The field data for this study was collected in 2009 (May–September) in four large mango farms within the same area as our 2008 study (Carvalheiro et al. 2010) in NE South Africa, which included orchards managed with conventional and organic farming practices. All orchards selected for this study within these farming regions were similar in size, tree age and density, and were subjected to similar management practices (for further details of location and management practices see Fig. S1, Supporting information). During mango flowering, all farms studied have managed honeybee Apis mellifera scutellata Lepeletier colonies (at least one hive per ha), although previous studies in Africa have found that honeybees are not very attracted to mango flowers (Free and Williams 1976; Carvalheiro et al. 2010). On all farms used in this study, soil nutrient and water content of the different orchards are monitored annually, and correction measures implemented to standardize conditions to those considered optimal for mango production (J. du Preez and H. Groove pers. comm.). These uniform management strategies minimize abiotic variability between sites, facilitating comparison of biotic variables.
In each of the four farms, an NFCAs was created at least 250 m away from natural habitat (a distance at which we expect less than half of the potential maximum abundance and diversity of visitors, see Carvalheiro et al. 2010). NFCAs were ca. 25 m2 in size, located in a corner of four mango orchards, and positioned so as not to interfere with normal management practices. Although ideally these NFCAs should comprise a diverse set of native plant species, we limited ourselves to two perennial native species that are present within the regional natural vegetation, flower before and during mango flowering season and have different floral structures. We used Aloe greatheadii Schönland (Asphodelaceae), which is attractive to many insects (Human & Nicolson 2006) and Barleria obtusa Nees (Acanthaceae), visited by flies and bees (Potgieter & Edwards 2005). We planted 30 individuals of each species, arranged alternately, in each NFCA. Flower visitation and production of orchards of the three cultivars located near NFCAs was recorded (see survey methods below). Although no mango pests were found in these plant species during preliminary surveys (see Table S1, Supporting information), 30 fruits from each plant species were collected within each NFCA (totalling 120 fruits per species) and kept under laboratory conditions to capture any insects emerging over a period of 6 months.
Effect of small patches of perennial native flora on mango flower visitor communities
Visitation surveys were conducted for both mango and the two planted species within the four NFCAs on warm, still, dry days (temperatures between 20 and 39 °C, wind speed <4 km h−1) between 08:00 and 16:00.
Timed focal point observations (2 × 10 min per plant species) were conducted in each NFCA, every month before and during the peak mango flowering season (i.e. from June until September), totalling 320 min of observation per species. In each 10-min observation period, the recorder observed all open flowers within a semi-circle of c. 50 cm (3–20 flowers for A. greatheadii and 2–10 flowers for B. obtusa). All flower visitors observed were recorded and collected for identification.
Mango flower visitation surveys were carried out in 21 orchards that were far from NFCAs (i.e. at least 250 m away) and in 11 orchards near NFCAs (i.e. orchards with centres 50–150 m from NFCAs). Fourteen orchards (of which five were near NFCAs) were organic and 18 (of which six were near NFCAs) conventional (see Table S2, Supporting information). To standardize conditions with respect to flower abundance, only orchards with more than 75% of inflorescences in flower were surveyed, yielding 62 surveys. Mango surveys were conducted only during peak mango flowering season (August and September 2009; two surveys per orchard about 4 weeks apart), in the centre of the orchard, along a 60 × 2 m transect comprised of two linear 30-m sections parallel to mango rows. In orchards near NFCAs, distance between the start of the transect and NFCAs was c. 40–120 m (depending on orchard size). Transects were walked slowly for 10–15 min; when insect activity was encountered, the observer would observe for a minimum of 5 s, recording all flower visitors that contacted the stigma or anthers. Immediately after observation, flower visitors were collected whenever possible for later identification. The total number of flowers in the transect was estimated by counting the number of open flowers of three randomly chosen inflorescences, averaging this value and then multiplying this by the total number of inflorescences in the transect. Flower visitor abundance data were divided by the total number of flowers observed, then multiplied by the overall average number of flowers per transect. As flower visitation rates to mango are low in this study area (Carvalheiro et al. 2010), and rare interactions might easily be missed by this survey method, species richness surveys were complemented with three 5-min collection surveys conducted simultaneously next to the transect by a second observer, using three randomly chosen mango inflorescences in which the observer captured all flower visitors observed. Any new interactions detected during this observation period were included in the data set as rare interactions (frequency of occurrence = 0·01). All flower visitors were collected and sorted to morphospecies and subsequently sent to professional taxonomists (see 'Acknowledgements') for identification.
As A. greatheadii is very attractive to honeybees (see 'Results'), and orchards have a high density of managed honeybees, we divided mango flower visitors into three subgroups for data analysis: ants, honeybees, and other flying visitors. Distances from the centre of each orchard to natural habitat were measured using maps based on aerial photographs taken in 2008 (corrected for later landscape changes), using arcgis 9.3® (2008; Environmental Systems Research Institute, Redlands, CA, USA).
To test whether spatial structure contributed significantly to the data variability, for all response variables, (log transformed to normalize residuals) we compared a null model with a model that includes a residual spatial correlation structure (linear and exponential variograms were considered, see Pinheiro & Bates 2000). No spatial autocorrelation was detected (see Table S3, Supporting information), and so, model selection procedures were carried out using generalized linear mixed models (GLMM) to assess how mango flower visitor abundance (honeybees and other visitors) and species richness (Poisson error distribution corrected for overdispersion when necessary, r package lme4) and the proportion of honeybees (Gaussian error distribution, with logit transformation) are affected by landscape context, management practices and NFCAs, using distance to natural habitat, use of pesticides (organic vs. conventional) and cultivar as fixed variables, and survey date within orchard as random variable. All possible combinations of explanatory variables and their interactions were considered. The most parsimonious model was selected as that with the lowest Bayesian information criterion (BIC). Whenever differences in BIC were not clear (<2 unit difference), we used the Akaike information criterion (AIC) with a second-order correction for small sample sizes, AICc (Burnham & Anderson 2002).
Effect of small patches of perennial native flora on mango production
To ascertain whether changes in the flower visitor community translate into changes in mango production, we gathered data on early fruit set (i.e. number of unripe fruits per tree, c. 6 weeks after the end of flowering ceased) for all 32 orchards used for flower visitor surveys. Within each orchard, thirty trees were randomly selected and all developing fruits counted. As carrying capacity might be limited (Bos et al. 2007), and fruit abortion might reflect pollination quality (Papadakis, Protopapadak & Therio 2009), early fruit set is only an indication of pollination efficiency but may not be a good indicator of pollination quality and, hence, of final crop production and economic value. Therefore, we also obtained information on final production (kg of commercially suitable mangoes as either fresh fruit or as processed products per tree) directly from farmers, for 225 orchards far from NFCAs (i.e. at least 250 m away) and 19 orchards near NFCAs (i.e. 50–150 m from NFCAs), which included the orchards used for flower and early fruit set surveys.
As with flower visitation data, we first evaluated spatial autocorrelation of all response variables (log transformed to normalize residuals and analysed with Gaussian error distribution). When spatial structure of the data contributed significantly to explain variation (i.e. lowered AIC values, Table S3, Supporting information), the residual spatial correlation structure was maintained in all subsequent analyses (using nlme r package, Pinheiro et al. 2012). We assessed effects of distance to natural habitat, pesticide use (presence/absence), cultivar and NFCAs on early fruit set and final production, by testing all possible combinations of variables and their interactions and selecting that which yielded the most parsimonious model (i.e. lowest BIC, AICc). For the vast majority of orchards (231 of 244) for which farmers provided final production information, we did not have information on flower visitation, so we could not directly test the influence of visitation on production. As an additional control, to confirm if production of orchards located near NFCAs was similar to all other orchards prior to NFCA creation, we repeated the analyses with data from the same orchards from the previous year (2008), when no NFCAs were present, and hence, no effect was expected. Costs and gains of NFCAs were evaluated based on 2009/2010 costs of material, labour, and mango commercial value.
All statistical analyses were performed using r (R Development Core Team 2010).