Knowledge From the Afrikan World, For the World

PMC Articles

The bii4africa dataset of faunal and floral population intactness estimates across Africa’s major land uses

PMCID: PMC10861571

PMID: 38346970

Abstract

Sub-Saharan Africa is under-represented in global biodiversity datasets, particularly regarding the impact of land use on species’ population abundances. Drawing on recent advances in expert elicitation to ensure data consistency, 200 experts were convened using a modified-Delphi process to estimate ‘intactness scores’: the remaining proportion of an ‘intact’ reference population of a species group in a particular land use, on a scale from 0 (no remaining individuals) to 1 (same abundance as the reference) and, in rare cases, to 2 (populations that thrive in human-modified landscapes). The resulting bii4africa dataset contains intactness scores representing terrestrial vertebrates (tetrapods: ±5,400 amphibians, reptiles, birds, mammals) and vascular plants (±45,000 forbs, graminoids, trees, shrubs) in sub-Saharan Africa across the region’s major land uses (urban, cropland, rangeland, plantation, protected, etc.) and intensities (e.g., large-scale vs smallholder cropland). This dataset was co-produced as part of the Biodiversity Intactness Index for Africa Project. Additional uses include assessing ecosystem condition; rectifying geographic/taxonomic biases in global biodiversity indicators and maps; and informing the Red List of Ecosystems.

Full Text

We address this data gap through a structured expert elicitation process (Fig. 1) involving 200 experts in Afrotropical biodiversity on mainland sub-Saharan Africa. Expert elicitation is used widely in conservation and natural resource management when data are insufficient or absent. Notable examples of where expert elicitation has been used include the International Union for Conservation of Nature (IUCN) Red List of Threatened Species, and assessments of the Intergovernmental Panels on Climate Change and on Biodiversity and Ecosystem Services. We employed the latest advances in expert elicitation to ensure data rigour and consistency. Expanding substantially on an earlier approach that involved 16 experts in Southern Africa, experts estimated ‘intactness scores’: the remaining proportion of an ‘intact’ reference population of a given species group (Table 1) in a particular land use (Table 2), on a scale from 0 (no remaining individuals) to 1 (same abundance as a reference population) and, in rare cases, 2 (populations that thrive in human-modified landscapes). The ‘intact’ reference is the population abundance that would likely have occurred in the area before alteration by modern industrial society. Because information on species populations from this era is virtually non-existent, standard protocol is to consider a remote wilderness area with a natural disturbance regime (a Hybrid-Historical approach) where necessary. The resulting bii4africa dataset contains standardised intactness scores representing terrestrial vertebrates (tetrapods: ±5,400 amphibians, reptiles, birds and mammals) and vascular plants (±45,000 forbs, graminoids, trees and shrubs) in sub-Saharan Africa across nine major land uses of varying intensity.
Overview of the structured expert elicitation process, based on the IDEA protocol (Investigate, Discuss, Estimate, Aggregate). This process was run for each broad taxonomic group (Table 1), between November 2020 and January 2022, to elicit from 200 experts the estimated impact of nine major African land uses (Table 2) on the relative population abundances of terrestrial vertebrates and vascular plants.
An expert elicitation process was run for each broad taxonomic group, with participants estimating intactness scores for each response group. Supplementary Table 1 includes details of each response group.
Supplementary Table 2 includes representative images of each land use.
The dataset was developed to enable the quantification of the Biodiversity Intactness Index for sub-Saharan Africa (https://bii4africa.org/). It provides an African-led alternative to a previous attempt to map the Biodiversity Intactness Index globally based on a model, which produced inaccurate results for Africa (likely in part because of the under-representation of African data in the model). The bii4africa dataset that we present here has a broad range of additional uses, including assessing and mapping ecosystem condition to inform national planning and reporting on Goal A in the post-2020 Global Biodiversity Framework; alleviating geographic and taxonomic biases in global biodiversity indicators and maps; parameterising models of biodiversity in a changing world; informing the IUCN Red List of Ecosystems; supporting the United Nations Decade on Ecosystem Restoration in identifying priority ecosystems and monitoring the impact of investments into restoration; identifying the properties of novel ecosystems; and informing future research and training in African biodiversity (Table 3). Importantly, the dataset—co-produced by 200 experts—embodies context-specific knowledge on African biodiversity that contributes to inclusivity in ecology. It is also a positive response to the recent call by the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) for African-led research that closes knowledge gaps by mobilising local data.
Supplementary Table 3 includes further details on these uses.
We implemented a published, modified-Delphi protocol (multiple rounds of individual, independent expert estimation interspersed with group discussion and review; Fig. 1), which has been shown to improve the rigour of elicitation outcomes. The elicitation was limited to terrestrial vertebrates (tetrapods) and vascular plants – groups that comprise species with a high diversity of attributes and functions. Experts were sought with knowledge of the degree to which human land uses impact populations of these species groups in sub-Saharan Africa (or a region therein). Such expert knowledge is typically limited to one taxonomic class (e.g., birds or reptiles), or in the case of mammals, often just one or several orders (e.g., primates or bats). As such, multiple expert elicitation processes were run between November 2020 and January 2022, to cover the various broad taxonomic groups of species (see Table 1). An expert in each broad taxonomic group was invited to lead the elicitation for that taxonomic group. Each ‘lead expert’ was identified based on their relevant expertise and existing network across the continent or willingness to develop such a network. For example, several lead experts serve in the IUCN Species Survival Commission working groups and other regional networks (e.g., https://ascaris.org/; https://www.birdlife.org/our-partners-africa/). The lead expert was responsible for identifying experts and inviting them to participate, as well as proposing a draft list of species response groups and providing input into the development of the land use categories. These tasks are detailed in the subsequent sections.
Of the 248 experts who agreed to participate, a total of 200 (81%) participated in the full elicitation process (https://bii4africa.org/category/experts/), with an average of 24 experts per broad taxonomic group (Fig. 2). These experts were from 39 countries, including 23 African countries (59% of experts were African nationals). Most experts (72%) were resident in Africa at the time of elicitation, across 26 countries. Their experience spanned sub-Saharan Africa, with Southern Africa being best represented, followed by East Africa, and West Africa having the lowest representation. Experts worked in a range of sectors, most commonly at universities, followed by research institutes and conservation organisations. Collectively, experts had over 3,300 years of relevant accumulated experience, with a mean (±SD) of 18 ± 10 years per expert.
Within each broad taxonomic group (Table 1), the lead expert proposed a draft set of species response groups, based on their knowledge of the key organismal attributes likely to determine the impact of different land uses on populations. These draft species response groups were presented to the participating experts for that taxonomic group during an introductory planning meeting (see Structured expert elicitation section below) and revised based on experts’ feedback. Key organismal attributes that informed species response group categorisation varied between the taxonomic groups (Table 1; Supplementary Table 1). For terrestrial vertebrates, attributes commonly included habitat requirements (e.g., forest, grassland, generalist), body size, diet and stratum (e.g., arboreal, fossorial, rupicolous). Terrestrial vascular plants were divided into three broad groups: trees and shrubs, graminoids, and forbs, and species response groups were described within each broad group based on a framework of life-history strategies shaped by disturbance. Species response groups were thus ecologically defined, not necessarily taxonomically defined. Large mammals were an exception to the response-group approach. The elicitation was done at species level for the seven large carnivores and 95 large herbivores. While the large herbivores were initially categorised into response groups, experts decided during the introductory meeting to estimate per species rather than per species response group. Thus, for large herbivores, we present species response groups, as well as species-level estimates. Details of each response group are included in the ‘Sp_Groups’ spreadsheet in the bii4africa dataset.
Intactness scores were estimated for nine distinct land uses (Table 2). These land uses with varying intensities of human modification were selected to capture the major land covers, uses and associated activities relevant to sub-Saharan Africa, while being broadly comparable with other land-use maps (e.g., Ellis et al.; Goldewijk et al.; Hurtt et al.). Experts made estimates at a ‘landscape’ scale (i.e., several square kilometres). A finer ‘patch’ scale (i.e., several square metres) would be inappropriate for an expert-elicited approach because experts need to consider multiple activities in an area that impact species. For example, a patch of vegetation in a city is likely to have a different impact on a species than a patch of vegetation in a smallholder cropland or a protected area (particularly for large-bodied or wide-ranging species). Instead, the landscape of land covers, uses and activities characteristic of each land use category were described to experts (Table 2), who were asked to consider the collective impact on the abundance of a given group of species within that land use.
Experts were provided with photographic examples of these land uses to help visualise them and promote consistency in scoring between experts (Supplementary Table 2). For each land use, experts were asked to visualise a specific landscape with which they are familiar that matched the description provided, or several such landscapes across which they could consider an average. Experts were instructed to consider the integrated impact of all characteristics of that landscape (e.g., habitat loss and fragmentation, land cover change, disturbance by people, their domestic animals, infrastructure, pollution, harvesting, persecution, introduced species and diseases, elimination of mutualistic species, altered fire and herbivory regimes) on each species response group. Integrated information at a landscape scale is also likely to be more useful for certain decision-maker needs than fine-scale, species-level data.
The IDEA (‘Investigate’, ‘Discuss’, ‘Estimate’, and ‘Aggregate’) structured expert elicitation protocol was used (Fig. 1). It is a modified-Delphi procedure that treats each step as a process of formal data acquisition, incorporating research from mathematics, psychology, and decision theory to help reduce the influence of biases and enhance the transparency, accuracy, and repeatability of the resulting estimates. The protocol was implemented as follows for each of the broad taxonomic groups:
Investigate: A one-hour online meeting was held to introduce the project and explain what was expected of participating experts. Experts had an opportunity to ask questions. The lead expert also presented the draft species response groups and received feedback from the participating experts that, in many cases, led to revisions of these groupings. Afterwards, experts were emailed a recording of the meeting, written instructions, and a survey spreadsheet in which to provide estimates. Experts could provide estimated intactness scores for all species response groups (ranging from six groups for primate experts to 33 for plant experts; Table 1) in all nine land uses (Table 2) or for a subset of species response groups and land uses, depending on the extent of their knowledge. Experts were encouraged to provide any comments relevant to each estimate (e.g., land use characteristics that could influence their score, assumptions that they made, uncertainties, the likely score range across species in a group, and any other explanatory information). Experts could also add general reflections to a comments box at the bottom of the spreadsheet. This qualitative information was useful when aggregating the data (step 2), to gain insight into the reasoning behind experts’ scoring, and detect potential inconsistencies between experts. Experts were asked not to talk to other participants about their estimates to ensure that they did not influence each other’s initial scoring. Experts were encouraged to use other means available to them to inform their scoring, such as talking to colleagues, consulting literature and relevant species reference lists, drawing on experience, and acquiring and interpreting data. They were given two weeks to email their ‘Round 1’ spreadsheet to the project lead.
Discuss: A one to 1.5-hour online meeting was convened, where aggregated (anonymised) results from Round 1 were presented to participating experts. These aggregated results included boxplots showing the range of estimates that experts provided for each response group and land use (and biome, for plants), and 95% confidence interval plots showing trends in mean intactness scores across experts for each response group–land use combination (see examples in the right-hand panels of Figs. 3, 4). Project and expert leads reflected on key trends and sources of variability, and any insights or discrepancies (e.g., when it was apparent from experts’ spreadsheet comments that they were interpreting a given land use in different ways). The project lead then facilitated a discussion among the experts, where they were encouraged to share their experiences from Round 1 (e.g., with what did they struggle; what helped them) and to reflect on the aggregated results (e.g., their insights for the species response groups they know well, or any results that surprised them). They were encouraged to discuss outlying (anonymous) expert estimates and why they may have occurred. The project lead emphasised that the purpose of the discussion was not to reach consensus. Rather, it was to interrogate sources of variability, improve the consistency with which experts were interpreting the species response groups and land uses, and cross-examine reasoning, assumptions and evidence, thereby sharing insights between experts to promote learning. (We found this discussion meeting to be a particularly useful step in the process, with many experts providing feedback saying that they found it an enjoyable opportunity to learn from each other.)
The bii4africa dataset is presented in a multi-spreadsheet .xlsx (Microsoft Excel Spreadsheet) file, which is freely accessible in Figshare. The raw data spreadsheet (‘Scores_Raw’) includes 31,313 individual expert estimates of the impact of a sub-Saharan African land use (Table 2) on a species response group of terrestrial vertebrates or vascular plants (Table 1). Estimates are reported as intactness scores – the remaining proportion of an ‘intact’ reference (pre-industrial or contemporary wilderness area) population of a species response group in a land use, on a scale from 0 (no individuals remain) through 0.5 (half the individuals remain), to 1 (same as the reference population) and, in limited cases, 2 (two or more times the reference population). For species that thrive in human-modified landscapes, scores could be greater than 1 but not exceeding 2 to avoid extremely large scores biasing aggregation exercises. Such truncation is common in standardised biodiversity metrics. Expert comments are included alongside respective estimates.
The raw dataset links, via unique species response group codes, to a spreadsheet (‘Sp_Groups’) describing the species response groups for which experts provided intactness scores (see summary in Table 1). For terrestrial vertebrates, the response group codes also link to a spreadsheet (‘Sp_Vert’) containing all IUCN-listed species in sub-Saharan Africa. Each species has been assigned to one species response group. As large herbivores were initially allocated to response groups but ultimately scored at the species level (see Identifying and describing species response groups section), each species has been assigned the appropriate response group code, with a unique letter differentiating species (e.g., African elephant and black rhinoceros, both in the ‘megaherbivore’ response group LH1 in Supplementary Table 1, are assigned codes LH1A and LH1B, respectively). For terrestrial vascular plants, the response group codes link to a spreadsheet (‘Sp_Plant’) containing aggregated expert-elicited estimates of the proportion of species in each biome that constitute each response group, within the broad groups of trees and shrubs, forbs and graminoids. Biome names also link to a spreadsheet (‘Biomes’) that includes a list of the 89 mainland Afrotropical ecoregions, each allocated to one biome (or two biomes if considered a mosaic).
The raw dataset links via unique, anonymous expert codes to a spreadsheet (‘Experts’) listing the region(s) of expertise of each expert (Central Africa, East Africa, Horn of Africa, Southern Africa, West Africa). A summary of the attributes of the 200 participating experts (including their regions of expertise) is provided in Fig. 2. Expert codes are also linked to a spreadsheet (‘Comments’) containing general comments made by experts. The number of intactness scores provided by each expert (‘Scores_Raw’) varied based on the elicitation that they participated in (e.g., primate experts were asked to score six species response groups, while bird experts were asked to score 17 such groups, see Table 1) as well as their extent of expertise (i.e., some experts provided scores for only a subset of species response groups and/or land uses). On average, each expert provided 155 intactness scores.
The number of experts estimating an intactness score for a given species response group in a given land use (and biome for plants) varied from one to 28, with an average (±SD) of ten (±7) experts providing independent scores for a particular combination. Arithmetic means are the most widely used form of data aggregation in applications of the IDEA expert elicitation protocol. For each species response group in each land use (per biome for plants), we report the mean intactness score across experts, the number of experts providing a score (sample size), as well as the variability in scores between experts (standard deviation, standard error, 95% confidence interval; ‘Scores_Agg’ spreadsheet). Figs. 3, 4 visualise the data at two levels of aggregation. The right-hand panels display examples of the aggregated (‘Scores_Agg’) data for reptiles and bats (Fig. 3) and mixed acacia-savanna plants (Fig. 4). These plots depict the variation in how different species response groups within a given taxonomic group are expected to be impacted by a given land use (differences between response group mean scores). They further depict variation between estimates of individual experts (95% confidence intervals around each response group mean score).
The left-hand panels in Figs. 3, 4 display a further level of data aggregation, in which intactness scores for each taxonomic group in each land use are presented. Each score is an average across the associated species response group means and confidence intervals (‘Scores_Agg’), weighted relative to the proportion of species in that response group (‘Sp_Vert’ and ‘Sp_Plant’). In other words, each species in sub-Saharan Africa counts equally in the aggregated intactness score. Such an aggregation could similarly be performed for all species in a given ecoregion, country, or other spatial unit. The dataset can be downloaded together with an R code script for performing such aggregations.
Notably, the ‘strictly protected areas’ land use does not always have an intactness score of 1 (i.e., equivalent to a pre-industrial ‘intact’ reference population; Figs. 3, 4). These lower scores are because (a) some species benefit from human-related disturbances that would have been present in a pre-industrial landscape but are no longer present in a strictly managed protected area and (b) for particularly large, wide-ranging species, even the best-managed protected area is unlikely to contain a ‘pre-industrial’ reference population (e.g., Prins et al.; Balme et al.). It is also worth noting that not all protected areas in the region are strictly protected in practice – these protected areas were not considered in this land use category, and intactness scores for such protected areas would likely be closer to the ‘near-natural lands’ land use or even some of the other land use categories (e.g., intensive rangelands or non-intensive smallholder croplands; Table 2), depending on the activities occurring in these areas.
The purpose of the facilitated group discussion is not to reach consensus. In our case, experts could disagree on how species are likely to respond to land uses, particularly for lesser-known species groups. Rather, the discussion in a modified-Delphi process aims to resolve linguistic ambiguity, promote critical thinking, share evidence, and improve the consistency with which experts interpret the questions. Thus, we would expect some but not all of the variability in expert scores to reduce in the second round of the elicitation process, subsequent to the group discussion. The standard error around expert scores for each species response group in each land use (and in each biome for terrestrial vascular plants) was lower in the second, compared with the first, round of the elicitation for 85% and 88% of terrestrial vertebrate and plant intactness scores respectively (Fig. 5). On average, there was a reduction in standard error of 0.01 for vertebrates and 0.03 for vascular plants (Fig. 5). Thus, the variability in estimates between experts was generally lower following the group discussions, indicating that the elicitation process resulted in improved scoring consistency between experts.
Validating this dataset using available field-collected data is limited by a spatial and temporal scale mismatch. Most field data are collected at the ‘patch’ scale, while the landscape scale was appropriate for producing this dataset, as explained in the methods. The experts considered diverse land covers and activities characteristic of each land use type (Table 2) to estimate their collective impact on a population in that land use. For example, in an agricultural landscape there is likely to be a higher abundance of fossorial reptiles in the habitat remnants than in the surrounding croplands. This landscape composition was considered by experts when estimating the overall remaining proportion of a reference population of these reptiles in a landscape characterised by non-intensive, smallholder croplands (or intensive, large-scale croplands, with fewer remnant habitat patches). In contrast, field studies tend to report land use at the scale of the habitat or cropland patch. For example, in the PREDICTS database—the largest global dataset of species abundances in different land uses—the only relevant cropland data for Afrotropical birds contains bird counts in ‘private farmhouse gardens surrounded by agricultural matrix’. This ‘patch’ scale, focused on a small subset of land covers and activities in an agricultural landscape, is incompatible with the cropland landscapes that experts considered (Table 2). The spatial and temporal scale of the reference site can also be mismatched. For example, the only relevant plantation data for Afrotropical birds in the PREDICTS database has a reference site of ‘forest fragments in timber plantations’, which is incompatible with the landscape-scale pre-industrial/large wilderness area reference state that experts considered in this study. While this scale mismatch limits validation using existing field data across multiple regions and taxa, future research aimed at validation could design landscape-scale data collection protocols that are more comparable (i.e., developing landscape-scale data collection protocols with multiple multi-taxa sampling sites evenly distributed across different land uses across the region). However, the absence of a ‘pre-industrial’ reference in field-collected data can still impede comparison.
A non-exhaustive list of potential uses of the bii4africa dataset is provided in Table 3. The data are best suited to broad-scale, multi-species applications, rather than finer-scale applications for which site-specific, field-collected data would be more appropriate. The standardised nature of the intactness scores (0–2 scale) means that the data (or a subset thereof) can be aggregated in several ways to meet a user’s needs, e.g., by taxonomic group, functional type, land use and/or spatial (e.g., biogeographical or political) unit (see R code provided for aggregating the data). To assist with data aggregation, scores for response groups can be linked back to individual species for terrestrial vertebrates (noting that large mammal scores are already at species level). Spatial distributions of these species are available from the IUCN Red List. A species list can thus be obtained for the area of interest, and those species scores then extracted from the bii4africa dataset. While the scores for terrestrial vascular plants are not connected to individual species, scores are provided per biome, meaning the data can be extracted at biome-scale, informed for example by the WWF ecoregion maps. The proportions of terrestrial vascular plant species in each response group per biome are also provided, enabling the weighting of score aggregations relative to the proportion of species that those scores represent. Data can also be aggregated according to the regional expertise of the contributing experts (e.g., using scores from only West African experts for a West African data application, or for testing differences between regions). For some species groups such as large mammals and birds, regional considerations such as whether bushmeat harvesting is prevalent can have an influence on experts’ intactness scores.

Sections

"[{\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig1\", \"Tab1\", \"Tab2\"], \"section\": \"Background & Summary\", \"text\": \"We address this data gap through a structured expert elicitation process (Fig.\\u00a01) involving 200 experts in Afrotropical biodiversity on mainland sub-Saharan Africa. Expert elicitation is used widely in conservation and natural resource management when data are insufficient or absent. Notable examples of where expert elicitation has been used include the International Union for Conservation of Nature (IUCN) Red List of Threatened Species, and assessments of the Intergovernmental Panels on Climate Change and\\u00a0on Biodiversity and Ecosystem Services. We employed the latest advances in expert elicitation to ensure data rigour and consistency. Expanding substantially on an earlier approach that involved 16 experts in Southern Africa, experts estimated \\u2018intactness scores\\u2019: the remaining proportion of an \\u2018intact\\u2019 reference population of a given species group (Table\\u00a01) in a particular land use (Table\\u00a02), on a scale from 0 (no remaining individuals) to 1 (same abundance as a reference population) and, in rare cases, 2 (populations that thrive in human-modified landscapes). The \\u2018intact\\u2019 reference is the population abundance that would likely have occurred in the area before alteration by modern industrial society. Because information on species populations from this era is virtually non-existent, standard protocol is to consider a remote wilderness area with a natural disturbance regime (a Hybrid-Historical approach) where necessary. The resulting bii4africa dataset contains standardised intactness scores representing terrestrial vertebrates (tetrapods: \\u00b15,400 amphibians, reptiles, birds\\u00a0and mammals) and vascular plants (\\u00b145,000 forbs, graminoids, trees\\u00a0and shrubs) in sub-Saharan Africa across nine major land uses of varying intensity.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab1\", \"Tab2\"], \"section\": \"\", \"text\": \"Overview of the structured expert elicitation process, based on the IDEA protocol (Investigate, Discuss, Estimate, Aggregate). This process was run for each broad taxonomic group (Table\\u00a01), between November 2020 and January 2022, to elicit from 200 experts the estimated impact of nine major African land uses (Table\\u00a02) on the relative population abundances of terrestrial vertebrates and vascular plants.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"MOESM3\"], \"section\": \"\", \"text\": \"An expert elicitation process was run for each broad taxonomic group, with participants estimating intactness scores for each response group. Supplementary Table\\u00a01 includes details of each response group.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"MOESM2\"], \"section\": \"\", \"text\": \"Supplementary Table\\u00a02 includes representative images of each land use.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab3\"], \"section\": \"Background & Summary\", \"text\": \"The dataset was developed to enable the quantification of the Biodiversity Intactness Index for sub-Saharan Africa (https://bii4africa.org/). It provides an African-led alternative to a previous attempt to map the Biodiversity Intactness Index globally based on a model, which produced inaccurate results for Africa (likely in part because of the under-representation of African data in the model). The bii4africa dataset that we present here has a broad range of additional uses, including assessing and mapping ecosystem condition to inform national planning and reporting on Goal A in the post-2020 Global Biodiversity Framework; alleviating geographic and taxonomic biases in global biodiversity indicators and maps; parameterising models of biodiversity in a changing world; informing the IUCN Red List of Ecosystems; supporting the United Nations Decade on Ecosystem Restoration in identifying priority ecosystems and monitoring the impact of investments into restoration; identifying the properties of novel ecosystems; and informing future research and training in African biodiversity (Table\\u00a03). Importantly, the dataset\\u2014co-produced by 200 experts\\u2014embodies context-specific knowledge on African biodiversity that contributes to inclusivity in ecology. It is also a positive response to the recent call by the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) for African-led research that closes knowledge gaps by mobilising local data.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"MOESM1\"], \"section\": \"\", \"text\": \"Supplementary Table\\u00a03 includes further details on these uses.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig1\", \"Tab1\"], \"section\": \"Elicitation planning and expert recruitment\", \"text\": \"We implemented a published, modified-Delphi protocol (multiple rounds of individual, independent expert estimation interspersed with group discussion and review; Fig.\\u00a01), which has been shown to improve the rigour of elicitation outcomes. The elicitation was limited to terrestrial vertebrates (tetrapods) and vascular plants \\u2013 groups that comprise species with a high diversity of attributes and functions. Experts were sought with knowledge of the degree to which human land uses impact populations of these species groups in sub-Saharan Africa (or a region therein). Such expert knowledge is typically limited to one taxonomic\\u00a0class (e.g., birds or reptiles), or in the case of mammals, often just one or several orders (e.g., primates or bats). As such, multiple expert elicitation processes were run between November 2020 and January 2022, to cover the various broad taxonomic groups of species (see Table\\u00a01). An expert in each broad taxonomic group was invited to lead the elicitation for that taxonomic group. Each \\u2018lead expert\\u2019 was identified based on their relevant expertise and existing network across the continent or willingness to develop such a network. For example, several lead experts serve in the IUCN Species Survival Commission working groups and other regional networks (e.g., https://ascaris.org/; https://www.birdlife.org/our-partners-africa/). The lead expert was responsible for identifying experts and inviting them to participate, as well as proposing a draft list of species response groups and providing input into the development of the land use categories. These tasks are detailed in the subsequent sections.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig2\"], \"section\": \"Identifying and recruiting experts\", \"text\": \"Of the 248 experts who agreed to participate, a total of 200 (81%) participated in the full elicitation process (https://bii4africa.org/category/experts/), with an average of 24 experts per broad taxonomic group (Fig.\\u00a02). These experts were from 39 countries, including 23 African countries (59% of experts were African nationals). Most experts (72%) were resident in Africa at the time of elicitation, across 26 countries. Their experience spanned sub-Saharan Africa, with Southern Africa being best represented, followed by East Africa, and West Africa having the lowest representation. Experts worked in a range of sectors, most commonly at universities, followed by research institutes and conservation organisations. Collectively, experts had over 3,300 years of relevant accumulated experience, with a mean (\\u00b1SD) of 18\\u2009\\u00b1\\u200910 years per expert.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab1\", \"Tab1\", \"MOESM3\"], \"section\": \"Identifying and describing species response groups\", \"text\": \"Within each broad taxonomic group (Table\\u00a01), the lead expert proposed a draft set of species response groups, based on their knowledge of the key organismal attributes likely to determine the impact of different land uses on populations. These draft species response groups were presented to the participating experts for that taxonomic group during an introductory planning meeting (see Structured expert elicitation section below) and revised based on experts\\u2019 feedback. Key organismal attributes that informed species response group categorisation varied between the taxonomic groups (Table\\u00a01; Supplementary Table\\u00a01). For terrestrial vertebrates, attributes commonly included habitat requirements (e.g., forest, grassland, generalist), body size, diet and stratum (e.g., arboreal, fossorial, rupicolous). Terrestrial vascular\\u00a0plants were divided into three broad groups: trees and shrubs, graminoids, and forbs, and species response groups were described within each broad group based on a framework of life-history strategies shaped by disturbance. Species response groups were thus ecologically defined, not necessarily taxonomically defined. Large mammals were an exception to the response-group approach. The elicitation was done at species level for the seven large carnivores and 95 large herbivores. While the large herbivores were initially categorised into response groups, experts decided during the introductory meeting to estimate per species rather than per species response group. Thus, for large herbivores, we present species response groups, as well as species-level estimates. Details of each response group are included in the \\u2018Sp_Groups\\u2019 spreadsheet in the bii4africa dataset.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab2\", \"Tab2\"], \"section\": \"Identifying and describing land use categories\", \"text\": \"Intactness scores were estimated for nine distinct land uses (Table\\u00a02). These land uses with varying intensities of human modification were selected to capture the major land covers, uses and associated activities relevant to sub-Saharan Africa, while being broadly comparable with other land-use maps (e.g., Ellis\\u00a0et al.; Goldewijk\\u00a0et al.; Hurtt\\u00a0et al.). Experts made estimates at a \\u2018landscape\\u2019 scale (i.e., several square kilometres). A finer \\u2018patch\\u2019 scale (i.e., several square metres) would be inappropriate for an expert-elicited approach because experts need to consider multiple activities in an area that impact species. For example, a patch of vegetation in a city is likely to have a different impact on a species than a patch of vegetation in a smallholder cropland or a protected area (particularly for large-bodied or wide-ranging species). Instead, the landscape of land covers, uses and activities characteristic of each land use category were described to experts (Table\\u00a02), who were asked to consider the collective impact on the abundance of a given group of species within that land use.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"MOESM2\"], \"section\": \"Identifying and describing land use categories\", \"text\": \"Experts were provided with photographic examples of these land uses to help visualise them and promote consistency in scoring between experts (Supplementary Table\\u00a02). For each land use, experts were asked to visualise a specific landscape with which they are familiar that matched the description provided, or several such landscapes across which they could consider an average. Experts were instructed to consider the integrated impact of all characteristics of that landscape (e.g., habitat loss and fragmentation, land cover change, disturbance by people, their domestic animals, infrastructure, pollution, harvesting, persecution, introduced species and diseases, elimination of mutualistic species, altered fire and herbivory regimes) on each species response group. Integrated information at a landscape scale is also likely to be more useful for certain decision-maker needs than fine-scale, species-level data.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig1\"], \"section\": \"Structured expert elicitation\", \"text\": \"The IDEA (\\u2018Investigate\\u2019, \\u2018Discuss\\u2019, \\u2018Estimate\\u2019, and \\u2018Aggregate\\u2019) structured expert elicitation protocol was used (Fig.\\u00a01). It is a modified-Delphi procedure that treats each step as a process of formal data acquisition, incorporating research from mathematics, psychology, and decision theory to help reduce the influence of biases and enhance the transparency, accuracy, and repeatability of the resulting estimates. The protocol was implemented as follows for each of the broad taxonomic groups:\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab1\", \"Tab2\"], \"section\": \"\", \"text\": \"Investigate: A one-hour online meeting was held to introduce the project and explain what was expected of participating experts. Experts had an opportunity to ask questions. The lead expert also presented the draft species response groups and received feedback from the participating experts that, in many cases, led to revisions of these groupings. Afterwards, experts were emailed a recording of the meeting, written instructions, and a survey spreadsheet in which to provide estimates. Experts could provide estimated intactness scores for all species response groups (ranging from six groups for primate experts to 33 for plant experts; Table\\u00a01) in all nine land uses (Table\\u00a02) or for a subset of species response groups and land uses, depending on the extent of their knowledge. Experts were encouraged to provide any comments relevant to each estimate (e.g., land use characteristics that could influence their score, assumptions that they made, uncertainties, the likely score range across species in a group, and any other explanatory information). Experts could also add general reflections to a comments box at the bottom of the spreadsheet. This qualitative information was useful when aggregating the data (step 2), to gain insight into the reasoning behind experts\\u2019 scoring, and detect potential inconsistencies between experts. Experts were asked not to talk to other participants about their estimates to ensure that they did not influence each other\\u2019s initial scoring. Experts were encouraged to use other means available to them to inform their scoring, such as talking to colleagues, consulting literature and relevant species reference lists, drawing on experience, and acquiring and interpreting data. They were given two weeks to email their \\u2018Round 1\\u2019 spreadsheet to the project lead.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig3\", \"Fig4\"], \"section\": \"\", \"text\": \"Discuss: A one to 1.5-hour online meeting was convened, where aggregated (anonymised) results from Round 1 were presented to participating experts. These aggregated results included boxplots showing the range of estimates that experts provided for each response group and land use (and biome, for plants), and 95% confidence interval plots showing trends in mean intactness scores across experts for each response group\\u2013land use combination (see examples in the right-hand panels of Figs.\\u00a03, 4). Project and expert leads reflected on key trends and sources of variability, and any insights or discrepancies (e.g., when it was apparent from experts\\u2019 spreadsheet comments that they were interpreting a given land use in different ways). The project lead then facilitated a discussion among the experts, where they were encouraged to share their experiences from Round 1 (e.g., with what did they struggle; what helped them) and to reflect on the aggregated results (e.g., their insights for the species response groups they know well, or any results that surprised them). They were encouraged to discuss outlying (anonymous) expert estimates and why they may have occurred. The project lead emphasised that the purpose of the discussion was not to reach consensus. Rather, it was to interrogate sources of variability, improve the consistency with which experts were interpreting the species response groups and land uses, and cross-examine reasoning, assumptions and evidence, thereby sharing insights between experts to promote learning. (We found this discussion meeting to be a particularly useful step in the process, with many experts providing feedback saying that they found it an enjoyable opportunity to learn from each other.)\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab2\", \"Tab1\"], \"section\": \"Data Records\", \"text\": \"The bii4africa dataset is presented in a multi-spreadsheet\\u00a0.xlsx (Microsoft Excel Spreadsheet) file, which is freely accessible in Figshare. The raw data spreadsheet (\\u2018Scores_Raw\\u2019) includes 31,313 individual expert estimates of the impact of a sub-Saharan African land use (Table\\u00a02) on a species response group of terrestrial vertebrates or vascular plants (Table\\u00a01). Estimates are reported as intactness scores \\u2013 the remaining proportion of an \\u2018intact\\u2019 reference (pre-industrial or contemporary wilderness area) population of a species response group in a land use, on a scale from 0 (no individuals remain) through 0.5 (half the individuals remain), to 1 (same as the reference population) and, in limited cases, 2 (two or more times the reference population). For species that thrive in human-modified landscapes, scores could be greater than 1 but not exceeding 2 to avoid extremely large scores biasing aggregation exercises. Such truncation is common in standardised biodiversity metrics. Expert comments are included alongside respective estimates.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab1\", \"MOESM3\"], \"section\": \"Data Records\", \"text\": \"The raw dataset links, via unique species response group codes, to a spreadsheet (\\u2018Sp_Groups\\u2019) describing the species response groups for which experts provided intactness scores (see summary in Table\\u00a01). For terrestrial\\u00a0vertebrates, the response group codes also link to a spreadsheet (\\u2018Sp_Vert\\u2019) containing all IUCN-listed species in sub-Saharan Africa. Each species has been assigned to one species response group. As large herbivores were initially allocated to response groups but ultimately scored at the species level (see Identifying and describing species response groups section), each species has been assigned the appropriate response group code, with a unique letter differentiating species (e.g., African elephant and black rhinoceros, both in the \\u2018megaherbivore\\u2019 response group LH1 in Supplementary Table\\u00a01, are assigned codes LH1A and LH1B, respectively). For terrestrial vascular\\u00a0plants, the response group codes link to a spreadsheet (\\u2018Sp_Plant\\u2019) containing aggregated expert-elicited estimates of the proportion of species in each biome that constitute each response group, within the broad groups of trees and shrubs, forbs and graminoids. Biome names also link to a spreadsheet (\\u2018Biomes\\u2019) that includes a list of the 89 mainland Afrotropical ecoregions, each allocated to one biome (or two biomes if considered a mosaic).\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig2\", \"Tab1\"], \"section\": \"Data Records\", \"text\": \"The raw dataset links via unique, anonymous expert codes to a spreadsheet (\\u2018Experts\\u2019) listing the region(s) of expertise of each expert (Central Africa, East Africa, Horn of Africa, Southern Africa, West Africa). A summary of the attributes of the 200 participating experts (including their regions of expertise) is provided in Fig.\\u00a02. Expert codes are also linked to a spreadsheet (\\u2018Comments\\u2019) containing general comments made by experts. The number of intactness scores provided by each expert (\\u2018Scores_Raw\\u2019) varied based on the elicitation that they participated in (e.g., primate experts were asked to score six species response groups, while bird experts were asked to score 17 such groups, see Table\\u00a01) as well as their extent of expertise (i.e., some experts provided scores for only a subset of species response groups and/or land uses). On average, each expert provided 155 intactness scores.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig3\", \"Fig4\", \"Fig3\", \"Fig4\"], \"section\": \"Data Records\", \"text\": \"The number of experts estimating an intactness score for a given species response group in a given land use (and biome for plants) varied from one to 28, with an average (\\u00b1SD)\\u00a0of ten (\\u00b17) experts providing independent scores for a particular combination. Arithmetic means are the most widely used form of data aggregation in applications of the IDEA expert elicitation protocol. For each species response group in each land use (per biome for plants), we report the mean intactness score across experts, the number of experts providing a score (sample size), as well as the variability in scores between experts (standard deviation, standard error, 95% confidence interval; \\u2018Scores_Agg\\u2019 spreadsheet). Figs.\\u00a03, 4 visualise the data at two levels of aggregation. The right-hand panels display examples of the aggregated (\\u2018Scores_Agg\\u2019) data for reptiles and bats (Fig.\\u00a03) and\\u00a0mixed acacia-savanna plants (Fig.\\u00a04). These plots depict the variation in how different species response groups within a given taxonomic group are expected to be impacted by a given land use (differences between response group mean scores). They further depict variation between estimates of individual experts (95% confidence intervals around each response group mean score).\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig3\", \"Fig4\"], \"section\": \"Data Records\", \"text\": \"The left-hand panels in Figs.\\u00a03, 4 display a further level of data aggregation, in which intactness scores for each taxonomic group in each land use are presented. Each score is an average across the associated species response group means and confidence intervals (\\u2018Scores_Agg\\u2019), weighted relative to the proportion of species in that response group (\\u2018Sp_Vert\\u2019 and \\u2018Sp_Plant\\u2019). In other words, each species in sub-Saharan Africa counts equally in the aggregated intactness score. Such an aggregation could similarly be performed for all species in a given ecoregion, country, or other spatial unit. The dataset can be downloaded together with an R code script for performing such aggregations.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig3\", \"Fig4\", \"Tab2\"], \"section\": \"Data Records\", \"text\": \"Notably, the \\u2018strictly protected areas\\u2019 land use does not always have an intactness score of 1 (i.e., equivalent to a pre-industrial \\u2018intact\\u2019 reference population; Figs.\\u00a03, 4). These lower scores are because (a) some species benefit from human-related disturbances that would have been present in a pre-industrial landscape but are no longer present in a strictly managed protected area and (b) for particularly large, wide-ranging species, even the best-managed protected area is unlikely to contain a \\u2018pre-industrial\\u2019 reference population (e.g., Prins et al.; Balme et al.). It is also worth noting that not all protected areas in the region are strictly protected in practice \\u2013 these protected areas were not considered in this land use category, and intactness scores for such protected areas would likely be closer to the \\u2018near-natural lands\\u2019 land use or even some of the other land use categories (e.g., intensive rangelands or non-intensive smallholder croplands; Table\\u00a02), depending on the activities occurring in these areas.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Fig5\", \"Fig5\"], \"section\": \"Technical Validation\", \"text\": \"The purpose of the facilitated group discussion is not to reach consensus. In our case, experts could disagree on how species are likely to respond to land uses, particularly for lesser-known species groups. Rather, the discussion in a modified-Delphi process aims to resolve linguistic ambiguity, promote critical thinking, share evidence, and improve the consistency with which experts interpret the questions. Thus, we would expect some but not all of the variability in expert scores to reduce in the second round of the elicitation process, subsequent to the group discussion. The standard error around expert scores for each species response group in each land use (and in each biome\\u00a0for terrestrial vascular plants) was lower in the second, compared with the first, round of the elicitation for 85% and 88% of terrestrial vertebrate and plant intactness scores respectively (Fig.\\u00a05). On average, there was a reduction in standard error of 0.01 for vertebrates and 0.03 for vascular\\u00a0plants (Fig.\\u00a05). Thus, the variability in estimates between experts was generally lower following the group discussions, indicating that the elicitation process resulted in improved scoring consistency between experts.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab2\", \"Tab2\"], \"section\": \"Technical Validation\", \"text\": \"Validating this dataset using available field-collected data is limited by a spatial and temporal scale mismatch. Most field data are collected at the \\u2018patch\\u2019 scale, while the landscape scale was appropriate for producing this dataset, as explained in the methods. The experts considered diverse land covers and activities characteristic of each land use type (Table\\u00a02) to estimate their collective impact on a population in that land use. For example, in an agricultural landscape there is likely to be a higher abundance of fossorial reptiles in the habitat remnants than in the surrounding croplands. This landscape composition was considered by experts when estimating the overall remaining proportion of a reference population of these reptiles in a landscape characterised by non-intensive, smallholder croplands (or intensive, large-scale croplands, with fewer remnant habitat patches). In contrast, field studies tend to report land use at the scale of the habitat or cropland patch. For example, in the PREDICTS database\\u2014the largest global dataset of species abundances in different land uses\\u2014the only relevant cropland data for Afrotropical birds contains bird counts in \\u2018private farmhouse gardens surrounded by agricultural matrix\\u2019. This \\u2018patch\\u2019 scale, focused on a small subset of land covers and activities in an agricultural landscape, is incompatible with the cropland landscapes that experts considered (Table\\u00a02). The spatial and temporal scale of the reference site can also be mismatched. For example, the only relevant plantation data for Afrotropical birds in the PREDICTS database has a reference site of \\u2018forest fragments in timber plantations\\u2019, which is incompatible with the landscape-scale pre-industrial/large wilderness area reference state that experts considered in this study. While this scale mismatch limits validation using existing field data across multiple regions and taxa, future research aimed at validation could design landscape-scale data collection protocols that are more comparable (i.e., developing landscape-scale data collection protocols with multiple multi-taxa sampling sites evenly distributed across different land uses across the region). However, the absence of a \\u2018pre-industrial\\u2019 reference in field-collected data can still impede comparison.\"}, {\"pmc\": \"PMC10861571\", \"pmid\": \"38346970\", \"reference_ids\": [\"Tab3\"], \"section\": \"Usage Notes\", \"text\": \"A non-exhaustive list of potential uses of the bii4africa dataset is provided in Table\\u00a03. The data are best suited to broad-scale, multi-species applications, rather than finer-scale applications for which site-specific, field-collected data would be more appropriate. The standardised nature of the intactness scores (0\\u20132 scale) means that the data (or a subset thereof) can be aggregated in several ways to meet a user\\u2019s needs, e.g., by taxonomic group, functional type, land use and/or spatial (e.g., biogeographical or political) unit (see R code provided for aggregating the data). To assist with data aggregation, scores for response groups can be linked back to individual species for terrestrial vertebrates (noting that large mammal scores are already at species level). Spatial distributions of these species are available from the IUCN Red List. A species list can thus be obtained for the area of interest, and those species scores then extracted from the bii4africa dataset. While the scores for terrestrial vascular\\u00a0plants are not connected to individual species, scores are provided per biome, meaning the data can be extracted at biome-scale, informed for example by the WWF ecoregion maps. The proportions of terrestrial vascular\\u00a0plant species in each response group per biome are also provided, enabling the weighting of score aggregations relative to the proportion of species that those scores represent. Data can also be aggregated according to the regional expertise of the contributing experts (e.g., using scores from only West African experts for a West African data application, or for testing differences between regions). For some species groups such as large mammals and birds, regional considerations such as whether bushmeat harvesting is prevalent can have an influence on experts\\u2019 intactness scores.\"}]"

Metadata

"{}"