Date of Award

5-2014

Level of Access Assigned by Author

Open-Access Thesis

Degree Name

Master of Science (MS)

Department

Ecology and Environmental Sciences

Advisor

Cynthia S. Loftin

Second Committee Member

Francis A. Drummond

Third Committee Member

Brian McGill

Abstract

Non-native honeybees historically have been used to pollinate many crops throughout the United States, however, recent population declines have revealed the need for a more sustainable pollination plan. Native bees are a natural resource that can play an important role in pollination. I used spatial modeling tools to evaluate relationships between landscape factors and native bee abundance, with a focus on the wild native bees that pollinate Maine’s lowbush blueberries. I applied the InVEST Crop Pollination ecosystem spatial modeling tool, which predicts pollinator abundance based on available floral resources and nesting habitat, to the Downeast Maine region. The InVEST model is a generic tool that can be adapted to any landscape with development of location specific parameters and a validation dataset. I surveyed botanists, entomologists and ecologists who are experts in native bee ecology and familiar with Maine’s landscape, and asked them to rank the suitability of landcover types as native bee habitat. I used previously collected bee abundance data to validate model assumptions. I evaluated the sensitivity and explanatory power of the InVEST model with four model parameterization methods: 1) suitability values assigned through the expert survey; 2) suitability values developed through a sensitivity analysis; 3) informed suitability values developed through an optimization based on the sensitivity analysis; and, 4) uninformed suitability values developed through machine-learning simulated annealing optimization. I evaluated the improvement in prediction gained from expert-informed and optimization-informed parameterization compared with prediction based on the relationship between proportion of landcover surrounding blueberry fields and native bee abundance as an alternative to the InVEST model. The InVEST model parameterized through expert opinion predicted native been abundance (r = 0.315; P = 0.047), whereas, the uninformed optimization improved model performance by 28% (r = 0.404; P = 0.010), and the informed optimization technique improved model performance by 58% (r = 0.486; P = 0.002). The landcover analysis found a significant relationship between the proportion of deciduous/mixed forest within a 2000 meter buffer around a field and native bee abundance within the field (r = 0.446; P = 0.004). Although the InVEST model reliably predicts bee abundance across a landscape, simpler models quantifying relationships between bee abundance and proportional land cover around focal fields may be suitable alternatives to the InVEST simulation model.

Download

Share