Development and comparison of Imaging FlowCytobot classifiers in coastal California

The Imaging FlowCytobot (IFCB) is emerging as a valuable new tool in phytoplankton ecology and HAB monitoring. Initial development of classifiers is time-consuming and labor intensive, requiring trained taxonomists with knowledge of local phytoplankton. However, after development and testing of a classifier for image identification, an IFCB provides round-theclock, real-time observation of the phytoplankton community in its deployment location. In this study, IFCBs were deployed at Pier 17 in the San Francisco Bay, and at the Santa Cruz Wharf in Monterey Bay. We investigate how the performance of the classifiers developed in these specific locations compare to one developed in the Gulf of Mexico for identifying common California HAB-forming taxa, including Pseudo-nitzschia spp., Alexandrium spp., and Dinophysis spp. Additionally, data collected from side-by-side IFCB deployments at the Santa Cruz Wharf were used to compare the performance of the instruments to one another. This allowed us to determine how well images from two different instruments represent the phytoplankton community at a single site, and evaluate whether a classifier is inherently tied to the instrument used to build it. Additionally, image classification methodologies were compared and here we present the results of a convolutional neural network (CNN) as an alternative to well-described supervised learning methods. This work provides valuable insight into the development of a HAB-monitoring network of IFCBs along the California coast.