Policy Research Working Paper 10821 Revisiting Global Biodiversity A Spatial Analysis of Species Occurrence Data from the Global Biodiversity Information Facility Susmita Dasgupta Brian Blankespoor David Wheeler Development Research Group Development Data Group & Environment, Natural Resources and Blue Economy Global Practice June 2024 Policy Research Working Paper 10821 Abstract This paper builds on recent advances in machine-based differences to technical differences in estimation methods pattern recognition to estimate species occurrence maps, or cases where the boundaries of existing expert maps could using georeferenced open-source data from the Global Bio- be revised to reflect species-level patterns in Global Bio- diversity Information Facility. With currently available data, diversity Information Facility reports. The paper uses the the estimation algorithm has produced maps for more than estimated Global Biodiversity Information Facility maps to 600,000 vertebrates, invertebrates, other animals, vascu- explore the global distributions of endemic species and spe- lar plants, and fungi. The algorithm is designed for rapid cies whose small occurrence regions increase their extinction map updates and estimation of new maps with continued risks. It finds a high overall incidence of endemism, with increases in Global Biodiversity Information Facility occur- significant variations across major species groups. It also rence reports. The paper compares the algorithm-produced identifies about 118,000 species whose small occurrence maps with species-matched sets of expert maps for mam- regions create significant extinction risks. For both endemic mals, ants, and vascular plants. Using comparative species and small-occurrence region species, the paper finds pat- density counts in a high-resolution grid, it finds close sim- terns of spatial clustering that identify candidate areas for ilarity in global distribution patterns. It also traces regional cost-effective protection. This paper is a product of the Development Research Group, the Development Data Group, Development Economics and the Environment, Natural Resources and Blue Economy Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at bblankespoor@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Revisi�ng Global Biodiversity: A Spa�al Analysis of Species Occurrence Data from the Global Biodiversity Informa�on Facility Susmita Dasgupta Brian Blankespoor David Wheeler JEL Classifica�on: Q57, C8 Keywords: Conserva�on planning, global biodiversity, species’ occurrence region, endemic and small- occurrence region, Kunming-Montreal Global Biodiversity Framework. Acknowledgments: This research was funded by a grant from the Global Environment Facility to a World Bank program managed by the authors and Dr. Nagaraja Rao Harshadeep, Global Lead, Disrup�ve Technology. We are also thankful to the par�cipants of the webinar conducted during Biodiversity Week, including our colleagues from the World Bank Sustainable Development Global Prac�ce and the Environment, Natural Resources and Blue Economy Global Prac�ce. 1. Introduc�on The world is rapidly losing biodiversity. Pimm et al. (2014) find that the current rate of species ex�nc�on is at least 1,000 �mes the background rate. Corrobora�ng evidence is provided by the Living Planet Index (LPI), an indicator of global biodiversity based on popula�on trends for vertebrate species in terrestrial, freshwater and marine habitats, which has declined by 69% since 1970. 1 The LPI informs the Conven�on on Biological Diversity (CBD) and its Conference of the Par�es (COP). In response to such indicators of rapid decline, 188 governments in the COP ra�fied the Kunming-Montreal Global Biodiversity Framework (GBF) at the COP 15 mee�ng in December 2022. Among other measures, the Framework commits par�cipants to protec�ng 30 percent of the planet by 2030. 2 Effec�ve implementa�on of the Framework will require addressing two difficult ques�ons: (1) What is the spa�al distribu�on of global biodiversity that should be protected? (2) How can protec�ng 30% of the planet best conserve this biodiversity, taking the opportunity value of protected areas into account? This paper atempts to contribute by revisi�ng the spa�al distribu�on of global biodiversity with species occurrence records from the Global Biodiversity Informa�on Facility (GBIF). During the past 15 years, the GBIF’s global repor�ng network has expanded to include occurrences for over 2 million species. During the past two years, the GBIF’s daily increase has been about 1.3 million occurrence records. 3 Since most of these records include loca�onal coordinates, the GBIF data can enable new es�mates of spa�al distribu�ons for previously-unmapped species, as well as improved es�mates for species with exis�ng maps. Species maps underpin spa�al analyses of global biodiversity. If all species are treated equally, then a map of global biodiversity can be created in three steps: (1) Choose the best available map for each species; (2) Overlay all chosen maps on a high-resolu�on global grid; (3) Count the total species incidence in each grid cell. In prac�ce, global biodiversity analyses can modify species coun�ng in three dimensions. First, species may be assigned different weights because they occupy different branches of the “tree of life” that describes overall gene�c varia�on in the global biome. Second, species may have different conserva�on priority weights because their vulnerability to human encroachment differs greatly. Third, species’ distribu�ons may be far from uniform in the spaces enclosed by their maps. If a species’ spa�al distribu�on density is known, its coun�ng weight for each cell can be made propor�onal to its likelihood of occurrence in that cell. The scien�fic literature features plen�ful examples of weighted species coun�ng (e.g., Guo et al. 2022; Jenkins et al. 2015; Pimm et al. 2014; Veach et al. 2017), but the requisite research can require detailed gene�c and environmental data, as well as expert analysis of their roles in the assignment of coun�ng weights. These are inevitably �me-consuming and intensive in technical resources that remain in short supply. In consequence, a large gap has emerged between the population of species with GBIF occurrence records and the population for which research-driven counting weights are available. This gap creates a dilemma for policy makers, who must act quickly to fulfill their Framework commitment to 30% global protec�on by 2030. Global biodiversity could be assessed with expert-mapped, well- researched species only, but this would ignore the huge popula�on of other species with recorded GBIF occurrences. Unfortunately, policy makers cannot automa�cally assume that well-documented species adequately represent the larger popula�on.4 1 For more detailed informa�on, see htps://www.livingplane�ndex.org/latest_results. 2 htps://www.unep.org/news-and-stories/story/cop15-ends-landmark-biodiversity-agreement 3 htps://www.gbif.org/analy�cs/global 4 For further discussion in the case of invertebrates, see Kass et al. (2022). 1 This paper addresses the informa�on gap by developing and applying an algorithm that generates species maps directly from GBIF occurrence data. The algorithm can automa�cally update maps as new occurrence data become available. We generate maps for all species whose data currently sa�sfy our computa�onal criteria (over 600,000). Overlaying these maps with a high-resolu�on grid provides a view of global biodiversity that broadens the tradi�onal focus on vertebrate animals to include more representa�on for invertebrates, other animals, plants, fungi and other non-plant and non-animal species. We also use the maps to develop new indicators of species endemism 5 and species with small occurrence regions. We believe that the new maps are complements rather than subs�tutes for in-depth biodiversity assessments. Where more informa�on is available, the policy process will undoubtedly benefit from incorpora�ng it. At the same �me, we believe that our approach can contribute in three ways. First, for currently-mapped species, fast updates using our algorithm can help iden�fy cases where newly-reported occurrences suggest altera�on of map boundaries. Second, for currently-unmapped species, our approach can provide useful new informa�on for global biodiversity assessments. Third, our mapping exercise can yield new insights about the global distribu�on of endemic species and species that are par�cularly vulnerable to human encroachment. The remainder of the paper is organized as follows. Sec�on 2 describes the GBIF occurrence data and the applica�on of our mapping algorithm. In Sec�on 3, we explore overall results for the 610,694 species maps generated by our exercise. Sec�on 4 extends the results to a new view of global endemism, while Sec�on 5 explores the distribu�ons of small-occurrence region species that are par�cularly vulnerable to human encroachment. Sec�on 6 summarizes and concludes the paper. 2. Mapping GBIF Occurrence Data 2.1 GBIF Data The Global Biodiversity Informa�on Facility provides a constantly-updated open-source repository of geolocated, date-stamped reports of species occurrences from many public ins�tu�ons and NGOs. These reports can be accessed directly from the GBIF’s Occurrence and Maps APIs. 6 In addi�on, the GBIF provides access to its full database through cloud-based services such as Google’s BigQuery and Amazon AWS. We use BigQuery because it provides convenient tools for reducing dataset size before downloading. We limit GBIF occurrence data to geolocated reports since 1970 for species that have at least three uniquely- iden�fied repor�ng loca�ons. 7 Our mapping algorithm operates reliably on several thousand points at most, so we allow a maximum of 20,000 randomly-selected reports per species. 8 We accept the GBIF’s protocols for occurrence report admissibility. 9 5 Species are tradi�onally iden�fied as endemic if they are 100% resident in one country. We also explore the implica�ons of lowering the endemism condi�on to 95% and 90%. 6 htps://www.gbif.org/developer/occurrence and htps://www.gbif.org/developer/maps, respec�vely. 7 Our mapping algorithm requires at least this many observa�ons to work reliably. 8 This dras�cally reduces the size of the download dataset, since some species have millions of occurrence records. For example the American Robin, Turdus migratorius, currently has 21,258,907 reported occurrences. 9 Detailed descrip�ons of the protocols and database elements are provided by the GBIF at htps://www.gbif.org/data-quality-requirements-occurrences. 2 2.2 Mapping Algorithm We begin with an illustra�on of the mapping algorithm that we have used for this exercise. Figure 1 displays GBIF occurrence reports and species maps for Lagidium viscacia, the Mountain Viscacha, whose range includes areas in Argen�na, Chile, Bolivia and Peru. Figure 1(a) presents an expert range map for Lagidium viscacia developed by Burgin et al. (2020). Figure 1(b) shows the loca�ons of the GBIF occurrence reports used for our mapping exercise. The two figures show that reported sigh�ngs of this species include a few points beyond the northern boundary of the expert map, along with many points beyond the southern boundary. Figure 1(c) displays the output of our mapping algorithm, which heavily overlaps with the Burgin map but extends its northern and southern boundary areas to incorporate these sigh�ngs. The mapping algorithm is a byproduct of recent research on machine-based patern recogni�on, cluster analysis and image processing. Formally, it addresses a problem in computa�onal geometry: efficient bounding of a spa�al set, given a subset of points that are actually observed. Sets with irregular shapes are poorly represented by tradi�onal algorithms that draw simple convex hulls. 10 In contrast, the algorithm we employ (alphahull, developed by Pateiro-López and Rodríguez-Casal (2010) 11) can construct con�nuous non-convex boundaries for efficient representa�on. This powerful property has mo�vated rapid adop�on of alphahull for species range analysis in the recent professional literature (Kass et al. 2022; Guo et al. 2022). Of the 610,694 species in our database, alphahull successfully es�mates occurrence maps for 92.9% (567,464). For each species of the remaining 7.1%, we employ a standard k-means algorithm to separate occurrence reports into spa�al clusters and draw a convex hull around each cluster. Overall, our algorithms es�mate occurrence maps for more than 560,000 terrestrial species and more than 41,000 marine species. To our knowledge, this is the largest set of species maps that have been es�mated from open-source data. Our es�ma�on algorithms are designed to support expansion of GBIF species maps with the con�nuing increase in occurrence reports. As an illustra�on, Figure 1(c) displays the algorithm’s applica�on to GBIF occurrence reports for Lagidium viscacia. The alphahull boundary follows the curvilinear north-south orienta�on of the point set, widening and narrowing as the point set expands and contracts. It closely resembles the expert map produced by Burgin et al., while extending it to incorporate new occurrence informa�on. 12 10 The convex hull is the smallest convex polygon that encloses all points in a set. A polygon is convex if it has no corner that is bent inward. 11 This algorithm is a func�on in the R programming language. 12 Two other powerful features of alphahull are not illustrated here, but play an important role in the produc�on of maps for all species covered. The first is exclusion of random outlier points that are distant from a point cluster. The second is produc�on of separate bounded areas for point clusters that are distant from one another. 3 Figure 1: Lagidium viscacia (Mountain Viscacha) – GBIF occurrences and maps (a) (b) (c) Cita�on (a) below) (a) Observed in Bolivia by Miglė Montrimaitė (licensed under http://creativecommons.org/licenses/by-nc/4.0/) 2.3 Overall Mapping Results We have used the alphahull algorithm to compute occurrence record maps for 610,694 GBIF species: 52,433 vertebrates (amphibians, birds, fish, mammals, rep�les); 213,268 arthropods; 32,355 molluscs; 24,109 other animals; 232,693 plants 38,122 fungi and 17,714 other non-plant and non-animal species. Table 1 provides a more detailed accoun�ng. We limit the exercise to GBIF species in the kingdoms Animalia, Plantae and Fungi that have at least five unique geolocated occurrences since 1970. We rely on two methods to remove spurious observa�ons from locales such as zoos and botanical gardens. The first is exclusion of isolated outlier occurrences before map es�ma�on, which happens automa�cally in alphahull. The second occurs a�er map es�ma�on, when we exclude alphahull-bounded point sets with fewer than three observa�ons. Many species maps have mul�ple bounded areas; in those maps, our approach imposes a conserva�ve interpreta�on of the evidence. We drop species maps with single bounded areas when they contain fewer than three observa�ons.13 3. Global Species Distribu�ons 3.1 Spa�al Selec�on Bias GBIF occurrence reports are o�en produced by voluntary exercises that do not employ scien�fic sampling methods. In consequence, spa�al point densi�es in species occurrence reports are posi�vely related to physical accessibility, popula�on density and income (Garcia-Rosello et al. 2023; Borgelt et al. 2022; Isaac and Silman 2011; Reddy and Dávalos 2003). Other things equal, more species sigh�ngs occur near transport arteries; in areas where there are more inhabitants to iden�fy species; and in areas where more inhabitants have disposable income sufficient to support species search and the cost of repor�ng. These factors complicate atempts to map species popula�on densi�es from occurrence reports (e.g., Kass et al. 2022). Alphahull es�ma�on of boundaries is different, because it focuses on exterior points in spa�al sets. However, accurate representa�on s�ll requires a cri�cal minimum number of sigh�ngs in areas that are not advantaged by transport access, popula�on density or disposable income. As Feeley and Smith (2011) note, the accuracy of boundary es�ma�on improves as more sigh�ngs occur in non-advantaged areas. In this context, it is useful to recall that the GBIF occurrence inventory is growing by about 1.3 million new reports per day. The following sec�on tests alphahull boundary mapping from the current inventory by asking whether it provides a view of global biodiversity that is consistent with the view provided by exis�ng expert maps. 3.2 Case Comparisons This paper develops a methodology for mapping the maximal number of species represented by GBIF occurrence reports, with frequent updates. We atempt to fill the gap between the huge number of GBIF species and the smaller inventory of expert range maps, with two primary objec�ves. The first is incorpora�ng more species into an expanded view of global biodiversity. The second is assessing the rela�onship between GBIF occurrence maps and expert range maps that may “undershoot” if their boundaries exclude GBIF occurrences, or “overshoot” if occurrences are persistently absent in part of their bounded areas. 13 This final condi�on may seem redundant, since the ini�al condi�on is exclusion of species with fewer than five unique occurrences since 1970. However, a species can pass the ini�al condi�on but fail the final one because the alphahull algorithm may exclude an outlier point or two from its computa�on of the bounded area. 6 Table 1: GBIF species occurrence maps Group Count Amphibians 5,055 Birds 11,064 Vertebrates [Class] Mammals 4,881 Reptiles 7,644 Fish 23,789 Subtotal 52,433 Araneae 10,438 Coleoptera 44,152 Diptera 23,567 Arthropods [Order] Hemiptera 13,272 Hymenoptera 26,159 Lepidoptera 50,675 Other 45,005 Subtotal 213,268 Molluscs 32,355 Other Animals 24,109 Asparagales 22,978 Asterales 17,439 Caryophyllales 9,554 Ericales 8,574 Fabales 16,277 Vascular Plants [Order] Gentianales 13,811 Lamiales 16,545 Malpighiales 12,790 Myrtales 10,336 Poales 16,845 Other 87,544 Subtotal 232,693 Fungi 38,122 17,714 Other species Total 610,694 7 In this sec�on, we pursue our two objec�ves with three cases that compare our GBIF species maps with expert maps from recently-published research. Our comparisons iden�fy thousands of species with GBIF maps that are also mapped by the research teams. Using these matched species, each comparison assesses the similarity in global biodiversity paterns produced by our GBIF maps and the expert research products. Where the paterns diverge, we explore the technical factors that explain the differences. The first case retains the tradi�onal focus on vertebrates in a comparison with mammal range maps es�mated by Marsh et al. (2022). The second extends the comparison to invertebrates, whose huge numbers are greatly underrepresented by exis�ng expert maps (Kass et al. 2022). As Table 1 shows, our exercise provides more ample representa�on by es�ma�ng maps for 213,268 Arthropods. In this sec�on, we focus on a comparison with maps for ants developed by Kass et al. (2022). Our third case focuses on vascular plants, for which we have es�mated 232,693 maps. Our comparison focuses on the more limited set of vascular plants mapped by Borgelt et al. (2022). 3.3 The Global Distribu�on of Mammals Marsh et al. (2022 – henceforth Marsh) map the na�ve ranges of global mammals using the authorita�ve taxonomy provided by the Mammal Diversity Database (Burgin et al. 2018). Their exercise harmonizes species maps from two major sources: the Checklist of the Mammals of the World (Burgin et al. 2020a, 2020b), and the Handbook of the Mammals of the World (HMW), published in nine volumes by Mitermeier et al. (2013); Wilson et al. (2016, 2017); and Wilson and Mitermeier (2009, 2011, 2014, 2015, 2018, 2019). In our GBIF occurrence maps database, we iden�fy 3,530 mammals that are also mapped by Marsh. We rasterize both sets of maps using a global grid with 0.05 degree (5 km) resolu�on. 14 We compute species densi�es by cellwise addi�on across 3,530 rasters for each set. Figure 2 provides comparisons of cell counts, which are ranked in 10 groups. The broad paterns in the two maps are quite similar. Both cases assign ranks in the highest two groups to Central America, northwest South America, West Africa, East Africa, the northern region of Central Africa, the eastern region of Southern Africa, Western Europe and Southeast Asia. However, they differ notably in some other regions. The GBIF map assigns higher ranks to large areas in Mexico, the western United States and eastern Australia, while it assigns lower ranks to the southeastern Amazon region and South Asia. Where the two paterns diverge, the technical explana�on seems clear. The Marsh maps es�mate native ranges of global mammals, taking account of recorded historical occurrences and biogeographical factors that correlate with the range of each mammal species. The GBIF mammal maps bound the areas where species occurrences have been reported since 1970. Occurrence regions with GBIF ranks higher than Marsh ranks have many species with reported occurrences beyond their es�mated na�ve ranges; regions with lower GBIF ranks have occurrence reports clustered in subareas within na�ve ranges. This difference could reflect underrepor�ng for GBIF species in lower-ranked areas, although many higher-ranked areas seem similarly-disadvantaged for species observa�on. The alterna�ve, and in our view more plausible, explana�on is that lower-ranked regions are populated by many species whose ranges have contracted over �me. Con�nuing accumula�on of GBIF occurrence reports should help resolve this ques�on. 14 For each species map, rasteriza�on assigns a value of 1 to grid cells that overlap with the map and 0 to the other cells. 8 Figure 2 provides a comparison for 3,530 mammals that have maps in both databases. Figure 3 does the same for all terrestrial mammals that are mapped (GBIF 4,881; Marsh 6,360). Comparison with Figure 2 reveals almost no difference for GBIF, while Marsh has generally-higher rankings for Indonesia and Papua New Guinea. Again, it is possible that mammal species are underrepresented in GBIF occurrence reports from the two countries, although this seems more likely for sparsely-populated Papua New Guinea than densely-populated Indonesia. In our view, the more likely explana�on is that the areas populated by many mammals have contracted. 3.4 The Global Distribu�on of Ants In 2002, Clark and May iden�fied a severe taxonomic bias in conserva�on research. Vertebrates accounted for 3% of described species but 69% of published papers, while invertebrates accounted for 79% of described species but only 11% of published papers (Leather, Basset and Hawkins 2008). A comprehensive study by Kass et al. (2022 – henceforth Kass) addresses this problem for ants using a variety of datasets and techniques (including alphahull, the algorithm we employ) to es�mate range maps. In our GBIF occurrence maps database, we iden�fy 5,445 ant species that are also mapped by Kass. We rasterize both sets of maps using a global grid with 0.05 degree (5 km) resolu�on and compute species densi�es by cellwise addi�on across 5,445 rasters for each set. Figure 4 provides a comparison of cell counts, which are ranked in 10 groups. Many areas have similar paterns (e.g., northern North America, Mexico, Central America, northwest South America, Eastern and Western Europe, West Africa, southern Africa, Madagascar, eastern Australia). However, three differences are notable. Both maps iden�fy a large high-ranking region in the Western Hemisphere, but it is further north for GBIF than for Kass. Both maps iden�fy a band of rela�vely high ranks across West and northern Central Africa, linking to a north-south band in East and southern Africa. However, the rankings are generally lower for GBIF than for Kass. The third difference is visible in Southeast Asia, which ranks uniformly higher for Kass than for GBIF. Since Kass also places heavy reliance on alphahull methodology, we atribute these differences to two technical factors. First, the Kass database comes from intensive processing and error checking of records drawn from the Global Ant Biodiversity Informa�cs (GABI) database in July 2020. In contrast, our records are drawn from GBIF occurrence data as of July 2023. Second, our criteria for inclusion of species and es�mated map regions are significantly more conserva�ve. For example, our maps exclude bounded subregions that have fewer than 3 occurrences, while Kass does not. We confine this comparison to matched species in Kass and GBIF because of another significant difference in methodology. Kass joins us in making plen�ful use of alphahull es�ma�on, but our approach is more conserva�ve. We exclude species with fewer than 3 unique occurrences, but Kass includes 5,168 ant species with fewer than 3 unique occurrences. Since alphahulls cannot be es�mated for these species, Kass es�mates their ranges by drawing 30 km buffer zones around the occurrence loca�ons. Given this difference, comparing full database results for GBIF and Kass would effec�vely be comparing apples and oranges. 9 Figure 2: Matched mammal species densi�es, Marsh et al. (2022) vs. GBIF occurrence reports Marsh GBIF Figure 3: Full mammal species densi�es, Marsh et al. (2022) vs. GBIF occurrence reports Marsh GBIF 11 Figure 4: Matched ant species densi�es, Kass et al. (2022) vs. GBIF occurrence reports Kass GBIF Figure 5: Matched plant species densi�es, Borgelt et al. (2022) vs. GBIF occurrence reports Borgelt GBIF 3.5 The Global Distribu�on of Vascular Plants Borgelt et al. (2022 – henceforth Borgelt) have recently developed spa�al density maps for vascular plants in the IUCN global Red List (IUCN 2021). They use maximum entropy (Maxent) models that predict the likelihood of species occurrences from the values of several environmental variables. For each species, they iden�fy “na�ve regions” from a web-scraping exercise using the Plants of the World Online (POWO) database, with regional iden�fica�on standardized from the World Geographical Scheme for Recording Plant Distribu�ons (WGSRPD). Typically, the resul�ng “na�ve regions” are the boundaries of small countries or provinces (GADM level-1 administra�ve units) in large countries. Borgelt es�mates the models using GBIF occurrence data with restric�ve prior condi�ons. The data are confined to the period 2000-2020 to preserve compa�bility with the environmental variables used for Maxent es�ma�on. For each species, georeferenced occurrence reports are pruned to exclude all observa�ons outside of pre-iden�fied “na�ve regions”. Maxent-es�mated species distribu�ons are also confined to na�ve regions. This approach has the advantage of guaranteeing the exclusion of spurious observa�ons from en��es like botanical gardens and private collec�ons in other regions. However, it also incurs the cost of excluding poten�ally-large numbers of occurrence observa�ons that lie outside pre- iden�fied na�ve regions that are arbitrarily defined by na�onal or provincial boundaries. Our exercise draws on a longer �me period (1970-2023) because we are not constrained by the need for compa�bility with environmental modeling variables. In addi�on, we impose no prior geographic restric�ons on the data. As we explained previously, our methodologies es�mate occurrence map boundaries a�er removing spurious single outliers and small isolated occurrence clusters. We have iden�fied 32,339 vascular plant species that are in both databases. As before, we rasterize our occurrence maps and compute cell counts at 5 km resolu�on. Borgelt provides species maps in a raster stack with much coarser resolu�on (50 km). We extract raster layers for the 32,339 common species and add across layers to obtain rela�ve incidence scores for the 50 km grid cells. Then we use mean smoothing to approximate the effect of higher resolu�on. Figure 5 provides the compara�ve results, displayed as ranks in 10 groups. Inspec�on of the two maps reveals essen�ally the same density patern, with the excep�on of somewhat more extensive high-ranking areas in South and Southeast Asia for the Borgelt maps and the United States for our maps. 3.6 Comparison Summary In this sec�on, we have compared our es�mated GBIF occurrence maps with species-matched sets of expert maps from recent publica�ons for mammals, ants and vascular plants. In all three cases, the compara�ve global paterns of species density are quite similar. Where the paterns diverge, the discrepancies can be traced to technical differences. For mammals, the differences between our GBIF maps and expert “na�ve range” maps are atributable to “undershoo�ng”, where the expert map boundaries exclude many GBIF occurrences, or “overshoo�ng”, where GBIF occurrences are persistently absent in parts of the na�ve range maps. For ants, where the research also employs aphahull es�ma�on, the differences are atributable to differences in the source databases and the rela�ve conserva�sm of our approach to map es�ma�on. For plants, where the research employs GBIF occurrences and the global patern similarity is most striking, the few discrepancies are atributable to temporal and spa�al restric�ons imposed by the expert research team. 13 4. The Global Distribu�on of Endemism Endemism plays an important role in the policy dialogue because countries have stewardship responsibili�es for species that are heavily concentrated within their borders. Endemic status is tradi�onally assigned to a species if its range lies en�rely within one country. However, a country’s stewardship could remain cri�cal for a species that resides almost en�rely within its borders. To explore the implica�ons, we overlay the World Bank’s country map with our GBIF species occurrence maps and compute country occurrence region percentages for each species. 4.1 Endemism by Species Group Table 2 summarizes our results when endemic status is assigned to species that are 100%, 95% and 90% resident in one country. With 610,694 species maps tabulated by country, 272,189 species (44.6%) are classed as endemic if the conven�onal 100% criterion is applied. When the criterion is changed to 95% and 90%, species classified as endemic increase to 317,815 (52%) and 339,656 (55.6%), respec�vely. As Table 2 shows, the incidence of endemism also varies widely by species group and endemicity criterion. With the 100% criterion, species groups’ global endemicity percents vary from a minimum of 29.5% for Fungi to a maximum of 54.5% for Mollusks. Reducing the criterion to 90% increases the minimum to 39.0% and the maximum to 62.1%. In this paper, we will employ the tradi�onal 100% criterion for subsequent analyses. However, the results in Table 2 are cau�onary because they show that countries assume stewardship responsibili�es for about 339,656 species if the criterion is reduced to 90%. Table 2: Global endemicity by species group Endemicity % Condition % Endemic Group 100 95 90 Total 100 95 90 Endemic Species Vertebrates 19,475 22,489 24,155 52,433 37.1 42.9 46.1 Arthropods 93,616 112,413 120,830 213,268 43.9 52.7 56.7 Molluscs 17,617 19,243 20,095 32,355 54.4 59.5 62.1 Other Animals 11,326 12,995 13,760 24,109 47.0 53.9 57.1 Plants 113,273 129,945 138,267 232,693 48.7 55.8 59.4 Fungi 11,232 13,645 14,875 38,122 29.5 35.8 39.0 Other Species 5,650 7,085 7,674 17,714 31.9 40.0 43.3 4.2 Endemism by Country Using the 100% criterion, we count endemic species by country and species group. Country scale plays a major role in raw counts, so we standardize by total country species to highlight the rela�ve importance of endemicity in each country and species group. Table 3 provides a summary for the top 30 countries in each species group, sorted in descending order by average ranking for the seven groups. 15 Our results assign overall top 10 status to Australia, the United States, Brazil, Mexico, South Africa, China, New Zealand, Madagascar, Japan and Costa Rica. Even for the top 30 among 297 en��es tabulated, endemicity varies enormously by group. For example, 62% of Vertebrates in Australia are endemic, while 15 Table 3 excludes small island territories that rank high in at least one group, including South Georgia, French Polynesia, Heard and McDonald Islands, Norfolk Island and the Malvinas/Falklands disputed territory. 15 the same is true for only 3% in the United Kingdom. For Plants, Madagascar, Australia and New Zealand have extremely high endemicity percents (89.6%, 88.1% and 84.4%, respec�vely). The maxima are even higher for Arthropods, with endemicity percents in New Zealand and Australia of 92.8% and 91.2%, respec�vely. Molluscs, Fungi and other species more closely resemble Arthropods and Plants in the rela�ve compactness of their occurrence regions. Table 3. Species endemicity %: Top-30 countries by species group Percent of Species That Are Endemic Other Other Country Vertebrates Arthropods Molluscs Animals Plants Fungi Species Australia 62.3 91.2 76.5 71.5 88.1 59.8 57.3 United States 41.3 41.7 57.5 49.5 48.4 27.9 26.3 Brazil 38.5 46.1 70 67.9 51.6 30.1 45.5 Mexico 45.1 47.7 41.4 48 58.1 44.1 40.7 South Africa 37 41.5 81.9 71 67.2 46.1 36.5 China 29.6 56.5 58.1 55.9 34 38.9 16.3 New Zealand 57.5 92.8 89.6 66.2 84.4 81.8 58.9 Madagascar 77.4 85.3 81.6 27.5 89.6 25.7 11.8 Japan 41 55.5 67 65.8 59.2 66.8 28.1 Costa Rica 33.7 48.6 50.2 63.8 42 43.2 35.7 Colombia 37.7 32.7 49.7 40 32.9 35.7 73.7 France 19.9 22 50.4 33.1 25.2 6 7.1 Spain 24.8 39.6 47.8 40.5 43.9 29.4 27.7 New Caledonia 50.8 64.1 54.2 49.1 94.7 76.9 55.6 Ecuador 51.9 60.5 65.6 72.4 46.8 56 80 Papua New Guinea 40.7 53.3 37.4 24.4 61.4 48.8 6.2 Indonesia 29 38.9 13.6 13.9 24.4 10.9 17.7 Peru 32.5 35.3 53.4 44.4 36 39.5 48.3 Canada 7.1 29.7 17.4 34.5 11.3 25.4 19.7 India 39.8 38.7 43.8 66.1 49.4 34.1 28.5 Chile 44.6 62.3 45.5 47.1 50.3 44.8 44.3 Russian Federa�on 14.1 15.5 27.5 48.5 22.2 12.3 51.8 Philippines 66.7 72.7 57 37.4 57.9 0 15.4 Sweden 13 10.1 10 51 42.1 5.1 23.4 United Kingdom 3 23.1 5.2 11.1 39.7 29.3 21.9 Argen�na 26.7 40.5 28.9 56.5 32.4 28 10 Cuba 31.2 32 10 0 70.4 4.5 6.9 Sri Lanka 95.9 88.9 63.2 100 89.3 72.7 88.1 Malaysia 34.2 41.1 61.3 44.4 25.3 51.9 80 Bolivia 27.4 53.6 50 28.6 50.3 46.4 50 16 4.3 Endemism within Countries In this sec�on, we increase spa�al resolu�on to assess the distribu�on of endemic percentages within countries. For each species group, we rasterize GBIF occurrence maps for endemic species and all species using a global grid with 0.25 degree (25 km) resolu�on. We perform cellwise addi�on across all rasters in each set. Then, for each cell, we compute the percentage of species in the cell that are endemic. Figures 6 and 7 display our results for Vertebrates, Plants and Arthropods, with three striking paterns apparent. First, all three groups exhibit huge global varia�on, from 0% to 60% endemic for terrestrial vertebrates and 0% to 100% for Plants and terrestrial Arthropods. Second, some countries have great varia�on within their borders (e.g., Brazil, United States, South Africa, India, China) while others do not (Madagascar, Australia, Papua New Guinea, New Zealand). Third, the overall geographic paterns are quite similar for the three species groups. In all three cases, Australia, New Zealand and Madagascar exhibit very high endemic intensi�es over their en�re areas. In Brazil, by contrast, all three groups exhibit progressively-declining endemic intensi�es from the Atlan�c coast to the western fron�er region. Similarly-broad paterns of varia�on across regions are visible in the United States, Mexico, Colombia, Argen�na, South Africa, Morocco, Ethiopia, India, China, the Russian Federa�on and Indonesia. Countries have added protec�on responsibili�es for endemic species, so large spa�al varia�ons in endemic intensity have significant implica�ons for protected-area policies. Scale economies are important in this context for two reasons. 16 First, monitoring and enforcement of protected-area status is lower-cost for smaller areas. Second, smaller protected areas have lower economic opportunity costs because they divert fewer hectares from economic use. For both reasons, spa�al clusters with high endemic intensity offer the poten�al for cost-effec�ve area protec�on. Figures 6 and 7 iden�fy candidate clusters in many countries. 5. The Global Distribu�on of Small-Occurrence Region Species Jenkins et al. (2015) note that “small range size is the best predictor of ex�nc�on risk and, thus, the first metric for conserva�on priority”. This factor has been studied extensively in the empirical literature (Kraus et al. 2023; Veach et al. 2017; Purvis et al. 2000; Jenkins et al. 2015; Manne and Pimm 2001; Manne, Brooks and Pimm 1999). It has par�cular significance because it is a widely-recognized indicator of ex�nc�on risk that is computable for any species that can be mapped. In contrast, expert assessments of ex�nc�on risks for individual species are limited by their resource intensity. The result is a gap between mappable species in the GBIF’s rapidly-growing occurrence inventory and species with expert risk assessments in the IUCN Red List 17 and other compendia. In this sec�on, we explore the size and global distribu�on of small-occurrence region species in our GBIF maps database. Then we consider the implica�ons for conserva�on policymakers, who must balance the demands of resource scarcity, securing protected territory, and incurring the economic opportunity cost of that territory. At the outset we should note that small-range status is not determinate; no single cri�cal minimum habitat size exists, given the myriad interac�ons between species and habitat characteris�cs that affect ex�nc�on risks. Instead, we begin by exploring the implica�ons of changing the species map area that qualifies for small-occurrence region status. 16 These factors are considered as general proposi�ons, ceteris paribus. Other factors may intervene in par�cular cases (e.g., the presence of valuable mineral deposits in a proposed protected area whose small size would otherwise be considered an advantage). 17 htps://www.iucnredlist.org/ 17 Figure 6: Global percent endemic species, terrestrial vertebrates (top panel) and plants (botom panel) Figure 7: Global percent endemic species, terrestrial arthropods 19 5.1 Species Counts by Occurrence Region Size Table 4 displays the cumula�ve global count for species groups and all species as occurrence region increases from 5 km x 5 km to 200 km x 200 km. 18 Even for occurrence regions of 10 km x 10 km or less, 57,765 species are iden�fied. This increases to 85,310 at 25 km x 25 km or less and 117,946 at 50 km x 50 km or less. Differences across species groups reflect their varying representa�on in the database and group-specific factors. Table 4: Species counts by group and grid scale Occurrence Region Other Vascular Other Categories Vertebrates Arthropods Mollusks Animals Plants Fungi Species 5 km x 5 km 3,029 17,587 3,336 2,843 12,908 3,410 2,046 10 km x 10 km 3,897 22,245 4,502 3,611 17,234 3,921 2,355 20 km x 20 km 5,385 29,016 6,166 4,575 24,611 4,674 2,948 25 km x 25 km 6,020 31,734 6,748 4,931 27,785 4,936 3,156 50 km x 50 km 8,580 42,894 8,976 6,214 41,285 6,125 3,872 100 km x 100 km 12,215 60,914 12,169 8,149 63,173 8,248 5,213 200 km x 200 km 17,522 88,204 16,425 10,927 94,036 11,755 7,303 We believe that the 50 km x 50 km upper bound on cri�cal scale in Table 4 is conserva�ve; many experts might well advocate a higher limit. Further extension increases the small-occurrence region species count to 170,081 at 100 km x 100 km and 246,172 at 200 km x 200 km. From a policy perspec�ve, we are aware that the feasibility and sustainability of species protec�on tend to decline with increases in the number of species protected. Since even the 50 km x 50 km limit qualifies nearly 117,946 species, we retain it for this exercise while recognizing that other analyses may well opt for higher limits. 5.2 Counts by World Bank Region and Income Group Using the species/country database for assessment, we extract the dominant country for all species map areas of 50 km x 50 km or less.19 We then match the data to World Bank assignments of countries to regions for understanding the regional distribu�on of the species with small occurrence regions. Overall, our results show that within the global total of 117,934 small-occurrence region species, 103,208 are in World Bank regions. In percentage of the global total, the majority are in East Asia & Pacific (28.0%) and La�n America & Caribbean (26.9%); two regions are mid-range: East Europe & Central Asia (17.3%), and 18 Our maximum grid scale of 50 km is conserva�ve and designed to focus aten�on on species with highly- vulnerable habitats. In contrast, Jenkins et al. (2015) assign small-range status to species with less-than-median range sizes. If we adopted the same criterion, the maximum grid scale would be 517 km, 10 �mes our maximum scale. At that grid scale small-range status would be assigned to over 180,000 species – far too many for feasible protec�on schemes to include. 19 In most cases small-range species are endemic, so the dominant country is chosen by default. In the other cases, it is the country with greatest area share in the species’ GBIF occurrence map. 20 Sub-Saharan Africa, 11.7%; and two regions have small representa�on (South Asia, 1.9% and Middle East and North Africa, 1.6%). The rest are in North America (10.7%) and other regions (1.8%). Table 5: Small-occurrence region species counts by region Region Count % East Asia & Pacific 33,017 28.0 Europe & Central Asia 20,428 17.3 La�n America & Caribbean 31,756 26.9 Middle East & North Africa 1,893 1.6 South Asia 2,257 1.9 Sub-Saharan Africa 13,857 11.7 North America 12,595 10.7 5.3 Counts by Country Other 2,131 1.8 Total 117,934 Table 6 counts small-occurrence region species by World Bank income group. These results are encouraging for conserva�on, since 85.1% of recorded small-occurrence region species are in High Income and Upper Middle Income countries that have significant resources for conserva�on. Table 6: Small-occurrence region species counts by World Bank income group 20 Income Group Count Percentage High income 54,902 48% Upper middle income 42,892 37% Lower middle income 11,920 10% Low income 5,128 4% Total 114,842 20 The total species count in Table 5 (117,934) is more than the total in Table 6 (114,842) because some poli�cal en��es with small-occurrence region species are not included in the World Bank’s income group assignments. 21 5.3 Counts by Country We find a highly-skewed interna�onal distribu�on when we also count small-occurrence region species by country. Table 7 reports the top-30 countries, led by Australia (12,200 species), United States (10,509), Brazil (6,058), Mexico (5,725) and France (5,261). Table 7: Top-30 countries – small-occurrence region species Country Count Australia 12,200 United States 10,509 Brazil 6,058 Mexico 5,725 France 5,261 South Africa 5,088 Costa Rica 4,278 China 3,516 Colombia 2,955 Madagascar 2,940 Japan 2,614 Spain 2,571 New Zealand 2,258 Ecuador 2,145 Canada 2,004 Indonesia 1,852 New Caledonia 1,848 Russian Federa�on 1,706 Peru 1,477 Papua New Guinea 1,430 Sweden 1,404 Malaysia 1,275 India 1,205 Chile 1,120 Panama 1,072 United Kingdom 1,009 Italy 990 Venezuela, RB 962 Philippines 945 Cuba 923 22 The top 30 have 89,340 of 117,946 species, or 75.7%, while they are only 10% of the 297 enumerated administra�ve areas. The skewed distribu�on principally reflects factors other than country size, since the rank correla�on of country area and small-occurrence region species count is only 0.40. 5.4 Global Varia�ons in Territorial Clustering Resources for habitat protec�on are scarce, so protec�on is most cost-effec�ve in areas inhabited by numerous small-occurrence region species. To explore this factor, we overlay the maps of 117,934 small- occurrence region species on the World Bank country map that includes 38,229 level-2 administra�ve units. In Brazil, for example, these are municipios within the level-1 units that are Brazilian states. We assign each species to the subna�onal administra�ve level 2 (henceforth WB2) unit that has the largest share of its habitat. A�er assigning all species, we compute WB2 species counts and order each country’s WB2s in descending order by count. We compute the cumula�ve count and iden�fy the WB2s that collec�vely account for 90% of the country’s small-occurrence region species. We divide the number of iden�fied WB2s by the total number of WB2s in the country to obtain a concentra�on measure for small- occurrence region species. Tables 8 and 9 summarize our results for 66 lower income and 101 higher income countries, respec�vely. 21 In each table, the rows divide countries into four groups by WB2 concentra�on. Table 8: Small-occurrence region species spa�al clustering by size class, lower middle and low income countries WB2 % Group Species Count Group [0-50] [50-250] [250+] Total 1 [0 - 5.0%] 6 3 0 9 2 [5.1 - 10%] 9 8 1 18 3 [10.1 - 20%] 1 10 5 16 4 [20.1+%] 4 11 8 23 Total 20 32 14 66 21 These are administra�ve en��es that have level-2 units mapped by the World Bank. Lower income countries are classified by the World Bank as low income or lower middle income; higher income countries are classified by the World Bank as upper middle income or high income. The Small occurrence region species counts and WB2 concentra�ons by World Bank income classifica�ons are available upon request. 23 Table 9: Small-occurrence region species spa�al clustering by size class, upper middle and high income countries WB2 % Group Species Count Group [0-50] [50-250] [250+] Total 1 [0 - 5.0%] 5 1 1 7 2 [5.1 - 10%] 4 10 2 16 3 [10.1 - 20%] 9 9 15 33 4 [20.1+%] 10 13 22 45 Total 28 33 40 101 Group 1 countries have the highest spa�al concentra�ons of small-occurrence region species and the greatest poten�al for cost-effec�ve protec�on, other things equal. In this group, 5% or less of WB2s account for 90% of small-occurrence region species. The remaining groups are par��oned as follows: 2[5.1-10%], 3[10.1-20%],4[20+%]. The columns divide countries into three groups by species counts: [0- 50], [50-250], [250+]. Tables 8 and 9 reveal the global significance of small-occurrence region species clustering. In the lower income group, nearly half the countries (27/66) are in WB2 group 1 or group 2 (no more than 10% of WB2s account for 90% of small-occurrence region species in a country). In the upper income group, more than 20% of the countries (23/101) have WB2 group 1 or 2 status. Together, the two tables show that 50 countries have strong spa�al clustering of small-occurrence region species. In an addi�onal 49 countries, 90% of small-occurrence region species are accounted for by 10%-20% of WB2s. To summarize, our results iden�fy a rela�vely small subset of countries where clustering of small- occurrence region species offers excep�onal opportuni�es for cost-effec�ve protec�on. In many cases, these areas may already be protected. We do not perform a global assessment in this paper, although we believe that such an assessment will be a useful future applica�on of our GBIF species maps database. However, we perform an illustra�ve exercise for Brazil, which has 6,058 small occurrence region species and high spa�al clustering (6.5% of municipios account for 90% of small-occurrence region species). Highligh�ng the degree of clustering, Table 10 shows that only 5 of Brazil’s 5,510 municipios account for 838 small-occurrence region species (13.8%). Figure 8 locates these municipios with the top 5 count along with other municipios with at least one species and their states on a map of Brazil. Table 10: Top-5 Brazilian municipios, small-occurrence region species counts State Municipio Count Para Altamira 314 Amazonas Barcelos 145 Amazonas Sao Gabriel Da Cachoeira 138 Espirito Santo Santa Teresa 124 Minas Gerais Diaman�na 117 24 838 Total Figure 8: Top-5 Brazilian municipios, small-occurrence region species counts To summarize, the Brazil case illustrates the poten�al of the GBIF database for iden�fying protec�on opportuni�es offered by spa�al clusters of small-occurrence region species, as well as the extent of exis�ng protec�on for these species. From a global perspec�ve, our overall results iden�fy many countries with high spa�al clustering where cost-effec�ve small-occurrence region species protec�on could be undertaken if it is not already in place. 6. Summary and Conclusions In December 2022, 188 governments ra�fied the Kunming-Montreal Global Biodiversity Framework that commits par�cipants to protec�ng 30 percent of the planet by 2030. Implementa�on with the limited resources available will require accurate iden�fica�on of areas that are both cri�cal for global biodiversity and suitable for cost-effec�ve protec�on. This paper atempts to inform the iden�fica�on process with new species habitat informa�on from the Global Biodiversity Informa�on Facility (GBIF). The GBIF’s global species database now includes occurrence reports for over 2 million species, with a daily increase of about 1.3 million reports. The database can support rapid expansion of occurrence maps for many previously-unmapped species, as well as improved es�mates for species with exis�ng maps. In the paper, we use recent advances in machine-based patern recogni�on to es�mate occurrence maps for over 600,000 species reported by the GBIF. To our knowledge, this is the largest set of species maps 25 that have been es�mated from open-source data. Our es�ma�on algorithm is designed to support expansion of GBIF species maps with the con�nuing increase in occurrence reports. We test the maps produced by our algorithm against recently-published expert maps for mammals, ants and vascular plants. In each case, we select GBIF maps that are also in the published map sets. We use the two map sets to compute species density counts for a high-resolu�on global grid and compare the resul�ng global distribu�ons. We find close similarity in the global comparisons for mammals, ants and vascular plants. Where there are regional differences, we trace them to technical differences in es�ma�on methods or cases where the boundaries of exis�ng expert maps could be revised to reflect species-level paterns in GBIF reports. We use our occurrence maps database to explore two important issues for priority-se�ng in the Global Biodiversity Framework. First, we es�mate the incidence of endemism across broad species groups (vertebrates, invertebrates, other animals, vascular plants, fungi). Endemism has policy significance because it iden�fies countries’ stewardship responsibili�es for species that reside en�rely or almost en�rely within their boundaries. We find a heavy incidence of endemism for 610,694 species in our database, with 272,189 (44.6%) classified as endemic for 100% habitat in one country and 339,656 (55.6%) for 90% habitat. We also find that endemism varies widely by species group. In a high-resolu�on spa�al analysis, we iden�fy candidate “hotspot” areas for protec�on within countries where endemic species are highly concentrated. Our second explora�on focuses on species whose small habitat areas increase their ex�nc�on risk. Our algorithm produces area es�mates for all mapped species, offering a low-cost complement to tradi�onal risk indicators whose species coverage is limited by their resource intensity. Using a conserva�ve standard for minimum cri�cal habitat size, we find 117,934 vulnerable species. Our global assessment finds a highly-skewed spa�al distribu�on that is only modestly correlated with country size. We find many skewed distribu�ons within countries as well, with 90% of small-occurrence region species concentrated in less than 10% of WB level-2 administra�ve units for 50 of 167 en��es analyzed. This creates a significant opportunity for species protec�on that incorporates both ex�nc�on risk and cost-effec�ve administra�on within limited areas. We illustrate the poten�al with a Brazilian case that iden�fies 5 of 5,510 Brazilian municipios that provide habitats for 13.8% of all small-occurrence region species in the country. To summarize, this paper atempts to inform the Global Biodiversity Framework with a large expansion of species maps es�mated from the GBIF’s open-source occurrence records database. Our es�ma�on algorithm is designed for rapid produc�on of new maps as more records become available. We find that the global biodiversity paterns produced by our mapping algorithm closely resemble paterns from exis�ng expert mapping exercises, as well as highligh�ng cases where exis�ng expert maps could be revised to incorporate new GBIF informa�on. In pilot applica�ons, we use our newly-es�mated maps to shed new light on the global distribu�ons of endemic and small-occurrence region species. We hope that many more applica�ons will accompany the con�nued rapid expansion of open-source GBIF occurrence reports. 26 References Borgelt, J., J. Sicacha-Parada, O. Skarpaas et al. 2022. Na�ve range es�mates for red-listed vascular plants. Nature Scien�fic Data, 9:117. Burgin, C. , J. Colella, P. Kahn, and N. Upham. 2018. How many species of mammals are there? Journal of Mammalogy, 99(1): 1–14. Burgin, C. , D. Wilson, R. Mitermeier, A. Rylands, T. Lacher and W. Sechrest. 2020a. Illustrated checklist of the mammals of the world: Vol. 2: Eulipotyphla to Carnivora. Lynx Edicions. Burgin, C. , D. Wilson, R. Mitermeier, A. Rylands, T. Lacher, and W. Sechrest. 2020b. Illustrated checklist of the mammals of the world: Vol. 1: Monotremata to Roden�a. Lynx Edicions. Burgin, C. , D. Wilson, R. Mitermeier, A. Rylands, T. Lacher, and W. Sechrest. 2020. Illustrated Checklist of the Mammals of the World. Lynx Nature Books. Clark, J. and R. May. 2002. Taxonomic bias in conserva�on research. Science, 297: 191–192. Feeley, K. and M. Silman. 2011. Keep collec�ng: accurate species distribu�on modelling requires more collec�ons than previously thought. Diversity and Distribu�ons, 17: 1132–1140. Garcia-Rosello, E., J. Gonzalez-Dacosta, C. Guisande and J. Lobo. 2023. GBIF falls short of providing a representa�ve picture of the global distribu�on of insects. Systema�c Entomology, 48(4): 489-497. Guo, W., J. Serra-Diaz, F. Schrodt F, et al. 2022. High exposure of global tree diversity to human pressure. Proceedings of the Na�onal Academy of Sciences, 119(25). Isaac, N. and M. Pocock. 2015. Bias and informa�on in biological records. Biological Journal of the Linnean Society. 115: 522–531. IUCN. 2021. The IUCN Red List of Threatened Species. Version 2021-1 htps://www.iucnredlist.org. Jenkins, C., K. Van Houtan, S. Pimm and J. Sexton. 2015. US protected lands mismatch biodiversity priori�es. PNAS, 112(16): 5081-5086. Kass, J. B. Guénard, K. Dudley et al. 2022. The global distribu�on of known and undiscovered ant biodiversity. Science Advances, 8(31). Kraus, D., A. Enns, A. Hebb et al. 2023. Priori�zing na�onally endemic species for conserva�on. Conserva�on Science and Prac�ce, 5(1). Leather, S., Y. Basset and B. Hawkins. 2008. Insect conserva�on: finding the way forward. Insect Conserva�on and Diversity, 1(1):67–69. Manne, L. and S. Pimm. 2001. Beyond eight forms of rarity: Which species are threatened and which will be next. Animal Conserva�on, 4:221–229. Manne, L., T. Brooks and S. Pimm. 1999. Rela�ve risk of ex�nc�on of passerine birds on con�nents and islands. Nature, 399: 258–261. Marsh, C. et al. 2022. Expert range maps of global mammal distribu�ons harmonized to three taxonomic authori�es. Journal of Biography, 49(5): 979-992. 29 Mitermeier, R., A. Rylands and D. Wilson. 2013. Handbook of the mammals of the world: Vol. 3: Primates. Lynx Edicions. Pimm, S. et al. 2014. The biodiversity of species and their rates of ex�nc�on, distribu�on, and protec�on. Science, 344(6187). Pateiro-López, B. and A. Rodríguez-Casal. 2010. Generalizing the Convex Hull of a Sample: The R Package Alphahull. Journal of Sta�s�cal So�ware, 34(5). Purvis, A., J. Gitleman, G. Cowlishaw and G. Mace. 2000. Predic�ng ex�nc�on risk in declining species. Proceedings of the Royal Society, Biological Sciences, 267: 1947–1952. Reddy, S. and L. Dávalos. 2003. Geographical sampling bias and its implica�ons for conserva�on priori�es in Africa. Journal of Biogeography, 30: 1719–1727. Veach, V, E. Di Minin, F. Pouzols and A. Moilanen. 2017. Species richness as criterion for global conserva�on area placement leads to large losses in coverage of biodiversity. Diversity and Distribu�ons, 23: 715–726. Wilson, D., T. Lacher and R. Mitermeier. 2016. Handbook of the mammals of the world. Vol. 6: Lagomorphs and Rodents. Lynx Edicions. Wilson, D., T. Lacher and R. Mitermeier. 2017. Handbook of the mammals of the world: Vol. 7: Rodents II. Lynx Edicions. Wilson, D. and R. Mitermeier. 2009. Handbook of the mammals of the world. Vol. 1: Carnivores. Lynx Edicions. Wilson, D. and R. Mitermeier. 2011. Handbook of the mammals of the world. Vol. 2: Hoofed mammals. Lynx Edicions. Wilson, D. and R. Mitermeier. 2014. Handbook of the mammals of the world. Vol. 4: Sea mammals. Lynx Edicions. Wilson, D. and R. Mitermeier. 2015. Handbook of the mammals of the world. Vol. 5: Monotremes and marsupials. Lynx Edicions. Wilson, D. and R. Mitermeier. 2018. Handbook of the mammals of the world. Vol. 8: Insec�vores, sloths and colugos. Lynx Edicions. Wilson, D. and R. Mitermeier. 2019. Handbook of the mammals of the world. Vol. 9: Bats. Lynx Edicions. 30