Introduction

Current global extinction rates are estimated to be a thousand times higher than background levels (Dirzo and Raven 2003; Barnosky et al. 2011; Pimm et al. 2014; Gray 2019), i.e., the level of extinction expected in the absence of human influence (Pimm et al. 2014). Under the United Nation (UN) Convention of Biological Diversity (CBD), countries pledged to prevent extinctions and improve or sustain the status of threatened species by 2020 (Aichi Biodiversity Target 12). Although this target has not been achieved, global efforts are continuing. Target 4 of the post-2020 Global Biodiversity Framework is to halt the increase in extinction rates or reverse them (Target 4, COP 15, 2022). Recognising the foundational role that plants play in ecosystem functioning, the Global Strategy for Plant Conservation puts specific emphasis on preventing plant extinctions (Goal A of 2050 Goals in Post-2020, 2021). To achieve the Global Strategy for Plant Conservation, empirical assessments aimed at understanding extinctions are required. Moreover, studies into the drivers threatening vulnerable taxa and how species’ traits interact with such drivers are also needed (Pimm et al. 2014; Le Roux et al. 2019; Sage, 2020).

The International Union for Conservation of Nature’s (IUCN) Red List of Threatened Species provides a standardised method to document and assess extinctions. However, only 15% of the world’s plant taxa have been assessed at the global scale (58 343 of an estimated 391 000 taxa) (Christenhusz and Byng 2016; Bachman et al. 2019). Of these assessed taxa, 165 taxa (< 1%) are considered Extinct (includes Extinct or Extinct in the Wild) (IUCN 2022a). In contrast, a recent estimate of plant extinctions by Humphreys et al. (2019), based on the consolidation of global and national Red Lists, estimated that 571 plant taxa are Extinct globally. The large difference between the number of Extinct plants between the IUCN global Red List and the recent work by Humphreys et al. (2019), is largely due to the latter also making use of national Red Lists, which, due to technical difficulties, capacity constraints, and assessment standardization, have not been incorporated into the global Red List. There is ongoing work to improve the taxonomic coverage of the global IUCN Red List, but present data are known to be incomplete and lagging behind for plant assessments (Rejmánek 2018; Bachman et al. 2019; Lughadha et al. 2020). These differences between the global and national assessments can stimulate updates and corrections as seen with the recent responses to Humphreys et al. (2019) and in the current study (Barnosky et al. 2011; Downey and Richardson 2016; Humphreys et al. 2019; Knapp et al. 2020).These updates and corrections mostly come from regions and countries which have more up-to-date information about their plants than what is currently on the global IUCN Red List (e.g., Knapp et al. 2021).

South Africa is a mega-diverse country with an estimated 22 142 plant taxa, including at least 14 149 endemics (SANBI 2021). This diversity is concentrated in three global biodiversity hotspots: the Cape Floristic Region (CFR), the Succulent Karoo, and the Maputaland-Pondoland-Albany hotspot (Myers et al. 2000; Marean et al. 2014). As a signatory to the UN CBD (CBD 1996), threats to plants are monitored and reported on through the South Africa national Red List (hereafter the Red List) (Raimondo et al. 2009; SANBI 2021), which follows the IUCN regional guidelines (IUCN 2012a).With 40 and 48 extinctions reported by Raimondo et al. (2009) and Humphreys et al. (2019) respectively, it is also a global hotspot of plant extinctions, however we still know little about the underlying processes that drove these extinctions (Ellis et al. 2014).

Understanding the traits that make a taxon more vulnerable to extinction can assist with predictions of which taxa are under threat (Gonzalez-Suarez et al. 2013; Gray 2019). Botanical studies have found that taxa from environments with low taxon richness and a poor ability to disperse to climatically suitable conditions in response to climate change are most vulnerable to extinction (Gray 2019). Further, Rejmánek (2018), based on his investigation of the drivers of plant extinction in California, found that taxa with small range sizes are much more vulnerable to extinction. These findings are supported by Gaston (1997) who investigated natural taxon distributions. More recently a global study examining plant extinctions found that the geographical location of the taxa and their life form determine the likelihood of a plant becoming Extinct (Humphreys et al. 2019). Taxa differ in how susceptible they are to various drivers of extinction based on the interaction of these drivers with their ecology (Fréville et al. 2007; Murray et al. 2014; Gray 2019). For example, the impact of habitat loss and range size, interact such that restricted species are more vulnerable than widespread species. Similarly, exploitation and different life-histories together impact extinction risk. For example, long-lived species with low reproductive output, such as cycads (Raimondo and Donaldson 2003), may be much more vulnerable to extinction than short-lived, quick reproducing species with many offspring.

Assessing a taxon as Extinct based on the IUCN Red List requires ‘no reasonable doubt that the last individual has died’ and that adequate searches relative to the taxon’s generation length have been conducted at appropriate times (IUCN 2012b). Establishing this can be extremely challenging as certain life histories allow long persistence times in seedbanks, which reduces detectability. In addition, plant extinctions can take longer than animal extinctions as interactions between life histories and threanening processes can delay extinctions (Cotto et al. 2017; Figueiredo et al. 2019). Plant taxa can remain present in a habitat after being impacted by threats, persisting at extremely low densities, thus becoming functionally, but not actually, Extinct (Gilbert and Levine 2013). Other taxa might only be known from a single collection with vague locality information, limiting the ability to locate populations to determine if they are extant. Other factors hindering extinction assessments include the difficulty detecting inconspicuous taxa and finding cryptic taxa with limited information about their ecology (Downey and Richardson 2016; Gray 2019).

To make assessments of extinction more consistent and transparent, and to address uncertainties, two approaches were developed (Akçakaya et al. 2017; Keith et al. 2017; Thompson et al. 2017). The first approach (search-effort) estimates the probability that a taxon has gone Extinct based on an iterative model, making use of presence records of the taxon and information about the timing, comprehensiveness, and adequacy of targeted surveys to detect the taxon after its last known record (Thompson et al. 2017). The second approach (argument-map) involves evaluating the interaction between the taxon’s life history traits and the severity, duration, and scope of threats (Keith et al. 2017). These approaches allow a more transparent evaluation of the evidence associated with extinction as defined by the IUCN (IUCN 2012b).

Here we profile plant extinctions [Extinct (EX) and Extinct in the Wild (EW) taxa] in South Africa based on the national Red List and review Critically Endangered Possibly Extinct (CR PE) taxa. By comparing threatened and not-threatened taxa, we then explore how factors related to plant traits, taxon distribution and anthropogenic threats correlate with a taxon being assessed as extinct. We aim to obtain a list of predictors of which plant taxa are most vulnerable to extinction. Further, we apply the survey-effort and the argument-map approaches to infer the probability of extinction. Since an extinction assessment has significant conservation consequences, we examine the usefulness of these two approaches in the context of a mega-diverse country, where many taxa need conservation attention.

Methods

Dataset description

In this study we make use of data collated through the South African Red List database (data extracted December 2021). This database includes information of all known taxa in South Africa. For the purposes of this study we first extracted information for Extinct (EX) and Extinct in the Wild (EW) plant taxa (hereafter Extinct taxa unless referred to separately using the abbreviations). Extinction was assessed from 1750, i.e., when the first widespread anthropogenic habitat transformation of South Africa began, roughly a century after Europeans established a colony in the Cape (Skowno et al. 2021). All Extinct taxa are endemic to South Africa, and hence these extinctions are global extinctions. Second, we extracted data for Critically Endangered taxa that were tagged as Possibly Extinct (CR PE). A taxon is tagged as CR PE if, based on available evidence, it is likely Extinct but there is a small chance that it is extant. Sufficient surveys in the correct season, over a time period relative to its ecology and in similar habitat as its known distribution, is required, and only if these fail to detect it, can the taxa be assessed as Extinct (IUCN, 2012b). As data on all ~ 22 000 extant taxa in South Africa are not available, we thirdly used a spatially and taxonomically representative sample—842 taxa that were not Data Deficient that were selected as part of the Sampled Red List Index for South Africa (Raimondo et al. 2019; Online resource 1 A1)—and classified these taxa as threatened or not-threatened. The majority of assessments (90%) have been updated post-2015 and any new occurrence data after these assessments that have been validated were included in the analysis. We specifically reviewed the CR PE taxa and identified those that need reassessment based on information on rediscoveries. The final extracted data of 944 species and infraspecific taxa (such as subspecies and varieties), includes 842 extant taxa, 33 Extinct taxa, and 69 CR PE taxa (Online resource 1 Table A1). Taxonomic changes can influence extinction assessments, as a taxon might not be valid anymore or one taxon might become multiple taxa- such cases will require reassessment. Therefore, the taxonomic status of all Extinct and CR PE taxa was reviewed as part of the current study, while a taxonomic update of all sampled taxa was conducted in 2020.

Selection of parameters and standardization of dataset

We used data readily available from Red List assessments, as well as information captured in floras, and floristic descriptions (See online resource 4). Parameters selected for inclusion are based on previous studies on drivers of extinction (Purvis et al. 2000; Humphreys et al. 2019; Staude et al. 2020; Skowno et al. 2021). Taxon distribution was included since restricted range taxa are naturally more vulnerable to extinction due to the limited area over which pressures need to act to affect the entire range of the taxon (Murray et al. 2014; Humphreys et al. 2019; Knapp et al. 2020). Certain plant traits may increase a taxon’s extinction vulnerability (Purvis et al. 2000; Mace et al. 2008), while anthropogenic threats were also included as they may have direct impacts on the taxon’s survival (Pimm et al. 2014; Di Marco et al. 2015). Different pressures can be important predictors of extinction and are thus included in the model to assess their role in extinctions in South Africa. Data extracted from the Red List were: taxonomic information of each taxon, life form (e.g., annual or perennial), growth form, distribution [biome occupied and range size (extent of occurrence)], date last seen, and human pressures [e.g., habitat loss, habitat degradation, invasive alien species, over-exploitation, pollution, other (including intrinsic factors, changes in native taxa dynamics, human disturbance, natural disasters), and unknown]. Parameter description and sources are summarised in Table 1. For Extinct taxa, the date last seen was based on the last record of the species. This information comes from the Red List assessment of the taxa done, and searches that did not find plants are generally not documented. In some assessments, search history was described, but this was not systematically. Most dates of the last sighting were based on specimen collections.

Table 1 Description of the explanatory variables used in the random forest model to classify taxa as Extinct, Threatened or Not-Threatened. Variables are organised in different classes: human pressure, biological traits, distribution state. The response variable was Extinction risk status grouped into Extinct, Threatened (CR, EN, VU) and Not-Threatened (NT, LC, Critically Rare, Rare). Habitat pressure information was sourced from the South African Red List database and reclassified to match the six major pressures on biodiversity identified by the Convention on Biological Diversity (CBD 2010)
Full size table

Growth form information not included in the Red List was extracted from the Botanical Records and Herbarium Management System (BRAHMS), a database maintained by the South African National Biodiversity Institute (SANBI 2021). Taxa were then broadly reclassified into growth forms based on the IUCN plant growth form classification scheme (IUCN 2022b), as well as being classified as annual or perennial based on information from taxon descriptions and other online sources such as floras and plant guides (Online Resource 4). Data extracted from the Red List included occurrence records, which were used to calculate the extent of occurrence of each taxon (IUCN 2012b). This was done by constructing minimum convex polygons around the occurrence records for each taxon (IUCN 2012b).

A caveat to the data is that no explicit measure of uncertainty has been used, but data was quality checked within each assessment and uncertainty was accounted for as described by the IUCN guidelines (IUCN 2012b). Uncertainty regarding occurrence record accuracy were captured using precision information and only records that were classified as validated were used in the analysis. All other uncertainties are captured within the different methods that have been used to collect the data, see Table 1.

Threat classification scheme

Threats were classified based on the South African threat classification scheme which is linked to the IUCN threat classification scheme (IUCN 2022c). Major drivers were simplified to align with the six principal drivers of threat to biodiversity as classified by the CBD: habitat loss, habitat degradation, invasive alien species, pollution, over-exploitation (CBD 2010) and other (to accommodate threats not captured within these categories). If multiple drivers were identified, we assumed additive impacts and that the taxa were driven to extinction due to the overall impacts, since it was impossible to know which threat took place first or which was the most severe. We then summarised the major drivers of extinction based on information from each Red List assessment. Threat data for the Extinct taxa are based on expert observations or judgement of the possible drivers of extinction. Information from the last known locality and which threats were active just before the taxon was last seen have been used: for example, where there are now towns or where observations or notes from surveys in the past noted “habitat loss to development” it was assumed that there was habitat loss due to settlements (Online Resource 3). If observed or inferred threats were identified for not-extinct taxa (threatened and not-threatened), these threats were documented following IUCN Red List guidelines (see Table 1 for sources of these threats). Threats can be categorised based on the timing of the threat (past, ongoing or future), here we only include past, ongoing and unknown threats, and excluded future and potential threats.

Modelling extinction predictors

Random forest models were used to estimate the probability that a taxon is classified as Extinct (EX and EW; n = 33), threatened (Critically Endangered Possible Extinct, Critically Endangered, Endangered and Vulnerable; n = 213) or not-threatened (Near Threatened, Least Concern, Critically Rare, and Rare; n = 700) (IUCN 2012b). Here we included CR PE in the threatened group, as taxa in this group have not formally met Red List criteria for classification as Extinct. Critically Rare and Rare are South African specific Red List categories that accommodate taxa that are very restricted but are not actively threatened (Online Resource 1 A2).

Random forest is a machine learning algorithm that produces an ensemble of classification trees each constructed on a bootstrap sample (a random draw from all taxa with replacement) (Breiman 2001; Strobl et al. 2009). This modelling technique has shown high predictive power in previous extinction risk related analyses (see Bland et al. 2015; Pelletier et al. 2018). The method tends not to overfit and is relatively robust to correlations between variables (Dimitriadis et al. 2018). Also, it fits complex relationships between the explanatory variables and output probabilities, potentially accounting for high order interactions between the explanatory variables (Basu et al. 2018). This is important in this study as we expect threats to be specific to a taxon’s life history traits, i.e., we expect an interaction between taxon traits and threats will predict extinction probability.

The models included three classes of predictor variables: (i) human pressure i.e., whether a driver was considered to adversely affect that taxon as per the Red List; (ii) biological traits including growth form and life form; and (iii) distribution state including range size based on extent of occurrence of the taxa and presence in a specific biome (Table 1). We used the ‘RandomForest’ (Breiman 2001; Liaw and Weiner, 2002) and ‘party’ (Hothorn, 2006) packages in R (R Core Team 2021) and set model parameters to build the classifiers based on growing 1 000 trees for each iteration (Online Resource 2). The R package ‘caret’ was used to implement the function ‘confusionMatrix’ to calculate model evaluation metrics accuracy (how many taxa the model classified into the correct group of the total number of predictions), sensitivity (proportion of true positives predicted), specificity (proportion of correctly predicted false taxa), and other statistics related to the model performance (Kuhn 2008). Because of the strong class imbalance in our dataset, we stratified the data into different groups and bootstrap sampled within the groups to a maximum of 32 samples (number of species in the scarcest class) when using the entire dataset. Here we aimed to identify the most important predictors of a taxon being classified as extinct.

We constructed four models: (a) the full model that included all 11 predictor variables and threat information for only EX-taxa and threatened taxa; (b) the full model excluding the most important predictor variable, to confirm its importance in the model; (c) the full model and sub-setting data into training and testing subsets (70% and 30% of the data respectively) to validate the model outputs; and (d) 11 predictor variables, with threat data for all taxa. Data on the threats to not-threatened taxa were excluded from the full model as such data were not collected systematically and as the impacts of such threats were considered negligible.

The importance of each variable for each model was determined by measuring the mean decrease in accuracy (MDA) and the Gini Index of the prediction after the removal from the predictive function. These values show how important each variable is in classifying the data (Martinez-Taboada and Redondo 2020). A large decrease in both measures characterises dominating predictors.

Extinction probability assessment

We assessed extinction probability of a taxon, firstly using the survey-effort approach (Thompson et al. 2017). This approach includes records and survey data, and accounts for data quality. All data records of the taxon were collated along with bounds estimating the reliability of the identification of the taxon. The survey-approach estimates whether the taxon’s full range was searched and the probability of detection in searched areas. This approach includes a scoring of effort during surveys conducted since the last record of the taxon, estimates of the proportion of the range that was sampled, the probability of the taxon being detected, and whether the observer could reliably identify the species (Thompson et al. 2017, see Online Resource 5, 6). Further, the years of passive searches, such as opportunistic surveillance that is assumed for every year without a record, is noted. These measures are estimated with bounds and used as input to a model that then calculates the probability of a global extinction based on survey-effort [P(E)s] (see Thompson et al. (2017) for full details of the methodology).

The second approach is based on how threats interact with the taxon life history traits (Keith et al. 2017). For each of the EX-taxa (excluding EW as some of these taxa have been reintroduced into the wild and are awaiting reassessment) we assembled argument-maps including information on the timing, scope, and severity of impacts and the interaction with life history traits (Fig. 1, Online Resource 1 Fig A1, Online Resource 3). We used a range of sources to build the arguments, including taxon descriptions, scientific publications, unpublished expert notes from the Red List, information about the area where the taxa were last seen, and current and historical land-use information based on satellite imagery. The argument-maps were used to estimate the following:

$$P\left( E \right)t{\rm{ }} = {\rm{ }}P{\rm{ }}\left( {local} \right){\rm{ }} \times {\rm{ }}P\left( {spatial} \right)$$
(1)

where P(local) is the probability that the combination of plausible threats occurred for a sufficient time and severity to cause local extinction; P(spatial) is the probability that the threats occurred across the entire range of the taxon; and P(E)t is the probability that a global extinction has taken place. If it is possible to use both approaches the overall P(E) is calculated which includes both P(E)t and P(E)s probabilities as per the Extinction Probability workbook (https://www.iucnredlist.org/resources/ex-probability).

Fig. 1
figure 1

One of the 26 argument maps constructed for extinct (EX) plant taxa in South Africa. Here for the claim that Lampranthus vanzijlae is extinct. Each statement is supported by evidence. All statements with a solid line border come from information in the Red List. Additional information of possible drivers, not included in the Red List assessment have dashed lines. Reasons are divided into three groups: (1) threats have been severe and persistent; (2) species is susceptible to the threats; and (3) threats operate throughout the species’ range. For each of these reasons, an objection is provided. For argument maps for the remaining 25 EX taxa (see Online Resource 3)

Full size image

Results

Summary of extinctions

In South Africa, 33 taxa [EX (26) and EW (7)] (Table 2) have gone Extinct between 1750 and 2021 at a rate of 0.12 taxa per year (Fig. 2). The most recent extinction was declared in 2015 (Conophytum herreanthus subsp herreanthus - EW). This taxon was over-exploited by collectors, it was not found at its last known location, and was searched for extensively (Hammer and Young 2015). More Extinct taxa were lost from the Fynbos biome (21 Extinct out of 5112 extant taxa, 0.41%), than the Grassland (6 Extinct out of 1246 extant taxa, 0.48%) and Indian Ocean Coastal Belt (IOCB) biomes (2 taxa Extinct out of 516 extant taxa, 0.39%), and the rest are scattered across the country as single Extinct taxa within a biome (Fig. 3, Online Resource 1 Fig A2). Most Extinct taxa had only been recorded at a single site (72% of taxa), some only collected once (n = 14). The 33 taxa were spread across 15 families, with the Ericaceae (6), Apocynaceae (4) and Fabaceae (4) having the highest representation. Most Extinct taxa were perennials, with only two annuals (Isolepis bulbifera and Nemesia micrantha). Extinct taxa tended to be small to medium shrubs (< 1–1.5 m). After reviewing taxa that are listed as CR PE, nine taxa have been rediscovered in recent years, these observations needs to be confirmed and the taxa reassessed accordingly. These taxa were excluded from our analysis as their Red List status is now out-of-date. Most of these rediscovered taxa, if confirmed, will be assessed as threatened as they are still only known from one locality and are facing active threats. This has reduced CR PE taxa from 78 to 69 taxa.

Fig. 2
figure 2

Plant extinctions in South Africa 1750-2021 (n = 33) with the trends in the drivers implicated in these extinctions. For the drivers, multiple taxa are stacked if the year “last seen” was the same (See Table 2)

Full size image
Fig. 3
figure 3

Localities of plant extinctions across South Africa 1750–2021 (n = 33) in the context of native plant species richness patterns (at a quarter-degree grid-cell scale). Approximate localities were added for taxa where precise locality information was not available. Background map is based on records from the Botanical Database of Southern Africa (BODATSA), private collections, iNaturalist, and other data sources. Extinct plants are indicated with white circles. A similar pattern is evident if number of plant specimen records and observations per quarter-degree grid-cell is mapped (see Online Resource 1 Fig A2)

Full size image
Table 2 Summary of plant taxa recorded as extinct in South Africa 1750-2021 (n = 33) and the drivers that caused these extinctions. The table is sorted by the date last seen. Each taxon is listed with the following information: their current Red List status as Extinct (EX) or Extinct in the Wild (EW); the drivers classified into major threats; the date last seen (extinction suspected after this date); the first date declared extinct; life form (perennial or annuals); province [Eastern Cape (EC), Gauteng (G), Kwazulu-Natal (KZN), Limpopo (LM), Mpumalanga (MP), Northern Cape (NC), Western Cape (WC)] and which biome(s) the taxon was present in. The final column includes the extinction probability P(E)t as estimated in this study following the argument-map approach (lower and upper bounds of uncertainty). IOCB - Indian Ocean Coastal Belt. Multiple values are separated by pipe-delimiters. For additional details see Online Resource 3 or visit http://redlist.sanbi.org/index.php
Full size table

Drivers of extinction

We found that 26 Extinct taxa were impacted by habitat loss, seven by invasive alien species, seven by habitat degradation, five taxa by over-exploitation, and two by other threats (Fig. 2; Table 2). Extinctions driven by habitat loss were due to urban expansion, agricultural expansion, afforestation, and dam construction. The causes of habitat degradation were inferred from recent field observations and were related to incorrect fire cycles (too frequent fires or fire suppression in fire prone vegetation) and impacts from livestock (e.g., trampling). The impacts of invasive alien species were identified along with other threats such as habitat loss and degradation but were not found to cause extinctions in isolation from other threats. Four taxa have been collected to extinction in the wild (EW), these include charismatic taxa in the genera of Encephalartos (cycads) and Conophytum (dwarf succulents highly prized as ornamentals). Fortunately, these are still present in ex-situ collections (cultivation) in botanical gardens or in private collections. For two taxa, the cause of extinction is unknown — they have not been found after extensive searches in intact habitat (Ceropegia antennifera and Disa forcipata), although there is high uncertainty about where they originally occurred.

Correlates of extinction

The full random forest model achieved a good fit with a global accuracy of 0.86 (Online Resource 2 for model results), indicating that the 11 explanatory variables discriminated between extinct, threatened, and not-threatened taxa. The full random forest model had a sensitivity of 1 to detect Extinct taxa and a specificity of 0.97. With the removal of range size (the most important predictor), the global accuracy decreased to 84%, with a sensitivity of 1 and a specificity of 0.91. We repeated the random forest model and in cross-validation similar results were found, with the classification accuracy of the validation model being 83%, and the sensitivity of 1 and a specificity of 0.96. Including threat information for both threatened and not-threatened taxa did not improve model fit [84% of taxa were correctly classified, with a sensitivity of 1 and a specificity = 0.96]. The full model correctly identified all the assessed EX-taxa but misclassified 26 threatened taxa and 3 not-threatened taxa as Extinct (based on the vote of all trees). These misclassifications were threatened taxa that had small ranges and at least one threat, while the not-threatened taxa had small ranges (1–50 km2) and were taxa that are assessed as Rare or Critically Rare. Of the 11 variables included in the full random forest model, the four most important for predicting extinction, were, in order, range size, habitat loss, habitat degradation, and presence of invasive alien species. To visualize the random forest results, a conditional inference tree (Hothorn, 2006), is used from a sampled tree (Fig. 4, Online Resource 1 Fig A3, Online Resource 2). When the source data (predictor variable values) is changed to include all threat information for both threatened and not-threatened taxa, we found that range size remains the most important predictor, followed by habitat loss and habitat degradation, but there is a change in the importance of the presence in certain biomes, which become more important than invasive alien species (Online Resource 1 Fig A3, Online Resource 2).

Fig. 4
figure 4

Visualization of random forest model classification for Extinct (EX), Threatened (Thr) and Not-Threatened (NotThr) taxa in South Africa using, conditional inference tree. For each node, predictor variable and P-value are given, the bar plots indicate the proportion of taxa within each class for each terminal node. IAS - Invasive alien species and IOCB - Indian Ocean Coastal Belt

Full size image

Structured extinction assessment: survey-effort and the argument-map approaches

Only two taxa had sufficient information to apply the survey-effort approach (Thompson et al. 2017). This increased the estimated probability of extinction, P(E), from 0.941 to 0.97 for Leucadendron grandiflorum and from 0.95 to 0.975 for Isolepis bulbifera, although the qualitative results were the same—these taxa are likely to be extinct. Other taxa had notes about surveys or mentioned extensive searches but no documentation about the timing and the number of sites searched was available.

Of the 26 Extinct taxa evaluated using the argument-map approach, nine (35%) scored P(E)t values of ≥ 0.9 (Fig. 5), the suggested threshold for declaring a taxon Extinct (Keith et al. 2017). These taxa had good threat information available, including how these threats could have interacted with the taxa, and good locality descriptions, which improved our understanding of the locality of the taxa relative to the threats. For example, Liparia graminifolia scored high for both threat interaction as well as whether these threats impacted the taxa across its range (Online Resource 3). Thirteen taxa (50%) had P(E)t scores in the range 0.5–0.9, indicating that the taxa should be classified as CR PE (Fig. 5). These taxa had higher uncertainty in their distribution range information and whether there are other areas in which they could be present. Three taxa had P(E)t scores of < 0.5, indicating that, if further surveys were conducted, the taxa could be found to be threatened or Data Deficient (12%) rather than CR PE. These were two Ceropegia taxa that are considered inconspicuous, and a third, a Lampranthus taxon, which had high uncertainty in its collection site information.

Even though these taxa scored a P(E)t score of below 0.5, these taxa have not been collected since they were first recorded. Disa forcipata was not evaluated due to poor information quality and concerns about the unreliability of results. It is only known from one collection without locality information and its distribution is based on estimates from the collector’s journal and the route travelled in 1870 (Liltved and Johnson 2012). Further investigation did not yield more information. Taxonomists have questioned Disa forcipata as a valid taxon, although it remains an accepted name (Liltved and Johnson 2012, SANBI 2021).

Fig. 5
figure 5

These probabilities were estimated using the argument-maps approach (Keith et al. 2017), based on the severity, extent, and timing of threat interaction with the taxon’s life history and range. Dotted line indicates the threshold between Threatened and Critically Endangered Possibly Extinct (CR PE) (0.5), and the solid line indicates the threshold between CR PE and Extinct (0.9)

Full size image

Discussion

Profile of Extinct plants

Based on the analysis presented here, 33 plant taxa have gone Extinct (26 EX and 7 EW) since 1750, fewer than previous estimates of 40 by Raimondo et al. (2009) and 48 by Humphreys et al. (2019). This difference is due to updates in the Red List through rediscoveries, reassessments, and taxonomic updates (SANBI 2021). Recent changes included here are the rediscovery in 2019 and 2020 of three taxa, previously assessed as Extinct (Raimondo et al. 2009, Online Resource 1 Table A2). Rediscoveries of CR PE taxa also highlights the importance of continued surveying of remnant natural areas (Fig. 6) as well as restoration of sites where native soil seed banks persist (Maynard-Bean and Kaye 2019). Extinct taxa were concentrated in the Cape Floristic Region, a global biodiversity hotspot (Myers et al. 2000; Humphreys et al. 2019). This is not surprising as the region is extremely species rich, has many endemics with narrow ranges (Goldblatt 1997; Rebelo et al. 2008; Rundel et al. 2016), and has experienced and continues to experience extensive habitat loss, especially in the lowlands (Rouget et al. 2003; Rebelo et al. 2008; Dayaram et al., 2022; Skowno et al. 2021). Other Extinct taxa are scattered across the country. Most Extinct taxa were last seen before the 1950s. Spatial information associated with these taxa may be less precise making it more difficult to relocate their records than more recent collections using GPS enabled devices and could also be due to lack of recent searches, however there is no systematic documentation of search effort. Notable, new plant species from South Africa are being described annually (Treurnicht et al. 2017; Mamathaba et al., 2022), and impacts of continued pressures within biodiversity hotspots are likely threatening undescribed taxa. Most newly described taxa are from very restricted habitats and are described from herbarium specimens and most could be found to be threatened with extinction once assessed (Treurnicht et al. 2017; Mamathaba et al., 2022). Species would still be lost in areas experiencing pressures, for example due to predicted continued urban expansion (Taylor et al. 2019; Alberti et al. 2020).

Fig. 6
figure 6

South African taxa thought to be extinct but were re-discovered by citizen scientists that are part of the Custodians of Rare and Endangered Wildflowers (CREW) project. Some of these discoveries were posted on the online platform iNaturalist to encourage further searches. Examples a and e illustrate how the ecology of the taxa might influence its detection, while a, b and c show how locality information accuracy and search effort influence taxa detection

Full size image

At a global scale more trees have been recorded as Extinct than other growth forms (Humphreys et al. 2019), while in the current study for South Africa most Extinct taxa were perennial shrubs, with only three tree taxa (cycads) assessed as Extinct (technically EW). The low incidence of extinction in smaller herbaceous annuals or perennials compared to shrubs could be due to their responsiveness to environmental changes (e.g., survival in the soil until favourable conditions are present) (Humphreys et al. 2019; Compagnoni et al. 2021). Alternatively, this may be a sampling artefact due to their lower detectability and short seasonal duration of visibility. Low detection introduces the risk that extinction of undiscovered taxa could go undetected. Our findings of differing extinction risk between annuals and perennial taxa needs further exploration.

Predictors of extinction

Most Extinct taxa had very restricted distribution ranges and in most cases these taxa were only known from a single site or collection. A single collection does not mean that the taxon’s full range was well understood and limits our ability to interpret these findings. However, small ranges have been shown in previous studies as a key characteristic of taxa that have gone Extinct (Bland and Böhm 2016; Darrah et al. 2017; Chichorro et al. 2019; Staude et al. 2020; Knapp et al. 2021). Small range size was found to be the most important predictor of taxa being classified as Extinct. This supports the criteria used in the IUCN Red List, in which criteria B and D include limited range size (Mace et al. 2008; IUCN 2012a). The dominant driver of extinction was habitat loss in line with studies of global extinctions (Pimm et al. 2014; Le Roux et al. 2019; Staude et al. 2020), though notably this is also the driver that is often easiest to diagnose and document. Habitat loss in South Africa was primarily related to urban and agricultural expansion and forestry activities (SANBI 2021).

Based on the number of Extinct taxa impacted by invasive alien species, this threat was not identified as a major driver of extinction on its own. This could be due to the difficulty in attributing impacts to this driver, as it is seldom clear whether the alien taxon was the driver of the extinction or there was another driver involved (e.g., a disturbance) that may have facilitated the invasion by the alien taxon (MacDougall and Turkington 2005). However, Bellard et al. (2016) found invasive alien plants to have a less important role in global plant extinctions than other threats. Further, habitat degradation, in the form of changes in fire cycle, over-grazing or trampling, were difficult to estimate if experts did not have a good understanding of the habitat, how it might have been disturbed and what its original state was. These patterns could indicate that habitat degradation and invasive alien species impacts are not sufficiently documented to attribute the level of their impact or that the impacts are not severe enough to cause plant extinction. Even though fewer taxa have gone Extinct due to impacts contributed by habitat degradation and invasive alien species, these remain the second and third most important drivers of extinction and threats to threatened taxa (Online Resource 1, Fig A4). Habitat degradation and invasive alien species were also found to be the third and fourth most important predictors of plant extinctions (Online Resource 1, Fig A3). To improve our ability to predict and detect the role of these drivers in extinction vulnerability, a robust understanding of the drivers and the impacted distribution ranges of the native taxa is needed. In the case of alien species impacts, standardised methods of measuring and documenting these impacts over time need to be developed and are required to support impact classification systems (Latombe et al. 2019). These classification systems include, for example, alien species impact documentation collated during Red List assessments as well as through the Environmental Impact Classification for Alien Taxa (EICAT) (Hawkins et al. 2015; IUCN 2020; van der Colff et al. 2020).

Surprisingly, life- and growth-form were found to be unimportant. However, in a global study, growth form was shown to be an important predictor of the type of taxa most likely to go Extinct (Humphreys et al. 2019). Contrary, a recent meta-analysis showed there is a bias in the traits and taxonomic groups considered when investigating important traits in predicting extinction risk, since only three traits had enough data on a global scale to conduct this analysis (Chichorro et al. 2019). It may be that traits not included in our analysis could be more informative or that different suites of traits are linked to different extinctions such that any single trait will have limited predictive power.

Structured extinction assessment approaches

Although two structured approaches for assessing extinction probability were applied, data constraints mean that only the argument-map approach by Keith et al. (2017) was applied to all EX-taxa. Based on the argument-map approach, only seven taxa had a probability score classifying them as Extinct due to the high threshold suggested by Keith et al. (2017) and Akçakaya et al. (2017), and most taxa fell within the CR PE group. This high threshold is to avoid instances where searches and conservation efforts are pre-maturely stopped, leading to an incorrect presumption of extinction (Collar 1998), which has the most severe cost for conservation (Akçakaya et al. 2017). However, obtaining a high score requires good knowledge about the taxon’s biology, its interaction with threats and its range. A limitation to this method is its dependence on knowledge about the taxa being assessed, and in most cases, we know very little about plant taxa. Many taxa were known only from their type locality, with limited biological information available and assumptions were made based on closely related taxa. Many taxa scored lower extinction probability scores with high uncertainty due to a lack of understanding of their distribution (see Online Resource 3).

Due to a lack in documentation of search effort, only two taxa were assessed using the survey-effort approach (Thompson et al. 2017). Many extinction assessments examined here mentioned extensive surveys, but no record of dates, number of searches, nor who conducted them were available. These searches are a prerequisite to assess a taxon as Extinct (Aedo et al. 2015; IUCN 2012a). Experts and volunteers spend time searching for Extinct, CR PE, and threatened taxa, but searches are not always reported when they are unsuccessful. These records, (i.e., null records) can be used to quantify the probability that a taxon is Extinct (using the survey-effort approach). Null records can also improve our understanding of a taxon’s range and, importantly, identify areas that need follow-up searches. Open access tools are needed to provide a centralized place where these null records can be submitted, for use in Red List assessments and to prioritize search areas. Including meta-data along with these records similar to the survey-effort approach, would provide the opportunity to assess whether credible searches were conducted (Thompson et al. 2017).

In a previous study, there was a strong correlation between the results of IUCN Red List assessment for birds and the outcome of using the extinction probability approaches (the search-effort and the argument map) (Butchart et al. 2018). In the current study, we found that 50% of EX-taxa were classified as CR PE, using the Extinction probability approaches due to uncertainty in range size and threat information, showing a weak relationship between the Red List assessments and using the two approaches. This difference between outcomes for birds and plants, may be related to birds being better known and more studied. The current study would benefit from further expert elicitation; however, we do support Butchart et al. (2018), that the two Extinction probability approaches should not be used as a rule, but as guidance, as these approaches allow for critical evaluation of assessment information before a final Red List assessment is obtained.

Conclusion

Creating argument maps to estimate the likelihood of extinction takes time. Such effort can be prohibitive for mega-diverse countries such as South Africa, where a multitude of threatened taxa require updating of Red List assessments and interventions. The argument map and the survey-effort approaches tested can be useful for identifying taxa that are suspected to be Extinct but may not ultimately be so, and hence for which in-field searches are required to understand potential conservation needs. Its application in conjunction with Red Listing is a priority for taxa currently assessed as CR PE, as well as those suspected to be Extinct. Misidentifying taxa as Extinct is detrimental to those extant taxa that require urgent taxon-specific conservation intervention (Akçakaya et al. 2017). The approach we propose supports evidence-driven assessments of taxon-level threats and conservation needs, thereby providing essential building blocks for effective biodiversity conservation.