Bayesian model averaging for harmful algal bloom prediction.

Area of Science:

• Marine biology
• Ecological modeling
• Environmental science

Background:

• Harmful algal blooms (HABs) pose a significant global threat to aquatic ecosystems and biodiversity.
• Existing predictive models for HABs often fail to adequately address parameter and model uncertainties, increasing management risks.
• Lyngbya majuscula blooms, a toxic cyanophyte, cause substantial ecological damage in affected areas.

Purpose of the Study:

• To develop and evaluate a statistical model for predicting the occurrence of Lyngbya majuscula blooms.
• To incorporate Bayesian model averaging (BMA) to account for uncertainties in ecological predictions.
• To assess the effectiveness of BMA as a management tool for HABs.

Main Methods:

• Utilized a probit time series model combined with Bayesian model averaging (BMA).
• Analyzed predictors for Lyngbya majuscula blooms in Deception Bay, Queensland, Australia.
• Compared different BMA strategies for predictive performance.

Main Results:

• Identified several significant predictors for HAB occurrence, with temperature being prominent.
• A model using average monthly minimum temperature as the sole covariate demonstrated the strongest posterior support.
• Both BMA approaches exhibited excellent predictive performance, with minimal differences between strategies.

Conclusions:

• Bayesian model averaging (BMA) is a powerful tool for predicting HAB occurrences, effectively managing parameter and model uncertainties.
• Temperature, specifically average monthly minimum temperature, is a critical factor in predicting Lyngbya majuscula blooms.
• BMA offers a robust and reliable approach for ecological management decisions, particularly for HABs.