Interacting pest control and pollination services in coffee systems

Bird activity alone did not change fruit set relative to excluding both birds and bees. In contrast, bee activity alone significantly increased proportional fruit set from 0.50 to 0.56 (representing an 11.0% increment) (Fig. 2A and Table 1). We found a significant interaction between bird and bee activity (Table 1), indicating synergistic effects between the pest control and pollination services. The combined effects of bird and bee activity on fruit set were significantly greater than their individual effects (Fig. 2A and Table 1), jointly increasing proportional fruit set from 0.50 to 0.62 (representing a 24.0% increment) compared to the neither activity treatment.

Both bee activity and fruit condition (i.e., bored vs. not bored) influenced average fruit weight, while bird activity alone did not (Table 1). Compared to treatments without bees, bees increased fruit weight from 1.41 to 1.47 g (representing a 4.2% increment). The same increment was found between bored and not-bored fruits. We found significant interactions between bird and bee activity for average fruit weight, as well as significant interactions between bee activity and fruit condition (Table 1 and Fig. 2 B and C). Fruits weighed more when they were not bored and came from branches where bees had access (1.53 g) compared to all other treatments (1.41 g), representing an 8.5% increment (Fig. 2B). Moreover, independently of fruit condition, average fruit weight was higher with bird and bee activity compared to all other treatments (0.09 g difference, representing a 6.6% increment) (Fig. 2C).

We also evaluated bird and bee activity effects on fruit weight uniformity, as measured by the fruit weight coefficient of variation (CV). We found no individual effects of bird activity, bee activity, or fruit condition on fruit weight uniformity but the interaction between bird and bee activity was significant (Table 1). We observed greater fruit weight uniformity on fruits from branches where both birds and bees had access, compared to all other treatments (Fig. 2D).

We found a significant interaction between birds, bees, and time affecting the observed proportion of bored fruits (Table 1). The proportion of bored fruits differed across treatments, and these differences changed as fruits matured (Fig. 3). In November, when the main harvest took place, the proportion of bored fruits was significantly lower in branches where birds were present (Fig. 3). Specifically, at this time, we observed the highest broca infestation rate when neither birds nor bees were present (0.14 proportion) and the lowest broca infestation rate when birds were present but bees were absent (0.07 proportion), resulting in a 52.8% reduction.

We estimated average coffee yield for our 30 study farms to be 12,889 kg/ha (±1,832) under natural conditions (i.e., bird and bee activity treatment). This potential yield translates to an expected gross income of US$4,317/ha (±614). Our experiments indicate that, with birds excluded, average yield would be reduced by 13.5% or 1,744 kg/ha (95% confidence interval [CI], 1,237 to 2,251), due to the combined effects on fruit set and fruit mass. This represents a financial loss of US$584/ha (95% CI, US$414 to 754). With bees excluded, average yield would be reduced by 24.5%, 3,161 kg/ha (95% CI, 2,242 to 4,080), representing a loss of US$1,059/ha (95% CI, US$751 to 1,367). Finally, we found the greatest reduction in average yield with both birds and bees excluded: 24.7% reduction, equivalent to losing 3,183 kg/ha (95% CI, 2,255 to 4,110) or US$1,066/ha (95% CI, 755 to 1,377). See SI Appendix for additional coffee yield and income calculations.

We conducted the study within the Volcanica Central Talamanca Biological Corridor (VCTBC) in Costa Rica. The VCTBC constitutes a heterogeneous landscape where productive lands interact with forest remnants within and outside protected areas across 1,146.3 km². In 2010, the area dedicated to coffee cultivation covered ∼8.5% of the total landscape, third in expanse after forests (51.1%) and pastures (25.3%) (Fig. 1A). We selected 30 coffee farms that represent local variation in shade, management and elevation (672 to 1,110 m above sea level) (Fig. 1A). Most farms were owned by smallholders with a coffee cultivated area ranging from 0.35 to 6 ha (only two farms had coffee cultivated area slightly greater than 20 ha). All farmers cultivated C. arabica L., most often the Caturra variety, but also varieties such as Obatá, Catimor, Catuai rojo, CR95, and Marsellesa (see SI Appendix for additional information on farm characteristics).

To assess the potential interacting contributions of pest control and pollination services to coffee productivity, we established a full-factorial experiment of bird and bee exclosure treatments (Fig. 1B). Before coffee flowering, between January and February 2019, we selected eight coffee plants of similar height and vigor in each coffee farm. All selected plants were located on the same coffee lot and planted at close distance, thus sharing location, and environmental and management conditions.

To assess bird pest control services, we enclosed four out of the eight coffee plants with a plastic mesh small enough to exclude foliage gleaning birds but large enough to allow bees and other small animals (20-mm mesh size) (23). A frame made of bamboo poles was used to maintain the plastic mesh in place, resulting in a permanent exclosure. The remaining four coffee plants were left open (i.e., not enclosed), allowing birds to forage freely. To assess bee pollination services, we selected four similar branches in each of the eight coffee plants (hereafter: experimental branches 1 to 4). We excluded bees using fine nylon mesh gauze bags in two out of these four branches (1 mm mesh size; handmade by a seamstress). We placed the bags over selected branches before the main flowering event and removed them once the flowering ended. In rare cases when we found opened flowers before setting the bags, we removed those flowers. The other two branches on each plant served as open controls to measure productivity under natural pollination. This full-factorial design resulted in four different treatments at the branch level (Fig. 1B): i) only birds were allowed access (bird activity alone), ii) only bees were allowed access (bee activity alone), iii) both were allowed access (bird and bee activity, i.e., control treatment), and iv) birds and bees were both excluded (neither activity). Furthermore, to make sure we accurately assessed broca infestation rates, we randomly selected four additional branches distributed in the upper and lower parts of the same eight plants (hereafter: branches 5 to 8) (Fig. 1B). We did not test for the effects from the experimental manipulations themselves from which there is inconsistent evidence (50). Finally, we checked bird exclosure nets regularly while the experiment was deployed to repair or replace nets if necessary and to make sure birds have been effectively excluded.

In May 2019, 12 to 18 wk after the main flowering event, we removed pollinator exclosures (i.e., nylon mesh gauze bags) and counted the number of early fruits at each experimental branch’s eight most distal nodes. We also estimated the number of flowers that opened at each node by adding the number of fruits to the number of flower scars and abnormal flowers (flowers that did not appear healthy/functional).

We visited every farm three more times between July and November 2019, and counted the total number of fruits and bored fruits present in all branches (i.e., experimental branches 1 to 4 and additional branches 5 to 8 used for broca assessment only). In branches 5 to 8, we counted fruits in all existing fruiting nodes (i.e., not limited to the eight most distal nodes). We collected all harvestable (ripe) fruits present in our experimental branches from the second visit onward. Immediately after harvesting, we measured fresh (wet) weight from each fruit individually and visually inspected for broca marks presence or absence. Due to weather conditions and differences in phenology, the period between visits differed among farms.

We calculated fruit set, average fruit weight, fruit weight CV (as a measure of fruit weight uniformity, a desirable condition for marketing and roasting), and proportion of bored fruits for each experimental treatment. Proportion of bored fruits was calculated based on information from branches 1 to 8. We calculated fruit set as the number of coffee fruits recorded in the second count (i.e., just before the initial harvest) divided by the number of initial flowers. We calculated the average and the CV of fruit weight for each treatment using the recorded mass of all harvested fruits of all experimental branches per treatment. Finally, we calculated broca infestation rate per treatment as the fraction of harvested fruits with broca marks. We acknowledge that broca might create a mark and be predated upon before causing seed damage; however, since we are focusing on differences among treatments, we assume this effect is consistent among them such that comparisons are valid. Also, we further inspected all fruits with broca marks and found that 81% showed evidence of effective broca colonization and seed damage.

To analyze the effect of bird and bee exclosure treatments on fruit set, we used generalized linear mixed models (glmer function in R package lme4). We included the number of initial flowers as an offset in this model to account for variation in flower and fruit counts and potential effects on fruit set, assuming a binomial distribution. We fitted the model with bird exclosure, bee exclosure, and their interaction as fixed effects, where a significant positive interaction is a signal of synergism (35). To meet the hierarchical sampling unit structure of the study design, we identified farm and bird exclosure sampling units nested within the farm as random effects.

To analyze the effect of bird and bee exclosure treatments and bored condition on the average fruit weight, we used linear mixed models (lme function in R package nlme). We identified farm and bird exclosure sampling units nested within the farm as random effects to meet the hierarchical study design. To assess the effects on fruit weight uniformity, we used the CV of fruit weight as a response variable following the same analytical approach.

To analyze the effect of bird and bee exclosure treatments on the proportion of bored fruits (i.e., fruits infested by broca from branches 1 to 8), we used generalized linear mixed models. Assuming a binomial distribution, we included the total number of fruits as an offset in this model to account for variation in fruit counts and potential effects on the proportion of bored fruits. Because we counted the number of bored fruits across time, we fitted this model with bird exclosure, bee exclosure, time, and all possible interactions as fixed effects. To account for nonindependence in the study design, we identified farm, bird and bee exclosure sampling units as nested random effects.

We calculated expected yield and farmer income based on fruit set, proportion of bored fruits, and fruit weight parameters from our exclosure experiment. This allowed us to calculate the economic benefits of current pest control and pollination services across coffee farms. To estimate average productivity per farm and exclosure treatment, we first estimated the average number of fruits on the four unmanipulated plants by counting all fruiting nodes in each plant. Then, we randomly selected three branches per plant and estimated the average number of fruits per node by counting the total number of nodes and the total number of fruits. Next, we estimated the average number of fruits per plant by multiplying the total number of nodes by the estimated number of fruits per node, and averaged using the information from all four plants. We then estimated average per plant productivity (in kilograms) per treatment considering the results from our exclosure experiment. To do so, we first calculated the number of harvestable fruits per plant in each farm by multiplying the average number of fruits per plant by the mean proportional fruit set from our statistical model for each treatment. Then, to calculate the number of bored harvestable fruits, we multiplied the total harvestable fruits per plant by the mean proportion of bored fruits from our statistical model for each treatment. The number of healthy fruits resulted from the difference between total harvestable fruits per plant and the number of bored harvestable fruits. To calculate kilograms of harvestable fruits per plant per treatment, considering differences in fruit weight between healthy and bored fruits, we individually multiplied the number of healthy and bored harvestable fruits per plant by their mean weight from our statistical model for each treatment, and summed them.

We then scaled up to farm-level estimates of coffee productivity (in kilograms per hectare), using plant density per hectare data estimated from a 10- × 10-m plot in each farm. We calculated yield in kilograms per hectare per farm for each exclosure treatment by multiplying the number of estimated plants per hectare by the estimated average per plant productivity (in kilograms) per exclosure treatment. Finally, we multiplied the average price that coffee mills paid per kilogram of coffee (US$0.33) in the Turrialba region during the year 2019 to obtain expected income per hectare for each treatment assuming all fruits were marketable, even if bored. This differs from methods used in other studies (25, 27) but reflects the reality of the Turrialba region where all fruits are weighed and paid for regardless of infestation (see SI Appendix for step-by-step calculations and formulas).