South China Sea Annual Meeting 2023 (SCSAM)

109 / 2023-07-31 09:42:45

Harmful algal bloom operational forecasting system in Fujian Province, China: Development and skill assessment

Operational forecasting system; Harmful algal bloom; Development; Skill assessment; Fujian

The increasing occurrence of harmful algae blooms (HABs) in coastal oceans poses significant threats to aquatic ecosystems, economic growth, and human health. Fujian Province in China has been particularly affected by these frequent HAB events. To address this issue, a comprehensive monitoring network consisting of over 20 automatic buoy systems was deployed in Fujian's coastal waters since 2002. These buoys continuously collect crucial meteorological parameters, such as air temperature, pressure, wind direction, and wind speed, alongside essential water quality parameters, including water temperature, salinity, pH, dissolved oxygen, chlorophyll a, and turbidity.



Utilizing the extensive dataset obtained from these observations, we have developed a short-term HAB forecasting model based on Backpropagation (BP) and Radial Basis Function (RBF) neural networks. Since 2019, this model has been operational in the Fujian coastal area, with the highest HAB identification rate reaching 60% with a 24-hour lead time.



To enhance its predictive capabilities, the model incorporates key techniques and characteristics. Firstly, rigorous quality control of buoy data is achieved through various statistical methods, while the use of Self-Organizing Feature Map networks (SOM) ensures accurate selection of HAB samples for model training. Genetic algorithms are applied to filter the model's input parameters and optimize the initial weights and thresholds, resulting in a significant improvement in the model's operational efficiency. Additionally, the composite forecasting approach of BP and RBF neural networks maximizes the high-frequency sampling potential of ecological buoys, generating multiple forecast results. These results are statistically analyzed to derive HAB occurrence probability levels, effectively mitigating potential errors associated with relying solely on a single method. Lastly, the model's adaptive nature is ensured by dynamically updating the training database with the latest 15-day data and historical HAB sample data before each forecast, ensuring continuous accuracy.



In conclusion, our comprehensive approach to HAB forecasting in Fujian coastal waters has yielded promising results, empowering us to effectively combat the harmful impacts of these blooms on the region's environment and socio-economic sectors.