Phylogenetic and geographic patterns
Under all future scenarios, reptiles had the highest median PCI scores, followed by mammals and amphibians, while birds had the lowest median scores (S3 Fig and S2 Table). We compared the distribution of average PCI scores and the proportion of threatened species according to the IUCN Red List in each family, for all classes together (S1 Data). 38% of families fall in the same quartile in both prioritization methods. We identified the families that are in the highest quartile for one prioritization method and in the lowest quartile for the other, to identify the taxonomic complementarity of both methods. These were mostly species-poor families, with 70% of them having five species or less (S1 Data). Among the families with high PCI scores and low proportion of threatened species are fossorial reptiles and amphibians, desert mammals, and island birds (S1 Data and S4–S7 Figs).
Broad geographical patterns in average PCI scores per ecoregion were consistent across most scenarios (S8 Fig) and increased mostly in tropical ecoregions under SSP 5.85 in 2100. We focus on SSP 5.85 in 2100 for our subsequent results and discussion of geographic patterns (results for other scenarios can be found in S8 Fig). When considering the average of all classes, arid shrublands in the island of Socotra had the highest PCI scores, followed by high altitude forests in western India and insular tropical forest ecoregions in the Caribbean. These regions were consistently the highest for individual classes, with some important exceptions. Amphibians did not have high scores for islands but had high scores for forests in China and Central Asia. Mammals also had very high scores for Madagascar. For birds, high score islands were mostly in Polynesia, instead of the Caribbean. Average PCI scores for each class in each ecoregion are found in Fig 3 and S2 Data.
The proportion of land vertebrates in threatened IUCN Red List categories for each ecoregion of the world was positively correlated with the average ecoregion PCI scores (PCI effect: 1.971, error 0.970, z-value: 2.031, p-value 0.04). We examined the residuals of this regression to identify regions where the PCI predicts higher conservation priority than expected by the IUCN Red List (Fig 4). These regions were mostly consistent with high PCI score regions discussed above, including Socotra, West India, and the Caribbean. However, arid and semiarid regions around the world emerge as having high PCI scores and low proportion of threatened species. Residuals are especially large in the Arabian Peninsula and northern Mexico (Fig 4 and S3 Data). The proportion of threatened species in an ecoregion was uncorrelated with the ecoregion’s protected area coverage, but this proportion was strongly negatively correlated with average PCI scores (Spearman’s rho: −0.29, p < 0.001). This result is expected, since the PCI incorporates protected area coverage in its calculation.
Regions where the residuals are high (yellow to red) are projected to be more threatened by future threats than current IUCN assessments suggest. Shapefile for ecorregions was obtained from Olson and colleagues 2001 [75]. The data underlying this figure can be found in https://zenodo.org/records/17080841.
https://doi.org/10.1371/journal.pbio.3003422.g004
Median PCI scores increased when assigning higher weights to land use change and artificial habitat intolerance, and decreased when assigning higher weights to climate change, biological invasions, human population density, protected range area, and geographic range size (S2 Fig). This pattern was consistent for most future scenarios, except for the 2100 SSP 5.85 scenario, in which increasing the weight of climate change increased PCI scores, and land use change made no difference (S2 Fig).
We identified four clusters of species (S9 Fig), characterized mostly by differences in their range areas and in levels of change in climate, land use, and human population density they will experience (S10 Fig). The distribution of threat variables between clusters was consistent across future scenarios, but in the year 2100 under the SSP 5.85 scenario, climate change contribution rose substantially for all clusters (S10 Fig). Amphibians and reptiles were proportionally more represented in the third and fourth clusters, which were most influenced by climate change (S11 Fig).
Complementarity of the PCI and the IUCN Red List
The IUCN Red List and the PCI revealed similar results for several species (e.g., the Hula painted frog—Latonia nigriventer and Gyrfalcon—Falco rusticolus; Fig 5), whereas for others the results are vastly different (e.g., the gecko Cyrtodactylus metropolis and the yellow-crested cockatoo—Cacatua sulphurea; Fig 5). High variability in PCI scores within each IUCN Red List category (SD range: 0.06–0.14 standard deviation, on a scale from 0 to 1, Fig 2) reveals that species with very different future conservation priorities are potentially grouped in the same IUCN Red List category (Fig 2). Modifying the weights given to land use change, biological invasions, and human population density did not increase PCI scores’ correlation to IUCN Red List categories. This indicates that PCI may provide different information than the IUCN Red List regarding these projected threats (S1 Fig), which have been forecasted to be major drivers of biodiversity loss in the near future [22]. Unexpectedly, the PCI and the IUCN Red List had higher correlation when increasing the weight of climate change (S1 Fig), which indicates currently threatened species will be especially vulnerable to climate change. Our continuous index reveals new information about future conservation needs that can complement the IUCN Red List when prioritizing species and regions for conservation.
Values in spider plots were scaled between 0 and 1. Values on the upper part of each photograph indicate our assigned PCI scores (inner, values closer to 1 indicate a higher conservation priority, relative to other species in the same class), and their IUCN Red List categories (outer, LC, Least Concern; CR, Critically Endangered; DD, Data Deficient; NE, Non-Evaluated). To keep the area of spider plots proportional to PCI scores, we have inverted variables that reduce PCI score, and changed their names here accordingly: inverted brood size is named “Reproductive Restriction,” inverted range size is named “Range Restriction,” and inverted protected range is named “Unprotected Range.” Cyrtodactylus metropolis photograph provided by L. Lee Grismer, Latonia nigriventer image by UR, all other photographs obtained from Wikimedia Commons (authors: Momofelit, Ólafur Larsen, Charles Lam, Omid Mozaffari). The data underlying this figure can be found in https://zenodo.org/records/17080841.
https://doi.org/10.1371/journal.pbio.3003422.g005
Our results show that species in the Near Threatened category (a “non-threatened” status) have PCI scores most similar to species in the “Vulnerable” category, considered threatened (Fig 2). Near Threatened Species may thus merit conservation attention similar to threatened species. In addition, Data Deficient and Non-Evaluated species, such as Val’s gundi (Ctenodactylus vali) and Papenfuss’s racerunner (Eremias papenfussi; Fig 5), have PCI scores that are, on average, similar to the scores of Endangered and Critically Endangered species (Fig 2). This result adds to accumulating evidence that Data Deficient and Non-Evaluated species are likely to be highly threatened [27–32]. The PCI can offer a way to prioritize the conservation of DD and NE species, which receive less conservation attention than species for which a threat category has been assigned [29], as it offers unique insights into threats these species are expected to face in coming decades. We suggest that conservation practitioners give due attention to conservation prioritization of Data Deficient, Non-Evaluated, and Near Threatened species, using complementary conservation prioritization tools such as the PCI.
A promising tool that might be useful in conjunction with the PCI is the IUCN Green Status of Species [33]. This future-focused score accounts for the future conservation dependence of a species and potential conservation gains both in the short and long term. Green Status assessments are currently available for only 115 species, including 78 terrestrial vertebrates, which makes it impractical to be incorporated in the PCI calculation now, but for the specific cases in which it is available, it could provide a more complete picture of the potential for conservation initiatives to address the threats highlighted by the PCI. For example, the Hainan black crested gibbon (Nomascus hainanus) is classified as Critically Endangered in the IUCN Red List and had the 41st highest PCI score among 5,658 mammal species for the year 2100, under scenario SSP 5.85. Its Green Status Assessment reveals the species is confined to a single protected area, due to land use change and overexploitation. It projects limited potential for increases in its population in the next 10 years, but a higher potential for population recovery in 100 years, based on a population viability analysis. The PCI, however, reveals that this species may also suffer from temperature extremes along all its range in the most extreme scenario for the year 2100. Thus, population recovery efforts could benefit from considering future climate projections, as climate change can affect this vulnerable species even in protected areas.
The PCI was most sensitive to the weighting of land use and human population density across future scenarios (S2 Fig). However, climate change became a dominant component in 2100 under the most extreme climate change scenario (SSP 5.85, S2 Fig). Under a milder emission scenario (SSP 2.45), average PCI scores in 2100 are much lower, especially in tropical regions (S8 Fig). This result highlights the importance of global efforts to mitigate carbon emissions, as climate change can become as great of a threat as land use change, if not addressed immediately. Our cluster analysis showed that species with small ranges will be more vulnerable to climate change, while species with less protected area coverage will be more vulnerable to land use change (S10 Fig). These results highlight that expanding the protected area network could be useful against land conversion, especially for species less tolerant to human presence, but this may be insufficient to protect small-range species, which may lose much of their ranges due to local climatic changes [34].
Phylogenetic and geographic patterns
Reptiles are the vertebrate clade with the highest conservation priority under all future scenarios examined, adding to previous evidence that reptiles might be more threatened than currently depicted in the IUCN Red List [32]. Despite being the most species-rich vertebrate class, reptiles are relatively neglected in conservation research and actions [31,35,36]. Fossorial and cryptic reptiles such as amphisbaenians and blind snakes had high average PCI scores but a low proportion of threatened species in the IUCN Red List (S1 Data). These are poorly studied taxa, which usually lack sufficient information for formal threat assessments [37]. The PCI can offer a practical way to evaluate the conservation needs of hard to study groups, informing conservation actions that may prevent future threats to these neglected species. Reptiles had highest average PCI scores in western India (Western Ghats and Deccan Plateau) and in the Caribbean (Fig 3). Targeted conservation interventions in these areas have a high potential to prevent species loss for this vulnerable and neglected group.
The IUCN Red List shows amphibians as the most threatened class [3]. The PCI may underestimate amphibian threat status since it does not include the threat of chytridiomycosis, a fungal disease considered by many to be a major threatening process for amphibians [38]. However, amphibian mortality associated with fungal diseases is often exacerbated by other threats, such as climate change [39], which are accounted for in the PCI. Regardless, our results show convergent conservation needs between reptiles and amphibians. Our cluster analysis suggests that amphibians and reptiles will be more vulnerable to climate change than mammals and birds under the worst climate change scenario (SSP 5.85; S11 Fig). As in reptiles, families of cryptic and fossorial caecilians had the highest unrecognized conservation needs (S1 Data), and the Western Ghats also emerged as one of the highest priority regions for amphibians. However, unlike reptiles, amphibians showed high average PCI scores in some forests in China and Central Asia (S1 Data).
Islands located mainly in the Caribbean, Polynesia, and the Indian Ocean (including Socotra and Madagascar) had the highest average PCI scores for most vertebrate classes, except amphibians (Fig 4 and S2 Data). Islands have high degrees of endemism and phylogenetic uniqueness [40], and their vulnerable small-ranged [41] fauna is likely to suffer some of the worst consequences of climate change [42]. In fact, 75% of recorded tetrapod species extinctions have been on islands [43–45]. Specialized conservation interventions for the islands we identified as more likely to be threatened, could prevent the loss of ancient and unique evolutionary lineages [42]. Tropical forests in montane regions such as the Western Ghats also had high PCI scores. Montane regions may experience isolation patterns similar to islands: high altitudes offer strong geographical constraints, acting as “sky-islands” [46], which house unique evolutionary lineages [47,48] and are especially vulnerable to climate change [49,50]. Islands and montane regions will be particularly important for the conservation of bird biodiversity. The ecoregions with highest average PCI scores for birds were mostly in Polynesia and western India (Fig 3 and S3 Data). Additionally, bird families with high PCI scores but low proportion of threatened species were also endemic to islands and montane forests around the tropics, including Cuban warblers, rockjumpers, and modulatricids (S1 Data).
Although the ecoregions with the highest average PCI scores for mammals were also concentrated in islands, mainly in the Caribbean and Madagascar, the mammal families with high PCI scores and low proportion of threatened species were mostly from arid regions in Africa and Asia, including brush-tailed mice, gundis, hyenas, and hyraxes. This coincides with our geographical comparison between the PCI and the IUCN Red List (Fig 4), which suggests deserts, steppes, and savannas in Africa, Asia, South America, and North America will require a level of conservation priority not currently reflected in the IUCN Red List. Arid regions are traditionally neglected for conservation efforts and are likely to face severe threats from climate change and land use change in the near future [51].
Using the PCI to inform the prioritization of taxa and regions for conservation may increase efficiency in allocation of limited conservation resources. It can enable targeting species and regions that are predicted to be most affected by global changes in coming decades, potentially informing preventive, rather than palliative, investment in conservation. By design, the PCI is negatively correlated with protected area coverage, and thus highlights areas not currently receiving conservation attention. It also has the advantage of considering different future scenarios of climate change, land use change, and human population density, which may help differentiate types of intervention with higher impact. Our results suggest preventive conservation measures could have a high impact if targeted at arid regions and at insular and montane tropical forests, with particular focus on reptiles and cryptic species.
Caveats and notes on implementation
In some cases, the IUCN Red List can include information on threats not captured by the PCI, as data on many important threat processes are not broadly available. For example, the Critically Endangered yellow-crested cockatoo (C. sulphurea; Fig 5) is heavily affected by pet trade, which was not accounted for in our study and thus has a low PCI score. Other species, such as the gecko C. metropolis (Fig 5), have a higher priority level in the PCI than in the IUCN Red List. Although this species has a very small range, and is located in the highly urbanized metropolitan area of Kuala Lumpur, it is reasonably well protected in the Batu caves, a cultural and religious site.
When calculating PCI scores, we consider only strict protected area categories (see Materials and methods), because the level of protection varies greatly between these categories. We provide data on protected area coverage, including less strict categories, in our R package, so users can incorporate it in the PCI.
The PCI reveals that the small range of C. metropolis will be highly susceptible to high levels of climate change: in the year 2100, under the most pessimistic scenario, C. metropolis is predicted to have all of its tiny range subject to extreme temperatures above historical records. In such scenarios, protected areas may be insufficient to protect this species. Our method is intended to complement the IUCN Red List, so we recommend that, in cases where the IUCN Red List and the PCI disagree substantially, conservation practitioners should examine threatening processes listed as most relevant under each method and use this complementary information to guide conservation actions. Comparisons made at the regional level, such as those displayed in Fig 4, highlight disagreements between the PCI and the IUCN Red List at larger scales, revealing broad patterns of information complementarity between the two methods, despite the limitations of either method at individual cases.
We recognize that the absence of global future projections for key threats, such as overexploitation or disease, can limit the application of the PCI. For species where these threats are critical, the PCI alone can be insufficient. In such cases, the PCI should be combined with other prioritization tools (e.g., the IUCN Red List) to ensure these threats are represented. Importantly, the PCI R package is designed so that additional information on additional threats can be incorporated as soon as global data becomes available. We try to approximate the threat of overexploitation using projections of human population density. Although human population density shows a strong correlation with overexploitation [52], it may underestimate the conservation needs of species affected mainly by this threat, but it may also reveal unrecognized threats which can interact with overexploitation, potentially compounding its threat [53]. Moreover, the variables chosen for our calculation do not consider local contexts or societal features, which may influence conservation priority levels. The PCI, as currently implemented, should be interpreted primarily as a measure of conservation prioritization in regard to the threats of climate change, land use change, biological invasions, and human population density, but users can incorporate any threat using our R package, if data are available.
A major challenge in designing and implementing unsupervised conservation prioritization schemes, such as the PCI, is determining the relative importance of threats and traits. We constructed the PCI and its associated online resources, to provide relative prioritization schemes customized to different sets of species, associated attributes, and weights, in different regions. We ensure users can implement any set of weights they deem appropriate for the context of their application. However, users should strive to determine a more appropriate set of weights a priori, based on empirical data, or on consensus in the scientific literature, instead of arbitrarily adjusting weights to achieve a desired result. Empirical data comparing the relative effect of threats could be used to set these weights in studies with more restricted taxonomic or geographic scope, for which this data could be available. We provide a weight optimization function with the PCI R package, which users can apply to optimize weights so PCI ranks match an external reference (e.g., the IUCN Red List). These optimized weights can then be applied to any set of species, including those not present in the reference prioritization rank or score. We do not apply this tool to our main results as the IUCN Red List is the only external reference available at global scale, and we make diverse comparisons between the PCI and the IUCN Red List.
We acknowledge that any prioritization framework, including the PCI and the IUCN Red List, involves arbitrary thresholds or weightings (e.g., range size thresholds for the IUCN criterion B, rates of decline for criteria A and C, and population sized for criteria C and D). To mitigate this, we make our weighting scheme explicit, provide sensitivity analyses, and offer users open tools to adapt weightings to the context of their use. This transparency allows users to understand where such arbitrariness may affect outcomes and adapt their decisions accordingly, rather than having them remain hidden. As long as the underlying data are available, PCI scores can easily be converted between different sets of weights, using our R package, making it simple to compare the results of different PCI implementations. We recommend that studies comparing different implementations of the PCI first convert values between different sets of weights used. We also recommend that users examine the sensitivity of results to different weight ranges, as we have done in our analyses (S1 and S2 Figs).
To maximize the longevity and accessibility of the PCI, we will maintain the tool through a publicly available GitHub repository (https://github.com/gabrielhoc/PCI), and archive it in the Comprehensive R Archive Network to ensure permanent access. As new global datasets, including other important threats such as overexploitation and diseases, become available, our R package and Shiny application will be updated to reflect the best information available. We will disseminate the PCI through collaborations with conservation organizations, workshops, and outreach efforts, which will be essential for ensuring that the PCI is adopted by a wider user community.
Index calculation
We included four main sources of threats in the calculation of the PCI: climate change, land use change, biological invasions, and human population density—across species ranges. Although not a direct threat, human population density is a driver of threats such as direct exploitation and pollution [7], and thus may serve as a proxy for these hard to measure threats. We also included five species-specific attributes that might modulate the effect of these threats: range size, body size, brood size, proportion of the range in protected areas, and tolerance to human-modified habitats. Although there are many other species attributes that may influence a species sensitivity to the threats, we limited our choice to variables that are widely available for most land vertebrate species. We focused on threats and traits more relevant for terrestrial organisms, as data on these are more readily available, but the index can be easily adapted to aquatic organisms by using our R package to input relevant threats and traits on the index’s calculation (see below). Taxonomic differences between datasets were standardized to match the species list in [5] using the R package “bdc” [54]. The PCI is a relative index, which varies from 0 (the score if a species is the least threatened with respect to every variable considered) to 1 (the score if the species is the most threatened with respect to every variable considered, compared to the other species included in the calculation).
To incorporate projected climate change effects on species, we used the proportion of each species’ range that will be subject to extreme temperature events [5]. These include higher intensity, frequency, or duration of extreme temperatures unprecedented in the species’ recent history, calculated for both 2050 and 2100 under either SSP 2.45 or 5.85 [5]. We chose to focus on extreme climatic events as they have more acute effects on biological systems than changes in mean climate [55,56]. To incorporate projected land use change effects on species, we used the proportion of a species range which will be under anthropogenic land uses (cropland, pasture, and urban areas). We used land use classes projections from the Land Use Harmonization 2 (0.25 degree resolution, [57] for 2050, and 2100 under SSP 2.45 and 5.85. To incorporate projected trends in human population density, we calculated population density (persons/km2) across a species range for the same future years and scenarios [58]. We calculated the percentage of each species range under high or very high future biological invasion threat, using the classification by Early and colleagues 2016 [59] for 2100 under scenario A2. We use the same value of biological invasion threat for every future scenario, as it was the only scenario available. Range sizes (km2) were obtained from the Global Assessment of Reptile Distributions (GARD) for reptiles [36], BirdLife International for birds (www.birdlife.org), and the IUCN for amphibians and mammals (www.iucnredlist.org). We calculated the exposure of species to future threats within their current ranges, as projecting future ranges is beyond the scope of this paper and could limit the number of species to which the method could be applied. However, range shifts may substantially change threat levels [20,60], and users interested in using threat exposure metrics based on future projected ranges can easily do so using our R package. We use future states of land use and human population density in our calculations, as we think it better reflects the exposure of species at that point in time. However, we also include in our R package metrics of current land use and human population density, so users may calculate rates of change for these variables, if they wish to.
For the body size metric, we used maximum adult body mass (grams) obtained from the GARD database for reptiles (https://gardinitiative.org, [61]) and the Global Amphibian Biodiversity Project for amphibians (https://amphibianbiodiversity.org), and mean adult body mass (grams) from AVONET for birds [62] and COMBINE for mammals [63]. Body size data were supplemented by imputed values for missing data (13% for amphibians, 20% for birds, 2% for reptiles, 13% for mammals) using phylogenetic imputation in the R package Rphylopars [64] with previously published phylogenies [65–69]. Likewise, brood size data [61,70] were supplemented by imputed values for missing data (70% for amphibians, 53% for birds, 46% for reptiles, 39% for mammals). For species with missing body size data (1,845) and missing brood size data (3,891) not represented in the phylogenies, we input them with the median values for these variables in their respective families. We further calculated the proportion of a species range which is covered by an IUCN protected area in categories I to IV (which are set aside strictly for biodiversity conservation [71]), obtained from the September 2022 version of the World Database on Protected Areas [72]. To incorporate intolerance of species to human-modified habitats, we used habitat preference information in the IUCN Red List [3]. Species for which artificial environments (those preceded by the tag “Artificial” in the IUCN habitat classification) were classified as “suitable” received a score of 0, species for which artificial environments were classified as “marginal” received a score of 0.5, and the remaining species (those with no “Artificial” tag in their IUCN habitat classification and those not in the IUCN Red List) received a score of 1. Although population size is an important metric to assess vulnerability, it is not available for most species. Some of the traits included in our calculations are correlated to population size and may approximate its effects, such as range size and body size [73]. The PCI is not designed to assess species already severely depleted in the past. We stress that, in such cases, the PCI should be used as complementary to frameworks such as the IUCN Red List, which focus on past and present threats. Used together, these approaches can provide a more complete picture of conservation priorities.
We calculated the PCI for 10,892 reptile, 5,658 mammal, 10,078 bird, and 6,932 amphibian species, for two future years (2050 and 2100), under two SSP (i.e., land use and climatic scenarios: SSP 2.45 predicting relatively mild climate changes and SSP 5.85 predicting more severe/drastic warming). Since the distributions of most variables in the PCI are right-skewed and vary by orders of magnitude (e.g., range size, body size), we added one to the values of each datum and log-transformed it. We then scaled each variable between 0 and 1, using a min-max normalization function ( , Equation 1).
(1)
In which x is a vector being scaled. We then calculated the weighted interaction products (ri) for each threat variable i (Equation 2):
(2)
where ti is a vector containing the scaled values of threat variable i for all species, vj is a vector containing the values of interacting variable j for all species, m is the number of interacting variables, and zi,j is the weight for the interaction between threat variable i and interacting variable j. The symbol ⊙ denotes element-wise multiplication. We then calculated the PCI as the weighted arithmetic mean of the scaled interaction products for each combination of threat variables (Equation 3):
(3)
In which p is the vector of PCI scores for all species, n is the total number of threat variables (we used n = 7, but users can input other numbers of variables using our dedicated R package—see below), and wi is the weight of threat variable i. The index’s calculation is illustrated in Fig 1.
The weighting scheme described above allows users to adjust the index according to which threatening process they consider more important for their specific taxa or location. This approach also enables users of the PCI to further adjust the interacting variables that might have a higher effect on each threatening process. Weights are scaled during the calculation, so weights with different absolute values may produce the same results, if they have the same proportion between each other. We have included a weight optimization feature in our R package, which allows the estimation of weights so PCI scores match an external reference (such as IUCN Red List categories) as closely as possible. This can be used, for example, to calibrate weights using a subset of species for which another ranking or conservation priority index is available, and then use these weights to calculate the PCI for the remainder of the set of species of interest. We did not use this feature in our geographical and taxonomic representations, instead assigning the same weight to each variable, as we aimed to produce scores that are independent of the IUCN Red List, so we can compare prioritization schemes based on the PCI and on the IUCN Red List.
