ALGAL BLOOMS FORECASTING WITH MACHINE LEARNING IN TOLO HARBOUR
Algal blooms pose a significant threat to coastal safety and the functioning of marine ecosystems. Since 1975, Tolo Harbour has experienced the highest frequency of red tides in Hong Kong, accounting for approximately one-third of all algal blooms in the region. This prevalence is primarily due to pollution and the enclosed nature of the bay. The complex environmental dynamics and limitations in data availability have consistently made forecasting algal blooms a challenging endeavor. Traditional models often rely on simplified assumptions and fixed parameters, which fail to capture the dynamic nature of ecosystems, leading to less accurate predictions. The rapid advancement of deep learning techniques offers a promising avenue for improving algal bloom forecasts. However, the typical weekly or biweekly intervals of monitoring data collection significantly constrain the capabilities of these models. To address this issue, we introduced the concept of data assimilation to account for model errors and incorporate new observations into a data-driven deep learning algal bloom prediction model. In this study,
we utilized remote sensing data to construct a foundational model for predicting chlorophyll-a, an indicator of algal blooms. We then trained a sub-model using biweekly monitoring water quality data. Finally, we applied an Ensemble Kalman Filter approach to assimilate the prediction results from both the main model and the sub-model, thereby enhancing the overall accuracy of our forecasts.
Additional information: Mr. LIU Zhangbo, zbliu@connect.hku.hk