Ocean Color Modeling in the Central Red Sea Using Oceanographical Observation and Simulated Parameters

The summer phytoplankton bloom events have been recently investigated using remote sensing observations over several geographical areas of the Red Sea and changed our impression of its oligotrophic characteristic. However, only limited blooms events were recorded due to active dust storms limiting the observations. This work focuses on predicting the potential bloom events in the central region of the Red Sea, indicated by the chlorophyll-a values, through the machine learning models built from the simulated and observed oceanographical parameters. Four subregions showing active eddy activities are selected to generate modeling datasets in the Case-1 waters (water depth > 300 meters) for each region. Automated model selection and tuning are performed among different candidate supervised models including linear regression, trees models, ensemble models and deep neural networks (101 in total). The ensemble models (random decision forest and bootstrap decision forest) outperform others in showing effective performance in estimating chlorophyll-a values with (\mathrm{R}^{2} > 0.82) of the training and (\mathrm{R}^{2} > 0.73) of the testing processes, respectively. This work shows the potential applications to use a machine learning model to reconstruct missing ocean color observations, as well as revealing the oceanographical mechanism to induce phytoplankton growth in the Red Sea.