The effectiveness of UK protected areas in preventing local extinctions

Protected areas (PAs) are a core component in global efforts to prevent further declines in biodiversity. We examine whether the United Kingdom's PA portfolio has reduced local extinctions of breeding birds and consider how the characteristics of different PA designations may have affected conservation outcomes. We use breeding bird atlas data to calculate the proportion of species in each 10 × 10‐km grid cell confirmed as breeding in 1988–1991 but not in 2008–2011. We find that the protection offered by PAs is little to no different from that of the wider landscape and, depending on the type of PA, attribute this to PAs being either too small or not managed for biodiversity outcomes. To meet the ambitious new global targets for expanding PA coverage set at the 15th UN Biodiversity Conference, all countries need to monitor and ensure PA effectiveness; high PA coverage does not guarantee that biodiversity is protected.

15th UN Biodiversity Conference (COP 15) to "Ensure and enable that by 2030 at least 30 per cent of terrestrial, inland water, and of coastal and marine areas, especially areas of particular importance for biodiversity and ecosystem functions and services, are effectively conserved and managed through ecologically representative, well-connected and equitably governed systems of protected area" (CBD, 2022).
Although they may originally have been designated for other reasons, PAs are principally intended to separate species from processes that threaten their persistence (Gaston et al., 2008), along with a potential role in facilitating range changes (Thomas et al., 2012).However, there is conflicting evidence that the expansion of PA coverage since 2010 has led to better biodiversity outcomes (Visconti et al., 2019).Indeed, species' populations continue to decline both inside and outside of PAs (Watson et al., 2014) and, globally, the rate of species extinctions has accelerated (CBD, 2020;WWF, 2020).Thus, while there has been good progress toward achieving the target to increase PA coverage percentages, the extent to which PAs are effective remains less clear.
The UK government is a signatory to the CBD, with an established portfolio (sensu Gaston et al., 2006) of designated sites that presently covers 28% of terrestrial and inland water (UNEP-WCMC, 2023).Despite an extensive PA portfolio, the United Kingdom's wildlife is among the most depleted in Europe (Newbold et al., 2015).Over 40% of UK monitored species have declined in abundance over the past five decades (Hayhow et al., 2019).The UK government's 25 Year Environment Plan (25YEP) to protect and restore wildlife includes the development of a 'Nature Recovery Network' (Defra, 2018).Evaluating the effectiveness of the current PA portfolio could usefully inform decisions on wildlife site expansion and management.Previous studies have assessed how well UK PAs perform in terms of meeting IUCN definitions (Starnes et al., 2021) and biodiversity representation (Cunningham et al., 2021;Jackson et al., 2009).It remains less clear how effective PAs are in retaining species in the landscape (i.e., preventing local extinctions) and whether effectiveness varies between different types of PAs and the wider landscape.
UK PAs consist of two major types that present contrasting characteristics from the perspective of conservation planning principles.The first set of PAs are those designated primarily for their biodiversity features and comprise Sites (or Areas) of Special Scientific Interest (hereafter we use the term SSSIs to refer to both).Most SSSIs are small, with a median size in England of 23.1 ha, although many individual sites are not contiguous, instead comprising smaller patches of habitat fragmented by roads and other features.This small size makes them highly vulnerable to external pressures and threats, which, even if otherwise in favorable condition, reduces their ability to support viable populations of many species in isolation, where extinction risk is negatively correlated with population size, and population size decreases with patch area (Hanski, 1998) and where the foraging ranges of many species in UK landscapes are known to be far greater than the size of most SSSIs (MacKenzie et al., 2013).
The second set of PAs are landscape-scale designations and include Areas of Outstanding Natural Beauty (AONBs), their equivalent National Scenic Areas (NSAs) in Scotland (hereafter AONB is used to refer to AONBs and NSAs), and National Parks (NPs).These PAs were established to conserve and enhance natural beauty, wildlife, and cultural heritage and promote public enjoyment and understanding, with the focus as much on the protection of landscape character and heritage as on biodiversity.Furthermore, although AONBs and NPs cover much larger and more aggregated areas of land than do SSSIs (and commonly include SSSIs within their boundaries), they often include built up areas and farmland, leaving less space for natural and semi-natural habitats of principal importance for biodiversity.While these designations count toward targets for PA coverage to benefit biodiversity, it is questionable whether they achieve conservation goals in practice due to conflicting management goals and the degree to which nature conservation is prioritized when resolving these conflicts.Consequently, questions have been raised about the credibility of nature protection offered by these protected sites (Starnes et al., 2021).
Using data from the two most recent British Trust for Ornithology (BTO) breeding bird atlases of Britain and Ireland at 10 × 10-km grid resolution, we assess whether PA coverage of a grid cell was related to the probability of species ' local extinctions between 1988-1991 and 2008-2011.For the United Kingdom, we provide a novel comparison of conservation outcomes between different PA designations (NPs, AONBs, and SSSIs) and assess PA effectiveness based on species persistence.We aim to answer the following questions: This provides a case study of how site design and quality can impact PA performance in a country with extensive PA coverage and could provide useful evidence on how to balance the prioritization of quality versus quantity of habitats in PAs to ensure that ambitious area-based targets following the COP 15 will deliver for biodiversity.

Bird data
Data on breeding bird distributions were obtained from the British Trust for Ornithology's "The New Atlas of Breeding Birds in Britain and Ireland: 1988-1991" (Gibbons et al., 1993) and "Bird Atlas 2008-2011: The breeding and wintering birds of Britain and Ireland" (Balmer et al., 2013).We extracted all available species records (243 species) for terrestrial hectads (i.e., a 10 × 10-km grid cell) in the United Kingdom (n = 3003) and, for each grid cell, determined whether a species was recorded as breeding in the 1988-1991 survey ("Breeding") and the 2008-2011 survey ("Confirmed").For each grid cell, we calculated the number of extinctions as the total number of species recorded as breeding in the grid cell in 1998-1991 that were not recorded in 2008-2011 (Figures S1-S4) (Figures S1-S4) and also calculated the total number of species present in 1998-1991.These values were used to model the proportion of extinctions per grid cell (see statistical analysis).There were 178,081 unique species-by-hectad presences recorded in the 1998-1991 atlas.

Environmental data
In order to evaluate the effect of PA coverage and type on local bird extinctions, we determined for each 10 × 10-km grid cell the proportion of the land area that was protected under NP, AONB, and/or SSSI status and the maximum contiguous area of SSSI that overlapped with the grid, which included the contiguous extent of SSSI outside of the cell using the geographical boundaries of each PA (source: Department of Agriculture, Environment & Rural Affairs, 2021; Natural England, 2021;Natural Resources Wales, 2021;Scottish Government, 2021).Throughout, we excluded SSSIs designated solely for geological features.
We also summarized additional environmental variables that might affect local extinction rates at the 10 × 10-km grid cell resolution.Elevation data were derived from the Registry of Open Data on AWS using the "get_dem" function in the microclima R package (Maclean et al., 2019).UK gridded human population data for 2011 were downloaded from the UK Centre for Ecology and Hydrology (Reis et al., 2017).We obtained land cover data from the 2015 UKCEH Land Cover Map for Great Britain (Rowland et al, 2017) and the Habitat Map of Scotland (HabMoS; Scottish Natural Heritage, 2018).We calculated the proportion of high-nature-value (HNV) habitat in NPs, AONBs, SSSIs, and the wider countryside (non-NP/AONB/SSSI land) and in each grid cell (see Supporting Information).HNV habitat was defined as major habitats known to be important for biodiversity in the UK context (see Table S2).These were broadleaved woodland, neutral grassland, calcareous grassland, Fen, marsh and swamp, heather, heather grassland, bog, inland rock, supralittoral rock, supralittoral sediment, littoral rock, saltmarsh, and Caledonian forest.

Statistical analysis
The effect of the proportion of NP (or AONB) in grid cell i on the proportion of bird species becoming locally extinct between the two time periods   was modeled as where   is the number of species present in cell i in the historic period,   is the expected probability of extinction in cell i,  is the dispersion parameter, and  is the overall intercept;  10   and  10   are the  10 ( + 1) transformation of elevation and human population densities in cell i, respectively, with the transformation applied to account for the many small and few very large values.These predictor variables were included to control for expected differences in local extinction rates along gradients of elevation and human population density associated with historical patterns of species loss and habitat management and not specifically related to the presence of PAs.For example, it is likely that nonurban areas have undergone greater change in the interval between the atlases, thus leading to greater species losses in areas of low population density.Similarly, land cover change and the expansion of human populations into higher elevations could have driven species extinctions at higher elevations.The proportion of SSSI (  ) in grid cell i was included in the model to account for differences in local extinction rates that might be associated with habitat or management rather than directly due to the presence of NPs or AONBs.Finally, one of   or   , which are the proportions of grid cell i covered by NP or AONB, respectively, but not designated as SSSI, was included in each model to test the effect of these variables on local extinctions.To reduce residual spatial autocorrelation, spatial coordinates were included as covariates in the model via a Gaussian Markov Process smoothing function with the number of knots set to 180.This consistently reduced the Moran's I from highly significant to nonsignificant.NP or AONB identity was modeled as a random intercept ([  ]).The proportion of a grid cell covered by land was included as a weight in the model to account for coastal cells having less land area than inland cells (i.e., a grid cell with only 50% land coverage contributes half as much to the log likelihood as a grid cell with 100% coverage).
The effect of SSSI on the expected proportion of local extinctions was evaluated with respect to either the total proportion of SSSI (  ) within a cell or the log 10 (x + 1) and transformed maximum contiguous size (ha) of SSSI overlapping a grid cell (  10   ), that is, including contiguous extent of SSSI outside of the cell:  (  ) =  +  10   +  10   +  [   10 ]  +  (  ,   ) .
To test whether habitat quality in NPs, AONBs, or SSSIs affected local extinctions independent of coverage or size, we fitted a model that included the amount of HNV habitat inside the PAs in a grid cell and the interaction between these variables and coverage of NPs, AONBs, or SSSIs.Furthermore, because PAs may target threatened species, we conducted the analysis only considering birds species Red-Listed in the United Kingdom.Details of methods and results are presented in the Supporting Information, but effect sizes were consistently small.
All analysis was conducted in R 4.1.1(R Core Team, 2021) and statistical models were fitted using the "mgcv" package (Wood & Wood, 2015).

Size distribution of PAs and proportional coverage of PAs in grid cells
The median size of SSSI is 29 ha but many individual sites adjoin others or are not contiguous and comprise smaller patches.The median size of contiguous SSSI "polygons" is 0.084 ha (Figure 1) but varies between constituent countries of the United Kingdom and has a median size of 7.15 ha in England (Table S1).The median size of NPs is 143,783.2ha and the median size of AONBs is 19,748.2ha (Table S1).
Proportional coverage of SSSI was >50% for only 6% of 10 × 10-km grid cells containing SSSIs, whereas proportional coverage of NPs was >50% for 54% of grid cells containing NPs and >50% for 47% of grid cells containing AONBs (Figure S5).

Effect of NP and AONB
Both NP (PropNP) and AONB (PropAONB) coverage were negatively correlated with the expected proportion of local extinctions, such that increasing the area designated as NP or AONB in a grid cell results in fewer species going extinct (Tables 1 and 2), but the effect sizes were small, equating to a 1% decrease in the proportion of local extinctions with an increase in NP or AONB coverage from 0% to 100%.The effect size of NP coverage was slightly larger for UK Red-Listed species (ca.3%; Table S7).The interaction between the proportion of NP classed as HNV habitat and NP coverage in a grid cell was not statistically significant (t = -1.159,p = 0.247), suggesting that for a fixed value of NP cover, increasing the proportion classified as HNV habitat did not affect the expected proportion of local extinctions.Similarly, the interaction between the proportion of AONB classed as HNV habitat and AONB cover was also not statistically significant (t = 1.361, p = 0.174).

Effect of SSSI
The proportion of SSSI coverage () was not significantly correlated with the expected proportion of local extinctions (Table 3), but the maximum area of contiguous SSSI overlapping the cell ( 10 ) had a significant negative correlation with the expected proportion of local extinctions (Table 4).Thus, an increase in the size of the largest contiguous area of SSSI overlapping a cell (including contiguous area of SSSI outside of the cell) resulted in a decrease in the expected number of local extinctions, although the effect size was again small.Effect sizes were similar for Red-Listed species (Tables S9 and S10).The interaction between the proportion of SSSI classed as HNV habitat and both SSSI coverage measures was not significant (: t = 0.357, p = 0.721;  10   : t = 1.06, p = 0.289), suggesting that for a fixed value of either maximum area of contiguous SSSI overlapping the cell or proportion of SSSI coverage in a cell, increasing the proportion classified as HNV habitat did not affect the expected proportion of local extinctions.

DISCUSSION
Here, we have shown weak effects of protected sites in the United Kingdom on the retention of bird species in the landscape.SSSIs are numerous, are managed for biodiversity, and contain a high proportion of HNV habitat, but were not found to protect breeding birds against local extinction.A possible explanation for this is their small median size, which, at 29 ha, is less than half that of the average farm holding in the United Kingdom of 81 ha (Defra, 2020) and much smaller than the home ranges (territory) of many birds of prey (e.g., barn owls: Bunn et al., 2010).While these species are rarely the features for which SSSIs are designated, this illustrates how many SSSIs are incapable of supporting even single pairs of larger bodied breeding species, let alone viable populations.Low SSSI coverage in a grid cell limits the effect that the protected land can have to reduce the risk of extinction in that cell.
Nevertheless, habitat quality could also be an issue in SSSIs and requires further investigation.
Our finding that SSSIs have more HNV habitat than the wider countryside and extinction risk was lower in grid cells with larger contiguous areas of SSSI-designated land within their borders provides support for species' declines on small protected sites being in large part due to their size.Making existing SSSIs bigger may thus have greater value than designating more SSSIs, especially if new sites were small, patchy, and unconnected to the current sites.Bigger SSSIs could support larger populations that would be less extinction prone and have a greater capacity to (re)colonize surrounding habitat (Lawson et al., 2012;Wilson et al., 2002).The finding that NPs and AONBs were only marginally effective at reducing extinctions might seem to conflict with the suggestion to make SSSIs bigger.NPs and AONBs protect large and aggregated areas and yet also fared little better than the wider environment in preventing local bird extinctions.However, many of these sites do not have nature conservation as their management priority and lack large areas of high-quality wildlife habitat (Lawton et al., 2010), thus making populations vulnerable to extinction (Thomas et al., 2011).Less than half the habitat they contain is HNV.We find little evidence of habitat type effects on extinction risk, but this may be due to HNV habitat classes and the data used to quantify these being unable to detect the particular aspects of quality affecting birds.Furthermore, we do not measure other aspects of quality, such as patch sizes and connectivity, matrix type, and management that might affect HNV habitat.Small "islands" of HNV habitat within larger low-nature-value habitat (e.g., agricultural land) might be more vulnerable to threats (e.g., predation).Source-sink effects may also undermine the benefits of HNV; NPs have quite low coverage, so much of the landscape could be acting as a sink.Quality effects may not manifest below a particular level of coverage.Quality effects associated with the time since a PA was designated may plausibly be associated with effectiveness, although evidence for this is mixed (Geldmann et al., 2013;Zhao et al., 2019), and our results suggest no effect on local extinctions (Figures S10 and S11).
We recognize the limitations of using land cover data to assess habitat quality but stand by our conclusions that quality is the most likely explanation of our results for NPs and AONBs.We show that high levels of local extinction have occurred across a broad spectrum of bird species and apart from raptor persecution and conditions on migration (e.g., hunting pressure) and wintering grounds for summer migrants, habitat loss and degradation are the most plausible drivers for these losses.NPs and AONBs cover several contiguous 10 × 10-km grid cells, and so it is highly unlikely that increases in coverage would reduce local extinctions and, therefore, improving the management of these sites for nature will be required.Both NPs and AONBs need to perform better to achieve long-term conservation of nature (Visconti et al., 2019) and to ensure they continue to count toward the United Kingdom's PA coverage (Starnes et al., 2021).
While PAs are only weakly associated with reducing local extinction risk for birds as a taxonomic grouping, a finding supported by other recent studies (Barnes et al., 2023;Sanderson et al., 2023), individual bird species and species from other taxonomic groups could still benefit from even small, high-quality protected sites (Barnes et al., 2023).For example, greater PA coverage is associated with sites that contain a large population of adders Vipera berus in the United Kingdom (Gardner et al., 2019) and the majority of butterfly and Odonata species maintain higher abundance in PAs than outside across their historical UK ranges (Gillingham et al. 2015).Broadly, evidence suggests that although terrestrial PAs provide some protective function, they are not universally effective at preventing population declines (Geldmann et al., 2013;Hallman et al., 2017).Ferraz et al. (2003) suggest that for forest birds, a 1000-fold increase in PA area is required for a 10-fold increase in the time it takes for a site to lose 50% of its species.Thus, although such scaling relationships are likely to vary across taxa, it is clear that stabilizing longterm rates of biodiversity loss in PAs requires substantial increases in the size of individual sites.
While our study focuses on the United Kingdom, it has important global implications.First, we present the United Kingdom as an example of a country with relatively high PA coverage and yet poor biodiversity protection by PAs.This is despite the United Kingdom having longestablished wildlife legislation, well-monitored biota, and some of the world's largest membership organizations dedicated to conservation.Second, we highlight the importance of creating large areas of high-quality habitat and emphasize the importance of strategic planning for PA networks to ensure that they are capable of meeting minimum requirements for supporting species populations and ecological processes.The UK PA networks have to a marked degree been established ad hoc and their performance likely reflects this lack of strategic oversight.Third, we note a lack of rigorous monitoring schemes designed specifically to allow for evaluation of PA performance relative to the counterfactual of no or different protection status and to evaluate performance relative to site management.This is despite the United Kingdom being one of the most well-surveyed countries and having multiple national systematic monitoring programs.Ambitious global targets to expand PA coverage were set at COP 15 and it will be important for all countries to monitor, evaluate, and, where necessary, modify their PA networks to ensure that species are protected against threatening processes and are able to maintain viable populations.

A C K N O W L E D G M E N T S
The work was part-funded by Natural England.ASG was funded by the Natural Environment Research Council (NERC) iCASE studentship grant NE/P01229/1, in partnership with Cornwall Council.DJB was funded by a United Kingdom Research and Innovation (UKRI) Landscape (i) How effective have NPs and AONBs been at preventing local extinctions?(ii) How effective have SSSIs been at preventing local extinctions?(iii) Does the contiguous size of SSSIs influence their effectiveness at preventing local extinctions?

F
I G U R E 1 Size distribution of UK Sites of Special Scientific Interest (SSSIs) (a) and of contiguous SSSI 'polygons' (b), excluding areas <0.001 ha (n = 233).Size is expressed on a log10 scale of area in hectares.Dashed vertical dark blue line represents the median size.Bar colors indicate values for constituent countries of the United Kingdom.F I G U R E 2 The proportion of total habitat in National Parks (NPs), Areas of Outstanding Natural Beauty (AONBs), Sites of Special Scientific Interest (SSSIs), and the wider countryside composed of high-nature-value habitat.

TA B L E 1
Abbreviations: Est.df, estimated degrees of freedom; Ref.df, reference degrees of freedom.