Sensing & Monitoring Pollinators: From Agent-Based Modeling to Live Sensing

Abstract

The introduction of digital technology has revolutionized agricultural systems by giving growers the ability to constantly monitor and manage their fields via comprehensive, large-scale data collection and decision support systems. While many avenues have been explored with different sensors, data management platforms, and machine learning, few works have taken a close look at incorporating pollinators into this new, growing infrastructure. With over 80\% of flowering plants requiring some form of animal pollination, the pollinator plays a key role in global agricultural food production. Pollinators in general adapt well to an evolving environment and adeptly navigate and forage by necessity. They have been shown to explore large, complex areas in search of food and resources. Of all pollinators, the honey bee is especially important with millions of colonies in the United States and contributing billions to the global economy. While the importance of honey bees has garnered it significant interest in the past, with many methods attempting to sense and track its behavior, and indirectly its effects on yield, prior works are invasive, large, or require line of sight. Other works have looked to bypass the requirement of instrumenting the pollinator itself by placing sensors such as video and acoustic devices in the environment to gather data. Although this method is limited by the number of devices deployed and area surveyed and is largely manual in data gathering and analysis, sensors placed in the environment have shown promise for gathering interesting information. In this thesis, I will propose solutions in line with both of these research directions. Adding to the tracking of managed pollinators, I will present a comprehensive foraging simulation platform that leverages an innovative flight recorder for monitoring managed pollinators. This simulation produces honey bee flight paths based on parameters found by an extensive literature search and live observation combined with modified robotic path planning algorithms. Paths are sampled by the sensor model and then processed to produce foraging activity and high-level obstacle maps. This simulation platform further provides a tool for modelling the real-life functionality of such a system and also explores sensor design tradeoffs. Adding to the tracking of pollinators by sensors in the environment, I will present two works: 1) a simulation of the feasibility of pollinator monitoring with typical agricultural robots. 2). a low-cost, portable, and user-friendly system for acoustically detecting pollinators in the field. The result of these works is a new direction for pollinator monitoring and digital agriculture in general that provides a new avenue for exploring system design in this space.