Trends in butterfly populations in UK gardens—New evidence from citizen science monitoring

Private gardens are recognised as potentially important refugia for butterflies. Yet little is known about how gardens might be contributing to butterfly conservation, as their restricted accessibility has meant that garden habitats are not well‐represented in traditional monitoring schemes. Garden BirdWatch (GBW) is the UK's largest structured bird survey, comprising over 25 years of weekly bird counts from more than 14,000 gardens, predominantly occupying suburban and rural locations. Since 2007, a subset of GBW participants have additionally recorded the weekly abundances of butterflies. Using data for 14 seasons (2007–2020) from 7971 gardens with consistent butterfly monitoring, we present the first garden‐specific, national trends for 22 widespread butterfly species (37% of all UK butterflies). Half of the species investigated increased significantly in abundance in gardens between 2007 and 2020. Conversely, only one species, Wall (Lasiommata megera), showed a marginal reduction, though this change was not statistically significant. A strong, positive association between these new, habitat‐focused trends and those for UK butterflies more broadly, previously reported by the UK Butterfly Monitoring Scheme (UKBMS), indicates that patterns of abundance in gardens are largely a reflection of the changes that are occurring nationally. However, butterflies do appear to be faring better in gardens compared to the wider landscape. Averaging trends across non‐migratory species revealed that GBW recorded significantly greater increases over time than UKBMS. Effective monitoring of butterflies in gardens can produce reliable and informative population trends, and it provides important evidence of the significant role gardens play in sustaining butterfly populations.


INTRODUCTION
Recent global analyses have drawn attention to the severe declines suffered by many invertebrate populations around the world (e.g., Dirzo et al., 2014;Hallmann et al., 2017), bringing conservation concern for these taxa into the public consciousness for the first time (Didham et al., 2020;Harvey et al., 2020).Almost all reported invertebrate declines have been linked to some form of habitat loss or degradation, particularly those associated with agriculture and urban development (Jones & Leather, 2012;Wagner, 2020).Privately owned, residential gardens could alleviate some of the pressures facing native biodiversity within the wider matrix of intensively managed and developed land that covers much of the UK's land surface, potentially offering a refuge for many species of invertebrate (Baldock et al., 2015;Goddard et al., 2010).However, there remains a paucity of evidence of the role gardens play in supporting invertebrate populations.
The richness, diversity and structural variability of plant life occurring in gardens are likely to be especially important for insects (Anderson et al., 2020;Samways et al., 2020), including butterflies (Fontaine et al., 2016;Ramírez-Restrepo & MacGregor-Fors, 2017).Within the UK, gardens are a major contributor to overall urban green space.For example, estimates for Great Britain indicate that residential gardens cover an area totalling approximately 521,872 ha and account for 29.5% of total urban land (Office for National Statistics, 2019).In more rural areas, gardens can similarly provide a focal source of nesting and foraging resources in comparison to the surrounding, largely homogeneous, agricultural landscape (Osborne et al., 2008;Samnegård et al., 2011).Even in suburban and urban gardens, insect communities are likely to be greatly influenced by nearby agricultural practices, since enclosed farmland is the biggest contributor to urban land cover (Office for National Statistics, 2019).To this end, calls have been made to consider gardens, and their owners, as vital components in the fight to stem butterfly population declines in the UK (Fox et al., 2015).Environmental non-governmental organisations, conservation activists and the mainstream media have inspired the growing wildlife gardening movement to improve provisions for garden wildlife, particularly butterflies (e.g., Butterfly Conservation, 2021; Goulson, 2019;The Wildlife Trusts, 2021).It is now essential that the scientific community builds the evidence base for determining whether, and how, gardens are contributing to butterfly conservation through long-term monitoring of butterfly populations in gardens.
Understanding changes in species abundances through time and along ecological gradients is a fundamental component of biodiversity conservation (Jetz et al., 2019;Pereira et al., 2013) and typically relies on volunteer-based monitoring schemes (citizen science), in order that data can be collected at large scales (Chandler et al., 2017).The UK is fortunate to hold some outstanding long-term data on butterfly numbers in the form of the UK Butterfly Monitoring Scheme (UKBMS), which has been used to produce robust and detailed long-term trends for 59 species of butterfly since 1976 (Fox et al., 2015;Middlebrook et al., 2021).However, UK butterfly monitoring has tended to focus on semi-natural and wider countryside habitats, such as grasslands, woodland and farmland (Brereton et al., 2011;Brereton et al., 2020).
While a subset of UKBMS sites is representative of urban habitats (Dennis, Morgan, Roy, & Brereton, 2017), there is a gap in the standardised monitoring of butterflies in gardens specifically.
Monitoring of garden biodiversity is challenging, as private ownership prohibits access by experienced surveyors to carry out conventional surveys.However, gardens are amenable to year-round monitoring by their owners, who are uniquely placed to develop a detailed knowledge of their gardens and the wildlife that they contain.
Harnessing this potential into structured, long-term monitoring has already proven successful for birds and mammals in UK gardens via the British Trust for Ornithology's (BTO) Garden BirdWatch (GBW) programme (Cannon et al., 2005;Toms & Newson, 2006), and for moths via the Garden Moth Scheme (Wilson et al., 2018), as well as through similar initiatives in other parts of the world (e.g., Maes et al., 2021;Veerman, 2002;Wells et al., 1998).Butterflies are a charismatic species group, readily attracting the interest of volunteer recorders (Asher et al., 2001).Indeed, mass-participation surveys involving members of the public opportunistically counting butterflies in gardens (e.g., Big Butterfly Count http://www.bigbutterflycount.org,Garden Butterfly Survey https://gardenbutterflysurvey.org), have received good levels of uptake and produced encouraging results (Dennis, Morgan, Brereton, et al., 2017).Therefore, it is feasible that garden owners could similarly be recruited to carry out repeated, standardised sampling of garden butterflies in order that credible garden population trends can be produced, complementing those produced for the countryside as a whole.
In 2007, the GBW was extended, giving volunteer surveyors the option to contribute weekly records of butterfly numbers in their gardens throughout the year, in addition to the records for the survey's primary target taxon, birds.Taking advantage of an established network of survey sites and volunteers, by adding-on secondary taxa to an existing structured monitoring scheme, can be an effective way of producing population trends for less well-sampled species groups (Massimino et al., 2018).
In this article, we examine the value of long-term butterfly monitoring in gardens through the GBW programme, by using the GBW data to estimate species population trends for UK gardens and by then comparing these with the butterfly population changes occurring simultaneously in the countryside more broadly.Specifically, we use established analytical methods, developed for the UKBMS (Dennis et al., 2016), to produce garden-specific trends in the relative abundance of 22 species of butterfly between 2007 and 2020.Then, using phylogenetically controlled analyses, we compare trends in gardens from GBW with the latest published whole countryside trends for the same suite of species and the same window of time (10 years; 2011-2020).In the discussion, we critically evaluate the reliability of the trends produced and explore the possible underlying causes of the patterns that we observe.

Butterfly monitoring
GBW is a structured, citizen science monitoring programme, established in 1995, wherein participants submit weekly presence and maximum counts of birds, with the option also to record a range of other wildlife, including mammals, amphibians, reptiles, bumblebees and butterflies.Butterfly recording was introduced as a standard component of GBW monitoring in 2007.Participants can simply record the butterfly species that are present, or they can also include the maximum number observed at any one time per species in a given week.We have focused our investigation exclusively on the records that include butterfly maximum counts, detailed below, in order that changes in abundance can be estimated.The majority of observations are collected through a dedicated JAVA-based web application, with built-in validation procedures, but paper recording forms are also available.
GBW involves some 14,000 participants from across the UK.The distribution of participants is closely linked to the pattern of human population density nationally, with greater levels of participation in the south-east of England than in northern Scotland, but the levels of coverage remain sufficient to produce robust measures of garden use by birds at the national or regional level (Cannon et al., 2005).Participants are asked to define a recording area within their garden, and to record the wildlife using this area within their self-defined weekly recording period.While participants differ in the amount of time that they devote to monitoring their garden each week, they are instructed to maintain a consistent level of monitoring effort from one week to the next, and to discard data from weeks that are under-or overrecorded.Thus, while geographical sampling coverage is neither random nor systematic, recording of presences and counts is structured and standardised.Although there will inevitably be some degree of variation in the amount of time spent by participants and, additionally, in their level of competence in counting and species identification, we contend that the use of site effects in data analysis to account for site-level variation in recorded counts, together with the very large sample size, facilitate the derivation of robust metrics of abundance change.

Data selection
We established data selection criteria to maximise the inclusion of available data across species, without compromising data quality, since the analytical approach described below is demonstrated to have high power to detect trends from monitoring data with reduced sampling effort (Schmucki et al., 2016).
Given that butterfly monitoring is not a mandatory component of GBW and that zero counts have not been recorded, it was necessary to identify genuine butterfly survey sites (where butterflies were consistently monitored) from within the full cohort of GBW gardens.It is reasonable to assume that garden sites conducting weekly butterfly counts may not observe butterflies in every week, due to their relatively small size (compared to a typical UKBMS monitoring site, for example) and the influences of phenology, mobility and habitat suitability on patterns of garden use.To this end, we retained sites that contributed five or more weeks of non-zero butterfly counts in a given season for analysis.Consequently, all sites submitting ad hoc records, which may not be a reliable measure of actual butterfly identity or abundance, were excluded.On average across years, the sites retained for analysis recorded data in 29.99 weeks (± 6.30 SD, median = 32), 86% of the survey season, with an average of 16.12 weeks (± 7.45 SD, median = 16) of non-zero butterfly counts.
We used Land Cover Map 2015 data (Rowland et al., 2017a(Rowland et al., , 2017b)), summarised as percentage covers of 21 land cover classes at a 1-km square scale, to ascertain the landscape coverage of sampled sites.
LCM2015 land cover classification is based on the UK Biodiversity Action Plan Broad Habitat definitions, and we considered the full suite of possible classes (see Table S1) to allow for a comprehensive assessment of the wider landscape context of the sites contributing to the results.Sites were assigned data for the 1 km in which they were situated, except for sites in the Channel Islands for which no data were available, in order that summaries of land cover type presence, dominance and percentage cover could be calculated.
Extreme outlying weekly counts, considered to be unrepresentative of broad-scale patterns of abundance and/or the result of surveyor error, were also removed using protocols adapted from Plummer et al. (2020).
Specifically, for species where the ratio of maximum to median count was greater than 20, all counts greater than twice the value of the 99th percentile were excluded (<0.2% of observations).
Data for species that had been recorded in at least 15 sites each season from 2007 to 2020 were retained for the analysis of species trends.This excluded less common, habitat-specialist species that are not major users of gardens.The distributions of the GBW sites included in the analysis for each individual species, and overall, were mapped at a 10-km 2 scale and cross-referenced with known species distribution patterns obtained through systematic recording (Fox et al., 2015) to ascertain the spatial representativeness of the data.

Annual abundance indices
Annual collated indices of relative butterfly abundance (collated indices hereafter) were estimated for all species following the Generalised Abundance Index (GAI) method developed by Dennis et al. (2016), and implemented using the rbms package (Schmucki et al., 2021) in R version 4.1 (R Core Team, 2021).In summary, for each species, first, we estimated the seasonal flight curve for every sampling year separately using a Poisson generalised additive model (GAM), with weekly counts fitted as a function of a site effect and a seasonal spline smooth.Flight curves were cross-referenced with published reports of seasonal butterfly abundance patterns (Butterfly Conservation, 2022) to further verify data reliability.Second, using a concentrated likelihood approach (Dennis et al., 2016), the flight curve was used to interpolate missing weekly count data in order that annual indices of abundance could be estimated for each site.Then, third, annual collated indices (the mean relative abundance of butterflies across sites per year) were computed using a Poisson generalised linear model (GLM) in which annual site indices (estimated in the previous step) were fitted as a function of site and year effects.The GLM was weighted by the proportion of the flight curve monitored per site and year (Middlebrook et al., 2021), such that sites with complete monitoring had a greater influence on the final indices than sites with multiple interpolated counts.This third step was repeated for 1000 bootstrap samples drawn from the annual site indices, with replacement, to derive 95% bootstrap confidence intervals around collated indices.For the benefit of computational efficiency, the bootstrap did not repeat the estimation of seasonal flight curves.Collated indices, and associated 95% confidence intervals, were transformed on to a log 10 scale, for comparability with UKBMS (Middlebrook et al., 2021), and centred relative to a value of 2.0 (100%) in the first year (2007).

Population trends
Log 10 (collated indices) were modelled as a function of year using generalised least squares (GLS) regression to calculate population trends over two timeframes, 2007-2020 (the complete time series of GBW butterfly monitoring) and 2011-2020 (the 10-year trend).The 10-year trends explicitly match the metrics routinely reported by other population monitoring schemes, including UKBMS.GLS is appropriate for time series estimations because it can address residual temporal structure in the data (Zuur et al., 2007).Examination of autocorrelation plots for species-specific model residuals found no evidence that temporal autocorrelation needed to be accounted for within the model structure.The coefficient of the year effect from each GLS model was used to derive the overall population trend for each species and timeframe, respectively, defined as the percentage change in the collated index between the first and last year in the time series.GLS modelling was replicated across all bootstrap samples to derive 95% confidence intervals around the trends.

Comparison of garden and UK-wide trends
Phylogenetic comparative analyses (Grafen, 1989) were used to examine the relationship and overall difference between the GBW 10-year trends and the trends reported by the UKBMS for the same time period and suite of species (Middlebrook et al., 2021).The UKBMS was established in 1976 and encompasses weekly volunteer data collection throughout the summer at over 3000 sample locations across the UK.The scheme comprises three methods of butterfly abundance sampling; traditional all-species butterfly transects (Pollard walks), Wider Countryside Butterfly Survey (WCBS) transects and species-specific reduced effort surveys.
Data are collated across all three survey components to derive the UKBMS national butterfly population trends.
A phylogenetic generalised least squares (PGLS) model was used to examine the association between the GBW and UKBMS 10-year trends, and a phylogenetic paired t-test was used to test the difference, implemented using the caper and phytools R packages respectively (Orme et al., 2018;Revell, 2012).Using phylogenetic comparative analyses allowed for the non-independence of related species, with respect to their life histories and responses to environmental gradients, to be accounted for within the modelling framework.Evidence of phylogenetic signal in the modelled relationships was estimated using Pagel's lambda, a measure of the phylogenetic covariation between the predictor and response which varies between zero and one.High Pagel's lambda values indicate a strong similarity in the predictor-response relationship between closely related species.A phylogenetic tree for the species of interest (Figure S1) was pruned from the majority rules consensus tree for European butterflies, developed by Wiemers et al. (2020).
Analyses were first conducted with all species considered, and then repeated with regular migrants (Painted Lady Vanessa cardui, Red Admiral Vanessa atalanta) removed.The all-species comparison was used to evaluate how closely patterns of change in gardens reflect changes in species abundances in the UK countryside more broadly.Since migrant numbers are not dependent on how the UK countryside is managed, the nonmigrant comparison was used to better evaluate whether trend differences could be explained by garden versus countryside habitat variation.

RESULTS
There were 7971 sites included in the analysis, distributed across the UK, Isle of Man and the Channel Islands, which were selected due to being monitored for butterfly abundance in five or more weeks in a season between 2007 and 2020 (Figure 1), resulting in 872,859 weekly observations from which inference could be drawn.On average, the most abundant land cover type within the 1 km 2 where sites were located (excluding sites in the Channel Islands) was suburban (Table S1; mean ± SD % coverage = 39.6 ± 29.1%), followed by improved grassland (27.6 ± 24.0%), and arable and horticulture (16.4 ± 23.1%).Only 51.6% of sites were in locations where the dominant land cover was either urban or suburban (Table S1).The number of sites monitored annually increased over time, from 749 in 2007 to 5113 in 2020 (Table S2).However, the overall spatial extent of monitoring activity remained largely unchanged, with sites increasing in density rather than spatial distribution across years (Figures S2 and S3).
There were sufficient data to estimate garden-specific population trends for 22 species of butterfly, 37% of the 59 species that occur regularly in the UK (n = 514-6763 sites per species; see Table S2 and Table S3 for data descriptive statistics).The spatial representations of species' records reflect their known distributions (Figure S4), and estimated flight curves reflect known seasonal patterns (Figure S5).
Abundance in gardens increased significantly for 11 species (50%) between 2007 and 2020 (Table 1, Figure 2).A further 10 species had positive, but non-significant, trends and only one species, Wall Lasiommata megera, was found to have a marginally negative trend, although this was also not significant (p = 0.313).
Marbled White Melanargia galathea and Large Skipper Ochlodes sylvanus showed the largest changes, both increasing by more than 200% (Table 1, Figure 2).A further five species (Holly Blue Celastrina argiolus, Small Skipper Thymelicus sylvestris, Ringlet Aphantopus hyperantus, Brimstone Gonepteryx rhamni and Orange-tip Anthocharis cardamines) doubled in abundance over the 14-year survey period.More moderately positive trends, of between 44% and 80%, were observed for Meadow Brown Maniola jurtina, Small White Pieris rapae, Green-veined White Pieris napi and Gatekeeper Pyronia tithonus.All species showed a high degree of inter-annual variation (Figure 2), contributing to the large confidence limits around species' trends (Table 1).
Comparison of the 10-year trends (2011-2020) from GBW with those similarly derived from UKBMS revealed a strong association between garden and UK-wide trends (r 2 = 0.899, F 1,20 = 177.8,p < 0.001, Figure 3a).A strong phylogenetic signal in the PGLS model residuals (λ = 0.952, p = 0.043) suggests that this relationship between the two trends is influenced by similarities among closely related species.Across all species, the average increase in abundance appeared to be greater in gardens (88.5 ± 20.4% SE) compared to the whole countryside (67.1 ± 27.0% SE), although this difference was not significant (t 19 = 1.49, p = 0.152, Figure 3b).Repeating the analysis, with the exclusion of regular migrants, resulted in a reduction in the strength of the association between the two trends (r 2 = 0.235, F 1,18 = 5.52, p = 0.030, Figure 3c).Furthermore, on average, increases in abundance were significantly greater for non-migrants in gardens (73.4 ± 12.2% SE) compared to in the countryside (41.1 ± 9.8% SE; t 17 = 2.12, p = 0.049, Figure 3c).

DISCUSSION
Our findings demonstrate that structured, volunteer-based data collection in private gardens can be used to produce viable trends for many of the UK's most widespread butterfly species, filling a gap in habitat coverage that exists across other national butterfly monitoring schemes.Butterfly numbers have increased in gardens since 2007 for 21 of the 22 species examined, with half of these species showing moderate to strong, statistically significant increases in abundance over time.Across species, the patterns to date appear to largely reflect the systematic changes that are occurring nationally.However, with many species showing larger increases in gardens than in the countryside, our findings suggest that garden habitats could be making a valuable contribution to conserving butterflies.We explore the likely validity and significance of our findings in more detail below.

Strengths and limitations of the monitoring approach
Concerns about the quality of the monitoring data underpinning highprofile reports of global insect declines have led to numerous calls for more comprehensive insect monitoring (e.g., Desquilbet et al., 2020;Habel et al., 2019;Wagner, 2020;Warren et al., 2021).Here, we present standardised, long-term data in a previously under-sampled, but conceivably important, habitat.As an optional extra for GBW participants, butterfly monitoring has become increasingly popular since its introduction in 2007 (see Table S2).Collecting butterfly data as part of an established bird monitoring programme has a number of T A B L E 1 The magnitude and significance of the population trends (% change in relative abundance) for 22 species of butterfly recorded in British Trust for Ornithology's (BTO) Garden BirdWatch sites between 2007 and 2020.Note: Confidence limits refer to bootstrapped 95% confidence intervals.The significance of the trends and associated test statistics are estimated using generalised least squares (GLS).The direction of change gives an overall summary of trend magnitude and significance, where strong = significant changes ≥ (±) 100% and moderate = significant changes of < (±) 100%.
advantages compared to initiating a new, stand-alone survey (Massimino et al., 2018).Most notably, it means that butterfly counts inherit the core benefits of the underlying GBW scheme-specifically, ongoing weekly recording via standardised protocols across a large number of privately owned and nationally distributed sites.Weekly recording of multiple taxa could be particularly worthwhile in improving the reporting of zero counts for the non-target or secondary taxa (Massimino et al., 2018), in this instance helping to deliver repeated, unbiased sampling of butterfly species presence/absence and abundance across the whole flight season.To this end, the sampling effort (within and across sites) underpinning the GBW butterfly dataset is well-suited to the modelling of species' phenologies and the production of credible long-term population trends (Schmucki et al., 2016;Wagner, 2020).The trends derived from the GBW data should, therefore, be a meaningful source of comparison against the established, national species trends derived from the UK's Butterfly Monitoring Scheme (Brereton et al., 2020).
We have demonstrated a high level of agreement between the 10-year trends we produced using GBW and those reported by the UKBMS, adding further support to the validity of GBW butterfly data.
The strong positive association evidenced by the all-species phylogenetic GLS model suggests that changes in butterfly abundance in gardens over time are reflective of broader, national patterns of widespread species.
Furthermore, closer inspection of inter-annual variations shows that GBW annual abundance indices have detected known events that have impacted on butterfly numbers in the UK in recent years.For example, the widely recorded 2009 and 2019 influxes of Painted Lady (e.g., Stefanescu et al., 2013), a migratory species which overwinters in North Africa (Asher et al., 2001), are also reflected in the GBW trend.The peaks in GBW trends in 2009 for Comma, Green-veined White and Speckled Wood had also been recorded by the UKBMS, with these species registering the highest index of their long-term trend up to that point (Botham et al., 2009).Conversely, the acutely low numbers of many species in gardens in 2012 (including Large White, Small White, Peacock and Common Blue) correspond with similarly poor indices in UKBMS, where 2012 was reported as the worst year in the, then, 37-year time series (Botham et al., 2013).Most species increased in number the following year, 2013, both in gardens (Figure 2) and in countryside habitats (Brereton et al., 2014).Although comparisons with UKBMS are not definitive evidence of the reliability of the GBW-derived indices, these similarities do support the interpretation that our findings accurately represent real patterns occurring in sampled gardens.Furthermore, though, the evidence of overall differences in the trends for non-migrant species suggests that we are also detecting the influence of additional ecological processes on garden butterfly numbers, over and above those driving major, widespread patterns.
While there are strengths associated with using GBW to estimate indices of butterfly abundance, the potential limitations of this monitoring approach should be evaluated.A common criticism of volunteer-based biodiversity monitoring, as employed by GBW, is the potential for inaccuracies in species records (Dickinson et al., 2010), which may also be further compounded for species that are not the survey's primary target.Although observer quality is not formally controlled per se, modelling protocols have accounted for count anomalies and inter-site differences; there is also no reason to suspect that trend results are compromised by temporal changes in data accuracy.
Restricting the analysis to gardens where there has been a consistent commitment to recording also excludes more casual, occasional submissions, which might be expected to be less reliable.Consequently, there was no evidence that species were being mis-identified outside of their expected range.Furthermore, the large sample sizes collected by GBW participants will have helped to minimise any influence of sampling error on modelling outcomes (Dickinson et al., 2010;Wauchope et al., 2019).Indeed, even species appearing in our study that are not traditionally considered to be common garden visitorsincluding Small Heath Coenonympha pamphilus, Marbled White, Wall, Small Skipper and Large Skipperwere independently observed at more than 500 sites.It is possible that the use of weekly monitoring by GBW will have increased the likelihood of capturing less common garden visitors, compared to alternative, less frequent monitoring strategies.
Another obvious concern is that GBW allows for the selfselection of sites, resulting in non-random temporal and spatial distributions of monitoring locations.Such biases could render unreliable abundance trends if species' ranges are not being sampled effectively or equally through time and space.However, we found no temporal shifts in the geographical coverage of GBW-monitored gardens, suggesting that temporal bias is not an issue that affects the outcomes of this study.Furthermore, cross-referencing of GBW butterfly sampling patterns with known UK distributions suggests that the spatial ranges of the species examined are well represented by the data.Spatial variation in sampling effort clearly also influences the representation of habitats and regions in outputs, but a bias towards areas with higher human population density and, by association, more gardens, as is the case for GBW, arguably provides an intrinsic, desirable stratification in the derivation of garden-specific trends.Garden sites included in this study were predominantly located in suburban and rural areas, where gardens tend to be larger and within the dispersal range of seminatural habitats, rather than in heavily urbanised areas.There is also likely to be a bias towards wildlife-friendly gardening practices within sampled GBW sites, compared to an average UK garden, since participation in environmental citizen science programmes can promote environmental stewardship (McKinley et al., 2017).This could further explain why we were able to produce trends for species more commonly associated with countryside habitats, including those of current conservation concern, such as Wall and Small Heath (Fox et al., 2022).
Further assessment to determine how representative GBW sites are of UK garden habitats more broadly would be of value in helping to fully contextualise GBW-derived findings.
There are some simple measures that could be taken to improve GBW data collection, aligning it more closely with established butterfly monitoring best practices.For example, (1) confirmation that weather criteria are met would enhance data validation, (2) options to submit and verify images of rare observations could reduce instances of species misidentification, and (3) adding date of recording, rather than using monitoring week, would increase the accuracy of abundance modelling and would generate absences, eliminating the need for these to be inferred.Nonetheless, with large numbers of sampled gardens, covering a broad cross-section of variability in garden localities and habitat types, we can be confident that GBW butterfly data, in its current form, can deliver useful, reliable inferences about populations.

The importance of gardens for butterflies
Our study focused on 22 of the most common and widespread butterfly species in the UK, many of which have increased in abundance in recent years (Brereton et al., 2020;Fox et al., 2015).Indeed, we similarly report positive 14-year trends for all but one of these species, Wall, in gardens.These findings contradict the broader long-term declines documented for many insects globally (e.g., Dirzo et al., 2014;Hallmann et al., 2017), including UK butterflies (e.g., Fox et al., 2015).
However, they are consistent with other reported increases for ecologically generalist species in recent years, which are more adaptable to human-dominated environments (Wagner, 2020) and are thought to be benefiting from the positive aspects of climate warming (MacGregor et al., 2019).Most of the species considered here are relatively mobile, have broad habitat and dietary requirements, and/or are multivoltine (Cook et al., 2021;Cowley et al., 2001), all traits associated with lower magnitudes of species declines (Wagner, 2020).
Gardens have the potential to enhance existing, or to provide new, habitat for butterflies by providing nectar sources for adults and, to a lesser extent, food plants for their caterpillars (Gaston et al., 2005).The mobile nature of many butterfly species means that trends in gardens may be indicative of patterns occurring at wider spatial scales.Indeed, the similarities between GBW and UKBMS trends we detected do suggest that butterfly numbers in gardens are not independent of wider, national population changes.However, discernible species-level differences in the overall scale of population changes between the two schemes suggest that butterflies are responding somewhat differently to environmental stimuli in gardens compared to in the wider landscape, potentially indicating the importance of garden habitats in providing resources for common, widespread butterflies, and possibly other insects by association (Thomas, 2005).
Butterflies appear to be faring better in gardens, with more positive trends being recorded by GBW than UKBMS, on average, across species.It was particularly illuminating to detect large increases in several species that are not typically associated with gardens, notably Large Skipper, Small Skipper and Marbled White.All three species have expanded in range in the UK in recent decades, probably in response to climate warming (Brereton et al., 2020).However, they were among several grassland species with noticeably greater gains in gardens over the past 10 years than elsewhere: others include Small Heath, Small Copper Lycaena phlaeas and Common Blue Polyommatus icarus.This raises an important question as to whether garden butterfly numbers could be contributing to population growth, at least for some species; some may be searching for nectar resources (as opposed to finding breeding habitat) or simply being reported in passing during dispersal, when the weather or resources elsewhere increase movements.Certainly, in France, butterfly-friendly gardening practices have been shown to mitigate the detrimental impacts of urbanisation on butterfly numbers at a local scale (Fontaine et al., 2016), and a growth in wildlife gardening in the UK may similarly have contributed to the butterfly abundance patterns detected in our study.It is also noteworthy that the trends we present contrast with the negative urban trends reported for UK butterflies previously for the period 1995-2014 (Dennis, Morgan, Roy, & Brereton, 2017).
Comparisons between these trends should be undertaken with caution due to the difference in the timeframes under investigation.
Nonetheless, the contrast is perhaps suggestive that gardens, specifically, could be a valuable refuge habitat within the wider urban landscape or that they may be becoming more important as surrounding urban land is built over.Importantly, the urban trends, derived from UKBMS (Dennis, Morgan, Roy, & Brereton, 2017), are based on transect routes that do not have direct access to private gardens, and instead are likely to almost exclusively sample alternative, public, urban green areas such as parks, road verges and cemeteries.Closer examination of the interdependence of garden, urban and countryside trends could be an important next step in determining the contribution garden habitats are making to the large-scale changes in UK butterfly populations.
Although the GBW scheme does not restrict the list of species that participants can record, we found that the number of non-zero counts reported for other, less common species did not meet our criteria for inclusion in the modelling of trends (i.e., those species are less likely to be observed in gardens and so GBW does not, at the time of this study, hold enough data for them).This has important implications for how we interpret and draw conclusions from our findings.While gardens appear to offer benefits to common, generalist butterflies, we know little about their utility for more specialist and more vulnerable species.By definition, gardens are unlikely to meet the critical requirements of seminatural habitat specialists, which comprise half of all UK butterfly species.However, by providing suitable nectar resources, they could still act as stepping stones for dispersal within functioning metapopulations.
Further research addressing garden butterfly community composition and its association with garden characteristics and wider landscape features would be especially informative in interpreting the value of gardens across the full range of UK butterfly species (Ellis & Wilkinson, 2021).An assessment of the extent to which ecological, morphological and life-history traits influence garden use could further evidence the potential role of gardens for butterfly conservation in general.
While the populations of the 22 species for which we were able to generate garden trends may not appear be undergoing declines at present, population status can change rapidly and unexpectedly in response to a host of environmental and anthropogenic drivers, such as disease or land-use changes (Robinson et al., 2010).Indeed, many formally abundant and widespread species are now in decline globally (Van Dyck et al., 2009;Wagner, 2020).Furthermore, many of the butterflies that have suffered the most severe long-term declines in the UK (since the 1970s) are also generalist species, including three species examined in the present study (Wall, Small Skipper and Small Tortoiseshell) (Fox et al., 2015).It is therefore prudent to monitor, and to understand, the drivers of population trends for common species.
Long-term monitoring is especially important for butterflies, since high inter-annual variability for this taxon, as reported here, can mask small population changes if data are not analysed over a sufficiently long timeframe (Didham et al., 2020;Fox et al., 2019;Habel et al., 2019).
Short-term population trend assessments are also sensitive to start and end-year values (Didham et al., 2020), therefore greater insight about the status of garden butterfly numbers is likely to be revealed

Concluding remarks
This study shows that the GBW programme can offer a robust and reliable means of capturing long-term data on butterfly numbers in gardens, providing an important addition to the existing suite of monitoring schemes for UK butterflies by covering a habitat that has previously been under-represented, but may be of growing importance.
Population trends to date have not differed hugely from those in the We have presented compelling evidence that gardens are a widely and frequently used habitat for many butterfly species.In particular, the detection of marked increases in several species considered to be uncommon garden visitors, such as Marbled White and Large Skipper, could be suggestive of novel range spread into gardens and supports earlier calls to recognise the potential contribution that gardens could make to butterfly conservation efforts in the UK (Fox et al., 2015).Even if garden use merely reflects transient individuals, garden monitoring could be equally important in providing a sensitive barometer for the effects of wider environmental change.In either case, it would appear that gardens represent a valuable resource for butterflies, particularly within highly urbanised and intensively managed, agricultural landscapes.
Importantly, our findings exemplify how actions implemented within individual gardens can scale-up to influence populations much more broadly (Goddard et al., 2010).Garden management decisionsincluding flowering plant number and diversity, and the presence of woody plants (scrubs and trees)-have been shown to have a strong positive influence on butterflies and other pollinators (Majewska et al., 2018).Our findings suggest, therefore, that individual garden owners have a role to play in protecting and enhancing UK butterfly populations through their gardening choices.With garden size and connectivity to natural sites also key to the functioning of gardens as havens for butterflies (Majewska et al., 2018;Majewska & Altizer, 2020;Olivier et al., 2016), there is equally a responsibility for developers and landscape planners, local authorities and policy makers to recognise the ecological importance of gardens as part of their decision-making.

F
I G U R E 1 Distribution of the 7971 BTO Garden BirdWatch sites included in the study, summarised by 10 km 2 .

F
I G U R E 2 Trends in relative abundance [log 10 (collated index)] for 22 butterfly species between 2007 and 2020.Annual indices are centred around a value of 2.0 [log 10 (100%)] to the first year, and grey shading around inter-annual changes corresponds to bootstrapped 95% confidence intervals.Linear trends are coloured according to significance (solid red lines indicate significant changes, broken blue lines for non-significant), with coloured shading corresponding to bootstrapped 95% confidence intervals.

F
I G U R E 3 Comparison of species 10-year trends derived from GBW (gardens) and UKBMS (whole countryside) for 2011-2020.(a) Linear association in the trends from the two schemes, where each data point represents an individual species and the GBW trends are accompanied by bootstrapped 95% confidence intervals.(b) Difference in the mean trend across species per scheme, presented using a box and whisker plot overlaid with individual species pairwise differences.(c) Linear association for trends of <200% (grey region of 3a) with species labels.
as the time series used to generate GBW trends is extended.Trends in gardens, specifically, could provide a useful indicator of broader patterns of population change.As a secondary habitat for most species, density dependence theory would suggest that a declining butterfly species should move out of gardens before reducing in number in their preferred habitat.In this scenario, a decline in the garden trend could act as an early warning for more concerning landscapescale changes in population status.Conversely, butterfly numbers might increase in gardens where the availability and/or suitability of their preferred habitat is compromised, for example due to land-use changes, disturbance or habitat fragmentation.Given the widely acknowledged adverse impacts of agricultural intensification onLepidoptera (Sánchez-Bayo & Wyckhuys, 2019;Wagner, 2020), patterns such as this may well be encountered in rural gardens, where local butterfly populations are under pressure from changing agricultural practices in the surrounding landscape.Continuing to monitor, and to evaluate trends in, butterfly numbers across a spectrum of garden habitat contexts could help towards improving our understanding of the drivers of population change.
countryside as a whole, but this could change over time and effective monitoring needs to be in place to inform about such changes.Trends derived from GBW data complement those from other ongoing butterfly monitoring and could usefully be integrated into regular reporting to enhance inference about patterns and causes of change.But furthermore, with the assurance that gardens can deliver a good impression of overall population trends for common species, it may even be worthwhile considering the potential for garden butterfly counts to replace traditional, transect-based monitoring for some of the UK's most widespread species.Although this might reduce understanding of causes of change for those species, feasibly, it could benefit species of conservation concern by allowing existing survey effort to be redirected and used to enhance their monitoring instead, in turn strengthening conservation prospects.