Plant species loss decreases arthropod diversity and shifts trophic structure
Nick M. Haddad, Gregory M. Crutsinger, Kevin Gross, John Haarstad, Johannes M.H. Knops, and David Tilman
Supplementary Material
Additional analyses within years
We conducted many additional, detailed analyses of arthropod species richness and abundance that are not reported in the main text. Herbivore species richness was significantly, positively related to plant species richness and to plant biomass, and the strength of this relationship varied across years (Table S1). Herbivore species richness increased most strongly in response to plant Shannon-Wiener index, and was reduced in the presence of C4 grasses (Table S2). No treatment variable significantly predicted herbivore abundance (Table S3), but plant composition did have significant effects, with forbs increasing herbivore abundances and C4 grasses decreasing herbivore abundances (Table S4).
Predator species richness increased as plant species richness and plant biomass increased (Table S5), and was unaffected by plant composition. Predator abundance also increased as plant biomass increased, and increased in the presence of C4 grasses (Table S6, S7).
To test whether our results within years were caused by differences in species detection across plots, we estimated species richness using Chao1 species richness estimator. Results of analyses of Chao1 estimated richness were consistent with results of analyses of observed species richness at each trophic level, and showed similarly weak responses. Thus, our results could not be attributed to artifacts of the way we sampled plots.
Cumulative results across all years
Cumulative herbivore species richness increased with higher plant species richness and plant biomass (Table S8). The stronger increase in cumulative herbivore species richness as compared with annual species richness was determined by higher numbers of species observed once or twice in diverse plots (i.e. singletons and doubletons) (Figure S2). Although there was no effect of the number of plant functional groups, herbivore species richness was higher when C3 grasses and legumes were present, and was better predicted by plant Shannon-Wiener index than by plant species richness (Table S9). None of the main treatment variables predicted herbivore abundances, but abundances were lower when C4 grasses were present, and higher when forbs were present (Table S10). We corrected herbivore richness for differences in herbivore abundances using rarefaction (using Primer v6 software), and we rarefied to 122 individuals per plot. Rarefied herbivore richness was positively related to plant species richness, the number of plant functional groups, and plant biomass (Table S11), and increased in the presence of C3 grasses, C4 grasses, and legumes. Analysis of cumulative estimated species richness (Chao1) for herbivores was consistent with results using observed richness. For predators, there was no relationship between plant species richness and Chao1, supporting our conclusion that the number of individuals was driving observed patterns in predator species richness.
As with herbivores, cumulative predator species richness increased as plant species richness and plant biomass increased (Table S12). The number of singleton and doubleton species also increased with plant species richness, but saturated more quickly than with herbivores (Figure S2). Plant Shannon-Wiener index was more strongly, positively related to predator diversity than plant species richness (Table S13). Predator abundance increased as plant biomass increased, and in the presence of C3 and C4 grasses (Table S14, 15). We rarefied predator richness to 32 individuals per plot. Only one factor, the presence of C4 grasses, was significant in reducing rarefied predator diversity.
Cumulative detritivore species richness was significantly, positively related to plant species richness and plant biomass, and significantly negatively related to plant functional group richness (Table S16). Specifically, cumulative detritivore species richness was positively related to the presence of C3 plants (Table S17). Detritivore abundance was positively related to plant biomass and to the presence of C3 grasses and C4 grasses (Table S18, S19). Because of low species richness and abundance, we did not compute rarefied detritivore richness.
Additivity analysis
To test whether we could predict arthropod species richness in plant mixtures based on arthropod species richness in component monocultures, we used the following additivity analysis that calculates the expected arthropod species richness in each polyculture under a collection of assumptions. To state these assumptions, let be the number of arthropods of species l occurring on plant species k in year t of the jth experimental replicate of plant species richness treatment i. First, we assume that the’s are Poisson distributed random variables with mean . Here, is a Poisson rate (in units of number of individuals per plot), and is the proportion of plot ij planted in plant species k (that is, = 1/(number of species planted in plot ij) . Second, the “additivity assumption” assumes that the rate at which arthropod species l occurs on plant species k in year t is constant across all experimental plots, e.g., . In the polycultures, the ’s are not observable. Instead, we are only able to observe , the total number of individuals of arthropods of species l in year t and plot ij. Thus, we also assume that and are conditionally independent for . Thus, the “additivity” model might be more precisely but verbosely described as the “additive independent Poisson” model.
To calculate the expected additive arthropod richness for a given plot, we followed these steps. First, we used the monocultures to estimate the rates . Then, for each polyculture, we used rate estimates to calculate the expected abundance of arthropod species l as . We then calculated the probability that arthropod species l was found in the plot by using one minus the zero term of the probability mass function of a Poisson distribution, 1 – exp(-E[Xijlt]). Finally, we calculated the total expected arthropod species richness by summing these probabilities over all arthropod species l. (Plugging in the expression for , dropping the i, j and t subscripts, and taking the sum over l gives eq. [1] in the main text.) If multiple monocultures were present for one or more of the plant species planted in the polyculture, then we repeated this procedure for each unique combination of monocultures, and averaged the resulting estimates of expected richness. Note that our analysis considers only species planted as part of the experiment, and does not consider unwanted species within experimental plots.
An important but subtle point is that the expected additive richness values calculated in this way are not directly comparable to the observed richness values. (To see this, suppose we observed 5 individuals of arthropod species Z in a monoculture of plant species Y, and we estimated the expected additive arthropod richness for this monoculture by using the same monoculture itself as a reference. The expected additive richness for the monoculture would be , which is less than the observed richness of 1). Thus, to make observed richness comparable to expected richness under the additive model, we applied the algorithm to each polyculture as well, using the observed arthropod abundance in each polyculture to estimate species-specific rates of occurrence. We used the adjusted observed richness to calculate the difference between observed and expected additive richness on a percentage basis.
On a small number of occasions, there were no monoculture data available for a given plant species in a given year. (For example, the lone monoculture of Panicum virgatum was not sampled for arthropods in 1996, 1997, or 1999.) In these cases, we examined the actual composition of all polycultures containing that plant species for that year (as opposed to the planted composition), and removed the polyculture from the analysis if the plant species in question made up more than 1% of the actual biomass or percent cover. If the plant species was less than 1% of the biomass or percent cover of the polyculture, we retained the polyculture in the analysis, set for the plant species without a monoculture, and inflated the remaining ’s equally so that they summed to 1.
Cumulative additivity analyses simply aggregated the observed arthropod abundances observed in each plot over all years. These cumulative analyses also required that we assume conditional independence among years. That is, we assumed that and were conditionally independent given the rates and for . Finally, rarefied additivity analyses rescaled Poisson rates by a factor equal to the ratio of the observed total arthropod abundance in a polyculture to the expected total arthropod abundance under the additivity model. This re-scaling controlled for changes in arthropod abundance by equating the expected arthropod abundance in the rarefied analysis with the actual arthropod abundance observed in each polyculture.
Spatial analysis
Because arthropods were not controlled in our experiments and are mobile, spatial factors could be important in controlling responses to our treatments. We have observed previously in this experiment in one year (1997) that the diversity of arthropods in the four plots bordering a focal plot is positively related to the diversity of arthropods in the focal plot. This neighborhood diversity explains about 10% of variation in arthropod species richness within plots, does not remove the effects of local factors, and does not extend beyond the four bordering plots to larger distances (Haddad, et al. 2001).
Unlike in 1997, we did not have data on arthropods in all neighboring plots (see methods). Because of this, we analyzed arthropod responses based on correlations with plant species richness, and developed measures of the average plant species richness, both planted and actual, in neighboring plots. We found no consistent relationship between the plant diversity in neighboring plots and arthropod species richness in focal plots in annual analyses. In analyses of cumulative data, the diversity of plants in neighboring plots had no effect on herbivore species richness, but was significantly, positively related to predator species richness. When plant species richness of neighboring plots was included in a full model for cumulative predator species richness with plant species richness, plant functional group richness, and plant biomass of the focal plot, model results were similar as in Table S12 but with the spatial variable also significant. Although we expect we would find stronger relationships if we had data on arthropods for all plots, we have no reason to believe that spatial effects obscure our responses to treatments variables.
Table S1. Repeated measures analysis, using a General Linear Model, of effects of plant species richness, plant functional group richness, and plant biomass on herbivore species richness within years.
Ln (plant species richness) / 1 / 359 / 7.13 / 0.008
# Plant functional groups / 1 / 98 / 1.94 / 0.166
Mean plant biomass / 1 / 393 / 7.81 / 0.006
Error / 148 / 7253
Within-subject effects
Year / 10 / 1406 / 10.32 / 0.001
Year * Ln (plant species richness) / 10 / 177 / 1.30 / 0.224
Year * # Plant Functional groups / 10 / 192 / 1.41 / 0.168
Year * Mean plant biomass / 10 / 461 / 3.39 / 0.001
Error (year) / 1480 / 20150
Table S2. Repeated measures analysis, using a General Linear Model, of effects of plant diversity, plant functional group composition, and plant biomass on herbivore species richness within years.
Source / df / Type III Sum of Squares / F / PMean plant Shannon-Wiener / 1 / 512 / 14.38 / 0.001
Presence C3 grasses / 1 / 60 / 1.67 / 0.198
Presence C4 grasses / 1 / 235 / 6.60 / 0.011
Presence forbs / 1 / 3 / 0.09 / 0.766
Presence legumes / 1 / 90 / 2.52 / 0.114
Mean plant biomass / 1 / 40 / 1.12 / 0.291
Error / 145 / 5161
Within-subject effects
Year / 10 / 704 / 5.36 / 0.001
Year * Mean plant Shannon-Wiener / 10 / 336 / 2.56 / 0.006
Year * Presence C3 grasses / 10 / 487 / 3.70 / 0.001
Year * Presence C4 grasses / 10 / 345 / 2.63 / 0.004
Year * Presence forbs / 10 / 214 / 1.63 / 0.097
Year * Presence legumes / 10 / 94 / 0.72 / 0.700
Year * Mean plant biomass / 10 / 690 / 5.26 / 0.001
Error (year) / 1450 / 19053
Table S3. Repeated measures analysis, using a General Linear Model, of effects of plant species richness, plant functional group richness, and plant biomass on herbivore abundance within years.
Plant species richness / 1 / 2.7 / 0.00 / 0.967
# Plant functional groups / 1 / 5444 / 2.87 / 0.092
Mean plant biomass / 1 / 1485 / 0.78 / 0.378
Error / 148 / 280565
Within-subject effects
Year / 10 / 127230 / 22.50 / 0.001
Year * Plant species richness / 10 / 632 / 1.12 / 0.349
Year * # Plant Functional groups / 10 / 2197 / 3.89 / 0.002
Year * Mean plant biomass / 10 / 634 / 1.12 / 0.347
Error (year) / 1480 / 565
Table S4. Repeated measures analysis, using a General Linear Model, of effects of plant species richness, plant functional group composition, and plant biomass on herbivore abundance within years.
Source / df / Type III Sum of Squares / F / PPlant species richness / 1 / 660 / 0.47 / 0.495
Presence C3 grasses / 1 / 2548 / 1.81 / 0.181
Presence C4 grasses / 1 / 60595 / 42.94 / 0.001
Presence forbs / 1 / 10293 / 7.29 / 0.008
Presence legumes / 1 / 2898 / 2.05 / 0.154
Mean plant biomass / 1 / 416 / 0.30 / 0.588
Error / 145 / 204628
Within-subject effects
Year / 10 / 57112 / 10.46 / 0.001
Year * Plant species richness / 10 / 4461 / 0.82 / 0.612
Year * Presence C3 grasses / 10 / 7535 / 1.38 / 0.184
Year * Presence C4 grasses / 10 / 34440 / 6.31 / 0.001
Year * Presence forbs / 10 / 8214 / 1.50 / 0.132
Year * Presence legumes / 10 / 19104 / 3.50 / 0.001
Year * Mean plant biomass / 10 / 12524 / 2.29 / 0.011
Error (year) / 1450 / 791844
Table S5. Repeated measures analysis, using a General Linear Model, of effects of plant species richness, plant functional group richness, and plant biomass on predator and parasitoid species richness within years.