Abstract
Although habitat loss is the predominant factor leading to biodiversity loss in the Anthropocene1,2, exactly how this loss manifests—and at which scales—remains a central debate3,4,5,6. The ‘passive sampling’ hypothesis suggests that species are lost in proportion to their abundance and distribution in the natural habitat7,8, whereas the ‘ecosystem decay’ hypothesis suggests that ecological processes change in smaller and more-isolated habitats such that more species are lost than would have been expected simply through loss of habitat alone9,10. Generalizable tests of these hypotheses have been limited by heterogeneous sampling designs and a narrow focus on estimates of species richness that are strongly dependent on scale. Here we analyse 123 studies of assemblage-level abundances of focal taxa taken from multiple habitat fragments of varying size to evaluate the influence of passive sampling and ecosystem decay on biodiversity loss. We found overall support for the ecosystem decay hypothesis. Across all studies, ecosystems and taxa, biodiversity estimates from smaller habitat fragments—when controlled for sampling effort—contain fewer individuals, fewer species and less-even communities than expected from a sample of larger fragments. However, the diversity loss due to ecosystem decay in some studies (for example, those in which habitat loss took place more than 100 years ago) was less than expected from the overall pattern, as a result of compositional turnover by species that were not originally present in the intact habitats. We conclude that the incorporation of non-passive effects of habitat loss on biodiversity change will improve biodiversity scenarios under future land use, and planning for habitat protection and restoration.
Similar content being viewed by others
Main
When habitat is lost, two different processes can lead to biodiversity decline. First, species are lost as habitat area is lost via the ubiquitous species–area relationship11,12; fewer species can persist in smaller habitats, and species that are rarer in the landscape will go extinct owing to passive sampling. Second, the demography of species within the remaining habitat can be altered, increasing extinction risk (or decreasing recolonization) subsequent to immediate species losses9,10. This process—which is referred to as ecosystem decay9—supposes that biological processes within intact habitats embedded within smaller habitat areas differ from those in larger patches (for example, owing to edge effects, lowered dispersal or demographic stochasticity). For a given sampling effort, we might expect fewer species, changes in total and relative abundances, and often altered ecosystem functions, in smaller relative to larger habitats as a result of ecosystem decay (see empirical examples in refs. 3,13,14,15). Both of these processes—biodiversity loss owing to passive sampling or to altered demography that leads to ecosystem decay—are well-known, but often confounded in empirical tests for which sampling efforts are variable. Additionally, most models that forecast biodiversity loss as a result of habitat loss implicitly assume passive sampling from larger to smaller habitats when developing scenario projections (for example, refs. 11,12,16,17), ignoring demographic effects. At present we are unable to synthesize the relative importance of the two main processes of biodiversity loss with habitat loss. Without this synthesis, we cannot address the necessity of incorporating loss owing to both passive sampling and ecosystem decay into forecasts of future biodiversity scenarios.
Passive sampling versus ecosystem decay
In Fig. 1, we visualize the primary processes by which biodiversity and its empirical estimates can respond to habitat loss when explicitly controlling for sampling effort to establish explicit testable hypotheses18,19,20. Although it is often impractical or impossible to count all of the individuals or species in a given habitat patch, we can readily establish a scheme to estimate the numbers of individuals, the numbers of species or other estimates of biodiversity (for example, relative abundances) within samples of the habitat (as illustrated by the square plots in Fig. 1). By standardizing this sampling effort, we can compare effort-standardized estimates from larger, more-continuous habitat patches to those from smaller habitat patches. With this, we can then explicitly test the importance of within-patch demographic changes that lead to the ecosystem decay of biodiversity measures amidst losses due to passive sampling20. Additionally, even if sampling effort among patches was not standardized in the original study, but is known, we can use post hoc methods for standardization (for example, various forms of rarefaction; Methods).
a, Small fragments (left) showing three hypotheses for comparisons made with the large fragment (right): (1) passive sampling, in which the densities of individuals and the relative abundances of species are similar to that of the large fragment; (2) ecosystem decay (evenness), in which the relative abundances of species are more uneven in the small fragment compared to the large fragment; and (3) ecosystem decay (individuals), in which the density of individuals is lower in the small fragment compared to the large fragment. Sampling is illustrated with boxes inside each fragment. In this example, the sampling effort increases from small (two samples) to large (four samples) fragments, but sampling effort can be standardized post hoc because sample-level data are available; other sampling methods are slightly different but can also be standardized (Methods). b, We compare the standardized number of individuals (average individuals per sample) (left), standardized species richness (average richness per sample) (centre) and standardized species evenness (average evenness per sample) (right) between a small fragment and a large fragment to illustrate how this information can be used to disentangle the passive sampling versus the ecosystem decay hypotheses. For the purposes of simple illustration, the large fragment has a completely even species abundance distribution, even though this is unrealistic. This figure was created for this paper by F. Arndt (www.Formenorm.de).
Several outcomes are possible when comparing sample-effort-controlled biodiversity patterns from larger habitat fragments to those from smaller habitat fragments20: three illustrative cases of these outcomes are shown in Fig. 1. The first is passive sampling: here, while there may be fewer total species in smaller relative to larger fragments, the numbers of individuals, species and relative abundances of species with a given standardized sampling effort is the same. That is, we expect there to be no relationship between any of these standardized estimates of biodiversity and habitat fragment size. In Extended Data Fig. 1, we use simulations to explore the robustness of this expectation and show that it is not sensitive to extreme non-randomness (that is, aggregation) in species distributions. Second, we consider ecosystem decay as a result of changes in the relative abundance of species. In this outcome, differential demographic responses of species in the assemblage lead to some species becoming relatively more abundant and others relatively less abundant (that is, relative abundances become more uneven) in smaller compared to larger fragments. As a result, the numbers of species for a given sampling effort declines, as do other measures of diversity based on relative abundances, with fragment size. The third example is ecosystem decay that results from reductions in the numbers of individuals. Here, there are fewer total species and also fewer species per standardized sampling effort in smaller relative to larger fragments, resulting from lower demographic rates of species in the small compared to the large fragments. In this scenario, while there are fewer individuals in samples from smaller fragments, all species have the same demographic response such that the relative abundances of species have not changed. As a result, the rarefied richness (standardized for the numbers of individuals) is similar in smaller and larger fragments. Although this is not shown in Fig. 1, it is also likely that ecosystem decay can result from changes both to the total numbers of individuals and to species relative abundances, from larger to smaller fragments. These are not the only possible outcomes, but other patterns—such as increases in abundance or evenness with decreasing habitat size—are less likely. Such increases in abundance or evenness would in turn lead to higher measures of diversity in standardized samples within smaller patches4,6. Some of these other outcomes are plausible, but we do not discuss them further as the bulk of our evidence indicates they are far less common than the three shown in Fig. 1.
Testing the hypotheses
Although the hypotheses illustrated in Fig. 1 are straightforward, appropriate syntheses have not heretofore been possible because of highly heterogeneous data that come from different sampling designs, as well as often-confounded analyses that do not control for sampling effort and grain21. However, if ecosystem decay in the form of changes to the numbers of species, individuals and/or relative species abundances in standardized samples is pervasive, this would mean that the majority of biodiversity loss scenarios—which assume passive sampling and ignore ecosystem decay—probably underestimate the effects of habitat loss on species loss. Here, we have developed an explicit and comprehensive evaluation of the passive sampling versus ecosystem decay hypotheses using 123 datasets in which species abundances were sampled from multiple habitat fragments of different sizes within once continuous landscapes22 (Fig. 2a). Our dataset was explicitly compiled for this effort, and we developed an analytical pipeline that controls for sampling effort to test the hypotheses of biodiversity loss stemming from habitat loss. Datasets had from 2 to 53 habitat fragments of varying size within a given ecosystem, ranging from less than 1 ha to more than 10,000 ha of continuous habitat cover, and included studies on plant, invertebrate and vertebrate assemblages. Approximately 85% of our studies came from forested landscapes embedded within anthropogenic matrices (for example, agriculture), and the other studies came from shrublands, grasslands and wetlands. Although our search for datasets was global, there are clear geographical patterns in terms of where studies were collected; the majority come from tropical areas, especially Central and South America (Fig. 2a) (further details are provided in the Methods and ref. 22).
a, Global map, indicating the taxon group and location of studies (n = 123) included in our analyses. b–d, Standardized samples show that number of individuals (b), species richness (c) and evenness (d) all increase as a function of fragment size. Solid black lines and shading show overall relationships and 95% credible intervals for each metric; the slope (β) coefficient for each metric and its 95% credible interval are shown at the top. Coloured lines show study-level relationships for different taxon groups (as shown in the map legend).
Overall, we found strong support for the ecosystem decay hypothesis across all studies (Fig. 2, Extended Data Fig. 2). After controlling for sampling effort (Methods), we found that each of the biodiversity variables increased as the size of the habitat fragments increased. This included increases in the standardized number of individuals with fragment size (Fig. 2a), increases in standardized species richness with fragment size (Fig. 2b), increases in the evenness (relative abundance) of species in the community with fragment size (estimated via the effective number of species conversion of the probability of interspecific encounter (SPIE), a measure of evenness that is relatively insensitive to sample size (Methods)) (Fig. 2c) and increases in several other sample-effort-controlled biodiversity measures with fragment size (Extended Data Fig. 2).
We developed a second analysis to explicitly account for the uncertainty surrounding the expected biodiversity estimates from the passive sampling hypothesis. Simulations of the passive sampling hypothesis show that there can be variation in the expected slopes of biodiversity responses to the size of habitat fragments (Extended Data Fig. 1), although our observed responses (Fig. 2, Extended Data Fig. 2) were considerably greater than any departures in our simulations. Nevertheless, to account for this variation, we calculated variation in the ‘expected’ value for a given smaller fragment by resampling from the largest fragment. We then included uncertainty in the expected value under passive sampling when making comparisons to the ‘observed’ value in each smaller fragment. The resulting z-score (Methods) allows us to discern whether and by how much biodiversity in smaller fragments deviates from that expected under passive sampling. With this analysis, we found that for each sample-standardized diversity response, the observed values deviated more from expected values (lower z-scores) for smaller patch sizes, which indicates strong support for ecosystem decay having a role in the response of taxa to habitat loss (Extended Data Fig. 3). Moreover, to make all 123 datasets comparable, we made decisions regarding response variables (for example, how to rarefy from different types of data), as well as the driver variables (for example, how to estimate the size of habitats classified as ‘continuous’). We present several analyses (Methods) that examine the sensitivity of our results, and show that our overall results are highly robust (Extended Data Fig. 4).
Variation in the effects
We were also interested in understanding how the response of biodiversity to fragment size varied across studies owing the taxa considered, the continent on which the studies were located, the time since fragmentation in the study, the quality of the matrix habitat surrounding the fragments in the study, and the latitude of the study. We focused on differences in the slopes of the response (with a slope of approximately zero expected from passive sampling) because differences in intercepts are influenced by the species pool and sampling methodology, and thus not meaningful in this context. We illustrate variation around the overall trend for sample-effort-standardized species richness in Fig. 3, and for sample-effort-standardized number of individuals and evenness in Extended Data Fig. 5.
a–d, Density plots of posterior distributions for study-level slope estimates of standardized species richness grouped by taxon group (a), continent (b), time since fragmentation (c) and matrix filter (d). The dashed vertical line shows the zero slope expected from passive sampling. Each density plot is based on 1,000 samples from the posterior distribution of each study-level slope estimate. Densities are shaded by quantiles and the black diamond shows the median for each group. Solid line and surrounding shading show the overall slope and 95% credible interval. n values next to each plot refer to the number of studies included in each category.
With a few exceptions, all results are qualitatively similar to those for species richness, which showed—first—that there are no clear differences across taxa (Fig. 3a). This is somewhat surprising, as it has often been suggested that larger taxa or those that disperse poorly might respond to habitat loss more markedly than taxa with the opposite traits (for example, refs. 19,23). However, the broad taxonomic groupings that were necessary for this global analysis do not allow for comparisons of species traits that vary within a group (for example, wind- versus animal-dispersed plants).
Second, we found little signal of continent, other than for temperate North America (which showed higher than average slopes) and Europe (which showed lower than average slopes) (Fig. 3b). One possibility is that these deviations reflect differences in other covariates. Specifically, habitats in 10 of the 16 studies from Europe were fragmented more than 100 years before the data were collected, whereas fragmentation occurred much more recently elsewhere. Likewise, studies from North America all had intermediate-to-harsh environmental filters. We explore time since fragmentation and matrix quality in more detail in the following two paragraphs.
Third, studies with the longest time since habitat loss had the shallowest slopes (Fig. 3c). This result might seem counterintuitive given the ‘extinction debt’ concept, which refers to the fact that species may persist in fragments after habitat loss, but eventually go extinct as populations are no longer viable24. However, alternative explanations for shallower slopes in older habitat fragments could be that populations eventually compensate for early losses (for example, via dispersal into the fragment or recovery of the landscape matrix), or that species turnover via immigration of more-tolerant species compensate for losses of more-specialized species25. We address the latter in the ‘Effects on species composition’ section below.
Fourth, studies in which fragments had habitat matrices that were categorized as harsh (for example, intense agriculture or urban areas) had steeper slopes (Fig. 3d). Fragments within matrices that were categorized as light (for example, secondary forests or shade-grown low-intensity agriculture) had shallower slopes (Fig. 3d). Such a result is consistent with expectations that the effects of ecosystem decay should be more pronounced when matrices are distinct from the habitat of interest14, and support the idea that less-intensive habitat use (and restoration) can reduce the effect of habitat loss on biodiversity loss within intact habitats.
Finally, we compared our results with those from a recent study that used very different methodology and scales to show greater responses to fragmentation in tropical compared to temperate areas26. We found a weak decline in the slope estimate with increasing (absolute) latitude (−0.0005; 95% credible interval of −0.001–−0.0001) (Extended Data Fig. 6), and there was no indication of variation among taxa in this result. Even though our study explicitly focused on local-scale variation in biodiversity within remnant fragments of different sizes (whereas ref. 26 examined landscape-level variation in biodiversity across fragmented habitats), these very different approaches both show stronger fragmentation effects in tropical areas.
A general feature of natural landscapes is that species are typically distributed non-randomly, and habitats are often destroyed non-randomly. As a result, it is possible that habitat fragments left remaining in the landscape are qualitatively different from those that were destroyed, and that diversity responses to habitat destruction can emerge from geometric, as well as demographic, effects27. For example, if a habitat fragment that remains after habitat destruction is of low quality, it may have less diversity even if there were no shifts in species demography following habitat destruction. If this were true, the patterns we observed would still be consistent with the ecosystem decay hypothesis, but the mechanism would be slightly different: they would be due to spatial rather than temporal variation in local demography after habitat loss. With the data that we have available, we have no indication of the quality of habitat patches before or after the surrounding habitats were destroyed, and thus we cannot differentiate whether spatial or temporal demographic variation (or both) underpin the observed ecosystem decay. Nevertheless, we argue that scenarios in which lower-quality patches are left behind (for example, if habitat destruction favours the removal of habitats with better resources to extract) are equally as likely as those in which higher-quality patches remain (for example, if regulations and habitat protection favour leaving particular higher-quality habitats). If true, then our null expectation that sample-controlled diversity estimates should be equal on average in larger versus smaller fragments should remain, and that any changes to this would be a result of shifts in demographic responses of species after habitat loss.
Effects on species composition
Our focus thus far has been on biodiversity change measured as numbers and relative abundances of species, but habitat fragmentation can also alter the composition of species. For example, habitat fragmentation often disadvantages habitat specialists but can benefit (or have a lower effect on) habitat generalists28. We examined compositional change and how it related to habitat loss by partitioning total community dissimilarity between large and small fragment pairs into: (i) the turnover component, which indicates the degree to which compositional dissimilarity results from different species; and (ii) the nestedness component, which indicates how much of the dissimilarity between fragments results from species in smaller habitats being subsets of more speciose fragments. On the whole, turnover among fragments was higher than nestedness components (Extended Data Fig. 7), consistent with the idea that there was an overall shift from more-specialist to more-tolerant species in smaller habitat fragments. However, there were enough cases of high nestedness between small and large fragments that there was no overall trend in the contribution of turnover to compositional change between large and small fragments (Extended Data Fig. 7a, b). On the other hand, nestedness components between fragments increased as the difference in size of habitat fragments increased, which indicates that biodiversity loss from larger to smaller fragments contributes to nested compositional differences among those fragments (Extended Data Fig. 7c, d).
Differences across studies in the species compositional turnover may also help to explain some of the variation in biodiversity responses to the size of habitat fragments. Studies with shallower slopes (for example, those with a long time since fragmentation or with less-harsh matrices) had greater contributions of species turnover to dissimilarity (Fig. 4a), potentially buffering the richness response (for example, if more-tolerant species replaced those species that were lost). On the other hand, studies with steeper slopes (for example, with less time since fragmentation or harsher matrices) had greater contributions of nestedness (Fig. 4b), indicating that species losses contributed more to the dissimilarity.
Studies with strong richness responses have higher turnover between small fragments (that is, a negative slope between turnover and fragment size) and the highest nestedness. a, Study-level slope estimates of the relationship with fragment size for species richness and the turnover component of Jaccard’s dissimilarity (Pearson’s correlation coefficient ρ = −0.42 (−0.56, −0.27; 95% confidence interval)). b, Study-level slope estimates of the relationship with fragment size for species richness and the turnover component of Jaccard’s dissimilarity (Pearson’s correlation coefficient ρ = 0.48 (0.34, 0.61; 95% confidence intervals)). Solid lines show linear models fit with ordinary least squares to aid visualization of relationships.
Discussion
Our results strongly support the notion of ecosystem decay; that is, the biodiversity loss that results from habitat loss is greater than that expected if species losses result from passive sampling with no changes in local demography. However, we also found that species turnover slightly buffers this effect, at least in some cases. Because our results include many taxa in many systems, we cannot discern among the many different mechanisms (for example, edge effects, demographic stochasticity or reduction of dispersal), or combinations thereof, by which ecosystem decay occurs. It is also important to emphasize that our analysis is conducted strictly within habitat patches. Thus, we cannot fully resolve the current debate as to whether and how habitat fragmentation per se (that is, the spatial configuration of habitat fragments when controlling for total habitat amount) influences biodiversity at the scale of the landscape3,4,5,6: the landscape-level data required to address this question are exceedingly rare. Even though our analyses point to a consistent decline in biodiversity in the habitats that remain after habitat loss and fragmentation, it is certainly possible that—at some scales—biodiversity might be unaffected or even increase at the landscape scale after fragmentation, depending on the balance between species that favour intact habitats and those that favour matrix habitats6. Nevertheless, assuming a given habitat type continues to be lost disproportionately, and that a number of species tend to specialize on the declining habitat type, then our results predict that the loss of biodiversity under an ecosystem decay scenario will extend beyond that expected from passive sampling.
Our results point to a critical limitation when forecasting the biodiversity loss expected due to habitat loss using the logic of the species–area relationship or the related relationship between endemic species and area (the endemics–area relationship)11,12,18. Any such modelling implicitly assumes that the underlying shape of the species–area or endemics–area relationships depends only on habitat geometry and the distribution of species before the habitats are destroyed, and thus implicitly assumes that species loss follows the passive sampling hypothesis (even if species distributions and habitat losses are not random per se). Improvements to these forecasts allow that species may persist in the matrix after habitat loss29, but still assume that the distribution and demography of species in the remaining habitats is the same regardless of the amount of surrounding habitat. Because we find support for the ecosystem decay hypothesis (showing that biodiversity loss from habitat loss exceeds that which would have been expected simply from passive sampling of lost habitats), forecasts that use tools based on the species–area or endemics–area relationships will underestimate longer-term patterns of biodiversity loss. We can apply the results of our global synthesis to quantify proportional species loss with habitat loss under the endemics–area relationship (passive sampling) and ecosystem decay models, which has been attempted previously with limited data and much speculation30. In Extended Data Fig. 8, we show how one could begin to predict species loss with habitat loss using a version of the endemics–area relationship parameterized with data from our synthesis, and compare the expectations of biodiversity loss to those that would emerge if these demographic changes were not incorporated. This provides an important step towards making more realistic forecasts of biodiversity loss, which will pave the way for the development of more accurate tools for conservation-based management. At the same time, our results suggest that restoration in habitat matrices may moderate the pervasive ecosystem decay that we observed in our synthetic analysis.
Methods
No statistical methods were used to predetermine sample size. The experiments were not randomized and investigators were not blinded to allocation during experiments and outcome assessment.
Literature search and data inclusion
Full details for the search criteria and process for developing the dataset analysed have previously been published22. In brief, the data needed for our analyses require data on the numbers of individuals of each species from each fragment of different sizes, as well as information on sampling design, numbers of samples (that is, plots, transects, traps, mist nets and so on) and sampling effort (that is, plot areas, transect length, trapping time, net lengths and so on) in each fragment. Because such data are exceedingly difficult to find, we did not restrict ourselves to formally standardized literature searches, but rather attempted to find as many datasets as possible. We also consulted important review papers (for example, refs. 4,23,29,31,32,33,34,35) for possible data sources, including forward citation searches. By extracting data from publications, appendices, and online repositories from a number of studies, and contacting authors directly for many more, we collated data from 117 studies for our purpose15,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142. In addition, we used data from six studies in the PREDICTS database143 that met our inclusion criteria144,145,146,147,148, giving us a total of 123 studies. Here, ‘study’ refers to sampling at a given time and place on a specific taxonomic group. If two or more taxonomic groups sampled in different ways were presented in the same paper, they were treated as two separate studies (hence the discrepancy between the number of studies and references).
Explanatory variables used for analysis
We also recorded metadata for each dataset, several of which we examined as covariates in the analysis. Specifically, we used absolute latitude and the following categories as covariates.
Taxon groupings
This included plants, invertebrates (mostly insects), birds, mammals and a category that included reptiles and amphibians (because they were often sampled together, and there were not enough studies of each for separate analyses).
Continent (or other geographical grouping)
We classified studies as occurring in Africa, Asia, Central America (we included Mexico here for biogeographical, rather than political, reasons), Europe, North America, Oceania or South America.
Time size fragmentation
We classified studies as ‘recent’ (fragmentation occurred less than 20 years previous to the study), ‘intermediate’ (fragmentation occurred 20–100 years previous to the study) and ‘long’ (fragmentation occurred more than 100 years previous to the study).
Matrix quality
We categorized studies on the basis of three qualitative categories of matrix quality: ‘light’, in which remnant habitat fragments were surrounded by relatively hospitable habitats (for example, secondary forest or shade-grown coffee); ‘intermediate’, in which remnant habitats occurred within a matrix of lightly used habitats (for example, forests surrounded by early successional habitat or livestock grazing); and ‘harsh’, in which remnant habitats occurred within a matrix of largely inhospitable matrix (for example, intense agriculture or water when reservoirs were built to create islands).
Vegetation type
We categorized the vegetation type within the habitat fragments as forest (n = 105), shrubland (n = 14), grassland (n = 3) and wetland (n = 1). As the majority of studies were from forested areas, we do not present the study-level variation in slope estimates for this categorization (which did not show any meaningful variation).
Standardization and estimation of biodiversity parameters
We made no attempt to standardize sampling across studies because of the inherent limitations of doing so when sampling methodology varies. Instead, we calculated a series of biodiversity indices that allow standardized comparisons of biodiversity differences among fragments within studies, which was crucial for evaluating the hypotheses overviewed in Fig. 1.
Sampling designs fell into one of three broad categories: (i) standardized sampling with equal numbers of samples (often one, but sometimes several) and thus equal sampling effort among habitat fragments (n = 50 studies with this kind of data structure); (ii) unequal number of standardized samples (typically more samples in larger fragments), but where data were available for each sample such that average per sample metrics could be calculated for each habitat fragment (n = 26 studies with this kind of data structure); (iii) unequal sampling effort among fragments, because the number of standardized samples varied among fragments, and the data were not available on a per sample basis (that is, they were pooled across all samples in a fragment), or because the effort per sample varied among the fragments (for example, different length of transects in different fragments) (n = 47 studies with this kind of data structure).
For each habitat fragment, we calculated several biodiversity indices that reflect different aspects of the total and relative species abundances, as well as aspects of sample completeness20,21. When there were several samples in a given fragment, we took the average of the calculated indices to get a fragment-level per-sample estimate.
The standardized number of individuals (Nstd) for each habitat fragment is typically the average number of individuals for a constant sampling effort. For studies in sample design (i), Nstd was the number of individuals per sample in each fragment, or the average number of individuals when there were multiple samples per fragment. For studies in sample design (ii), Nstd was the average number of individuals per sample in each fragment. For studies in sample design (iii), we calculated Nstd by first calculating the relative sampling effort (SErel) for each fragment by dividing the total sampling effort in that fragment by the minimum sampling effort per fragment across all fragments in the study, and then second by dividing the observed number of individuals across all samples in that fragment (Nobs) by the relative sampling effort of that fragment (that is, Nstd = Nobs/SErel, rounded to the nearest integer).
For each fragment, we calculated the number of species expected for a given constant sampling effort, which we call standardized species richness (Sstd). For studies in sample design (i) or (ii), Sstd was the average of the observed richness per sample in each fragment. For studies in sample design (iii), an individual-based rarefaction curve was created for each fragment (based on, for example, refs. 149,150), and Sstd was calculated as the expected species richness for the standardized number of individuals (Nstd) in each fragment.
To measure standardized evenness, we quantified the effective number of species using the conversion of Hurlbert’s151 probability of interspecific encounter (SPIE) (PIE = 1 – Simpson’s diversity)152. Here, we calculate SPIE per sample for each fragment (sample design (i) and (ii)). For studies in sample design (iii), we calculated SPIE using the full data of each fragment. Although PIE (and SPIE) can be influenced by sample size at very low sampling effort, it quickly stabilizes to provide a relatively unbiased estimate of relative species abundances in a community153. We hereafter refer to SPIE as ‘standardized evenness’ for simplicity to indicate that it is influenced by the relative abundances of species151,152,153, although we recognize that the term evenness is variously used in the literature. Furthermore, we compare results when only including studies with sample designs (i) and (ii) (n = 76 studies) to those that also include sample design (iii) (n = 47 studies) (see ‘Sensitivity analysis’).
Rarefied richness (Sn), the number of species for a standardized number of individuals149, was calculated according to a previously published protocol150. The reference number of individuals for Sn was the minimum of two times the smallest number of individuals per sample and the maximum number of individuals per sample among all fragments in the study.
Coverage standardized richness (Scov); this measure standardizes comparisons of species richness with respect to sample completeness154. We derived the reference coverage for comparisons among fragments following previously published suggestions150. Accordingly, the reference coverage for Scov was the minimum of the maximum coverage per sample and the minimum of the sample-scale coverage with two times the observed numbers of individuals per sample.
Asymptotic richness for each fragment was calculated using the Chao 1 estimator (Schao)155.
SPIE, Sn and Schao were calculated using the R package mobr156, and Scov was calculated from the R package iNEXT157.
To account for uncertainty in our biodiversity indices that results from sampling of fragments, we resampled 200 abundance distributions by resampling the observed number of individuals with replacement from the observed species abundance distributions and calculated all 5 biodiversity indices from the 200 resamples. For each analysis, we used the mean of each biodiversity index across the 200 resamples for each fragment in each study.
Dissimilarity partitioning
We calculated the pairwise dissimilarity in community composition between pairs of fragments within a study158 by partitioning the incidence-based dissimilarity metric (Jaccard) into turnover and nestedness components159. For each fragment pair, we also partitioned the abundance-based (Ruzicka) dissimilarity metric into balanced variation in species abundances (the abundances of some species decline and other species increase from one fragment to the other) and abundance gradient (the abundances of all species decline or increase equally from one fragment to the other) components160. Fragment pair calculations of the dissimilarity partitions were calculated for standardized sampling effort. For studies with the same number of samples in each fragment (study design (i)), we directly calculated the dissimilarity and its partitioning for all fragment pairs. When there were different numbers of samples in fragments and sample-level data (study design (ii)), we implemented a resampling approach for fragment pairs that differed in their number of samples. We randomly selected one sample from each fragment and calculated dissimilarity and its partitioning, and repeated this procedure 200 times to calculate the average values for the dissimilarity partitions across the resamples for each fragment pair. For fragment pairs in study design (iii), we used our calculations of the relative sampling effort (SErel) as well as the standardized number of individuals (Nstd) for each habitat fragment. According to the calculation of standardized species richness (see ‘Standardization and estimation of biodiversity parameters’), we resampled Nstd individuals from the abundance distributions of each fragment in the pair and calculated the dissimilarity partitions from the resampled data. This procedure was repeated 200 times to calculate the average values for the dissimilarity partitions across the resamples for each fragment pair. We used the function beta.div.comp() function from the R package adespatial for dissimilarity partitioning calculations161.
Establishment of the null expectation
Although Fig. 1 illustrates the expectation for passive sampling and the alternative hypotheses when habitat size varies, the uneven distribution of species across the landscape can influence habitat fragmentation effects27. This might lead some to wonder whether the same types of spatial structuring might also influence patterns expected at the within-patch level where sampling effort was held constant. We used a spatial simulation model to show that our null expectation under passive sampling (that is, a zero slope) was robust across different types of spatial aggregation pattern (Extended Data Fig. 1). The regional species pool was set to have 40,000 individuals, 2,000 species and a log-normal species abundance distribution with a coefficient of variation equal to 1. We examined three different levels of within species aggregation via a Thomas cluster process; each species had a single cluster, and the degree of intraspecific aggregation was controlled by setting the displacement of individuals of each species from the single cluster (or mother point) to values that resulted in either an approximately random distribution (sigma = 1), or increasingly aggregated spatial distributions (sigma = 0.1, 0.02). To avoid edge effects, fragments were randomly placed (without overlap) within the central square unit of the landscape (fragment area = {0.0005, 0.005, 0.05, 0.5}), and samples were taken from a single quadrat (area = 0.0001) randomly placed within each fragment. We quantified total numbers of individuals, species richness and evenness (SPIE) for each sample, and fit ordinary least-squares linear models to log-transformed diversity (or total numbers of individuals) as a function of log-transformed fragment size. This process was repeated 2,000 times, and we plotted the slope coefficients from the linear models to illustrate the expectation of the passive sampling hypothesis (Extended Data Fig. 1). All simulations were run using the R package mobsim162.
Statistical analyses
Although many ecological syntheses use meta-analytic tools to compare published estimates163, our approach using underlying assemblage-level data enabled us to directly analyse data across studies164. To examine the relationship between biodiversity indices and fragment size, we fit two sets of hierarchical generalized linear models to the data: one to model each of the biodiversity indices described in ‘Standardization and estimation of biodiversity parameters’ as a function of fragment size; the second to explicitly account for uncertainty in the expected levels of biodiversity under the passive sampling hypothesis (see ‘Estimating and analysing uncertainty’). As the results were qualitatively consistent between the two sets of models, we present the relationships between biodiversity indices and fragment size in the main text, and show the robustness of our results to the inclusion of uncertainty surrounding the passive sampling hypothesis in Extended Data Fig. 3.
In several studies, large habitat fragments were labelled as continuous, without provision of the actual size of these areas. For these cases, we imputed the size of the continuous habitat to be ten times as large as the next largest habitat fragment size reported in the respective study. However, we also used different multiplicative factors to assess the robustness of our results to this assumption (see ‘Sensitivity analysis’).
Fragment size was log-transformed and centred by subtracting the mean (log) fragment size of each study from the size of each fragment before fitting models, and back-transformed for presentation in the figures. For each biodiversity measure, we estimated the overall relationship with fragment size, and also allowed the slope and intercept of the relationship to vary for each study in the dataset (that is, the slope and intercept were estimated as fixed effects, as well as random effects for each study). Following standardization, none of our response variables remained with exclusively integer values, which precluded the use of Poisson or the negative binomial error distributions. Because all responses were constrained to be positive, we used the log-normal error distribution and an identity link function in the analysis of biodiversity indices.
We visualized variation in effects of fragment area on biodiversity associated with taxon group, geographical region, time since fragmentation and matrix quality using the posterior distributions of study-level slope estimates. Alternatively, we could have fit models with interactions between the covariates and fragment area, but—owing to data limitations—these would have been restricted to two-way interactions. However, it is likely that multiple covariates combine and contribute to variation in the effects of fragmentation on biodiversity. Because each study represents a combination of all these covariates in our analyses, examining study-level variation allows us to efficiently simultaneously adjust for and characterize relationships among multiple covariates and fragmentation effects on biodiversity. However, to assess the robustness of patterns found in the study-level variation for the different groups, we fit models with two-way interactions between fragment size and taxon group, geographic region, time since fragmentation and matrix quality, respectively (Extended Data Tables 1–3).
Estimating and analysing uncertainty
The passive sampling hypothesis posits that biodiversity in an effort-standardized sample in a small fragment is the same as the biodiversity in a corresponding effort-standardized sample in a large fragment (Fig. 1). That is, the expected value of the standardized biodiversity metrics under the passive sampling hypothesis in smaller fragments is equal to that in the largest fragment for each study (Fig. 1). To account for uncertainty associated with the passive sampling hypothesis, we calculated z-scores ((observed − expected)/s.d.(expected)) for each biodiversity index and for each fragment smaller than the largest in each study. To derive the z-score, we calculated the expected value and its s.d. for each biodiversity index from the resampling process for the largest fragment in each study. We used the observed value for each of the biodiversity indices for each smaller fragment to calculate the z-scores. As the standardized number of individuals was defined as Nstd = Nobs/SErel, there was no estimate of uncertainty, and z-scores were calculated only for the five biodiversity indices.
We modelled the z-scores as a function of fragment size using models similar to those described in ‘Statistical analyses’. Fragment size was log-transformed and centred by subtracting the mean (log) fragment size from each observation before fitting models. For the z-score of each measure, we estimated the overall relationship with fragment size, and also allowed the slope and intercept of the relationship to vary for each study in the dataset (that is, the slope and intercept were estimated as fixed effects, as well as random effects for each study). z-scores were distributed around zero for all metrics, and comparisons of models fit with Gaussian, asymmetric Laplacian (back-to-back exponential) or Student’s error distribution showed strong support for assuming error to follow a Student distribution, and it was used for all analyses. Similar to the models of biodiversity indices, we found positive slopes (95% credible intervals did not overlap zero) for the relationship between z-scores of all metrics and fragment size (Extended Data Fig. 3). This suggests that our inference of pervasive ecosystem decay from Fig. 2 is robust to the inclusion of uncertainty in the passive sampling hypothesis.
Latitudinal variation
To examine whether there was a latitudinal gradient in the slope of the relationship between fragment size and biodiversity indices, we modelled the study-level slope estimates as a function of absolute latitude. We used a random-effects meta-analytic model that incorporated uncertainty associated with the study-level slope estimates (that is, the standard error of the posterior distributions) into our analysis; study was included as a random effect, and the model assumed a Gaussian error distribution and an identity link function. To examine whether the latitudinal gradients differed between taxa, we fit a second model that included an interaction between absolute latitude and taxon group.
Compositional dissimilarity
To examine the relationship between dissimilarity components among fragments and differences in fragment size, we modelled the turnover and nestedness components of Jaccard dissimilarity159, and the balanced abundance and abundance gradient components of Ruzicka dissimilarity160, as a function of the log-transformed ratio of the sizes for each fragment pair (that was centred before model fitting). We fit models that estimated the overall relationship, and also allowed both the slope and intercept to vary for each study. Models assumed zero–one inflated beta distributions and logit-link functions for the turnover components, and zero inflated beta distributions (and logit-link functions) for the nestedness components.
Model fitting
For Bayesian inference and estimates of uncertainty, all models were fit using the Hamiltonian Monte Carlo (HMC) sampler Stan165 and coded using the brms package166. All models were fit with 4 chains and 2,000 iterations, with 1,000 used as a warm-up. We used weakly regularizing priors and visual inspection of the HMC chains showed excellent convergence.
Sensitivity analysis
Forty-seven studies had unequal numbers of samples across fragments and data were pooled (that is, sample design (iii)). This pooling of samples means that data from larger fragments, which encompass a larger sampling extent, will also encompass more heterogeneity. This can, in turn, influence diversity estimates. To examine whether our inferences of ecosystem decay were sensitive to the inclusion of studies with sample design (iii), we removed those studies and refit the models to the remaining data (n = 79 studies for which sampling extents were comparable).
For all analyses, we needed to assign a fragment size to fragments labelled as continuous without a specified size. In the main results, we calculated their size as ten times larger than the next largest fragment in a given study, which was an arbitrary—but reasonable—assumption (that is, no habitat is ever continuous). However, to assess the sensitivity of our results, we varied this multiplier so that continuous fragments were either 2 or 100 times larger than the next largest fragment.
The calculations of the biodiversity parameters used here typically require counts of individuals of each species (that is, integer abundance values). However, a few of the studies in our dataset report non-integer abundance values for samples (for example, based on averages for subsamples within each sample, where the subsample data were not available). In the results discussed in the main text, we calculated the biodiversity parameters directly from these non-integer abundances. To assess the potential sensitivity of our findings to the use of non-integer abundances, we implemented two additional methods. In one, we rounded all abundances to the next largest integer. In this way, we favoured rare species with respect to their relative abundance. The second method divided all abundances by the minimum abundance in the respective study. This resulted in the lowest abundance equalling 1 after scaling, but all relative abundances were maintained.
Adding ecosystem decay to extinction forecasts
Many published models of biodiversity loss implicitly or explicitly assume that scaling relationships of biodiversity are constant even when habitat is lost (that is, assume passive sampling). For example, forecasts based on the species–area relationship or the endemics–area relationship implicitly assume that passive sampling is the primary process by which habitat size influences species diversity (or, rather, only consider immediate extinctions, and not extinction debt). Here we illustrate how estimates such as ours—which quantify the degree to which ecosystem decay leads to extinctions beyond passive sampling—could improve forecasts of biodiversity loss due to habitat loss.
If smaller fragments are simply passive samples from the assemblages found in larger areas, then the expected number of species lost with a loss of area a (from a total area, A) can be calculated using the endemics–area relationship12:
with species lost,
Ni is the total abundance of species i, and S is the total number of species in the region A.
We calculate the additional species lost under ecosystem decay using the slope estimated from our model fit to standardized species richness (β = 0.06). We assume that species lost owing to passive sampling equals Spassive, and that additional species lost owing to ecosystem decay are proportional to the area lost a, to calculate Sdecay = Spassive + βa.
To illustrate the difference between species loss owing to random expectations and the empirically derived loss including ecosystem decay, we simulated a regional assemblage using the mobsim R package162. We calculated Spassive and Sdecay using the equations described above. We set A = 10,000, S = 1,000 and simulated species in A using a log-normal species abundance distribution (total community N = 10,000, coefficient of variation = 1) randomly distributed in space. We examined area lost, a, that ranged from 1 to A, and show the number of extinctions as a function of the area (a) lost (Extended Data Fig. 8).
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.
Data availability
All of the data used in this analysis are open access and available in ref. 22 (117 of the datasets) and ref. 143 (5 of the datasets). Raw data (before standardization) are available from GitHub (https://github.com/FelixMay/FragFrame_1), and are mirrored on Zenodo (https://doi.org/10.5281/zenodo.3862409).
Code availability
The R code used for standardizing the data and doing the analyses presented here are available from GitHub (https://github.com/FelixMay/FragFrame_1), and are mirrored on Zenodo (https://doi.org/10.5281/zenodo.3862409).
References
Pimm, S. L. et al. The biodiversity of species and their rates of extinction, distribution, and protection. Science 344, 1246752 (2014).
Díaz, S. et al. Pervasive human-driven decline of life on Earth points to the need for transformative change. Science 366, eaax3100 (2019).
Haddad, N. M. et al. Habitat fragmentation and its lasting impact on Earth’s ecosystems. Sci. Adv. 1, e1500052 (2015).
Fahrig, L. Ecological responses to habitat fragmentation per se. Annu. Rev. Ecol. Evol. Syst. 48, 1–23 (2017).
Fletcher, R. J. Jr et al. Is habitat fragmentation good for biodiversity? Biol. Conserv. 226, 9–15 (2018).
Fahrig, L. et al. Is habitat fragmentation bad for biodiversity? Biol. Conserv. 230, 179–186 (2019).
Connor, E. F. & McCoy, E. D. The statistics and biology of the species–area relationship. Am. Nat. 113, 791–833 (1979).
Yaacobi, G., Ziv, Y. & Rosenzweig, M. L. Habitat fragmentation may not matter to species diversity. Proc. R. Soc. Lond. B 274, 2409–2412 (2007).
Lovejoy, T. E. et al. in Extinctions (ed. Nitecki, M. H.) 295–325 (Univ. of Chicago Press, 1984).
Hanski, I., Zurita, G. A., Bellocq, M. I. & Rybicki, J. Species-fragmented area relationship. Proc. Natl Acad. Sci. USA 110, 12715–12720 (2013).
Pimm, S. L. & Askins, R. A. Forest losses predict bird extinctions in eastern North America. Proc. Natl Acad. Sci. USA 92, 9343–9347 (1995).
He, F. & Hubbell, S. P. Species–area relationships always overestimate extinction rates from habitat loss. Nature 473, 368–371 (2011).
Terborgh, J. et al. Ecological meltdown in predator-free forest fragments. Science 294, 1923–1926 (2001).
Laurance, W. F. et al. The fate of Amazonian forest fragments: a 32-year investigation. Biol. Conserv. 144, 56–67 (2011).
Gibson, L. et al. Near-complete extinction of native small mammal fauna 25 years after forest fragmentation. Science 341, 1508–1510 (2013).
Thomas, C. D. et al. Extinction risk from climate change. Nature 427, 145–148 (2004).
Tilman, D. et al. Future threats to biodiversity and pathways to their prevention. Nature 546, 73–81 (2017).
Halley, J. M., Sgardeli, V. & Monokrousos, N. Species–area relationships and extinction forecasts. Ann. NY Acad. Sci. 1286, 50–61 (2013).
Bueno, A. S. & Peres, C. A. Patch-scale biodiversity retention in fragmented landscapes: reconciling the habitat amount hypothesis with the island biogeography theory. J. Biogeogr. 46, 621–632 (2019).
Chase, J. M. et al. A framework for disentangling ecological mechanisms underlying the island species–area relationship. Front. Biogeogr. 11, e40844 (2019).
Chase, J. M. et al. Embracing scale-dependence to achieve a deeper understanding of biodiversity and its change across communities. Ecol. Lett. 21, 1737–1751 (2018).
Chase, J. M. et al. FragSAD: a database of diversity and species abundance distributions from habitat fragments. Ecology 100, e02861 (2019).
Ewers, R. M. & Didham, R. K. Confounding factors in the detection of species responses to habitat fragmentation. Biol. Rev. Camb. Philos. Soc. 81, 117–142 (2006).
Tilman, D., May, R. M., Lehman, C. L. & Nowak, M. A. Habitat destruction and the extinction debt. Nature 371, 65–66 (1994).
Jackson, S. T. & Sax, D. F. Balancing biodiversity in a changing environment: extinction debt, immigration credit and species turnover. Trends Ecol. Evol. 25, 153–160 (2010).
Betts, M. G. et al. Extinction filters mediate the global effects of habitat fragmentation on animals. Science 366, 1236–1239 (2019).
Chisholm, R. A. et al. Species–area relationships and biodiversity loss in fragmented landscapes. Ecol. Lett. 21, 804–813 (2018).
Matthews, T. J., Cottee-Jones, H. E. & Whittaker, R. J. Habitat fragmentation and the species–area relationship: a focus on total species richness obscures the impact of habitat loss on habitat specialists. Divers. Distrib. 20, 1136–1146 (2014).
Koh, L. P., Lee, T. M., Sodhi, N. S. & Ghazoul, J. An overhaul of the species–area approach for predicting biodiversity loss: incorporating matrix and edge effects. J. Appl. Ecol. 47, 1063–1070 (2010).
Halley, J. M., Monokrousos, N., Mazaris, A. D., Newmark, W. D. & Vokou, D. Dynamics of extinction debt across five taxonomic groups. Nat. Commun. 7, 12283 (2016).
Saunders, D. A., Hobbs, R. J. & Margules, C. R. Biological consequences of ecosystem fragmentation: a review. Conserv. Biol. 5, 18–32 (1991).
Andrén, H. Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable habitat: a review. Oikos 71, 355–366 (1994).
Debinski, D. M. & Holt, R. D. A survey and overview of habitat fragmentation experiments. Conserv. Biol. 14, 342–355 (2000).
Fahrig, L. Effects of habitat fragmentation on biodiversity. Annu. Rev. Ecol. Evol. Syst. 34, 487–515 (2003).
Jones, I. L., Bunnefeld, N., Jump, A. S., Peres, C. A. & Dent, D. H. Extinction debt on reservoir land-bridge islands. Biol. Conserv. 199, 75–83 (2016).
Aguiar, W. M. D. & Gaglianone, M. C. Euglossine bee communities in small forest fragments of the Atlantic Forest, Rio de Janeiro state, southeastern Brazil (Hymenoptera, Apidae). Rev. Bras. Entomol. 56, 210–219 (2012).
Aizen, M. A. & Feinsinger, P. Habitat fragmentation, native insect pollinators, and feral honey bees in Argentine “Chaco Serrano”. Ecol. Appl. 4, 378–392 (1994).
Almeida-Gomes, M. & Rocha, C. F. D. Diversity and distribution of lizards in fragmented Atlantic Forest landscape in southeastern Brazil. J. Herpetol. 48, 423–429 (2014).
Andresen, E. Effect of forest fragmentation on dung beetle communities and functional consequences for plant regeneration. Ecography 26, 87–97 (2003).
Báldi, A. & Kisbenedek, T. Orthopterans in small steppe patches: an investigation for the best-fit model of the species–area curve and evidences for their non-random distribution in the patches. Acta Oecol. 20, 125–132 (1999).
Baz, A. & Garcia-Boyero, A. The SLOSS dilemma: a butterfly case study. Biodivers. Conserv. 5, 493–502 (1996).
Bell, K. E. & Donnelly, M. A. Influence of forest fragmentation on community structure of frogs and lizards in northeastern Costa Rica. Conserv. Biol. 20, 1750–1760 (2006).
Benedick, S. et al. Impacts of rain forest fragmentation on butterflies in northern Borneo: species richness, turnover and the value of small fragments. J. Appl. Ecol. 43, 967–977 (2006).
Benítez-Malvido, J. et al. The multiple impacts of tropical forest fragmentation on arthropod biodiversity and on their patterns of interactions with host plants. PLoS ONE 11, e0146461 (2016).
Berg, Å. Diversity and abundance of birds in relation to forest fragmentation, habitat quality and heterogeneity. Bird Study 44, 355–366 (1997).
Bernard, E. & Fenton, M. B. Bats in a fragmented landscape: species composition, diversity and habitat interactions in savannas of Santarém, Central Amazonia, Brazil. Biol. Conserv. 134, 332–343 (2007).
Bolger, D. T. et al. Response of rodents to habitat fragmentation in coastal southern California. Ecol. Appl. 7, 552–563 (1997).
Bossart, J. L. et al. Richness, abundance, and complementarity of fruit-feeding butterfly species in relict sacred forests and forest reserves of Ghana. Biodivers. Conserv. 15, 333–359 (2006).
Bossart, J. L. & Antwi, J. B. Limited erosion of genetic and species diversity from small forest patches: sacred forest groves in an Afrotropical biodiversity hotspot have high conservation value for butterflies. Biol. Conserv. 198, 122–134 (2016).
Bragagnolo, C. et al. Harvestmen in an Atlantic forest fragmented landscape: evaluating assemblage response to habitat quality and quantity. Biol. Conserv. 139, 389–400 (2007).
Brosi, B. J., Daily, G. C., Shih, T. M., Oviedo, F. & Durán, G. The effects of forest fragmentation on bee communities in tropical countryside. J. Appl. Ecol. 45, 773–783 (2008).
Brosi, B. J. The effects of forest fragmentation on euglossine bee communities (Hymenoptera: Apidae: Euglossini). Biol. Conserv. 142, 414–423 (2009).
Cabrera-Guzmán, E. & Reynoso, V. H. Amphibian and reptile communities of rainforest fragments: minimum patch size to support high richness and abundance. Biodivers. Conserv. 21, 3243–3265 (2012).
Cadotte, M. W., Franck, R., Reza, L. & Lovett-Doust, J. Tree and shrub diversity and abundance in fragmented littoral forest of southeastern Madagascar. Biodivers. Conserv. 11, 1417–1436 (2002).
Carneiro, M. S., Campos, C. C., Ramos, F. N. & Dos Santos, F. A. Spatial species turnover maintains high diversities in a tree assemblage of a fragmented tropical landscape. Ecography 7, e01500 (2016).
Cayuela, L., Golicher, D. J., Benayas, J. M. R., González-Espinosa, M. & Ramírez-Marcial, N. Fragmentation, disturbance and tree diversity conservation in tropical montane forests. J. Appl. Ecol. 43, 1172–1181 (2006).
Chiarello, A. G. Effects of fragmentation of the Atlantic forest on mammal communities in south-eastern Brazil. Biol. Conserv. 89, 71–82 (1999).
Cosson, J. F. et al. Ecological changes in recent land-bridge islands in French Guiana, with emphasis on vertebrate communities. Biol. Conserv. 91, 213–222 (1999).
Dami, F. D., Mwansat, G. S. & Manu, S. A. The effects of forest fragmentation on species richness on the Obudu Plateau, south-eastern Nigeria. Afr. J. Ecol. 51, 32–36 (2013).
Dauber, J., Bengtsson, J. & Lenoir, L. Evaluating effects of habitat loss and land-use continuity on ant species richness in seminatural grassland remnants. Conserv. Biol. 20, 1150–1160 (2006).
Davies, R. G. et al. Environmental and spatial influences upon species composition of a termite assemblage across neotropical forest islands. J. Trop. Ecol. 19, 509–524 (2003).
de La Sancha, N. U. Patterns of small mammal diversity in fragments of subtropical interior Atlantic forest in eastern Paraguay. Mammalia 78, 437–449 (2014).
de Souza, O. F. F. & Brown, V. K. Effects of habitat fragmentation on Amazonian termite communities. J. Trop. Ecol. 10, 197 (1994).
Dickman, C. R. Habitat fragmentation and vertebrate species richness in an urban environment. J. Appl. Ecol. 24, 337–351 (1987).
Didham, R. K., Hammond, P. M., Lawton, J. H., Eggleton, P. & Stork, N. E. Beetle species responses to tropical forest fragmentation. Ecol. Monogr. 68, 295–323 (1998).
Ding, Z., Feeley, K. J., Wang, Y., Pakeman, R. J. & Ding, P. Patterns of bird functional diversity on land-bridge island fragments. J. Anim. Ecol. 82, 781–790 (2013).
Dixo, M. & Metzger, J. P. Are corridors, fragment size and forest structure important for the conservation of leaf-litter lizards in a fragmented landscape? Oryx 43, 435 (2009).
Dominguez-Haydar, Y. & Armbrecht, I. Response of ants and their seed removal in rehabilitation areas and forests at El Cerrejón coal mine in Colombia. Restor. Ecol. 19, 178–184 (2011).
Echeverría, C. et al. Impacts of forest fragmentation on species composition and forest structure in the temperate landscape of southern Chile. Glob. Ecol. Biogeogr. 16, 426–439 (2007).
Edwards, D. P. et al. Wildlife-friendly oil palm plantations fail to protect biodiversity effectively. Conserv. Lett. 3, 236–242 (2010).
Estrada, A. & Coates-Estrada, R. Bats in continuous forest, forest fragments and in an agricultural mosaic habitat-island at Los Tuxtlas, Mexico. Biol. Conserv. 103, 237–245 (2002).
Estrada, A. & Coates-Estrada, R. Dung beetles in continuous forest, forest fragments and in an agricultural mosaic habitat island at Los Tuxtlas, Mexico. Biodivers. Conserv. 11, 1903–1918 (2002).
Filgueiras, B. K. C., Iannuzzi, L. & Leal, I. R. Habitat fragmentation alters the structure of dung beetle communities in the Atlantic forest. Biol. Conserv. 144, 362–369 (2011).
da Fonseca, G. A. B. & Robinson, J. G. Forest size and structure: competitive and predatory effects on small mammal communities. Biol. Conserv. 53, 265–294 (1990).
Fujita, A. et al. Effects of forest fragmentation on species richness and composition of ground beetles (Coleoptera: Carabidae and Brachinidae) in urban landscapes. Entomol. Sci. 11, 39–48 (2008).
Gavish, Y., Ziv, Y. & Rosenzweig, M. L. Decoupling fragmentation from habitat loss for spiders in patchy agricultural landscapes. Conserv. Biol. 26, 150–159 (2012).
Giladi, I. et al. Scale-dependent determinants of plant species richness in a semi-arid fragmented agro-ecosystem. J. Veg. Sci. 22, 983–996 (2011).
Giraudo, A. R. et al. Comparing bird assemblages in large and small fragments of the Atlantic forest hotspots. Biodivers. Conserv. 17, 1251–1265 (2008).
Gonçalves-Souza, T., Matallana, G. & Brescovit, A. D. Effects of habitat fragmentation on the spider community (Arachnida, Araneae) in three Atlantic forest remnants in southeastern Brazil. Rev. Iber. Aracnol. 16, 35–42 (2008).
Goodman, S. M. & Rakotondravony, D. The effects of forest fragmentation and isolation on insectivorous small mammals (Lipotyphla) on the Central High Plateau of Madagascar. J. Zool. 250, 193–200 (2000).
Guadagnin, D. L., Peter, Â. S., Perello, L. F. C. & Maltchik, L. Spatial and temporal patterns of waterbird assemblages in fragmented wetlands of southern Brazil. Waterbirds 28, 261–272 (2005).
Halme, E., Niemela, J. & Haime, E. Carabid beetles in fragments of coniferous forest. Ann. Zool. Fenn. 30, 17–30 (1993).
Henry, M., Pons, J.-M. & Cosson, J.-F. Foraging behaviour of a frugivorous bat helps bridge landscape connectivity and ecological processes in a fragmented rainforest. J. Anim. Ecol. 76, 801–813 (2007).
Horváth, R. et al. Spiders are not less diverse in small and isolated grasslands, but less diverse in overgrazed grasslands: a field study (East Hungary, Nyirseg). Agric. Ecosyst. Environ. 130, 16–22 (2009).
Jauker, F., Jauker, B., Grass, I., Steffan-Dewenter, I. & Wolters, V. Partitioning wild bee and hoverfly contributions to plant–pollinator network structure in fragmented habitats. Ecology 100, e02569 (2019).
Jung, J. K. et al. A comparison of diversity and species composition of ground beetles (Coleoptera: Carabidae) between conifer plantations and regenerating forests in Korea. Ecol. Res. 29, 877–887 (2014).
Jyothi, K. M. & Nameer, P. O. Birds of sacred groves of northern Kerala, India. J. Threat. Taxa 7, 8226–8236 (2015).
Kapoor, V. Effects of rainforest fragmentation and shade-coffee plantations on spider communities in the Western Ghats, India. J. Insect Conserv. 12, 53–68 (2008).
Kappes, H. et al. Response of snails and slugs to fragmentation of lowland forests in NW Germany. Landsc. Ecol. 24, 685–697 (2009).
Klein, B. C. Effects of forest fragmentation on dung and carrion beetle communities in central Amazonia. Ecology 70, 1715–1725 (1989).
Knapp, M. & Řezáč, M. Even the smallest non-crop habitat islands could be beneficial: distribution of carabid beetles and spiders in agricultural landscape. PLoS ONE 10, e0123052 (2015).
Lambert, T. D. et al. Rodents on tropical land-bridge islands. J. Zool. 260, 179–187 (2003).
Lasky, J. R. & Keitt, T. H. Abundance of Panamanian dry-forest birds along gradients of forest cover at multiple scales. J. Trop. Ecol. 26, 67–78 (2010).
de Lima, M. G. & Gascon, C. The conservation of linear forest remnants in central Amazonia. Biol. Conserv. 91, 241–247 (1999).
Lima, J. et al. Amphibians on Amazonian land-bridge islands are affected more by area than isolation. Biotropica 47, 369–376 (2015).
Lion, M. B., Garda, A. A. & Fonseca, C. R. Split distance: a key landscape metric shaping amphibian populations and communities in forest fragments. Divers. Distrib. 20, 1245–1257 (2014).
Lion, M. B., Garda, A. A., Santana, D. J. & Fonseca, C. R. The conservation value of small fragments for Atlantic forest reptiles. Biotropica 48, 265–275 (2016).
Lövei, G. L. & Cartellieri, M. Ground beetles (Coleoptera, Carabidae) in forest fragments of the Manuwatu, New Zealand: collapsed assemblages? J. Insect Conserv. 4, 239–244 (2000).
Mac Nally, R. & Brown, G. W. Reptiles and habitat fragmentation in the box-ironbark forests of central Victoria, Australia: predictions, compositional change and faunal nestedness. Oecologia 128, 116–125 (2001).
Manu, S., Peach, W. & Cresswell, W. The effects of edge, fragment size and degree of isolation on avian species richness in highly fragmented forest in West Africa. Ibis 149, 287–297 (2007).
Martensen, A. C., Ribeiro, M. C., Banks-Leite, C., Prado, P. I. & Metzger, J. P. Associations of forest cover, fragment area, and connectivity with Neotropical understory bird species richness and abundance. Conserv. Biol. 26, 1100–1111 (2012).
McCollin, D. Avian distribution patterns in a fragmented wooded landscape (North Humberside, U.K.): the role of between-patch and within-patch structure. Glob. Ecol. Biogeogr. Lett. 3, 48–62 (1993).
McIntyre, N. E. Effects of forest patch size on avian diversity. Landsc. Ecol. 10, 85–99 (1995).
Meyer, C. F. J. & Kalko, E. K. V. Assemblage-level responses of phyllostomid bats to tropical forest fragmentation: land-bridge islands as a model system. J. Biogeogr. 35, 1711–1726 (2008).
Nemésio, A. & Silveira, F. A. Orchid bee fauna (Hymenoptera: Apidae: Euglossina) of Atlantic Forest fragments inside an urban area in southeastern Brazil. Neotrop. Entomol. 36, 186–191 (2007).
Nemésio, A. & Silveira, F. A. Forest fragments with larger core areas better sustain diverse orchid bee faunas (Hymenoptera: Apidae: Euglossina). Neotrop. Entomol. 39, 555–561 (2010).
Neuschulz, E. L., Botzat, A. & Farwig, N. Effects of forest modification on bird community composition and seed removal in a heterogeneous landscape in South Africa. Oikos 120, 1371–1379 (2011).
Nogueira, A. & Pinto-da-Rocha, R. The effects of habitat size and quality on the orb-weaving spider guild (Arachnida: Araneae) in an Atlantic forest fragmented landscape. J. Arachnol. 44, 36–45 (2016).
Nufio, R. C., McClenahan, L. J. & Thurston, G. E. Determining the effects of habitat fragment area on grasshopper species density and richness: a comparison of proportional and uniform sampling methods. Insect Conserv. Divers. 2, 295–304 (2009).
Nyeko, P. Dung beetle assemblages and seasonality in primary forest and forest fragments on agricultural landscapes in Budongo, Uganda. Biotropica 41, 476–484 (2009).
Nyelele, C. et al. Woodland fragmentation explains tree species diversity in an agricultural landscape of Southern Africa. Trop. Ecol. 55, 365–374 (2014).
Owen, C. L. Mapping Biodiversity in a Modified Landscape. MSc thesis, Imperial College London (2008).
Paciencia, M. L. B. & Prado, J. Effects of forest fragmentation on pteridophyte diversity in a tropical rain forest in Brazil. Plant Ecol. 180, 87–104 (2005).
Pardini, R. Effects of forest fragmentation on small mammals in an Atlantic forest landscape. Biodivers. Conserv. 13, 2567–2586 (2004).
Pineda, E. & Halffter, G. Species diversity and habitat fragmentation: frogs in a tropical montane landscape in Mexico. Biol. Conserv. 117, 499–508 (2004).
Raheem, D. C. et al. Fragmentation and pre-existing species turnover determine land-snail assemblages of tropical rain forest. J. Biogeogr. 36, 1923–1938 (2009).
Rocha, R. et al. Consequences of a large-scale fragmentation experiment for Neotropical bats: disentangling the relative importance of local and landscape-scale effects. Landsc. Ecol. 32, 31–45 (2017).
Sam, K., Koane, B., Jeppy, S. & Novotny, V. Effect of forest fragmentation on bird species richness in Papua New Guinea. J. Field Ornithol. 85, 152–167 (2014).
Savilaakso, S., Koivisto, J., Veteli, T. O. & Roininen, H. Microclimate and tree community linked to differences in lepidopteran larval communities between forest fragments and continuous forest. Divers. Distrib. 15, 356–365 (2009).
Schnitzler, F. R. Hymenopteran Parasitoid Diversity and Tri-Trophic Interactions: The Effects of Habitat Fragmentation in Wellington, New Zealand. PhD thesis, Victoria Univ. of Wellington (2008).
Senior, M. J. M. Assessing Biodiversity and Ecosystem Functioning in Fragmented Tropical Landscapes. PhD Thesis, Univ. of York (2014).
Silva, M. P. P. & Porto, K. C. Effect of fragmentation on the community structure of epixylic bryophytes in Atlantic forest remnants in the northeast of Brazil. Biodivers. Conserv. 18, 317–337 (2009).
Silva, R. J., Storck-Tonon, D. & Vaz-de-Mello, F. Z. Dung beetle (Coleoptera: Scarabaeinae) persistence in Amazonian forest fragments and adjacent pastures: biogeographic implications for alpha and beta diversity. J. Insect Conserv. 20, 549–564 (2016).
Silveira, G. C. et al. The orchid bee fauna in the Brazilian savanna: do forest formations contribute to higher species diversity? Apidologie 46, 197–208 (2015).
Slade, E. M. et al. Life-history traits and landscape characteristics predict macro-moth responses to forest fragmentation. Ecology 94, 1519–1530 (2013).
Sridhar, H., Raman, T. S. & Mudappa, D. Mammal persistence and abundance in tropical rainforest remnants in the southern Western Ghats, India. Curr. Sci. 94, 748–757 (2008).
Stireman, J. O. III, Devlin, H. & Doyle, A. L. Habitat fragmentation, tree diversity, and plant invasion interact to structure forest caterpillar communities. Oecologia 176, 207–224 (2014).
Storck-Tonon, D. & Peres, C. A. Forest patch isolation drives local extinctions of Amazonian orchid bees in a 26 years old archipelago. Biol. Conserv. 214, 270–277 (2017).
Struebig, M. J. et al. Conservation importance of limestone karst outcrops for Palaeotropical bats in a fragmented landscape. Biol. Conserv. 142, 2089–2096 (2009).
Tellería, J. L. & Santos, T. Effects of forest fragmentation on a guild of wintering passerines: the role of habitat selection. Biol. Conserv. 71, 61–67 (1995).
Tonhasca, A., Blackmer, J. L. & Albuquerque, G. S. Abundance and diversity of euglossine bees in the fragmented landscape of the Brazilian Atlantic forest. Biotropica 34, 416–422 (2002).
Uehara-Prado, M., Brown, K. S. & Freitas, A. V. L. Species richness, composition and abundance of fruit-feeding butterflies in the Brazilian Atlantic forest: comparison between a fragmented and a continuous landscape. Glob. Ecol. Biogeogr. 16, 43–54 (2007).
Ulrich, W., Lens, L., Tobias, J. A. & Habel, J. C. Contrasting patterns of species richness and functional diversity in bird communities of east African cloud forest fragments. PLoS ONE 11, e0163338 (2016).
Usher, M. B. & Keiller, S. W. J. The Macrolepidoptera of farm woodlands: determinants of diversity and community structure. Biodivers. Conserv. 7, 725–748 (1998).
Vallan, D. Influence of forest fragmentation on amphibian diversity in the nature reserve of Ambohitantely, highland Madagascar. Biol. Conserv. 96, 31–43 (2000).
Vasconcelos, H. L., Vilhena, J. M., Magnusson, W. E. & Albernaz, A. L. M. Long-term effects of forest fragmentation on Amazonian ant communities. J. Biogeogr. 33, 1348–1356 (2006).
Vieira, M. V. et al. Land use vs. fragment size and isolation as determinants of small mammal composition and richness in Atlantic forest remnants. Biol. Conserv. 142, 1191–1200 (2009).
Vulinec, K. et al. Dung beetles and long-term habitat fragmentation in Alter do Chão, Amazônia, Brazil. Trop. Conserv. Sci. 1, 111–121 (2008).
Wang, Y., Wang, X. & Ding, P. Nestedness of snake assemblages on islands of an inundated lake. Curr. Zool. 58, 828–836 (2012).
Williams, M. R. Habitat resources, remnant vegetation condition and area determine distribution patterns and abundance of butterflies and day-flying moths in a fragmented urban landscape, south-west Western Australia. J. Insect Conserv. 15, 37–54 (2011).
Zartman, C. E. Habitat fragmentation impacts on epiphyllous bryophyte communities in Central Amazonia. Ecology 84, 948–954 (2003).
Ziter, C., Bennett, E. M. & Gonzalez, A. Functional diversity and management mediate aboveground carbon stocks in small forest fragments. Ecosphere 4, 85 (2013).
Hudson, L. N. et al. The database of the PREDICTS (Projecting Responses of Ecological Diversity In Changing Terrestrial Systems) project. Ecol. Evol. 7, 145–188 (2017).
Cáceres, N. C., Nápoli, R. P., Casella, J. & Hannibal, W. Mammals in a fragmented savannah landscape in south-western Brazil. J. Nat. Hist. 44, 491–512 (2010).
Ewers, R. M., Thorpe, S. & Didham, R. K. Synergistic interactions between edge and area effects in a heavily fragmented landscape. Ecology 88, 96–106 (2007).
Fernández, I. C. & Simonetti, J. A. Small mammal assemblages in fragmented shrublands of urban areas of Central Chile. Urban Ecosyst. 16, 377–387 (2013).
Garmendia, A., Arroyo-Rodríguez, V., Estrada, A., Naranjo, E. J. & Stoner, K. E. Landscape and patch attributes impacting medium- and large-sized terrestrial mammals in a fragmented rain forest. J. Trop. Ecol. 29, 331–344 (2013).
Stouffer, P. C., Johnson, E. I., Bierregaard, R. O. Jr & Lovejoy, T. E. Understory bird communities in Amazonian rainforest fragments: species turnover through 25 years post-isolation in recovering landscapes. PLoS ONE 6, e20543 (2011).
Gotelli, N. J. & Colwell, R. K. Quantifying biodiversity: procedures and pitfalls in the measurement and comparison of species richness. Ecol. Lett. 4, 379–391 (2001).
Chao, A. et al. Rarefaction and extrapolation with Hill numbers: a framework for sampling and estimation in species diversity studies. Ecol. Monogr. 84, 45–67 (2014).
Hurlbert, S. H. The nonconcept of species diversity: a critique and alternative parameters. Ecology 52, 577–586 (1971).
Jost, L. Entropy and diversity. Oikos 113, 363–375 (2006).
Olszewski, T. D. A unified mathematical framework for the measurement of richness and evenness within and among multiple communities. Oikos 104, 377–387 (2004).
Chao, A. & Jost, L. Coverage-based rarefaction and extrapolation: standardizing samples by completeness rather than size. Ecology 93, 2533–2547 (2012).
Chao, A. Nonparametric estimation of the number of classes in a population. Scand. J. Stat. 11, 265–270 (1984).
McGlinn, D. J. et al. Measurement of Biodiversity (MoB): a method to separate the scale-dependent effects of species abundance distribution, density, and aggregation on diversity change. Methods Ecol. Evol. 10, 258–269 (2019).
Hsieh, T. C., Ma, K. H. & Chao, A. iNEXT: an R package for rarefaction and extrapolation of species diversity (Hill numbers). Methods Ecol. Evol. 7, 1451–1456 (2016).
Marion, Z. H., Fordyce, J. A. & Fitzpatrick, B. M. Pairwise beta diversity resolves an underappreciated source of confusion in calculating species turnover. Ecology 98, 933–939 (2017).
Baselga, A. Partitioning the turnover and nestedness components of beta diversity. Glob. Ecol. Biogeogr. 19, 134–143 (2010).
Baselga, A. Separating the two components of abundance-based dissimilarity: balanced changes in abundance vs. abundance gradients. Methods Ecol. Evol. 4, 552–557 (2013).
Dray, S. et al. adespatial: multivariate multiscale spatial analysis. R package version 0.3-4 https://CRAN.R-project.org/package=adespatial (2019).
May, F., Gerstner, K., McGlinn, D. J., Xiao, X. & Chase, J. M. mobsim: an R package for the simulation and measurement of biodiversity across spatial scales. Methods Ecol. Evol. 9, 1401–1408 (2018).
Gurevitch, J., Koricheva, J., Nakagawa, S. & Stewart, G. Meta-analysis and the science of research synthesis. Nature 555, 175–182 (2018).
Purvis, A. et al. Modelling and projecting the response of local terrestrial biodiversity worldwide to land use and related pressures: the PREDICTS project. Adv. Ecol. Res. 58, 201–241 (2018).
Carpenter, B. et al. Stan: a probabilistic programming language. J. Stat. Softw. 76, 1–32 (2017).
Bürkner, P. C. brms: an R package for Bayesian multilevel models using Stan. J. Stat. Softw. 80, 1–28 (2017).
Acknowledgements
All authors were supported by the German Centre of Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig (funded by the German Research Foundation; FZT 118). The contribution of T.M.K. was also supported by the Helmholtz Association and by the Alexander von Humboldt Foundation. We thank the many authors who supplied the data that went into the core analyses of this paper, and A. Sagouis and M. Liebergesell for help with data acquisition, collation and harmonization. A. Sagouis helped with the preparation of the simulations in Extended Data Fig. 1, and Fig. 1 was created by F. Arndt (Formenorm.de) for the express use in this paper. Finally, we thank R. Colwell and J. Hortal for important comments and criticisms that helped us to improve the manuscript.
Author information
Authors and Affiliations
Contributions
J.M.C. and T.M.K. conceived the project; J.M.C., K.G. and F.M. developed the initial protocol for data collation and hypothesis tests; J.M.C., F.M. and S.A.B. organized and cleaned the data; F.M. and S.A.B. performed the analyses; J.M.C. wrote the first draft, and all authors contributed to revisions.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Peer review information Nature thanks Robert Colwell, Joaquin Hortal, James O’Dwyer and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data figures and tables
Extended Data Fig. 1 Simulations of the null expectation under the random sampling hypothesis for varying degrees of within-species aggregation.
To evaluate the robustness of the null expectation of a zero slope of biodiversity patterns in standardized samples with fragment size, we took different-sized samples from a simulated landscape and estimated biodiversity patterns. a, Examples of landscapes with different levels of intraspecific aggregation (left to right is from completely random to most-aggregated). From within each of these landscapes, we illustrate four different fragment sizes (shaded squares), and then take standardized (constant-sized) samples from each fragment. b, Density plots showing slope estimates of linear models fit to numbers of individuals (top), species richness (centre) and evenness (bottom) as a function of fragment areas for 2,000 simulated landscapes of each level of aggregation. Both the response and fragment area were log-transformed before model fitting. Densities are shaded by quantiles and the black diamond shows the median for each combination of aggregation and metric; vertical dashed line shows the zero expectation and the median result is given. In some cases, the median result lies very slightly above or below zero (though this does not seem to be associated with levels of aggregation). This is an outcome of the stochastic simulation we performed, and is sensitive to parameters and numbers of iterations, and thus we do not perform statistical tests.
Extended Data Fig. 2 Different measures of species richness related to the size of habitat fragments.
For each case, richness metrics were positively associated with the size of habitat fragments, supporting ecosystem decay as the predominant driver. a, Richness standardized to a common number of individuals. b, Richness standardized to a common sample completeness. c, Asymptotic richness. Solid black lines and shading show overall relationships and 95% credible intervals for each metric. The slope (β) coefficient and its 95% credible interval are shown at the top. Coloured lines show study-level relationships for taxon groups.
Extended Data Fig. 3 Incorporation of uncertainty by calculating z-scores of observed versus null-expected outcomes.
Because there is always uncertainty surrounding the expected outcomes based on passive sampling, we repeated analyses of all metrics recalculated as z-scores. After standardization, z-scores ((observed − expected)/s.d.(expected)) were calculated for the following. a, Standardized species richness. b, Standardized evenness (SPIE). c, Richness standardized to a common number of individuals. d, Richness standardized to a common sample completeness. e, Asymptotic richness. As with the direct measurements, analyses of z-scores show—in total—greater biodiversity loss than expected from passive sampling in smaller habitat fragments, and thus support the ecosystem decay hypothesis. Solid black lines and shading show overall relationships and 95% credible intervals for each metric. Inset shows β-slope coefficient and its 95% credible interval. Coloured lines show study-level relationships for taxon groups. The β-coefficients are not directly comparable to the results from Fig. 2 and Extended Data Fig. 2, owing to differences in the response scale of the z-scores.
Extended Data Fig. 4 Testing of robustness of results to alternative methods.
a, Testing the sensitivity of our results to the exclusion of 47 studies in which data were pooled, and thus sample extent could not be controlled (Methods). Analyses on this subset of data (n = 79) were consistent with those from the full dataset (that is, the 95% credible intervals did not overlap zero). b–d, Testing the sensitivity of the results to decisions we made when imputing missing sizes of habitats labelled as continuous and the treatment of non-integer species abundances (Methods). In each case, the reference scenario imputed the size of continuous habitats to have 10× the area of the next largest fragment and calculated the biodiversity metrics using the non-integer abundance values. Alternate combinations were: (1) imputed area for continuous habitats assumed to be 2× that of next biggest fragment, and non-integer values unchanged; (2) imputed area for continuous habitats assumed to be 100× that of the next biggest fragment, and non-integer values unchanged; (3) imputed area for continuous habitats assumed to be 10× that of the next biggest fragment, and non-integer values rounded up; (4) imputed area for continuous habitats assumed to be 10× that of the next biggest fragment and all abundance values divided by the lowest value within each study, resulting in the lowest abundance equalling one, but retaining the same relative abundances. b, Slope estimates for the relationship between standardized numbers of individuals and fragment size. c, Slope estimates for the relationship between standardized species richness and fragment size. d, Slope estimates for the relationship between standardized evenness and fragment size. Colours depict different auxiliary decisions and imputation required in the analysis (Methods). Small points represent study-level estimates, and large points and error bars are the overall estimates and their 95% credible intervals.
Extended Data Fig. 5 Study-level variation in the number of individuals and evenness.
a–h, Density plots of posterior distributions of study-level slope estimates for total abundance (a–d) and for evenness (SPIE) (e–h). Groupings are by taxon group (a, e), continent (b, f), time since fragmentation (c, g) and matrix filter (d, h). Each density plot is based on 1,000 samples from the posterior distribution of each study-level slope estimate, and is accompanied by the number of studies for each group. Densities are shaded by quantiles and the black diamond shows the median for each group. Solid black line and surrounding shading show the overall slope estimate and its 95% credible interval.
Extended Data Fig. 6 Study-level slope estimates with latitude.
We found that there was a weak negative signal between study-level slope and the absolute value of latitude, which suggests that the influence of ecosystem decay becomes stronger towards the tropics. Each point shows the study-level slope estimate from the standardized species richness as a function of fragment size. Vertical bars are the 95% credible interval associated with each study-level slope estimate. Solid black line and shading shows the relationship and 95% credible interval between the slope estimates and absolute latitude.
Extended Data Fig. 7 Relationship between size of habitat fragment and species composition.
Overall, we found that turnover contributes more than nestedness to pairwise dissimilarity between fragments within a study, but shows contrasting patterns with increasing fragment size differences. a, Turnover component of Jaccard dissimilarity. b, Turnover or balanced abundance component of Ruzicka dissimilarity. c, Nestedness component of Jaccard dissimilarity. d, Nestedness or abundance gradient component of Ruzicka dissimilarity. Solid black lines and shading show overall relationship and 95% credible interval between each dissimilarity component and the ratio of fragment sizes. Coloured lines show study-level relationships for taxon groups.
Extended Data Fig. 8 Endemics–area relationships.
Here, we illustrate the number of species expected to be lost as a function of area of habitat lost under the typically assumed passive sampling hypothesis (purple), and the number of species expected to be lost with habitat lost by inputting our observed parameters from the effect of ecosystem decay (orange).
Supplementary information
Rights and permissions
About this article
Cite this article
Chase, J.M., Blowes, S.A., Knight, T.M. et al. Ecosystem decay exacerbates biodiversity loss with habitat loss. Nature 584, 238–243 (2020). https://doi.org/10.1038/s41586-020-2531-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41586-020-2531-2
This article is cited by
-
Cost-effective biodiversity protection through multiuse-conservation landscapes
Landscape Ecology (2024)
-
Habitat loss reduces abundance and body size of forest-dwelling dung beetles in an Amazonian urban landscape
Urban Ecosystems (2024)
-
Small forest patches in Ethiopian highlands uniquely support high plant biodiversity
Biodiversity and Conservation (2024)
-
Forest edges increase pollinator network robustness to extinction with declining area
Nature Ecology & Evolution (2023)
-
Species diversity of arbuscular mycorrhizal but not ectomycorrhizal plants decreases with habitat loss due to environmental filtering
Plant and Soil (2023)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
