Using a naive Bayes classifier to explore the factors driving the harmful dinoflagellate Karenia selliformis blooms in a southeastern Mediterranean lagoon

Abstract

The blooms of the toxic dinoflagellate Karenia selliformis can be predicted with accuracy derived from knowledge of the main forcing variables. A naive Bayes classifier modeling framework, a member of the Bayesian network family, is used to identify the phytoplankton community and the physical and meteorological parameters involving K. selliformis blooms in the eutrophic Boughrara Lagoon (BL), Tunisia. The proposed model takes into account the physical environment parameters (salinity, water temperature, tide amplitude), meteorological constraints (evaporation, air temperature, insolation, rainfall, atmospheric pressure, and humidity), phytoplankton groups (diatoms, dinoflagellates, cyanobacteria, Euglenophyceae), and the sampling months on K. selliformis blooms. The shift to highest salinity and atmospheric pressure, associated with reduced tide, are the most favorable conditions for K. selliformis blooms in BL. The results show that K. selliformis formed nearly monospecific blooms. A shift in species composition was pointed out between the bloom and non-bloom conditions. The Euglenophycea and some dinoflagellates like Peridinium minimum, Prorocentrum minimum, Prorocentrum micans, Prorocentrum gracile, Protoperidinium minutum, and Scrippsiella trochoidea appeared during blooms, whereas diatoms, diazotrophic cyanobacteria, and dinoflagellates (Akaskiwo sanguinea, Karlodinium veneficum, Gyrodinium spirale, Oxyrrhis marina, Polykrikos kofoidii) were observed under non-bloom conditions. This study highlighted the most favorable physical and meteorological conditions for K. selliformis bloom occurrences and also pointed out species indicators for bloom establishment and others for non-bloom conditions. Monitoring the dynamics of these species with the associated physical and meteorological variability offers valuable information to plan for the best options for prediction of potentially toxic blooms of K. selliformis and associated dystrophic consequences.