As we move towards shipboard-underway and automated systems for monitoring water quality and assessing ecological status, there is a need to evaluate how effective the existing monitoring systems are, and how we could improve them. Considering the existing limitations for processing numerous and complex data series generated from automated systems, and because of processes involved in phytoplankton blooms, this paper proposes a data-driven evaluation of an unsupervised classifier to optimize the way we track phytoplankton, including harmful algal blooms (HABs), and to identify the main associated hydrological conditions. We used in situ data from a portable flow-through automatic measuring system coupled with a multi-fixed-wavelength fluorometer implemented in the eastern English Channel during a bloom of Phaeocystis globosa (high biomass, non-toxic HAB species). This combination of technologies allowed high resolution online hydrographical and biological measurements, including spectral fluorescence as a means of quantifying phytoplankton biomass and simplifying the phytoplankton community structure inference. An unsupervised spectral clustering method was applied to this multi-parameter high-resolution time series, which allowed discrimination under near real-time of 6 to 33 contrasting water masses based on their abiotic and biotic characteristics. In addition, areas subject to extreme events such as HABs could be precisely identified, so controlling factors or their direct and indirect effects could be hierarchized. Considering the benefits and limitations of such a strategy, future applications of such methods will be important in the context of implementing the Marine Strategy Framework Directive.