Nested species subsets and geographic isolation: a case study
A major goal of community ecology is to identify and explain nonrandom patterns of species composition. In the past, a variety of “neutral models” have been developed and applied to test whether observed patterns of species distribution deviate from random expectations (Diamond 1975, Simberloff and Connor 1979, Strong et al. 1979, Nitecki and Hoffman 1987). Neutral models have been particularly used for analyzing patterns of species distribution among islands and other insular habitats (Simberloff 1978, Connor and McCoy 1979, Simberloff and Connor 1979, 1981, Diamond and Gilpin 1982, Connor and Simberloff 1984, Gilpin and Diamond 1984). Recently, several authors have used neutral models to test the hypothesis that insular systems contain “nested subsets” of species composition (Simberloff and Levin 1985, Patterson 1987, Ryti and Gilpin 1987, Blake 1991, Bolger et al. 1991, Patterson and Brown 1991, Simberloff and Martin 1991). Under such circumstances, each island of successively fewer species contains a subset of those species found on the next richer one (Patterson and Atmar 1986). The results obtained in these studies indicated that a variety of organisms including vertebrates, reptiles, amphibians, insects, and plants exhibit nested subsets of species composition, and that such nonrandom patterns of species distribution are particularly common among insular systems where species number is highly correlated with island size (Schoener and Schoener 1983, Simberloff and Levin 1985, Patterson and Atmar 1986, Patterson 1987, 1990, Ryti and Gilpin 1987, Blake 1991, Bolger et al. 1991, Patterson and Brown 1991, Simberloff and Martin 1991). The latter finding has led to the conclusion that differences among species in the effect of island size on the probability of extinction constitute the main mechanism producing nested subsets of species composition (Patterson and Atmar 1986, Patterson 1987, Soule et al. 1988, Bolger et al. 1991, Patterson and Brown 1991). Several authors (Darlington 1957, Ryti and Gilpin 1987, Patterson 1990) have proposed that differences among species in dispersal ability may interact with geographic isolation to produce nested species subsets as well, but supporting evidence is extremely rare (Schoener and Schoener 1983, Ryti and Gilpin 1987).
In this paper I use neutral models to test the hypothesis that isolation effects, coupled with selective dispersal, may produce nested subsets of species composition. I focus on woody plants (trees and shrubs) inhabiting a set of seven islands that were created in 1954 by the filling of the Clarks Hill Reservoir, Georgia, USA, and a single mainland site on its shore. Results from a previous study (Kadmon and Pulliam 1993) indicated that all of the studied islands, as well as the mainland site, had been logged and cleared of woody plants prior to their isolation from the mainland. It has also been found that the current number of woody species on the studied islands is negatively correlated with their distance to the mainland, and that islands at similar distances from the mainland have more species in common than islands differing from each other in their distance to the mainland (Kadmon and Pulliam 1993). On the basis of these findings I expected that species assemblages of islands at different distances from the mainland should represent nested subsets of each other.
Selection of the islands and the mainland site
Only small islands ([less than]1 ha) were selected for the study in order to allow a complete survey of all tree and shrub species on each island. Since the main focus of the study was on the role of geographic isolation, islands were selected to represent a wide range of distances from the mainland. To minimize the effects of post-isolation disturbance due to flooding or erosion by water-level fluctuations, only islands that had most of their area [greater than or equal to]30 cm above the maximum water level were included. An attempt was made to select relatively flat islands of relatively similar elevation and size. A total of seven islands fulfilling the above requirements were located. Distances between the selected islands and the mainland ranged from 20-540 m (average: 194 m), island area ranged from 0.3 to 0.8 ha (average: 0.47 ha) and the maximum elevation of islands ranged from 1 to 3 m (average: 1.4 m). The mainland site was a 100 x 25 m plot located along the water line on the shore of the reservoir.
Nestedness was assessed using two different approaches: the conventional Wilcoxon test as used in several previous studies, and a modification of this test based on computer randomization tests. The two approaches differed from each other in the probability structure of their null models.
Conventional tests. – The first approach was based on the conventional Wilcoxon two-sample test (also known as the Mann-Whitney U test) as proposed by Simberloff and Martin (1991). In this method (see also Schoener and Schoener 1983 and Simberloff and Levin 1985) the whole data set is arranged in the form of a presence/absence matrix with sites (columns) rank-ordered in terms of decreasing number of species and species (rows) rank-ordered in terms of decreasing number of sites occupied. Then, for each row, the degree to which the sequence of presences and absences is “ordered” is determined using the Wilcoxon two-sample rank-sum statistic. This statistic gives the null probability for deviation of the observed sequence of presences and absences from a sequence in which the same numbers of presences and absences are randomly arranged. The advantage of this approach is that it allows one to assess the degree to which the whole system is nested (by assuming the species are independent and combining the tail probabilities for the individual species), as well as to determine which species conform or do not conform to a nested pattern (Simberloff and Levin 1985, Simberloff and Martin 1991). In the present study, Wilcoxon statistics were calculated for three hierarchical levels: (1) individual species, (2) groups of species representing different dispersal properties, and (3) the whole species assemblage. Species were categorized into three dispersal categories: species with adaptations for bird dispersal, species with adaptations for wind dispersal, and species lacking any adaptations for bird or wind dispersal. Fleshy fruits were considered an adaptation for bird dispersal and wings or pappus were considered adaptations for wind dispersal. Species showing zero absences were excluded from the analysis, as recommended by Simberloff and Martin (1991). Scores of the Wilcoxon statistics were transformed to standard deviations from the mean to use the normal approximation and their significance levels were determined using one-tailed tests. To directly test the hypothesis that geographic isolation was important in producing nested patterns of species composition the data were reanalyzed using the same analytic approach, but with sites ranked according to their distance from the mainland, rather than with respect to their species number. Similarly, by ranking the sites according to their area I tested the degree to which differences among sites in area were important in producing nested species subsets.
Randomization tests. – In terms of the presence/absence matrix, the null model used by the conventional Wilcoxon test is one in which the numbers of presences and absences within each row are kept at the observed values but their order is completely random. Thus, the fact that the observed data are ranked in a specific way prior to the calculation of the Wilcoxon statistic is not taken into account in the probability structure of the null model. Arranging the sites of the matrix with respect to species number can be expected to introduce some nestedness to any matrix (even a completely random one), simply because species presences are shifted into one side of the matrix and absences are shifted into the other side. A similar artifact is expected if sites are arranged with respect to any variable that is correlated with species number per island (e.g., island size or distance from the mainland). To correct for this bias, the ordering process associated with the calculation of the Wilcoxon statistic has to be taken into account in the probability structure of the relevant null model. In terms of the presence/absence matrix this can be done by using an ordered version of the randomized matrix (rather than the completely randomized one) as the null model with which the observed patterns are compared.
In the study reported here I used randomization experiments to test whether the results of the conventional Wilcoxon tests were biased by the ordering process. Each permutation run included two steps: (1) randomization of the original matrix under the constraint that the number of islands occupied per species is kept at the observed level (as assumed by the conventional Wilcoxon test), and (2) arrangement of the sites with respect to the variable whose effect on the distribution pattern was tested (i.e., species number, distance from the mainland, or area). Deviations of observed distribution patterns from the ordered null model were assessed by comparing their Wilcoxon scores with the distribution of scores obtained for 10,000 matrices that were randomized and ordered in the fashion described above.
Thus, for each Wilcoxon score I determined two alternative significance levels, one based on a conventional test (a completely random matrix as the null model) and the other based on a randomization test (an ordered version of the random matrix as the null model). Differences between the probability values obtained by the two methods were tested using pairwise t tests.
Conventional tests (random matrix as the null model)
The whole matrix was significantly ordered with respect to both species number and distance to the mainland, but not with respect to area (Table 1). However, when the same analysis was repeated for each dispersal category separately, statistically significant deviations from the random expectation were obtained for only two categories; species with adaptations for bird dispersal and species lacking adaptations for wind or bird dispersal (Table 1). Both groups exhibited patterns of distribution that were significantly ordered with respect to species number and distance from the mainland. The patterns of island occupancy obtained for wind dispersed species were not significantly different from the random expectation in all cases (Table 1).
Analysis of distribution patterns at the level of individual species indicated that eight species (Carya tomentosa, Crataegus marshallii, Crataegus spathulata, Crataegus viridis, Prunus umbellata, Quercus rubra, Q. stellata, and Robinia pseudoacacia) exhibited presence/absence sequences that were significantly ordered with respect to species number. All of these species plus one additional species (Quercus phellos) exhibited significantly ordered sequences with respect to distance from the mainland. The percentage of species showing significantly ordered sequences of occurrence with respect to distance from the mainland was lowest in the case of the wind dispersed species, intermediate in the case of the bird dispersed species, and highest in the case of species with no adaptations for wind or bird dispersal [ILLUSTRATION FOR FIGURE 1 OMITTED]. Two species (Betula nigra and Morus alba) showed presence/absence sequences that were significantly ordered with respect to area. However, considering that in a data set consisting of 43 species two species may show significantly ordered sequences by chance, the area variable cannot be considered significant in this analysis.
TABLE 1. Significance levels for deviations of observed Wilcoxon
two-sample test scores from the random null model, calculated for
sites ranked with respect to their species number, distance from the
mainland, and area. Boldface values represent significant (P [less
than] 0.05) deviations (see Methods: Data analysis for details about
the probability structure of the null model).
Differences among species in dispersal ability may result in selective colonization of islands at different distances from the mainland (MacArthur and Wilson 1967, Gilpin and Diamond 1981), but also in selective rates of local extinction via the “rescue effect” (Brown and Kodric-Brown 1977). Both processes may contribute to the development of nested subsets of species composition (Patterson 1990). Which of these processes was more important in producing the patterns observed in this study cannot be determined from the data available.
If limited dispersal ability facilitates the development of nested species compositions we would expect that strongly nested patterns should be more common among taxa characterized by poor mobility (e.g., mammals or reptiles) than among highly mobile taxa (e.g., birds). Suprisingly, data from two previous studies (Schoener and Schoener 1983 and Ryti and Gilpin 1987) suggest exactly the opposite trend. Schoener and Schoener (1983) found that migrant birds distributing among the Bahamian islands were more nested with respect to isolation indices than resident birds or lizards. Similarly, Ryti and Gilpin (1987) found that patterns of bird distribution among islands in a variety of archipelagos were usually more ordered than those of mammals. In both of these studies, highly nested patterns were associated with high, rather than with low dispersal ability. Ryti and Giipin (1987) proposed two types of hypotheses to explain the patterns they found: (1) taxa with limited dispersal ability may be more sensitive to the biotic and abiotic factors affecting the probability of successful colonization, and (2) taxa with low dispersal ability may have a larger random component in their colonization process. However, none of these hypotheses was supported by empirical evidence.
It is important to emphasize that the island system analyzed in this study differs from many previously studied systems in both the spatial and temporal scales over which the processes producing the observed distribution patterns have operated. As to the spatial scale, most previous studies have focused on archipelagos where mainland-island distances were greater by several orders of magnitude than those characterizing the Clarks Hill system. In such systems even species with relatively good dispersal ability can be expected to show distance-related patterns of island occupancy. In the system studied here the distance between the mainland and the most isolated island was [less than]600 m and it is therefore not surprising that species with adaptations for long-range dispersal were not sensitive to among-island differences in the degree of isolation.
As to the temporal scale, most previous tests of the nestedness hypothesis have focused on insular systems that were isolated for long periods of time (in many cases [greater than or equal to][10.sup.3] yr). In systems with such a long history of isolation extinction events can be expected to play an important role in shaping patterns of island occupancy. The island system analyzed in this study was isolated for [less than]40 yr. Considering that the median longevity of trees and shrubs is usually [greater than]50-100 yr (Harper 1977), it is not surprising that no evidence for extinction-related patterns of island occupancy was found.
In the study reported here, observed patterns of island occupancy were tested against two types of null models; a random one (the conventional Wilcoxon two-sample test as was used in several previous studies, e.g., Schoener and Schoener 1983, Simberloff and Levin 1985, Simberloff and Martin 1991), and an ordered one. In both types, the number of islands occupied per species was kept at the observed value. However, in one type (the random model) the sequence of presences and absences within each row was completely random, while in the other type (the ordered model) the sites were rank-ordered with respect to the variable whose effect was tested. The results indicated that deviations of the observed patterns of island occupancy from the ordered null model were significantly smaller than the corresponding deviations from the random model. In addition, several distribution patterns were significantly different from the random model, but not from the ordered one. These findings indicate that the ordering process associated with any test of the nestedness hypothesis has to be taken into account in structuring the null model against which the observed data are tested.
TABLE 3. Matrices representing two hypothetical systems with
two species (rows) and 20 islands (columns). Islands are
assumed to be arranged with respect to their distance from the
mainland. Both matrices would show statistically significant
Wilcoxon scores, but neither of them is nested.
Harper, J. L. 1977. The population biology of plants. Academic Press, London, England.
Kadmon, R., and R. H. Pulliam. 1993. Island biogeography: effect of geographic isolation on species composition. Ecology 74:977-981.
MacArthur, R. H., and E. O. Wilson. 1967. The theory of island biogeography. Princeton University Press, Princeton, New Jersey, USA.
Nitecki, M. H., and A. Hoffman, editors. 1987. Neutral models in biology. Oxford University Press, New York, New York, USA.
Patterson, B. D. 1987. The principle of nested subsets and its implications for biological conservation. Conservation Biology 1:323-334.
—–. 1990. On the temporal development of nested subset patterns of species composition. Oikos 59:330-342.
Patterson, B. D., and U. Atmar. 1986. Nested subsets and the structure of insular mammalian faunas and archipelagoes. Biological Journal of the Linnean Society 28:65-82.
Patterson, B. D., and J. H. Brown. 1991. Regionally nested patterns of species composition in granivorous rodent assemblages. Journal of Biogeography 18:395-402.
Ryti, R. T., and M. E. Gilpin. 1987. The comparative analysis of species occurrence patterns on archipelagoes. Oecologia 73:282-287.
Schoener, T. W., and A. Schoener. 1983. Distribution of vertebrates on some very small islands. 1. Occurrence sequences of individual species. Journal of Animal Ecology 52:209-235.
Simberloff, D. 1978. Using island biogeographic distributions to determine if colonization is stochastic. American Naturalist 112:713-726.
Simberloff, D., and E. F. Connor. 1979. Q-mode and R-mode analysis of biogeographic distributions: null hypotheses based on random colonization. Pages 123-138 in G. P. Patil and M. L. Rosenzweig, editors. Contemporary quantitative ecology and related ecometrics. International Cooperative Publishing House, Fairland, Maryland, USA.
Simberloff, D., and E. F. Connor. 1981. Missing species combinations. American Naturalist 118:215-239.
Simberloff, D., and B. Levin. 1985. Predictable sequences of species loss with decreasing island area – land birds in two archipelagos. New Zealand Journal of Ecology 8:11-20.
Simberloff, D., and J. L. Martin. 1991. Nestedness of insular avifaunas: simple summary statistics masking complex species patterns. Ornis Fennica 68:178-192.
Soule, M. E., D. T. Bolger, A. C. Alberts, J. Wright, M. Sorice, and S. Hill. 1988. Reconstructed dynamics of rapid extinctions of chaparral-requiring birds in urban habitat islands. Conservation Biology 2:75-92.
Strong, D. R., L. A. Szyska, and D. Simberloff. 1979. Tests of community-wide character displacement against null hypotheses. Evolution 33:897-913.
COPYRIGHT 1995 Ecological Society of America
COPYRIGHT 2004 Gale Group