An estimator of winter loss of honey bees in Sweden

Abstract: The aim of this thesis is to suggest a statistical model to estimate the loss rates of honey bees during winter in Sweden. The estimator is to be based on annual summary statistics, collected by the Swedish beekeeping organisation. Regional specific estimates for winter loss rate is derived by a spatial and temporal hierarchical model with binomial response. The model is updated by Bayesian inference using Integrated Nested Laplace Approximations (INLA). Winter loss estimates were derived with probability intervals and presented on a map of Sweden. The analysis shows that the average winter loss rates ranges between 10.2 and 19.7 % across the 21 regions, while the differences in average loss rates between years were increasing from ranges 9.3 to 18.3 % in 2015, 9.9% to 19.4 % in 2016 and 11.0 to 21.3 % in 2017. Regional differences were not linked to cultivation zones, when comparing different models with the information criterions WAIC and DIC. The analysis included summary statistics from three years. It is possible to expand the model to include spatial and temporal interaction and trends over time by including summary statistics from more years. Estimates of winter loss rates based on data from several years have stronger properties compared to properties due to changes between regions, especially since the contributing beekeepers may vary a lot from year to year and in each region.