PS 16-228 - Assessing Insect-Plant Interactions Using Autonomous Camera Trap Video and Deep Learning Methods in San Diego, California

Monday, August 7, 2023

5:00 PM – 6:30 PM PDT

Location: ESA Exhibit Hall

Presenting Author(s)

JK

Jessica Keatly

Student Researcher
California State University San Marcos
San Marcos, California, United States

Coauthor(s)

JO

Jasmine O'hara

Student Researcher
California State University San Marcos, United States
CS

Christina Simokat

Faculty Mentor
California State University San Marcos, United States
EF

Elizabeth Ferguson

Faculty Mentor
California State University San Marcos, United States

Abstract: Understanding pollination requirements for native plant species is critical for wildlife conservation and management purposes. There are various methods of identifying insects and collecting data on their behaviors. One such method involves the use of pan traps placed near a plant that kill insects, allowing identification by an entomologist. Another common method is focal monitoring, where a human researcher observes and records insect species occurrence. However, these approaches either harm insects, or introduce observer bias and restrict monitoring to daytime hours. The goal of this study was to assess the effectiveness of autonomous camera trap video for monitoring insect-plant interactions while developing deep learning methods of detection for future datasets. The development of deep neural networks for large dataset detection reduces the need for human-mediated annotation while also retaining reliable insect group or species identification. We constructed two camera systems as described by the PICT system in July 2022, which were used in a pilot study to collect video footage of insect pollinators of Encinitas Baccharis (Baccharis vanessae), an endemic plant to the San Diego region. To compare the efficacy of the system to traditional methods, we conducted focal monitoring with the Pollinator Monitoring Program at California State University San Marcos, which collects data in preserves and gardens. Video was annotated by human researchers in VIAME, a deep learning object tracking software that has freely accessible, cloud-based access. Annotations of approximately 120 hours of footage revealed a total of 229 insect interactions, which could be broken into approximately five groups of insect pollinators (bees, ants, flies, moths, and insects unable to be identified). Precise identifications of insect pollinators were unable to be achieved due to technical problems we aim to rectify for the 2023 bloom period. In comparison, our focal data, collected in nine 15 minute intervals recorded 244 total interactions over the course of 3 days of monitoring. Autonomous camera trap video shows promise in matching the efficacy of other monitoring methods, and was successful in establishing robust data on pollinator interactions and previously unknown nighttime pollination events. We also share methods for documenting the occurrence of insect pollinators and training the software’s neural network, allowing for automated tracking and identification of insects in future datasets. This work is important as the survival of native habitats are dependent on pollinators, and new methods of monitoring such as this one can better understand local pollinator networks to inform conservation methods.