Local biodiversity is higher inside than outside terrestrial protected areas worldwide — TEI-tagged view

Claudia L. GraypersName, Samantha L. L. HillpersName, Tim NewboldpersName, Lawrence N. HudsonpersName, Luca BörgerpersName, Sara ContupersName, Andrew J. HoskinspersName, Simon FerrierpersName, Andy PurvispersName, Jörn P. W. ScharlemannpersName (2016date). “Local biodiversity is higher inside than outside terrestrial protected areas worldwide.” Nature Communicationstitle·j.

Object of study — four local biodiversity measures at sites inside and outside terrestrial protected areasterm·key worldwide

Research question — whether protectionterm·dom benefits biodiversity beyond what reduced land-use change explains

Contribution — global evidence that protectionterm·dom raises local richness and abundanceterm·key, mostly through land useterm·key

Full text — tagged

Local biodiversity is higher inside than outside terrestrial protected areasterm·key worldwide Abstract Protected areasterm·key are widely considered essential for biodiversity conservation. However, few global studies have demonstrated that protectionterm·dom benefits a broad range of species. Here, using a new global biodiversity database with unprecedented geographic and taxonomic coverage, we compare four biodiversity measures at sites sampled in multiple land usesterm·key inside and outside protected areasterm·key. Globally, species richnessterm·key is 10.6% higher and abundanceterm·key 14.5% higher in samples taken inside protected areasterm·key compared with samples taken outside, but neither rarefaction-based richness nor endemicityterm·mth differ significantly. Importantly, we show that the positive effects of protectionterm·dom are mostly attributable to differences in land useterm·key between protected and unprotected sites. Nonetheless, even within some human-dominated land usesterm·key, species richnessterm·key and abundanceterm·key are higher in protected sites. Our results reinforce the global importance of protected areasterm·key but suggest that protectionterm·dom does not consistently benefit species with small ranges or increase the variety of ecological niches.

Protected areasterm·key are considered an essential strategy for habitat and species conservation 1 . Parties to the Convention on Biological Diversity have committed to increase the terrestrial area currently under protectionterm·dom 2 from 15.4% to at least 17% in ‘effectively and equitably managed, ecologically representative and well connected' protected areasterm·key by 2020 (Aichi biodiversity target 11) 3 . A recent assessment of progress suggests that the coverage target will be met, and that protected areaterm·key representativeness and management are improving 4 .

However, there is some doubt over the success of protected areasterm·key. Management effectiveness reports suggest only 22% of protected areasterm·key have ‘sound management' 5 , experts estimate that only half of all tropical reserves are effective 6 and human pressures are increasing in Latin American, African and Asian protected areasterm·key 7 . Declines in animal and plant abundanceterm·key have been documented inside protected areasterm·key 6 , 8 , 9 and in many countries the effectiveness of protectionterm·dom is being compromised by external pressures and inadequate government support 10 , 11 . Protecting all terrestrial sites of conservation significance could cost US$ 76 billion annually 12 , plus associated opportunity costs (up to US$ 6,500 ha −1 for productive agricultural land 13 ). Quantifying the effectiveness of protected areasterm·key is therefore crucial to justify maintaining and expanding the network.

Evidence from remote sensing suggests that protected areasterm·key slow change from ‘natural' to ‘human-modified' land cover 14 and successfully retain forests 1 . In the tropicsterm·key, protectionterm·dom reduces deforestationterm·dom 15 , 16 , 17 , loss of carbon 18 and forest fires 19 . There is some evidence that changes in land cover and human pressure vary with IUCN Protected Areaterm·key Management Categories (henceforth ‘IUCN categoryterm·key') 20 , although protected areasterm·key that have been assigned categories with more restrictive management objectives have not consistently experienced less land-use change, as their location may be more important than their IUCN categoryterm·key 16 , 17 , 19 , 21 , 22 , 23 , 24 .

Importantly, preventing land-use change does not necessarily conserve species within protected areasterm·key (for example, hunting may still occur 25 ) and there are few regional or global assessments of how protectionterm·dom affects species and assemblages. Geldmann et al . 1 found that most of the 42 studies in their meta-analysis reported that abundancesterm·key of species were higher inside protected areasterm·key, but acknowledged the limited evidence base. Coetzee et al . 22 , reviewing 86 studies, found a positive effect of protectionterm·dom on species richnessterm·key and abundanceterm·key. These two meta-analyses are informative but their use of effect sizes—rather than primary data—constrained the biodiversity measures that could be considered. Consequently, it is not yet clear whether protectionterm·dom solely maintains greater numbers of individuals (and therefore higher species richnessterm·key) or additionally maintains a greater variety of niches. If the latter is true, we expect protectionterm·dom to be associated with more species for a given number of individuals (that is, a higher rarefaction-based richness 26 —hereafter ‘rarefied richnessterm·key').

A further limitation of some previous studies is that the apparent success of protected areasterm·key may be caused by their location rather than protectionterm·dom per se 27 . To account for this bias, several analyses of land cover have matched protected locations or pixels to unprotected counterparts having similar values of potential confounding variables such as elevation or slope 14 , 17 , 19 , 28 , 29 , 30 . Alternatively, potential confounding variables have been included as covariates in models of land-use change 31 , 32 , including some analyses that have compared matched pairs of protected and unprotected locations as differences between paired locations remained even after matching 24 . Both previous meta-analyses of protected-area effects on species 1 , 22 were limited in their ability to account for potential confounding factors using either of these approaches as they lacked geographic coordinates for individual sample sites and data on specific land useterm·key types. Therefore, it remains undecided whether protectionterm·dom offers benefits to biodiversity beyond those attributable to reduced land-use change, that is, whether protectionterm·dom can benefit biodiversity even within a particular land useterm·key.

Here we assess the effect of protectionterm·dom on species and assemblages using collated primary data rather than effect sizes, quantifying the effects of protectionterm·dom both among and within land usesterm·key, while controlling for potentially confounding variables (see Methods). Using the PREDICTSterm·mth (Projecting Responses of Ecological Diversity In Changing Terrestrial Systemsterm·mth) database 33 , which collated data on species' presence or abundanceterm·key at sampled sites from peer-reviewed spatial comparisons of different types and intensities of anthropogenic pressures, we calculate four biodiversity measures based on sampled abundancesterm·key and occurrences at each site (henceforth ‘within-sample' biodiversity measures). The database records geographic coordinates of sampled sites, allowing selection of comparably surveyed sites located inside and outside protected areasterm·key by intersecting with the World Database on Protected Areasterm·dom (WDPAterm·dom) 34 . We extract data from 156 studies, including 13,669 species of vertebrates, invertebrates and plants, that had sites both inside ( n =1,939 sites) and outside ( n =4,592 sites) 359 terrestrial protected areasterm·key ( Fig. 1 ) and use mixed-effects modelsterm·key to assess the effects of protectionterm·dom while accounting for among-study differences in sampling methodology. Although this represents a small fraction of the protected areaterm·key network (0.18%), it is substantially larger than previous meta-analyses 1 , 22 and spans 48 countries, 101 ecoregions and 13 of the 14 terrestrial biomes. The sampled protected areasterm·key show a similar distribution to that of all terrestrial protected areasterm·key in the WDPAterm·dom 34 in terms of size, year of establishment and IUCN categoryterm·key and are reasonably representative in terms of how their total area is divided among land usesterm·key, ecoregions and biomes ( Supplementary Fig. 1 ). We find that within-sample species richnessterm·key and total abundanceterm·key are significantly higher at sites inside than outside protected areasterm·key. Importantly, significant interactions between protectionterm·dom and land useterm·key in our models indicate that protectionterm·dom does more than merely prevent land-use change (suggesting protectionterm·dom benefits biodiversity even within human-dominated land usesterm·key). However, in contrast to our expectations, protectionterm·dom has no effect on either rarefied richnessterm·key (suggesting protectionterm·dom has little effect on the number of species present for a given number of individuals, and therefore does not increase the variety of viable niches available) or endemicityterm·mth (suggesting protectionterm·dom has little effect on the proportion of individuals within a community that have narrow geographic ranges). Finally, we estimate that on average the global protected areaterm·key network is 41% (95% confidence interval (CI): 2.0 to 81%) effective at retaining within-sample species richnessterm·key and 54% (95% CI: 0 to 136%) effective at retaining local abundanceterm·key.

Spatial and taxonomic structure of data. ( a ) Location of 4,592 sites outside and 1,939 inside 359 protected areasterm·key (pale green; all terrestrial protected areasterm·key from ref. 34 ) from 156 studies in the full (all sites) dataset; horizontal lines indicate equator and tropicsterm·key of Cancer and Capricorn; world map from ref. 51 . ( b ) Diagram of a single study, showing sampling sites inside and outside a protected areaterm·key (green outline) and different land usesterm·key (shades of grey). ( c ) Latitudinal and taxonomic distribution of sites by 6° latitudinal bands for all sites (left) and subset of sites matched by land useterm·key (right); colours correspond to taxonomic groups (green=plants, blue=invertebrates, orange=vertebrates). ( d ) Diagram of the same study as ( b ) showing subsets of sites matched by land useterm·key (matched-sites data; this included 144 studies with 3,296 sites outside and 1,719 inside 313 protected areasterm·key). Sampled protected areasterm·key have similar distribution to all terrestrial protected areasterm·key in terms of size, year of establishment and IUCN Protected Areasterm·key Management Category (henceforth IUCN categoryterm·key) and broadly similar proportion of total area in each of the land usesterm·key, ecoregions and biomes represented ( Supplementary Fig. 1 ).

Results Local biodiversity inside and outside protected areasterm·key Samples from protected sites ( Fig. 1b ) contained 10.6% more species (95% CI: 4.1 to 17.6%; χ 2 =9.99, df =1, P= 0.002; Fig. 2a ) and 14.5% more individuals (95% CI: 2.0 to 28.7%; χ 2 =5.09, df =1, P= 0.024; Fig. 2b ) than samples from unprotected sites. If protectionterm·dom maintains a wider set of viable niches, we would also expect rarefied richnessterm·key (that is, the number of species expected if each site within a study had yielded the same number of individuals) to be higher inside protected areasterm·key. However, there was no significant effect of protectionterm·dom on rarefied richnessterm·key (95% CI: −31.06 to 13.5%; χ 2 =1.33, df =1, P >0.2; Fig. 2c and Supplementary Table 1 ), suggesting that the higher within-sample species richnessterm·key of protected sites largely reflects higher overall abundanceterm·key (as the number of species detected increases with the number of individuals sampled 26 ). Assemblages in protected areasterm·key might also be expected to have a higher proportion of endemic species and a smaller proportion of widespread species. However, we found the effect of protectionterm·dom on endemicityterm·mth (a measure we calculated as the inverse of community-weighted mean geographic range sizeterm·mth) was marginally nonsignificant (95% CI: −0.6 to 11.1%; χ 2 =2.99, df =1, P= 0.08; Fig. 2d ).

Effects of terrestrial protected areasterm·key on four local biodiversity measures. ( a ) Species richnessterm·key, ( b ) total abundanceterm·key, ( c ) rarefied richnessterm·key and ( d ) endemicityterm·mth at sites inside (filled circles) relative to sites outside protected areasterm·key (open). Estimates are given separately for protected areasterm·key in different management category groups (grey circles; least restrictive (IUCN categories III–VI), unknown (missing IUCN categoryterm·key, potentially mixed set of categories) and most restrictive land management regimes (IUCN categories I and II)). Bars indicate 95% CIs; ** P <0.01 and * P <0.05. Number of sampled sites in each category is shown; sample sizes vary between panels due to differences in the use of occurrence and abundanceterm·key data to calculate biodiversity measures (see Supplementary Information ). Separate generalized linear mixed effects modelsterm·key were run for each response variable (see Supplementary Information for further information). Supplementary Fig. 3 gives the corresponding results for analyses where sites were matched across the protected areaterm·key boundary by land useterm·key.

Analyses of spatial comparisons cannot reveal the mechanism behind differences between protected and unprotected sites. Protected areasterm·key may have been selected based on pre-existing biodiversity gradients, or protected areaterm·key management may result in the preservation of populations lost from surrounding areas. These mechanisms are not mutually exclusive. Time-series data on species and assemblages inside and outside protected areasterm·key would be helpful for determining their relative importance. Additionally, we may underestimate the overall contribution of protectionterm·dom, as many protected areasterm·key aim to protect biodiversity features not included in our analyses, such as beta diversity, particular rare species, migratory routes or ecological processes 20 .

Protected areasterm·key with more restrictive management objectives are expected to retain more biodiversity. Although IUCN categories are not necessarily applied consistently across countries 20 , several studies have used them to compare biodiversity across management objectives 16 , 17 , 19 , 21 , 22 , 23 . As most of these studies focus on land cover (particularly deforestationterm·dom), there is little information on how effects on species differ among protected areasterm·key in different IUCN categories; the limited available evidence indicates a positive effect size for both the most restrictive IUCN categories and for one category that allows some extraction 22 . We found no significant differences among protected areaterm·key management category groups (three groups, in order of lowest to highest restriction on human activity: IUCN categoryterm·key III–VI, IUCN categoryterm·key unknown (a mixture of categories, considered an intermediate category on average) and IUCN categoryterm·key I and II; Fig. 2a ; all comparisons among protected groups gave P >0.2), but samples from protected sites in each management category group had higher species richnessterm·key than those from unprotected sites ( Fig. 2a ; χ 2 =11.18, df =3, P= 0.011. There was large variation but no significant difference in abundanceterm·key, rarefied richnessterm·key or endemicityterm·mth among management category groups and unprotected sites ( Fig. 2 and Supplementary Table 1 ). The heterogeneity within groups could reflect conservation objectives not captured in a protected areaterm·key's IUCN categoryterm·key 35 or differences in how these categories have been applied 20 .

Effects of protectionterm·dom within and among land usesterm·key As all analyses above accounted for differences in elevation, slope and agricultural suitability, differences in land useterm·key are the most likely explanation for higher species richnessterm·key and total abundanceterm·key in samples from protected sites. We used two approaches to test this hypothesis. First, we included site-specific land useterm·key as a fixed effect in our models; and second, we restricted our dataset to sites matched by land useterm·key across the protected areaterm·key boundary ( Fig. 1d ). In both cases, we also explored whether the effect of protectionterm·dom varied with latitudinal zone and taxonomic group.

Land useterm·key and the effects of protectionterm·dom The effect of protectionterm·dom on within-sample species richnessterm·key, abundanceterm·key and endemicityterm·mth varied among land usesterm·key ( Fig. 3 ; χ 2 =15.26, df =7, P =0.033; χ 2 =19.12, df =7, P= 0.008; χ 2 =25.05, df =7, P= 0.001, respectively) but again the effect on rarefied richnessterm·key did not ( Fig. 3c and Supplementary Table 3 ). The last result reinforces the finding that samples from protected sites do not have more species for a given number of individuals, even across land usesterm·key. Protectionterm·dom had little effect on biodiversity measures at sites within primary and secondary vegetation, probably because such sites tend to experience limited human pressure whether outside or inside protected areasterm·key. However, in human-dominated land usesterm·key—particularly plantation and cropland—samples from protected sites contained significantly more individuals and species than those from unprotected sites ( Fig. 3a,b ), but only in the tropicsterm·key ( Supplementary Table 2 , Supplementary Fig. 2e and f ). The greater effect within the tropicsterm·key is encouraging given they are often considered a high conservation priority (for example, ref. 36 ). However, given recent acceleration in human activity in the tropicsterm·key 37 , tropical landscapes may be experiencing an extinction debt that has already been repaid in temperate protected areasterm·key. Perhaps surprisingly, endemicityterm·mth was lower at protected than unprotected sites in human-dominated land usesterm·key, particularly in cropland ( Fig. 3d and Supplementary Table 2 ) and for vertebrates and plants ( Supplementary Table 2 and Supplementary Fig. 2d ). This effect suggests that protected areasterm·key in human-dominated land usesterm·key may either benefit species with wide ranges or may have been located specifically to protect them (for example, migratory birds).

Effects of protectionterm·dom on four biodiversity measures across eight land useterm·key types. ( a – d ) Sites outside (open circles) and inside (filled circles) protected areasterm·key in different land usesterm·key (colours: from left to right: primary vegetation; mature, intermediate and young secondary vegetation; plantation; cropland; pasture; urban). Error bars show 95% CIs. The number of sites in each type of land useterm·key and protectionterm·dom is given underneath each data point. Separate generalized linear mixed effects modelsterm·key were run for each response variable (see Supplementary Information for further information). Supplementary Fig. 2 gives corresponding analyses for taxonomic group, latitudinal zone and use intensity.

If protectionterm·dom ameliorates human pressures even within land usesterm·key, differences in land-use intensity could explain the higher biodiversity at protected than unprotected sites within the same land-use type. The PREDICTSterm·mth database specifies three levels of use intensity within each land-use type 33 : minimal (for example, very limited levels of disturbance for natural land usesterm·key, low-intensity agriculture), light (for example, some extraction of timber, hunting or pesticide application) and intense (for example, clear-felling, high level of hunting, intensive agriculture, highly urbanized). Even when comparing unprotected and protected sites experiencing the same use intensity in human-dominated land usesterm·key, samples from protected sites consistently contained more individuals and species than those from unprotected sites ( Supplementary Table 2 and Supplementary Fig. 2i,j ). These results suggest an influence of factors not captured in our measure of use intensity (for example, the condition of the wider landscape, or habitat restoration).

Effect of protectionterm·dom when sites are matched by land useterm·key Analysing only the sites within each study for which land useterm·key could be matched across the protected areaterm·key boundary ( Fig. 1d ), we found no significant effect of protectionterm·dom on any biodiversity measure for any management category group, taxonomic group or latitudinal zone ( Supplementary Table 3 and Supplementary Fig. 4 ). However, samples from protected areasterm·key that were both young (<20 years) and small (<400 km 2 ) had higher species richnessterm·key than samples from unprotected sites ( Fig. 4 ; χ 2 =16.22, df =4, P= 0.003); rarefied richnessterm·key again did not differ significantly ( Supplementary Table 3 ). It is possible that more recently designated protected areasterm·key have targeted areas of high species richnessterm·key more precisely through use of spatial prioritization algorithms (for example, ref. 38 ), or that older protected areasterm·key have declined in species richnessterm·key. The effect of protected areaterm·key size/age class on abundanceterm·key and endemicityterm·mth also varied among taxonomic groups ( χ 2 =25.38, df =8, P= 0.001; χ 2 =16.64, df =8, P= 0.034, respectively, Supplementary Fig. 4 ), with higher invertebrate abundanceterm·key, vertebrate abundanceterm·key and plant endemicityterm·mth in samples from larger, older protected areasterm·key. These results suggest that larger, older protected areasterm·key may have been designated to protect, or have retained or increased the local abundanceterm·key of animals and less geographically widespread plants.

Effects of protected areaterm·key age and size class on within-sample species richnessterm·key. The number of sites sampled outside and inside protected areasterm·key is given for each age and size combination. Effects were tested using a generalized linear mixed effects model with a Poisson error distribution (log link). See also Supplementary Fig. 4 .

Estimating the global effectiveness of protected areasterm·key We extrapolated from our models of local biodiversity inside and outside protected areasterm·key ( Fig. 3 ) to estimate the effectiveness of the current global protected areaterm·key network at retaining site-level biodiversity. Our measure of effectiveness would be 0% if sites within protected areasterm·key are, on average, as diverse as unprotected sites, and 100% if the biodiversity of protected sites is, on average, as high as for ‘pristine' sites (minimally-impacted primary vegetation; note that the scale is not bounded at either 0% or 100%). On this scale, we estimate that on average the global protected areaterm·key network is 41% (95% CI: 2 to 81%) effective at retaining within-sample species richnessterm·key and 54% (95% CI: 0 to 136%) effective at retaining local abundanceterm·key.

How to improve the global protected areaterm·key network An important question for environmental decision makers is whether biodiversity will benefit more from increasing restrictions on human activity in existing protected areasterm·key or from expanding the network 39 , 40 . Although the trend towards higher species richnessterm·key and abundanceterm·key at sites in protected areasterm·key with more restrictive management objectives was not statistically significant, the coefficients suggest the effects of management restrictiveness may be large ( Fig. 2a,b ). The large uncertainty seen in these models reflects both a lack of precise data on the objectives of protected areaterm·key management and on effectiveness, and a lower number of sites with the most restrictive management. Better information on protected areaterm·key management intent and effectiveness is needed to confidently quantify the biodiversity outcome of increasing management restrictions. However, to demonstrate the importance of having improved information on management, we used our model coefficients ( Fig. 2a,b ) to estimate that the effectiveness of the protected areaterm·key network could be increased to 94% (95% CI: 50 to 139%) for average within-sample species richnessterm·key and 167% (95% CI: 0 to 392%) for average within-sample abundanceterm·key if all protected areasterm·key enforced the most restrictive management objectives (that is, IUCN categories I or II). To raise average within-sample species richnessterm·key worldwide by the same amount solely through expanding the current protected areaterm·key network (i.e., with the existing distribution among IUCN categories) would require protectionterm·dom of 22% (95% CI: 12 to 63%) of terrestrial area; for within-sample abundanceterm·key, the corresponding figure is 31% (95% CI: 13 to 299%). The wide CIs on these estimates highlight the urgent need for improved data on the management of protected areasterm·key. To help decide whether to expand or change the management of the protected areaterm·key network requires information on the costs of expansion versus changes in management, the representation of species not currently under protectionterm·dom 41 and the extent to which these options achieve other globally agreed conservation targets 42 . Nevertheless, other recent studies also suggest that increasing the performance, rather than the total coverage of protected areasterm·key, may achieve the desired outcomes for biodiversity more efficiently 39 , 40 , 43 , 44 : this is an important issue requiring further study.

Discussion In summary, these first detailed global analyses of site-level data on a large, taxonomically broad set of species show that, overall, samples from protected sites contained more individuals and species than samples from unprotected sites. By contrast, protected sites did not consistently have higher rarefied richnessterm·key or levels of endemicityterm·mth—both measures of community characteristics that are often considered when setting conservation priorities. The greatest differences in species richnessterm·key and abundanceterm·key occurred across land usesterm·key: protected areasterm·key are most effective where they minimize human-dominated land useterm·key, especially where they safeguard primary or mature secondary vegetation. However, the positive effect of protectionterm·dom within human-dominated land usesterm·key, particularly in the tropicsterm·key, shows that land useterm·key is not the only cause of higher biodiversity within protected areasterm·key. Better data on management is needed to quantify biodiversity benefits of restricting human activity inside protected areasterm·key but, if the trend in our data is confirmed, more restrictive protected areaterm·key management across the current network could be as important as extending the network. Importantly, we cannot discern whether protectionterm·dom has prevented losses in site-level biodiversity seen in surrounding areas, increased numbers of individuals, or retained a pre-existing biodiversity gradient. Nonetheless, these analyses represent a substantial advance in knowledge about several measures of biodiversity inside versus outside protected areasterm·key. Our results reinforce recent calls 45 for increased support and recognition of the importance of protected areasterm·key worldwide 10 , 11 , but highlight that the network is not currently effective for all measures of local biodiversity.

Methods Data For each sampling location or site in the PREDICTSterm·mth database (November 2014 version), we calculated within-sample species richnessterm·key, total abundanceterm·key of individuals, rarefied richnessterm·key (based on the fewest individuals at any site within each study) and community weighted mean log 10 geographic range size—the inverse of which was then plotted to give our endemicityterm·mth measure. Each species' range size was derived from its global occurrence in the Global Biodiversity Information Facility database. We recognise biases in the Global Biodiversity Information Facility data, but these are mitigated to some extent by our hierarchical modelling approach and our estimates compare reasonably well with estimates based on other data sources, listed in full in the Supplementary Information . Land useterm·key was classified using the study authors' description for each site; this method has been shown to be repeatable 33 . Sites were considered to be protected if their geographical coordinates fell inside protected areasterm·key from the World Database on Protected Areasterm·dom 34 (see Supplementary Methods ). We then derived two datasets: the first included all studies with sites inside and outside protected areasterm·key (all-sites data; Fig. 1b ); the second retained only those sites from each study for which land useterm·key could be matched across the protected areaterm·key boundary (matched-sites data 2; Fig. 1d ). All sources of biodiversity data are listed in the Supplementary Information .

Analyses We used generalized linear mixed-effects modelsterm·key to account for differences in response variables due to study-specific methodologies and the spatial structure of sites 46 . The PREDICTSterm·mth data present a rare opportunity to compare sites inside and outside protected areasterm·key, but do not have the geographic coverage required for a stricter counterfactual approach 14 , 17 in which sites are individually matched. To reduce the risk that any differences observed between sites inside and outside protected areasterm·key were caused by biases in the location of protected areasterm·key 27 , we considered elevation 47 and derived slope at c. 1 km 2 resolution and agricultural suitability 48 at 10 km 2 resolution as covariates in all models (see Supplementary Information for further details). To ensure independence of all variables in the model, we intentionally included only these three confounding variables that we considered to be fully independent of the presence of a protected areaterm·key. For example, distances to roads and markets are affected by the presence of protected areasterm·key so are not independent confounding factors (see Supplementary Information for details). We sequentially compared models with and without each fixed effect and at each step dropped the term with the highest P -value, until all terms had P <0.05 (ref. 49 ).

Assessing protectionterm·dom effects We tested for biodiversity differences between sites inside and outside protected areasterm·key using the all-sites data, treating protectionterm·dom status (inside vs outside a protected areaterm·key) as a fixed effect. We then tested whether biodiversity measures differed between management category groups by re-coding IUCN categoryterm·key as a four-level factor: unprotected, IUCN categoryterm·key III–VI, IUCN categoryterm·key unknown, and IUCN categoryterm·key I and II.

Assessing protectionterm·dom effects within and among land usesterm·key We used two approaches to test whether biodiversity differences between protected and unprotected sites varied with land useterm·key. First, using the all-sites data, we modelled the response of each biodiversity measure to protectionterm·dom status, land useterm·key, and their interaction. We also tested for the three-way interaction between land useterm·key, protectionterm·dom and either use intensity, latitudinal zone or taxonomic group. Second, using the matched-sites data, we re-ran models with protectionterm·dom status, and then with management category group as a fixed effect. We also split the matched-sites data by latitudinal zone and taxonomic groups to assess whether these factors influenced the effect of protectionterm·dom. Finally, we tested whether the site-level biodiversity response to protectionterm·dom varied with the size/age class of the protected areaterm·key [four-level factor with all combinations of young (<20 years), old (20–85 years), small (<400 km 2 ) and large (400–12,000 km 2 ); these thresholds between categories were selected to give a similar number of sites in each group].

Estimating global protected areaterm·key effectiveness The global effectiveness of protected areasterm·key ( e ) was estimated from e =1−(1− i )/(1− o ), where modelled site-level biodiversity inside ( i ) and outside ( o ) protected areasterm·key are expressed as a proportion of that under ‘pristine' conditions. We calculated the ratio of i / o from the model estimates for biodiversity inside relative to outside protected areasterm·key in each land useterm·key ( Fig. 3 ), where each land-use parameter was weighted by the proportion of global terrestrial area within that land-use type. This value of i / o could then be used to solve an equation expressing the global state of site-level biodiversity: 1− r = ai +(1− a ) o , where r is the estimated global average loss of site-level biodiversity relative to pristine 46 and a is the fraction of the total land area that is protected 50 . Solving this equation for i and o allowed us to estimate e . Finally, by using estimates for the effect of protectionterm·dom in IUCN categories I and II ( Fig. 2a,b ) to give i / o , we estimated e under the more restrictive management scenario. By rearranging the equations we estimated the total protected areaterm·key ( a ) needed to obtain the same average local biodiversity outcome (1− r ) inferred under this more restrictive management scenario. See Supplementary Information for more details.

Data availability The biodiversity data that support the findings of this study are available in the Natural History Museum data portal ( data.nhm.ac.uk ) with the identifier dx.doi.org/10.5519/0095544. R scripts are available at http://github.com/claudialouisegray/PREDICTSterm·mth_WDPAterm·dom .

Additional information How to cite this article: GraypersName, C. L. et al . Local biodiversity is higher inside than outside terrestrial protected areasterm·key worldwide. Nat. Commun. 7:12306 doi: 10.1038/ncomms12306 (2016).

Supplementary Material Supplementary Figures 1-4, Supplementary Tables 1-3, Supplementary Note 1, Supplementary Methods and Supplementary References.

Acknowledgments We thank the hundreds of data contributors; all PREDICTSterm·mth project volunteers, masters and PhD students that collated records; and Adriana De Palma, Helen Phillips, Diego Juffe-Bignoli, Neil Burgess, Max GraypersName, Daniel Ingram, Valerie Kapos, Naomi Kingston, Sarah Luke and the protected areasterm·key team at UNEP-WCMC for comments and assistance. We thank the School of Life Sciences at the University of Sussex for support and the Natural History Museum for a GIA travel award. The PREDICTSterm·mth project is funded by the UK Natural Environment Research Council (NERC, grant number: NE/J011193/2). PREDICTSterm·mth is endorsed by the Group on Earth Observations Biodiversity Observation Network (GEO BON). This is a contribution from the Imperial College Grand Challenges in Ecosystem and the Environment Initiative, and the Sussex Sustainability Research Programme.

Footnotes References Associated Data Supplementary Materials Supplementary Figures 1-4, Supplementary Tables 1-3, Supplementary Note 1, Supplementary Methods and Supplementary References.

Data Availability Statement The biodiversity data that support the findings of this study are available in the Natural History Museum data portal ( data.nhm.ac.uk ) with the identifier dx.doi.org/10.5519/0095544. R scripts are available at http://github.com/claudialouisegray/PREDICTSterm·mth_WDPAterm·dom .