Gardens can benefit pollinators living in surrounding farmland landscapes, but the reason for their value is not clear. Gardens are no different from many semi-natural farmland habitats in terms of the quantity of floral resources (pollen and nectar) they produce, but the timing of their resource supply is very different, which may explain their value. We show that gardens provide 15% of overall annual nectar in farmland landscapes in Southwest UK, but between 50% and 95% during early-spring and late-summer when farmland supplies are low. Gardens can therefore reduce seasonal nectar gaps experienced by farmland bumblebees. Consistent with this pattern, bumblebee activity increased in gardens relative to farmland during early spring and late summer. An agent-based model reinforces this point, showing that timing, not quantity, of garden nectar supply enhances bumblebee colony growth and survival in farmland. We show that over 90% of farmland in Great Britain is within one kilometre of a garden and therefore positive actions by gardeners could have widespread spillover benefits for pollinators across the country. Given the widespread distribution of gardens around the world, we highlight their important interplay with surrounding landscapes for pollinator ecology and conservation.

There are four datasets in this Dryad submission which are listed below along with information on how they were collected and processed:

1) Garden_nectar_production

How it was collected:

Garden floral abundance was recorded during 2019 in 59 urban gardens in the city of Bristol, South West UK. Although gardens were sampled in a different year to farmland, the mean annual temperature, rainfall and growing season of these two years were similar and previous data from this region show a similar pattern of nectar phenology between years (Timberlake et al. 2019). Residential gardens included the land adjacent to and associated with each domestic property (both front and back gardens), irrespective of whether it was paved, vegetated, or used as a driveway. In total, gardens ranged in size from 31.3 m² to 407.7 m² (mean 156.4 m² ±12.7 SE) and their selection was stratified by both geographical location and neighbourhood income (see Tew et al. 2022 for further details). From March to October 2019, each garden was visited once per calendar month to record floral abundance by counting the number of open floral units of each flowering plant species within or directly above each garden. In contrast to farmland habitats, sampling along a 50 m transect is not practical in residential gardens, which exist as small discrete patches of a single land use type. Instead, floral units were either counted individually using a handheld tally counter or estimated by sub-sampling (e.g. for flowering shrubs and trees and flower-rich lawns). The floral unit values were then multiplied by the mean floral sugar production of each species to obtain an estimate of the grams of sugar per unit area per 24-hour period for each garden. Values for the nectar sugar production of garden species were from Baude et al. (2016), Hicks et al. (2016) and Tew et al. (2021, 2022) who quantified the sugar production of all 636 flowering plant species present in these gardens using the methods described in the previous section. Variation in garden nectar supply stabilises at the point of around 20 gardens (Tew et al. 2022), suggesting that 59 gardens is a sufficient sample from which to characterise the average nectar profile for Bristol gardens. Furthermore, values of average garden nectar production show no significant differences between UK cities (Tew et al. 2021) suggesting the patterns recorded in our study are likely to be broadly representative of gardens at a national scale.

How it was processed:

For each study garden on each sampling occasion (monthly), values for the nectar sugar production of all species are summed and divided by the area of garden (in m2) to give the total nectar sugar production of the garden in grams per m2 per day. A separate column shows the number of square metres of garden that are typically present in a 1km2 area of farmland landscape so that values for each garden can be scaled up to predict how much sugar they would produce in a typical farmland landscape (see paper methods for more information).

2) Farm_garden_nectar_production

How it was collected:

Raw values of nectar sugar production per m2 of farmland (from Timberlake et al. 2019) and garden habitats (from Tew et al. 2022; see dataset above) in units of grams of nectar sugar/m²/day allowed us to characterise the phenology of nectar supply in high resolution for each landcover and to compare between them. A generalised additive model (GAM) was used to model this smooth, non-linear trend in nectar availability over time for each landcover (gardens and farmland). A thin-plate regression spline was used to model day of the year, with the degree of smoothing selected using the default generalised cross-validation method (Wood 2011). 

How it was processed:

The outputs from these models (grams of nectar sugar/m²/day) were multiplied by the area of each habitat in a 1 km² area of farmland landscape (including gardens which comprised 1.9% ±0.01 SE of our study landscapes), to compare their contributions to landscape-level nectar supply. 

3) BeeWalk_data_South-West

How it was collected:

To establish seasonal patterns of bumblebee activity in farmland and gardens, we analysed occurrence records of four common farmland bumblebee species (Bombus terrestris, B. pascuorum, B. lapidarius & B. pratorum) from BeeWalk transect data collected between 2008 and 2019 (Comont 2020). BeeWalk is a standardised monitoring scheme which collects data on the abundance and distribution of bumblebees in Britain. Fixed transects with clearly defined habitat composition are systematically surveyed by trained volunteer recorders once per month from March-October.

How it was processed:

We included only records from the South West region of the UK (counties of:  Bristol, Cornwall, Dorset, Devon, Gloucestershire, Somerset and Wiltshire) to minimise any confounding effects of latitudinal differences in phenology. Records were filtered to include transect sections whose primary habitat designation was strictly related to either farmland or garden. Any transect sections classified as primarily agricultural but secondarily garden (99 records in total) were removed to avoid ambiguity; this left us with a total of 493 km of purely farmland transect sections and 168 km of garden-associated transect sections. For each of the two habitat types (i.e., farmland and garden), we calculated the total number of worker and queen bumblebees recorded in each month of the year for each of the four bumblebee species (hereafter referred to as bumblebee activity). Male bumblebees were excluded from the analysis as their foraging patterns and phenology of activity are different from those of females. Total observations for each species were divided by the total length of transect recorded in each habitat type so that results were scaled by sampling effort. Activity values were therefore expressed as the mean number of bees recorded per kilometre of transect. For each month of the year, we divided bumblebee activity values in gardens by activity values in farmland (i.e. the ratio of garden: farmland bumblebee activity) to identify which periods of the year bumblebees were relatively more active in gardens, and vice versa.

4) BEE-STEWARD_RESULTS

How it was collected:

To test the influence of garden nectar supply on farmland bumblebee population dynamics and disentangle the effects of resource quantity (how much nectar) from the effects of resource phenology (when the nectar is available) we ran an in-silico experiment using the agent-based model ‘BEE-STEWARD’ developed by Twiston‐Davies et al. (2021) implemented in Netlogo. The model simulates the foraging behaviour, life history and colony growth of bumblebees foraging for nectar and pollen in a temporally-dynamic, spatially explicit, user-defined landscape. Bee foraging activity is modelled based on the detection probability and attractiveness of individual forage patches (influenced by their size and distance from nest sites), flower handling time (affected by corolla depth and resource depletion) and sugar concentration (this can be defined by the user). Colony growth is an emergent property of floral resource acquisition (as well as other environmental factors) and is based on empirical data (see Fig. S6 for an overview of model structure). BEE-STEWARD, and related models, have been widely used in pollination ecology research as they enable users to alter the spatial, temporal and nutritional features of floral resources (as well as other aspects of the environment) and investigate the resulting impacts on bee colony dynamics (see https://beehave-model.net/publications/).

Our model landscapes were based on 12 real circular farmland landscapes with a 1 km radius (3.14 km² area) in Southwest UK which were characterised and mapped by Timberlake et al. (2021) (Fig. S2 & S7). Each of these study landscapes had at least some garden present, ranging from 0.2 – 5.9% coverage, with a mean of 1.9% ±0.01 SE (Table S1). The quantity and phenology of nectar supply in each habitat was based on empirical data from our study (see methods above).

In each of the 12 model landscapes we ran four alternative simulations to test the importance of gardens. These were: 1) gardens present as normal; 2) gardens removed completely; 3) gardens replaced with pasture (the most common land use in this region, which could also be considered an analogue of a simple lawn garden); and 4) gardens replaced with an alternative landcover of identical total floral resource value but with the phenological pattern of pasture (i.e., the same total resources but distributed differently through the year), to test the role of phenology per se. All other features of the landscape, including pollen availability and nesting site availability, were kept identical amongst treatments. In real landscapes, there will inevitably be other population-limiting factors such as pollen availability, nesting sites, predation and weather. However, we focus only on the impact of nectar availability in otherwise benign conditions. Experiments were run with Bombus terrestris only and began with 500 queens, as is the default for BEE-STEWARD. Mean emergence date and standard deviation were kept consistent between all sites, using default values from BEE-STEWARD (01 April ±28 days SD).

We selected five different population response variables to provide us with a broad view of the mechanisms driving population change. These were: 1) colony density throughout the flight season (recorded on the 15^th day of each month); 2) maximum number of colonies produced per year; 3) total number of bees produced per year; 4) number of new queens produced per year; and 5) percentage spring queen survival. 

How it was processed:

The model was run 20 times for each treatment-farm combination (using a different random seed in each case), and for each simulation run we calculated the mean value for each response variable over the first five years of the simulation. The great majority of populations persisted longer than five years, so five years was taken as a cut-off for the period of relative population stability.