Delivery of crop pollination services is an insufficient argument for wild pollinator conservation

There is compelling evidence that more diverse ecosystems deliver greater benefits to people, and these ecosystem services have become a key argument for biodiversity conservation. However, it is unclear how much biodiversity is needed to deliver ecosystem services in a cost-effective way. Here we show that, while the contribution of wild bees to crop production is significant, service delivery is restricted to a limited subset of all known bee species. Across crops, years and biogeographical regions, crop-visiting wild bee communities are dominated by a small number of common species, and threatened species are rarely observed on crops. Dominant crop pollinators persist under agricultural expansion and many are easily enhanced by simple conservation measures, suggesting that cost-effective management strategies to promote crop pollination should target a different set of species than management strategies to promote threatened bees. Conserving the biological diversity of bees therefore requires more than just ecosystem-service-based arguments.

was not related to the average distance between the sites that had been sampled in each study (F 1,80 = 2.77, P = 0.100). Within studies, sites were at least 1 km apart. Studies with average between site distances larger than 100 km (21-23, 76-79, 86; see Extended Data Table 1) are not shown for clarity. c, The number of dominant species was not related to the number of observed individuals (F 1,83 = 2.13, P = 0.148). South African studies were excluded from this B particular analysis as in South Africa it is impossible to distinguish wild from managed honey bees. Studies with more than 3,000 individuals are not shown (studies 9, 23; Supplementary Table 1). Dominant species contributed more than 5% to the total number of individuals of the bee community on crop flowers in each study. For all three analyses, regression on the full dataset revealed heteroscedasticity. This variability was caused by a small number of outliers (i.e. studies with very high between site distances). Transformations of the response variable, or analyses assuming binomial or Poisson error distributions, did not produce homoscedastic datasets. Analyses without these studies produced constant error variance. The results of these analyses were similar to those of analyses using the full dataset, and so in this figure, we therefore present the results using the full dataset. For illustrative purposes only, the nonsignificant relationships are depicted by means of a dashed trend-line based on these analyses.

Supplementary Table 1 | A summary of the studies providing the data on bee species visiting flowers of insect pollinated crops.
Unpublished datasets are identified by the name of the first author. Yield and price statistics based on FAO data 25 . N.a.: not available. survey, each transect was surveyed for 15 minutes by slowly walking its length. Surveys were conducted between 09:00 and 18:00 at temperatures above 15 °C on days with no rain, low or nonexistent cloud cover and low wind speeds. Honey bees and bumble bees were identified to species in the field, all other bees were collected for identification in the lab. Method: In 2011 bee densities were assessed on 16 flowering sunflower fields. In each field, bees were surveyed between 4 July and 6 August, 2011, in two 150 m long and 1 m wide transects; one located along the edge of the field and the other in the centre of the field. Surveys were conducted between 09:00 and 18:00 at temperatures above 16 °C on days with no rain, low or non-existent cloud cover and low wind speeds. On two fields, edge and centre transects were surveyed four times. On three more fields, edge transects were surveyed three times but center transects only two times. All other fields were surveyed completely three times, each time for 15 minutes, while slowly walking along the transect and recording only pollinators on sunflower heads. Honey bees and bumble bees were identified to species in the field, all other bees were collected for identification in the lab.

Methods:
In June and July, 2006 bees were surveyed in 23 oilseed rape fields. Bees were surveyed in up to three sampling points within each field (depending on the field size). Each survey lasted ten minutes and took place between 10:00 and 17:00 on sunny days with little wind. Each sampling point was surveyed up to three times, depending on flowering phenology. Surveys were carried out by two experienced ecologists. Bees were either identified in the field or collected for identification in the laboratory. Methods: Bees were surveyed using standardized transect walks in 10 field bean fields. Surveys were carried out between 09:00 and 17:00 from May to August, 2005 under good weather conditions with temperatures at least 15°C, no precipitation and a wind speed below 40 km.h -1 . Bees were surveyed for 30 min in a 150 m transect line identifying visiting insects at species level and catching unidentified species within a 4 m wide corridor. Each field was surveyed four times during the main flowering period of the crop. Collected specimens were pinned, labeled, and subsequently identified to species. 62.5% of the bumble bee (Bombus) individuals were only identified to genus level because field workers did not catch them. For these studies, as Bombus is a wellknown genus with most individuals identified to species level, the unidentified specimens were assigned to species based on the proportions of actual species level identifications within each study.

Methods:
In 2011, bee visitation to field beans and strawberry fields was surveyed. For each crop, 8 fields were selected and 2*150m transects were walked between rows. For recording purposes, the transects were sub-divided into 3*50m transects, each of which was walked for 10 minutes. Any pollinators observed carrying out floral visits (legitimate only for beans) were recorded, if the pollinator could not be identified in the field, it was collected and identified in the laboratory. Three rounds of bean surveys were carried out at each field between the 10th and 25th of May and 3 rounds of strawberry surveys between the 18th of May and the 14th of June. All surveys were conducted only when temperatures exceeded 15 °C and when wind was light or non-existent. 12.8 % of the bumble bee (Bombus) individuals were only identified to genus level because field workers did not catch them. For these studies, as Bombus is a well-known genus with most individuals identified to species level, the unidentified specimens were assigned to species based on the proportions of actual species level identifications within each study.

Methods:
In 2010 and 2011 bee visitation rate on crop flowers was examined in 6 apple and 6 pear orchards. The same apple and pear orchards were used in both years. Each orchard was surveyed twice per year, once in the morning and once in the afternoon with at least three and at most seven days separating surveys. Surveying was conducted by four experienced entomologists between 23 April and 6 May, 2010 and between 8 and 20 April, 2011 under sunny conditions or scattered clouds. Temperatures ranged between 15 °C and 20 °C with calm wind to moderate breeze. Bees were surveyed using a single transect between two rows of trees along the length of each orchard with the transect subdivided into 25 m long plots (mean number of plots per orchard ± s.e.: 8.5±1.0 for apple and 9.7±0.5 for pear). In each plot all bees observed on apple or pear flowers during a 10 minuteperiod were identified to species. Easily recognizable species were generally identified in the field; all other species were collected and identified in the lab.

Methods:
In 2011, bees were surveyed in 8 oilseed rape fields. One field was surveyed only once on 30 April, while all others were surveyed twice between 30 April and 30 May, once in the morning and once in the afternoon. In each field, bees were surveyed in two 1 x150 m transects located at the edge and in the interior of the field (>25 m from field edge). Transects were subdivided into three 1 m x 50 m plots. In each plot, bees visiting crop flowers were collected during a period of 5 minutes. Easily recognizable species were generally identified in the field; all other species were collected and identified in the lab. Surveys were carried out under dry weather conditions, with low to moderate wind speeds and temperatures above 15 °C. Methods: In 2012, bees were surveyed at 10 leek fields in the province of Foggia, Italy. Each field was surveyed once between 19 and 21 June, 2012 under sunny weather conditions with temperatures above 20 °C and light or non-existent winds. In each field, bees were surveyed in a single 5 m long transect between two crop rows. During a period of 10 minutes (net observation time), all bees visiting leek umbels were noted. Easily recognizable species were generally identified in the field; all other species were collected and identified in the lab. Study no reference Crop; Location; year; yield (tonnes.ha -1 ); price ($.tonne -1 ); no. sites; no. wild bees; ratio wild/honey bee Methods: Data were extracted from reference (63). Bees were surveyed on presumably 8 alfalfa fields in 8 different areas between 21 June and 22 August, 1954. Surveys were repeated from 11 July and 9 September, 1955 in the same 8 areas along with one additional area, presumably on 9 alfalfa fields. Finally, a total of 11 fields in the same 9 areas were surevey between 5 July and 21 August, 1956. The study areas were scattered across Hungary and in each area and year bees were surveyed during approximately 10 days. Total number of survey days per year were 67 in 1954 and 80 in 1955. In 1956, bees were surveyed for a total of 127 hours. Wild bees were collected with nets from alfalfa flowers. Individual surveys lasted 30 minutes and were conducted both in the morning and in the afternoon. Temperatures ranged between 16°C and 35°C and wind speeds were below 15 km/h. Each site was surveyed once by capturing bees along a 50 m long and 1 m wide transect, within the main flowering crops, oilseed rape and sunflower and flowering alfalfa. All species (other than honey bees) were identified in the lab by specialists.

Methods:
In 2007 and 2008, bees were collected with sweepnets on oilseed rape flowers in two regions of France. Bees were always captured in the first meter of the fields, except in 2008 in Pleine-Fougères where bees were also captured in the middle of the fields. Each field was surveyed three times during the oilseed rape flowering period, with between 3 and 5 days separating surveys. Surveying was conducted under sunny conditions or with scattered clouds. Temperatures ranged between 15 °C and 20 °C with at most a moderate breeze. Depending on the field, a survey round comprised 3 to 12 points. On each point, 3 strikes of a sweepnet were used to catch bees on oilseed rape flowers. All bees were collected and identified in the lab.  (25 × 25 m) were located at field edges; in some fields (sunflower-9, watermelon-12) an additional interior plot was located 100 m from the edge. Sampling sites were separated by at least 1 km from one another. Field work was conducted under standardized weather conditions (sunny to light overcast skies, temperatures >18 ºC and mean wind velocity <5 m.s -1 ). Each plot was sampled between one and three times (mostly twice), each time on a separate day. In each sampling day, two sampling sessions (2-3 hours apart) were carried out. Each session included 10 min (or 15 min in sunflower in 2010) of bee netting (the stopwatches were stopped when handling bees that were caught). Bee sampling was conducted between 8:00 and 16:00 in the sunflower study, and between 7:00 and 11:00 in the watermelon study. Each orchard was surveyed once or twice during the apple bloom, on days with temperature > 15°C between 10:00 and 15:30. We required that there be enough sun to cast a shadow. At each site, multiple transects of 15-minute aerial netting surveys were conducted along blooming tree rows. During each survey, collectors walked a steady pace along 50 m of each side of two-adjacent tree rows and netted all bees observed to be visiting apple blossoms. Transects were spaced, at least, 50 m apart and were placed where trees were in highest bloom, within 150 m from the orchard edge. Distance between orchards was at least 1.9km. The number of timed net collections per site varied according to farm size. Methods: In June and July, bees were surveyed in tomato fields on 15 study farms in 2004 and 13 study farms in 2005. Surveys used one 50 meter transect per farm within which all data were collected. All non-Apis bees visiting crop flowers were collected by hand net along the entire length of the transect. Total minutes of sampling effort varied across years but was always standardized across all farms within a given year. One sample day per farm was conducted per year. Honey bees were observed visiting flowers in timed samples, but not netted, so data used for honey bees include only observed visitors. Data collection was only conducted on days suitable for bee activity (sunny, partly cloudy or bright overcast; wind speeds <2.5 m.s -1 ; >18 °C). Bees were identified by professional taxomists. Methods: Bees were surveyed in 16 commercial cranberry bogs. Within each bog, two 60 m transects were located; one in the interior of the cranberry bog, one parallel to the edge next to forest. Two sample days per farm were conducted per year, and within each day data sampling was conducted once in the morning and once in the afternoon. All wild bees visiting cranberry flowers within the transect were collected for a total of 60 minutes per collection day. Data were collected from June to July, in each of 2009 and 2010. Honey bees were observed visiting flowers in timed samples, but not collected, so data used for honey bees include only observed visitors. Sampling was only conducted during weather suitable for bee activity (>15 °C, wind <3.5 meters.s -1 , not dark overcast). Bees were identified by professional taxomists.  In years with multiple sampling days per year, sampling was organized into rounds, with the rounds temporally stratefied throughout the period of bloom. Data were collected between June and August in each year. Data collection was only conducted on days suitable for bee activity (sunny, partly cloudy or bright overcast; wind speeds <2.5 m/s; >18 C), with a few exceptions. Honey bees were observed visiting flowers in timed samples, but not netted, so data used for honey bees include only observed visitors. Bees were identified by professional taxomists. Methods: Data were extracted from reference (70). Bees were surveyed from mid-June to mid-July on cranberry bogs. Eight bogs were surveyed in 1990 and nine in 1991, three of which were the same as in 1990. In each bog, as many bees as possible were captured with either an insect net or jar as the individual collecting moved through the bog over 15 min in 1990 and 10 min in 1991. Collections were carried out three times during cranberry bloom. All bees, including honey bees, foraging on cranberry bloom were collected. The bees were collected and pinned and identified to species in the laboratory.  In each orchard we observed flower visitors on five trees at the orchard edge closest to semi-natural habitat. At each tree, eight groups of flowers were observed for three times 20 seconds each, two each in the inner top, inner bottom, outer top and outer bottom quadrants of the tree (total of around 13 min per orchard). Species were identified mainly by close observations of the flower bundles or caught for identification in the lab. Methods: In July and August, 2001, bees were surveyed in 11 tomato fields in northern California. In each tomato field, bees were surveyed by walking transects at the rate of 10 m/min, covering each row twice, once in each direction, and recording all bee visits to tomato flowers. In small fields, transects were walked along all rows. In larger fields, surveys were carried out at up to four transects, each 80m long. Each field was sampled between 8:30 and 12:30 on three different days, in the early, mid, and late morning, respectively. In 2011, surveys were conducted once each at 7 sites. In 2012, surveys were conducted twice, once in spring and once in summer at each of 17 sites (though one site was dropped from summer sampling because strawberries were pulled up). Bees were collected via netting in good weather conditions (at least partially sunny with low wind between 0.4 and 3.5 m.s -1 , and temperatures above 14°C). At each site, six 10-minute sample periods took place in the same good weather conditions. Clocks were stopped when handling specimens. All bees collected were identified by a professional taxonomist to the lowest taxonomic level possible. For a number of Lasioglossum species, only females could be identified to species level. We therefore allocated all unidentified Lasioglossum males in proportion to the numbers that were identified as female Lasioglossum specimens. Honeybees were not collected and observed counts were recorded.

Methods:
In 2010 and 2011 bees were netted during standardized time periods on production watermelon fields in central California. Some farms were sampled in both years but never the same field. Within a year each site was visited three times during peak bloom at 4-5 day intervals. Sites were visited between 8 June and 3 August, 2010 and between 22 June and 16 August, 2011 during sunny conditions with temperatures between 22 °C and 33 °C and with wind speeds below 3 m.s -1 . On each sampling date, bees were netted during four 10-minute periods during the day along the same 50 m transect (40 min total). All specimens were collected and identified to species in the lab. For a number of Lasioglossum species, only females could be identified to species level. We therefore allocated all unidentified Lasioglossum males in proportion to the numbers that were identified as female Lasioglossum specimens. Methods: In 2000 to January, 2001 bees visitating highland coffee flowers were surveyed in 24 agroforestry systems and bees visiting lowland coffee flowers were surveyed in 15 agroforestry systems. Agroforestry systems were dominated by coffee and cacao and located in the buffer zone of the Lore-Lindu National Park in Central Sulawesi. Bee flower visitation was observed for 25 minutes on a full-blooming coffee plant per agroforestry system and this was repeated three times for a total of 75 minutes of observation time for each of the 24 agroforestry systems. Each day, a different fullblooming coffee plant than that used the day before was observed. Sampling was carried out between 9:00 and 14:00 on sunny to slightly overcast days. Easily recognizable species were identified in the field while others were collected and identified with the help of trained locals in the lab. Three observations were carried out during the day at each point. These were at 10-11 am, 12-1 pm and 2-3 pm. Only umbels with more than 30% of flowers open were observed. Umbels were examined along rows within the confines of the marked observation points. Approximately 15 minutes were required to complete the observations at each point and 60 minutes to complete observations across an entire field. When the identity of the bee species was unknown, specimens were collected using vials, containers and nets for identification in the laboratory.

Methods:
In March and April, 2009 bees were surveyed in commercial sunflower farms by capturing all visitors of flowerheads. Surveys were conducted in 33 plots (4 x 4 m, sunflower density of 10 plants m2) within sunflower fields and plots were at least 350 m away from each other. Each plot was surveyed on two different days (once in the morning and once in the afternoon) during the week of peak flowering. In each survey, three locations (one for each of three observers) were randomly selected within the plot and all the sunflower heads that could be reached were observed for 4 minutes, during which all insects that touched the reproductive parts of the flowers were recorded. In total, each plot was observed for 24 minutes. All bees collected were identified to the lowest possible taxonomic level by an expert entomologist. Methods: During 16-23 March, 2011 insect pollinators were surveyed on ten commercial sunflower farms. Five field study sites were selected adjacent to natural vegetation (<200 m) and five were selected at a distance >2000 m from natural vegetation. On each field study site, 100 flower heads in each of four parallel transects spaced 20 m apart (total 400 flower heads) were surveyed in the morning (09:00 -12:00) and afternoon (14:00 -16:00). Surveys were conducted by walking along a transect between rows of plants and recording the number of insect pollinators seen on individual flower heads, one by one. Voucher specimens for all insect flower visitors that touched the reproductive structures of surveyed sunflower heads were collected. From this, all bee specimens were identified to the lowest possible taxonomic level.  Methods: In April 2006, bees were surveyed on coffee flowers at 13 different sites in cofffee plantations in Nueva Alemania in the southern highlands of Chiapas. In the study region, Coffea arabica and Coffea robusta are planted (approximately 4000 coffee bushes per hectare) under a canopy of overstorey trees. Surveys were conducted in 15 minute periods between 8:00 and 14:00. At each site, four fully flowering branches (minimum of 20 blossoms) were randomly chosen from a randomly selected cofffe bush. During observation periods, the identity of the visitor was noted and, when possible, bees were captured after the observation period for identification.
† Includes managed bees hired for pollination as well.
Supplementary Table 2 | The top 100 bee species with the highest mean contribution to crop production value and their occurrence in different studies and crops. Mean contribution to crop production is based on the 53 studies for which contribution to production value could be calculated (i.e. data on crop production value and the relative contribution between wild and managed bees were available). Occurrences also include crops and studies for which no contribution to crop production value could be calculated.

Supplementary Table 4 | The relationship between flower visitation frequency and crop pollination.
For each crop and year we show the number of pollinator species groups analysed (n; number of species are in parentheses), the Pearson correlation (r) between visitation frequency and total pollination, the correlation between mean per visit pollen deposition and total pollination, the correlation between visitation frequency and mean per visit pollen deposition, and the ratio of the standard deviations of the logarithm of the visitation frequency and logarithm of the per visit pollen deposition (the parameter R of reference 28). Crop