Research ArticleAPPLIED ECOLOGY

# A third of the tropical African flora is potentially threatened with extinction

See allHide authors and affiliations

Vol. 5, no. 11, eaax9444

## Abstract

Preserving tropical biodiversity is an urgent challenge when faced with the growing needs of countries. Despite their crucial importance for terrestrial ecosystems, most tropical plant species lack extinction risk assessments, limiting our ability to identify conservation priorities. Using a novel approach aligned with IUCN Red List criteria, we conducted a continental-scale preliminary conservation assessment of 22,036 vascular plant species in tropical Africa. Our results underline the high level of extinction risk of the tropical African flora. Thirty-three percent of the species are potentially threatened with extinction, and another third of species are likely rare, potentially becoming threatened in the near future. Four regions are highlighted with a high proportion (>40%) of potentially threatened species: Ethiopia, West Africa, central Tanzania, and southern Democratic Republic of the Congo. Our approach represents a first step toward data-driven conservation assessments applicable at continental scales providing crucial information for sustainable economic development prioritization.

## INTRODUCTION

Major threats to biodiversity, especially in areas of exceptional plant diversity, primarily in the tropics, are often linked to industrial-scale activities such as timber exploitation or large plantations, mining, and agriculture (1). Article 14 of the United Nations Convention on Biological Diversity (www.cbd.int/) explicitly indicates that environmental impact assessments (EIAs) should be conducted before implementing these projects. To reduce risks linked to environmental concerns, EIAs should identify adverse impacts on biodiversity by projects and indicate measures to avoid, minimize, and offset impacts. A growing realization that environmental impacts represent risks not only to biodiversity but also to operational, financial, and reputational aspects of projects has led extractive industries, agro-business, financial institutions, governments, and civil society, inter alia, to identify and adopt best practices for managing biodiversity. Threatened species are one of the key elements [e.g., (2)] that may be affected by these proposed projects.

## MATERIALS AND METHODS

### Flora of tropical Africa dataset

We used a taxonomically verified database of tropical African vascular plant species distribution: RAINBIO (8, 12). This database contains 590,231 georeferenced records representing distribution information for 25,222 native species in sub-Saharan Africa, excluding Madagascar and southern Africa. Invasive or planted or cultivated species were not included [see (12) for details on cleaning and quality-checking the data].

Global assessments require knowledge of all known occurrences for a given taxon. Non-endemics to our study area, i.e., species with occurrences both within and outside tropical Africa, are problematical because RAINBIO does not include occurrences records from outside the continent (12). A species with a few records in tropical Africa might be assessed using the PACA approach as potentially threatened, although it is widely distributed outside Africa and therefore likely not threatened. To reduce this bias, we first excluded occurrences of species only found in South Africa, Swaziland, and Lesotho because only a small portion of the available records for these countries were included in the RAINBIO database (this concerns 2254 species and 3332 occurrences). We then applied the following procedure to identify species whose range extends beyond our study area by comparing the distribution as described by the RAINBIO database to that based on records from GBIF. Specifically, we searched GBIF for occurrences of 22,968 species in the RAINBIO database using the “rgbif” R package (44) and extracted occurrences for each from GBIF (excluding those with georeferencing issues, R code available at https://github.com/gdauby/stevart_el_al_PACA). On the basis of these GBIF occurrence records, the number of occupied cells at 10-km resolution was calculated, as well as the number of continents in which the species has been recorded. Species identified by PACA as potentially threatened under Criterion B based on the RAINBIO dataset (see the method explaining the preliminary assessment below) and occupying more than 10 10 km × 10 km cells based on the GBIF data were tagged. GBIF occurrences were found for 21,345 species (96.9% of those in the RAINBIO dataset). Using RAINBIO, the number of species classified as potentially threatened under Criterion B (i.e., belonging to CR, EN, or VU categories, see below) was 15,470. Among these species, 1220 were recorded from more than 10 10 km × 10 km cells in the GBIF dataset, indicating that subpopulations for these species are missing in the RAINBIO database. When scrutinizing these species, it was found that the additional subpopulations are often artifacts of georeferencing errors or involved doubtful records. Hence, because the rationale of PACA is to provide preliminary conservation assessments, we adopted a conservative approach and only removed species occurring in more than two continents (20 species) or occupying more than 15 10 km × 10 km cells based on the GBIF dataset (912 species) to reduce the risk of incorrectly removing species that are truly threatened. The final dataset used in this paper thus comprised 580,208 distribution records for 22,036 species. Hence, from the initial list of 25,222 species, a total of 3186 were judged to have their known occurrences poorly covered by the RAINBIO database and/or have distributions likely to lie primarily outside our tropical African study area and were thus not assessed here.

### IUCN-based PACA

Depending on the type of information available for a species, an IUCN conservation assessment can be undertaken using any or all of five criteria, A to E. Criterion A is based on estimates of population (number of mature individuals) reduction over 10 years or three generations, Criterion B is based on geographic range, Criterion C is based on population size, Criterion D mainly concerns very small or restricted populations based on the number of mature individuals and AOO, and Criterion E is based on a quantitative analysis of extinction probability within a given number of years. In the absence of detailed information about population size (i.e., the number of mature individuals), which is commonly the case for plant taxa, standard practice calls for using estimates of geographic range obtained from occurrence records (e.g., georeferenced herbarium specimens) for assessments using Criterion B (26). As part of the PACA process, we generated a framework for estimating whether a species faces potential population reduction (required for Criterion A) or future decline in key geographic parameters (needed for Criterion B). Thus, PACA is aligned with two key elements of the IUCN Red List: Criterion A, relating to population size reduction, and Criterion B, relating to geographic range. Approaches aligned with the three other Red List criteria were not implemented as they require data that are unavailable for most plant species, especially across the tropics.

All IUCN parameters needed for preliminary assessments of taxa based under Criterion B (see below) were calculated using the R package ConR ver. 1.2.1 (38). Using an as-yet unreleased version (https://github.com/gdauby/stevart_el_al_PACA), we also implemented assessments aligned with the parameters of Criterion A (see below).

Decline in habitat quality. When using both Criteria A and B, an assessment of observed, estimated, inferred, suspected or projected decline in AOO, Extent of Occurrence (EOO) (see definitions below), and/or habitat quality is needed for each taxon [subcriteria A2 and/or A3 under Criterion A; subcriterion (b) under Criterion B; see below, Table 1]. This is generally assessed on a taxon-by-taxon basis using detailed knowledge about land-use changes and the impacts of threats on the species, which is not possible when simultaneously assessing thousands of taxa. Our method is able to treat a large number of species by estimating population reduction and decline in geographic parameters indirectly based on two sources of information that are likely to be informative. For the first of these, we used land cover characterized by a moderate to high human influence to develop a “human affected” layer. This was done using the land-cover map of Africa produced by (45), which describes 27 land cover types based on remote sensing data at 1-km resolution. We constructed a raster layer for each land cover type by aggregating the original raster at 10-km resolution and computing the proportion of each land cover type within 10-km2 cells. We identified seven land cover types indicative of moderate to high level of human impact: degraded evergreen forests, mosaic forests/croplands, croplands (>50%), croplands with open woody vegetation, irrigated croplands, tree crops, and cities. These layers are directly linked to the main threats on African flora, viz. small- and large-scale agriculture, urbanization, roads, and logging. A given occurrence was considered to be facing a decline in habitat quality if it occurred within a cell where the summed proportion of the seven land cover types indicative of human impact was higher than 50%. The second source of information on human impact was based on the prediction that mining activities will increase significantly in the next few decades across Africa (14). Using a map of major mineral deposits where industrial mining activities are taking place or will likely take place in near future, according to (46, 47), we estimated a decline in habitat quality for any site located within a 10-km radius of such a deposit, based on the inferred scale of the environmental effects of mining activities given in (48).

We identified a decline in habitat quality for a taxon [which is applicable under subcriteria A2 and A3, and invokes subcriterion (b) under Criterion B] when at least one of its known occurrences was found either in an area characterized by moderate to high human impact and/or in a 10-km radius around a major mineral deposit. In the specific case where all known occurrences of a species were found within one or more protected areas, we assumed no decline in habitat quality.

PACA aligned with IUCN Red List Criterion B. Assessing species under Criterion B of the IUCN Red List relies on two subcriteria: B1 and B2, based on the EOO and the AOO, respectively. In order for a species to be assessed as threatened, threshold values under at least two of three subcriteria must also be met as follows: (a) number of locations, (b) inferred/projected decline in various parameters including habitat quality, and (c) extreme fluctuation of populations (Table 1). Subcriterion (c) is rarely applicable for plants and was not considered here.

The EOO is the smallest surface contained in a polygon drawn from an imaginary boundary encompassing all known occurrences of a taxon (namely, the hull convex). At least three points are needed for calculating this parameter, so the EOO was not computed for taxa with less than three unique occurrences.

The AOO is the area within the EOO occupied by the taxon. The AOO is estimated by calculating the sum of occupied cells after superimposing a grid with cells of 2 km2. In ConR, the AOO is estimated by four different positions of the grid cell, and the one resulting in the minimum number of occupied cells is retained (38). A single record per taxon can be used to estimate AOO and thus undertake the assessment, a common practice in tropical plant Red Listing workshops.

The number of locations, as defined by IUCN, is a “geographically or ecologically distinct area in which a single threatening event can affect all individuals of the taxon.” This parameter is difficult to estimate automatically [see (38) for discussion about how it is estimated]. For the purpose of the PACA approach, the number of locations was estimated on the basis of two considerations. First, all the occurrences found within a single protected area were considered to represent a single location, based on the rationale that these occurrences would be equally affected by a single event such as downgrading, downsizing, or degazetting the protected area [e.g., (37)], although we acknowledge that these events can have various impacts that are hard to estimate at broad scales. The number of locations was thus estimated using a shapefile of terrestrial protected areas for tropical Africa downloaded from the World Database on Protected Areas (www.protectedplanet.net). Second, for occurrences located outside protected areas, the number of locations was estimated as the number of occupied cells within a superimposed grid of 10 km2. This grid cell size is suggested as a suitable proxy for detecting a threat that would equally affect all individuals of a taxon contained therein [e.g., mining activities; see (48)].

On the basis of the calculations of EOO and AOO, the estimate of the number of locations and whether or not potential past or future decline was inferred, we automatically assigned each taxon to one of three preliminary threat categories (see IUCN guidelines):

• 1) Potentially CR: EOO < 100 km2 or AOO < 10 km2 and locations = 1 and at least one of its occurrences subjected to decline in habitat quality because it is found in a cell classified as human affected or identified as actually or potentially subjected to mining.

• 2) Potentially EN: EOO < 5000 km2 or AOO < 500 km2 and locations ≤5 and at least one of its occurrences found in a human affected or mining cell.

• 3) Potentially VU: EOO < 20,000 km2 or AOO < 2000 km2 and locations ≤10 and at least one of its occurrences found in a human affected or mining cell.

PACA aligned with IUCN Red List Criterion A. As a complement to inferring PACA using key parameters of IUCN Criterion B, we also inferred parameters for Criterion A based on the observed, estimated, inferred, suspected, or projected reduction in the population size of a taxon, i.e., in number of mature individuals. For most plant species, little, if any, information about population dynamics through time is available. We thus inferred population reduction by making a quantitative estimate of the percentage of population decline using the AOO for each taxon that meets either subcriterion A2 or A3. A2 relies on observed, estimated, inferred, or suspected population reduction in the past, “where the causes of reduction may not have ceased or may not be understood or may not be reversible.” A3 is based on projected, inferred, or suspected population reduction to be met in the near future (up to 100 years or three generations, whichever is smaller).

Using the human impact and mining layers described above, we estimated a population reduction percentage for each taxon by inferring a potential decrease in AOO (AOODEC). This was done by using the ratio between an AOO estimated from all occurrences (AOOFULL) and an AOO estimated only from occurrences situated outside human affected or mining areas (AOORED). The value of AOODEC = ([AOOFULL − AOORED]/AOOFULL) × 100. AOODEC thus represents the decrease in AOO if all occurrences occurring within the human affected or mining areas were lost in the near future. Using AOODEC as computed here and applying the threshold values for IUCN Criterion A, we assigned each taxon to one of the following preliminary threat categories:

• 1) Potentially CR: AOODEC ≥ 80%;

• 2) Potentially EN: AOODEC ≥ 70%;

• 3) Potentially VU: AOODEC ≥ 50%.

Names for PACA categories and subcategories. Assessments obtained using the PACA method must be regarded as “preliminary” because they are not the result of a full IUCN assessment, a taxon-by-taxon procedure based on exhaustive information (e.g., bibliographic, remote sensing data, and in situ observations) on the threats affecting each individual taxon. Species were therefore assigned to three PACA-derived categories based on the automatic batch output from ConR (see Table 1 and Fig. 1): (i) LT, encompassing species flagged as potentially CR or potentially EN, (ii) PT, encompassing species flagged as potentially VU, and (iii) PNT, encompassing species that potentially do not fall into one of the three IUCN threatened categories (i.e., which would correspond to the IUCN Red List categories of NT or LC).

For species assessed as PNT, we further distinguished three subcategories: (i) LR, which includes species whose EOO, AOO, and number of locations all fall within the thresholds for CR and EN but for which we did not infer a decline in the quality of habitat; (ii) PR, which is similar to LR but for species whose EOO, AOO, and number of locations fall within the limits for VU; and (iii) LNT, which includes all other species that do not belong to the categories described above.

### Geographic distribution of threatened species

After species were assigned to a PACA category (Table 1), we compiled distributional data to summarize and map the estimated level of threat across Africa. A gridded spatial representation of threat was undertaken using an “adaptive resolution” SU method (49). This approach adapts the size of the SU as a function of a user-defined threshold of minimum occurrence records. A shapefile of the adaptive SU grid was created by uploading the RAINBIO database to the Infomap Bioregions application (49) using the following parameters: maximum cell capacity = 1000, minimum cell capacity = 250, maximum cell size = 8°, and minimum cell size = 0.5°. We also represented our results for each country and each terrestrial ecoregion (33) found in Africa. In each case (grids, adaptive SU, and ecoregion), we estimated the total number of taxa recorded and the proportion of taxa assessed as LT and PT, under Criteria A and B separately, and by combining both criteria, i.e., a taxon would, for example, be categorized as LT if it is assessed as LT by at least one of the two criteria.

### Threat per habit

We assessed the proportion of threatened species corresponding to each of four major habits: herbs, trees, lianas, and shrubs. Information about species habit was extracted from the RAINBIO database (12).

### Comparison with full published IUCN assessments

For each species, the full published IUCN assessment was downloaded from the IUCN website using the API tools of the “rredlist” R package (version 2018-2, www.iucnredlist.org). To evaluate the extent to which the results of the PACA approach matched those of full Red Listing, we computed a Kappa coefficient for evaluating agreements between full published IUCN categories and PACA categories.

## SUPPLEMENTARY MATERIALS

Fig. S1. Total number of species assessed as Likely/Potentially Threatened following criterion A, B, and both A and B.

Fig. S2. Total number of species preliminarily assessed as LR/PR following the Criterion B.

Fig. S3. Summary statistics for full conservation assessments of plant species published on the IUCN Red List portal (version 2018-2, www.iucnredlist.org).

Fig. S4. Proportion of species preliminarily assessed as Likely/Potentially Threatened following Criterion A and Criterion B for four habit types.

Table S1. Proportion (in %) of LT/PT species assessed under Criteria A, B, and both A and B for all countries within our study area (tropical Africa).

Table S2. Number of LT/PT species assessed under Criteria A, B, and both A and B for all countries within our study area (tropical Africa).

Table S3. Proportions of likely/potentially threatened species assessed under Criteria A, B, and both A and B across ecoregions.

Table S4. Number of likely/potentially threatened species assessed under both Criteria A and B and total number of species for four different habits for each ecoregion.

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

## REFERENCES AND NOTES

Acknowledgments: Funding: This work was cofunded by the French Foundation for Research on Biodiversity and the Provence-Alpes-Côte d’Azur région through the Centre for Synthesis and Analysis of Biodiversity Data Program, as part of the RAINBIO research project through funding to T.L.P.C. (http://rainbio.cesab.org/). G.D. was partly funded by the Belgian Fund for Scientific Research (F.R.S.-FNRS). J.-C.S. and A.B.-O. acknowledge economic support from VILLUM FONDEN through the VILLUM Investigator project “Biodiversity Dynamics in a Changing World” (grant 16549). Author contributions: T.S., G.D., and T.L.P.C. conceived the study; T.S., G.D., P.P.L., V.D., G.E.S., B.A.M., M.S.M.S., J.J.W., A.B.-O., J.-C.S., and T.L.P.C. contributed ideas to the methods; V.D., B.S., B.A.M., D.J.H., and J.J.W. contributed data; G.D. analyzed the data and prepared the figures; T.S., G.D., P.P.L., and T.L.P.C. led the writing; all authors read and approved the final manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. Data used in this study are available at this website: http://rainbio.cesab.org. All analyses can be redone using code available at https://github.com/gdauby/stevart_el_al_PACA.
View Abstract