14 BEE INTERACTION STUDY
14.2 Bee Preference Study Methods
During the early stages of the glasshouse experiment it was decided that bees would be observed and their plant preferences recorded. Several methods were devised for this including introducing bees to the plants at certain periods, and knowledge was sought from experts in this field (Chapter 16). However, once the plants had started flowering it was noticed that bees were entering the glasshouse and using the plants of their own accord. Due to technical and logistical difficulties with the originally proposed methods, it was decided that the bees already freely using the plants would give good insight into bee preference and should therefore be observed.
During the peak flowering times in 2012 and 2013, bee foraging preferences were observed during 1-hour periods (Goulson & Hanley, 2004). During each survey, individual bees were selected and their activity recorded during a 5- minute period. Where possible, each bee was identified to species level and the identification number of the plant used by the bee was recorded (‘bee visits’) and total time spent on flowers in a given plant (from landing to departure) was recorded (‘time spent’) in seconds (Herrera, 1989) using a stopwatch. The total number of plants with flowers visited over the survey period was also recorded (Herrera, 1989). Interference and disturbance of the bee activity by the recorder was avoided by limiting fast movements, avoiding creating shadow onto each bee and keeping at least 1m distance. If plant identification numbers could not be identified at this distance they were checked as soon as the bee vacated it. Notes were made if the bee was only resting on the plant rather than collecting pollen and nectar. Ambient
temperature and outside weather conditions were recorded at the start of each survey, and surveys were only undertaken between 10.00 and 17.30, during dry weather when ambient temperature was above 17oC (Heard et al., 2007).
Bees which went out of sight either during or after their allotted 5 minutes and then reappeared might be recorded more than once (Fussel and Corbet, 1992).
Survey details and conditions are tabulated in Appendix XVIII.
The Herbivore Requirements Results (Chapter 12.8) were used in conjunction with bee observations to determine whether any interactions existed between these response variables.
14.2.1 Data Analysis
All parameters were first tabulated with means, standard deviations and standard errors calculated. Bar charts were produced in Excel (Microsoft, 2013) (version 15.0.4551.1005) to see initial differences between ecotypes and treatments.
Data were then saved as CSV files for use in R statistical software version 3.1.0 (R Core Team, 2014). Histograms were created using Rcmdr (Fox & Bouchet-Valat, 2014) (version 2.0-4) to help determine the distribution of each data set, as data was found to be of non-parametric distribution the Kruskall- Wallis test was used to determine initial significant differences.
Scattergraph matrices and boxplots were created using Rstudio (RStudio, 2013) (version 0.98.994) to identify similarities and correlation between datasets.
14.2.2 Data Analysis: Generalized Linear Mixed Model (GLMM) Generalized linear mixed models (GLMM) were used to statistically predict ecotypic distributions with specific sub-sets of response variables, as this statistical technique is good for species-specific models (Guisan et al., 1999). GLMM tuition and written aids were used in the application and interpretation of the models in R (Field et al., 2009; Zuur et al., 2009; Winter, 2013; Smith, 2014) in R Studio (RStudio, 2013) (version 0.98.994). It was originally hoped that one model could be used (with the response variable changed for each dataset). However due to the high number of variables and interactions, the sample size prevented a full model analysis (Smith, 2014). Partial models were therefore processed and resulting effects plots along with the Akaike Information Criterion (AIC) and P values from ANOVA comparisons values used to determine the most important factors for the final model (Winter, 2013;
Smith, 2014). For the first model (Model A) it was hoped to include each ecotype, the treatment soil and management. The partial models outlined in Chapter 11.6 were tried with the following models found to be the most suitable for ‘bee visits’ and ‘time spent’ (the bold text is the response variable which was changed to that being investigated at the time of analysis). The standard used in these models was Cockey Down (a calcareous loam/grazed ecotype), in calcareous loam and grazed treatment.
<-lmer(response_variable~EcotypeSeed+soil+management+(1|f.replic), data= response_ variable)
This model uses replication as a random effect and ecotype, soil treatment and management treatment as fixed effects.
The interaction and ecotype site model was simpler (with no significant difference) without the interaction factor, therefore the simpler model was used:
<-lmer(response_variable~EcotypeSoil+soil+EcotypeMgmt+management+ (1/f.replic),data=reponse_variable)
As before, these models were repeated with ANOVA comparison of null (partial) models to establish P value significance of factors.
All of the above models were calculated using a Poisson distribution due to the nature of the data (count data and non-parametric data), ‘family = poisson’ was entered into the model equation before ‘data=’.
To identify whether any of the previously measured response variables would act as predictor factors for the bee variables two further models were
calculated, these were split between vegetation factors and flower factors. As all factors were numeric the predictor variable was defined as continuous. Vegetation factors model:
<-lmer(response_variable~+ nitrogen + HCN + hirsuteness + vegetation_dry_biomass + (1/f.replic),data=reponse_variable)
Flower factors model:
<-lmer(response_variable~+ flower_number + flower_dry_biomass + pre_harvest_flower_scent + relative_moisture_content +
(1/f.replic),data=reponse_variable)
14.2.3 Distance from Bee Home-site (Bath Spa University)
To aid in interpreting the results of the bee data, geographical distances were calculated between the glasshouse location and the bees foraging there (referred to as ‘test foraging area'), to each ecotype site (Table 34). These distances were studied in conjunction with results to identify whether there were any bee preferences for sites geographically closer to the glasshouse.
Table 34. Geographical distance (km) between the glasshouse (home-site of bees) and ecotype sites, in ascending order of distance
Bath Spa University Glasshouse Ecotype sites OS Grid
refs. ST693637 Southstoke ST737610 5.16 Folly Farm ST611606 8.77 Hellenge Hill ST345572 35.4 Berrow Dunes ST292532 41.45 Woodborough SU117614 42.46
Salisbury Plain SU192481 52.28
Cockey Down SU173320 57.52
Dawlish Warren SX983789 110.6
Woolacombe