Recent climate change has significantly influenced agricultural practices worldwide (Wheeler and Braun, 2013;Gruda et al., 2019;Rising and Devineni, 2020). In particular, crops sensitive to changes in average temperature, such as temperate species, are becoming increasingly difficult to cultivate in their traditional growing regions (Moore and Lobell, 2015;Bento et al., 2021). Consequently, the area devoted to greenhouse cultivation has steadily increased as a strategy to maintain crop productivity and stability (Vatistas et al., 2022;Dohlman et al., 2024). In Korea, the total greenhouse cultivation area expanded from 90,468 ha in 2015 to 96,642 ha in 2025, representing an increase of approximately 6.82% (Statistics Korea, 2025). Similarly, in the United States, greenhouse cultivation has grown consistently by approximately 4-5% every five years since 2009 (USDA-NASS, 2009, 2014, 2019). Although greenhouse farming requires higher initial construction and maintenance costs compared to open-field cultivation, its gradual expansion is expected to continue in the future.

Successful greenhouse crop production requires not only optimal environmental control but also effective pollination management (Gudowska et al., 2024). In strawberries, selfpollination alone often results in low fruit set and reduced market quality; thus, managed pollinators such as honey bees (Apis spp.) or bumble bees (Bombus terrestris) are essential (Dimou et al., 2008;Wietzke et al., 2018;Cao et al., 2024). Consequently, numerous studies have investigated the role of pollinators in greenhouse strawberries and sought to identify the most efficient species (Zaitoun et al., 2006;Lee et al., 2008;Abrol et al., 2019). In addition, in the winter of 2020, South Korea experienced a nationwide loss of approximately 17.6% of overwintering Apis mellifera colonies (Jung and Bae, 2022;Kang et al., 2024), leading to shortages in pollinator supply and increased costs. Although B. terrestris has been widely adopted as an alternative, its use can increase fruit weight but is also associated with a higher incidence of fruit deformities, necessitating careful management (Zaitoun et al., 2006;Cao et al., 2024).

As a potential solution, the Asian honey bee (A. cerana) has been proposed (Osterman et al., 2021; Kim et al., 2024). This species reportedly exhibits lower mortality rates than A. mellifera and comparable pollination performance (Gudowska et al., 2024), although quantitative evidence remains limited. However, no prior study has directly compared colony strength dynamics and pollination efficiency of A. cerana and A. mellifera in greenhouse strawberry production.

Addressing this gap, the present study aimed to evaluate the potential of A. cerana as an alternative pollinator to A. mellifera by comparing colony strength dynamics, daily foraging activity, and pollination efficiency under controlled greenhouse conditions. To this end, we monitored changes in colony strength over the same period following pollinator introduction, compared daily activity patterns and flower visitation rates between the two species, and quantified the proportion of normally developed achenes per fruit, which is closely linked to marketable fruit quality.

Materials and Methods

Study site and insect

The study was conducted from November 2024 to March 2025 in greenhouse strawberry cultivation complexes located in Nonsan, Chungcheongnam, South Korea. Three farms (N1, N2, and N3) within the region were selected as study sites (Table 1), and a total of 13 greenhouses were used in the experiment. Each greenhouse was stocked with a single honey bee colony, either A. cerana or A. mellifera. In total, 10 greenhouses were stocked with A. mellifera and three with A. cerana. In all greenhouses, the entrances were oriented toward the southwest, and strawberry beds were arranged longitudinally, perpendicular to the entrance direction. The greenhouses were covered with double-layer plastic film, with polyolefin used as the outer layer and polyethylene as the inner layer. The strawberry cultivar grown in all experimental greenhouses was Fragaria × ananassa.

All colonies were installed on 28 November 2024, prior to the start of monitoring. The colonies used in the experiment were obtained from the experimental apiaries of the National Institute of Agricultural Sciences, Wanju, Jeollabuk-do, South Korea (35°49'49'' N, 127°02'14'' E; 35°49'32'' N, 127°02'41'' E). All colonies contained a healthy, naturally mated queen and no queen replacement occurred during the study period; however, information on queen weight was not available. The A. mellifera colonies were housed in Langstroth hives containing a total of three frames (honeycombs), comprising two brood frames and one food frame. Colony strength was adjusted so that brood coverage occupied approximately 60% of the surface area of each frame. The A. cerana colonies were maintained in hives of the same design, with a total of three frames, consisting of two brood frames and one food frame, and their colony strength was adjusted to approximately 40% brood coverage per frame. These target brood coverages were chosen to reflect typical winter management practices in Korea, where A. mellifera colonies are usually maintained at higher brood area than A. cerana because A. cerana naturally forms smaller colonies with lower brood area and egg-laying capacity (Zhang et al., 2019). Consequently, the 60% versus 40% broodcoverage settings were designed to represent realistic strong production colonies for each species under commercial greenhouse conditions rather than to equalize absolute adult bee numbers.

Colony strength monitoring

The strength of each honey bee colony was measured throughout the experimental period. Measurements commenced on 28 November 2024 and were conducted once per month, with the final measurement taken on 28 February 2025. Colony strength was estimated using a modified version of the method described by Delaplane et al. (2013). For each colony, the proportion of the honeycomb surface covered by adult bees was visually assessed for both sides of each frame, and the average percentage per frame was calculated. A frame completely covered by bees on one side was considered 100% coverage for that side. The same estimation procedure and the same conversion factor (2,200 bees per fully covered side) were applied to both A. mellifera and A. cerana colonies. Although A. cerana workers are smaller in body size, no species-specific conversion factors are currently available, and using an identical standard allows for direct comparison of relative changes in colony strength under shared environmental conditions. The resulting percentage values for each of the three frames were multiplied by 2,200 to estimate the number of bees, and colony strength was expressed as the sum of these estimates.

Honey bee foraging activity

Honey bee foraging activity was quantified using hive entrance counts. At each hive and observation hour, the numbers of bees entering and exiting the hive were recorded using three consecutive 3-min counts. For each count, the numbers of incoming and outgoing bees were recorded separately. The three replicate counts for entering bees and for exiting bees were averaged, and the sum of these two averages was used as the hourly foraging activity index for that hive. When more than one record occurred within the same hour, these values were further averaged to obtain a single activity value per hive and hour for statistical analyses. Individual flower visits by honey bees typically last only a few seconds, allowing multiple foraging events to be captured within short observation windows (Frazier et al., 2015; Wierzchowski et al., 2024). Entering/exiting activity was recorded simultaneously by three trained observers (one per farm). All observers underwent joint training before the study to standardize observation procedures and minimize inter-observer and handling variation. On the same six monitoring days, floral visitation was recorded within each greenhouse. At three time points per day (09:00, 12:00, and 16:00), the numbers of honey bees visiting strawberry flowers were counted within a 1 m² area located within 30 m of each hive. At each time point, three consecutive 3-min counts were conducted, yielding three replicate visitation records per greenhouse and time point. In the subsequent analyses, each 3-min count was treated as an individual observation, and greenhouse identity was included as a random intercept in the linear mixed-effects models to account for non-independence among replicates from the same greenhouse. For analyses of pollination efficiency, foraging activity was defined as the combined number of hive entering/exiting events and strawberry flower visits recorded during the same observation period. This combined index integrates both outbound/inbound for aging trips and direct floral visitation.

Assessment of achene set

Pollination efficiency was assessed at the fruit level by calculating the proportion of normally developed achenes per fruit. A normally developed achene was defined as one with a longitudinal length of approximately 0.6 mm, while smaller achenes were classified as abnormal (Ariza et al., 2011;Cekic et al., 2018). In each greenhouse, ten flower buds were randomly selected and individually tagged, and each bud subsequently produced one mature fruit. Achenes were counted for each harvested fruit, and the numbers of normal and abnormal achenes were recorded. Thus, the pollination-efficiency dataset consisted of ten fruits per greenhouse, collected from seven A. mellifera greenhouses and three A. cerana greenhouses (100 fruits in total), providing 100 fruit-level observations for analysis.

Data analysis

All statistical analyses were conducted using R version 4.5.1 (R Core Team, 2025). To evaluate species-specific changes in colony strength over time, linear mixed-effects models (LMMs) were fitted with species, days after installation, and their interaction as fixed effects, and colony identity as a random intercept. Because colony strength was measured at only four time points per colony, we initially evaluated an autoregressive correlation structure (AR(1)); however, the estimated autocorrelation parameter was unstable and did not improve model fit based on AIC comparisons. Therefore, the simpler randomintercept model without AR(1) was selected. Colony strength values were right-skewed and showed increasing variance with the mean; thus, colony strength was log-transformed to stabilize residual variance. Model assumptions of normality, homoscedasticity, and independence were checked using Q-Q plots, residual-versus-fitted plots, and inspection of random-effect distributions. No substantial violations of assumptions were detected, and the transformed models showed improved residual structure.

For foraging activity (entering/exiting counts), an LMM was fitted with species, hour of day, and their interaction as fixed effects, and hive identity as a random intercept to account for among-greenhouse variation. Standardized foraging activity was obtained by dividing raw entering/exiting counts by the estimated total number of adult bees in each colony. The estimated number of adult bees per colony was calculated from the frame coverage assessments described above using the conversion factor of 2,200 bees per fully covered side and then averaged across all colony-strength measurements during the study period. The same LMM structure was applied to both the raw and standardized datasets. Flower visitation counts were analyzed using separate LMMs for raw and standardized values, with greenhouse identity included as a random intercept to account for non-independence among observations from the same greenhouse

Pollination efficiency, expressed as the proportion of normal achenes per fruit, was analyzed using both ordinary least squares (OLS) regression and beta regression with a logit link. In both models, species was included as the main effect, and foraging activity and colony strength were included as covariates. The foraging activity covariate used in these models corresponded to the combined activity index described above (entering/exiting counts + flower visits per observation period). Because these regression models were fitted at the fruit level and did not include greenhouse identity as a random effect, fruits sampled within the same greenhouse may not be fully independent, and the fixed-effect estimates should be interpreted as greenhouse-level trends summarized from withingreenhouse variation. Estimated marginal means (EMMs) and 95% confidence intervals were calculated for species-level comparisons.

The colony-strength model included 40 measurements (10 colonies measured at four time points), the foraging-activity model used 600 hourly observations, the flower-visitation model comprised 180 observations, and the pollinationefficiency model was based on 100 fruit-level measurements. Due to incomplete records, colony-strength analyses were restricted to 10 colonies that were consistently monitored across all four time points.

Results

Colony strength dynamics

Colony strength declined over time after hive installation in A. mellifera and A. cerana (Fig. 1). Colony strength was measured four times per colony between late November and late February. Linear mixed-effects models (LMM) with random intercepts for colony showed that the decline rate was significant for A. mellifera (p < 0.001; Table 2), corresponding to an estimated 50% reduction of the initial strength at approximately 52 days post-installation. In contrast, the decline rate for A. cerana was not statistically significant (p = 0.1460), and the estimated time required for colony strength to reach 50% of its initial level was approximately 79 days. Because only four repeated measurements were available for each colony, temporal autocorrelation could not be reliably estimated; therefore, we report the results from the simpler random-intercept LMM without an explicit AR(1) correlation structure.

Comparative foraging activity

Foraging activity showed distinct diurnal patterns between A. cerana and A. mellifera (Fig. 2). A. cerana exhibited low activity in the early morning (08:00–09:00) and peaked at 11:00–12:00, whereas A. mellifera displayed a sharp increase from 10:00, with the highest activity observed at 12:00– 13:00. Linear mixed-effects models revealed significant species × hour interactions at 12:00 (p = 0.035) and 13:00 (p = 0.0063), with A. mellifera showing higher entering/exiting counts than A. cerana. No significant differences were detected at other time points, although both species exhibited elevated activity at 11:00 and 14:00. When standardized by colony strength, the magnitude of species differences in entering/ exiting activity was substantially reduced. The species × hour interaction was not statistically significant (p = 0 .435), indicating that the diurnal patterns of relative activity were broadly similar between A. cerana and A. mellifera. This suggests that the higher raw entering/exiting counts of A. mellifera can be largely explained by their larger colony sizes rather than intrinsic differences in foraging intensity per bee.

Raw flower visit counts tended to be higher for A. mellifera than for A. cerana, but the difference was not statistically significant (β = 0.810 ± 0.824, p = 0.346; Fig. 3a). When standardized by colony strength, the species difference further diminished and remained non-significant (β = -0.0002 ± 0.0003, p = 0.456; Fig. 3b), indicating that part of the difference in raw counts was attributable to the larger colony sizes of A. mellifera. The number of flowers visited per hive entering/exiting trip was, on average, higher for A. mellifera (0.589 ± 3.968) than for A. cerana (0.0289 ± 0.0736), although the high variability in A. mellifera led to overlapping error ranges (Fig. 3c).

Pollination efficiency by normal achene set

Across all sampled fruits (n = 100), the total number of achenes per fruit averaged 190.3 ± 74.0 (range: 0–465), indicating substantial variation in potential seed set per fruit. Pollination efficiency, expressed as the percentage of normally developed achenes, was compared between species (A. cerana and A. mellifera) using both linear regression and beta regression models, with foraging activity and colony strength included as covariates (Fig. 4). In both models, the species effect was not statistically significant (OLS: β(AM-AC) = -3.93, 95% CI -23.25 to 15.38, p = 0.651; beta regression: β = -0.248, 95% CI -0.946 to 0.449, p = 0.486). Among the covariates, foraging activity was positively associated with the percentage of normal achenes (p < 0.01), whereas colony strength showed a negative association (p < 0.05). Estimated marginal means (EMMs) indicated no significant difference between species (A. cerana: 75.0%, 95% CI 57.8-92.2; A. mellifera: 71.1%, 95% CI 64.1-78.0; p = 0.651). In an additional analysis based on efficiency per visit (percentage of normal achenes per number of visits), A. mellifera was estimated to be lower (p = 0.071).

Discussion

In this study, we compared colony strength dynamics, daily foraging activity patterns, and pollination efficiency between the western honey bee (A. mellifera) and the Asian honey bee (A. cerana) in greenhouse-grown strawberries. The comparatively faster decline of A. mellifera colonies may reflect their higher energetic demands and lower tolerance to cold or variable winter greenhouse conditions, which intensify metabolic stress relative to A. cerana. Differences in foraging activity patterns and absolute visitation numbers between the two species were largely explained by colony size, while pollination efficiency per flower visit did not differ significantly.

In greenhouse strawberry cultivation, temperatures inside the greenhouse can fluctuate widely, ranging from 5°C to 30°C (Ahn et al., 2021;Khammayom et al., 2022). In Korea, strawberries are typically cultivated from November to March, coinciding with the wintering period of both honey bee species (Döke et al., 2015;Lee et al., 2022). These environmental conditions may therefore impose physiological stress and may contribute to colony decline for both species. As observed in this study, colony strength gradually decreased after hive installation, with a steeper rate of decline in A. mellifera. Previous studies indicate that A. cerana demonstrates greater thermal adaptability, exhibiting flight activity under lower temperatures, lower light intensities, and reduced solar radiation (Katuwal et al., 2023). Earlier reports suggested that A. cerana wax may melt at slightly higher temperatures than A. mellifera (Ruttner, 2013), although more recent analyses of Korean colonies found no significant species difference (Dekebo & Jung, 2023). The functional implications of these minor differences remain unclear, but these intrinsic traits collectively suggest that A. cerana may be comparatively more tolerant under variable or suboptimal winter greenhouse conditions. Moreover, A. cerana workers are smaller and consume less sugar per day than A. mellifera (Zhang et al., 2019), which may offer additional energetic advantages under resourcelimited or low-temperature environments.

The similarity in overall foraging patterns between species suggests that temperature, rather than species identity, is the dominant factor shaping diurnal activity rhythms in winter greenhouse environments. The slightly earlier activity of A. cerana is consistent with its ability to forage under lower thermal thresholds. This pattern is consistent with previous findings that A. cerana can remain active under relatively lower temperatures and light levels compared with A. mellifera (Zhang et al., 2019;Katuwal et al., 2023). In addition, differences in the absolute number of flower visits between the two species, based on colony strength and foraging activity, were largely explained by colony size. This suggests that strawberry flower visitation rates are more strongly influenced by colony strength than by interspecific differences. In other words, since flower visitation rates are determined more by the level of activity than by species-specific traits, it is presumed that the visitation rate of each bee species could be regulated by adjusting foraging conditions, such as temperature, in accordance with species-specific preferences.

The comparable pollination outcomes between species indicate that, when foraging intensity is similar, both honey bees make equivalent contact with floral reproductive structures, leading to similar achene development. Although strawberries can be partially self-pollinated, pollination by insect pollinators leads to higher-quality fruit with greater market value (Wietzke et al., 2018). This can be explained by understanding the pollination mechanism: when pollination is sufficiently effective, achene formation becomes complete, which in turn promotes the synthesis of auxin, a plant growth hormone, resulting in larger and more uniformly shaped fruits (Guo et al., 2022). Moreover, uniform achene development is associated with a higher rate of well-formed fruit. Ultimately, since no significant difference in pollination outcomes was observed when both bee species visited flowers at similar rates, it is necessary to focus on enhancing overall foraging activity through appropriate management strategies. Although per-visit efficiency tended to be lower in A. mellifera, this difference was not statistically significant, suggesting that further studies with larger sample sizes are needed to confirm this trend.

In interpreting the pollination-efficiency results, it is important to note that fruit-level measurements were obtained from ten fruits within each greenhouse. Fruits collected from the same greenhouse are likely to share microclimatic and management conditions, and the regression models did not include greenhouse identity as a random effect. As a result, the fruit-level observations are not strictly independent. Consequently, the negative association between colony strength and the proportion of normal achenes should be interpreted with caution, as it may partially reflect within-greenhouse covariance or temporal mismatches between colony assessments and fruit sampling rather than a direct causal effect of larger colonies reducing pollination efficiency. Future studies adopting hierarchical sampling designs and mixed-effects models that explicitly account for greenhouse-level structure would help to clarify these relationships.

Although the number of greenhouses differed among farms (three to four houses per farm), hive identity was included as a random effect in all mixed-effects models, which accounts for potential farm-level clustering. In addition, greenhouse structures (orientation, internal layout, and bed arrangement) were highly consistent across farms, reducing the likelihood that structural differences biased species comparisons. Nevertheless, variation in greenhouse number may still contribute to unexplained site-level heterogeneity, and future studies with balanced replication across farms are warranted.

These findings, derived from winter greenhouse strawberry production in South Korea, provide an empirical foundation for selecting and managing honey bee species under controlled environments. Future work should incorporate more targeted experimental approaches to elucidate the mechanisms that drive species-specific performance. Controlled manipulation of initial colony strength would help clarify whether differences in visitation rate and colony decline arise from colony size or from intrinsic physiological traits of each species. Experimental adjustments to greenhouse microclimate, including stepwise changes in temperature, regulated humidity, or the timing of ventilation, would further identify the thermal and environmental thresholds that optimize foraging behavior. Spatially explicit monitoring of foraging movements using technologies such as RFID tagging or automated video analysis could provide insights into within-greenhouse movement patterns and flower visitation efficiency. In addition, larger-scale production trials that directly compare fruit yield and marketable quality across species, colony sizes, and environmental settings would offer more practical guidance for commercial greenhouse pollination. Together, these research directions will enhance our understanding of honey bee performance in protected agriculture and support evidence-based pollination management.