Machine learning-assisted identification of Hoverfly plant-pollinator interactions using crowdsourced iNaturalist image observations
Faculty Mentor Name
Ezra Kottler
Research or Creativity Area
Natural Sciences
Abstract
Plant–pollinator interactions are important for maintaining biodiversity and supporting
ecosystem stability. Although hoverflies (Family Syrphidae) are widespread and effective pollinators, there is limited research identifying which plant species they pollinate. iNaturalist is a community science application/platform where users upload images which are recorded as species observations. In this study, we are investigating which plants are being pollinated by which insect taxa in the hoverfly family. Images of hoverflies on plants were collected and processed from the iNaturalist database. AI models were trained to recognize and remove hoverflies from iNaturalist images, enabling identification of plant species using the plant-identification iNaturalist algorithm. Plant genera were confirmed by comparing iNaturalist findings with images and descriptions from the CalFlora database of the plants of California. We will present early findings via a map of plant pollinator interactions characterizing what hoverfly genera in California were associated with and potentially pollinating which plant taxa. In subsequent work, we will implement an additional machine learning model to identify a larger sample of plant species from citizen-science-generated iNaturalist observations. Additionally, using climate data from the time & GPS location of the observations, we will use statistical models to test whether environmental variables are predictive of particular pollinator relationships.
Machine learning-assisted identification of Hoverfly plant-pollinator interactions using crowdsourced iNaturalist image observations
Plant–pollinator interactions are important for maintaining biodiversity and supporting
ecosystem stability. Although hoverflies (Family Syrphidae) are widespread and effective pollinators, there is limited research identifying which plant species they pollinate. iNaturalist is a community science application/platform where users upload images which are recorded as species observations. In this study, we are investigating which plants are being pollinated by which insect taxa in the hoverfly family. Images of hoverflies on plants were collected and processed from the iNaturalist database. AI models were trained to recognize and remove hoverflies from iNaturalist images, enabling identification of plant species using the plant-identification iNaturalist algorithm. Plant genera were confirmed by comparing iNaturalist findings with images and descriptions from the CalFlora database of the plants of California. We will present early findings via a map of plant pollinator interactions characterizing what hoverfly genera in California were associated with and potentially pollinating which plant taxa. In subsequent work, we will implement an additional machine learning model to identify a larger sample of plant species from citizen-science-generated iNaturalist observations. Additionally, using climate data from the time & GPS location of the observations, we will use statistical models to test whether environmental variables are predictive of particular pollinator relationships.
