Harmful algal blooms (HABs) are a serious threat to ecosystems and human health. The accurate prediction of HABs is crucial for their proactive preparation and management. While mechanism-based numerical modeling, such as the Environmental Fluid Dynamics Code (EFDC), has been widely used in the past, the recent development of machine learning technology with data-based processing capabilities has opened up new possibilities for HABs prediction. In this study, we developed and evaluated two types of machine learning-based models for HABs prediction: Gradient Boosting models (XGBoost, LightGBM, CatBoost) and attention-based CNN-LSTM models. We used Bayesian optimization techniques for hyperparameter tuning, and applied bagging and stacking ensemble techniques to obtain the final prediction results. The final prediction result was derived by applying the optimal hyperparameter and bagging and stacking ensemble techniques, and the applicability of prediction to HABs was evaluated. When predicting HABs with an ensemble technique, it is judged that the overall prediction performance can be improved by complementing the advantages of each model and averaging errors such as overfitting of individual models. Our study highlights the potential of machine learning-based models for HABs prediction and emphasizes the need to incorporate the latest technology into this important field.

Cyanobacterial Harmful Algal Blooms (CyanoHABs) produce toxins and odors in public water bodies and drinking water. Current process-based models predict algal blooms by modeling chlorophyll-a concentrations. However, chlorophyll-a concentrations represent all algae and hence, a method for predicting the proportion of harmful cyanobacteria is required. We proposed a technique to predict harmful cyanobacteria concentrations based on the source codes of the Environmental Fluid Dynamics Code from the National Institute of Environmental Research. A graphical user interface was developed to generate information about general water quality and algae which was subsequently used in the model to predict harmful cyanobacteria concentrations. Predictive modeling was performed for the Hapcheon-Changnyeong Weir–Changnyeong-Haman Weir section of the Nakdong River, South Korea, from May to October 2019, the season in which CyanoHABs predominantly occur. To evaluate the success rate of the proposed model, a detailed five-step classification of harmful cyanobacteria levels was proposed. The modeling results demonstrated high prediction accuracy (62%) for harmful cyanobacteria. For the management of CyanoHABs, rather than chlorophyll-a, harmful cyanobacteria should be used as the index, to allow for a direct inference of their cell densities (cells/mL). The proposed method may help improve the existing Harmful Algae Alert System in South Korea.

Process-based modeling for predicting harmful cyanobacteria is affected by a variety of factors, including the initial conditions, boundary conditions (tributary inflows and atmosphere), and mechanisms related to cyanobacteria growth and death. While the initial conditions do not significantly affect long-term predictions, the initial cyanobacterial distribution in water is particularly important for short-term predictions. Point-based observation data have typically been used for cyanobacteria prediction of initial conditions. These initial conditions are determined through the linear interpolation of point-based observation data and may differ from the actual cyanobacteria distribution. This study presents an optimal method of applying hyperspectral images to establish the Environmental Fluid Dynamics Code-National Institute of Environment Research (EFDC-NIER) model initial conditions. Utilizing hyperspectral images to determine the EFDC-NIER model initial conditions involves four steps that are performed sequentially and automated in MATLAB. The EFDC-NIER model is established using three grid resolution cases for the Changnyeong-Haman weir section of the Nakdong River Basin, where Microcystis dominates during the summer (July to September). The effects of grid resolution on (1) water quality modeling and (2) initial conditions determined using cumulative distribution functions are evaluated. Additionally, the differences in Microcystis values are compared when applying initial conditions using hyperspectral images and point-based evaluation data. Hyperspectral images allow detailed initial conditions to be applied in the EFDC-NIER model based on the plane-unit cyanobacterial information observed in grids, which can reduce uncertainties in water quality (cyanobacteria) modeling.