Even small forest patches increase bee visits to flowers in an oil palm plantation landscape

Pollination sustains biodiversity and food security, but pollinators are threatened by habitat degradation, fragmentation, and loss. We assessed how remaining forest influenced bee visits to flowers in an oil palm‐dominated landscape in Borneo. We observed bee visits to six plant species: four crops (Capsicum frutescens L. “chili”; Citrullus lanatus (Thunb.) Matsum & Nakai “watermelon”; Solanum lycopersicum L. “tomato”; and Solanum melongena L. “eggplant”); one native plant Melastoma malabathricum L. “melastome”; and the exotic Turnera subulata Smith “turnera”. We made one local grid‐based and one landscape‐scale transect‐based study spanning 208 and 2130 m from forest, respectively. We recorded 1249 bee visits to 4831 flowers in 1046 ten‐min observation periods. Visit frequency varied among plant species, ranging from 0 observed visits to S. lycopersicum to a mean of 0.62 visits per flower per 10 min to C. lanatus. Bee visitation frequency declined with distance from forest in both studies, with expected visitation frequency decreasing by 55% and 66% at the maximum distance from forest in each study. We also tested whether the distance to the nearest oil palm patch, with a maximum distance of 144 m, influenced visitation, but found no such associations. Expected visitation frequency was 70%–77% lower for plants close to a 200 ha forest fragment compared with those near large continuous forests (>400 ha). Our results suggest that, although found throughout the oil palm‐dominated landscape, bees depend on remaining forests. Larger forests support more bees, though even a 50 ha fragment has a positive contribution.


| 19
POWER Et al. incorporating forest and other vegetation has become an important issue for plantation owners and planners (Meijaard et al., 2016;Yahya et al., 2017). One motivation for this is to maintain a range of different taxa, including wild pollinators, within the landscape.
Although pollinators are mobile, they can be greatly affected by habitat fragmentation. The conversion of native forests to cultivated land has the potential to cause the loss of both feeding sites (sources of pollen and nectar) and suitable nesting habitat for pollinators (Patrício-Roberto & Campos, 2014). The habitat requirements of pollinators can be complex (e.g., different nesting and foraging habitats; Antoine & Forrest, 2020;Westrich, 1996), which makes them particularly sensitive to habitat loss and fragmentation. Most (an estimated 94%) tropical flowering plants are animal-pollinated (Ollerton et al., 2011). A decline in pollinators thus impacts the reproduction of wild plants and consequently entire ecosystems (Burkle et al., 2013;IPBES, 2016;Potts et al., 2010). Landscape changes, including those within the remnant fragmented areas, may cause loss of genetic variability and population stability, potentially leading to the disappearance of populations (Patrício-Roberto & Campos, 2014;Sodhi et al., 2004) and having severe effects on pollinator services (Potts et al., 2010).
Many food crops require animal pollination to maximize fruit set, size, and quality (Ollerton et al., 2011). Even oil palm is primarily pollinated by African weevils (Elaeidobius kamerunicus Faust) that increase fruit set and yield (Caudwell, 2001;Zulkefli et al., 2021).
Wild pollinators in particular contribute to the productivity and viability of many crops (Garibaldi et al., 2013) and thus to food security and nutrition (Ellis et al., 2015;Garibaldi et al., 2016). Evidence of this exists for many crops including watermelon (Sawe et al., 2020), tomatoes (Cooley & Vallejo-Marin, 2021;Neto et al., 2013), and chilies (Landaverde et al., 2017). Land-use conversion can disrupt pollination services on which the crops depend (see, e.g., Klein et al., 2003;Ricketts et al., 2008). Such disruptions add to the pollinator declines already seen worldwide, reflecting not just habitat loss but also damaging land-use practices (e.g., pesticide use) and climate change (Potts et al., 2010). While oil palm has raised incomes and living standards in Indonesia (Qaim et al., 2020), there are concerns that diet quality may have declined (Food Security Council et al., 2015;Ickowitz et al., 2016). One possible reason is the difficulty of producing highly nutritional, pollination-dependent food crops in landscapes with insufficient pollination services.
We recognized that a better understanding of pollinator activity may contribute to improved planning and management of the landscape to maintain local pollinators and their beneficial role. Here, we assess bee activity within a landscape of industrial oil palm and remnant forest in West Kalimantan, Indonesia. Our main aim was to determine whether the number of bee visits to flowers is affected by distance to forest. We hypothesize that the frequency of bee visits to flowers of the selected plant species will decrease with distance from forest. A further aim of our study was to test whether flower visitation rates by bees were dependent on the distance to the nearest planted oil palm patch. We expect palm trees to have conditions more similar to forest and to provide more resources for bees than agricultural fields, fallows, and other open land. Therefore, we hypothesize there will be an increase in bee visits when flowers are closer to planted palm compared with non-forested areas. To answer our study questions, we observed bee visits to six plant species and related the visitation frequencies to distance from forest while controlling for weather and time of day. To our knowledge, this is the first study to assess bee visitation frequency to both wild plants and food crops within an oil palm-dominated landscape.

| Study area
This study was conducted from June to November 2017 within the concession of PT Kayung Agro Lestari (KAL) in Kabupaten Ketapang in the province of West Kalimantan, Borneo, Indonesia (1°26′00.0″S 110°13′00.0″E; Figure 1a). The plantation is owned and managed by PT Austindo Nusantara Jaya (ANJ), a member of the Roundtable on Sustainable Palm Oil (RSPO) (PT Austindo Nusantara Jaya Tbk, 2016). Before conversion to oil palm plantation, the landscape was primarily logged-over natural forest (~8600 ha) and degraded land (Meijaard et al., 2016). Conversion started in 2010 and by 2016, 12,061 ha out of the 17,998 ha had been planted (Meijaard et al., 2016;PT Austindo Nusantara Jaya Tbk, 2016). Here, we use the terms "oil palm plantation" and "plantation" when referring to the entire area inside of the concession and "planted oil palm" when referring to sections of monoculture planted oil palm within the plantation.
The majority of the planted oil palm grows on shallow peat (~63%), with less on mineral soil (~33%) and sands (~4%). The palms are planted about 9 m apart resulting in a mostly closed canopy. Intensive maintenance, including regular physical clearance of ground vegetation and application of herbicides, results in little understory vegetation among the planted palms. Sixteen forested areas (20-2333 ha), 21% (3884 ha) of the concession, have been identified as having High Conservation Value (HCV) and are regularly monitored by the company (Meijaard et al., 2016).

| Study species
We studied bee visits to six angiosperm species (Table S1). Four of these (Capsicum frutescens L. "chili," Citrullus lanatus (Thunb.) Matsum & Nakai "watermelon," Solanum lycopersicum L. "tomato," and Solanum melongena L. "eggplant") are common local crops. The other two, the native Melastoma malabathricum L. and the introduced Turnera subulata Smith, have a wide distribution throughout the plantation. All the focal species are non-native except M. malabathricum, which is a common colonizing plant that occurs in cleared, degraded areas near forest edges within the plantation. T. subulata is planted as a method of bio-control for leaf-eating caterpillars (fireand bagworms), common pests to oil palm (Rashid et al., 2014).

| Study design
We conducted two studies: The first (hereafter referred to as the "Grid Study") was a systematically planned study of crop plants within an extensively cleared area of several hectares and the second was a transect study (hereafter referred to as the "Transect Study"). The Grid Study spanned up to a maximum distance of 208 m from forest and 144 m from planted palm (Figure 2a), while the Transect Study spanned from the forest edge up to 2130 m from natural forest.
Transect locations were chosen to represent a range of forest sizes and distances from these (Table S2). Forest 1 is a large, continuous forest that extends beyond the plantation boundary; Forest 2 is a conserved forest within the plantation that extends beyond the boundary; and Forest 3 is an isolated secondary forest hill surrounded by oil palm. The Grid Study was conducted in relation to Forest 4, an isolated hill that has been classified as a high conservation value area.

| Grid Study
We conducted the Grid Study between 22 July and 5 September

| Transect Study
We conducted the Transect Study from 15 to 29 October 2017. We

| Direct flower visit observations
To estimate flower visitation frequencies, we observed pollinator visits to flowers on all the above-mentioned plant species. We define a visit as a pollinator making apparent contact with the stigma or anthers of the preselected flowers. Each observation period lasted 10 min, and all were conducted by the same observer. During the observation period, all observed pollinator visits were recorded, but later we focused on analyzing only the bee data. Due to taxonomic challenges and the low numbers of visits from some morpho-species, we analyzed all bees as one group. Specimens in an adjacent study were collected and photographed (Hessen, 2020), and visual identification of these specimens was carried out by John S. Ascher based on diagnostic characters documented in Soh and Ascher (2020).

| Flower visit observations with cameras
We used Brinno BCC200 Pro cameras to perform additional flower visit observations. We used a T1 Clamp tripod to attach the camera to a wooden pole that would stand vertically when placed into the ground ( Figure 1b). We adjusted the focus of the cameras manually during each setup. The cameras were set to have a frame rate of one picture per second and with a resolution of 1280 × 720 pixels  (Table S3). The photographs did not allow for identification of pollinators below order level.

| Environmental variables
For each observation period, we recorded time of day, temperature, and relative humidity with a Suncare thermo-hydrometer (model 303C). We subjectively categorized wind, wetness of the vegetation, and sun exposure (Table S4). We also obtained data on daily rainfall from a weather station at the plantation, and we used a weather logger (UA-002 HOBO) placed at a fixed point within the plantation to record light intensity and air temperature at 3-hour intervals. We obtained additional weather data from a meteorological station in Ketapang (~50 km from the study site) and in Pontianak (~188 km from the study site; Table S4).
In the Grid Study, the mean temperature of the observation periods was 28.8°C (23.8-34.0°C) and the mean humidity was 72.4% (50%-96%). In the Transect Study, the mean temperature of the observation periods was 28.5°C (25-32.4°C) and the mean humidity was 79.4% (60%-94%).

| Variables
We collected data on various factors that might influence pollinator activity. We placed the variables into five categories: weather (including temperature, humidity, precipitation, and sunlight), temporal  Table S4 for more details on all of the variables we identified.

| Analyses
All data analyses were performed using R (version 3.5.1 with macOS version 10.14.6; R Core Team, 2018). We conducted initial data exploration following Zuur et al. (2010) on all variables (Table S4).
To analyze the relationship between flower visitation frequencies and a number of explanatory variables, we generated generalized linear mixed models (GLMMs) with a Poisson error (log link) distribution. Number of visits was used as the response variable and the number of flowers observed was included as an offset variable in all models, following Reitan and Nielsen (2016). Observation ID was included in each model as a random effect to account for overdispersion (Harrison, 2014). Other variables considered as random effects include transect, plot, day, and recording ID. All models were generated using the "glmer" function in the R package "lme4" version 1.1-15 (Bates et al., 2015) with the "bobyqa" optimizer. Continuous variables were centered and scaled using the "scale" function (R Core Team, 2018). We used an information-theoretic approach to identify the most parsimonious model using the Bayesian information criterion (BIC). Variance inflation factors (VIFs) were assessed using the "vif" function in the R package "car" (Fox & Weisberg, 2011). Dispersion, zero-inflation, and uniformity were tested using "testDispersion," "testZeroInflation," and "testUniformity" functions in the R package "DHARMa" version 0.3.2.0 (Hartig, 2020). Confidence intervals were calculated using the Wald method with the "confint.merMod" in the R package "lme4" version 1.1-15 (Bates et al., 2015). Pseudo R 2 values (delta method) were generated for each model using the "r.squaredGLMM" function in the R package "MuMIn" version 1.42.1 (Bartoń, 2018). Figures 3 and 4 were created using "ggplot" function in the R package "ggplot2" version 3.3.2 (Wickham, 2016). Effect of predictors for each model was generated using "allEffects" function in the R package "effects" version 4.1.0 (Fox, 2003(Fox, , 2019.

| RE SULTS
The field observations revealed a diversity of flower-visiting insects with bees, the most frequently observed visitor to all species, making up 81.4% of the total number of observed visits ( Figure S2a,b).
Specimens collected in an adjacent study (Hessen, 2020) show As we recorded no visits to S. lycopersicum flowers (n = 56 observation periods), we excluded this species from further analyses. The flower visits recorded in the two studies were analyzed separately.

| Factors explaining variation in visit frequency in the Grid Study
Our best model (Model 1) explaining how visitation frequency to flowers varied in the Grid Study (R 2 m = 0.520, R 2 c = 0.947) included distance from forest, plant species, sun, time of day, and sampling method (camera or manual observation) as fixed effects ( Table 2).
The estimated relative contribution of explained variation for each variable in the model is listed in Table S5a. The estimated effect of each predictor (based on Model 1) with all other variables being held constant is listed in Table S6a. VIF for each variable is <2. Temperature and humidity were highly correlated with time of day, and thus, we were unable to disentangle the effects of these three variables. Therefore, although we expect temperature and humidity to play an important role, they were not included in the best model.

Citrullus lanatus
We did not anticipate the observation method would influence the number of observed visits. However, analyses showed that the cameras revealed a higher visit frequency than human observations.

| Factors explaining variation in visit frequency in the Transect Study
We developed two models to describe how visit frequency to  (Table 3) Model 2.2 included distance from forest (as a factor: <450 m or >800 m from forest), forest ID, sun, time of day, and camera as fixed effects. Visitation frequencies were significantly higher near forests, with expected visitation frequency being 66.2% lower at distances greater than 800m from the forest edge than at distances less than 450 m (Figure 4d). The other fixed effects showed similar patterns as for Model 2.1 (Table 4).

| DISCUSS ION
Once environmental factors were accounted for, flower visitation frequency by bees was influenced by the distance to the nearest forest. This relationship was observed in both studies despite the difference in spatial scale.
The results from each study indicate expected visitation frequency to decrease by 55.4% at 208 m from forest, and 66.2% at >800 m from forest, respectively. Declines in bee visit frequency with distance from natural habitats have been found in agricultural systems elsewhere, for example, with coffee flowers having higher visitation frequency near native forests in Costa Rica (Ricketts, 2004); mustard and radish flowers having higher bee visitation near natural grasslands within an agricultural landscape in Germany (Steffan-Dewenter & Tscharntke, 1999); and watermelon flowers having higher visitation near oak woodland and chaparral habitat on farms in California (Kremen et al., 2002). The relationship between flower visitation and distance from forest suggests the forests act as a source of pollinators which may forage among the oil palms but reside in more natural habitats. The lack of relationship between flower visitation and distance from planted oil palm suggests the plantation does not provide resources comparable to the native forest.
Flower visitation frequency was affected by the nearest forest as seen in the Transect Study, where flowers in proximity to Forest 3 (the smallest fragment) had an expected visitation frequency 70.3% lower than flowers near Forest 1 and 76.5% lower than flowers near Forest 2 (the larger forests). Studies elsewhere have indicated that various pollinators are more abundant in or near large primary forests than small forest fragments and plantations (Beck et al., 2002;Liow et al., 2001;Lucey & Hill, 2011;Mayfield, 2005). Despite this, we found that even a 54 ha forest patch boosts flower visitations in the surrounding plantation landscape. Similar findings of small forest fragments positively affecting potential pollinators have been observed in other landscapes, for example with forests as small as 0.24 ha supporting a diverse bee assemblage in an agricultural landscape in Costa Rica (Brosi et al., 2008). Although conserving large intact forests remains crucial (Edwards et al., 2011), small damaged forests should also be protected where practical as this will help maintain bees and other taxa (Benedick et al., 2006). This was likely due to the height-biased selection of observed flowers (due to T. subulata bushes being taller than the camera setup), camera placement, and more observations outside of peak activity time. These findings suggest results are highly dependent on camera setup and flower selection.
The variation in visit frequency among the plant species indicate their differing levels of attraction for available pollinators. Strikingly, S. lycopersicum -from a genus known for its pollen-only, nectar-free flowers that rely on specialized "buzz-pollination" (Vallejo-Marín,

2019) -received no visits. This plant originates in the Americas
where there appear to be effective pollinators (Franceschinelli et al., 2013;Rosi-Denadai et al., 2020). In addition, C. frutescens -another neotropical Solanaceae, known for effective self-pollination (though also pollinated by bees in the Americas [Knapp, 2010]) -had the lowest visitation aside from S. lycopersicum. In contrast, C. lanatus, though also an exotic (originating in Africa), is less specialized and had the highest visit frequency.
However, fruit set and quality were not assessed in this study due to high plant mortality, partly as a result of the harsh conditions that emerged in the study site (high temperatures and droughts).
We, therefore, can only make conclusions about visits and not pollination adequacy, since visits may not translate into effective pollination events. Here, we compare our observed visits to the known requirements of C. lanatus to speculate about the pollination adequacy of our study location. It has previously been shown that C. lanatus flowers require 6-8 honey-bee visits, or just one bumble bee visit, in order to achieve optimum fruit set (Adlerz, 1966;Bomfim et al., 2016). This translates to a visit frequency of at least 0.11 visits per 10 min per flower for the day the flower is open, which is lower than our observed mean visit frequency for C. lanatus (0.62) (mean at <50 m distance from forest: 0.83 visits per 10 min per flower, mean at >200 m from forest: 0.36 visits per 10 min per flower). This suggests sufficient visits in our study for optimum fruit set and quality for C. lanatus. In contrast, S. lycopersicum clearly receives insufficient visits to achieve optimum yields (zero visits in the entire study), but this is not linked to distance from forest. We stress that pollination adequacy for crops within the oil palm dominated landscape requires further investigation.
As mentioned above, the study period was dry and hot with plants requiring watering twice a day to avoid wilting and death.  (Hardwick et al., 2015;Luskin & Potts, 2011;Ramdani et al., 2014) and the increased heat and reduced rain already seen across the island of Borneo (McAlpine et al., 2018). These climate-driven impacts are making small-scale agriculture harder and riskier. The local and island level impacts are likely to become more severe as plantations spread, forest cover declines, and the global climate gets warmer and less predictable . This may ultimately impact not only small holder agriculture but also the plantations themselves .
Pollinator density and richness have been shown to improve yields in various pollinator-dependent crop systems across different ecosystems, with flower-visitor density being the most important predictor of crop yield globally .
Land-use intensification can disrupt pollinators and pollinator services by causing declines in both pollinator species and functional richness (Rader et al., 2014). A wide range of Southeast Asian taxa including bees (Liow et al., 2001), butterflies , and moths (Beck et al., 2002) are experiencing declines in species richness and population density due in part to increasing forest disturbance . While pollination limitation is a concern, especially for specialized crops such as S. lycopersicum, adequate pollination is more likely for crops grown sufficiently close to natural forests (Klein et al., 2003). In this study, bees were found throughout the oil palm plantation, but with significantly higher bee visitation to flowers near forests. This relationship was observed even with the smallest forest fragment, though flowers near the larger forests had the highest visitation frequency. Our results emphasize the importance of maintaining as much native forest as possible within and around the agricultural landscape to sustain pollinator availability. We encourage further research to focus on pollination adequacy within oil palm landscapes.

CO N FLI C T O F I NTE R E S T
The project was developed as a formal cooperation with PT Austindo Nusantara Jaya Agri. The company hosted and facilitated our work on the basis that we would have freedom to publish our results without interference. We agreed in advance that they could, based on their review of our drafted articles, request us to withhold any specific details judged sensitive for commercial reasons. This review has been completed, and no such requests were made.

AUTH O R CO NTR I B UTI O N S
DS conceived the project in close cooperation with CIFOR, PT Austindo Nusantara Jaya Agri, Tanjungpura University and Borneo Futures. DS and AN designed, managed, and supervised the larger project. DS, AN, and CP designed the specific study described here. CP collected and analyzed the data and wrote the original manuscript with DS and AN providing supervision and critical inputs. All coauthors reviewed and edited the final manuscript and approve the final version.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in the Dryad Digital Repository: https://doi.org/10.5061/dryad. s4mw6 m96h (Power et al., 2021).