1 Universidad Francisco de Miranda, Coro, Venezuela. 2 Centro Agronómico Tropical de Investigación y Enseñanza, CATIE, Costa Rica. 3 Universidad ...
Discriminating community assembly processes through functional diversity indices Laura Pla1, Fernando Casanoves2, María del Carmen Romero3 and Julio Di Rienzo4 1 Universidad Francisco de Miranda, Coro, Venezuela 2 Centro Agronómico Tropical de Investigación y Enseñanza, CATIE, Costa Rica
3 Universidad Nacional del Centro de la Provincia de Buenos Aires, Argentina 4 Facultad de Ciencias Agropecuarias, UNC, Córdoba, Argentina
OUR CONTRIBUTION
RESULTS
Compare the ability of functional diversity indices to detect patterns of traits values in a community: 1. Is a functional diversity index able to reflect the differences among dispersion patterns? 2. Does the ability of the index depends on richness? 3. Is there an optimum subset of functional diversity indices that best discriminate among dispersion patterns?
As other authors have shown (Mouchet et al 2010), functional diversity increases when assembly process goes from environmental filtering to neutral and from them to limiting similarity (Fig. 1). + richness
Aims
Why is this important? Functional diversity measures have been used to quantify functional diversity assuming they are able to reflect different aspects or facets of functional diversity in the community. All functional diversity measures are based on functional traits evaluated at individual level for each species and some of them include information about how abundant the species are. Besides statistical properties of indices, its ability to distinguish among species coexistence processes and community assembly rules have received relatively little attention (Mouchet et al. 2010).
+ functional diversity
Figure 1: Biplot for functional diversity indices (variables represented with vectors) and simulated samples (points not shown). Ellipses of prediction for 95% confidence are supper impose with continuous lines for limiting similarity (red), neutral (green) and environmental filtering (blue), and for different richness from 20 species (light green) to 200 species (dark green) with dotted lines.
HOW WE DID IT? We organized a simulation to generate three patterns of trait dispersion: even dispersion, clustered dispersion and random dispersion. We used three assembly processes to generate the patterns from a theoretical species pool (Kraft et al. 2007, Mouchet et al. 2010) assuming that traits are phylogenetically conserved: limiting similarity (MacArthur and Levins 1967) to produce even dispersion (Stubbs and Wilson 2004), habitat filtering (Zobel 1997) to produce clustered dispersion (Peres-Neto 2004) and neutral assembly (Gotelli and Graves 1996) to produce random dispersion. We used lognormal abundance distribution to model relative abundance of species.
The species pool was generated as a set of 400 species with five traits. The trait values were randomly generate using a multivariate normal distribution with cero mean and identity covariance matrix (function rmvnorm in R library mvtnorm). We explore different sample sizes (100, 250, 500 and 1000) to select the minimum number of replications that produces a CV less than 10% for all indices. We determined that 250 replications were enough to make stable estimates within the richness range of our simulation.
We estimated three indices based on trait values without abundance loading (FD: functional diversity, MFAD: modified functional attribute diversity and FRic: functional richness) and five indices with abundance loading (wFD: weighted functional diversity, Rao: Q-entropy, FDis: functional dispersion, FDiv: functional divergence and FEve: functional evenness). We use FDiversity software (Casanoves et al. 2011).
Functional diversity indices
There are more than one subset of indices to choose in order to detect the main assembly process driven the community expression (Table 1).
Confidence intervals are not enough to conclude about ability of indices to detect main assembly processes. All functional diversity indices that include the relative contribution of each species do not show a tendency to increase with richness (see example in Fig.2). On the other hand indices that do not include abundance increase linearly (FD and MFAD) or nonlinearly (Fric) with richness (see example in Fig. 3). 18
Table 1: Classification error rate using discriminant rules
Limiting similarity Neutral Environmental filtering
15
Rao (mean, 2.5-97.5 EPD)
Simulation
Assembly process
12
9
6
3
0 0 10 20 30 40 50 60
80
100
150
200
Number of species in the community
Statistical analysis
Figure 2: Confidence intervals for Rao index. Similar shapes were obtained with wFD, and FDis. Intervals were estimated with the 250 samples by richness using the empirical probability distribution (EPD).
To have a first approximation of the relationships among functional diversity indices, assembly process and richness we performed a principal component analysis using the values of the set of indices (8 indices by 12 richness by 3 assembly rules by 250 simulated samples).
300 Limiting similarity Neutral Environmental filtering
Answers to the first and second question can be explored together. We assume that environmental filtering produces communities that, in average, are significantly less diverse than those obtained with no restriction from the same species pool. On the other hand, limiting similarity process produces more diverse communities than those obtained without restrictions. We compare confidence intervals (CI) estimated for the three assembly processes. For each index, we have plotted the mean, the lower band and the upper band of empirical probability (2.5%-97.5%). Mass ratio hypothesis (Grime 1998) states that ecosystem functional properties are related to most abundant species, so we also have plotted CI for functional indices calculated only with the species needed to recover at least the 90% of abundance.
FD (mean, 2.5-97.5 EPD)
Confidence intervals 225
150
75
0 0 10 20 30 40 50
100
150
200
For those indices that detect changes in assembly processes, its ability increases as species richness increases. The classification errors tend to cero with richness above 80 species (results not shown).
Number of species in the community
Discriminant analysis To answer the third question we used step-wise linear discriminant analysis using the set of indices for each richness. The results obtained with 25 samples were randomly selected to define the discriminant function. The rest were used to calculate the classification error rate (Table 1).
REFERENCES
Figure 3: Confidence intervals for FD index. Similar shapes were obtained with MFAD, and FRic. Intervals were estimated with the 250 samples by richness using the empirical probability distribution (EPD).
CONCLUSION
Casanoves et al. (2011) Meth Ecol Evol 2: 233-237.
McArthur and Levins (1967) Amer. Natur. 101: 377-385
Gotelli and Graves (1996) Null models in ecology. Smithsonian Institution, Washington.
Mouchet et al. (2010) Func Ecol 24: 867–876.
Grime (1998) J Ecol 86: 902-910.
Stubbs and Wilson 2004 J. Ecol 92: 557-567.
Kraft et al. (2007) Amer Natur 170: 271-283.
Zobel (1997) Trend Ecol Evol 12: 266-269.
Peres-Neto (2004) Oikos 93: 110-120.
Acknowledgments To Inter-American Institute for Global Change Research, IAI-CRN 2015 (supported by NSF, Grant GEO-0452325), and to Núcleo DiverSus, supported by CONICET, FONCYT, Universidad Nacional de Córdoba, DIVERSITAS and the IGBPGlobal Land Project
We may conclude that if the aim is to detect possible differences in assembly processes, indices based on trait and abundance values perform better than those that do not include estimation of species contribution to the community. One way to avoid field evaluation of abundance is to include trait for all the species, even for rare species. These results may help ecological scientists to plan field surveys and to decide how to applied mass ratio hypothesis.
