The Sampled Red List Index for Plants, phase II: ground-truthing specimen-based conservation assessments

Abstract

The IUCN Sampled Red List Index (SRLI) is a policy response by biodiversity scientists to the need to estimate trends in extinction risk of the world's diminishing biological diversity. Assessments of plant species for the SRLI project rely predominantly on herbarium specimen data from natural history collections, in the overwhelming absence of accurate population data or detailed distribution maps for the vast majority of plant species. This creates difficulties in re-assessing these species so as to measure genuine changes in conservation status, which must be observed under the same Red List criteria in order to be distinguished from an increase in the knowledge available for that species, and thus re-calculate the SRLI. However, the same specimen data identify precise localities where threatened species have previously been collected and can be used to model species ranges and to target fieldwork in order to test specimen-based range estimates and collect population data for SRLI plant species. Here, we outline a strategy for prioritizing fieldwork efforts in order to apply a wider range of IUCN Red List criteria to assessments of plant species, or any taxa with detailed locality or natural history specimen data, to produce a more robust estimation of the SRLI.

1. Introduction

Failure to meet the 2010 Biodiversity Target of the Convention on Biological Diversity (CBD) to achieve ‘a significant reduction in the current rate of loss of biodiversity’ [1] galvanized the world's decision makers at the Xth Conference of the Parties to the CBD to approve a set of even more ambitious targets to be met in 2020 (the CBD ‘Aichi’ Targets; Decision X/2). In turn, these targets have led to the further development of a series of biodiversity indicators designed to measure progress towards the Aichi Targets, whose activities are coordinated through the Biodiversity Indicators Partnership (www.bipindicators.net). As with many decision-making processes, however, the capacity to implement the decisions lags behind the decisions themselves, and many of the CBD Targets do not yet have comprehensive indicators in place to measure their relative successes, while some are still lacking indicators at all. As the half-way point for meeting the Aichi Targets by the 2020 deadline approaches, the scientific development of these indicators in order to measure progress towards their respective targets urgently needs to increase in pace and in scope.

(a). Targets and biodiversity indicators

One of the key global CBD Biodiversity Indicators is the IUCN Red List Index (hereafter, RLI) [2,3], which is one of four that measure progress towards Aichi Target 12: ‘by 2020, the extinction of known threatened species has been prevented and their conservation status, particularly of those most in decline, has been improved and sustained’. The RLI is produced using a cumulative formula that expresses the overall conservation status of a large taxonomic group as a single value [4]. It is analogous to a stock market index, where the stocks (or companies) are equivalent to species and their share price is equivalent to a numerical value corresponding to their rating on the IUCN Red List (www.iucnredlist.org), the world's most comprehensive system for assessing the conservation status of species [5,6]. The value of the RLI in a given year is scaled between 0 and 1, where a value of 1 would mean that no species of that group was considered threatened and a value of 0 would mean that every species of that group had gone extinct [4]. The more threatened the IUCN category, the greater the numerical weight assigned to that species, while species assessed as being of Least Concern are weighted as zero and thus do not contribute to the value of the index unless they were to move into a threatened or the Near Threatened (NT) category; therefore, the RLI is not quoted as simply a percentage of species threatened but instead captures, to some extent, the degree of threat facing that taxonomic group as a whole. Where comprehensive assessment of the individual conservation status of every species in a taxonomic group is unfeasible, the RLI is instead estimated from a randomly selected sample of species—hence Sampled Red List Index (hereafter, SRLI) [7]. The minimum sample size required for correctly estimating the direction of change of the SRLI on at least 95% of occasions is 900 species per taxonomic group, although in recognition that most large taxa contain a high proportion of poorly known species, a sample size of 1500 species has been recommended to allow for up to 40% of species being assessed as Data Deficient (DD) [8].

(b). Taxonomic scope of the Sampled Red List Index for Plants and work completed to date

The Royal Botanic Gardens, Kew, and the Natural History Museum, London together coordinate the plant side of the SRLI [8–11]. The SRLI for Plants includes randomly selected samples of 1500 species from four major plant taxa, bryophytes, pteridophytes, monocots and the legume family, plus a complete assessment of a fifth major taxon, gymnosperms, which contains more than 900 but fewer than 1500 species (1032 species) [8,9]. At the time of sampling, a complete, authoritative checklist of all dicot species was not available; the legume family was therefore chosen as a surrogate for the dicots, as of all angiosperm families this single family best represents patterns of overall angiosperm diversity [12]. Of the gymnosperms, assessments of conifers are coordinated and undertaken by the IUCN Conifer Specialist Group, and for cycads by the IUCN Cycad Specialist Group. A sample of approximately 7000 species in total, including all gymnosperm species, has thus been selected at random from authoritative global plant checklists; data sources sampled for each group are given in table 1. In undertaking IUCN Red List assessments for sampled species, taxonomic consistency has been improved by contributions from numerous taxonomic and regional experts, by consulting national checklists, flora accounts and taxonomic treatments, and through peer-review of the assessments as they were completed. Full IUCN Red List assessments have to date been completed and submitted to IUCN for almost four thousand (3990) species, from samples of pteridophytes, monocots and legumes from around the world and, supplementing the work of the IUCN Conifer and Cycad Specialist Groups, for all remaining gymnosperm species. Each of these assessments is either now available on the Red List or is in the process of being evaluated by IUCN [11]. It has not yet been possible to gather sufficient data with which to assess the remaining species from the sample; these are likely to be fully assessed as DD, although these DD assessments have not yet been formally submitted to IUCN as they do not contribute to the value of the index.

Table 1.

Sources of taxonomic resources from which samples of species (pteridophytes, monocots and legumes) or complete checklists (gymnosperms) were taken.

major taxon	data source(s)
bryophytes	Index of Mosses (also includes much information on liverworts) (http://www.mobot.org/MOBOT/tropicos/most/iom.shtml)
pteridophytes	Ferns and Fern Allies of the World (http://homepages.caverock.net.nz/~bj/fern/list.htm)
gymnosperms	Farjon A. 2010 A handbook of the world's conifers. Leiden, Netherlands and Boston, USA: Brill
	The Cycad Pages (http://plantnet.rbgsyd.nsw.gov.au/PlantNet/cycad/)
	Ephedra and Gnetum: the World Checklist of Selected Plant Families (http://apps.kew.org/wcsp/) (now incorporated into The Plant List (http://www.theplantlist.org/))
monocots	Monocot Checklist (http://apps.kew.org/wcsp/) (now incorporated into The Plant List (http://www.theplantlist.org/))
legumes	International Legume Database and Information Service (ILDIS; http://www.ildis.org/)

(c). Calculating future values for the index

As with the RLI, the SRLI is re-calculated at regular intervals by re-assessing the conservation status of the same species, and the expert judgement of the assessor decides whether any change in the conservation status of each species is genuine or is merely due to increased knowledge of the species' true conservation status. Only genuine changes in status contribute to the change in the overall value of the index [3], which thus reveals the degree of change in overall conservation status for that taxonomic group, in the same way as a stock market index reflects the overall value of changes in the share prices of many different individual stocks—except that for practical reasons the SRLI will be re-calculated once every 10 years or so (this differs between taxonomic groups depending on the feasibility of re-assessing species' conservation status) rather than every few minutes as stock market indices are. However, for many species, any apparent change in status following a re-assessment of its IUCN Red List status may be attributable to a combination of both genuine changes in status and increases in the assessor's knowledge of the status of that species, either for the Red List parameters previously estimated or for additional parameters for which there was not sufficient information available at the time of the last assessment.

It is thus crucial to the effectiveness of the SRLI as a global indicator of the status of biological diversity that any and all genuine change in each assessment contributing to the index can be adequately detected and that such genuine change can be effectively distinguished from an increase in knowledge that does not reflect the actual change in status. In this paper, we outline a strategy for undertaking re-assessments of the sample of plant species for the SRLI in order to detect effectively this genuine change, using spatial analysis by Geographic Information Systems, remotely sensed vegetation data and species distribution modelling that highlights precise localities for on-the-ground fieldwork to ground-truth the species' re-assessment and provides additional data with which then to apply further IUCN criteria. The approach outlined here is applicable to any taxon for which there is a body of natural history collections data underlying each Red List assessment or for any species with a detailed knowledge of individual locality occurrences across the species range. As such, it is particularly applicable for species that have not yet received a Red List assessment and for which readily available data may seem scarce, but for which multiple natural history collections exist.

2. Conservation assessments: methods

Assessments of the conservation status of species for the IUCN Red List should be based on a detailed evaluation under all of the five IUCN criteria (A–E, table 2); however, the appropriate rating may be established if any one of the criteria is met [13]. The criteria only apply to species assessed as threatened, and in the case of species meriting different ratings under different criteria, the more threatened rating is accepted (the Precautionary Principle) [13]. As detailed and reliable population-level data are generally lacking for plant species, especially those from the tropics, species conservation assessments for the SRLI for Plants have predominantly been based on herbarium specimen data from natural history collections assessed under IUCN criterion B [9]: an evaluation of range size measured as either extent of occurrence (EOO, criterion B1) or area of occupancy (AOO, criterion B2) followed by an assessment of population fragmentation, decline or fluctuation to satisfy (or not) at least two out of three subcriteria [13] (table 2), and the identification of possible threatening processes. Natural history collections usually represent the best available, and often the only available, data for poorly known species [9,12], for which there is often little published beyond their original taxonomic description. The methodology outlined here and subsequent discussion are therefore equally applicable to any taxonomic group, such as insects or indeed most groups of invertebrate animals, for which ready-made sources of range or population data are scarce but for which a sufficient body of natural history collections data could be used as a proxy for range estimates, allowing Red List assessments to be carried out under IUCN criteria.

Table 2.

Summary of the five criteria (A–E) and quantitative thresholds used to evaluate threatened categories (Vulnerable, VU; Endangered, EN; Critically Endangered, CR) for the IUCN Red List (after [13]).

use any of the criteria A–E	Vulnerable	Endangered	Critically Endangered
A. Population reduction: declines measured over the longer of 10 years or three generations
A1	≥50%	≥70%	≥90%
A2, A3 and A4	≥30%	≥50%	≥80%
A1. Population reduction observed, estimated, inferred or suspected in the past where the causes of the reduction are clearly reversible AND understood AND have ceased, based on and specifying any of the following: (a) direct observation, (b) an index of abundance appropriate to the taxon, (c) a decline in AOO/EOO and/or habitat quality, (d) actual or potential levels of exploitation and (e) effects of introduced taxa, hybridization pathogens, pollutants, competitors or parasites.
A2. Population reduction observed, estimated, inferred or suspected in the past where the causes of reduction may not have ceased OR may not be understood OR may not be reversible, based on (a) to (e) under A1.
A3. Population reduction projected or suspected to be met in the future (up to a maximum of 100 years) based on (b) to (e) under A1
A4. An observed, estimated, inferred, projected or suspected population reduction (up to a maximum of 100 years) where the time period must include both the past and the future, and where the causes of reduction may not have ceased OR may not be understood OR may not be reversible, based on (a) to (e) under A1.
B. Geographical range in the form of either B1 (EOO) AND/OR B2 (AOO)
B1. EOO	<20 000 km2	<5000 km2	<100 km2
B2. AOO	<2000 km2	<500 km2	<10 km2
AND AT LEAST 2 of the following:
(a) severely fragmented, or no. of locations	≤10	≤5	=1
(b) continuing decline in any of: (i) EOO, (ii) AOO, (iii) area, extent and/or quality of habitat, (iv) number of locations or subpopulations and (v) number of mature individuals
(c) extreme fluctuations in any of: (i) EOO, (ii) AOO, (iii) number of locations or subpopulations and (iv) number of mature individuals
C. Small and declining population size
number of mature individuals	<10 000	<2500	<250
AND EITHER C1 or C2:
C1. An estimated decline of at least	10% in 10 years or three generations	20% in 5 years or two generations	25% in 3 years or one generation
C2. A continuing decline AND (a) and/or (b):
(a) (i) # mature individuals in each subpopulation	<1000	<250	<50
(a) (ii) or % individuals in at least one subpopulation	100%	95%	90%
(b) extreme fluctuations in the number of mature individuals
D. Very small or restricted population
Either:
D1. Number of mature individuals	≤1000	≤250	≤50
AND/OR
D2. Restricted AOO	AOO <20 km2 or # locations ≤5	n.a.	n.a.
E. Quantitative analysis indicating the probability of extinction in the wild to be:
	≥10% in 100 years	≥20% in 20 years or five generations (max. 100 years)	≥50% in 10 years or three generations (max. 100 years)

Specimens without precise latitude and longitude coordinates on the label are geo-referenced by the SRLI for Plants team, wherever possible, and data from external data sources carefully checked and/or geo-referenced as necessary, in order to estimate range size and produce a preliminary conservation rating under criterion B through automated spatial algorithms [9,14]; desktop applications such as GeoCAT make this process straightforward (http://geocat.kew.org) [15]. Having identified species with range sizes below the IUCN threshold for threatened species (EOO < 20 000 km², AOO < 2000 km²) [13], the desktop assessment process then focuses on qualitative and, where possible, quantitative evaluation of the threats facing that species, if any, to determine if the subcriteria can also be satisfied, and thus the establishment of the appropriate Red List category [9,15,16]. All species are assessed against the subcriteria, but species not meeting the subcriteria will not be assessed as threatened, regardless of range size, whereas those meeting some but not all of the subcriteria may be assessed as NT. Many plant species are naturally rare with ranges under the threshold for a Red List rating of Vulnerable (20 000 km), but are assessed as Least Concern in the absence of any threats to their survival. The IUCN Red List thus distinguishes between threat, a cause of range (or population) decline that may be natural but is generally anthropogenic [11], and rarity, a consequence of natural ecological and evolutionary processes.

(a). Herbarium specimen data

The question thus arises as to how well natural history collections represent species' ranges, and spatial patterns of biodiversity more generally. For individual species, EOO is estimated accurately enough from a convex hull—the smallest polygon encompassing all known points in which no angle exceeds 180°—to assign the correct Red List category from relatively few specimen records, judging by a simulation study of the endemic legume and orchid species of Madagascar [17] that measured EOO and AOO at different sample sizes against ratings under criteria B1 and B2, respectively. These are two very contrasting groups of plants: legumes are classic ‘ecosystem engineers’ providing many of the dominant components of different ecosystems; orchids by contrast are always thought to be classic ‘niche fillers’, sporadically occurring in very restricted localities. With samples of 25 specimens, there was no difference between the preliminary rating produced and the final rating produced using all specimens, with 15 specimens the rating produced was the same as that from the full dataset almost 97% of times, while with 10 specimens this was true in almost 90% of cases and with only five specimens true more than two-thirds of the time (although in reality, with so few records good evidence would be needed that such a species had not been under-collected, and additional criteria applied if at all possible to strengthen the assessment) [17]. The estimate of range size (and thus Red List category preliminarily assigned) is robust to the order in which the specimens are sampled, implying that early collections of a species quickly establish the limits of a species' range and additional collections mostly occur within that species' range. There was no significant correlation between numbers of specimens and rating produced under criterion B beyond four specimens for legumes and eight specimens for orchids, showing that range size estimates are not merely a factor of the number of collections [17].

(b). Using species distribution modelling

Perhaps surprisingly, natural history collections, which represent the collective efforts of multiple independent collectors over extended periods of time, may offer a better representation of the distribution of species than more structured, stratified survey designs. When species distribution models for nine species of tree ferns from New Zealand from two separate datasets, one comprising a relatively small number of fewer than 100 herbarium records and the other more than 5000 tree fern records from New Zealand's National Vegetation Survey, were compared with an independent presence/absence dataset of 1300 forest plots in which absences were also recorded, models based on the National Vegetation Survey data were found to be more biased, falsely predicting tree ferns to be absent 12% of the time but falsely present 30% of the time, while the sparser herbarium data falsely predicted tree ferns to be either absent or present about 20% of the time [18]. The Precautionary Principle places a bigger emphasis on the assumption that a species is no longer there than on the assumption that a species is present when in fact it is not; therefore, from a conservation point of view, inferring distributions from herbarium data may potentially be more reliable than using survey data.

Species distribution modelling is a powerful set of techniques to estimate better range size for species with few records, and thus help to indicate potential additional peripheral populations of a species, leading to an increase in knowledge and potentially a re-evaluation of the original conservation assessment. Most species distribution models assume an equilibrium between a species' range and environmental conditions; however, prevalent non-equilibrium factors such as source–sink dynamics, dispersal distances, propagule viability and interspecific competition with closely related or otherwise ecologically convergent species may be preventing the establishment of new populations of a given species outside its current known range. Application of species distribution modelling to Red List assessments therefore requires careful selection of variables and judicious interpretation of predictions in the light of IUCN criteria [19], but recent demonstrations [20] show the promise of such techniques, particularly for species with 10 specimen records or fewer. EOOs derived from convex hulls based on herbarium specimens are significantly correlated with EOOs from convex hulls fitted around predicted presences from a distribution model for that species when the model threshold for probability of presence is set at the maximum geographical congruence between the two convex hulls [20]. There is no significant difference in range size between the EOO predicted by a species distribution model and the EOO shown by herbarium specimens, although EOOs from species distribution models tend to over-predict ranges of any size when there are few specimens; again, a conservative approach to conservation assessments would favour using the slightly smaller EOO from herbarium specimens [20].

3. The Sampled Red List Index for Plants: results

Results from the SRLI for Plants show that more than one in five (22%) of species of plants worldwide are threatened with extinction under IUCN criteria (table 3 [11]). Gymnosperms are the most threatened of the major plant taxa with 40% of species threatened; excluding gymnosperms and including only those species sampled from larger taxonomic groups, 14% of species are threatened, but one in five species is threatened or NT under IUCN criteria. Of the 816 threatened species from all groups, 484 (59%) were assessed under criterion B (either criterion B1 or criterion B2) and 342 (42%) were assessed under criterion B1; including only species sampled from larger taxonomic groups, almost half (49%) of species were assessed under criterion B1 (table 4; note that each assessment seeks to apply all criteria and any species may be assessed under multiple criteria). The baseline value of the SRLI for Plants is 0.86, equal to that for mammals [21]; plants are less threatened than amphibians (0.74) [22] but are more threatened than birds (0.92) [23]. The number of specimens per species assessed under criterion B1 is shown in figure 1. As the number of collections for a given species tends to increase with the size of that species' range, the great majority (87%) of threatened species assessed under criterion B1, with ranges less than 20 000 km² by definition, are known from only 20 records or fewer (figure 1; cf. [17, fig. 1]). This scaling of numbers of collections with range size also creates difficulties in the effective application of criterion B2 (AOO), which is seldom used as the sole basis for a plant assessment.

Table 3.

Numbers of species in each IUCN Red List category for each group of plants assessed for the SRLI, together with upper, best and lower estimates of percentage threatened species, if all species assessed as DD were threatened, or if all species assessed as DD were in the same proportions as species actually assessed as threatened, or if no species assessed as DD was threatened, respectively (after [11]). CR, Critically Endangered; EN, Endangered; VU, Vulnerable; NT, Near Threatened; LC, Least Concern; DD, Data Deficient; NE, Not Evaluated.

rating	monocotyledons	legumes	gymnosperms	pteridophytes	total
CR	32	14	80	18	144
EN	55	51	163	46	315
VU	71	39	156	91	357
threatened	158	104	399	155	816
lower estimate % threatened	15.46	10.53	39.58	15.95	20.45
best estimate % threatened	17.59	11.43	40.38	16.01	21.68
upper estimate % threatened	27.59	18.42	41.57	16.36	26.12
NT	67	74	167	55	363
LC	677	732	418	758	2585
DD	124	78	20	4	226
subtotal	1026	988	1008	972	3990
NE	478	512	24	528
total	1500	1500	1032	1500

Table 4.

Frequency of criteria used in assessments of plant species for the SRLI. Numbers of species are given for all four groups combined (samples of pteridophytes, monocots and legumes together with all gymnosperm species) and for only samples of pteridophytes, monocots and legumes without gymnosperms. Percentages are rounded to the nearest %; they will not sum to 100% as species may be assessed with multiple criteria.

	four plant groups combineda		sampled groups excluding gymnosperms
criterion	no. threatened species	percentage of threatened species	no. threatened species	percentage of threatened species
A1	2	<1	0	0
A2	165	20	16	4
A3	10	1	4	1
A4	54	7	2	<1
B1	342	42	205	49
B2	254	31	72	17
C1	78	10	3	1
C2	33	4	2	<1
D1	27	3	14	3
D2	135	17	103	25
E	0	0	0	0

^aSamples of species of pteridophytes, monocots and legumes together with all gymnosperms.

Figure 1.

Numbers of specimens per species assessed under criterion B1; note that each species assessed under criterion B1 may also have been assessed under additional criteria. (Online version in colour.)

4. Discussion

Loss of populations of threatened species in isolated localities leading to a genuine up-listing of that species to a more threatened category will be especially influential in driving changes to the value of the SRLI. Confirmation of the persistence, or not, of peripheral subpopulations (potential genuine change), the confirmation of potential additional subpopulations on the periphery of the species' range (increase in knowledge), the careful distinction between such genuine change versus increases in knowledge and robust evaluation of the subcriteria under IUCN criterion B are of fundamental importance to the SRLI and hence our understanding of the global conservation status of plants or of any other taxonomic group.

(a). Re-assessment of automated criterion B assessments

The principal issue with measuring change in conservation status from assessments based on criterion B1 (EOO) is that genuine decline in range cannot be estimated reliably purely from existing specimen data. Through targeting particular species in the SRLI sample, especially those assessed as threatened, there will potentially be an observable increase in EOO and/or AOO from newly accessioned specimens from additional localities (which may be from beyond the current known range, hence increasing EOO and AOO, or may be from within the current known range, hence increasing only AOO), all of which would be considered increase in knowledge, not as genuine change. However, even an increase in knowledge is unlikely given that the stock of herbarium specimens mined for the SRLI for Plants to date represents the collective efforts of thousands of botanists over hundreds of years. In any case, for the SRLI to be representative of the status of plants as a whole, no species in the sample ought to receive attention or conservation actions far in excess of that for un-sampled species, or else the value of the index is liable to be changed positively out of proportion with the change in conservation status of the rest of plant diversity worldwide, by improving the conservation status of threatened SRLI species alone, the overall status of plants as a whole will appear to be better than it really is. This is known as Goodhart's law [24] and could equally be true for any species listed on the Red List. However, due to the reliance of measures of habitat loss in order to satisfy the subcriteria under criterion B, any positive change in the status of SRLI species will also have a positive change on the status of many other species in the vicinity [11].

The future loss of peripheral populations of threatened species, representing a genuine change in conservation status, however, might be inferred from satellite imagery, on the assumption that an obvious loss of known habitat of that species will mean a corresponding extinction of local subpopulations. Decline in range can be estimated from standardized products derived from remotely sensed satellite data (e.g. [25]; Global Forest Watch (www.globalforestwatch.org/)), allowing comparisons between different points in time for individual species, and of declines in range between different species, to help ensure comparability between assessments. Information on the ecology of the species in the sample, gathered as part of the assessment process, can be used to create species-specific habitat suitability maps within the convex hull defining the EOO of the species; loss of habitat associated with peripheral localities leads to an estimate of rate of decline of EOO (figure 2), loss of habitat associated with specimen localities in general (peripheral or non-peripheral) leads to an estimate of rate of decline of AOO and loss of any suitable habitat within the convex hull, whether associated with a known locality for that species or not, leads to an estimate of rate of decline of area, extent and in many cases also the quality of habitat. The approach outlined above thus satisfies one of the three subcriteria for an IUCN Red List assessment under criterion B: continuing decline in the range of a species measured in terms of EOO, AOO, extent and/or quality of habitat, numbers of mature individuals and potentially number of subpopulations (subcriterion b, table 2). Although spatial methods to estimate number of subpopulations have been developed [16], usually the three subcriteria are qualitatively judged by the assessor to have been met or not; a species cannot be assessed as being threatened without also meeting two of the subcriteria, irrespective of its range size. Consistent use of remotely sensed data to detect range decline can thus start to quantify the subcriteria for criterion B to improve the repeatability of automated, specimen-based assessments of any taxon, and would be one way to begin the process of re-assessing the sample of plant species in order to calculate future values of the SRLI for Plants.

Figure 2.

Strategies for prioritizing ground-truthing of criterion B assessments, illustrated with Aloe pubescens from the Ethiopian Highlands; loss of suitable habitat within the convex hull leads to an estimate of rate of decline of area and extent of habitat. (a) Circles are known specimen collections, with the EOO (shaded) fitted as a convex hull around the peripheral points and the AOO calculated as the sum of 2 × 2 km cells in which the species is found (not visible at this scale). (b) Loss of habitat associated with peripheral localities leads to an estimate of rate of decline of EOO (hatched), if the western-most subpopulation were to go locally extinct—this would be a priority for ground-truthing. (c) Further reduction in EOO if the southern-most subpopulation were to go locally extinct, leaving only the northeasterly locations and resulting in a revised rating for A. pubescens of Endangered under criterion B1 as well as criterion B2. (d) Remaining localities of A. pubescens that would be a priority for collecting population counts for assessments under criteria A, C and/or D; AOO cells are now partially visible behind the circles, and loss of habitat associated with specimen localities in general (peripheral or non-peripheral) leads to an estimate of rate of decline of AOO. These images were prepared with the GeoCAT tool developed by the Royal Botanic Gardens, Kew (http://geocat.kew.org [15]). (Online version in colour.)

(b). A strategy for future iterations of the Sampled Red List Index for Plants

It is crucial that the change observed in the value of an SRLI is real and that there is sufficient confidence this is representative of all species. Current assessments can be authenticated and re-assessments produced by establishing or inferring the presence of each species in threatened locations, either by visiting known localities directly or ground-truthing predictions from species distribution modelling, or through analysis of more recent satellite imagery, respectively. In this way, species assessed under criterion B could be re-assessed. Of course, species should also be assessed under other criteria, wherever possible, to give increased confidence in the conservation rating at any particular point in time, and any genuine change in status between time points. Changes in habitat cover and quality from remotely sensed data can be continually measured, and from this it can be inferred which species may no longer be present in particular localities. Used in conjunction with an appropriate species occupancy model, together with knowledge of species' generation length, this then allows the application of criterion A—percentage population reduction—to species previously assessed only under criterion B, assuming that the decline in population size detected is sufficiently large to meet the thresholds for the IUCN Red List categories, through using subcriterion c: continuous decline in EOO, AOO and/or extent and quality of habitat across a species' range. The automated capture of habitat change across the range of each species thus helps to quantify both subcriterion b of criterion B and subcriterion c of criterion A and can be used as the basis for assessments under criterion A, so long as there is sufficient knowledge of how the population occupancy of that species scales with its range, the species' generation length and the distribution of the number of mature individuals between each of the subpopulations of that species [19].

Population reduction under criterion A may be estimated, inferred, projected or suspected as well as being directly observed, however, so without having a true knowledge of the effect of habitat loss on the persistence of different subpopulations of the species sampled for the SRLI for Plants, the resulting index—were it to be based on a large proportion of criterion A assessments—might not be reliable. Targeted fieldwork is therefore an essential component of future iterations of the SRLI for Plants. Criterion B assessments will be authenticated and re-assessed with ground-truthing, and at the same time counts of the number of mature individuals in subpopulations studied in the field will augment these criterion B assessments with assessments under criterion A. Knowledge of the precise location and persistence of different subpopulations and repeat visits to collect the population data that underpins assessments and re-assessments under criteria A and B also then allows the application of criterion C—a small and declining population size—or criterion D—a very small or restricted population, so that by the third time point a series of assessments and re-assessments can be produced using multiple Red List criteria.

This proposed approach highlights two emerging issues. Firstly, there is a practical question of logistics: it is unlikely to prove feasible to gather new field data for every species, let alone every locality, and even if it were possible much of the resulting data would have little or no effect on the SRLI in the short-term. In order to estimate accurately changes in the status of plants globally, it will be crucial to be sure that those particular species and those specific locations in which the observable change is occurring have been accurately identified, and such change then measured effectively in those locations. The second issue is more theoretical: in order to measure observable change, the emphasis should move from assessments under criterion B at time point 1 (when the value of the index was first calculated) to assessments applying criteria A, C or D at time point 2. Assessments under criterion B1 at time point 2 will likely show little change from time point 1 if the same specimen data is re-analysed with the same automated algorithms. Assessments applying criteria A, C or D then need to be backcast to time point 1 so that any genuine change is correctly observed between assessments at different points in time under the same criterion. However, without any population data for assessments at time point 1, these assessments under criteria A, C and D will need to be backcast using expert knowledge, available satellite imagery and the judgement of the assessor to estimate past population sizes.

(c). Identifying genuine change and backcasting assessments

In order to quantify the change in conservation status between one time point and the next, repeat assessments under the same criterion are needed [3], meaning that at any one point in time there will be multiple assessments of the species under different criteria, some of which are fresh assessments under that criterion, others of which are backcast assessments under that criterion to an earlier point in time when that criterion could not previously be applied. The issue of identifying genuine changes in the conservation status of species—the prerequisite for calculating future readings of the SRLI—is thus intimately connected with the issue of backcasting. As assessments under multiple criteria do not yet exist for many plant species, and future assessments for the SRLI should not rely exclusively on criterion B, future assessments under other criteria need to be complemented by assessments at earlier points in time that will need to be backcast under those criteria, in order to ensure that changes to the value of the index reflect only genuine changes in the conservation status of species. At each locality that can be visited, ground-truthed population counts can be collected that would therefore together allow the authentication and re-evaluation of existing assessments under criterion B and new assessments under criterion A, and additional assessments applying criterion C or criterion D. New assessments at the second time point using criterion A will need to be backcast to previous assessments of those species under criterion A at the first time point, using satellite data from the corresponding time period to measure the change in EOO, AOO and/or habitat quality under subcriterion c (table 2). The application of criteria C and D is considerably easier for species with small ranges (when the total population size can be estimated with greater confidence), so is more likely to be used for species assessed as Endangered or Critically Endangered under criterion A and/or B. These new assessments based on criterion A and/or C or D conducted at the second time point need to be backcast to the first time point, and these related to the original assessments carried out under criterion B at the first time point.

With increasing knowledge of a species conservation status at subsequent re-assessments, it may be necessary to backcast a revised assessment of IUCN Red List status to an earlier point in time, reflecting what the earlier assessment would have been had that additional knowledge been available when the species was previously assessed. This issue is not unique to the species comprising the SRLI sample, but is common to every species on the Red List that receives a re-assessment. However, this does mean that earlier values for the SRLI may need to be revised over time in the light of new knowledge, and the key measure for the SRLI is therefore the amount of change between time periods rather than the value of the index itself at any one point in time. Backcasting can thus be undertaken using satellite imagery from the time in question, with a standardized comparison of habitat change from one time period to the next measured as decline in EOO, AOO and/or habitat quality, integrating an understanding of population sizes and occupancy ratios from ground-truthing in the field. Harvesting of data from satellite imagery can be automated across each species range, increasing both efficiency and comparability between assessments of different species.

(d). Optimizing effective fieldwork

In addition to refining guidelines for ground-truthing the assessment of a given species effectively once it has been located in the field, a series of further questions that impact on the planning and resourcing of fieldwork effort needs to be borne in mind. When documented localities for which there are verified specimen vouchers are re-visited, there is always the possibility that the species will not be re-found or correctly identified. However, failure to detect the species does not necessarily mean that it is not present: depending on the season, the plant may be present only in a vegetative state (lacking flowers and fruits and thus much more difficult to observe and to identify correctly) or even as seeds in the soil (in the case of annual species), or other subterranean structures (in the case of geophytic plants and many herbaceous perennials). Even when these possibilities have been eliminated and a species can be confidently reported to be absent, it does not necessarily mean that it has undergone local extinction in that locality within the time period for which the SRLI is being calculated. The loss of this subpopulation could have happened decades previously and to assume that it happened more recently would risk exaggerating the rate of decline in range between the first (desktop) assessment and the second (ground-truthed) assessment. If the species can be found, any observed change in the status of that species between one SRLI time period and the next would then ideally be based on the same criteria, otherwise distinguishing between genuine change and an increase in knowledge becomes intractable. Although the IUCN Red List criteria are intended to be comparable between Red List categories (i.e. a species assessed as Vulnerable under criterion B1 should also be assessed as Vulnerable under criterion A2), using different criteria by definition requires an increase in knowledge or otherwise that the criterion would have been applied initially, and more work is needed to explore whether the thresholds set for each criterion correlate well with results from specimen-based assessments [26] and how that increase in knowledge can be effectively distinguished from any genuine change in status.

A combination of strategies will be needed in planning effective fieldwork. Plans need to be logistically feasible, not covering too great a distance or completely within inaccessible or inhospitable terrain. It will also be important to maximize the return on the time and effort invested through combining potential opportunities to ground-truth several species within a reasonably small area. Finally, fieldwork needs to produce data that will not only test current assessments under criterion B but also allow for the re-assessment of these species under criterion A, C and/or D, which will necessitate more time consuming procedures, such as population counts, in the field. However, in the event that it does not prove feasible to produce new assessments of SRLI species under criterion A, C or D, it will be important at least to produce re-assessments under criterion B1, EOO. In this context, the peripheral localities for each species (which define the convex hull used to measure EOO) become the most critical for a re-assessment for the SRLI for Plants: changes in the status of individual subpopulations at peripheral localities will affect the rating of a conservation assessment under criterion B1 far more than will changes at other localities. Moreover, changes in peripheral localities most distant from other localities will have the greatest effect on Red List status. This becomes particularly important for species whose range size is already close to a category threshold for EOO (e.g. just over 5000 km², so any loss of range would likely move that species from Vulnerable to Endangered).

(e). Targeting priority localities

It is possible to identify priority species and priority target localities for ground-truthing expeditions by measuring and highlighting localities with large inter-point distances in species with EOOs close to a threatened category threshold (as Least Concern species have zero weight in calculating the SRLI). Targeting these distant localities as well as clusters of localities close together (those with small inter-point distances) where numbers of mature individuals may be greater and hence population size can be estimated with greater confidence will provide the data required to both validate and/or re-assess conservation status under criterion B, as well as produce an assessment under criterion A. Species distribution models can also be used to guide future field expeditions for reassessments, and furthermore offer the opportunity to identify potential new localities with suitable environmental conditions for the species, to infer the ecology of poorly known species and to highlight data points that may be possible errors or inaccuracies in taxonomy or geo-referencing. Further investigation will then confirm if high probabilities of occurrence outside the known range reveal new, previously unknown localities for the species, or whether some known localities have particularly low probabilities associated with them because of inaccuracy of the data being used. Species distribution models can thus play an important role in exposing increases in knowledge for a species re-assessment, which must be distinguished from any genuine change in that species' status.

Having a means of targeting high-priority species and localities that are going to have the greatest impact on changes to the value of the SRLI will be a great step forward, as it will not be feasible to visit every locality and efforts and resources will have to be concentrated where they are most needed. Another crucial element in distinguishing genuine change is the expert opinion of taxonomic and/or regional specialists who themselves have good knowledge of the species in question in the field. These may be local in-country counterparts accompanying field trips or partner organizations with which logistical arrangements are made, or any of the hundreds of botanists contacted to provide their expert opinion during the evaluation of each species conservation assessment before it is submitted to IUCN. Through our proposed approach, the critical input and subjective component of those individuals with an intimate knowledge of the species in question is not lost, but is complemented by automated techniques that make use of available specimen and satellite data in an integrated and standardized way between different species to produce comparable species conservation assessments with the same degree of rigour and repeatability. This strategy will be implemented in the ground-truthing of conservation assessments of species in the SRLI for Plants that is now underway in plant diversity hotspots in Europe, in Africa and in other areas of the world with large numbers of threatened species. Being able to target ground-truthing efforts in the field is the first step in being able to gather the population data to apply the full range of IUCN Red List criteria, allowing the transition between criteria at different time points while relying on the same underlying specimen data, and better quantifying the uncertainty at each time point associated with assessments under different criteria, thus increasing confidence in the assessment of each species, in the value of the index at each point in time and the change in value of the index between time points.

5. Conclusion

In order to adequately detect genuine change in the status of species to produce future readings of the SRLI for Plants, it will be essential to bring more IUCN Red List criteria to bear on the assessment of each species in addition to criterion B, so that genuine change can be detected between re-assessments under the same criterion. To do this, as well as to provide re-assessments under criterion B, automated use of remotely sensed satellite imagery provides a first stage that produces data applicable under both criteria A and B. This, together with additional spatial analysis of known specimen information, also highlights priority localities for efficient, targeted ground-truthing expeditions. Population data collected in the field then allow for assessments under criterion A if the species generation length and occupancy ratio is known, as well as providing the basis for assessments under criteria C and D. Backcasting these assessments to earlier points in time, again using appropriate satellite imagery coupled with expert knowledge, allows for genuine change to be identified and measured between time points using the same criteria. The value of the SRLI, and hence our estimate of extinction risks facing the world's biota and the effectiveness of international conservation policy, can thus be estimated with greater confidence.

Funding statement

Work on the Sampled Red List Index for Plants has been funded through the Charles Wolfson Charitable Trust, the Department of Environment, Food and Rural Affairs (Defra), the Esmée Fairbairn Foundation, the Fondation Prince Albert II of Monaco, Rio Tinto plc, the 7th Framework Programme of the European Union, the UK government's World Collections Programme, and with support from the Royal Botanic Gardens, Kew, and the Natural History Museum, London.