An ensemble modeling framework to elucidate the regulatory factors of chlorophyll-a concentrations in the Nanji wetland waters of Poyang Lake

Abstract

Chlorophyll-a (Chl a) is an important indicator of algal biomass frequently used to evaluate the severity of cultural eutrophication. Identifying the key covariates of Chl a concentrations is essential to understand the mechanisms that drive eutrophication and to develop forecasting tools that guide the restoration process. In this study, we present a novel ensemble modeling framework founded upon the complementary features of Random Forest (RF) and Generalized Additive modeling (GAMs). A series of RF models are first developed to forecast Chl a concentrations based on the antecedent values of a multitude of environmental predictors. GAMs are then used to explore the presence of non-linearities in the seasonal relationships between Chl a and the identified predictors. The optimal RF models using a 0–8 day time lag displayed high predictive skills with adjusted R2 values consistently above 0.80. Analyses of the RF models revealed that the modulating factors of Chl a display significant seasonality. Dissolved oxygen (DO) and turbidity were the key covariates of Chl a in the spring, while the water level fluctuations predominantly regulated phytoplankton biomass in the summer and winter. The occurrence and severity of algal blooms in the summer and autumn were associated with threshold levels of 0.06 and 1.50 mg/L for total phosphorus (TP) and total nitrogen (TN) concentrations, respectively. These results reveal the potential of the introduced modeling framework to shed light on the regulatory factors of algal biomass as well as to establish real-time predictions in the Nanji wetland waters of Poyang Lake.

Keywords

• Chlorophyll-a
• Random forest
• Generalized additive modeling
• Eutrophication
• Poyang Lake