Cyanobacterial biomass prediction in a shallow lake using the time series SARIMAX models

To address the challenge of forecasting harmful algal blooms caused by excessive growth of cyanobacteria in a shallow lake, we investigated cyanobacterial biomass prediction models based on a Seasonal Autoregressive Integrated Moving Average model with Exogenous Regressors (SARIMAX). Cyanobacteria cell density (CCD) was chosen as the indicator of cyanobacterial biomass. Ten years monthly observations from 2009 to 2018 collecting from a typical eutrophication shallow lake (Meiliang Bay, Lake Taihu) with frequent occurrence of algae blooms were used to construct and validate the proposed models. The mathematical experiments show that the variability and fluctuations of cyanobacterial biomass can be well-predicted by the models with appropriate exogenous variables. The selection of exogenous variables is critical for the model performance. Feature importance analysis of exogenous variables reveals that main variables influencing the prediction of cyanobacterial blooms are chlorophyll a, chemical oxygen demand, total phosphorus and temperature. These variables should be considered in actual prediction of cyanobacterial blooms. This study provides an efficient tool for forecasting algal bloom with ecological interpretability and can improve our ability to deal with bloom-induced environmental issues in shallow lakes.