Local-scale tree and shrub diversity improves pollination services to shea trees in tropical West African parklands

1School of Natural Sciences, Trinity College Dublin, Dublin 2, Ireland; 2RSPB Centre for Conservation Science, The Royal Society for the Protection of Birds, The Lodge, Sandy, UK; 3Department of Culture, Heritage and the Gaeltacht of Ireland, National Parks and Wildlife Service, Dublin 2, Ireland; 4Naturama, Ouagadougou, Burkina Faso; 5Laboratoire de Biologie et Ecologie Végétales, Institut des Sciences, Université Ouaga I Pr Joseph KI-ZERBO, Ouagadougou, Burkina Faso; 6Village of Torem, Po, Centre-Sud, Burkina Faso; 7BirdLife International, Cambridge, UK and 8University of Cambridge Institute for Sustainability Leadership, Cambridge, UK


| INTRODUC TI ON
An estimated 96% of flowering plants in the tropics benefit from animal-mediated pollination (Ollerton, Winfree, & Tarrant, 2011). On a global scale, there is concern about decline of pollinating insects, particularly bees, although most evidence for pollinator decline comes from North America and Europe. Despite their cultural, nutritional and economic value (IPBES, 2016), relatively little is known of the status of pollinators and pollination services in lower latitudes (Brown & Paxton, 2009). Pollinator declines are frequently linked to land use (Potts et al., 2010), and agriculture has expanded and intensified in the tropics in recent decades (Gibbs et al., 2010;Stéphenne & Lambin, 2001). Landscape changes, particularly landscape homogenization, have been shown to affect pollinator abundance and diversity in tropical systems, with negative effects on certain crop yields (Dainese et al., 2019;Klein et al., 2012).
Loss and fragmentation of natural habitat in the landscape affects the availability of resources such as food and shelter, alters the impacts of competing species and pathogens (Kremen et al., 2007) and, ultimately, influences the communities of pollinating insects.
Impacts of reduced area of natural habitats can lead to reduced delivery of pollination service (Banks et al., 2013). Furthermore, in highly modified landscapes with few natural areas, mobile, generalist species such as honeybees tend to persist, while smaller or less mobile bee species are scarcer (Klein et al., 2012).
On a local scale, relationships between floral diversity and pollinator abundance and diversity in subtropical and tropical systems are inconsistent, particularly where woody species grow in mixed cropping systems. Norfolk, Eichhorn, and Gilbert (2016) found a positive relationship between ground flora diversity, pollinator density and yield within almond orchards in Egypt, but no relationship between flower visitors and proximity to natural habitats. Here, orchards provide more resources for pollinating insects than the arid habitats surrounding them, demonstrating the importance of landscape context. Conversely, higher species richness of non-crop trees has been linked with lower pollinator abundance in coffee farms, possibly due to competition for pollinators (Boreux, Krishnan, Cheppudira, & Ghazoul, 2013). On a landscape scale, maintaining natural habitats close to agricultural areas has been shown to increase pollinator diversity by supporting non-Apis bees (Klein, Steffan-Dewenter, & Tscharntke, 2003), an important factor influencing the effectiveness of pollination services (Garibaldi et al., 2011;Woodcock et al., 2019).
The importance of local-and landscape-scale factors in delivery of pollination services can be linked, for example Motzke, Klein, Saleh, Wanger, and Tscharntke (2016) found that bee abundance in home gardens in Indonesia was greatest when flower cover within the garden was high and distance to a forest fragment was <100 m.
In the West African Sudano-Sahel, shea parkland is a traditional agroforestry system practised from Senegal in the west to Kenya in the east (Figure 1). The parkland is composed of cultivated fields where trees valued by local communities, primarily shea Vitellaria paradoxa Gaertn, are planted or regenerate naturally during fallow periods (Boffa, 2015;Gijsbers, Kessler, & Knevel, 1994). Shea is nutritionally valuable to people as the fruits ripen when food is scarce, and the pulp is a source of protein, fibre, calcium and potassium (Maranz, Kpikpi, Wiesman, De Saint Sauveur, & Chapagain, 2004). Shea nuts are also commonly processed to produce shea butter, which is an important source of dietary fats (Victoria & Ajay, 2015). The nuts and butter are commercially valuable: traded locally providing women with a rare source of cash income and internationally for use in the cosmetics and confectionary industries (Pouliot, 2012). Shea is heavily dependent on bees for pollination, and a range of bee species have been recorded visiting flowers, although honeybees are thought to be more effective pollinators than smaller bees (Lassen, Nielsen, Lompo, Dupont, & Kjaer, 2016;Stout et al., 2018).
Agricultural intensification and expansion in shea parklands in recent decades (Stéphenne & Lambin, 2001) has resulted in longer cultivation periods within fields, shorter fallow periods and, in some cases, abandonment of the practice of fallowing (Augusseau, Nikiéma, & Torquebiau, 2006). Trees regenerate during fallow periods, and these changes have led to reduced diversity and abundance of trees in parkland fields (Kaboré, Bastide, Traoré, & Boussim, 2012), and loss of semi-natural habitats including scrub and woodland from parklands (Gijsbers et al., 1994). Such environmental perturbations can result in changes to plant-pollinator interactions, few pollination events or the transfer of less pollen, and thus pollination limitation and decreases in fruit and/or seed production (Knight et al., 2005).
Given the intensification of agriculture and encroachment on natural areas in West African shea parklands, and the association between habitat loss and fragmentation with decreased pollinator abundance and diversity, pollination services in shea parklands may be negatively affected by human activities. As shea relies heavily on bees for pollination (Lassen et al., 2016;Stout et al., 2018), changes in agricultural practices may have a negative effect on pollinating insects, which could contribute to threats to food security and the economic well-being of an estimated 45.1 million rural people living in the shea belt (Naughton, Lovett, & Mihelcic, 2015). We aimed to quantify the relationship between species diversity in fields in shea parkland, the area of semi-natural habitat in the wider landscape and provision of pollination services to shea trees by testing the following hypotheses: 1. Shea trees in West Africa are pollination limited.
2. The proportion of natural habitat surrounding sites and plant diversity within sites influence the visitation rate of bee species that Pollination limitation, bee visitation, and fruit set and weight were used as proxies to measure the effectiveness of pollination services.

| Study species
Shea, or Vitellaria paradoxa, is the only member of its genus in the family Sapotaceae. Its range is restricted to the Sudano-Sahelian zone of western and central Africa ( Figure 1) and it comprises two subspecies: Vitellaria paradoxa ssp. nilotica in the east and Vitellaria paradoxa ssp. paradoxa in the central and western parts of its range (Naughton et al., 2015). Shea trees bloom during the dry season, between January and April, with peak fruiting occurring by the end of July, although anecdotal evidence suggests flowering and fruiting periods can vary yearly. Flowers are born on inflorescences located at the tips of branches. The flowers are short lived, but buds, receptive flowers and older blooms occur simultaneously over a long flowering period (Stout et al., 2018). The flowers are protogynous, with the stigma emerging from the bud and becoming receptive before the sepals part to reveal the petals and stamens (Hall, Aebischer, Tomlinson, Osei-Amaning, & Hindle, 1996). Although this reduces the likelihood of a flower self-pollinating, the presence of flowers at different stages of maturity on the same inflorescence indicates that within-inflorescence pollination is possible.
The success of pollination by flowers on the same inflorescence is not well understood. Shea produces fruit with sweet pulp surrounding a large seed (the shea 'nut'), and occasionally more than one nut may be produced in a single fruit.

| Site selection
This study was conducted in the Centre-Sud region of Burkina Faso, south of the Kaboré Tambi National Park (KTNP) and east of Nazinga Game Ranch ( Figure 1). All data were recorded between February and June 2017.
We selected ten 1-ha sites (100 m × 100 m) in fields which had been cultivated during the previous rainy season. Sites were organized into five pairs located 2-3.5 km apart, and the minimum distance between pairs was 4 km. The paired design allowed us to minimize the impacts of external factors (e.g. distance to a large town or nature reserve). Within each pair, we selected one site with low diversity of woody species and one site with higher diversity. All trees and shrubs, including shea, >3 m in height were identified and counted in each site, and the mean inverse Simpson's index (1/D) was calculated to represent site diversity. F I G U R E 1 Northern Africa with shea parkland adapted from Naughton et al. (2015) represented by hatching (main map). The inset shows the 10 study sites in southern Burkina Faso. Unshaded areas are primarily managed as agroforestry while shaded areas are under management for conservation

Ghana Burkina Faso
We did not include small trees and shrubs (<3 m in height) because local knowledge indicated they were unlikely to flower (an observation that proved true). Herbaceous species were also excluded from site diversity calculations because we worked in ploughed fields that were open to grazing animals, with low cover of herbaceous plants (<0.01% cover). We mapped and calculated the area of semi-natural habitat within a radius of 1 km from the centre of each site in ArcMap 10.4 using satellite imagery provided by ESRI world maps and Google Earth (accessed in September 2017, photographs dated to 2013). Restricting the radius to 1 km avoided an overlap in mapped habitat between sites. Although the foraging range of A. mellifera can greatly exceed this distance, studies based on waggle dances and mark-recapture indicate that African honeybee races, including A. mellifera adansonii, focus the majority of their foraging within distances of 1 km of the hive or previous feeding stations (Roubik, 1999;Schneider, 1989). Diversity (1/D) of woody species within our 10 sites (site diversity) and area of semi-natural habitats then acted as the independent variables (see data analysis).
We collected data regarding pollination services from 10 shea trees within each site that met the following criteria: selected trees had a single trunk, were >3 m tall, appeared healthy and had developing flower buds. If more than 10 trees in a site met the criteria, the 10 selected trees were distributed as evenly as possible through each site. The number of trees and inflorescences varied between sites because three comparable pairs of inflorescences were not present on every tree (Table 1). In total, 2,248 flowers were included in the analysis: 1,096 untreated flowers and 1,152 flowers with pollen supplementation by hand. On average, the difference in the number of hand pollinated and control flowers at each site was 2.4%. The greatest difference was 8.6%. All results relating to flowers and fruit set are from these selected flowers.

| Flower visitors
We surveyed flower visitors during two 10-min recording episodes per tree, one between 6 and 8 a.m. and another between 4 and 6 p.m., when insect visitors are most active, on each of three dates during the flowering season (January-March 2017; Stout et al., 2018). If there were no receptive flowers on a tree, no attempt was made to record visitation (Table 1). All bees observed visiting inflorescences during recording episodes were caught. The native honeybee, Apis mellifera adansonii Latreille could be identified easily, but other bees were placed in 70% alcohol solution and sent for expert identification.

| Fruit set
We counted fruit set in early May 2017, when the selected flowers had developed small fruits, but before fruits ripened and fell. Ripe fruits were collected and their nuts were weighed in the field in June to August 2017. Initially, fruits were hand harvested (n = 35), but proved to be unripe. The remaining (n = 133) fruits were collected by placing bags around them, so they could fall naturally as they ripened without being lost. Only the ripe fruits collected in bags were included in the analysis of fruit weight.

| Pollination limitation
We assessed pollination limitation in a mixed effects model using the proportion of flowers that formed fruit per inflorescence ('fruit set') as the dependent variable and pollen supplementation as a fixed factor (expressed as pollen supplementation = 1, no pollen supplementation = 0). Site was specified as a random factor, the binomial distribution was used and the model was weighted by the number of flowers originally selected on each inflorescence (Table 2).

| Spatial autocorrelation and correlation among independent variables
We tested the relationship between pollination limitation and distance between sites using the Incremental Spatial Autocorrelation tool in ArcMap 10.4. This tool indicates the minimum distance between independent sites by comparing the Moran's index of spatial correlation for data recorded over a range of distances with the expected values if no clustering occurs. Spatial autocorrelation was detected in sites <4 km apart (I = 0.85, z = 2.2, p = 0.03) but not in sites >4 km apart (I < 0.05, z < 0.4, p = 0.35). Because the distance between pairs of sites was >4 km, each pair can be considered independent of the others. There was no evidence of correlation between area of semi-natural habitat and site diversity (data not shown).

| Influence of site diversity and area of semi-natural habitat on pollination services
Five mixed effects models were used to assess the influence of site diversity and area of semi-natural habitat on pollination services (as assessed through bee visitation, degree of pollen limitation and fruit set; Table 2). Apis mellifera and non-Apis bees have been shown to act as two different functional groups in Burkina Faso (Lassen et al., 2016), and were treated separately in the analysis.
The visitation rate (number of bees recorded/number of recording episodes per tree) was therefore assessed in separate models for Apis mellifera and non-Apis bees. The degree of pollination limitation was expressed as the difference between the proportion of pollen-supplemented and untreated flowers that set fruit per tree (proportional difference). This approach reduced zero inflation and TA B L E 2 Design of mixed effects models used in the analysis. Hurdle models were used for the models investigating Apis mellifera visitation, non-Apis visitation and fruit set. In these two-part models, the dependent variable is broken into presence/absence and a quantitative measure for non-zero values before the analysis is conducted allowed calculations to be made when none of the flowers with supplementary pollen bore fruit. Fruit set was expressed as the mean proportion of untreated flowers that bore fruit per tree. The impact of pollen supplementation on nuts was assessed by comparing the weights of nuts from treated and untreated flowers (n = 133). In each of the five models addressing the influence of site diversity and area of semi-natural habitat on pollination services, site diversity (expressed as diverse = 1, less diverse = 0) and area of semi-natural habitat were specified as fixed effects. The area of semi-natural habitats was expressed as a proportion of the total area within 1 km of the site and scaled (the mean was subtracted from each value and the result was divided by the standard deviation) to enable the data to meet the requirements of the analysis.
Site was placed within pair as a random factor to account for the paired design.
As the data collected for bee visitation and fruit set contained a large proportion of zeros and were over-dispersed, hurdle models were the most appropriate tool for these analyses. In a hurdle model, the dependent variable is broken into two components:  (Table 2).
Because the model has two parts, two p-values are generated per hurdle model.

| Environmental variables
In total, 285 trees and shrubs >3 m tall, belonging to 19 species, were found in the sites (range: 1-7 species per site). Shea was the commonest tree (222 individuals), followed by Parkia biglobosa (27 individuals) and Lannea microcarpa (seven individuals). The area of semi-natural habitats surrounding each site ranged from 6% to 48% with an average of 31% (Table 3).

| Visitation by Apis mellifera and non-Apis bees
We In total, 347 non-Apis bees were caught at diverse sites and 164 were caught at less diverse sites. The best fit model for non-Apis bee visitation included diversity and omitted the area of semi-natural habitat, as for A. mellifera. However, for non-Apis bees, diversity was not related to the likelihood of bees being present at trees (β = 0.14 ± 0.62, p = 0.82), while visitation rate where non-Apis bees were present was significantly related to diversity (β = 0.57 ± 0.28, p = 0.04, Figure 3).

| Degree of pollination limitation
Pollination limitation was greater at less diverse sites (β = 0.44 ± 0.18, p = 0.015, Figure 3), but area of semi-natural habitat was not related to the degree of limitation observed. The best fit model included diversity but not area of semi-natural habitat.

| Fruit set and nut weight
Area of semi-natural habitat (but not diversity) had a significant relationship with the presence of fruit on untreated flowers (β = −0.95 ± 0.31, p = 0.003, Figure 4). The best fit model included area of semi-natural habitat but not diversity. Fruit abundance was not related to site diversity or area of semi-natural habitat, but a power analysis indicated that an effect size comparable to that observed for the degree of limitation was unlikely to be detected.
Of the 413 fruits that were counted in May 2017, 168 matured to collection. Fruit loss was attributed primarily to fructivory, although late-acting self-abortion is also possible, and some fruits may simply have matured and fallen before collection. Thirty-five fruits were harvested before they had reached maturity (see Section 2). Only 22 fruits of untreated inflorescences were collected for weighing and the remaining 111 fruits were from flowers that received supplementary pollen (Table 3). Mean weight of nuts of flowers with supplementary pollen was 18.4 g and of 16.6 g for untreated flowers, but the difference was not significant.

TA B L E 3
Summary of variables recorded at study sites. Degree of limitation is the mean of the proportion of fruit set of flowers with supplementary pollen minus that of untreated flowers per tree

| D ISCUSS I ON
We found strong evidence of pollination limitation in production of shea fruits. More visitation by bee species (the proportion of trees visited by A. mellifera and the number of non-Apis bees recorded per recording episode) was observed in sites with greater diversity of woody species, and pollination limitation was lower in these sites.
Additionally, we found a negative relationship between the proportion of untreated inflorescences that bore fruit on trees and the area of semi-natural habitat within 1 km.
Pollen limitation of fruit set is common and can be a consequence of habitat degradation (Knight et al., 2005)  Although A. mellifera occurred at more trees in diverse sites, their abundance where they occurred was not related to site diversity or area of semi-natural habitat. This may relate to foraging behaviour: A. mellifera tended to be abundant on only one tree at a site, even when other trees had a profusion of receptive flowers. This focus on a single tree may result from the within-nest recruitment of foragers to specific floral resources and the use of attractant scent marks on flowers (Free & Williams, 1983;Stout & Goulson, 2001).
Non-Apis bees were widespread in occurrence, but more abundant in more diverse sites. H. ruspoli was the bee species recorded most frequently and they drove this relationship. Lobreau-Callen, Le Thomas, Darchen, Darchen, and D'Amico (1990) (Lassen, Raebild, Hansen, Brødsgaard, & Eriksen, 2012;Stein et al., 2018), and we noted that H. ruspoli colonies were very numerous at our sites (>200 were noted within 10 m of study plots, data not shown). However, Lobreau-Callen et al. (1990)   Several lines of evidence of herbivory by insect larvae, mammals and birds were commonly observed (personal observations), but it is not clear if herbivores were attracted from nearby semi-natural habitats to developing shea fruit. This deserves further investigation, and the contribution of shea to supporting overall biodiversity in parklands should be taken into account when establishing a sustainable level of exploitation.
The lack of a relationship between semi-natural habitats and either bee visitation or pollination limitation could be due to the quality or fragmented nature of semi-natural habitats (fallow fields and fragments on rocky hills or in river corridors), which may have limited their value for pollinators. Alternatively, shea may not have been attractive enough floral resource to draw bees from semi-natural areas, particularly during the dry season, when bees prefer to forage in more natural habitats, moving to agricultural ones during the rainy season (Stein et al., 2018). In addition, we only looked at landscape composition, and Semi-natural habitat (ha) Proportion of trees with fruit as landscape configuration is important for pollinators elsewhere (e.g. Freitas et al., 2014), this deserves further attention.

| CON CLUS IONS
Our results show that woody species diversity in shea parklands supports pollination services of shea. We recommend that a range of native woody species are retained or planted into cultivated fields when fallows are cleared to support pollinators and pollination services.
Fragments of semi-natural habitat in the surrounding landscape may not be sufficient to provide the necessary pollination services. Shea is one of several insect-pollinated species that provide non-timber forest products and flower during the dry season in shea parklands. In a region that is subject to nutritional poverty and where the finance required to supplement local foods is lacking for most families, maintaining natural ecosystem services that support food provision must be a priority. Coulthard for project support and Ruth Kelly for statistical advice.

ACK N OWLED G EM ENTS
We thank the Naturama staff in Burkina Faso, the farmers and shea harvesters near Po for their hospitality and good-will, and Connal Eardley for providing expertise in African bee identification. We also thank Florence Hecq for reviewing the translated abstract.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available via the Dryad Digital Repository https://doi.org/ 10.5061/dryad.69p8c z8z0 .