Some of the favoured traits may promote pest control or pollination services in adjacent fields , but other traits may not . Even if these particular bee and fly species are not contributing substantially to pollination or pest control services now, they could become important in the future if environmental conditions change – for example, as a result of changes in farm management, climate or altered biotic relationships . Further work is needed to elucidate how small-scale restoration influences pollination services via their effects on species’ response and effect traits . Meanwhile, this study shows that these habitat enhancements provide clear conservation benefits for sensitive species in flower visitor communities, even in highly intensively managed agricultural landscapes.Land use change is a main driver of biodiversity declines. Most land conversions are associated with the expansion of croplands, habitat loss and fragmentation, and biodiversity loss . Currently, agricultural land conversion is concentrated in the tropics, where most new agricultural lands – especially between 1980 and 2000 – came at the expense of undisturbed and disturbed forests raising important global concerns about biodiversity conservation . Furthermore, agricultural intensification, or changes in the actual management within farms may exacerbate the impacts of land use conversion for biodiversity . Thus, factors acting at multiple spatial scales may have strong impacts on diversity and alter processes structuring biotic communities . Yet, the effects of environmental change on community composition are not random . Hence, changes in biotic communities depend on the abundance of different taxonomic groups and on particular traits that mediate species’ responses to the magnitude, frequency and spatial patterns of disturbance . In light of environmental change,draining pots communities can undergo biological homogenization whereby sensitive species are lost from a regional pool of species or experience range contraction and tolerant species increase their ranges and abundance .

These non-random changes can cascade to affect the functional traits within a community, and thereby affect ecosystem functioning, with important implications for ecosystem services . Bees provide ecosystem services, but bee communities and populations are affected by environmental change. Most tropical crop plant species require or benefit from visits by native and non-managed bees for successful reproduction . Thus, conservation of diverse bee communities is important for both food production and tropical plant communities . Bee communities and populations are affected by land use modifications at both local and landscape scales . Bee diversity increases with flowering plant diversity and the availability of nesting sites . Agricultural practices such as tillage and sowing that reduce available resources, along with pesticide use, negatively affect bees and drive population declines . At the landscape scale, land use diversity, connectivity and proximity to undisturbed forest fragments benefits bees . In simplified landscapes, local factors are more important predictors for bee community composition, whereas these same factors are less important in highly diverse landscapes . Furthermore, local and landscape factors differentially influence bee species with specialist and low-dispersal ability species being more strongly affected by intensification and fragmentation compared with generalist, social, and high-dispersal ability species such as Apis mellifera . Most research evaluating how local and landscape factors influence patterns of bee diversity in agricultural landscapes focuses on temperate latitudes, where farms tend to be large and homogeneous but see . However, the effects of local and landscape factors on tropical mountainous bee communities is still under explored . Understanding how local and landscape disturbance affects bees in heterogeneous agricultural landscapes is important for designing conservation strategies in areas with high dependence on non-managed bees. In this study, we ask how differences in local habitat structure and landscape configuration affect bee communities across a heterogeneous, mountainous agricultural landscape in Anolaima, Colombia. We asked Which local and landscape factors influence bee abundance and diversity ?Which local and landscape factors drive changes in generic and tribe abundance and composition across farms in Anolaima?

We predicted that farms with a higher percent of undisturbed habitat, more complex vegetational structure, lower agricultural disturbance, and surrounded by a higher percent of complex habitat at the landscape scale will host higher abundance and richness of bees; local factors will have greater influence on bee community composition compared with landscape factors; and bee generic richness and abundance of specific tribes will vary depending on availability of land use types. We conducted this study in Anolaima, in the eastern slope of Andes mountains in Colombia . This municipality extends between 900 and 2800 m.a.s.l., with an average elevation of 1650 m.a.s.l.. Most lands in the municipality have steep slopes . The traditional precipitation regime is bimodal, with marked dry seasons between Dec – Mar and Jul – Sept, mean annual precipitation of 1232 mm, and average relative humidity between 70% and 80% . Life zones in the municipality transition between cloud-submontane forest and tropical dry forest, but most land cover in the area is comprised of cattle ranching and cropland . Coffee is the most extensive crop covering 10% of the total area. Small farms represent 92.6% of private landholdings in the area, and cover 53% of the total land area in the municipality . We worked in seventeen farms chosen to represent a gradient of management intensification. Farms were separated by a minimum of 2 km and represented the full range of agricultural management types present in Anolaima. Land uses included secondary forests; permanent crops arranged as agroforests ; shaded crops with simplified shade ; unshaded staple crops ; unshaded commercial short-cycle crops ; fallow lands or unmanaged areas undergoing natural regeneration; and pastures. Permanent shaded crops and traditionally managed staple crops are managed in diversified systems seldom treated with synthetic biocides. In contrast, conventional short-cycle crops are monocultures or polycultures intensively managed with synthetic biocides and with short fallow periods. Because of the average farm size , monocropping seldom extends over large areas in this region . We measured local and landscape habitat features for each study farm. To survey vegetation, we established a 1-ha plot centered on a random point within each farm and divided it into sixteen 25 m x 25 m quadrants . We classified land use types and measured canopy cover in each 25 m x 25 m quadrant. Within each quadrant, we established 4 random 2 m x 2 m sub-plots, 64 in total per farm, in which we measured ground cover and flower abundance.

In addition, we established a 200 m-radius circle around the center of the 1-ha plot and divided it into six pie pieces. In each pie piece we randomly established a 15 m x15 m plot in which we measured arboreal vegetation. We conducted landscape analyses within circles of 200 m, 500 m and 1 km radii around the 1-ha plot. We measured local vegetation features within each farm. Within each 2 m x 2 m subplot we estimated ground cover , measured height of the tallest herbaceous vegetation, and counted the number of flowers on herbs and shrubs. Within 25 m x 25 m quadrants we counted the number of flowering trees, and measured canopy cover with a concave spherical densitometer by averaging measurements at the center, and 10 m to the east, west,large plastic garden pots north and south of the quadrant center. We also observed and registered the land use of each 25 m x 25 m quadrant and then grouped them in one of seven categories: forest/agroforest; crops with simplified shade; unshaded crops with traditional management; fallowed lands; pastures; unshaded crops with conventional management; constructions ; and border of roads. We collected this data on the same days that bees were collected in each site. Within each 15 m x 15 m plot, we estimated the vertical structure of the vegetation , counted the number of trees , and registered tree morpho-species, tree height, and tree diameter at breast height . We measured tree diversity, tree size, and the vertical structure of the canopy between Jun – Aug 2015. We analyzed the configuration and composition of the landscape surrounding each farm with SPOT satellite images and digitalized aerial photographs from Instituto Geográfico Agustín Codazzi. To estimate landscape composition and determine the landscape context of each site, we classified images and created four land cover categories: complex habitat ; unshaded crops; pastures; and eroded soils. We estimated the percent area of each land cover category within 200 m, 500 m and 1000 m of the center of each farm. We also calculated the nearest distance from the center of the bee survey plot to complex habitat, unshaded crops, and to water. We conducted these analyses in ArcGis 10.3. We used aerial nets and observations to survey bees. We netted and observed bees between 0-3 m above ground in each 25 m x 25 m quadrant during 10 min. and walked all quadrants four times during the same day, for a total of 40 mins. per quadrant. Overthe four visits to each 1-ha plot, we varied the time of day each quadrant was visited to capture bees under different temperature, humidity, and sunlight conditions. We netted all bees except for Apis, Trigona , Tetragonisca and Eulaema bees that we identified and counted in the field. We killed bees with ethyl acetate, placed them in dry containers, and pinned them. We determined bees to the genus level using identification keys for bees in Colombia, Panama and Brazil at Laboratorio de Abejas in Universidad Nacional de Colombia. We sampled bees in the dry and wet seasons of 2016. We registered the type of land use in which we captured each bee.

All bee netting and observations took place between 7 AM and 2 PM on sunny days with low wind speed and with no rain. We took data on relative humidity, temperature, and wind speed at 8:00am and 12:00m as covariates. We selected five bee abundance variables, two community similarity variables, and three bee diversity variables for inclusion in model analysis. We sampled bees in the 25 m x 25 m quadrants but aggregated bee data at the farm scale for all analysis. For abundance, we used total bee abundance, partial abundance after excluding the two most common genera, and abundance of the three most common tribes. For community similarity, we used axis 1 of a non-metric multidimensional scaling analysis based on Bray-Curtis similarity for bee genera and for bee tribes. For bee diversity we used estimators of bee richness, evenness and dominance using rarefied Hill numbers. Hill numbers convert basic diversity measures to “effective number of species” numbers that obey a duplication principle. We calculated Hill numbers at three different orders of diversity. Order q=0 is equal to species richness, giving more weight to rare species; when q=1 the weight of each species is based on its relative abundance; and when q=2 abundant species have a higher weight in the community . We used 0D numbers as estimators of richness, the Hill estimator of evenness , and the Hill inequality factor as estimator of dominance across study sites . Because sample size differed across farms, we rarefied Hill numbers at q=0, q=1 and q=2 to assemblages of 72 individuals with all genera, and to 31 individuals for analysis without the two most common genera. We calculated rarefied Hill numbers with the iNEXT package and plotted diversity profiles with the Entropart package . To select explanatory variables for analyses, we grouped local and landscape features as separate groups and then ran Pearson’s correlations to identify non-correlated variables within each group. Some variables did not fit within any group and were included. Other variables had high numbers of zeros and were excluded. We used 12 explanatory variables in our models . To test whether local and landscape factors influence bee variables, we ran generalized linear models in R using the glmulti package . We tested all combinations of explanatory factors and compared Akaike Information Criterion values to select for the best models. We report conditional averages for significant model factors, AICc values, p-values and multiple linear R2 values for the best predicting models. When more models were within 2 AIC points of the next best model, we averaged models using the R MuMIn package and used conditional averages to account for significant model factors . To test whether factors influenced community similarity, we ran a permutational multivariate analysis of variance on bee genera and tribe similarity matrices using the R vegan package .