Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Changes in human footprint drive changes in species extinction risk

Changes in human footprint drive changes in species extinction risk ARTICLE DOI: 10.1038/s41467-018-07049-5 OPEN Changes in human footprint drive changes in species extinction risk 1,2 3 1,4 1,5 Moreno Di Marco , Oscar Venter , Hugh P. Possingham & James E.M. Watson Predicting how species respond to human pressure is essential to anticipate their decline and identify appropriate conservation strategies. Both human pressure and extinction risk change over time, but their inter-relationship is rarely considered in extinction risk modelling. Here we measure the relationship between the change in terrestrial human footprint (HFP)— representing cumulative human pressure on the environment—and the change in extinction risk of the world’s terrestrial mammals. We find the values of HFP across space, and its change over time, are significantly correlated to trends in species extinction risk, with higher predictive importance than environmental or life-history variables. The anthropogenic con- version of areas with low pressure values (HFP < 3 out of 50) is the most significant predictor of change in extinction risk, but there are biogeographical variations. Our framework, cali- brated on past extinction risk trends, can be used to predict the impact of increasing human pressure on biodiversity. 1 2 Centre for Biodiversity and Conservation Science, The University of Queensland, 4072 Brisbane, QLD, Australia. CSIRO Land & Water, EcoSciences Precinct, 41 Boggo Road, Dutton Park, QLD 4102, Australia. Natural Resource and Environmental Studies Institute, University of Northern British Columbia, 3333 University Way, Prince George V2N 4Z9, Canada. The Nature Conservancy, 4245 North Fairfax Drive, Suite 100, Arlington, VA 22203-1606, USA. Global Conservation Program, Wildlife Conservation Society, 2300 Southern Boulevard, Bronx, NK 10460, USA. Correspondence and requests for materials should be addressed to M.D.M. (email: moreno.dimarco@gmail.com) NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 1 1234567890():,; ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 pecies are disappearing at rates that are 1000 times faster Here we compare a 16-year trend in HFP (1993–2009) with a than those registered in the fossil record , and accurate 12-year trend in the extinction risk of 4421 terrestrial mammal Spredictions of extinction risk are necessary to anticipate species (1996–2008). Our goal is to test the existence of a direct declines under past, current, and projected levels of human pres- relationship between changing human pressure, as represented by sure. Understanding the relationship between changes in human the HFP, and changing risk status of species over a comparable pressures and the decline of individual species is necessary for time frame. This allows for the dynamic, as opposed to static, identifying those species at highest risk, and for prioritising the modelling of species extinction risk , and takes advantage of a 2,3 actions and policies required to combat their decline .Com- single, cumulative, representation of how human pressure has parative extinction risk modelling, which builds on the relationship changed over time . We focus on terrestrial mammals as they between species threat status, their life histories, and the pressure have had their extinction risk measured over a similar period as mapped within their ranges, is increasingly used to predict the risk HFP , and they have served as a focal group in several previous 4–8 17 of extinction . This approach allows inferring the extinction risk extinction risk analyses . We classified species into two groups, of a large number of species based on readily available data, and following earlier work : ‘low-risk’ transitions and ‘high-risk’ predictions can be updated more often than expert-based assess- transitions (Fig. 1). The low-risk group included species that 9,10 ments, given the substantially lower resources requirement . retained a category of least concern and species that moved from However, a major limitation in these analyses is the absence of a any higher category of threat to a lower category during the study link to spatial and temporal changes in human pressure and how period. The high-risk group included all species that retained a these lead to change in the risk of species declines .This isfurther category of threatened or near threatened, together with species complicated by two types of change in human pressure, the change that moved from any lower category of threat to a higher cate- in extent of pressures (e.g. road building in a new area), and the gory. We also test a more conservative classification of risk intensification of existing pressures (e.g. increase in deforestation change, where species are considered either ‘uplisted’, if they rates). The missing linkage between pressure and extinction risk moved from any Red List category to a higher category during the means comparative extinction risk analysis has struggled to inform study period, or ‘not uplisted’. We measured the proportion of policy and management . each species’ range overlap with high HFP values, and how this As a species’ conservation status is sensitive to changes in overlap has changed through time, testing all possible definitions 12,13 human pressure , more dynamic extinction risk modelling has of what constitutes ‘high HFP’. We used these values, and other the potential to elucidate links between trends in pressures and known human pressure, environmental, and life-history pre- trends in extinction risk. The recent publication of a temporally dictors of risk (Table 1), to provide estimate of the extinction risk inter-comparable map of human footprint (HFP) presents an transitions of species as a function of change in human pressure important advance in the global representation of changing within their distributions. human pressure on the terrestrial environment. The map, which Our results show the importance of HFP as a predictor of incorporates eight pressure layers standardised into a cumulative extinction risk transition in terrestrial mammals, and suggest the index (see Methods for details), is calculated at two time points conversion of natural and semi-natural areas (those with low HFP and provides an opportunity to investigate the relationship values) has the strongest association with high-risk transitions in between changes in human pressure and changes in the status of species conservation status. We also identified biogeographical biodiversity. HFP provides a spatially explicit index of cumulative differences in the best HFP threshold to determine areas of ‘high human pressure ranging from 0 to 50, where a value of zero pressure’, which can be used for regional monitoring of extinction corresponds to ‘wilderness areas’ free from any significant human risk change. influence , a value of four corresponds to low pressure levels (e.g. pasture lands), and values above 20 typically represents very high pressure levels (e.g. densely populated semi-urban and urban Results areas) . Yet, the HFP is not necessarily a direct measure of threat Global change in human pressure and species extinction risk. to species, and it would be inappropriate to assume that all Much of Earth’s terrestrial surface (30.8%) has undergone an species respond to human activities in the same way. Conse- increase in human pressure, as indicated by HFP values that have quently, the relationship between HFP and species extinction risk increased since 1993 (Fig. 2a; Supplementary Fig. 1a). Two thirds requires testing, in the context of environmental and life-history of those areas already had relatively high HFP values in 1993 characteristics of each species. ( ≥ 4) which became even higher by 2009. At the other end of the a b Low-risk transitions High-risk transitions CR CR CR CR EN EN EN EN VU VU VU VU NT NT NT NT LC LC LC LC Past Present Past Present Fig. 1 Classification of species extinction risk transitions, based on past and present IUCN Red List categories*. Low-risk transitions include those species that were of least concern throughout the study period, together with species that moved from any higher category of threat to a lower one. High-risk transitions include all species that were originally threatened or near threatened and retained their category throughout the study period, together with species that moved from any lower category of threat to a higher category. *Acronyms for the Red List categories: Least Concern (LC); Near Threatened (NT); Vulnerable (VU); Endangered (EN); Critically Endangered (CR) 2 NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 ARTICLE Table 1 Description of the variables used to predict extinction risk transitions in the random forest model Class Variable Description and source Source Pressure High HFP extent Proportion of species range overlapping with high human footprint values in 2009. Pressure High HFP change Difference in the proportional overlap between species range and high HFP during 1993–2009. Pressure Human population Density of human population in year 2000. density Pressure Human population Human population growth, proportional increase in human population between 1990 and 2000. growth Pressure Travel time to cities Accessibility from major cities, measured as travel time. Life history Taxonomic order Species taxonomic orders. 49,50 Life history Gestation length Gestation length, a proxy of species reproductive output. Life history Weaning age Weaning age, a proxy of species reproductive timing. as above 49–51 Life history Body mass A generic proxy of species life history and energetic requirements Life history Diet Dietary category: carnivore ( > 90% animal matter ingested), omnivore (10–90%), herbivore ( < 10%). 35,53 Life history Habitat class Species preferences of macro-habitat categories: aquatic, artificial, caves, desert, forest, grassland, rocky areas, savanna, shrubland, generalists (two or more of the previous categories). Environment NDVI Normalized difference vegetation index, proxy of primary productivity, registered from year 2013. Environment Tree cover Percentage tree cover values registered in year 2000. Environment Habitat prevalence Proportion of suitable habitat within species range. Variables are aggregated into three main classes (human pressure, life-history, environmental characteristics) a b HFP 1993 RL 1996 Min EX CR EN VU NT LC 0 1020304050 LC NT VU EN CR EX Max Fig. 2 Recent changes in terrestrial human footprint and species extinction risk. a Shows a transition matrix in which any position represents the initial (x axis) and final (y axis) human footprint value (from 0 to 50) of global 1 km terrestrial grid cells; the colour scheme represents the number of individual cells in each particular transition state. b Shows a transition matrix in which any position represents the initial (x axis) and final (y axis) extinction risk category (from Least Concern to Extinct) of terrestrial mammal species; the colour scheme represent the number of individual species in each particular transition state spectrum, most of the areas that did not face an increase in Measuring human pressure within species ranges. The cumu- human pressure (41.5% of the total terrestrial surface) are char- lative distribution of HFP values within species geographic ranges acterised by a relatively low HFP value ( < 4). Half of these low- followed similar patterns across the two species groups (high-risk HFP areas have been identified as the last remaining terrestrial and low-risk) and across years (Fig. 3). However, high-risk spe- ‘wilderness’, which is free of any significant human disturbance cies had on average a larger proportion of their range overlapping (HFP = 0). with high HFP values, compared to low-risk species. The level of When looking at the transitions in species extinction risk, we overlap with those areas classified as wilderness was comparable found that 69% of species faced a low-risk transition, while 31% between high-risk and low-risk species, while the biggest differ- faced a high-risk transition (Fig. 2b; Supplementary Fig. 1b). This ences among the two groups was observed for HFP values in the is largely due to 1,229 (27.8%) threatened and near-threatened range 3–15, which correspond to moderate or high levels of species retaining their Red List category, and in minor part to 159 human pressure . This was reflected in significantly higher mean uplisted species (3.6%) that moved towards higher Red List HFP values within the range of high-risk species compared to −12 −16 categories. Only 22 (0.5%) species moved towards lower Red List low-risk species (p-value = 2*10 in 1993 and 2* 10 in 2009; categories during the study period. Wilcoxon signed rank test, one-sided; Supplementary Fig. 2). NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 3 HFP 2009 RL 2010 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 Low-risk High-risk Low-risk 1993 Low-risk 2009 High-risk 1993 High-risk 2009 0 1020304050 0 1020304050 HFP HFP threshold Fig. 3 Cumulative extent of human footprint values within mammal species ranges. The lines represent the cumulative percentage of species range that Fig. 4 Changes in the overlap between species ranges and high human overlaps with increasing values of human footprint, averaged among all footprint values over time. The lines report the average change in the species. Different lines refer to species in the low-risk or high-risk overlap between species ranges and human footprint values bigger than categories, for the period 1993 or 2009, as reported in legend any given threshold. Different lines refer to species in the low-risk or high- risk categories, as reported in legend. The shaded areas around the lines High-risk species typically faced a larger change in the extent of represent the 95% standard credible interval measured across a total of high HFP values compared to low-risk species, with the exceptions 4421 species of very low HFP thresholds (Fig. 4). When looking at HFP thresholds between 0 and 2, low-risk species had a larger modified landscapes. We found that pressure variables had higher proportion of their range moving from below- to above- predictive performance compared to life-history and environ- threshold values compared to high-risk species. This might be mental variables, highlighting the magnitude of human influence related to many threatened and declining species retaining little on environmental trends , and the two HFP variables were the natural areas within their range at the beginning of the study most important predictors in the model (Fig. 5). The model period, with consequent little chances of observing an increasing showed good overall classification ability during cross-validation HFP in natural areas during the study period. For HFP in the (species correctly classified = 82.9%), albeit the accuracy in pre- range 3–49, high-risk species consistently showed higher propor- dicting high-risk species (sensitivity = 60.4%) was lower than the tions of their range moving to above-threshold values, with a accuracy in predicting low-risk species (specificity = 92.4%; True difference that was significant for thresholds in the range 6–44 Skill Statistics = 0.53). −6 −2 (p-values= 2*10 –2*10 ; Wilcoxon signed rank test, one- sided). Overall, the largest effect size for the difference in extent The biogeography of human pressure within species ranges. of high HFP values between low-risk vs high-risk species was We observed some differences among realms and biomes in observed for a HFP threshold of 3 (Cohen’sd = 0.43) and terms of HFP change patterns (Fig. 6). For example, the HFP decreased afterwards, while the effect size for change in the extent value at which high-risk species had the largest proportions of of high HFP values increased up to a threshold of 6 and then their ranges moving above threshold was < 6 in the Nearctic, stabilized (with Cohen’s d values in the range 0.20–0.22; Neotropical and Afrotropical realms, and > 6 in the Palearctic, Supplementary Fig. 3 and 4). Indomalay and Australasian realms. Most realms showed general When looking at the difference in extent of high HFP values for consistency with the global analysis in that high-risk species had a uplisted vs. not uplisted species, we found even larger differences higher proportion of their range moving toward higher HFP than those described for low-risk vs. high-risk species (Supple- values compared to low-risk species, especially when looking at mentary Fig. 5), with substantially higher values for uplisted intermediate and high HFP thresholds. However there were species when looking at HFP thresholds between 2 and 20. exceptions in the Afrotropical and Indomalay realms. In the Afrotropical realm, the exception emerged for grassland biomes Modelling transitions in species extinction risk. We measured (Supplementary Fig. 7), where low-risk species showed larger the performance of HFP in predicting low- vs. high-risk transi- conversions to high HFP values than high-risk species, when tions in species extinction risk, using a random forest model for considering thresholds in the range 7–12. In the Indomalay classification . In this analysis, we compared the predictive realm, low-risk species had similar (or even higher) proportions performance of HFP with a number of other pressure, life-history, of their ranges moving toward higher HFP values compared to and environmental variables (Table 1). We measured HFP both high-risk species. This was especially the case for species living in as the current extent of high HFP values within species ranges, the moist forest biome, which contrasted with the results obtained and as the change in high HFP values over the time period for the same biome in other realms. When looking at dry forest (1993–2009). We adopted all possible thresholds to determine species in the Indomalay, we found low-risk species had faced low vs. high HFP values (from HFP > 0 to HFP > 49), and found larger increase in the extent of HFP values in the range 9–18, that the importance of HFP variables as predictors decreased with while high-risk species have faced a higher change for HFP values increasing thresholds (Supplementary Fig. 6). Overall, a HFP above 18. threshold of ≥ 3 resulted in the highest prediction performance Despite the differences in HFP change patterns observed across across the two HFP variables (current extent and change over biogeographical domains, we still found that relatively low time), indicating that human pressure intensification in intact and thresholds resulted in the highest predictive performance of near-intact areas is globally more relevant, for explaining “HFP change” as a variable in random forest models developed extinction risk transitions, than intensification within already- for separate realms (Supplementary Fig. 8). The lowest threshold 4 NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications Cumulative range % % change in high HFP extent NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 ARTICLE High HFP extent High HFP change Travel time to cities Order Realm Human pop. dens. Tree cover Body mass NDVI Human pop. growth Weaning age Gestation length Habitat class Habitat prevalence 0 20 40 60 80 100 Mean decrease accuracy High HFP extent High HFP change Human pop. dens. Body mass Weaning age Travel time to cities Human pop. growth Tree cover Habitat prevalence NDVI Gestation length Pressure Order Life-history Realm Environment Habitat class 0 50 100 150 200 Mean decrease gini Fig. 5 Predictive importance of variables for the prediction of extinction risk transitions in terrestrial mammals. Variables are colour-coded accordingto their broad class (human pressure, life-history, environmental characteristics). Different plots refer to different measures of variable importance: a variable effect on the overall decrease in prediction accuracy, and b contribution of the variable to decrease Gini Index during the classification routine. A description of all the variables can be found in Table 1. In this analysis, “high HFP" included values of 3 or above (HFP ≥ 1) was observed in the Nearctic realm, while the highest Identifying a threshold of human pressures beyond which threshold was observed in the IndoMalay realm (HFP ≥ 5). species show negative response is essential for monitoring land conversion rates in the context of international biodiversity tar- 15,23 gets . Yet the definition of “high pressure” levels has remained elusive in global analyses so far. Our results show that averting Discussion In order to proactively inform monitoring and management, it is the conversion of natural and semi-natural areas, those with HFP necessary to know the conditions under which a species is likely to values ≥ 3, is the most effective strategy to prevent species from retain an unsustainable high risk of extinction, or to face increased undergoing a high-risk transition in their conservation status risk over time . We found that the extent of high human pressure when accounting for environmental and life-history traits. These (as defined by the HFP index) within species ranges, and the results are in line with recent findings that deforestation within change in this extent over time, were strong correlates of extinc- intact landscapes is the strongest correlate of decline in forest tion risk transitions. These two variables (state and change of high species , opening up the path to a number of direct threat HFP values) were found to be the strongest predictors of risk mechanisms (such as hunting, diseases spread, and invasive when compared to an array of other variables, including species’ species). However, protecting natural and semi-natural land- traits, environmental conditions, and individual pressure layers. scapes is not sufficient to improve the status of species which are already at a high risk, some of which have little natural habitat left This result contrasts with the findings from previous extinction risk modelling for mammals, where the importance of human within their distributions and will require habitat restoration to reduce their risk of extinction. In fact, high-risk and low-risk pressure as predictors was found to be lower than environmental 3,17,21 or life-history variables . This shows that temporal cumula- species showed markedly different changes in their overlap with tive pressure mapping is a powerful tool for improving extinction intermediate HFP values with high-risk species facing con- risk modelling and forecasting, coupling changes in human sistently larger increase in HFP levels. This was confirmed, with pressure with changes in biodiversity state. an even stronger pattern, when looking at uplisted vs not uplisted NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 PA NA 4 IM 2 10 0 5 10 15 20 25 30 0 5 10 15 20 25 30 HFP threshold HFP threshold 0 5 10 15 20 25 30 HFP threshold HFP threshold 12 AA NT AT 0 5 10 15 20 25 30 0 5 1015202530 HFP threshold 0 5 10 15 20 25 30 HFP threshold HFP threshold >20 Fig. 6 Changes in the overlap between species ranges and high human footprint (HFP) values for each biogeographic realm*. The underlying map reports, for each biome within each realm, the threshold at which change in HFP values is the highest for high-risk species. The plots report the average change in the overlap between species ranges and HFP values bigger than a given threshold within each realm (with high-risk species in red and low-risk species in blue). The shaded areas around the lines in the plots represent the 95% standard credible interval measured across a total of: 493 AA species, 854 AT species, 604 IM species, 259 NA species, 949 NT species, 442 PA species*. *Realm acronyms: AA Australasia, AT Afrotropical, IM Indomalay, NA Nearctic, NT Neotropical, PA Palearctic species, albeit this latter classification could not be used for biogeographic realms and in combination with other variables, we extinction risk modelling due to very high imbalance in species found that low thresholds (in the range 1–5, depending on the numbers between the two classes. The identification of this HFP realm) still performed the best in separating low from high HFP threshold (i.e. 3), and what happens when changes occur in the values. This demonstrates that the conservation of intact areas, HFP around this threshold, provide simple guidelines for iden- and areas with little human modification, is relevant at the scale tifying tipping points beyond which human activities might lead of individual realms and not only globally. to species decline. Our results showed that high-risk species had faced larger We found some biogeographical differences in the way low-risk increases in pressure levels within areas of moderate HFP values, and high-risk species overlap with HFP values. Particular while low-risk species had faced larger pressure increases in areas exceptions were found in the grassland biomes of the Afrotropical of former low HFP (those < 2). This includes the loss of wild- realm and in the moist forest biome of the IndoMalay realm, erness areas, which was more likely to occur within the range of where low-risk species faced similar (or higher) increase in HFP low-risk species than high-risk species. This finding is probably values compared to their high-risk counterparts. These exceptions related to the fact threatened species are less likely to overlap with might indicate that species living in those environments are wilderness areas compared to non-threatened species, as a relatively resilient to human pressure as measured in the HFP reflection of pressure operating within their past ranges. This index, and respond more strongly to other pressures not incor- finding means that the continuous conversion of intact and near- porated in the index such as fire regimes, especially relevant in intact areas will likely result in species that are currently classified African grasslands , and overexploitation, relevant in Southeast as low-risk to become high-risk in the future. The loss of intact Asia . However there is also the possibility that some species lands within the ranges of low-risk species should thus act as an currently classified as low-risk might actually be facing a higher early indication of a trajectory of increasing species endanger- risk of extinction than previously thought, especially in forested ment, and points to the need of identifying, and securing, those biomes, as it seems to result from recent, rapid, deforestation . remaining intact ecosystems. These results support the call for a status re-assessment of these Our model was better able to correctly classify low-risk species species within these regions. Interestingly however, when HFP than high-risk species. This might be related the fact that some change was considered as a predictor of risk within separate high-risk species are responding to different components of 6 NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications % change in range extent % change in range extent % change in range extent % change in range extent % change in range extent % change in range extent NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 ARTICLE pressures, not well represented here, while low-risk species are were assigned a score of 4, which decayed exponentially out to 15 km away from the waters. After each pressure layer was standardised within the same values just not facing significant pressures. However this same pattern range, they were summed together to create a cumulative map of human pressure. was also found (at various degrees) in other extinction risk The results are two globally standardised HFP maps, with values ranging from 0 to 3,7,8,28 modelling exercises , indicating that the condition under 2 50 and a spatial resolution of 1 km , one for the year 1993 and one for 2009 (based which species are likely to undergo a low-risk transition are likely on pressure layers referred to the different periods). In this analysis we use the integer version of the HFP maps, to only represent integer changes in the index easier to identify compared to the conditions leading to high-risk (+ 1, + 2, + 3 etc.). transitions. When human pressure is operating, the risk for We measured the change in HFP values for each 1 km terrestrial grid cell species is determined by a complex combination of pressure between 1993 and 2009, and contrasted this change with species geographic levels, species’ sensitivity and their potential for adaption , which distributions to understand changes in perceived treat levels for the species. HFP change can result in an increase in pressure level, i.e. from lower to higher values, determine higher levels of uncertainty in the predictions. This or a decrease in pressure, i.e. the opposite. Here we only accounted for increases in adds a level of complexity in understanding the relationship HFP values, as decreases in pressure levels (e.g. abandonment of agricultural land) between pressure change and change in species extinction risk. are likely to take time before having a measurable effect on species threat status, The current availability of HFP maps (for the years 1993 and especially for species with a long generation time such as some of the high-risk 2009) allowed us to test the relationship between human pressure species in our dataset (average generation time for high-risk species is 7 years, compared to 4.5 years for low-risk species). We used HFP and its change over time and extinction risk over a similar time frame, that encompasses as a predictor of extinction risk change, in combination with previously identified three or more generations for 73% species in our analyses . This variables (Table 1). relationship between change in HFP and change in extinction risk can serve as a basis for future research. For example, our mod- Species extinction risk change. We represented the extinction risk of terrestrial elling framework, calibrated on observed trends, can allow pro- mammal species using the information available from the IUCN Red List , and jecting future extinction risk transitions under alternative the retrospective Red List Assessments published in Hoffmann et al. . We con- sidered the 2010 IUCN Red List categories of each species and the retrospectively scenarios of socio-economic development . This will require assigned categories for 1996. These latter categories were defined using the same generating future projections of the base pressure layers that methodology as the 2010 assessments, but referred to the past condition of the constitute the HFP map, as well as predicting the shift in species species. We considered the following IUCN Red List categories, assigned using a set distributions due to future climate and land-use change . Also, 36 of five quantitative criteria with associated sub-criteria and thresholds : Least the established relationship between HFP and extinction risk Concern (LC); Near Threatened (NT); Vulnerable (VU); Endangered (EN); Cri- tically Endangered (CR); Extinct in the Wild (EW); Extinct (EX). We excluded transitions can be used to estimate the risk faced by ‘data defi- species not evaluated in the Red List, those without a defined risk of extinction 28,33 cient’ species , but this would require resolving the taxonomic category (Data Deficient), and those already extinct at the beginning of the study and geographical uncertainty characterising species in this period. We retained 4421 species of terrestrial mammals with a defined extinction category. risk category for the years 1996 and 2010, corresponding to 83% of all species in the group. Efforts to integrate human pressure maps and extinction risk 9 We recorded the initial (1996) and final (2010) Red List category of each has the potential to change the way we assess species risk and species, and followed Di Marco et al. in classifying species into two main groups proactively inform conservation action in a way that minimises (Fig. 1): low-risk transitions and high-risk transitions. The low-risk group included the number of species that will face a high risk of decline. Con- species that were LC throughout the study period, together with species that moved from any higher Red List category to a lower category (i.e. category ‘downlisting’). servation organisations that have a mission to prevent the decline The high-risk group included all species that were originally threatened or near of species can use our approach to prioritise actions for mini- threatened and retained their category throughout the study period, together with mizing extinction risk. These include both species that are already species that moved from any lower Red List category to a higher category (i.e. threatened with extinction and species that are likely to become category ‘uplisting’). The method behind this classification has been statistically justified , and reflects the fact that remaining within the same Red List category so if current rates of intensification in the human footprint through time does not necessarily imply that a species is in a stable condition. For continue into the future. The HFP index presents a standardised instance, while a species that retains a LC category is not undergoing a significant representation of human pressure levels, combining different population decline (or loss of geographic range), a species retaining a threatened human activities that represent potential sources of impact for category implies substantial continued decline . species. While this index represents a comprehensive and easy We also tested the use of a more conservative approach for classifying extinction risk transitions, where species were separated into two groups: ‘uplisted’ tool for estimating change in species extinction risk and guide species, those that had a deterioration in their Red List category (eg from Least broad-scale conservation efforts, we acknowledge that it cannot Concern to Near Threatened), and ‘not uplisted’ species, those that retained the substitute local-scale assessments of the conservation needs of same category or improved it. This classification can be seen as a more conservative each species. Instead, knowing which species and which areas are approach for defining extinction risk transitions, because in this case the ‘high-risk’ group only includes transitions that are of sufficient magnitude to generate an most likely to face a high risk can guide the prioritisation of local- upward shift in Red List categories. This classification however generated a large scale assessments by conservation practitioners. imbalance between species groups, with only ~4% of species being included in the uplisted class. Methods Human footprint state and change. We used the recent release of the global HFP Human footprint as a driver of extinction risk change. Several methods are 14,16 map , to represents the cumulative human impact on the environment. This available to measure the level of overlap between a spatial pressure layer and a map is built from eight base layers: (i) the extent of built environments; (ii) crop species’ geographic range . These include both measures of central tendencies, e.g. land; (iii) pasture land; (iv) human population density; (v) night-time lights; (vi) the mean/median pressure level observed within the range, and measures of spatial railways; (vii) roads; and (viii) navigable waterways. Following the approach ori- extent, e.g. how much of the species range is covered with high pressure levels. ginally proposed by Sanderson and colleagues , each layer was placed in a Measuring the extent of high pressure levels within a species’ range has been shown 1–10 scale with a value weighted according to the relative intensity of human to be a more sensitive way to predict extinction risk than using mean pressure 14 22 pressure (see Venter et al. for full justification and validation): (i) all built levels and was often a preferred choice in comparative extinction risk model- 21,37 environments were assigned a score of 10 while non-built environment had a score ling . However, identifying the best way for separating low and high pressure of zero); (ii) areas mapped as croplands were assigned a score of 7; (iii) areas levels requires testing multiple thresholds. mapped as pasture lands were assigned a score of 4; (iv) areas with a high human We measured the cumulative overlap between 1993 and 2009 HFP values population density of > 1,000 people/km received a score of 10, while areas with within species ranges, generating curves to represent how much of a species’ range lower density received a lower log-scaled score; (v) areas were divided into 10 overlaps with increasing values of HFP (from 0 to 50). We used the same species quantiles of increased night-time light intensity associated to score of 1 to 10, while distribution range maps for these measures, since past range maps for the areas with no lights were assigned a zero; (vi) railways and their immediate 500 m ~4500 species included in our analyses were not available. Given our study period buffers were given a score of 8, with a value of zero elsewhere (i.e. assuming no was reasonably restricted (16 years), we assumed change in the extent of species indirect impact); (vii) roads and their immediate 500 m buffers were given a score geographic range was overall limited. We generated separate curves to represent of 8 (direct impact), while nearby areas up to 15 km had score that decayed the average accumulation of HFP values in low-risk and high risk species, both in exponential to zero (indirect impact); (viii) areas adjacent to navigable water bodies 1993 and 2009. We then tested all possible thresholds of HFP (from HFP > 0 to NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 HFP > 49) to define the change in the overlap between species ranges and high HFP 5. Fisher, D. O., Blomberg, S. P. & Owens, I. P. F. Extrinsic versus intrinsic values between 1993 and 2009. We measured the proportional difference, between factors in the decline and extinction of Australian marsupials. Proc. R. Soc. B 1993 and 2009, in the overlap of species ranges with HFP values smaller or equal 270, 1801–1808 (2003). than the defined threshold. We also measured the changes in the extension of high 6. Cardillo, M. et al. Multiple causes of high extinction risk in large mammal HFP values after discarding areas where HFP values were lower in 2009 than in species. Science 309, 1239–1241 (2005). 1993 (assuming ‘no change’ in those cases). We reported the effect size of the 7. Davidson, A. D., Hamilton, M. J., Boyer, A. G., Brown, J. H. & Ceballos, G. extent of high HFP values in low-risk vs. high-risk species, and the effect size of the Multiple ecological pathways to extinction in mammals. Proc. Natl. Acad. Sci. change in the extent of high HFP values in low-risk vs. high-risk species, using U. S. A. 106, 10702–10705 (2009). Cohen’sd statistic . 8. Di Marco, M., Collen, B., Rondinini, C. & Mace, G. Historical drivers of We used HFP change to predict species extinction risk transitions (low-risk vs extinction risk: using past evidence to direct future monitoring. Proc. R. Soc. B 3,7,8 high-risk), using a multi-variable Random Forest model . Following previous 282, 20150928 (2015). 8,39 works , we included important intrinsic and extrinsic predictors of risk (see 9. Rondinini, C., Di Marco, M., Visconti, P., Butchart, S. & Boitani, L. Update or Table 1 for a description). In identifying extrinsic variables, we favoured those outdate: long term viability of the IUCN Red List. Conserv. Lett. 2, 126–130 datasets with a good temporal match with the period over which extinction risk (2014). transitions were observed. We did not include species’ range size as a predictor, in 10. Bland, L. M. et al. Cost-effective assessment of extinction risk with limited order to prevent potential circularity in the estimation of extinction risk . We used information. J. Appl. Ecol. 52, 861–870 (2015). as predictors both the current extent of high HFP values within species’ ranges, and 11. Cardillo, M. & Meijaard, E. Are comparative studies of extinction risk useful the proportional change in the extent of high HFP values through time. For for conservation? Trends Ecol. Evol. 27, 167–171 (2012). example, we measured how much of the distribution range of the lion (Panthera 12. Di Marco, M. et al. A Retrospective evaluation of the global decline of leo) is currently in overlap with HFP values of x or above, and which proportion of carnivores and ungulates. Conserv. Biol. 28, 1109–1118 (2014). the lion’s range has undergone a change from low ( < x) to high ( ≥ x) values of 13. Hoffmann, M. et al. The impact of conservation on the status of the world’s HFP during 1993–2009. We measured the importance of HFP as a predictor vertebrates. Science 330, 1503–1509 (2010). relative to other variables, using two standard metrics for Random Forest models . 14. Venter, O. et al. Global terrestrial human footprint maps for 1993 and 2009. The first metric is the decrease in classification accuracy, reporting the decrease in Sci. Data 3, 273–281 (2016). the model’s ability to correctly classify data if the values of a predictor variable are randomly permuted. The second metric is the decrease in Gini coefficient, 15. Watson, J. E. M. et al. Catastrophic declines in wilderness areas undermine global environment targets. Curr. Biol. 26, Volume 26, 2929–2934 reporting the total decrease in Gini coefficient from splitting the data based on a predictor variable, averaged over all classification trees in the Random Forest. We (2016). also report the overall performance of our Random Forest model during cross- 16. Venter, O. et al. Sixteen years of change in the global terrestrial human validation, in terms of: proportion of correctly classified species, proportion of footprint and implications for biodiversity conservation. Nat. Commun. 7, correctly classified high-risk species (sensitivity), proportion of correctly classified 1–11 (2016). low-risk species (specificity), and true skill statistic (TSS = sensitivity + 17. Verde Arregoitia, L. D. Biases gaps, and opportunities in mammalian specificity −1). extinction risk research. Mamm. Rev. 46,17–29 (2016). We tested the use of a Random Forest model to predict uplisted vs not uplisted 18. Watson, J. E. M. et al. Persistent disparities between recent rates of habitat species, but found completely biased results with almost all species being classified conversion and protection and implications for FUture Global Conservation as not uplisted. This implies the model is unable to correctly classify uplisted Targets. Conserv. Lett. 9, 413–421 (2016). species, due to the very large imbalance between number of uplisted species (4% of 19. Breiman, L. Random Forests. (University of California, Berkeley, USA, 2001). the total) and number of not uplisted species (96% of the total). We thus only 20. Santini, L., González-Suárez, M., Rondinini, C. & Di Marco, M. Shifting report the main results on the low-risk vs high-risk model. baseline in macroecology? Unravelling the influence of human impact on mammalian body mass. Divers. Distrib. 23, 640–649 (2017). 21. Cardillo, M. et al. The predictability of extinction: biological and external Measuring human footprint impact across biogeographic realms. We repre- sented the biogeographical variation in HFP change, by measuring the change in correlates of decline in mammals. Proc. R. Soc. B 275, 1441–1448 (2008). overlap of species ranges with high HFP values (again testing all HFP thresholds) 22. Di Marco, M., Rondinini, C., Boitani, L. & Murray, K. A. Comparing multiple within separate biogeographical domains. We run separate analyses for separate species distribution proxies and different quantifications of the human biogeographic realms, and for separate biomes within each realm (i.e. biome- footprint map, implications for conservation. Biol. Conserv. 165, 203–211 realms), following the biogeographic classification of the world proposed by Olson (2013). et al. . In this case we only retained species with > 50% of their distributions 23. Jones, K. R. et al. One-third of global protected land is under intense human within a realm, or a biome-realm, and we discarded all biome-realms which did not pressure. Science 360, 788–791 (2018). have at least five low-risk and five high-risk species. We also run separate Random 24. Betts, M. G. et al. Global forest loss disproportionately erodes biodiversity in Forest models for species restricted to separate biogeographic realms, following the intact landscapes. Nature 547, 441–444 (2017). same settings as in the global model (see previous section). 25. Giglio, L., Randerson, J. T. & Van Der Werf, G. R. Analysis of daily, monthly, All spatial analyses were performed in GRASS GIS , statistical analyses were and annual burned area using the fourth-generation global fire emissions 43 44 45 performed in R , using the packages ‘effsize’ and ‘randomForests’ . database (GFED4). J. Geophys. Res. Biogeosciences 118, 317–328 (2013). 26. Sodhi, N. S., Koh, L. P., Brook, B. W. & Ng, P. K. L. Southeast Asian biodiversity: an impending disaster. Trends Ecol. Evol. 19, 654–660 (2004). Data availability 27. Tracewski, Ł. et al. Toward quantification of the impact of 21st-century The Human Footprint dataset used in this study is available from the Dryad Digital deforestation on the extinction risk of terrestrial vertebrates. Conserv. Biol. 30, Repository with the identifier doi:10.5061/dryad.052q5 . The other datasets that 1070–1079 (2016). support the findings of this study derive from published sources, cited in the 28. Bland, L. M., Collen, B., Orme, C. D. L. & Bielby, J. Predicting the Methods section and listed in Table 1. conservation status of data-deficient species. Conserv. Biol. 29, 250–259 (2015). Received: 10 May 2018 Accepted: 9 October 2018 29. Di Marco, M. et al. A novel approach for global mammal extinction risk reduction. Conserv. Lett. 5, 134–141 (2012). 30. Pacifici, M. et al. Generation length for mammals. Nat. Conserv 5,89–94 (2013). 31. Riahi, K. et al. The shared socioeconomic pathways and their energy, land use, and greenhouse gas emissions implications: an overview. Glob. Environ. Chang 42, 153–168 (2017). References 1. De Vos, J. M., Joppa, L. N., Gittleman, J. L., Stephens, P. R. & Pimm, S. L. 32. Visconti, P. et al. Projecting global biodiversity indicators under future development scenarios. Conserv. Lett. 9,5–13 (2016). Estimating the normal background rate of species extinction. Conserv. Biol. 29, 33. Jetz, W. & Freckleton, R. P. Towards a general framework for predicting threat 452–462 (2015). 2. Brook, B. W., Sodhi, N. S. & Bradshaw, C. J. A. Synergies among extinction status of data-deficient species from phylogenetic, spatial and environmental information. Philos. Trans. R. Soc. B 370, 20140016 (2015). drivers under global change. Trends Ecol. Evol. 23, 453–460 (2008). 34. Sanderson, E. W. et al. The human footprint and the last of the wild. 3. Murray, K. A., Verde Arregoitia, L. D., Davidson, A., Di Marco, M. & Di Bioscience 52, 891–904 (2002). Fonzo, M. M. I. Threat to the point: improving the value of comparative extinction risk analysis for conservation action. Glob. Chang. Biol. 20, 483–494 35. IUCN. IUCN Red List of threatened species version 2012.1. Version 20104 2008, (2012). (2014). 36. IUCN. IUCN Red list categories and criteria, version 3.1. (IUCN Gland, 4. Purvis, A., Gittleman, J. L., Cowlishaw, G. & Mace, G. M. Predicting extinction risk in declining species. Proc. R. Soc. Lond. B 267, 1947–1952 (2000). Switzerland and Cambridge, UK, 2001). 8 NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 ARTICLE 37. Di Marco, M. et al. Drivers of extinction risk in African mammals: the 54. NASA Earth Observatory Group (2018) Normalized difference vegetation interplay of distribution state, human pressure, conservation response index. Retrieved from http://neo.sci.gsfc.nasa.gov/view.php? and species biology. Philos. Trans. R. Soc. Lond. B Biol. Sci. 369, 20130198 datasetId=MOD_NDVI_M. (2014). 55. Hansen, M. C. et al. High-resolution global maps of 21st-century forest cover 38. Cohen, J. Statistical Power Analysis for the Behavioral Sciences (2nd ed.). change. Science 342, 850–853 (2013). (Academic Press, New York, 1988). 39. Di Marco, M. & Santini, L. Human pressures predict species’ geographic range Acknowledgements size better than biological traits. Glob. Chang. Biol. 21, 2169–2178 (2015). M.D.M. acknowledges the support of the ARC Centre of Excellence for Environmental 40. Allouche, O., Tsoar, A. & Kadmon, R. Assessing the accuracy of species Decision. distribution models: prevalence, kappa and the true skill statistic (TSS). J. Appl. Ecol. 43(6), 1223–1232 (2006). 41. Olson, D. M. et al. Terrestrial ecoregions of the world: a new map of life on Author contributions earth. Bioscience 51, 933–938 (2001). M.D.M. and J.E.M.W. conceived the study; M.D.M. performed the analyses; M.D.M., 42. GRASS Development Team. Geographic Resources Analysis Support System O.V., H.P.P., J.E.M.W. discussed the results; M.D.M., O.V., H.P.P., J.E.M.W. wrote the (GRASS) Software, Version 7.0. (2016). manuscript. 43. R. Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing. (2015). 44. Torchiano, M. effsize: Efficient Effect Size Computation. R package version Additional information 0.7.1. (2017). Available at: https://cran.r-project.org/package=effsize. Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467- 45. Liaw, A. & Wiener, M. The randomforest package. R. News 2,18–22 (2002). 018-07049-5. 46. CIESIN & CIAT. Gridded Population of the World, Version 3 (GPWv3): Competing interests: The authors declare no competing interests. Population Density Grid. Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC) (2005). Available at: http://sedac.ciesin. columbia.edu/data/set/gpw-v3-population-density. (Accessed: 1st November Reprints and permission information is available online at http://npg.nature.com/ 2013) reprintsandpermissions/ 47. CIESIN, FAO & CIAT. Gridded Population of the World, Version 3 (GPWv3): Population Count Grid, Future Estimates. Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC) (2005). Available at: Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in http://sedac.ciesin.columbia.edu/data/set/gpw-v3-population-count-future- published maps and institutional affiliations. estimates. (Accessed: 1st November 2013) 48. Nelson, A. Travel time to major cities: a global map of Accessibility. Global Environment Monitoring Unit - Joint Research Centre of the European Commission, Ispra Italy. (2008). Available at: http://bioval.jrc.ec.europa.eu/ Open Access This article is licensed under a Creative Commons products/gam/index.htm. Attribution 4.0 International License, which permits use, sharing, 49. Jones, K. E. et al. PanTHERIA: a species-level database of life history, ecology, adaptation, distribution and reproduction in any medium or format, as long as you give and geography of extant and recently extinct mammals. Ecology 90, 2648 appropriate credit to the original author(s) and the source, provide a link to the Creative (2009). Commons license, and indicate if changes were made. The images or other third party 50. Tacutu, R., Craig, T. & Budovsky, A. Human ageing genomic resources: material in this article are included in the article’s Creative Commons license, unless integrated databases and tools for the biology and genetics of ageing. Nucleic indicated otherwise in a credit line to the material. If material is not included in the Acid. Res 41, 1027–1033 (2013). article’s Creative Commons license and your intended use is not permitted by statutory 51. Verde Arregoitia, L., Blomberg, S. & Fisher, D. Phylogenetic correlates of regulation or exceeds the permitted use, you will need to obtain permission directly from extinction risk in mammals: species in older lineages are not at greater risk. the copyright holder. To view a copy of this license, visit http://creativecommons.org/ Proc. R. Soc. B 280, 20131092 (2013). licenses/by/4.0/. 52. Wilman, H. et al. EltonTraits 1. 0: species-level foraging attributes of the world’s birds and mammals. Ecology 95, 2027 (2014). 53. Rondinini, C. et al. Global habitat suitability models of terrestrial mammals. © The Author(s) 2018 Philos. Trans. R. Soc. Lond. B. Biol. Sci. 366, 2633–2641 (2011). NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 9 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nature Communications Springer Journals

Changes in human footprint drive changes in species extinction risk

Loading next page...
 
/lp/springer-journals/changes-in-human-footprint-drive-changes-in-species-extinction-risk-b4Pmf3tMuh

References (54)

Publisher
Springer Journals
Copyright
Copyright © 2018 by The Author(s)
Subject
Science, Humanities and Social Sciences, multidisciplinary; Science, Humanities and Social Sciences, multidisciplinary; Science, multidisciplinary
eISSN
2041-1723
DOI
10.1038/s41467-018-07049-5
Publisher site
See Article on Publisher Site

Abstract

ARTICLE DOI: 10.1038/s41467-018-07049-5 OPEN Changes in human footprint drive changes in species extinction risk 1,2 3 1,4 1,5 Moreno Di Marco , Oscar Venter , Hugh P. Possingham & James E.M. Watson Predicting how species respond to human pressure is essential to anticipate their decline and identify appropriate conservation strategies. Both human pressure and extinction risk change over time, but their inter-relationship is rarely considered in extinction risk modelling. Here we measure the relationship between the change in terrestrial human footprint (HFP)— representing cumulative human pressure on the environment—and the change in extinction risk of the world’s terrestrial mammals. We find the values of HFP across space, and its change over time, are significantly correlated to trends in species extinction risk, with higher predictive importance than environmental or life-history variables. The anthropogenic con- version of areas with low pressure values (HFP < 3 out of 50) is the most significant predictor of change in extinction risk, but there are biogeographical variations. Our framework, cali- brated on past extinction risk trends, can be used to predict the impact of increasing human pressure on biodiversity. 1 2 Centre for Biodiversity and Conservation Science, The University of Queensland, 4072 Brisbane, QLD, Australia. CSIRO Land & Water, EcoSciences Precinct, 41 Boggo Road, Dutton Park, QLD 4102, Australia. Natural Resource and Environmental Studies Institute, University of Northern British Columbia, 3333 University Way, Prince George V2N 4Z9, Canada. The Nature Conservancy, 4245 North Fairfax Drive, Suite 100, Arlington, VA 22203-1606, USA. Global Conservation Program, Wildlife Conservation Society, 2300 Southern Boulevard, Bronx, NK 10460, USA. Correspondence and requests for materials should be addressed to M.D.M. (email: moreno.dimarco@gmail.com) NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 1 1234567890():,; ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 pecies are disappearing at rates that are 1000 times faster Here we compare a 16-year trend in HFP (1993–2009) with a than those registered in the fossil record , and accurate 12-year trend in the extinction risk of 4421 terrestrial mammal Spredictions of extinction risk are necessary to anticipate species (1996–2008). Our goal is to test the existence of a direct declines under past, current, and projected levels of human pres- relationship between changing human pressure, as represented by sure. Understanding the relationship between changes in human the HFP, and changing risk status of species over a comparable pressures and the decline of individual species is necessary for time frame. This allows for the dynamic, as opposed to static, identifying those species at highest risk, and for prioritising the modelling of species extinction risk , and takes advantage of a 2,3 actions and policies required to combat their decline .Com- single, cumulative, representation of how human pressure has parative extinction risk modelling, which builds on the relationship changed over time . We focus on terrestrial mammals as they between species threat status, their life histories, and the pressure have had their extinction risk measured over a similar period as mapped within their ranges, is increasingly used to predict the risk HFP , and they have served as a focal group in several previous 4–8 17 of extinction . This approach allows inferring the extinction risk extinction risk analyses . We classified species into two groups, of a large number of species based on readily available data, and following earlier work : ‘low-risk’ transitions and ‘high-risk’ predictions can be updated more often than expert-based assess- transitions (Fig. 1). The low-risk group included species that 9,10 ments, given the substantially lower resources requirement . retained a category of least concern and species that moved from However, a major limitation in these analyses is the absence of a any higher category of threat to a lower category during the study link to spatial and temporal changes in human pressure and how period. The high-risk group included all species that retained a these lead to change in the risk of species declines .This isfurther category of threatened or near threatened, together with species complicated by two types of change in human pressure, the change that moved from any lower category of threat to a higher cate- in extent of pressures (e.g. road building in a new area), and the gory. We also test a more conservative classification of risk intensification of existing pressures (e.g. increase in deforestation change, where species are considered either ‘uplisted’, if they rates). The missing linkage between pressure and extinction risk moved from any Red List category to a higher category during the means comparative extinction risk analysis has struggled to inform study period, or ‘not uplisted’. We measured the proportion of policy and management . each species’ range overlap with high HFP values, and how this As a species’ conservation status is sensitive to changes in overlap has changed through time, testing all possible definitions 12,13 human pressure , more dynamic extinction risk modelling has of what constitutes ‘high HFP’. We used these values, and other the potential to elucidate links between trends in pressures and known human pressure, environmental, and life-history pre- trends in extinction risk. The recent publication of a temporally dictors of risk (Table 1), to provide estimate of the extinction risk inter-comparable map of human footprint (HFP) presents an transitions of species as a function of change in human pressure important advance in the global representation of changing within their distributions. human pressure on the terrestrial environment. The map, which Our results show the importance of HFP as a predictor of incorporates eight pressure layers standardised into a cumulative extinction risk transition in terrestrial mammals, and suggest the index (see Methods for details), is calculated at two time points conversion of natural and semi-natural areas (those with low HFP and provides an opportunity to investigate the relationship values) has the strongest association with high-risk transitions in between changes in human pressure and changes in the status of species conservation status. We also identified biogeographical biodiversity. HFP provides a spatially explicit index of cumulative differences in the best HFP threshold to determine areas of ‘high human pressure ranging from 0 to 50, where a value of zero pressure’, which can be used for regional monitoring of extinction corresponds to ‘wilderness areas’ free from any significant human risk change. influence , a value of four corresponds to low pressure levels (e.g. pasture lands), and values above 20 typically represents very high pressure levels (e.g. densely populated semi-urban and urban Results areas) . Yet, the HFP is not necessarily a direct measure of threat Global change in human pressure and species extinction risk. to species, and it would be inappropriate to assume that all Much of Earth’s terrestrial surface (30.8%) has undergone an species respond to human activities in the same way. Conse- increase in human pressure, as indicated by HFP values that have quently, the relationship between HFP and species extinction risk increased since 1993 (Fig. 2a; Supplementary Fig. 1a). Two thirds requires testing, in the context of environmental and life-history of those areas already had relatively high HFP values in 1993 characteristics of each species. ( ≥ 4) which became even higher by 2009. At the other end of the a b Low-risk transitions High-risk transitions CR CR CR CR EN EN EN EN VU VU VU VU NT NT NT NT LC LC LC LC Past Present Past Present Fig. 1 Classification of species extinction risk transitions, based on past and present IUCN Red List categories*. Low-risk transitions include those species that were of least concern throughout the study period, together with species that moved from any higher category of threat to a lower one. High-risk transitions include all species that were originally threatened or near threatened and retained their category throughout the study period, together with species that moved from any lower category of threat to a higher category. *Acronyms for the Red List categories: Least Concern (LC); Near Threatened (NT); Vulnerable (VU); Endangered (EN); Critically Endangered (CR) 2 NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 ARTICLE Table 1 Description of the variables used to predict extinction risk transitions in the random forest model Class Variable Description and source Source Pressure High HFP extent Proportion of species range overlapping with high human footprint values in 2009. Pressure High HFP change Difference in the proportional overlap between species range and high HFP during 1993–2009. Pressure Human population Density of human population in year 2000. density Pressure Human population Human population growth, proportional increase in human population between 1990 and 2000. growth Pressure Travel time to cities Accessibility from major cities, measured as travel time. Life history Taxonomic order Species taxonomic orders. 49,50 Life history Gestation length Gestation length, a proxy of species reproductive output. Life history Weaning age Weaning age, a proxy of species reproductive timing. as above 49–51 Life history Body mass A generic proxy of species life history and energetic requirements Life history Diet Dietary category: carnivore ( > 90% animal matter ingested), omnivore (10–90%), herbivore ( < 10%). 35,53 Life history Habitat class Species preferences of macro-habitat categories: aquatic, artificial, caves, desert, forest, grassland, rocky areas, savanna, shrubland, generalists (two or more of the previous categories). Environment NDVI Normalized difference vegetation index, proxy of primary productivity, registered from year 2013. Environment Tree cover Percentage tree cover values registered in year 2000. Environment Habitat prevalence Proportion of suitable habitat within species range. Variables are aggregated into three main classes (human pressure, life-history, environmental characteristics) a b HFP 1993 RL 1996 Min EX CR EN VU NT LC 0 1020304050 LC NT VU EN CR EX Max Fig. 2 Recent changes in terrestrial human footprint and species extinction risk. a Shows a transition matrix in which any position represents the initial (x axis) and final (y axis) human footprint value (from 0 to 50) of global 1 km terrestrial grid cells; the colour scheme represents the number of individual cells in each particular transition state. b Shows a transition matrix in which any position represents the initial (x axis) and final (y axis) extinction risk category (from Least Concern to Extinct) of terrestrial mammal species; the colour scheme represent the number of individual species in each particular transition state spectrum, most of the areas that did not face an increase in Measuring human pressure within species ranges. The cumu- human pressure (41.5% of the total terrestrial surface) are char- lative distribution of HFP values within species geographic ranges acterised by a relatively low HFP value ( < 4). Half of these low- followed similar patterns across the two species groups (high-risk HFP areas have been identified as the last remaining terrestrial and low-risk) and across years (Fig. 3). However, high-risk spe- ‘wilderness’, which is free of any significant human disturbance cies had on average a larger proportion of their range overlapping (HFP = 0). with high HFP values, compared to low-risk species. The level of When looking at the transitions in species extinction risk, we overlap with those areas classified as wilderness was comparable found that 69% of species faced a low-risk transition, while 31% between high-risk and low-risk species, while the biggest differ- faced a high-risk transition (Fig. 2b; Supplementary Fig. 1b). This ences among the two groups was observed for HFP values in the is largely due to 1,229 (27.8%) threatened and near-threatened range 3–15, which correspond to moderate or high levels of species retaining their Red List category, and in minor part to 159 human pressure . This was reflected in significantly higher mean uplisted species (3.6%) that moved towards higher Red List HFP values within the range of high-risk species compared to −12 −16 categories. Only 22 (0.5%) species moved towards lower Red List low-risk species (p-value = 2*10 in 1993 and 2* 10 in 2009; categories during the study period. Wilcoxon signed rank test, one-sided; Supplementary Fig. 2). NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 3 HFP 2009 RL 2010 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 Low-risk High-risk Low-risk 1993 Low-risk 2009 High-risk 1993 High-risk 2009 0 1020304050 0 1020304050 HFP HFP threshold Fig. 3 Cumulative extent of human footprint values within mammal species ranges. The lines represent the cumulative percentage of species range that Fig. 4 Changes in the overlap between species ranges and high human overlaps with increasing values of human footprint, averaged among all footprint values over time. The lines report the average change in the species. Different lines refer to species in the low-risk or high-risk overlap between species ranges and human footprint values bigger than categories, for the period 1993 or 2009, as reported in legend any given threshold. Different lines refer to species in the low-risk or high- risk categories, as reported in legend. The shaded areas around the lines High-risk species typically faced a larger change in the extent of represent the 95% standard credible interval measured across a total of high HFP values compared to low-risk species, with the exceptions 4421 species of very low HFP thresholds (Fig. 4). When looking at HFP thresholds between 0 and 2, low-risk species had a larger modified landscapes. We found that pressure variables had higher proportion of their range moving from below- to above- predictive performance compared to life-history and environ- threshold values compared to high-risk species. This might be mental variables, highlighting the magnitude of human influence related to many threatened and declining species retaining little on environmental trends , and the two HFP variables were the natural areas within their range at the beginning of the study most important predictors in the model (Fig. 5). The model period, with consequent little chances of observing an increasing showed good overall classification ability during cross-validation HFP in natural areas during the study period. For HFP in the (species correctly classified = 82.9%), albeit the accuracy in pre- range 3–49, high-risk species consistently showed higher propor- dicting high-risk species (sensitivity = 60.4%) was lower than the tions of their range moving to above-threshold values, with a accuracy in predicting low-risk species (specificity = 92.4%; True difference that was significant for thresholds in the range 6–44 Skill Statistics = 0.53). −6 −2 (p-values= 2*10 –2*10 ; Wilcoxon signed rank test, one- sided). Overall, the largest effect size for the difference in extent The biogeography of human pressure within species ranges. of high HFP values between low-risk vs high-risk species was We observed some differences among realms and biomes in observed for a HFP threshold of 3 (Cohen’sd = 0.43) and terms of HFP change patterns (Fig. 6). For example, the HFP decreased afterwards, while the effect size for change in the extent value at which high-risk species had the largest proportions of of high HFP values increased up to a threshold of 6 and then their ranges moving above threshold was < 6 in the Nearctic, stabilized (with Cohen’s d values in the range 0.20–0.22; Neotropical and Afrotropical realms, and > 6 in the Palearctic, Supplementary Fig. 3 and 4). Indomalay and Australasian realms. Most realms showed general When looking at the difference in extent of high HFP values for consistency with the global analysis in that high-risk species had a uplisted vs. not uplisted species, we found even larger differences higher proportion of their range moving toward higher HFP than those described for low-risk vs. high-risk species (Supple- values compared to low-risk species, especially when looking at mentary Fig. 5), with substantially higher values for uplisted intermediate and high HFP thresholds. However there were species when looking at HFP thresholds between 2 and 20. exceptions in the Afrotropical and Indomalay realms. In the Afrotropical realm, the exception emerged for grassland biomes Modelling transitions in species extinction risk. We measured (Supplementary Fig. 7), where low-risk species showed larger the performance of HFP in predicting low- vs. high-risk transi- conversions to high HFP values than high-risk species, when tions in species extinction risk, using a random forest model for considering thresholds in the range 7–12. In the Indomalay classification . In this analysis, we compared the predictive realm, low-risk species had similar (or even higher) proportions performance of HFP with a number of other pressure, life-history, of their ranges moving toward higher HFP values compared to and environmental variables (Table 1). We measured HFP both high-risk species. This was especially the case for species living in as the current extent of high HFP values within species ranges, the moist forest biome, which contrasted with the results obtained and as the change in high HFP values over the time period for the same biome in other realms. When looking at dry forest (1993–2009). We adopted all possible thresholds to determine species in the Indomalay, we found low-risk species had faced low vs. high HFP values (from HFP > 0 to HFP > 49), and found larger increase in the extent of HFP values in the range 9–18, that the importance of HFP variables as predictors decreased with while high-risk species have faced a higher change for HFP values increasing thresholds (Supplementary Fig. 6). Overall, a HFP above 18. threshold of ≥ 3 resulted in the highest prediction performance Despite the differences in HFP change patterns observed across across the two HFP variables (current extent and change over biogeographical domains, we still found that relatively low time), indicating that human pressure intensification in intact and thresholds resulted in the highest predictive performance of near-intact areas is globally more relevant, for explaining “HFP change” as a variable in random forest models developed extinction risk transitions, than intensification within already- for separate realms (Supplementary Fig. 8). The lowest threshold 4 NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications Cumulative range % % change in high HFP extent NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 ARTICLE High HFP extent High HFP change Travel time to cities Order Realm Human pop. dens. Tree cover Body mass NDVI Human pop. growth Weaning age Gestation length Habitat class Habitat prevalence 0 20 40 60 80 100 Mean decrease accuracy High HFP extent High HFP change Human pop. dens. Body mass Weaning age Travel time to cities Human pop. growth Tree cover Habitat prevalence NDVI Gestation length Pressure Order Life-history Realm Environment Habitat class 0 50 100 150 200 Mean decrease gini Fig. 5 Predictive importance of variables for the prediction of extinction risk transitions in terrestrial mammals. Variables are colour-coded accordingto their broad class (human pressure, life-history, environmental characteristics). Different plots refer to different measures of variable importance: a variable effect on the overall decrease in prediction accuracy, and b contribution of the variable to decrease Gini Index during the classification routine. A description of all the variables can be found in Table 1. In this analysis, “high HFP" included values of 3 or above (HFP ≥ 1) was observed in the Nearctic realm, while the highest Identifying a threshold of human pressures beyond which threshold was observed in the IndoMalay realm (HFP ≥ 5). species show negative response is essential for monitoring land conversion rates in the context of international biodiversity tar- 15,23 gets . Yet the definition of “high pressure” levels has remained elusive in global analyses so far. Our results show that averting Discussion In order to proactively inform monitoring and management, it is the conversion of natural and semi-natural areas, those with HFP necessary to know the conditions under which a species is likely to values ≥ 3, is the most effective strategy to prevent species from retain an unsustainable high risk of extinction, or to face increased undergoing a high-risk transition in their conservation status risk over time . We found that the extent of high human pressure when accounting for environmental and life-history traits. These (as defined by the HFP index) within species ranges, and the results are in line with recent findings that deforestation within change in this extent over time, were strong correlates of extinc- intact landscapes is the strongest correlate of decline in forest tion risk transitions. These two variables (state and change of high species , opening up the path to a number of direct threat HFP values) were found to be the strongest predictors of risk mechanisms (such as hunting, diseases spread, and invasive when compared to an array of other variables, including species’ species). However, protecting natural and semi-natural land- traits, environmental conditions, and individual pressure layers. scapes is not sufficient to improve the status of species which are already at a high risk, some of which have little natural habitat left This result contrasts with the findings from previous extinction risk modelling for mammals, where the importance of human within their distributions and will require habitat restoration to reduce their risk of extinction. In fact, high-risk and low-risk pressure as predictors was found to be lower than environmental 3,17,21 or life-history variables . This shows that temporal cumula- species showed markedly different changes in their overlap with tive pressure mapping is a powerful tool for improving extinction intermediate HFP values with high-risk species facing con- risk modelling and forecasting, coupling changes in human sistently larger increase in HFP levels. This was confirmed, with pressure with changes in biodiversity state. an even stronger pattern, when looking at uplisted vs not uplisted NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 PA NA 4 IM 2 10 0 5 10 15 20 25 30 0 5 10 15 20 25 30 HFP threshold HFP threshold 0 5 10 15 20 25 30 HFP threshold HFP threshold 12 AA NT AT 0 5 10 15 20 25 30 0 5 1015202530 HFP threshold 0 5 10 15 20 25 30 HFP threshold HFP threshold >20 Fig. 6 Changes in the overlap between species ranges and high human footprint (HFP) values for each biogeographic realm*. The underlying map reports, for each biome within each realm, the threshold at which change in HFP values is the highest for high-risk species. The plots report the average change in the overlap between species ranges and HFP values bigger than a given threshold within each realm (with high-risk species in red and low-risk species in blue). The shaded areas around the lines in the plots represent the 95% standard credible interval measured across a total of: 493 AA species, 854 AT species, 604 IM species, 259 NA species, 949 NT species, 442 PA species*. *Realm acronyms: AA Australasia, AT Afrotropical, IM Indomalay, NA Nearctic, NT Neotropical, PA Palearctic species, albeit this latter classification could not be used for biogeographic realms and in combination with other variables, we extinction risk modelling due to very high imbalance in species found that low thresholds (in the range 1–5, depending on the numbers between the two classes. The identification of this HFP realm) still performed the best in separating low from high HFP threshold (i.e. 3), and what happens when changes occur in the values. This demonstrates that the conservation of intact areas, HFP around this threshold, provide simple guidelines for iden- and areas with little human modification, is relevant at the scale tifying tipping points beyond which human activities might lead of individual realms and not only globally. to species decline. Our results showed that high-risk species had faced larger We found some biogeographical differences in the way low-risk increases in pressure levels within areas of moderate HFP values, and high-risk species overlap with HFP values. Particular while low-risk species had faced larger pressure increases in areas exceptions were found in the grassland biomes of the Afrotropical of former low HFP (those < 2). This includes the loss of wild- realm and in the moist forest biome of the IndoMalay realm, erness areas, which was more likely to occur within the range of where low-risk species faced similar (or higher) increase in HFP low-risk species than high-risk species. This finding is probably values compared to their high-risk counterparts. These exceptions related to the fact threatened species are less likely to overlap with might indicate that species living in those environments are wilderness areas compared to non-threatened species, as a relatively resilient to human pressure as measured in the HFP reflection of pressure operating within their past ranges. This index, and respond more strongly to other pressures not incor- finding means that the continuous conversion of intact and near- porated in the index such as fire regimes, especially relevant in intact areas will likely result in species that are currently classified African grasslands , and overexploitation, relevant in Southeast as low-risk to become high-risk in the future. The loss of intact Asia . However there is also the possibility that some species lands within the ranges of low-risk species should thus act as an currently classified as low-risk might actually be facing a higher early indication of a trajectory of increasing species endanger- risk of extinction than previously thought, especially in forested ment, and points to the need of identifying, and securing, those biomes, as it seems to result from recent, rapid, deforestation . remaining intact ecosystems. These results support the call for a status re-assessment of these Our model was better able to correctly classify low-risk species species within these regions. Interestingly however, when HFP than high-risk species. This might be related the fact that some change was considered as a predictor of risk within separate high-risk species are responding to different components of 6 NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications % change in range extent % change in range extent % change in range extent % change in range extent % change in range extent % change in range extent NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 ARTICLE pressures, not well represented here, while low-risk species are were assigned a score of 4, which decayed exponentially out to 15 km away from the waters. After each pressure layer was standardised within the same values just not facing significant pressures. However this same pattern range, they were summed together to create a cumulative map of human pressure. was also found (at various degrees) in other extinction risk The results are two globally standardised HFP maps, with values ranging from 0 to 3,7,8,28 modelling exercises , indicating that the condition under 2 50 and a spatial resolution of 1 km , one for the year 1993 and one for 2009 (based which species are likely to undergo a low-risk transition are likely on pressure layers referred to the different periods). In this analysis we use the integer version of the HFP maps, to only represent integer changes in the index easier to identify compared to the conditions leading to high-risk (+ 1, + 2, + 3 etc.). transitions. When human pressure is operating, the risk for We measured the change in HFP values for each 1 km terrestrial grid cell species is determined by a complex combination of pressure between 1993 and 2009, and contrasted this change with species geographic levels, species’ sensitivity and their potential for adaption , which distributions to understand changes in perceived treat levels for the species. HFP change can result in an increase in pressure level, i.e. from lower to higher values, determine higher levels of uncertainty in the predictions. This or a decrease in pressure, i.e. the opposite. Here we only accounted for increases in adds a level of complexity in understanding the relationship HFP values, as decreases in pressure levels (e.g. abandonment of agricultural land) between pressure change and change in species extinction risk. are likely to take time before having a measurable effect on species threat status, The current availability of HFP maps (for the years 1993 and especially for species with a long generation time such as some of the high-risk 2009) allowed us to test the relationship between human pressure species in our dataset (average generation time for high-risk species is 7 years, compared to 4.5 years for low-risk species). We used HFP and its change over time and extinction risk over a similar time frame, that encompasses as a predictor of extinction risk change, in combination with previously identified three or more generations for 73% species in our analyses . This variables (Table 1). relationship between change in HFP and change in extinction risk can serve as a basis for future research. For example, our mod- Species extinction risk change. We represented the extinction risk of terrestrial elling framework, calibrated on observed trends, can allow pro- mammal species using the information available from the IUCN Red List , and jecting future extinction risk transitions under alternative the retrospective Red List Assessments published in Hoffmann et al. . We con- sidered the 2010 IUCN Red List categories of each species and the retrospectively scenarios of socio-economic development . This will require assigned categories for 1996. These latter categories were defined using the same generating future projections of the base pressure layers that methodology as the 2010 assessments, but referred to the past condition of the constitute the HFP map, as well as predicting the shift in species species. We considered the following IUCN Red List categories, assigned using a set distributions due to future climate and land-use change . Also, 36 of five quantitative criteria with associated sub-criteria and thresholds : Least the established relationship between HFP and extinction risk Concern (LC); Near Threatened (NT); Vulnerable (VU); Endangered (EN); Cri- tically Endangered (CR); Extinct in the Wild (EW); Extinct (EX). We excluded transitions can be used to estimate the risk faced by ‘data defi- species not evaluated in the Red List, those without a defined risk of extinction 28,33 cient’ species , but this would require resolving the taxonomic category (Data Deficient), and those already extinct at the beginning of the study and geographical uncertainty characterising species in this period. We retained 4421 species of terrestrial mammals with a defined extinction category. risk category for the years 1996 and 2010, corresponding to 83% of all species in the group. Efforts to integrate human pressure maps and extinction risk 9 We recorded the initial (1996) and final (2010) Red List category of each has the potential to change the way we assess species risk and species, and followed Di Marco et al. in classifying species into two main groups proactively inform conservation action in a way that minimises (Fig. 1): low-risk transitions and high-risk transitions. The low-risk group included the number of species that will face a high risk of decline. Con- species that were LC throughout the study period, together with species that moved from any higher Red List category to a lower category (i.e. category ‘downlisting’). servation organisations that have a mission to prevent the decline The high-risk group included all species that were originally threatened or near of species can use our approach to prioritise actions for mini- threatened and retained their category throughout the study period, together with mizing extinction risk. These include both species that are already species that moved from any lower Red List category to a higher category (i.e. threatened with extinction and species that are likely to become category ‘uplisting’). The method behind this classification has been statistically justified , and reflects the fact that remaining within the same Red List category so if current rates of intensification in the human footprint through time does not necessarily imply that a species is in a stable condition. For continue into the future. The HFP index presents a standardised instance, while a species that retains a LC category is not undergoing a significant representation of human pressure levels, combining different population decline (or loss of geographic range), a species retaining a threatened human activities that represent potential sources of impact for category implies substantial continued decline . species. While this index represents a comprehensive and easy We also tested the use of a more conservative approach for classifying extinction risk transitions, where species were separated into two groups: ‘uplisted’ tool for estimating change in species extinction risk and guide species, those that had a deterioration in their Red List category (eg from Least broad-scale conservation efforts, we acknowledge that it cannot Concern to Near Threatened), and ‘not uplisted’ species, those that retained the substitute local-scale assessments of the conservation needs of same category or improved it. This classification can be seen as a more conservative each species. Instead, knowing which species and which areas are approach for defining extinction risk transitions, because in this case the ‘high-risk’ group only includes transitions that are of sufficient magnitude to generate an most likely to face a high risk can guide the prioritisation of local- upward shift in Red List categories. This classification however generated a large scale assessments by conservation practitioners. imbalance between species groups, with only ~4% of species being included in the uplisted class. Methods Human footprint state and change. We used the recent release of the global HFP Human footprint as a driver of extinction risk change. Several methods are 14,16 map , to represents the cumulative human impact on the environment. This available to measure the level of overlap between a spatial pressure layer and a map is built from eight base layers: (i) the extent of built environments; (ii) crop species’ geographic range . These include both measures of central tendencies, e.g. land; (iii) pasture land; (iv) human population density; (v) night-time lights; (vi) the mean/median pressure level observed within the range, and measures of spatial railways; (vii) roads; and (viii) navigable waterways. Following the approach ori- extent, e.g. how much of the species range is covered with high pressure levels. ginally proposed by Sanderson and colleagues , each layer was placed in a Measuring the extent of high pressure levels within a species’ range has been shown 1–10 scale with a value weighted according to the relative intensity of human to be a more sensitive way to predict extinction risk than using mean pressure 14 22 pressure (see Venter et al. for full justification and validation): (i) all built levels and was often a preferred choice in comparative extinction risk model- 21,37 environments were assigned a score of 10 while non-built environment had a score ling . However, identifying the best way for separating low and high pressure of zero); (ii) areas mapped as croplands were assigned a score of 7; (iii) areas levels requires testing multiple thresholds. mapped as pasture lands were assigned a score of 4; (iv) areas with a high human We measured the cumulative overlap between 1993 and 2009 HFP values population density of > 1,000 people/km received a score of 10, while areas with within species ranges, generating curves to represent how much of a species’ range lower density received a lower log-scaled score; (v) areas were divided into 10 overlaps with increasing values of HFP (from 0 to 50). We used the same species quantiles of increased night-time light intensity associated to score of 1 to 10, while distribution range maps for these measures, since past range maps for the areas with no lights were assigned a zero; (vi) railways and their immediate 500 m ~4500 species included in our analyses were not available. Given our study period buffers were given a score of 8, with a value of zero elsewhere (i.e. assuming no was reasonably restricted (16 years), we assumed change in the extent of species indirect impact); (vii) roads and their immediate 500 m buffers were given a score geographic range was overall limited. We generated separate curves to represent of 8 (direct impact), while nearby areas up to 15 km had score that decayed the average accumulation of HFP values in low-risk and high risk species, both in exponential to zero (indirect impact); (viii) areas adjacent to navigable water bodies 1993 and 2009. We then tested all possible thresholds of HFP (from HFP > 0 to NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 HFP > 49) to define the change in the overlap between species ranges and high HFP 5. Fisher, D. O., Blomberg, S. P. & Owens, I. P. F. Extrinsic versus intrinsic values between 1993 and 2009. We measured the proportional difference, between factors in the decline and extinction of Australian marsupials. Proc. R. Soc. B 1993 and 2009, in the overlap of species ranges with HFP values smaller or equal 270, 1801–1808 (2003). than the defined threshold. We also measured the changes in the extension of high 6. Cardillo, M. et al. Multiple causes of high extinction risk in large mammal HFP values after discarding areas where HFP values were lower in 2009 than in species. Science 309, 1239–1241 (2005). 1993 (assuming ‘no change’ in those cases). We reported the effect size of the 7. Davidson, A. D., Hamilton, M. J., Boyer, A. G., Brown, J. H. & Ceballos, G. extent of high HFP values in low-risk vs. high-risk species, and the effect size of the Multiple ecological pathways to extinction in mammals. Proc. Natl. Acad. Sci. change in the extent of high HFP values in low-risk vs. high-risk species, using U. S. A. 106, 10702–10705 (2009). Cohen’sd statistic . 8. Di Marco, M., Collen, B., Rondinini, C. & Mace, G. Historical drivers of We used HFP change to predict species extinction risk transitions (low-risk vs extinction risk: using past evidence to direct future monitoring. Proc. R. Soc. B 3,7,8 high-risk), using a multi-variable Random Forest model . Following previous 282, 20150928 (2015). 8,39 works , we included important intrinsic and extrinsic predictors of risk (see 9. Rondinini, C., Di Marco, M., Visconti, P., Butchart, S. & Boitani, L. Update or Table 1 for a description). In identifying extrinsic variables, we favoured those outdate: long term viability of the IUCN Red List. Conserv. Lett. 2, 126–130 datasets with a good temporal match with the period over which extinction risk (2014). transitions were observed. We did not include species’ range size as a predictor, in 10. Bland, L. M. et al. Cost-effective assessment of extinction risk with limited order to prevent potential circularity in the estimation of extinction risk . We used information. J. Appl. Ecol. 52, 861–870 (2015). as predictors both the current extent of high HFP values within species’ ranges, and 11. Cardillo, M. & Meijaard, E. Are comparative studies of extinction risk useful the proportional change in the extent of high HFP values through time. For for conservation? Trends Ecol. Evol. 27, 167–171 (2012). example, we measured how much of the distribution range of the lion (Panthera 12. Di Marco, M. et al. A Retrospective evaluation of the global decline of leo) is currently in overlap with HFP values of x or above, and which proportion of carnivores and ungulates. Conserv. Biol. 28, 1109–1118 (2014). the lion’s range has undergone a change from low ( < x) to high ( ≥ x) values of 13. Hoffmann, M. et al. The impact of conservation on the status of the world’s HFP during 1993–2009. We measured the importance of HFP as a predictor vertebrates. Science 330, 1503–1509 (2010). relative to other variables, using two standard metrics for Random Forest models . 14. Venter, O. et al. Global terrestrial human footprint maps for 1993 and 2009. The first metric is the decrease in classification accuracy, reporting the decrease in Sci. Data 3, 273–281 (2016). the model’s ability to correctly classify data if the values of a predictor variable are randomly permuted. The second metric is the decrease in Gini coefficient, 15. Watson, J. E. M. et al. Catastrophic declines in wilderness areas undermine global environment targets. Curr. Biol. 26, Volume 26, 2929–2934 reporting the total decrease in Gini coefficient from splitting the data based on a predictor variable, averaged over all classification trees in the Random Forest. We (2016). also report the overall performance of our Random Forest model during cross- 16. Venter, O. et al. Sixteen years of change in the global terrestrial human validation, in terms of: proportion of correctly classified species, proportion of footprint and implications for biodiversity conservation. Nat. Commun. 7, correctly classified high-risk species (sensitivity), proportion of correctly classified 1–11 (2016). low-risk species (specificity), and true skill statistic (TSS = sensitivity + 17. Verde Arregoitia, L. D. Biases gaps, and opportunities in mammalian specificity −1). extinction risk research. Mamm. Rev. 46,17–29 (2016). We tested the use of a Random Forest model to predict uplisted vs not uplisted 18. Watson, J. E. M. et al. Persistent disparities between recent rates of habitat species, but found completely biased results with almost all species being classified conversion and protection and implications for FUture Global Conservation as not uplisted. This implies the model is unable to correctly classify uplisted Targets. Conserv. Lett. 9, 413–421 (2016). species, due to the very large imbalance between number of uplisted species (4% of 19. Breiman, L. Random Forests. (University of California, Berkeley, USA, 2001). the total) and number of not uplisted species (96% of the total). We thus only 20. Santini, L., González-Suárez, M., Rondinini, C. & Di Marco, M. Shifting report the main results on the low-risk vs high-risk model. baseline in macroecology? Unravelling the influence of human impact on mammalian body mass. Divers. Distrib. 23, 640–649 (2017). 21. Cardillo, M. et al. The predictability of extinction: biological and external Measuring human footprint impact across biogeographic realms. We repre- sented the biogeographical variation in HFP change, by measuring the change in correlates of decline in mammals. Proc. R. Soc. B 275, 1441–1448 (2008). overlap of species ranges with high HFP values (again testing all HFP thresholds) 22. Di Marco, M., Rondinini, C., Boitani, L. & Murray, K. A. Comparing multiple within separate biogeographical domains. We run separate analyses for separate species distribution proxies and different quantifications of the human biogeographic realms, and for separate biomes within each realm (i.e. biome- footprint map, implications for conservation. Biol. Conserv. 165, 203–211 realms), following the biogeographic classification of the world proposed by Olson (2013). et al. . In this case we only retained species with > 50% of their distributions 23. Jones, K. R. et al. One-third of global protected land is under intense human within a realm, or a biome-realm, and we discarded all biome-realms which did not pressure. Science 360, 788–791 (2018). have at least five low-risk and five high-risk species. We also run separate Random 24. Betts, M. G. et al. Global forest loss disproportionately erodes biodiversity in Forest models for species restricted to separate biogeographic realms, following the intact landscapes. Nature 547, 441–444 (2017). same settings as in the global model (see previous section). 25. Giglio, L., Randerson, J. T. & Van Der Werf, G. R. Analysis of daily, monthly, All spatial analyses were performed in GRASS GIS , statistical analyses were and annual burned area using the fourth-generation global fire emissions 43 44 45 performed in R , using the packages ‘effsize’ and ‘randomForests’ . database (GFED4). J. Geophys. Res. Biogeosciences 118, 317–328 (2013). 26. Sodhi, N. S., Koh, L. P., Brook, B. W. & Ng, P. K. L. Southeast Asian biodiversity: an impending disaster. Trends Ecol. Evol. 19, 654–660 (2004). Data availability 27. Tracewski, Ł. et al. Toward quantification of the impact of 21st-century The Human Footprint dataset used in this study is available from the Dryad Digital deforestation on the extinction risk of terrestrial vertebrates. Conserv. Biol. 30, Repository with the identifier doi:10.5061/dryad.052q5 . The other datasets that 1070–1079 (2016). support the findings of this study derive from published sources, cited in the 28. Bland, L. M., Collen, B., Orme, C. D. L. & Bielby, J. Predicting the Methods section and listed in Table 1. conservation status of data-deficient species. Conserv. Biol. 29, 250–259 (2015). Received: 10 May 2018 Accepted: 9 October 2018 29. Di Marco, M. et al. A novel approach for global mammal extinction risk reduction. Conserv. Lett. 5, 134–141 (2012). 30. Pacifici, M. et al. Generation length for mammals. Nat. Conserv 5,89–94 (2013). 31. Riahi, K. et al. The shared socioeconomic pathways and their energy, land use, and greenhouse gas emissions implications: an overview. Glob. Environ. Chang 42, 153–168 (2017). References 1. De Vos, J. M., Joppa, L. N., Gittleman, J. L., Stephens, P. R. & Pimm, S. L. 32. Visconti, P. et al. Projecting global biodiversity indicators under future development scenarios. Conserv. Lett. 9,5–13 (2016). Estimating the normal background rate of species extinction. Conserv. Biol. 29, 33. Jetz, W. & Freckleton, R. P. Towards a general framework for predicting threat 452–462 (2015). 2. Brook, B. W., Sodhi, N. S. & Bradshaw, C. J. A. Synergies among extinction status of data-deficient species from phylogenetic, spatial and environmental information. Philos. Trans. R. Soc. B 370, 20140016 (2015). drivers under global change. Trends Ecol. Evol. 23, 453–460 (2008). 34. Sanderson, E. W. et al. The human footprint and the last of the wild. 3. Murray, K. A., Verde Arregoitia, L. D., Davidson, A., Di Marco, M. & Di Bioscience 52, 891–904 (2002). Fonzo, M. M. I. Threat to the point: improving the value of comparative extinction risk analysis for conservation action. Glob. Chang. Biol. 20, 483–494 35. IUCN. IUCN Red List of threatened species version 2012.1. Version 20104 2008, (2012). (2014). 36. IUCN. IUCN Red list categories and criteria, version 3.1. (IUCN Gland, 4. Purvis, A., Gittleman, J. L., Cowlishaw, G. & Mace, G. M. Predicting extinction risk in declining species. Proc. R. Soc. Lond. B 267, 1947–1952 (2000). Switzerland and Cambridge, UK, 2001). 8 NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07049-5 ARTICLE 37. Di Marco, M. et al. Drivers of extinction risk in African mammals: the 54. NASA Earth Observatory Group (2018) Normalized difference vegetation interplay of distribution state, human pressure, conservation response index. Retrieved from http://neo.sci.gsfc.nasa.gov/view.php? and species biology. Philos. Trans. R. Soc. Lond. B Biol. Sci. 369, 20130198 datasetId=MOD_NDVI_M. (2014). 55. Hansen, M. C. et al. High-resolution global maps of 21st-century forest cover 38. Cohen, J. Statistical Power Analysis for the Behavioral Sciences (2nd ed.). change. Science 342, 850–853 (2013). (Academic Press, New York, 1988). 39. Di Marco, M. & Santini, L. Human pressures predict species’ geographic range Acknowledgements size better than biological traits. Glob. Chang. Biol. 21, 2169–2178 (2015). M.D.M. acknowledges the support of the ARC Centre of Excellence for Environmental 40. Allouche, O., Tsoar, A. & Kadmon, R. Assessing the accuracy of species Decision. distribution models: prevalence, kappa and the true skill statistic (TSS). J. Appl. Ecol. 43(6), 1223–1232 (2006). 41. Olson, D. M. et al. Terrestrial ecoregions of the world: a new map of life on Author contributions earth. Bioscience 51, 933–938 (2001). M.D.M. and J.E.M.W. conceived the study; M.D.M. performed the analyses; M.D.M., 42. GRASS Development Team. Geographic Resources Analysis Support System O.V., H.P.P., J.E.M.W. discussed the results; M.D.M., O.V., H.P.P., J.E.M.W. wrote the (GRASS) Software, Version 7.0. (2016). manuscript. 43. R. Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing. (2015). 44. Torchiano, M. effsize: Efficient Effect Size Computation. R package version Additional information 0.7.1. (2017). Available at: https://cran.r-project.org/package=effsize. Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467- 45. Liaw, A. & Wiener, M. The randomforest package. R. News 2,18–22 (2002). 018-07049-5. 46. CIESIN & CIAT. Gridded Population of the World, Version 3 (GPWv3): Competing interests: The authors declare no competing interests. Population Density Grid. Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC) (2005). Available at: http://sedac.ciesin. columbia.edu/data/set/gpw-v3-population-density. (Accessed: 1st November Reprints and permission information is available online at http://npg.nature.com/ 2013) reprintsandpermissions/ 47. CIESIN, FAO & CIAT. Gridded Population of the World, Version 3 (GPWv3): Population Count Grid, Future Estimates. Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC) (2005). Available at: Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in http://sedac.ciesin.columbia.edu/data/set/gpw-v3-population-count-future- published maps and institutional affiliations. estimates. (Accessed: 1st November 2013) 48. Nelson, A. Travel time to major cities: a global map of Accessibility. Global Environment Monitoring Unit - Joint Research Centre of the European Commission, Ispra Italy. (2008). Available at: http://bioval.jrc.ec.europa.eu/ Open Access This article is licensed under a Creative Commons products/gam/index.htm. Attribution 4.0 International License, which permits use, sharing, 49. Jones, K. E. et al. PanTHERIA: a species-level database of life history, ecology, adaptation, distribution and reproduction in any medium or format, as long as you give and geography of extant and recently extinct mammals. Ecology 90, 2648 appropriate credit to the original author(s) and the source, provide a link to the Creative (2009). Commons license, and indicate if changes were made. The images or other third party 50. Tacutu, R., Craig, T. & Budovsky, A. Human ageing genomic resources: material in this article are included in the article’s Creative Commons license, unless integrated databases and tools for the biology and genetics of ageing. Nucleic indicated otherwise in a credit line to the material. If material is not included in the Acid. Res 41, 1027–1033 (2013). article’s Creative Commons license and your intended use is not permitted by statutory 51. Verde Arregoitia, L., Blomberg, S. & Fisher, D. Phylogenetic correlates of regulation or exceeds the permitted use, you will need to obtain permission directly from extinction risk in mammals: species in older lineages are not at greater risk. the copyright holder. To view a copy of this license, visit http://creativecommons.org/ Proc. R. Soc. B 280, 20131092 (2013). licenses/by/4.0/. 52. Wilman, H. et al. EltonTraits 1. 0: species-level foraging attributes of the world’s birds and mammals. Ecology 95, 2027 (2014). 53. Rondinini, C. et al. Global habitat suitability models of terrestrial mammals. © The Author(s) 2018 Philos. Trans. R. Soc. Lond. B. Biol. Sci. 366, 2633–2641 (2011). NATURE COMMUNICATIONS | (2018) 9:4621 | DOI: 10.1038/s41467-018-07049-5 | www.nature.com/naturecommunications 9

Journal

Nature CommunicationsSpringer Journals

Published: Nov 5, 2018

There are no references for this article.