Citizen science data reveals the need for keeping garden plant recommendations up-to-date to help pollinators

Widespread concern over declines in pollinating insects has led to numerous recommendations of which “pollinator-friendly” plants to grow and help turn urban environments into valuable habitat for such important wildlife. Whilst communicated widely by organisations and readily taken up by gardeners, the provenance, accuracy, specificity and timeliness of such recommendations remain unclear. Here we use data (6429 records) gathered through a UK-wide citizen science programme (BeeWatch) to determine food plant use by the nations’ bumblebee species, and show that much of the plant use recorded does not reflect practitioner recommendations: correlation between the practitioners’ bumblebee-friendly plant list (376 plants compiled from 14 different sources) and BeeWatch records (334 plants) was low (r = 0.57), and only marginally higher than the correlation between BeeWatch records and the practitioners’ pollinator-friendly plant list (465 plants from 9 different sources; r = 0.52). We found pollinator-friendly plant lists to lack independence (correlation between practitioners’ bumblebee-friendly and pollinator-friendly lists: r = 0.75), appropriateness and precision, thus failing to recognise the non-binary nature of food-plant preference (bumblebees used many plants, but only in small quantities, e.g. lavender—the most popular plant in the BeeWatch database—constituted, at most, only 11% of records for any one bumblebee species) and stark differences therein among species and pollinator groups. We call for the provision and use of up-to-date dynamic planting recommendations driven by live (citizen science) data, with the possibility to specify pollinator species or group, to powerfully support transformative personal learning journeys and pollinator-friendly management of garden spaces.


Scientific Reports
| (2020) 10:20483 | https://doi.org/10.1038/s41598-020-77537-6 www.nature.com/scientificreports/ showed that different bumblebee species favoured different food plants. Whilst using the label 'pollinator-friendly' brings to attention the plight of pollinating insects and will sway many to purchase plants that may be used as respective food plants, acknowledging greater complexity, including the existence of different levels of pollinator friendliness and variation among species therein, could create learning journeys and behavioural change 18 . Human-Computer-Interaction approaches can be used to develop interactive 'planting for pollinators' tools that support such learning. Such planting recommendations would remain up-to-date if based on species interaction data from active citizen science programmes.

Results
Evaluating practitioners' pollinator-and bumblebee-friendly planting recommendations. Across all 23 UK practitioner sources consulted (Suppl. Appendix Table 1), 465 different plant species were listed as good for pollinators and 376 plants were recorded as good for bumblebees. Numerous species occurred in both lists, and using those reoccurrences allowed us to reveal a remarkable similarity between those plants regularly mentioned as pollinator-friendly and those repeatedly identified as being bumblebee-friendly plants (r = 0.75, p < 0.001; Fig. 1a). This implies a generality in practitioners' assumptions about pollinatorfriendly plants for bumblebees, i.e. that plants identified as being good for pollinators in general are also considered to be the most suitable for the pollinator subgroup, bumblebees or visa-versa. Relationship between the number of practitioner sources that recommended a plant as being good for pollinators in general (total sources = 9; Suppl. Appendix Table 1 sources 1-9) and the number of practitioner sources that recommended a plant as being good for bumblebees specifically (total sources = 14; Suppl. Appendix Table 1, sources 10-23); (b) relationship between the number of practitioner sources that recommended a plant as being good for pollinators in general (total sources = 9; Suppl. Appendix Table 1, sources 1-9) and the number of times that plant was recorded as being used by bumblebees by BeeWatch participants, and; (c) relationship between the number of practitioner sources that recommended a plant as being good for bumblebees specifically (total sources = 14; Suppl. Appendix www.nature.com/scientificreports/ The citizen science programme BeeWatch gathered 6429 records of plant visitation by bumblebees from gardens and public greenspaces across the UK (Fig. 2a). These records associated UK bumblebees with 334 different plant species. Almost three-quarters of those (250 plant species) occurred in the practitioners' sources identifying species as being good for pollinators in general. Nearly two-thirds of the plants recorded in the BeeWatch database (235) occurred in the practitioners' list of plants identified as good for bumblebees specifically (Fig. 2b). A high percentage of shared species may suggest a reasonable level of agreement. Yet, when taking into account the frequencies at which plants occurred across all practitioner sources and comparing those recommendations with actual reported levels of plant use by BeeWatch participants, a different picture emerged: relationships were significant but with rather limited predictive power, for both pollinator-friendly (r = 0.52, p < 0.001; Fig. 1b) and bumblebee-friendly plant sets (r = 0.57, p < 0.001; Fig. 1c). Only 12 of the top 25 most frequently practitionerrecommended bumblebee-friendly plants appeared in the top 25 food plants used by bumblebees, as reported by BeeWatch participants (bold font in Suppl. Appendix Table 2). Vice versa, less than half of the top 25 BeeWatchidentified forage plants were in the top 25 bumblebee-friendly plants recommended by practitioners (bold font in Suppl. Appendix Table 2). Thus, there appears to be a discrepancy between plants identified as being important for pollinators in general, or even specifically for bumblebees, by practitioners and those regularly recorded by BeeWatch participants as having been used by bumblebees.
Making the assumption that BeeWatch participants had access to the majority of recommended pollinator-and bumblebee-friendly plants, many of these recommended plants were not observed by BeeWatch participants as being exploited by bumblebees (Fig. 2a). An absence, and under-reporting, of specific plant groups was apparent from the BeeWatch data. For instance, more than a quarter (27%) of the 207 plants which either did not appear in the BeeWatch database (e.g. crab apple species (Malus sylvestris, Malus sargentii); Fig. 2b) or were infrequently reported (e.g. other trees such as birch (Salix spp.), 23 records, and pear (Pyrus communis), 2 records) were large shrubs or trees known to exceed 2.5 m in height, where the flowers would be above the line of sight of most people and hence scarcely searched for bumblebees, photographed and submitted to BeeWatch. Many wildflowers were also absent from the BeeWatch database, e.g. monk's-hood (Aconitum napellus) and sweet clovers (Melilotus spp.) (Fig. 2b), perhaps because they are poisonous or regarded as 'weeds' and therefore not widely grown in gardens (where the majority (80%) of BeeWatch records originated from). Conversely, since there are over 70,000 plants in cultivation in the UK 19 , it is feasible that some species recorded by BeeWatch users as being used by bumblebees were missing from the practitioner-recommendation lists. Some of these absent plants were common garden species, such as lobelia (Lobelia spp.), petunia (Petunia spp.), red-hot poker (Kniphovia spp.), and lily-of-the-Nile (Agapanthus spp.) (Fig. 2b), perhaps overlooked by practitioners because www.nature.com/scientificreports/ of their commonality. Plants that flower outside the normal flight times of bumblebees, e.g. cyclamen (Cyclamen spp.) and pasqueflower (Pulsatilla spp.) were also reported as having been used by bumblebees by BeeWatch users but were not included on the practitioner-recommended lists. This is presumably an example of these lists being out of sync with the recently developed phenomenon of active over-wintering buff-tailed bumblebees increasingly observed in southern areas of the British Isles 20,21 as winter temperatures increase.

Range and diversity of plants used by different bumblebee species. The notion of 'bumblebee-
friendly' plants implies a high level of commonality among the many different species of bumblebees. Indeed, this appeared to be the case for cuckoo bumblebees-a relatively rare group specialised in parasitizing other bumblebee species-as BeeWatch data for all six cuckoo species were found to rely heavily on a single food plant (marsh thistle (Cirsium palustre); Suppl. Appendix Table 3). Yet, only four true bumblebee species [common carder (Bombus pascuorum), white-tailed (B. lucorum agg.), buff-tailed (B. terrestris) and red-tailed bumblebees (B. lapadarius)], relied relatively heavily on the most commonly used plant-lavender (Lavendula spp.) ( Table 1). For the other 12 true bumblebee species, the most frequently used plants were found across the rank order of most commonly used plants (Table 1). Thus, most bumblebee species had distinct food plant preferences, i.e. different species of bumblebee preferred different species of plants. In fact, the BeeWatch data indicates that only the morphologically rather similar white-tailed (B. lucorum agg.) and buff-tailed bumblebee (B. terrestris) had diets that showed considerable similarity (Table 1). Although certain plant species were more 'popular' than others, the relative proportions accounted for by the top plants for individual bumblebee species in the BeeWatch data were relatively low. For instance, lavender (the main food plant for many bumblebees) and a plant in the top 3 practitioner recommendations (Suppl. Appendix Table 2), only accounted for 11%, at most, of the records for any bumblebee species, and the 7 plants that stood out as being most frequently reported by BeeWatch participants constituted only 3-8% of the total number of plants used by all true bumblebees (Table 1 and Suppl. Appendix Table 2). The bumblebees that had slightly higher percentages of records against their 'most favourite food plant species' , such as 17% and 16% of records referring to heather (Calluna vulgaris) for the billberry bumblebee (B. monticola) and heath bumblebee (B. jonellus) respectively, were those considered to have a more 'specialist' diet 22 . The high number of records for cranesbill (Geranium spp.) use by the early bumblebee (B. pratorum) and foxglove (Digitalis spp.) by the garden bumblebee (B. hortorum), 16% and 13% respectively, may be accounted for by their morphology 23open geranium flowers used by the small (and small-tongued) early bumblebee and deep foxgloves used by the long-tongued garden bumblebee.
Most bumblebee species, however, were found to use a wide range of plants (Table 1). This is in contrast to the recommendations by practitioners, where more than a third of all plant species were recommending by four or more of the 14 sources, showing considerable levels of agreement between different sources as to which plants would be most useful. The use of many different plant species by the most common bumblebee species could be because observations by BeeWatch participants were made over many months: there was a high correlation between the number of months in which different bumblebee species were active and the predicted number of www.nature.com/scientificreports/ plant species they could potentially use, r = 0.73, p = 0.005. This indicates that these common (and long-flying) bumblebee species had the opportunity to feed on many plants that were in flower at different times during the season. By extension, the rarefaction model predicted that as the sample size of any bumblebee species increased, so did the mean number of plants any one bumblebee species would be expected to use (Fig. 3). This reveals that the label 'good for bumblebees' is very likely to be valid for a large number of plant species, however, it may have limited utility as in reality most plant species will be little used (Table 1). In other words, there are degrees of 'friendliness' , with only a few plant species being used to a greater extent than most others. Furthermore, the set of most-used plants are different for many of the different species of bumblebee.

Discussion
The need to conserve pollinating insects has captured the public's attention 24,25 . One method of doing so is through the planting of pollinator-friendly plants. The labelling of such plants is known to work well in certain settings. For instance, in garden centres where people are faced with a huge amount of choice, schemes advocating particular plants as good for pollinators have been very successful 8 . However, most of the "pollinator-friendly" planting recommendations available to the general public are remarkably similar, with the absence of reference to specific evidence detailing when, where and by whom the information was collected, making it difficult to ascertain on what basis such recommendations are made. The omission of reference to underlying data leads to speculation that many of these recommendations lack independence, i.e. that new lists are based on older ones 16 . This may lead to many suggestions being relatively old, with no account taken of changes in popularity of particular garden plants, availability of new plant species, or arrival of new cultivars 17 with different nectar or pollen qualities (these being the focus of foraging pollinating insects 26 ). Mismatches, or more opportune overlaps, of flowering times with insect emergence flight times may now occur due to phenological changes, making planting recommendations out-of-date [27][28][29] . A further drawback of such schemes lies in the overt communication of a simple dichotomy, i.e. either good for pollinators or not, whilst BeeWatch data-and various other studies-make it clear that there are gradations and pollinator-specific patterns therein 22,23 . Whereas the labelling of plants in a dichotomous manner may be a sufficient entry-point to begin engaging people with pollinators, the self-perpetuation of recommendations on which this dichotomy is based will continue to lead to confusion for individuals who seek out such plants and find that they either do not attract the pollinators they were hoping for or are little used by pollinators. This, in turn, would result in a lack of more in-depth engagement with these societally important insects.
Whilst being more up-to-date, detailed and bumblebee-species specific, data from the citizen science programme BeeWatch is not without problems. Arguably, the most difficult aspect of this citizen science data is that it does not necessarily represent plant attractiveness per se, but instead informs us which plants different bumblebee species used across the UK. Here, the popularity of certain plants, and thus their abundances, will contribute to what species appear attractive (based on their proportion of all BeeWatch data; Fig. 1). Resolving this dilemma at the spatial scales required, along with the number of plants concerned is very difficult, and arguably out of reach for a country-wide citizen science scheme that needs to engage widely with the public. Abundance data of plant species used (and, in the case of BeeWatch, photographed) by bumblebees could be requested from scheme Figure 3. Estimated mean numbers of plants predicted to be visited by bumblebee species when assessing an increasingly large number of records. (a) mean number of plant species predicted to be used by each of the seven 'common' UK true bumblebee species; (b) mean number of plant species predicted to be used by each of six 'rarer' UK true bumblebee species. Results were generated using data gathered from the citizen science programme BeeWatch, where 10 plant choices were randomly sub-sampled, without replacement, using 100 repeats of the procedure. Predictions for bumblebee species which had less than 10 records, i.e. the brownbanded carder and the red-shanked carder bees and the ruderal bumblebee, were unable to be made. www.nature.com/scientificreports/ participants, but in the absence of abundance data of alternative forage plants such information is difficult to interpret 30 . More absolute measures of relative attractiveness can be determined through standardised protocols involving timed observations of flowers, ideally in a paired manner within the same locations (i.e. individual gardens). However, this can likely only be done for a limited number of species, and not the many hundreds of recommended (n = 376) or used (n = 334) plants our study deals with. Although realising that the BeeWatch data is not preference data pur sang, we expect it to be a reasonable signal thereof. Testing this assumption, for example by conducting an investigation of the 7 plant species identified from BeeWatch data as the most widely used by bumblebees in UK gardens (constituting > 3% of plant species used; see Suppl. Appendix Table 2), using a standardised protocol involving timed observations of specific numbers of flowers [30][31][32] , would be productive. This could take the form of a new citizen science initiative, professional investigation, or possibly conducted using data from other studies, where such data is available. Earlier work evaluating the BeeWatch citizen science scheme 33 drew out the importance of identification tools offered as a starting point for wider learning and exploration of one's own gardening or nature space 18,34 . We suggest that simple (pollinator) labelling can be used as an entry point for this. Yet, creatively acknowledging levels of pollinator friendliness and complexities, by using intuitive visual interfaces for exploring pollinator-plant interactions or deploying recommender systems for suggesting collections of plants that will support a group of pollinators through the season, as pioneered by the Planting for Pollinators scheme 33,34 , would open the opportunity for individual transformative learning journeys 35 . It is these journeys that connect people to pollinating insects and, by extension, to their plight. Furthermore, we know that this process can lead to remarkable changes in garden management and initiation of positive environmental actions by individuals 33 .
We advocate the use of dynamic systems using filter-based data presentation approaches, which make use of up-to-date information to foster positive ecological thinking and action. Moreover, by using citizen science data captured at the national level to populate such systems, this would provide planting recommendations that reflect current knowledge and the complexities of pollinator friendliness, and ultimately help pollinating insect populations around the globe.

Methods
Pollinator-and bumblebee-friendly practitioner plant data. Practitioner plant suggestions were compiled from 23 widely available UK sources (Suppl. Appendix Table 1 -sources that were readily available to the general public through internet and library searches for "pollinator-friendly plants" and "bumblebeefriendly plants") that detailed plants identified as being good for pollinators in general or bumblebees in particular. To evaluate the usefulness of pollinator-and bumblebee-friendly practitioner plant lists we ranked each plant based on the number of literature sources where it occurred as a recommended species. In some instances, plants have been broken down to the species level, whereas in other cases only genera have been used, reflecting the degree of specificity we found in practitioner recommendations. For similar reasons, we did not include all the cultivars available for certain plant species that have undergone horticultural breeding and selection. This enabled us to produce two ranked plant lists derived from literature sources: (i) plants which were identified as pollinator-friendly (9 sources; see Suppl. Appendix Table 1 [sources numbered 1-9]); and (ii) plants which were identified as being specifically good for bumblebees (14 sources, see Suppl. Appendix Table 1 [in bold font, sources numbered [10][11][12][13][14][15][16][17][18][19][20][21][22][23]). Some plants classed as being good for bumblebees may have been missed from our combined list if they were recorded in academic journal sources, which we did not access as we concentrated on sources that were widely available to the public. However, we believe this number of plant species to be small, particularly as we found many of the same plants mentioned with increasing frequency the more plant lists we consulted.
BeeWatch-an online bumblebee recording citizen science scheme. BeeWatch is an online citizen science initiative developed by the University of Aberdeen and the Bumblebee Conservation Trust in 2011, and has been co-ordinated by the University of Aberdeen. Members of the public submit photographs of bumblebees with details on location, date of sighting and plant species used (if applicable) via the online interface. Further details about BeeWatch can be found in Van der Wal et al. 2016 30 . When specifying which plant species was being utilised by a bumblebee, recorders had the option of free data entry or they could choose a plant from the compiled list generated from 23 different sources (see Suppl. Appendix Table 1). Since the emphasis of BeeWatch has been on learning, exploration and positive action, its biological records are opportunistic rather than gathered through structured recording processes that take into account factors such as recording effort (timed searches 36 ), plant-species specific sampling, prevailing weather conditions and quantification of the floral resources (to determine whether flower abundance could affect plant use). However, BeeWatch data do represent a 'snapshot' of plant use by bumblebees across UK gardens and greenspaces as observed by participants at any one time over a 6-year period.
Range and diversity of plants used by bumblebees. Data was downloaded from the BeeWatch system at the beginning of July 2017 and covered the period from the initiation of the 'live' BeeWatch program on 25 August 2011 until 30 June 2017. For the purposes of these analyses, we separated our data into two groupings: the 16 true bumblebee species (see Table 1 for details) and the six cuckoo bumblebee species (see Suppl. Appendix Table 3 for details). We separated true bumblebee species and cuckoo bees, because cuckoos are known to employ a different life strategy to that of true bumblebees and hence use forage plants in a different way 37  www.nature.com/scientificreports/ We collated all BeeWatch data where a plant species that had been used by a bumblebee was recorded. We ranked each plant by the number of times it was mentioned in the BeeWatch database, then counted the number of times a specific plant was mentioned as being use by a specific bumblebee. By then dividing these numbers by the total number of records for each bumblebee species we were able to produce proportion data on plant use by individual bumblebee species. We also utilized the overall numbers of plants used to derive the proportional use of individual plants by all bumblebee species combined.
Statistical analyses. Pearson's correlations were performed to examine relationships between: (i) the number of sources-for each of the plants mentioned in one or more of the 23 available practitioner recommendation sources-which identified a plant as pollinator-friendly (9 sources) with the number of sources recommending it (i.e. recommendation frequency) as bumblebee-friendly (14 sources); (ii) the recommendation frequency of plants as pollinator-friendly (9 sources) with the frequency of plant occurrence in the BeeWatch database, and; (iii) the recommendation frequency of plants as bumblebee-friendly (14 sources) with the frequency of plant occurrence recorded in the BeeWatch database. This enabled us to determine how similar the lists of all pollinator-recommended plants and those specifically recommended as bumblebee-friendly were, whilst also comparing practitioners' recommendations with actual patterns of plant use recorded by BeeWatch participants.
We estimated the mean number of different plant species that could be expected to be used by the different bumblebee species by using rarefaction analysis, as this procedure accounts for the varying sample sizes apparent in the BeeWatch data. We randomly sub-sampled 10 bumblebee plant choices, without replacement, from our BeeWatch plant use database, using 100 repeats of the procedure. Using a sample size of 10 meant that we were unable to make predictions for bumblebee species in which we had less than ten records, i.e. the brown-banded carder and the red-shanked carder bees and the ruderal bumblebee. Pearson's correlations were used to determine if there was a correlation between the number of months in which different bumblebee species were active and the mean number of different plant species they would be expected to use, as determined by the rarefaction analysis.
All analyses were carried out in program R version 3.4.0 38 , using the vegan package version 2.4-3 39 for rarefactions. The map in Fig. 1a contains data from BeeWatch (2011-2017) and OS data Crown copyright and database right (2019) and was generated by H. Anderson using ArcGIS Desktop 10.7 Esri Inc. 1999-2018.