Introduction

Making informed decisions on the design of multi-functional landscapes for biodiversity conservation requires an accurate knowledge of the relative conservation value of the different land-uses [1]. The ability to accurately determine the relative conservation values of undisturbed primary and human-modified environments is of particular importance. However, quantifying these values can be extremely difficult in complex species-rich ecosystems. This problem is particularly acute in the humid tropics where most species are locally rare, many remain undescribed, and there is a lack of ecological or biological information available for those species that are known to science [2], [3], [4].

Given the critical lack of detailed information on species ranges, preferred habitats and functional roles, comparisons of the biodiversity value of different land-uses invariably depend upon relatively simple metrics. In recent years, a growing number of tropical ecologists have evaluated conservation value using the number (or proportion) of species that are unique to primary or old-growth forests when compared to other land-uses within the wider landscape [5], [6], [7], [8]. Despite its simplicity, this method has many advantages as it facilitates comparisons between studies [6], [9], and is intuitive and easier for non-scientists to understand than many alternative measures of conservation value like similarity indices (see Materials and Methods).

However, it is possible that this metric is highly sensitive to occasional species in samples. There are good reasons why species may be naturally rare [10], but in many cases this rarity may be a sampling artifact. For example, some of the individuals recorded in biodiversity samples could represent transient or sink populations and spill-over effects could mean that occasional species are recorded in modified land-uses even though they are unable to sustain viable populations in the long-term when isolated from source populations in primary habitat. This could lead scientists to underestimate the conservation value of primary forests compared to alternative land-uses [5], [11], [12]. These occasional or rare species are thought to dominate species sample data from systems around the world [10], and their presence in samples could distort indices of uniqueness.

In an ideal world, we would have direct assessments of the viability of all species in biodiversity samples, and could use this to inform our estimates of conservation value of different land-uses. However, we are highly unlikely to ever fulfill this for the vast majority of tropical forest species, and are thereby forced to use species sample-abundances as a proximate estimate of its viability in a given environment (while acknowledging the abundance does not always provide a good proxy of habitat quality; [13]). This presents a dilemma regarding occasional species, as all else being equal we have much greater confidence about the viability of species that are abundant that species that are rare in samples. One common approach used to address this problem is to simply remove occasional species from analyses [14], so inferences are limited to those species that we have most confidence in. This is equivalent to discarding species that were only recorded using a particular sampling technique that is known to be unreliable. However, in this case there is a risk of losing valuable information regarding patterns of conservation value. Furthermore, determining which of these occasional species are genuinely rare (i.e. having either a small geographic range, narrow habitat breadth or a low local density; [15] is difficult and highly context-dependent [16] and requires detailed biological information that remains unavailable for the vast majority of species [17], and especially for those found in tropical forests.

An alternative method is to remove occasional or rare species according to their rank abundance or occupancy [10]. However, rules regarding how many species to remove, or where to remove them from, are very subjective. A search of the literature quickly reveals a bewildering array of criteria for classifying or excluding rare and occasional species from ecological analyses, including removing species with <3 occurrences [18]; species that are not present in at least 5% of the samples [19]; species representing ≤1% of any sample [20]; and species that fail to represent at least 2% (in term of abundance and biomass) of every monthly sample from each sampling station [21].

Given the frequency with which these removal rules have been applied, it is alarming that so little is known about how these essentially arbitrary decisions affect estimates of conservation value. This question has particular applied relevance to the current tropical conservation research agenda as it has recently been argued that agricultural abandonment and subsequent recovery of secondary forests on degraded land could help offset the biodiversity lost from the clearance of old-growth tropical forests [22]. Understanding the extent to which secondary forests and modified forest land-uses such as plantation forestry may provide a “safety net” for tropical forest species depends on having a robust understanding of their conservation value [2], [6], [23].

We use a comprehensive dataset of 15 taxa sampled in primary, secondary and plantation forests in the northeastern Brazilian Amazon to explore how occasional species affect estimates of conservation value. By sequentially removing different abundance classes of occasional species, we test the hypothesis that occasional species can significantly decrease estimates of the conservation value of primary forests (or, conversely, increase the estimated conservation value of modified land-uses). The rationale for this is based on the possibility that spill-over effects could lead to the detection of a disproportionately high number of occasional species in modified land-uses, although the lack of biological data on almost all tropical forest species means we are unable to examine the mechanistic causes behind the observed patterns. We use two different removal rules (removing occasional species from all forest types simultaneously, or removing occasional species from modified land-uses only) and two alternative metrics of uniqueness (the proportion of the entire species pool unique to primary forest, and the proportion of species that were recorded in primary forest that were also unique to primary forest). Although most tropical forest biologists are aware that the removal of the rarest species from analyses will inevitably affect estimates of conservation value based on unique species, this is the first time the sensitivity of these commonly used conservation value metrics has been assessed in any systematic way.

Materials and Methods

Results

Overall, the removal of occasional species consistently increased estimates of the conservation value of primary forest (Figure 1a and 1b) while reducing estimates of the conservation value of both secondary forests and Eucalyptus plantations (Figure 1c–f). Irrespective of the removal rule or uniqueness metric, there was a fundamental difference among different taxa in the proportion of species considered to be unique to primary forest (Figure 2, and see [5]). The consequences of removing occasional species were also highly variable between taxa. Arachnids, fruit-flies and fruit-feeding butterflies were either strongly affected by the removal rule, the uniqueness metric, or both, while other taxa were less sensitive (Figure 2)

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. The conservation value of forests in the Jari landscape.

Panels show how the metric of conservation value changes in primary forest (a & b), secondary forest (c & d) and Eucalyptus plantations (e & f) as an increasing number of abundance classes of occasional species are sequentially removed from the analysis. Values represent the mean proportion (± SE) from 15 vertebrate, invertebrate and plant taxa sampled. Panels on the left (a, c & e) represent values when occasional species are removed from all forest types simultaneously (symbols as circles), while panels on the right (b, d, & f) represent values when occasional species are only removed from modified forest types (symbols as triangles). Solid and open symbols (panels a & b) represent the proportion of the entire species pool, and the proportion of primary forest species (i.e. only species that were recorded at least once in primary forest), respectively.

https://doi.org/10.1371/journal.pone.0009609.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. The conservation value of primary forest (measured as the proportion of species unique to primary forest samples) for 15 vertebrate, invertebrate and plant taxa, following the sequential removal of abundance classes.

Lines within each panel show different selection criteria and removal rules, and the grey shaded area indicates the region between the upper and lower estimates of uniqueness. Circles represent values when occasional species are removed from all forest types simultaneously, and triangles represent the removal of occasional species from modified forest types only. Solid and open symbols represent the proportion of the entire species pool, and the proportion of primary forest species (i.e. only species that were recorded in primary forest), respectively. Order of panels follows Barlow et al. (2007).

https://doi.org/10.1371/journal.pone.0009609.g002

The removal of singletons had a greater effect on conservation value estimates than the sequential removal of doubletons, trebletons, etc (Figures 1 and 2). The removal of additional occasional species (those recorded more than three times) had little or no overall effect on the most unbiased uniqueness metrics which excluded occasional species from both primary and modified land-uses, regardless of whether we considered the proportion of all primary forest species or the proportion of the entire species pool (Figure 1a). Unsurprisingly, when occasional species were only removed from modified land-uses, a larger proportion of species were considered to be unique to primary forests (Figure 1b), but this rise approached an asymptote after the removal of species recorded five times or less.

Sample Representation

Occasional species accounted for a highly variable proportion of the species richness and total abundance of the different taxa (Figure 3). Overall, singletons accounted for 3% of the total abundance and 28% of the species from the forest treatments, while doubletons accounted for 2% and 12%, and trebletons 2% and 6%, respectively. There was substantial variation between taxa and between forest types: singletons alone accounted for between 55–60% of all species in the three forest types in species rich taxa like moths, but just 12–14% of the species of scavenger flies (Figure 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Patterns of rarity in the species data used in this paper.

Panels show the cumulative proportion of species (black symbols) and individuals (open symbols) removed from primary forest (circles), secondary forests (squares) and tree plantations (triangles) (i.e. following the removal of singletons, doubletons, and so on, up to those species that were recorded ≤10 times in each forest type).

https://doi.org/10.1371/journal.pone.0009609.g003

Sample representation was estimated by dividing observed richness in each forest type by the average of three abundance-based richness estimators (Chao 1, Jack 1 and ACE; see Colwell, 2005). Overall, estimated sample representation was generally high across taxa (>70% for 11, 12, and 13 of the 15 taxa sampled in primary, secondary and plantations, respectively; see [24]). The mean sample representation for all forest types was not correlated with the proportion of species in each taxa that were considered unique to primary forest (r = −0.02, n = 15, P = 0.9). Within treatment sample representation can also be assessed visually from species accumulation curves shown in previous manuscripts using the same data (see [5]).

Discussion

The development of effective conservation management plans that encompass entire landscapes requires a detailed understanding of the biodiversity consequences of land-use change and landscape modification [33], [34]. Within the humid tropics, this understanding is severely limited by our poor knowledge of species biology, making it almost impossible to determine whether occasional species recorded in modified land-uses represent viable populations.

As a result, the biodiversity uniqueness of undisturbed primary forests is shrouded by many layers of uncertainty. Estimates are particularly sensitive to decisions researchers make regarding the inclusion or exclusion of occasional species, or how many to exclude (Figures 1 and 2); whether occasional species are excluded from all land-uses, or only modified land-uses (Figure 1, comparing panels a, c and e with b, d and f); whether the metric of conservation value considers the entire species pool for that landscape, or only those species known to exists within the land-use of conservation interest (e.g. primary tropical forests in our case) (Figures 1a & b); and the choice of study taxa (Figure 2). Moreover, estimates of uniqueness are also likely to be affected by the level of taxonomic resolution achieved by researchers, particularly when apparently abundant genera are composed of many rare species (as is likely to be the case for many genera of trees and lianas). If primary forests hold more rare species then anthropogenic land-uses, then a low taxonomic resolution could also lead researcher to underestimate the unique conservation value of primary forests.

Our results show that the lowest proportion of species are considered to be unique to the land-use of interest if no occasional species are removed and the proportion is calculated based on the entire species pool [5]. By contrast, estimates of uniqueness are much higher if an increasing number of occasional species are removed (from either pool), if occasional species are only removed from modified land-uses, and if the proportion is based on the pool of species that are known to inhabit primary forest.

By presenting the uncertainty inherent in almost all biodiversity sample data, we provide a more transparent estimate of the importance of areas of primary forest in multiple-use landscape mosaics. Yet the metrics that make up this upper and lower bound should not be weighted equally. The lower bound of uniqueness will almost certainly underestimate the biodiversity uniqueness of primary forests, because the metric includes many species that are not found in undisturbed land-uses (i.e. open-area and generalist species), and because the removal rule removes many of the rare species that are so characteristic of tropical primary forests.

In our opinion, the upper bound is much more likely to provide an accurate estimate of the conservation value of primary forests. This is because there is much stronger theoretical support for 1) removing occasional species from modified forest sites only (removing species from primary environments is likely to remove viable populations of threatened or rare species) and 2) only valuing species that were recorded at least once in primary forest (it makes little sense to include species that are recorded outside of primary forest but never within it, as these are likely to be species of lowest conservation concern). Although this metric could overestimate conservation value, as some of the occasional species recorded outside primary forest inevitably represent viable populations, we believe it is the most useful metric to use under a precautionary approach, especially when considered against the full range of uncertainty. Finally, it is important to note that we did not remove occasional species from the primary forest only, while leaving them in the modified land-uses. We chose not adopt this approach as we can see no rational biological justification for assuming that occasional species would be able to maintain viable populations in modified land-uses, but would not be able to so in native primary forests.

Acknowledgments

We thank Jari Celulose, Grupo Orsa and the staff of Orsa Florestal for logistical support and access to the Jari forest landscape. We are very grateful to the people who collected data and identified specimens: I.S. Araujo, T.C. Avila-Pires, A.B. Bonaldo, J.E. Costa, M.C. Esposito, L.V. Ferreira, J. Hawes, M.I.M. Hernandez, M.S. Hoogmoed, R.N. Leite, N.F. Lo-Man-Hung, J.R. Malcolm, M.B. Martins, L. .M. Mestre, R. Miranda-Santos, A.L. Nunes-Gutjahr, W.L. Overal, L. Parry, S.L. Peters, M.A. Ribeiro-Junior, M.N.F. da Silva, and C. da Silva Motta. Joe Wright and Jerome Chave provided comments that helped improve the manuscript. This is publication number 22 of the Land-Use Change and Amazonian Biodiversity Project.