Quantifying evidence-based threat assessments for plant species
We find that of the 350,699 accepted plant species (TPL 2013), 25.6% have been subject to some form of evidence-based threat assessments, of which 9.1% have been designated as threatened by either the IUCN-RL or a RRL. The accumulation of plant species assessments in recent years has been impressive, rising from 36,618 in 2003, to 89,810 species-level assessments in 2015 (Fig. 3). However, this accumulation has been largely driven by RRL, as evidenced by the fact that in 2003, 47% were IUCN-RL assessments, whereas in 2015, only 23% are IUCN-RL assessments (Fig. 3). Indeed, by 2015 there are five times more plant species assessed solely by RRL than solely by IUCN-RL, with 69,443 and 14,555 species respectively, equating to 19.8 and 4.2% of total plant species diversity respectively (Fig. 4). In other words, efforts to assess plant species by RRL have extended our knowledge to 69,443 additional species that have no current IUCN-RL assessments. Moreover, our results demonstrate a considerable proportion of non-overlap between both taxa assessed by the IUCN-RL and RRL (Fig. 3). From the 89,810 species-level assessments, only 5812 have been assessed in common by IUCN-RL and RRL (Fig. 4).
Plants are unarguably under-represented in the IUCN-RL 2015-4 data relative to vertebrates, as evidenced by the fact that only 6% of accepted plant species have had global IUCN-RL assessments. The extent of this under-representation is not a new finding, although our analyses suggest that the situation remains severe 5 years on from the GSPC report, which estimated 10% coverage of plant diversity by IUCN-RL criteria and protocols in 2010. Previous explanations for this under-representation have suggested that, whereas metazoan assessments have been a top-down global process led by specialist groups, global plant assessments have been bottom-up and organized primarily by countries that already have well-developed RRL for plant species (Zamin et al. 2010). Consistent with this explanation, our analyses show that the bulk of plant species assessments do indeed stem from RRL, where the greater attention to plant diversity may reflect the priorities of local conservation practitioners, policy makers and the public. Our analyses certainly support the contention that RRL have emerged as the major contributor to the threat assessment process within the plant kingdom. Furthermore, there is valuable complementarity between the RRL and IUCN-RL datasets, as only 6.4% of species have been assessed in common (Fig. 4), supporting in principle, the integration of these data sources to better assess GSPC targets.
Methodological challenges in normalising RRL and IUCN-RL data
Nonetheless as our methodology highlights, in practice, there are notable procedural challenges in normalizing RRL and the IUCN-RL, in order to compare the data sources, and with a view to their eventual integration. Ultimately, as discussed below, in order to normalize the datasets for comparison, 3.2% of all assessments were discarded in our analyses (Fig. 2). Of the initial 98,844 assessments of plant species, there were also problems arising from assessments based on inadequately described taxa, which resulted in 129 discarded assessments. Aside from ‘species novum’ taxa there were also assessments for nomina nuda such as ‘Rughidia milleri’ and ‘Oreofraga morrisiana’, neither of which have ever been published to our knowledge, or that of The Plant List (TPL 2013). Finally, changes in taxonomy since the time of assessment, together with on-going taxonomic uncertainty and taxonomic disagreement clearly affect a substantial number of published assessments. A total of 2131 assessments were discarded because, according to TPL, they were no longer considered to be a species-level taxon, or because they were considered to be synonymous with another assessed taxon. Again with the bulk of this discrepancy coming from the RRL, with 1837 examples of these. In total, issues with systematics caused us to discard 388 (1.8%) of IUCN-RL assessments, and 2834 (3.6%) of the published RRL assessments (Fig. 2). Evidently the degree of discrepancy found within RRL data sources, relative to TPL, appears to be considerably higher than the IUCN-RL, owing in part to the IUCN-RL usage of TPL since its release in 2010.
Gaps in taxonomic and phylogenetic coverage
Bias in taxonomic coverage across conservation assessments is a well-established phenomenon, and plants have previously been identified as under-represented in conservation science and policy. Our analyses reveal similar biases in coverage within assessments of plant species at the level of the four major groupings of land plants, angiosperms, gymnosperms, pteridophytes, and bryophytes. Considering all assessments, both RRL and IUCN, coverage per group is; 80,942 species or 26.6% of angiosperms, 988 species or 89.3% of gymnosperms, 3677 species or 34.4% of pteridophytes, and 4303 species or 12.2% of bryophytes. Although by raw count there are vastly more angiosperm assessments, in terms of percentage coverage, due to the high species diversity of angiosperms the overall percentage coverage of the group is low (26.6%) relative to gymnosperms and pteridophytes. The coverage of gymnosperms is exceptionally good, as there is an active group (Conifer Specialist Group) working on the conservation assessments of these taxa and there are relatively few species to assess. As might be expected, given smaller research communities and more limited pools of taxonomic expertise, the number of assessments for bryophytes is the lowest both as an absolute number and as a proportion.
Perhaps more instructive, is the comparison of the taxonomic coverage between the IUCN-RL and RRL data sets. The IUCN-RL has better coverage of gymnosperms, but much poorer coverage of angiosperms. Relative to RRL, IUCN-RL coverage of bryophyte species is exceptionally poor, where just 102/34,556 species (0.3%) of accepted bryophyte species have an IUCN-RL assessment. With over 4000 bryophyte assessments, the RRL are providing key coverage for the early diverging land plant lineages, in absence of a coherent international program. Although a predictable deficiency, the lack of coverage of bryophyte species fails to recognize that bryophytes form dominant communities across a significant fraction of the Earth’s surface. Bryophytes typically occur in tundra, boreal, and peatland ecosystems (Lindo and Gonzalez 2010), where they provide a habitat to a highly diverse microbiota (Kauserud et al. 2008), and play a central role in the detrital networks and nutrient cycling in these ecosystems (Turetsky 2003).
Higher numbers of assessments in RRL datasets versus IUCN-RL hold across lower taxonomic hierarchies. At the ordinal level, mean percentage coverage relative to all accepted species by RRL is 32.5% per order, whereas the IUCN-RL assessments cover an average of 14.9% of species per order (Fig. 5). Consistent with these mean values, the same data mapped across a consensus topology of the land plants in Fig. 5, show that RRL have more even and proportional sampling across the major orders of land plants. RRL species assessments equal or exceed the number of IUCN-RL species assessments for all but the Cycadales, Pinales, Gnetales, Magnoliales, Dilleniales, Canellales, Berberidopsidales, and Mettenusiales (Fig. 5). That the IUCN-RL assesses far more species in the order Magnoliales is due to active specialist groups focusing on their assessments, such as the publication of the Red List of Magnoliaceae (Rivers et al. 2015), evidence of the success of focused taxonomic intervention of global assessment in key groups.
Overall, the results clearly suggest that in terms of enriching taxonomic coverage, much could be gained from merging RRL and the IUCN-RL into one single pool of plant species threat assessments as suggested by Zamin et al. (2010). This has now been turned into reality in the publication by BGCI of a global database ThreatSearch (www.bgci.org/threat_search.php), which has assembled regional and global assessments of plants in a single portal.
Geographic coverage
As with taxonomic bias, the geographical bias in conservation activity, including biodiversity data collection and publication has been well documented (Reddy and Dávalos 2003; Boakes et al. 2010; Trimble and van Aarde 2012; Oliveira et al. 2016). In general, despite high species diversity, tropical areas are understudied and underrepresented in biodiversity data (Kier et al. 2005; Collen et al. 2008). Consistent with this observation, previous studies have highlighted similarly large gaps in evidence-based conservation assessments across all continents that transect the tropics: Africa, Asia, Australasia and South America (Collen et al. 2008; Zamin et al. 2010). The RRL for Ecuador (León-Yánez et al. 2011), Malaysia (Chua et al. 2010) and the Philippines (Fernando et al. 2008) are notable exceptions to this tropical deficit. Limited availability of financial resources, using Gross Domestic Product as an indicator, is also strongly correlated with limited biodiversity data, again contributing to gaps in conservation assessments across these same continents (Zamin et al. 2010). In addition, specifically with respect to RRL coverage, previous studies have identified substantial gaps in data from West Africa, Central Africa, Australasia, and the Caribbean (Zamin et al. 2010).
Our analyses of plant focused conservation assessments are consistent with these previous studies, as depicted cartographically, conservation assessment coverage does not correlate well with the natural geographic distribution of plant species diversity (Fig. 7). The average percentage of total plant biodiversity assessed across Africa (including the Middle East), Asia (including Oceania and the Pacific Islands), and South America is 13.3, 21.8, and 7.6% respectively, compared to Europe (including Central Asia) and North America (including the Caribbean and Central America) at 27.7 and 16.0% respectively (Fig. 7). However there are notable differences in the extent of geographic coverage between IUCN and RRL assessments of plant species, with IUCN having at least one species in 249/249 countries while the RRL cover 89/249 countries (Fig. 6). There are also clear differences in geographic focus. The RRL have a disproportionate bias towards assessments of plants in the USA and China, but are still notably absent from West Africa, Central Africa, and the Middle East (Fig. 6). The IUCN-RL thus often offers the only conservation assessments across large tracts of continent for which RRL are apparently absent (Fig. 6).
Clearly the globally operating IUCN-RL is providing key biodiversity data in areas where resources are limiting the production of regional red lists. Simply put, with respect to plant species assessments, IUCN-RL often usefully complements the geographic gaps in RRL coverage, and vice versa.
Threat assessment between and among RRL and IUCN-RL datasets
Taxonomic and geographic gaps can undermine the efficacy and utility of evidence-based conservation assessments for biological conservation. The absence of certain species from the assessment process also has the potential to lead species to extinction, and discrepancies between different sources can undermine the credibility of both global and regional red lists. Discrepancies are commonly ascribed to a number of causes, including: regional inequalities in the availability of financial resources and expertise, regional differences in the value attached to different units and types of biodiversity, politicization of bureaucracy associated with regional and global assessments, the application of different protocols, the misapplication of common protocols, taxonomic uncertainty and disagreement, and differences in abundance of individual species at different geographical scales (Collen et al. 2008; Kozlowski 2008; Brito et al. 2010; Zamin et al. 2010).
Taking the RRL and IUCN-RL dataset for plant species as a whole, there are clearly considerable differences in the proportion of species considered as imperiled: 59% of IUCN-RL assessed taxa are considered threatened compared with only 33% in digitally-available RRL assessments. We explicitly recognize that due to the biases operating within these compiled datasets, this estimate cannot be meaningfully compared with, or substituted for, the recent figures obtained by the relatively unbiased Sampled Red List Index for Plants, which estimates that more than 20% of plant species are threatened with extinction. It is likely that the IUCN-RL overestimates the overall extinction of species mainly because over many years most effort has been focused on assessing the species that scientists expect to be threatened, due both to the limited resources available for assessments and to a natural tendency to select species considered to be most in need of assessment.
The 4524 species that have been assessed by both IUCN-RL and RRL with adequate data (i.e. not data deficient), give us a unique window on the comparability of IUCN-RL and RRL assessments of plant species. We observe (Fig. 8a) that 49.1% of IUCN-RL and RRL threat assessments do not agree on the exact category of extinction risk that has been assigned. That discord is just as common as agreement between IUCN-RL and RRL assessments has been noted before in the context of fish (Helfman 2013), but not plants. We find that 15.7% (709) of our sample is composed of species that are not recognized as threatened (Least Concern or Near Threatened) by the IUCN-RL, yet considered threatened or extinct per RRL. In this regard our results differ strongly from that of Brito et al. (2010) who analysed just four RRL but across all taxa, not just plants. Brito et al. (2010) found that only 2% of species assessed in common were not considered threatened by IUCN-RL, but were considered threatened on an RRL. In the same analysis, Brito et al. (2010) estimated that 14% of plants were listed as globally threatened but not nationally threatened, whereas our analysis finds that 8.3% (377) of plant species are considered globally threatened by IUCN-RL, but not considered threatened by RRL assessment. But here we analyze results in a more fine-grained way than a simple binary comparison of threatened and non-threatened, and consequently expose a greater range of discord than revealed by Brito et al. (2010).
The overall level of agreement (0 difference in assessment rank) is particularly driven by agreement on assessment of species that are Least Concern; such assessments make-up 42% (962/2301) of all the agreement assessment pairs. A two-tailed Fisher’s exact test (p < 0.0001) demonstrates then when the IUCN-RL assessment is Least Concern, the RRL assessment has a significant association with this assessment. So, for Least Concern assessments, agreement between IUCN-RL and RRL assessments is significantly better than random. This is unsurprising—it is relatively easier to agree on species which are not at risk of extinction. But even hidden within the assignment of ‘Threatened’ statuses there is disagreement, namely that 19.3% or 873 of our assessment pairs disagree in terms of what exact category of threat is assigned to particular species e.g.VU versus EN, EN versus CR, or VU versus CR (similar to that shown by Rivers et al. 2014). Our data shows a positive skew towards RRL assigning a greater extinction risk to species, than the corresponding IUCN-RL assessment (Fig. 8a).
Generally this positive skew fits null expectation as, due to reasons of geographic scale, the most common explanation is that a species is much more likely to become regionally extinct than to become globally extinct. However there are two more troubling aspects to these analyses of level of agreement. Firstly, when just endemic species are considered, the overall pattern remains largely similar with a similar positive skew (Fig. 8e). More than half (1314/2623, 50.1%) of the endemics-only data present some form of assessment disagreement albeit typically only one rank of threat status up or down (Fig. 8e). Procedural factors and differences in assessment protocol may therefore be a major contributor, perhaps including a national bias towards overvaluing culturally important plants, political interference to elevate or reduce threat status, application of different assessment protocols, misapplication or alteration of IUCN protocols, and taxonomic disagreement. Second, although the overall skew is towards the positive, there are a large number of disagreements, which fall to the negative i.e. the RRL assign a lower risk of extinction to species than the corresponding IUCN-RL assessment. But it is logically impossible for a species to be Near Threatened or Vulnerable nationally and Extinct globally. These ‘negative’ tail disagreements are the more dangerous in terms of species conservation, because most species conservation efforts are delivered locally or regionally, and so an underestimation of extinction risk by RRL may undermine these conservation efforts. However here, our comparisons and analyses provide a useful tool for targeting interventions, as a way of prioritizing species re-evaluation for which one of the assessments, either Regional or IUCN-RL, might be in need of updating an accurate.
Observed differences in assessment have different causes
Although just over 50% of the taxa have directly equivalent assessments between IUCN-RL and RRL, the remaining half of taxa have assessments that are different by one rank or more. The causes of difference in each case are not easy to pinpoint and cannot generally be identified or predicted by statistical modeling with the data we have available—they appear to require explanation and examination on a case-by-case basis. However, we can point to some general reasons, with examples from species assessments that are among those included in this study.
-
1.
Likely to form the bulk of the differences between IUCN-RL and RRL is the simple fact that regional extinction is inherently more likely than global extinction. In the case of the 3 species which have the maximal (7) difference in assessment possible, it is unsurprising to find that the countries lie at the extremities of the species range (Brito et al. 2010). For assessments of Cyperus glaber (1), Ranunculus ophioglossifolium (2), and Ulex europaeus (3) the IUCN-RL has them all as Least Concern, whilst at the regional level they have all been assessed as ‘Extinct’ (1) by a Croatian NL, (2) by an Israeli NL, and (3) by a Norwegian NL. For these species it is clear that they are geographically widespread species and the difference in assessment merely reflects regional extinction at the edge of the native species range, rather than global extinction.
-
2.
Real change in extinction risk occurring in the time elapsed between two assessments. In our pool of data we have some assessment pairs in which one assessment may be published up to 30 years after the other assessment for the same species (incidentally in all three of these maximal instances the assessment stays the same [Least Concern] between the two assessments). In the intervening time between assessments it is conceivable that the extinction risk of a species may genuinely change rank and thus difference in assessment could be purely down to just time elapsed between assessments. The statistical modeling demonstrates that time elapsed is only weakly predictive of difference in assessment. To add a complication, time elapsed also increases the likelihood of the taxon circumscription being revised, so time elapsed is co-correlated and not independent of issues owing to taxonomic change. Examination of assessment pairs that are separated in publication date by 5 years of less (e.g. Fig. 8f), show a similar trend to the overall data (Fig. 8a).
-
3.
Differences owing to changes in our understanding of the circumscription of taxa. In 1998, a global IUCN-RL assessment for Chrysophyllum claraense was published assessing it as Critically Endangered. The Plant List v1.1 treats Chrysophyllum claraense as a synonym of Chrysophyllum oliviforme What was assessed in 1998 under C. claraense by IUCN-RL is no longer considered a discrete separate taxon worthy of its own species-rank name. That taxon is considered a synonym of Chrysophyllum oliviforme and this species is regionally assessed as Least Concern by the Cuba Red List (Berazain Iturralde et al. 2005). As our analyses show, changes to the circumscription of taxa that affect conservation status are relatively rare (Fig. 8a, b). Most synonyms are orthographic or nomenclatural synonyms that do not change the circumscription of the taxon, unlike taxonomic synonyms which do change taxon circumscription (sensu Remsen 2016).
-
4.
Possible local rediscoveries, reflected in local assessments, that have not yet filtered back to global assessment processes. There are a noticeably large number of taxa (830) for which the RRL assessment is much less threatened than the global IUCN-RL assessment. Of particular interest, four species, namely Adiantum lianxianense, Melicope haleakalae, Chrysophyllum januariense, and Ormosia howii have regional assessments that indicate they are extant and are either Near Threatened or Vulnerable. However, the IUCN-RL has assessed all four species as globally extinct—a difference of unarguable significance. Three of those local assessments are noticeably more recent that their IUCN-RL counterparts which suggests they may have been rediscovered in the wild after the IUCN-RL ‘Extinct’ assessments were made. However, for Melicope haleakalae the publication date of NatureServe’s ‘G2’ assessment and IUCN-RL’s ‘Extinct’ assessment is the same: 1998, suggesting a possible taxonomic error. We note that Melicope haleakalae is listed as a ‘Species of Concern’ but not one that is Extinct (Wood et al. 2016), which suggests that the IUCN-RL assessment is perhaps too pessimistic. These sorts of disagreements need urgent re-evaluation to ensure appropriate conservation action is taken.
Concluding perspectives
We have found that RRL assessments are increasing at a rate much faster than global IUCN-RL assessments of plant species. The collective contribution of RRL is invaluable in terms of broadening the scope of threat assessments and providing much needed depth in certain taxonomic areas such as the bryophytes. The Regional and IUCN assessments are largely complementary in terms of taxonomic and geographic coverage, and there would be much to be gained by combining the work of IUCN-RL and RRL to support subsequent GSPC targets on the in situ and ex situ conservation of plants, which has now happened with the recent release of BGCI ThreatSearch (https://www.bgci.org/threat_search.php). However, comparisons of regional-level and global IUCN-RL assessments of the same species reveal interesting discrepancies in the exact level of extinction risk between assessments, something that could be usefully resolved in specific cases with further investigation. As expected, our data show that RRL tend to report higher risk of extinction relative to global assessments, but more concerning are the great many RRL assessments that report lower extinction risk for species than the global assessments Our analyses serve to highlight these instances and to direct specific efforts for re-assessment. In extreme cases this could lead to official ‘rediscovery’ of species that were previously considered extinct by global assessment groups, but found to be extant by regional assessments. These discrepancies that we highlight, together with the taxonomic and geographic gaps, should help guide focused intervention, ensuring both breadth and depth in the threat assessment coverage of all plant species.