Factors influencing nest site selection in a rapidly declining shorebird, the Eurasian curlew
Abstract
In this period of rapid human-induced environmental change, it is vital that influences of habitat on the distribution and productivity of threatened species are understood. Ground-nesting birds are declining more rapidly across Europe than any other group, with large-bodied birds at the greatest risk of extinction. Productivity and adult survival cannot both be maximised concurrently, and individuals will make decisions during the nest-placement phase which will favour one outcome or the other; however, under conditions of accelerating change, these decision processes may become decoupled from positive fitness outcomes. The Eurasian curlew Numenius arquata is Europe's largest wader and is showing steep declines in breeding productivity. Curlews are known to use a diverse range of habitats for nesting, where anthropogenic or natural features may influence distribution. There is an urgent need to understand the spatial scales of these impacts, and whether habitat characteristics have a positive or negative impact on nest survival. In our study site in southern England, curlew showed a marked preference for nesting in wetter habitats, primarily mire, and a weaker selection of dry heathland. Nest survival improved in wetter habitats, and the area of mire round a nest site was positively associated with increased levels of nesting success, whilst area of scrub had a negative association. Woodland significantly excluded curlew from potentially suitable breeding habitat, with an impact observed up to 2 km from the nest site, but nest survival did not improve with distance from woodland. Curlews strongly avoided nesting near a major road passing through the study site, despite seemingly suitable habitat nearby. Understanding landscape effects will assist in planning future habitat management for curlew, impacts of forestry and avoidance of additional pressures on a species of high conservation concern.
Introduction
Ground-nesting birds are declining more rapidly across Europe than other species (McMahon et al. 2020), and when phylogeny is controlled for, large-bodied birds with long generation times are at the greatest risk of extinction (Bird et al. 2020). Habitat alteration or degradation is detrimental for many species, and magnified by synergistic impacts of climate change and changes in predator assemblages (McMahon et al. 2020, Lees et al. 2022). Nest site selection is the product of decisions which aim to maximise fitness as a function of one of two possible outcomes during the nesting phase: protection of the incubating adult and protection of the clutch (Nguyen et al. 2003, Allen Smith et al. 2007, Walpole et al. 2008, Anteau et al. 2012a, b2012b, Miller et al. 2015). Productivity and adult survival cannot both be maximised concurrently, and individuals will make decisions during nest-placement which will favour one outcome or the other (Seltmann et al. 2013). However, the relationship between phenotypic responses to habitat characteristics and fitness may be interrupted by rapid anthropogenic change to environments, so once-effective choices decouple from positive fitness outcomes. Additionally, strong site fidelity increases the risk that attractive habitats become an ecological trap through poor productivity or reduced adult survival (Ekroos et al. 2012).
The trade-off between clutch and adult survival is acute in ground-nesting waders since they are exposed to a wide and potentially rapidly changing suite of predators (Macdonald and Bolton 2008, Laidlaw et al. 2021, Douglas et al. 2023). Waders employ visual and auditory detection of predators as a defence allowing early escape responses, and restriction of this mechanism can result in higher levels of adult predation (Amat and Masero 2004) since dense cover can reduce escape opportunities and attenuate detection of predators (Schneider and Griesser 2013), despite potentially offering greater protection to the clutch (Ost and Steele 2009). Exposure to predators quantifiably enhances the ‘neural circuitry of fear'(Rosen and Schulkin 2004) in the brains of animals (Clinchy et al. 2012), and this may impact nest site decisions.
Some species of bird demonstrate adaptive plasticity in nest site selection in apparent response to predator densities (Spaans et al. 1998, Larsen 2000, Forstmeier and Weiss 2004), but a lack of phenotypic plasticity may cause some animals to tolerate unsuitable habitats in the face of novel threats, simply because they never previously encountered these during their behavioural evolution (McNamara et al. 2006). Choice of nest site habitat in Kentish plovers Charadrius alexandrinus can vary between nesting attempts, with vegetative cover levels changing if the earlier attempt had been predated (Amat et al. 1999). However, on a longer decision-time interval, Eurasian oystercatchers Haematopus ostralagus did not exhibit any measurable phenotypic plasticity in response to catastrophic flooding events, and continued to nest in vulnerable habitats (Bailey et al. 2017). In a climatically unstable world, this may be a threat which could disproportionately impact populations of long-lived species, which do not benefit from fast-turnover generations adapting to new circumstances (Bailey et al. 2017). Some evidence suggests that high-quality nesting patches include optimal foraging habitat, since this reduces associated fitness costs of energy and predation risk from travelling to a feeding site (Olsson and Bolin 2014). However, quality of available foraging habitat did not guide nest site selection in several tundra-nesting shorebirds, which displayed non-random avoidance of dry heath habitats with the highest local invertebrate abundance (Allen Smith et al. 2007). This indicates foraging requirements may sometimes be separate from decisions which drive nest site selection, and home range usage may not necessarily define the area of strongest habitat selection for clutch or adult safety during incubation.
The Eurasian curlew (Numenius arquata, hereafter ‘curlew') is a long-lived (longevity record in the wild from ringing data, 32.5 years – BTO 2023) philopatric bird which can breed beyond 30 years (Rivers unpubl.), so their nest-site selection decisions may be refined through perception of predator pressure, or degree of success in avoiding previous encounters. Curlews are classed as ‘Near Threatened' on the International Union for Conservation of Nature (IUCN) Red List (Birdlife 2017), and the UK breeding population of ca 59 000 pairs has declined 62% between 1970 and 2012, with a 46% loss between 1995 and 2013 (Harris et al. 2015, Hayhow et al. 2014). The New Forest in Hampshire represents one of five remaining strongholds for breeding curlew in southern England, yet even during the study period, declines in the numbers of breeding pairs have been observed despite intensive monitoring (a reduction of ca 25–33% between 2020–2022, Rivers unpubl.). It is difficult to accurately quantify local declines in the New Forest, since historical surveys have not been performed with a comparable methodology to recent work, and numbers were based on extrapolation from 33% of the area of suitable habitat, reaching a total of 132 pairs of breeding curlew (Tubbs and Tubbs 1996). More recent comprehensive survey work determined a population of ca 40–45 pairs (HOS 2001). Despite the methodological differences, these figures point to possible severe declines in breeding pairs. There is a pressing need to determine the key heathland habitats used, and influences on breeding productivity for this southern lowland curlew population: whose loss would represent a significant UK range contraction. We urgently need a better understanding of habitat requirements in curlews, and in particular their response to woodland, because net-zero carbon targets in the UK are driving legislation to increase tree cover in the UK by 16.5% of the total available land by 2050 (ca 1 million ha – The Environmental Targets (Woodland and Trees Outside Woodland) (England) Regulations 2023). In Wales, the Sustainable Farming Scheme proposes grant payments to farmers who plant 10% of their land with trees (Department for Environment, Food and Rural Affairs 2023), a policy which will require sensitive understanding of curlew responses to tree presence if it is not to impact their distribution on otherwise suitable land. With climate change an increasingly dominant factor in the political agenda: driving policy to greatly accelerate tree planting to three times current rates, there is £20 m in funding for woodlands and commercial forestry, and a target of approximately 30 000 ha of new woodland every year by 2025 (Environment, Food and Rural Affairs Committee 2022). Forestry England is a Government executive body, and England's largest land manager, and has stated that tree planting should aim for the principle of ‘right tree, right place for the right reasons' (Forestry Commission 2017). If these tenets are to be fulfilled alongside protection of breeding curlew, then there is an urgent need to understand their behavioural and demographic responses to woodland. If curlew are excluded from suitable nesting habitat by woodland planting, this may result in breeding attempts which are affected by poor-quality sites with higher predator pressure (Silva-Monteiro et al. 2021), lower levels of camouflage for the adult and clutch, and poor feeding resources. This risks further limiting curlews' ability to mitigate the pressures they are currently facing. For a species potentially so close to extinction in the UK (Brown et al. 2015), this could be catastrophic.
Previous studies have found commercial afforestation was associated with declines in breeding pairs (Douglas et al. 2014, Franks et al. 2017), and loss of curlew breeding habitat through fragmentation and direct reduction of area (Ratcliffe 2007). Two studies found only weak evidence of an effect of woodland on curlew distribution (Żmihorski et al. 2018, Zielonka et al. 2020), but both were in grassland habitats, with one in Poland which has overall greater availability of open space than the UK and a very different suite of predators. Hence, these may be limited in their generalisability to heathland-nesting populations as in the New Forest. The variability of habitats selected for nesting in published literature suggests that preferences may need to be studied at the level of local populations, to ensure correct identification of potential ecological sinks and appropriate protection measures for local landscapes. Nevertheless, wider inferences may be possible using the growing body of evidence from studies across different regions with similar habitats available.
Therefore, in this study we sought to identify the key characteristics of habitats selected during the discrete stage of incubation by a declining curlew population (ca 40–45 pairs at the initiation of the study in 2020) breeding in a heathland environment in the south of England, UK. We considered the habitat composition of the immediate and wider environment, and the distance of nest placement from selected natural and anthropogenic landscape features of potential importance, for example woodland. Since the area over which birds make decisions related to nest and adult safety are unknown, and likely to be independent to the foraging range captured by GPS tracking, we tested this systematically.
Material and methods
Study site
The New Forest National Park (Fig. 1) is principally managed by Forestry England, whose jurisdiction covers 29 214 ha (JNCC 2015), and holds one of the largest populations of curlew breeding in the English lowlands. It is designated as a special area of conservation, Site of Special Scientific Interest and Ramsar site, and forms the largest extent of lowland heathland found in the UK (JNCC 2015). The maintenance of the heathland habitat on this scale is largely due to centuries of grazing by cattle, ponies, donkeys, and occasionally sheep owned by ‘Commoners' with rights to turn animals out on the Forest (Tubbs 1965, Putman 2022, Newton 2020).

Map of the New Forest National Park (colour shading) and its location within the UK. National Park boundary (dashed line) obtained from: https://naturalengland-defra.opendata.arcgis.com/datasets/national-parks-england. Squares represent curlew nesting areas, heavy black lines major roads (A31 (north) and A35 (south)), and light lines minor roads.
The dry heaths comprise ling heather Calluna vulgaris, dwarf gorse Ulex minor, and bristle bent grass Agrostis curtisii, and in many places transition into Atlantic wet heath, holding the largest extent found in southern England. Atlantic wet heath is composed primarily of cross-leaved heath Erica tetralix (Sphagnum compactum type); Schoenus nigricans–Narthecium ossifragum mire, and interspersed in many places with bog myrtle Myrica gale. Frequently, wet heath transitions into mire (quaking bogs and alkaline fens) with sedge (Carex spp.), rush (Juncus spp. and Schoenus nigricans), and purple moor-grass Molinia caerulea dominating (Fig. 2 – nests shown in each habitat). Additionally, acid grassland and wet grassland with Molinia feature widely, together with blocks of mixed deciduous and coniferous evergreen woodland (JNCC 2015).

Nests in the three key habitats used by breeding curlew in the New Forest – (a) dry heath, (b) mire and (c) wet heath.
The New Forest was selected because of its geographical importance to breeding curlew in the UK (largest most southerly population – Colwell et al. 2020), whilst being subjected to the combined pressures of habitat fragmentation, human recreation, and the broad predator suite facing many other declining curlew populations (Berg 1992, Grant et al. 1999, Valkama et al. 1999, Baines et al. 2023), as well as intensive human recreation (Grant and Edwards 2008). Located on the south coast of the UK, GPS-tagged birds breeding in the New Forest are known to winter in nearby coastal areas including Beaulieu, Southampton, Portsmouth, Brownsea and Weston-super-Mare.
Location of nests
Curlew nests were located between 2020 and 2022. Surveys covered all areas where curlew were known to breed, based on long-term observations which were shared by Forestry England and independent bird observers who have studied this population for many years (Potts et al. 2009, Wynn and Page 2018, Page and Currie unpubl., HOS 2006). Curlews are highly site-faithful (Berg 1994, Valkama et al. 1998, Saurola et al. 2013, Pakenen and Kylmänen 2023), and New Forest breeding pairs have been observed to consistently occupy the same nesting areas, identified by repeated visits across multiple seasons. GPS-tagging and colour-marking prior to and during the study period has provided further evidence of breeding site fidelity (Rivers unpubl.). We have confidence that if there were curlews breeding outside of our surveyed areas, their breeding behaviour would have been reported to us by other wildlife personnel working here. Breeding site surveys were conducted by one or two individuals, in settled weather, to maximise the chances of detecting early-season breeding behaviour; and consisted of walk-through and point observations (de Jong et al. 2013), of approximately 1 h per historic territory initially. Nest visits are not known to cause an increase in the likelihood of nest predation in breeding birds; therefore, these methods were considered unlikely to bias the outcomes of the monitored nesting attempts in this curlew population (Galbraith 1987, Verboven et al. 2001, Fletcher et al. 2005, Ibáñez-Álamo et al. 2015, Salewski and Schmidt 2022). If breeding behaviour was identified, birds were observed until a nest was located, often requiring another visit at a different time of day when birds were more active, i.e. at dawn. During all observation periods (other than walk-throughs), surveyors remained in a position where their presence did not cause any disturbance or consequential alteration of the birds' behaviour, for example concealed in scrub vegetation or in a vehicle from a track.
Across three years of fieldwork, 76 active nest locations were recorded to 1 m accuracy. These were pooled to form one dataset. This meant that uncommon year effects, such as the Covid lockdowns, would not have a disproportionate impact on the analyses. Habitat classifications and local features (including minor roads and other infrastructure) were taken from Forestry England data, and trunk roads were included using Ordnance Survey shapefiles. Factors which were investigated for their potential to influence curlew nest site selection included: dominant habitat types at the precise nest site and within four size classes of buffer (50–2000 m radius) surrounding the point; and distance to natural or anthropogenic features (habitat types = woodland, dry heath, wet heath, mire, dry grassland, wet grassland and pond; anthropogenic features = campsites, car parks, minor B roads, two major roads – the A31 and A35). Additionally, some replacement clutches were also investigated for any evidence of switching habitats following a failed attempt, and mean distance to the next nearest curlew nest was also calculated by year to clarify approximate core territory sizes.
Nest detection probability
A binomial generalised linear mixed model was performed in R (www.r-project.org) on the survey dataset (binomial outcome = bird seen/not seen). The same potential effects on detectability were considered as fixed effects as in the analysis of nest placements: dominant habitat type, mean distances to car parks, campsites, minor roads, woodland, and to the A35 and A31. Additional fixed effects were survey number and effort level; but since all nests were located through fieldwork rather than the GPS tags, factors relating to tag performance were not included. Random factors were the individual sites where pairs were surveyed. Nest placements were determined through intensive fieldwork (ca 2 months of 10–12 h days each year) by experienced nest-finders, but it is conceivable that there were differences in the detectability of the birds in different habitats; or behavioural differences in relation to different levels of human activity, which could influence the number of nests identified in a certain habitat and thus the observed results. As detailed visit data were gathered for the 2022 field season, covering who performed the survey, effort (scale of 1–3 from incidental record to full walk-through and point survey) and the adjoining habitat and landscape variables, it was possible to further analyse these data to understand whether any factors were impacting the detection probability of a nesting bird. All visit data were included up to and including the survey where a nest was detected.
GPS-tracking, home ranges, and distances travelled by incubating adults
Since 2018, adult curlews were caught for GPS tagging and colour-marking in the New Forest and nearby Solent coastline (Potts et al. 2009). Birds were caught using a variety of methods: on the nest, using either a walk-in trap or a hand-net; or using a mist-net or cannon-net. Initially, tags manufactured by Pathtrack or Movetech Telemetry were used, and latterly Ornitela. They were attached using leg-loop harnesses from 1.5 mm EPDM rubber cord passed though Silastic tubing or from 2.0 mm silicone cord. Both allowed a small amount of expansion for weight gain, and tags were known to have been shed safely, typically after ca 18 months (Hoodless unpubl.). During the breeding season the tags were programmed to take hourly fixes of the birds' positions, which were uploaded every two days. For the seven GPS-tagged individuals who were carrying a tag during eleven known complete or near-complete incubation periods (determined by the use of a trail camera at the nest or intensive observations), shapefiles of real 95 and 50% home ranges were produced. Where birds had been nest trapped, this usually took place near the end of incubation so their data were not included until the subsequent year, when the whole nesting period could be considered; and four individuals have yielded data for two years' full observation where they were mist-netted early in the season or caught overwinter.
As animal movement data are inherently temporally and spatially autocorrelated, we considered it inappropriate to use kernel density estimators to predict the extent of core home range usage (Noonan et al. 2015) and, instead, the home ranges were estimated using continuous time movement modelling (CTMM in running in the web app provided at https://ctmm.shinyapps.io/ctmmweb, Dong et al. 2018), a method which accounts for this inherent autocorrelation (Fleming and Calabrese 2017). Here data were extracted directly from Movebank (Wikelski et al. 2006) and manipulated in CTMM to produce shapefiles of 95% home range estimations, with upper and lower confidence intervals applied. To satisfy the requirements of the movement model, the stationary portion of the data (incubating on the nest) was segmented out, leaving only excursions from the nest site (Fleming unpubl.). The ‘stationary points' were defined as those within a 9.0 m buffer of the nest site, to allow for GPS location error. This 9.0 m error was calculated by averaging across measures for the three types of tag used: for Ornitela, we calibrated the tag data with trail camera images of the birds on the nest with the tag visible, which yielded an average error of 6.0 m. Movetech tags on an equivalent 60-min fix schedule were found to have a 6.50 m horizontal error (Acácio et al 2022). Error data gathered by Bowgen et al. (2012a) on equivalent Pathtrack tags, where any fixes with fewer than six satellites were removed, gave an error of 14.95 m. Averaged together this gave an error of 9.15 m. Finally, any duplicate fix records for all manufacturers were removed. The remaining data were fitted to the best CTMM foraging model, a continuous-velocity Ornstein-Uhlenbeck-F model (OUF).
These home ranges informed understanding of the area covered by incubating curlew, and whether there were strong inter-individual differences indicative of diverse strategies. The distances birds must cover to fulfil their resource acquisition needs will be dictated by both biotic and abiotic features of local patches to their nest (Brown 1984). Therefore small home ranges would suggest birds have easy access to good foraging, whilst meeting the need for safety of adult/clutch in the nesting habitat, whilst large home ranges suggest the birds must travel further to meet their foraging needs. To further define space-use during incubation, the minimum, mean, median and maximum Euclidean distances travelled from the nest were also calculated for each tagged bird during each incubation period recorded.
Habitat within buffers around nests
Habitat composition was considered within four circular buffer sizes (Kenward et al. 2001, Olivier and Wotherspoon 2006, Klug et al. 2009, Mueller et al. 2009) set at 50, 500 m, 1 and 2 km to sample biologically informative areas over which decision making may take place during nest site selection. As these areas have not to our knowledge been defined by previous research, habitat composition around real nests in incremental buffer sizes were tested against the habitat composition in the random equivalently buffered nest site sets, using Spearman's rank correlation to assess differentness. This established a potential area of greatest influence in habitat selection. To identify the alternative habitat options available to the birds in the area where they placed their nest, the proportion of each usable nesting habitat (dry heath/ wet heath/mire) in 500 m buffers was averaged across sets of nests placed in each habitat. For example, the average proportion of dry heath, wet heath and mire was calculated surrounding nests where wet heath was selected, and this process repeated for nests located in mire and dry heath.
Comparison of home ranges with composition of circular buffers around nests
The habitat composition in the home ranges was additionally compared with that of the four buffers used in this study, to understand if there was a good correlation between any size class despite the inherent disparity in shape. Since there was a known nest site associated with each bird equipped with a GPS tag, a circular buffer was added to this unique point, and the habitat composition within the buffers was compared with that of the real home range. This comparison was performed including proportion of woodland, which although not used for nesting was a habitat of significance during the breeding phase (Pearce-Higgins and Grant 2003, Douglas et al. 2014). All buffer sizes were then tested against 95 and 50% home ranges using Spearman's rank correlation.
Distance of nests from natural landscape and anthropogenic features
Natural and anthropogenic features were selected for investigation based on their theoretical potential to deter or encourage nesting. For example, some may represent decreased visual scope for predator detection or greater abundance (i.e. woodland, Douglas et al. 2014, Franks et al. 2017), or enhanced levels of human disturbance (i.e. car parks or campsites) which is known to impact curlews during the non-breeding season (Burton 2007, Navedo and Herrera 2012, Scarton 2018). Minor and major roads were included for their potential to increase noise pollution and disturbance, as well as human littering, which again may attract generalist predators such as carrion crow Corvus corone and red foxes Vulpes vulpes (Katlam et al. 2018, Preininger et al. 2019). Finally, as curlew have been frequently observed feeding around the edges of New Forest ponds (Rivers unpubl.), permanent ponds were included in the analysis. The mean of the shortest distance from the real nest sites to each of the features was calculated. Additionally, estimation of approximate territory size was approached by considering the mean distance to the next nearest curlew nest within year sets.
Comparison of real nests to randomly located nests
Our dataset showed that during the study period, curlew in the New Forest were only observed to nest in mire, wet heath and dry heath. To determine whether nest placement within the three used habitats was non-random, the habitat and distance metrics calculated for the real nests were compared to the equivalent values for the same number (76) of replicate randomly distributed nests within these habitats. The location of random nests was determined in R ver. 4 (www.r-project.org) using the st_sample function of the ‘sf' package (Pebesma 2018, Pebesma and Bivand 2023). For each replicate, this generated 76 randomly located nests within the combined area of mire, wet heath and dry heath. We generated 10 000 replicate sets of randomly located nests with habitat and distance metrics calculated for each. The habitat and distance metrics measured for the real nests would be expected to be similar to those determined for the randomly located nests if the real nests were randomly distributed across mire, wet heath and dry heath. Observed habitat and distance metrics were then compared to the 0.5, 2.5, 97.5 and 99.5% quantiles for randomly located nests to determine whether observed habitat and distance metrics were greater or lower than expected.
Habitat influences on nest survival
During the study period, Forestry England took the decision to make changes in predator management in certain areas of the New Forest, in response to some of the early findings in our study regarding nest predation. Therefore those nests that were subject to enhanced levels of predator control were removed from the dataset to avoid confounding any relationships between habitat and nest survival, leaving a sample of 49 nests. The R package ‘RMark' ver. 3.0.0 (Laake 2013) was used to perform the nest survival analysis, with the candidate model covariates including the nest site habitat (mire, dry heath and wet heath); area of habitats in the 500 m buffers; area of scrub and woodland inthe 500 m buffers; distance from scrub; and distance from woodland. Estimates of daily survival rate (DSR) without a time component (assuming constant daily survival) for each habitat type and covariate were tested; as well as additionally including a time component to consider changes in nest survival rate in each habitat through the breeding period. The data collection period covered by the remaining nests in the dataset was 67 days, and DSR across this time period was modelled in relation to each of the covariates to understand if nests are more vulnerable to predation in certain habitats as the season progresses.
Results
Location of nests and detection probability
Of the 76 curlew nests analysed across three years, 25 were placed in mire, 31 in dry heath and 20 in wet heath. No effect of habitat type or distance to features was found on detection probability beyond survey number (p < 0.05; Supporting information) as a function of the increased likelihood of detection as surveys progressed and observers gained understanding of curlew locations.
Across the study area there was substantially more dry heath available than mire or wet heath (4237 versus 1249 ha and 1224 ha, respectively), yet the proportion of real nests placed in dry heath did not reflect this. The mean numbers of random nests in mire, dry heath and wet heath were 14, 48 and 14. The number of real nests in dry heath was lower than the 0.5% quantile of the random nests, suggesting this habitat was being avoided by the birds (Fig. 3). The number of real nests in mire was higher than the 99.5% quantile of the random nests, suggesting that this habitat was being selected by the birds. The number of real nests in wet heath was between the 2.5 and 97.5% quantiles of the random nests, but closer to the 97.5% quantile, indicating some selection for this habitat.

Comparison of the habitats occupied by real and randomly located nests. Solid symbols show the number of real nests located in each habitat. Error bars show the range of random nest numbers in each habitat: left short bar = 0.5% quantile; left long bar = 2.5% quantile; right long bar = 97.5% quantile; right short bar = 99.5% quantile.
Of the nine pairs of curlews that produced replacement clutches following nest predation, just two persisted in the same habitat for subsequent attempts. Twice as many nests were placed in a wet habitat on the first attempt than dry; with the pairs switching to a dry habitat if that first attempt failed (Supporting information).
Home ranges during incubation
Curlew 95% home ranges during the incubation period ranged widely in size from 17.52 to 1244 ha (mean 182.72 ha, SE 107.26, Fig. 4, also see Supporting information), and 50% home ranges varied from 2.07 to 208 ha (mean 30.18 ha, SE 18.02). Broken down by sex, the male home ranges varied from 32.88 to 173 ha (mean 109.87 ha, SE 25.72), and the females from 30.62 to 1244 ha (mean 243.43 ha, SE 200.3). Both 95 and 50% home range sizes were dominated by the three habitats exclusively exploited for nest placement – wet heath, dry heath and mire. The greatest proportional habitat in the real 95% home ranges was wet heath (39.5%), followed by dry heath (31.6%), woodland (8.6%) and mire (8.3%). Less area was composed of dry grassland (7.2%), wet grassland (3%) and pond (1.9%). The 50% home ranges were mainly dominated by the three habitats – dry heath, wet heath and mire (35.6, 34.5.3 and 15.8%, respectively), with lesser coverage of woodland (5.3%) and dry grassland, pond and wet grassland (4.7, 2.1 and 2%, respectively) (Fig. 4).

Mean proportional distribution of habitats within all 95 and 50% home ranges, shown with mean proportional distribution of habitats within all 500 m buffers round the nest locations of the seven birds over 11 GPS tracked periods of incubation. Error bars represent 95% confidence intervals.
When the proportional habitat compositions of the 95% home ranges of GPS-tagged birds were compared with 50-, 500-, 1000- and 2000-m buffers around the associated year's nest site, the greatest overall correlation was between the 95% home range and the 500 m buffer, with a rho of 0.83 (p < 0.001), followed by the 1000 m buffer (rho 0.81, p < 0.001). The lowest correlations with the 95% buffer were the 50 m and the 2000 m buffer (rho 0.58 and 0.71, respectively, p < 0.001). The greatest positive correlation with the 50% home ranges was also the 500 m buffer with a rho of 0.83 (p < 0.001), followed by the 1000 m buffer (rho 0.77, p < 0.001). The lowest correlations again were with the 50 and 2000 m buffers (0.65 and 0.66, respectively, p < 0.001 – Fig. 4). Therefore, although the 500 m buffers were shaped differently to the home ranges, they were broadly representative of the habitat distribution selected by incubating curlew (Fig. 5).

All 95% home ranges included in the study – please note image h) is at a different scale to the preceding home ranges due to the much greater area it covers. White circles show 500 m buffers centred on nest sites (white dot).
Habitat within buffers around all nests
Most notably, the proportion of woodland within all size class buffers around real nests was lower than the equivalent 0.5% quantiles for all buffer classes around randomly located nests, suggesting that birds were markedly avoiding woodland at all studied scales. Conversely, the proportion of wet habitats (mire and wet heath) within the 50 m buffers around real nests was higher than the equivalent 99.5% quantiles for buffers around randomly located nests, suggesting that birds were strongly selecting these wet habitats; at 500–2000 m buffer scale this relationship becomes weaker for mire but remains more consistent for wet heath. Birds showed a non-random avoidance of dry heath within the 50 m buffers, but at 500–2000 m this relationship weakens and reverses, with the mean proportion of dry heath in the real buffers sitting between the 97.5 and 99.5% quantile distribution of the random nests, and then becoming a higher proportion than random at 1–2000 m. At both 50/500 m dry grassland comprised a lower proportion of the nest buffers than random, only reaching random distribution by the extent of the 1000 m buffer. At all size classes, the proportion of wet grassland sits within the lower end of the random distribution (Fig. 6–7).

Habitat composition of all size classes of circular buffers around real and randomly located nests. Due to variation in the areas of different habitats within buffers, two groups of habitats are plotted with different ranges of the horizontal axis. Solid symbols show mean habitat composition around real nests; error bars show the range of habitat composition around random nests: left short bar = 0.5% quantile; left long bar = 2.5% quantile; right long bar = 97.5% quantile; right short bar = 99.5% quantile.
When the mean habitat composition of the four size classes of buffer was tested for correlation with the equivalently sized buffers around the randomly distributed nests, the real nest 50 and 500 m buffers were the least correlated with random nest buffers (rho 0.79 and 0.74, respectively), and the real nest 1000 and 2000 m buffers were increasingly correlated with the random nest buffers (rho 0.83 and 0.93, respectively).
When the availability of each possible nesting habitat within the averaged 500 m buffer surrounding nest sets in each habitat was examined, it was evident other habitats were present within close proximity with their selected habitat, and high levels of heterogeneity existed: where the primary nesting habitat was wet heath, the mean 500 m buffered areas contained proportions of 0.11 mire, 0.44 dry heath and 0.35 wet heath. Where it was dry heath, the mean buffers contained 0.14 mire, 0.55 dry heath and 0.18 wet heath; and where it was mire, 0.22 mire, 0.49 dry heath and 0.16 wet heath (Fig. 7).

Habitat composition in a random sample of real nests showing habitat heterogeneity: (a) up to 2000 m buffer; and zoomed in (b) up to 1000 m buffer. Nest location shown as black dot, then 50, 500, 1000 m buffers (in (a) and (b)) and 2000 m buffer (in (a)) surrounding shown as black lines, and white areas are human settlement or non-Forestry England land.
Distance of nests from landscape and anthropogenic features
The mean distance from woodland to real nests was greater than the 99.5% quantile for random nests. Real nests had a mean distance to woodland of 438 m, whereas the equivalent distance for randomly located nests was 249 m. Similarly, the mean distance from the A31 to real nests was greater than the 99.5% quantile for random nests. Real nests had a mean distance to the A31 of 8246 m whereas the equivalent distance for randomly located nests was 6029 m. Additionally, there was evidence for some avoidance of B roads (mean distance from real and random nests 692 and 579 m, respectively), with the mean distance to B roads from real nests being greater than the 97.5% quantile of the equivalent value for randomly located nests. The distances from real nests to the remaining features (car parks, campsites, ponds and the A35) were all within the 2.5 and 97.5% quantiles of the random nests, not providing evidence that real nests were closer or further away from these features than expected (Fig. 8).

Comparison of the mean distance to landscape and anthropogenic features from real and randomly located nests. Due to variation in the distance to different features, two groups of features are plotted with different ranges of the horizontal axis. Solid symbols show mean distances from real nests. Error bars show the range of mean distances from random nests: left short bar = 0.5% quantile; left long bar = 2.5% quantile; right long bar = 97.5% quantile; right short bar = 99.5% quantile.
Distance to nearest nest
Average distance to the next nearest curlew nest was 824 m in 2020, 1454 m in 2021 and 984 m in 2022; giving an overall mean nearest neighbour distance of 1087 m (SD 327).
Distance travelled from the nest during incubation
The GPS-tagged curlews in the study were highly variable in the distances they travelled during incubation, with an average minimum distance of 4 m (range 0.15–32 m); mean distance of 520 m (range 154–1495 m); and maximum distance of 2771.5 m (range 498–12 286 m).
Nest survival and habitat
The most parsimonious habitat model within the nest survival analysis was MireArea (Table 1), with a positive relationship between the proportional area of mire in each 500 m buffer around the real nests (Fig. 9c). The next two strongest models were DryHeath and Mire which were the habitats demonstrating the lowest and the highest daily nest survival, respectively, of the three analysed (Fig. 9a). The interaction model between Habitat and Time was the next strongest (Fig. 9b), followed by ScrubArea (Fig. 9d) and WoodDist (Fig. 9e).

(a) Mean ± SE daily nest survival rate in dry heath, mire and wet heath, without a time component; (b)–(e) daily nest survival rate: (b) in the three habitats throughout the nesting period (day 1–67): mire = red, wet heath = blue and dry heath = black; (c) in relation to proportional area of mire in 500 m buffer round nests; (d) in relation to proportional area of scrub in 500 m buffer round nests; (e) in relation to distance from woodland (m).
Model | npar | AICc | DeltaAICc | weight | Deviance |
---|---|---|---|---|---|
S(~Time) | 2 | 170.1676 | 0 | 0.152485 | 166.1376 |
S(~1) | 1 | 170.6938 | 0.526295 | 0.117205 | 168.6839 |
S(~MireArea) | 2 | 170.8246 | 0.65702 | 0.109789 | 166.7946 |
S(~DryHeath) | 2 | 170.9568 | 0.78923 | 0.102766 | 166.9268 |
S(~Mire) | 2 | 171.2331 | 1.0656 | 0.089503 | 167.2031 |
S(~Habitat ×* Time) | 6 | 171.4376 | 1.270051 | 0.080805 | 159.2255 |
S(~ScrubArea) | 2 | 171.6772 | 1.50969 | 0.071681 | 167.6472 |
S(~WoodDist) | 2 | 172.3807 | 2.21319 | 0.050424 | 168.3507 |
S(~WHthArea) | 2 | 172.4998 | 2.33225 | 0.04751 | 168.4698 |
S(~habitat) | 3 | 172.5306 | 2.36305 | 0.046784 | 166.4704 |
Mean DSR in all habitat types was 0.9287 (SE 0.012, CI 0.9011–0.9490; Supporting information), therefore nest survival for the full incubation period of ca 28 days (Robinson 2023) was estimated at 0.1260 (ca 13%). In the simple comparison of DSR between the three habitats, assuming a constant survival rate throughout incubation, mire nests were found to have the highest DSR at 0.9465 (SE 0.0167, CI 0.9026–0.97122), followed by wet heath at 0.9289 (SE 0.0201, CI 0.8779–0.9596) and then dry heath at 0.8995 (SE 0.0289, CI 0.8273–0.9436), giving overall survival rates of 0.21, 0.13 and 0.05. When DSR was plotted through time (the period of observation of all nesting attempts = day 1–67) in the three habitats, there were strong divergences in the likelihood of survival between the three habitats, with mire nests maintaining a much more constant likelihood of survival throughout the breeding season, followed by wet heath and then dry heath. Nest survival decreased very slightly with distance from woodland, although the confidence levels widen at the greater distances due to the sample size at the larger distances. Overall, the trend does not suggest any clear relationship between proximity to woodland and nest failure.
Discussion
Despite the area of dry heath available to curlews in the New Forest, they showed a marked preference for mire and wet heath, and these relationships were mirrored in higher rates of estimated nest survival in these wetter habitats. Birds strongly avoided woodland in all buffer sizes; rejected dry heath at the smallest buffers; and preferred coverage of wetter habitats. Grassland was not an important nesting habitat for curlew in the New Forest. Of the distance measures, just woodland and the A31 trunk road were avoided more than expected from a random distribution of nests.
Despite some localised studies describing an association of breeding curlew with mire habitats (Henderson et al. 2002, Ewing et al. 2018), the strong relationship with mire in the New Forest has not been consistently described in other populations of breeding curlew. For example, in Sweden just 38% of the densities found in arable regions were found in mire (de Jong et al. 2013), and national declines have been observed in breeding curlew abundance in bog habitats in the UK (Franks et al. 2017), with some suggestion they may be moving toward drier grassland habitats. Many studies have described curlews' marked selection of grassland for nesting (Berg 1992, Valkama et al. 1998, Zielonka et al. 2020, Leprince et al. 2022), yet this is a habitat almost completely avoided by nesting curlews in the New Forest. The New Forest is grazed year-round, which assists in maintaining the heathland and mires in suitable condition for breeding waders, but equally this grazing will likely render grassland swards too short to offer any visual protection to either the clutch or the incubating bird, presumably causing their rejection of it as a nesting habitat.
It was notable that overall nest survival in areas without predation management was estimated at a very low level of ca 13%, a situation echoed in other areas of the UK where multi-year/site studies were undertaken. In North Wales, Bowgen et al. (2012a) estimated overall nest survival of ca 15% with reported daily nest survival rate of 0.935, comparable to our findings based on ca 28 day incubation. In a Northern Irish population, just 3.6–19.0% nests hatched each year (Grant et al. 2001); and a more recent study in Breckland in East Anglia reported nest survival rates of ca 25% (Ewing et al. 2022). These levels are somewhat lower than those reported in Sweden (35.6% – Berg 1992), and Finland (32% – Valkama and Currie 1999). Therefore, there is likely pressure on the birds to make habitat selection decisions that maximise clutch survival, whilst taking greater perceived risks with adult survival. An interesting finding in this study was that the order of preferences of nesting habitat closely mirrored the estimates of nest survival rates – with mire nests showing an elevated likelihood of surviving during the breeding season, followed by wet heath, and finally dry heath. This suggests that individuals who are prepared to risk the trade-off between adult safety and clutch safety – such as reduced visual range to maximise clutch survival – are playing the game successfully and seeing greater productivity in densely vegetated mire habitats, an outcome which is supported in some other localised studies in a population in Wales (Johnstone et al. 2017) and Sweden (Berg 1992, Fig. 11).
Nest survival in relation to the proportional area of mire in the 500 m buffers around the nest was the strongest habitat model in the analysis and supports the initial indication that the wet mire habitat may afford some kind of protection, particularly against mammalian predators. One GPS-tracking study in Germany showed foxes tend to avoid wet swampy habitats, and the authors suggested this may alleviate predation risk on ground-nesting birds (Fiderer et al. 2019); and Arctic geese nesting in mesic tundra and wetlands experienced higher nest survival in the wetland habitats with greater structural complexity (Lecomte et al. 2008). Typically, dry heaths in the New Forest support dense networks of narrow pathways created by deer and livestock, and trails like these are often used by foxes and badgers Meles meles (Muñoz-Igualada et al. 2010, Short et al. 2012). Fox GPS-tracking has shown that foxes travel and forage along linear features (Bischof et al. 2019, Schwemmer et al. 2024) and dry heaths include myriad tracks used by walkers and vehicles. Conversely, due to their wetness, wet mires are largely unnavigable to vehicles and pedestrians, which can reduce their accessibility to foxes and badgers. Nest survival declines with an increase in proportional area of scrub in the 500 m buffer, and this may relate to scrubby areas harbouring mammalian predators. The reasons for this avoidance of dry heath as a nesting habitat cannot be fully addressed by this study; however, in an area impacted by high numbers of visitors, dry heath is likely to receive greater levels of human disturbance, potentially increasing predation risk and nest exposure time. The relatively darker coloured habitat may not afford the same levels of visual camouflage to the incubating adult as the lighter and more heterogeneous vegetation of mire and wet heath, and ground-nesting birds may make fine-scale decisions to select habitat which most accurately matches their plumage (Stevens et al. 2017). Therefore, birds nesting deep in Molinia-dominated mire may sacrifice the ability to easily detect predators in favour of superior camouflage, and be able to more effectively employ the strategy of ‘sitting tight' on the nest when threatened (Fig. 11). Curlews can vary their flight initiation distance (FID) based on factors of incubation stage and diurnal period (de Jong et al. 2013), but their behaviour in different habitats has not been studied in detail to date. Feeling safe enough to maintain a short FID could minimise the intensity of scent trails around the nest, by reducing the number of times they flush and then walk back to the nest, and potentially reduce attention from olfactory-hunting predators. The safety of the nest when the adult has flushed may also be lower in dry heath habitats, since the nest frequently is more visible from above than when placed in mire (Rivers unpubl., also see Fig. 2), and for a species with a nest concealment strategy this could impact upon survival (Laidlaw et al. 2020).
Comparing the habitat composition of the four classes of buffer size with randomly distributed buffers suggested that the area of strongest selection takes place within approximately 500 m radius of the nest. When the proportions of available nesting habitats were calculated in the 500 m buffers, this showed that there were high levels of habitat heterogeneity within this nominal sample of their territory area. This suggested there were alternative nesting habitats available to the birds, and thus it was possible for them to make fine-scale decisions about habitat selection within their core territory areas. Therefore, it seems reasonable to assume that birds were able to make fine-scale selection choices within the habitat suite available to them, without needing to move to an unfamiliar area outside of their core territory. This possible behaviour is supported by our observation that 7/9 pairs selected alternative habitats following a nest failure. Whilst curlew are considered to be ‘site-faithful', they may maximise their potential fitness by employing a combination of a site familiarity strategy (Piper 2011), where knowledge of feeding opportunities, cover and neighbour interactions are beneficial, with a fine-scale version of the win-stay: lose-switch (Switzer 1997, Schmidt 2004) strategy: changing habitat if a first nesting attempt is predated, or persisting in a habitat year-on-year if attempts are successful. Therefore, although this dataset will inevitably contain repeat nesting attempts by the same pairs, either within years or across years, it was still considered appropriate to analyse it as independent datapoints since they arguably represent short-term decisions and re-evaluations of the fitness benefits conferred by any one habitat within their territory. However, this behaviour may not be apparent in birds who breed in more homogenous landscapes where the opportunity to switch habitat whilst maintaining site familiarity is less available. The implications of this may be that birds who are forced to nest in anthropogenically modified tracts of land such as improved grassland or arable farmland have fewer opportunities to employ these adaptive behavioural tactics and improve their chances of successful reproduction.
Several measures of habitat and space use during incubation suggested that the area of approximately 500 m surrounding the nest were of most critical importance when selection decisions are being made. The average median distance between nests was 544 m; the average distance travelled from the nest was 520 m; and the 500 m radius buffer size was the point where the differentness of habitat composition began to dissolve to become more correlated with random selection of available habitat. Additionally, the average 95% home range was 187 ha (which if transposed to a circle would have a radius of 771 m), which sits between the size of 500 and 1000 m buffers, and correlated most closely with habitat composition of the real home ranges. Whilst the home range shapes are inherently variable and influenced by foraging strategies dependent on local resource availability, constraints and individual priorities, it seems that for the purposes of this study the 500 m buffers adequately represent the habitat patches which are most important to the curlews, even if it is not possible to define a shape which could substitute a real curlew home range. Although one female in the study commuted considerable distances during incubation to feed on pasture, the majority of the individuals foraged nearer to the nest. The smallest home ranges were observed where a pond was used by all birds in the vicinity to regularly feed, supporting a link between home range size and food availability. In this case although they maintained ca < 700 m distance between their nests, their home ranges significantly overlapped at the pond (Fig. 10).

Differing adult concealment strategies in heathland habitat – (a) adult is in low dry heath vegetation with good visibility, and (b) adult is more concealed in dense Molinia-dominated vegetation but sacrificing some visual range (in red dashed circle).

Three overlapping incubation territories surrounding a pond system (marked with dashed red circle).
Curlews' avoidance of the major road (A31) compared to random distribution was also marked, despite being flanked by belts of apparently suitable nesting habitat. It is possible that this habitat is being additionally degraded in its suitability by the tracts of woodland which mirror the line of the road from one side, and the noise shadow created by traffic on the other (Supporting information). The effect decreased with the size of the road (A35/local roads) which create relatively lower levels of auditory disturbance as opposed to the four carriageways of the A31. Many species of bird are negatively impacted by road noise (Reijnen et al. 1996, Hirvonen 2001, Halfwerk et al. 2010, Blickley et al. 2012a, b2012b, Injaian et al. 2018, Slabbekoorn 2024). In ground-nesting species that are vulnerable to predation on the nest during incubation, sound that interferes with acute levels of hearing as a predator defence strategy may cause excessive vigilance behaviour, or simply compel them to abandon suitable habitat in favour of quieter areas which allow full perception and awareness of approaching threats (Barber et al. 2010). This is clearly an area which requires further investigation, and possibly trials of sound-reducing barriers alongside the road which could serve a dual function of pushing birds in flight up above the height of passing vehicles.
Finally, curlews in the New Forest showed a strong avoidance of woodland. The tendency for nests to be placed away from woods corroborated existing literature (Gunnarsson et al. 2006, Bertholdt et al. 2016, Holmes et al. 2020). Their avoidance of woodland even within 2 km buffers is notable in a landscape so heavily fragmented by trees as the New Forest. Birds are strongly selecting habitat that minimises their overall exposure to woodland, even in circumstances where habitat is limited and other factors such as disturbance may exclude them from otherwise suitable areas. A similar pattern was previously observed in Sweden where the median distance to forest edge was significantly greater than in randomly selected sites (Berg 1992), and UK-wide trends have been linked to reductions in curlew abundance with increased woodland cover (Franks et al. 2017). However, some studies found weak or no evidence for a link between woodland cover and curlew abundance (Douglas et al. 2014, Johnstone et al. 2017, Żmihorski et al. 2018, Zielonka et al. 2018). As with nesting habitats there may be wide inter-regional differences in avoidance of woodland, and curlew may vary in tolerance depending on additional pressures not accounted for by measuring just this correlate of distribution. In this study there is no clear relationship between woodland proximity and nest survival, suggesting the impacts of predation are more nuanced, and their avoidance of woodland might be driven by perceived presence of predators and impeded visual range. Inconsistent impacts of woodland on nest survival are found across other studies, from no effect (Berg 1992, Valkama et al. 1999, Zielonka et al. 2020), to an inverse relationship between tree cover and nest survival (Douglas et al. 2014).
The degree to which woodland impacted curlew distribution in this study has implications for the current agendas of tree-planting to mitigate impacts of climate change. If new woodland blocks are placed in open habitat being used by curlews, this may force them to nest in areas that are exposed to unsustainable pressures from other factors. Since a very marked exclusionary effect of woodland was observed even within 1–2 km buffers of a nest, this suggests a substantial open area around curlew breeding habitat must be spared from tree planting. This is a strong precautionary call echoed in other regions (e.g Iceland – see Pálsdóttir et al. 2022) where similar effects have been observed; and when taken into consideration alongside the multiple pressures already faced by breeding curlew, it is vital this avoidable concern is mitigated before they are lost as a breeding bird in this country.
Conclusion
Curlews were found to select their nesting habitat at varying spatial scales, which can inform management decisions made in areas where curlew breed. Whilst the ca 500 m area immediately surrounding the nest site appears to be critical in its characteristics, woodland was found to have an exclusionary impact up to 2000 m, a factor with important implications. Breeding curlews in the New Forest displayed a strong selection for wetter habitats for nesting, in particular mire. Since nest survival was higher in wet habitats than dry, provision of more of this habitat may confer some protection against predation, and disturbance from high visitor pressure, which is currently intensive at this and many other lowland sites. These findings support the extensive investment from Forestry England under the Higher Land Stewardship scheme to restore wetlands and reduce conifer blocks (Smith 2006, 2018), which will improve conditions for nesting curlew and other waders, as well as the wider assemblage of wildlife. The impact of the A31 trunk road on curlew distribution highlights a need for further investigation into the impacts of major roads – and perhaps other transport infrastructure such as railways – on breeding birds in the UK, as suitable open land for nesting becomes increasingly squeezed by human developments and tree-planting initiatives. Similarly, if it is traffic noise that deters curlews, careful thought should also be given to the siting of wind farms in open areas of suitable breeding habitat, where they may cause both an auditory and visual deterrent to nesting. Curlews' clear avoidance of woodland found in this study calls for caution to be exercised when new tree-planting initiatives are being planned, and realistic consideration be given to the risk of excluding curlews from high-value breeding habitat.
Acknowledgements
– Many thanks to Hampshire Ornithological Society, Simon Currie, and Russell Wynn and Marcus Ward of Wild New Forest for previous surveys and nest finding; Chris Heward for support catching and tagging curlews; and, importantly, the New Forest Keepers Austin Weldon, Lee Knight, Maarten Ledeboer, Jordan Thomas, Patrick Cook, Alan Stride and Sandy Shore; and Rupert Brewer from Bisterne Estate for all their support, cooperation and dedication through the last four years.
Funding
– Elli Rivers: PhD jointly funded by Game and Wildlife Conservation Trust (GWCT) and Bournemouth University; Mike Short: GWCT; Andy Page: Forestry England; Peter Potts: Independent wader researcher, Farlington Ringing Group; Kathy Hodder: Bournemouth University; Andrew Hoodless: GWCT; Richard Stillman: Bournemouth University. Tags have been funded by Hampshire Ornithological Society, Associated British Ports, Forestry England and GWCT.
Permits
– All fieldwork was carried out with landowner permissions from Forestry England and Bisterne Estate. Nest visits were conducted under license from the British Trust for Ornithology (BTO) Adult curlews were caught and GPS-tagged under a special methods license from the BTO, and SSSI consent from Natural England.
Author contributions
Eleanor Marie Rivers: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Project administration (lead); Writing – original draft (lead). Mike J. Short: Conceptualization (supporting); Investigation (supporting); Methodology (equal); Project administration (supporting); Resources (equal); Supervision (equal); Validation (equal); Writing – review and editing (supporting). Andy Page: Funding acquisition (supporting); Investigation (supporting); Methodology (supporting); Resources (supporting); Supervision (supporting). Peter M. Potts: Conceptualization (supporting); Funding acquisition (supporting); Investigation (supporting); Writing – review and editing (supporting). Kathy Hodder: Formal analysis (supporting); Methodology (supporting); Supervision (supporting); Writing – review and editing (supporting). Andrew Hoodless: Conceptualization (equal); Data curation (equal); Formal analysis (supporting); Funding acquisition (lead); Investigation (supporting); Methodology (equal); Project administration (supporting); Resources (lead); Software (supporting); Supervision (equal); Validation (supporting); Writing – review and editing (supporting). Rob Robinson: Formal analysis (supporting); Methodology (supporting); Supervision (supporting); Validation (supporting); Writing – review and editing (supporting). Richard Stillman: Data curation (supporting); Formal analysis (supporting); Funding acquisition (lead); Investigation (supporting); Methodology (supporting); Project administration (supporting); Resources (equal); Software (supporting); Supervision (supporting); Writing – review and editing (equal).
Open Research
Data availability statement
Data are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.1zcrjdg24 (Rivers et al. 2024).