Improvement of harmful algal bloom models through advanced data assimilation and adaptive modeling

Abstract

Even though much progress has been made recently in modeling harmful algal blooms, uncertainties remain much larger than in ocean circulation models. Fundamental equations governing the dynamics of phytoplankton species and their interactions remain relatively unknown. As a consequence, many formulations in harmful algal bloom models are based on empirical relationships, which imply an excessive degree of parameterization and restrictive assumptions. Here, we present adaptive modeling as an extension of data assimilation to the selection of model state variables, linkages, parameterizations and mathematical formulations. The regular model state variables are combined with model formulations, forcing fields and parameters in an augmented state vector, which is updated through an ensemble Kalman filter. Based on misfits between model predictions and real data, adaptive modeling identifies model structures and formulations that need to be improved, estimates those improvements, and corrects the model accordingly. The model changes and learns from data providing more realistic predictions and selecting the most adequate model formulations that describe the system. We use this adaptive modeling approach to identify the factors (e.g. life cycle strategy, nutrient availability, species competition, or zooplankton grazing) that determine Alexandrium minutum blooms in Arenys de Mar harbor (northwestern Mediterranean Sea)