Short-Term Early Warning Method for Algal Bloom Risk Based on a Neural Network and a Two-Dimensional Hydrodynamic Model

Abstract

Eutrophication of lakes and reservoirs and the occurrence of cyanobacteria blooms are two of the major environmental problems facing the whole world. However, if cyanobacteria bloom outbreaks can be predicted in advance, there is enough time to implement various measures to reduce ecological harm and health risks and greatly reduce economic losses. Yuqiao Reservoir was a typical shallow-water lake reservoir that was also faced with a greater risk of cyanobacteria bloom outbreaks in summer and autumn. In this study, a BP neural network model and a two-dimensional hydrodynamic numerical model were constructed based on meteorological, hydrological, water quality, and terrain data for Yuqiao Reservoir. The neural network model was used to simulate the biogeneration and extinction processes of cyanobacteria bloom biomass, while the two-dimensional hydrodynamic numerical model was used to simulate the migration and accumulation processes of cyanobacteria. The two models were coupled by an algal attenuation coefficient for the shortterm warning of algal bloom. The results of the Nash coefficient evaluation showed that the coupled model had good overall efficiency and could effectively simulate the algal density both in time and space. The results of scenario analysis showed that the change in flow field in a shallow water reservoir had a great influence on algae migration and accumulation, and water transfer, wind speed, and direction all affected the flow field distribution and then the formation of bloom. The model application results showed that the short-term bloom prediction method constructed in this study had advantages in forecasting accuracy and displayed an effect.