 0 suggests trait divergence and SES < 0 suggests trait convergence. Analyses were carried out by including all traits (dispersal mechanism, life span, main pollen vector, mean plant height, mean seed weight, mode of reproduction, specific leaf area, terminal velocity of diaspores). In box plots, the median for each data set is indicated by the heavy center line, and the first and third quartiles are the lower and upper edges of each box, which is known as the interquartile range (IQR). The extreme values (within 1.5 times the IQR from the upper or lower quartile) are the ends of the lines extending from the IQR. Points at a greater distance from the median than 1.5 times the IQR are plotted individually.

Reference for S4

12 de Bello F (2012) The quest for trait convergence and divergence in community assembly: are

null‐models the magic wand? Glob Ecol Biogeogr 21:312–317

13