To understand the mechanisms driving the population dynamics of pollinators, the research team will develop technologies for deeply embedded hardware/software learning systems capable of remote, long term, autonomous operation; and will analyze the resulting new data to better understand pollinator activity.
Plant-pollinator interactions are key promoters of terrestrial biodiversity and crucial for our society’s food security. Insect pollination directly affects the yield and quality of 75% of globally important crop types and annually contributes $235–$577 billion in food production. Several stressors including habitat deterioration, diseases, species invasions, climate change, and pesticides are producing a rapid worldwide pollinator decline; U.S. beehives have decreased by 50% in the past 50 years and flying insects in German nature reserves have decreased by 75% in the past 25 years. The goal of this project is to discover the mechanisms responsible for the function and persistence of pollination systems. Furthermore, beekeepers, growers and land managers need information and technology to help them modify their practices to mitigate stressors and support pollinator health.
The project will make contributions in the areas of energy-efficient un- and semi-supervised machine learning algorithm design for resource-constrained distributed wireless sensing systems, automated on-line adaption of features to current environmental conditions without losing the use of prior machine learning training state, evaluation of a multi-field based approach to representing and carrying out non-linear classification on concurrent audio signals and of techniques for compressing the field representations to enable faster and more energy efficient execution/classification, combining the machine learning process with the automated generation of lossy and lossless compression algorithms tailored to the data, study of the space- and time-dependent diversity and behavior of pollinators, and (6) field data gathering in pollination systems unachievable by traditional sampling methods (e.g. time series of pollinators’ visits to an individual plant, visitation rates of pollinators over time under different background of flowering-plant diversity).