Observing and understanding harmful algal blooms using autonomous underwater imaging

Harmful Algal Bloom (HAB) events pose threats to local aquaculture, fisheries and public health. In many cases the mechanisms triggering HABs remain poorly understood, in part because of insufficient environmental and ecological monitoring. Here, we utilize an in-situ underwater microscope to develop a high temporal resolution, multi-year record of the abundance of harmful protists at Scripps Pier, a coastal location in the Southern California Bight. We convened a panel of taxonomy experts to identify and enumerate the population dynamics of 9 major HAB species recorded at Scripps Pier and their major grazers, including heterotrophic protists and copepods, to construct a training set. Those images are used to build a convolutional neural network that classified roughly 12 million images of particles within 30 µm to 1000 µm size range, taken from 2015 to 2019. We compare the temporal dynamics of these HAB taxa to key environmental drivers, including irradiance, sea temperature, macronutrients, salinity and weather conditions, using a Random Forest approach to select the variables most closely tied to population dynamics for each taxa. Here, we discuss the utility of autonomous underwater imaging for monitoring HABs and species-specific associations between bloom events and environmental and biotic conditions at Scripps Pier.