Deploying the Imaging FlowCytobot in the Mediterranean Sea and developing a Trait-Based Classifier for High-Frequency Screening of Phytoplankton Traits and Dynamics

Due to its potential to monitor phytoplankton taxa at a rate that even the best experts can't equal, the Imaging FlowCytobot (IFCB) has rapidly gaining popularity and has revolutionized the way researchers are studying phytoplankton communities. However, the huge amount of images it generates makes manual identification impractical. Therefore, there is a need to develop machine learning algorithms to automate the analysis of IFCB images. The present study explored the possibility to build a trait-based classifier rather than using the traditional taxonomic approach. IFCB images were collected in the Mediterranean Sea from January 2023 to June 2024 and used to develop a random forest classifier to classify phytoplankton images based on their morphological traits. After training, the classifier was able to classify 28 classes of phytoplankton with an overall F1-score of 0.90. While still subject to improvement, the present classifier did a relatively good job and demonstrated that the trait-based approach can be used as an alternative to study phytoplankton communities with an IFCB.