Conceived and designed the experiments: TEL. Performed the experiments: TEL. Analyzed the data: AKL NH. Contributed reagents/materials/analysis tools: AKL NH TEL. Wrote the paper: AKL NH TEL.
Current address: Department of Plant Science and Landscape Architecture, University of Maryland, College Park, Maryland, United States of America
CitiCorp is a commercial funder, but this organization had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Having this source of funding does not alter the authors' adherence to PLoS ONE policies on sharing data and materials.
Habitat fragmentation is a major driver of biodiversity loss. Yet, the overall effects of fragmentation on biodiversity may be obscured by differences in responses among species. These opposing responses to fragmentation may be manifest in higher variability in species richness and abundance (termed hyperdynamism), and in predictable changes in community composition. We tested whether forest fragmentation causes long-term hyperdynamism in butterfly communities, a taxon that naturally displays large variations in species richness and community composition. Using a dataset from an experimentally fragmented landscape in the central Amazon that spanned 11 years, we evaluated the effect of fragmentation on changes in species richness and community composition through time. Overall, adjusted species richness (adjusted for survey duration) did not differ between fragmented forest and intact forest. However, spatial and temporal variation of adjusted species richness was significantly higher in fragmented forests relative to intact forest. This variation was associated with changes in butterfly community composition, specifically lower proportions of understory shade species and higher proportions of edge species in fragmented forest. Analysis of rarefied species richness, estimated using indices of butterfly abundance, showed no differences between fragmented and intact forest plots in spatial or temporal variation. These results do not contradict the results from adjusted species richness, but rather suggest that higher variability in butterfly adjusted species richness may be explained by changes in butterfly abundance. Combined, these results indicate that butterfly communities in fragmented tropical forests are more variable than in intact forest, and that the natural variability of butterflies was not a buffer against the effects of fragmentation on community dynamics.
Theoretical and empirical research provides overwhelming evidence that habitat fragmentation reduces biodiversity
One potential explanation for contrasting results among studies and taxa is that fragmentation may increase variability of community diversity or composition. Laurance
There are at least two potential explanations for increased variability in species richness within fragmented landscapes. First, resident populations may exhibit increased variability in abundance after fragmentation. Second, fragmentation may change community composition, favoring species that fluctuate more in their population sizes. For example, in the same experimental landscape as the tropical tree study, beetles that were rare or had naturally fluctuating abundances were more likely than species with stable abundances to persist in forest fragments a decade after fragmentation
Insects, the most diverse higher taxon
Here, we test the long-term effects of forest fragmentation on the dynamics of species richness and community composition of butterflies. We analyzed a dataset on butterfly species richness that spans 11 years from an experimentally fragmented landscape in the central Amazon. Rather than comparing forest fragments to continuous forest at a single point in time, we evaluated the effects of fragmentation on changes in species richness through time, focusing on three main questions:
Does habitat fragmentation reduce species richness, and do changes persist over time?
Does fragmentation increase spatial and temporal variation in species richness?
Does fragmentation alter butterfly community composition, and how might differences in community composition impact variability in species richness?
The Biological Dynamics of Forest Fragments Project (BDFFP) is a 1,000 km2 experimental landscape located in the central Amazon, approximately 80 km north of Manaus, Brazil. Starting in the early 1980s, tropical forest was cleared to create five 1 ha, four 10 ha, and two 100 ha forest fragments (referred to as fragmented forest plots). The same number of plots, in matching sizes, was established in nearby continuous forest (referred to as intact forest plots). Initially, the matrix surrounding forest fragments was maintained as cattle pasture, but over time some pastures were abandoned and subsequently invaded by shrubs and secondary forest species. A detailed description of the BDFPP experiment, including a map and plot history, can be found in Laurance and Bierregaard
The data we analyzed were collected during butterfly surveys in the fragmented and intact forest plots between 1980–1986 and in 1991 (
Surveys were typically conducted on one day per year. Because of this, we restricted all of our analyses to data from one day each year. In the occasional years when there were additional surveys, we chose surveys across plots that were closest in time of year and observer hours (typically seven hours). We did not pool surveys for a given plot within a year because multiple surveys within a year were rare and incomplete, and because of high species turnover across seasons (see
As survey duration varied by plot and year (
We used ANOVA to test for effects of log-transformed plot size, fragmentation status (fragmented or intact), and their interaction on the adjusted species richness for each plot averaged across years. We then conducted analyses for each plot size, and compared the values of adjusted species richness post-fragmentation to average adjusted species richness in intact forest. We used a Dunnett's test to compare adjusted species richness of fragmented plots in each year after fragmentation to adjusted species richness of intact forest. Dunnett's test is a post-hoc test that accounts for multiple comparisons. It is more powerful than a Bonferroni test because the treatments are only compared to the control (intact forest), as opposed to every possible pairwise comparison.
To test for the effects of fragmentation on temporal variation in adjusted species richness, we measured the average change in adjusted species richness within a plot over time. For each plot, we measured the change in adjusted species richness between two surveys using the formula:
We measured spatial variation by examining the difference in adjusted species richness between plots within a given calendar year (intact forest plots) or number of years since fragmentation (fragmented forest plots). Separately, for each year in the intact and fragmented forest plots, we calculated the coefficient of variation (
As species richness and species extinction rates depend on abundance, we used the only abundance data available, estimated to the nearest power of two (1, 2, 4, 8, 16, etc.), to calculate rarified species richness using Primer (Plymouth Routines in Multivariate Ecological Research, v.6). Rarefaction controls for the effects of butterfly abundance on their diversity
We analyzed community composition based on the habitat association of species present in fragmented and intact forests. Previous analysis
We grouped butterflies into the same four habitat association categories as Brown and Hutchings
We analyzed data from 40 butterfly surveys of 11 intact forest plots and 48 butterfly surveys of 11 fragmented forest plots. A total of 414 butterfly species was observed.
Log-transformed plot size, and the interaction between size and fragmentation status, were significant predictors of adjusted species richness (overall model F3,18 = 7.810,
Values of adjusted species richness were regressed against plot size for intact plots (solid squares, solid regression line) and fragmented plots (gray diamonds, dashed regression line). For fragmented plots, larger plots have more species. Intact plots are offset for visual emphasis only.
Comparing intact forest plots to fragmented forest plots based on time since fragmentation showed an inconsistent response of adjusted species richness to fragmentation. In 1-ha plots, there was no difference in adjusted species richness between intact forest plots and fragmented forest plots at any time post-fragmentation (F6,32 = 2.389,
Adjusted species richness (+SD) for intact plots (black bars) and fragmented plots (gray bars) for (A) 1 ha plots, (B) 10 ha plots, and (C) 100 ha plots. Data from intact plots are combined across all years. Bars with asterisks indicate time points with a significant difference (
Temporal variation in adjusted species richness was significantly greater in fragmented forest plots relative to intact forest plots (
Variation in adjusted species richness for fragmented and intact forest plots for (A) average change within plots over time (temporal variation) and (B) average variation between plots at a given point in time (spatial variation). (C) A regression of average
Fragmented forest plots of all sizes harbored different butterfly communities relative to intact forest plots (
For each habitat association, differences were taken as average proportion in intact plots–average proportion in fragmented plots. Therefore, negative values represent cases where a habitat association had a higher average proportion in fragmented forest plots relative to intact forest plots. Asterisks denote habitat associations for which fragmentation status was significant and plus signs denote associations for which size was significant (see
Habitat association/variable | df | F | ||
Edge ( |
2,19 | 68.022 | <0.001 | |
ln(size) | 1 | 4.493 | 0.047 | |
fragmentation status | 1 | 131.550 | <0.001 | |
Canopy and clearing ( |
2,19 | 1.307 | 0.294 | |
ln(size) | 1 | 1.856 | 0.189 | |
fragmentation status | 1 | 0.758 | 0.395 | |
Understory sun ( |
2,19 | 24.507 | <0.001 | |
ln(size) | 1 | 5.852 | 0.026 | |
fragmentation status | 1 | 43.163 | <0.001 | |
Understory shade ( |
2,19 | 5.733 | 0.012 | |
ln(size) | 1 | 1.918 | 0.182 | |
fragmentation status | 1 | 9.548 | 0.006 |
Our results show that tropical forest fragmentation increased the temporal and spatial variability of species richness adjusted for survey duration in butterfly communities. We examined three mechanisms to explain this higher variability. First, the variability in adjusted species richness was not associated with a change in average species richness between fragmented and intact forests, but was associated with a change in community composition. Second, variability was associated with changes in butterfly community composition. Butterfly communities in fragmented and intact forest had similar proportions of species associated with high light environments, such as canopies and clearings. However, fragmentation increased the proportion of edge species and decreased the proportion of shade-dwelling species associated with closed canopy forest. Third, when we accounted for abundance in estimates of rarefied species richness, using the limited data we had available (indices of abundance), fragmented forest plots were no more spatially or temporally variable than intact forest plots. Combined, these results suggest that the increased variability in butterfly species richness in forest fragments may be explained by changes in community composition and butterfly abundance. Consequently, the natural variability of butterflies was not a buffer against the effects of fragmentation on community dynamics
The response of butterfly species richness to habitat fragmentation and plot size demonstrates how fragmentation effects can be obscured by opposing responses of different species groups (
Our results show evidence that hyperdynamism in forest fragments may be caused by changes in abundance. One cause for changes in abundance may be changes in resource availability
The change in community composition may also contribute to hyperdynamism in fragmented forests, as the community could have shifted toward species or species groups that are more variable in abundance. The diversity of butterfly species associated with closed canopy forest (understory shade species) and understory sun species were the most severely reduced by fragmentation. Although average adjusted species richness did not differ between fragmented and intact forests, the proportion of understory sun and shade species significantly declined in fragmented forest plots, indicating local extinctions of these butterflies. The decline in shade species likely reflects a decline in their available habitat in proportion to fragment size. Species associated with high light environments, such as canopy and clearing species, would be less affected by changes in habitat structure and microclimate near habitat edges because the new habitat more closely resembles their natural habitat.
Uncontrolled aspects of the experimental design could contribute to our findings of higher variability of adjusted species richness in forest fragments. Variation in landscape features, including the distance to intact forest and the type of matrix vegetation, could contribute to the higher spatial variability of species richness in fragmented forest plots. Although these are other interesting consequence of fragmentation, the differences we observed are not confounded by them, as our measure of temporal variability compared the average change
The logistical and statistical hurdles faced in this study are representative of many large-scale, long-term studies that analyze multiple species. Ideally but unrealistically, all plots would have been completely surveyed multiple times per year, with precise measures of species abundance for monitoring population trends. One alternative, focusing intensively on just a few charismatic or rare species, may give misleading information about overall community dynamics, particularly given the number of rare species in tropical forests. By grouping species based on life history traits, feeding guilds, or habitat associations, we can use more targeted methods to assess the effects of fragmentation. In turn, these data could inform conservation strategies designed to protect the most vulnerable species.
Our findings of equal adjusted species richness in fragmented and intact forest plots are consistent with those from a study of trees in the same experiment
Ecological theory provides compelling evidence that the spatial structure of landscapes can affect the stability of community dynamics
Two main conservation recommendations for monitoring insects in fragmented habitats come from our findings that butterfly communities display characteristics of hyperdynamism. First, long time-series of community data are critical to understanding the effects of fragmentation. Increased spatial and temporal variation of adjusted species richness in fragmented plots was not associated with a change in overall species richness. However, a survey at any given time after fragmentation could have found either higher or lower species richness in fragmented plots, relative to intact plots, simply by chance. By looking at changes in richness over time, we found that fragmentation affected the dynamics of species richness, as opposed to static values of species richness. Second, given limited time and resources, monitoring based on community composition, rather than species richness, provides quick feedback to conservation planners on the effects of fragmentation. Analyses based on percent composition may reduce the effect of inaccurate species identification and help in cases where there are many rare species in an extremely diverse fauna.
Supporting text.
(0.03 MB DOC)
Schematic of butterfly surveys used in the analyses. For each survey, the date and duration of the survey (in hours) is listed. The year a plot was fragmented (top of figure) is outlined in black.
(0.57 MB TIF)
The effect of seasonality on species richness. The residuals from the regression of combined species richness against the number of months between surveys was not significant for intact forest plots (solid squares, solid regression line), but was significant for fragmented plots (gray diamonds, dashed regression line).
(0.14 MB TIF)
Regression of species richness against survey hours for intact forest plots. Surveys are coded by plot size for visual emphasis only, as plot size was not a significant variable in determining species richness.
(0.14 MB TIF)
Regression of species richness against survey hours for fragmented forest plots. (A) All plots combined. Surveys are coded by plot size for visual emphasis only. (B) The 1 ha regression was not significant, but the 10 ha (C) and 100 ha (D) regressions were significant.
(0.19 MB TIF)
ANOVA results for species richness in intact forest plots with significant variables for the effects test bolded.
(0.01 MB PDF)
ANOVA results for species richness in fragmented forest plots with significant variables for the effects test bolded (
(0.04 MB PDF)
We thank R. Hutchings and K. Brown for the use of the butterfly dataset and R. Hutchings for insights about the interpretation of the results and comments on the manuscript. We also thank L. Borda de Água, R. R. Dunn, and S. L. Pimm for additional insights into the analysis, K. Gross for help with statistics, and R. R. Dunn and two reviewers for helpful comments on the manuscript. This manuscript is number 553 of the BDFFP technical series.