Bias and incompleteness in biodiversity inventories
Although parasitic wasps are key components of nearly all terrestrial ecosystems (LaSalle & Gauld, 1993), their macroecological and evolutionary patterns have been scarcely studied outside of a few temperate and tropical areas, mainly because of the inherent difficulty of working with an hyperdiverse group with complex biological interactions and whose taxonomy is far from complete. Further, it is likely that unevenness in the effort devoted to their inventory and systematics (see, e.g. Gaston, 1993; Jones et al., 2009; Baselga et al., 2010) has prevented from developing large-scale analyses of their diversity patterns (but see, e.g. Hawkins, 1994 and references therein).
Many factors can affect the process of inventorying and describing species, and therefore the quality of taxonomic databases. The characteristics of the species affect their probability of being inventoried. For example, the body size, abundance, geographical range and ecological requirements (e.g. trophic and habitat ranges) are all known to influence species discovery (Gaston, 1991, 1993; Gaston & Blackburn, 1994; Patterson, 1994; Blackburn & Gaston, 1995; Gaston et al., 1995; Cabrero-Sañudo & Lobo, 2003; Collen et al., 2004; Adamowicz & Purvis, 2005; Baselga et al., 2007, 2010; Guil & Cabrero-Sañudo, 2007; Jiménez-Valverde & Ortuño, 2007; Jones et al., 2009). In addition, geographical biases in survey effort are the rule rather than the exception (see, e.g. Dennis et al., 1999; Dennis & Thomas, 2000; Hortal et al., 2007, 2008; Lobo et al., 2007; Baselga et al., 2010). In general, northern temperate areas have been more thoroughly studied than the tropics or south temperate regions (Gaston, 1994; Allsop, 1997; Medellín & Soberón, 1999; Cabrero-Sañudo & Lobo, 2003; Collen et al., 2004; Adamowicz & Purvis, 2005; Gibbons et al., 2005; Baselga et al., 2007; Guil & Cabrero-Sañudo, 2007). This bias also seems to be common to the parasitoids (e.g. Gaston, 1993; Jones et al., 2009; Baselga et al., 2010), and is further confirmed by our results (see Fig. 5 and Appendices S1 and S2). Surveys may also be biased at smaller spatial scales. Survey effort is usually higher near recorders’ home ranges, work centres, roads and railway stations or simply in more accessible natural areas (Prendergast et al., 1993; Allsop, 1997; Dennis et al., 1999; Dennis & Thomas, 2000; Kadmon et al., 2004; Diniz-Filho et al., 2005; Jiménez-Valverde & Ortuño, 2007; Sánchez-Fernández et al., 2008; Baselga et al., 2010). These biases seriously compromise the description of species distributions, as well as the representation of their environmental responses (Lobo et al., 2007; Hortal et al., 2008; Jiménez-Valverde et al., 2008; D. Rocchini, J. Hortal, S. Lengyel, J.M. Lobo, A. Jiménez-Valverde, C. Ricotta, G. Bacaro & A. Chiarucci, unpubl.).
Identifying evenly inventoried areas
Several methods have already been developed to identify and account for different types of bias and limitations of biodiversity data. The most developed ones make use of several measures of sampling effort, such as the number of survey records, individuals or traps, in combination with species accumulation curves (e.g. Soberón & Llorente, 1993; Lobo & Martin-Piera, 2002; Hortal et al., 2004, 2008; Hortal & Lobo, 2005), or other relationships with survey effort (Hortal et al., 2001, 2007; Lobo & Martin-Piera, 2002; Garcillán et al., 2003), including species richness estimators (Petersen et al., 2003; Soberón et al., 2007; Lobo, 2008). However, these methodologies usually involve the use of detailed data on the surveys, which is not always accessible, especially in taxonomic databases, such as in the case study presented here. This hampers analyses of survey completeness, thereby limiting the reliability and usefulness of some databases for macroecological studies. In such cases, it is necessary to develop new methods that allow meaningful comparisons of species inventories from different areas without the need for detailed information on the recording process.
Here, we presented a protocol based in three criteria, covering the three main aspects that we believe that, ideally, characterise a reliable inventory: (i) lack of evident biases towards particular taxa, (ii) congruence with well-established ecological relationships, and (iii) origination from works involving enough sampling effort to be potentially complete. The criterion of completeness at higher taxonomic levels accounts for the effort made in describing and inventorying species from different high-level taxa (in this case, subfamilies), taking into consideration that each region has its own colonisation and evolutionary history, and therefore its own taxonomic composition (see, e.g. Ricklefs, 2007). The most important drawback in the use of this criterion relates to how we determine which particular components an inventory must have to be considered reliable. Our sequential approach of first determining how many subfamilies are widespread in the island faunas of the region, and then establishing the minimum number of widespread subfamilies an island should have to be considered as evenly inventoried from the decay in their recorded numbers (see Fig. 1) is a plausible and easy-to-implement approach. However, identifying the point at which this decay pattern changes from being the outcome of biogeographical processes to being a consequence of undersampling can prove difficult sometimes (as evidenced by, e.g. the case of the ichneumonids from Australasia and Palearctic, Figs 2h and 2l). This has the unfortunate effect of adding some undesired subjectivity to this criterion. Also, exceptionally, some islands might truly host less widespread subfamilies due to other causes than being poorly inventoried (e.g. biogeographical factors), and therefore fail to comply with this criterion despite being, in fact, well-inventoried. Nevertheless, we believe these cases are uncommon; the consistency with the islands selected with the SAR criterion provides some support to the adequacy of the choices made.
The rationale for our second criterion comes from the assumption that obvious outliers in well-established ecological relationships are unlikely to have been completely inventoried or nearly so. Perhaps, the most adequate of these relationships is the SAR, due to its generality. Several authors have used the SAR to determine the reliability of the observed species richness from a territory by comparing it with the general relationship found for other areas or well-sampled territories (e.g. Garcillán et al., 2003; Petersen et al., 2003; Roos et al., 2004; Adamowicz & Purvis, 2005; Nikolićet al., 2008). This seems especially appropriate for islands, where land area is known to be one of the most important, although not universal, determinants of species richness (reviewed in, e.g. Whittaker & Fernández-Palacios, 2007). However, this method requires that observed species richness is compared with the extrapolation from the SAR of a well-studied area. Since, in our case study, there was no a priori knowledge of which areas are well sampled, it was not possible to extrapolate the number of species that might be missing from a particular island. Given the large body of knowledge on ISARs, our alternative solution is to use theoretical thresholds (e.g. lower and upper SAR ratio thresholds of 0.2 and 0.65, respectively) to determine when the inventory from any island is poorer, or richer, than should be expected given its area. We use an upper threshold to give cautionary advice about oversampled areas, which might not be comparable to the rest of the less-well (but sufficiently) inventoried ones (see Lobo & Martin-Piera, 2002). Thus, although we recommend discarding data from areas with a SAR ratio of <0.2, we also flag those that have a SAR ratio of >0.65 and would recommend that, rather than omitting them entirely, they should only be discarded if they appear as outliers in other analyses. Islands that are truly species poor could be incorrectly excluded by this criterion. However, since parasitoids typically show ISAR slopes higher than 0.3 (see, e.g. Santos et al., 2010), such incorrect exclusions should be rare or non-existent.
The publication effort criterion is intended to act as a proxy for sampling effort (see, e.g. Hortal et al., 2007; Soberón et al., 2007). However, implementing this kind of criterion using taxonomic databases can be more difficult than expected. In our case, Taxapad only provides information on the total number of pages per publication, but no detailed record on the specific number of pages that refer to a particular island or territory. This could explain why we were unable to identify any patterns of decreasing rate of species accumulation with increasing number of published pages, except for a few islands that have received an important amount of effort in relation to their area. These problems are probably common to many databases and evidence that the implementation of a criterion based on the intensity of inventory effort needs more detailed information on the surveys than that available in most taxonomic databases (see Discussion in Hortal et al., 2007).
Although we have specifically applied our protocol to islands, the generality of its principles may make it easy to adapt to mainland areas, such as countries or biogeographical provinces, and/or to other taxa. Its importance lies in the fact that only requires information on the species inventory and a few general characteristics of the areas, allowing the use of checklists that normally would be considered unsuitable for macroecological studies. Furthermore, by scoring areas instead of simply discarding some of them, this protocol can be useful for identifying different levels of uncertainty, that could be used to weight the areas in regressions or other analyses. Such scoring could also be used to allocate field and taxonomic resources. Of course, our method, like any other, has its limitations. It only allows the identification of which inventories are comparable in terms of taxonomic effort, rather than identifying well-surveyed areas. Therefore, its use within large-scale conservation assessments should be discarded (or used with caution), because these analyses need detailed and accurate results if they are to be used for decision-making. Nevertheless, we believe this protocol might be adequate as a previous step for many analyses of macroecological patterns, as evenly inventoried areas identified this way can be reliably used for large-scale analyses.