Quantification of phytoplankton groups using in-situ multi-excitation chlorophyll fluorescence measurements and machine learning (mf-ML)

Abstract

This study introduces a novel framework for the high-resolution quantification of phytoplankton communities using machine learning, integrating in situ multi-excitation chlorophyll fluorescence data with traditional high-performance liquid chromatography (HPLC) methods. The framework overcomes limitations of conventional sampling techniques by providing continuous, high-resolution profiling of phytoplankton distributions, capturing diel vertical migration (DVM) patterns, and analyzing responses to environmental factors such as irradiance and nutrient concentrations. The results from an XGBoost (Extreme Gradient Boosting)-based model demonstrated strong predictive accuracy across eight phytoplankton groups, effectively identifying complex nonlinear interactions that traditional methods struggle to resolve. Specifically, the model successfully traced DVM in dinoflagellates, offering insights into harmful algal bloom (HAB) dynamics. This study highlights multi-wave length excited fluorescence spectrometry as a cost-effective, accurate, and robust tool for monitoring phytoplankton distributions in situ, offering a significant advancement over remote sensing and discrete sampling techniques. By providing continuous monitoring of phytoplankton behavior and community structure, this approach can enhance the management of marine ecosystems, particularly in the context of HABs. Future work could expand this framework's applicability to other marine regions and phytoplankton communities, with the potential for real-time monitoring systems.