Forecasting Prorocentrum minimum blooms in the Chesapeake Bay using empirical habitat models

Frontiers in Marine Science

Marine Science

Christopher W. Brown¹  Raleigh R. Hood²  Dante M. L. Horemans³  Marjorie A. M. Friedrichs³  Pierre St-Laurent³ 

[1] Center for Satellite Applications and Research, National Oceanic and Atmospheric Administration, College Park, MD, United States;Horn Point Laboratory, University of Maryland Center for Environmental Science, Cambridge, MD, United States;Virginia Institute of Marine Science, William & Mary, Gloucester Point, VA, United States;

关键词: harmful algal bloom;    Prorocentrum minimum;    forecasting;    Chesapeake Bay;    logistic regression;    generalized linear models;    generalized additive models;   

DOI  :  10.3389/fmars.2023.1127649

 received in 2022-12-19, accepted in 2023-02-27,  发布年份 2023

来源: Frontiers

【 摘 要 】

Aquaculturists, local beach managers, and other stakeholders require forecasts of harmful biotic events, so they can assess and respond to health threats when harmful algal blooms (HABs) are present. Based on this need, we are developing empirical habitat suitability models for a variety of Chesapeake Bay HABs to forecast their occurrence based on a set of physical-biogeochemical environmental conditions, and start with the dinoflagellate Prorocentrum minimum (also known as P. cordatum).To identify an optimal set of environmental variables to forecast P. minimum blooms, we first assumed a linear relationship between the environmental variables and the inverse of the logistic function used to forecast the likelihood of bloom presence, and repeated the method using more than 16,000 combinations of variables. By comparing goodness-of-fit, we found water temperature, salinity, pH, solar irradiance, and total organic nitrogen represented the most suitable set of variables. The resulting algorithm forecasted P. minimum blooms with an overall accuracy of 78%, though with a significant variability ~ 30-90% depending on region and season. To understand this variability and improve model performance, we incorporated nonlinear effects into the model by implementing a generalized additive model. Even without considering interactions between the five variables used to train the model, this yielded an increase in overall model accuracy (~ 81%) due to the model’s ability to refine the regions in which P. minimum blooms occurred. Including nonlinear interactions increased the overall model accuracy even further (~ 85%) by accounting for seasonality in the interaction between solar irradiance and water temperature. Our findings suggest that the influence of predictors of these blooms change in time and space, and that model complexity impacts the model performance and our interpretation of the driving factors causing P. minimum blooms. Apart from their forecasting potential, our results may be particularly useful when constructing explicit relationships between environmental conditions and P. minimum presence in mechanistic models.

【 授权许可】

Unknown   
Copyright © 2023 Horemans, Friedrichs, St-Laurent, Hood and Brown