Prediction on daily spatial distribution of chlorophyll-a in coastal seas using a synthetic method of remote sensing, machine learning and numerical modeling

Harmful algal blooms (HABs) pose a severe environmental issue and have significant economic and ecological consequences on coastal oceans. Predicting the occurrence of these blooms has become increasingly vital for coastal communities. To facilitate this, chlorophyll-a (Chl-a) levels have been widely used to forecast algal blooms. Although Hydro-biogeochemical (HBGC) process-based models display reasonable accuracy in predicting hydrodynamic variables and nutrients, they are not as effective in predicting Chl-a. Purely data-driven machine learning techniques also have limitations in accurately predicting Chl-a of high spatio-temporal resolutions. In this study, a coupled HBGC-Convolutional Neural Network (CNN) model was developed to predict the daily surface Chl-a distribution. The HBGC-CNN model integrates the information gathered by the HBGC model on temperature, salinity, dissolved inorganic nitrogen, dissolved organic phosphorus, and zooplankton with the remote sensing Chl-a products for the CNN model training. The results revealed that the HBGC-CNN model can effectively reproduce both daily and seasonal Chl-a variations, and interpret spatiotemporal information related to an HAB event triggered by the heavy rainfall during typhoon Lekima in 2019. Furthermore, this method can be used for data reconstruction, producing gap-free Chl-a products for historical reanalysis, especially in nearshore regions. The successful implementation of the HBGC-CNN model in predicting Chl-a highlights its potential in being incorporated into an operational forecasting system from a regional scale to a global scale, reducing the adverse impact of HAB disasters and facilitating emergency treatment.