Combined Generalized Additive model and Random Forest to evaluate the influence of environmental factors on phytoplankton biomass in a large eutrophic lake

Abstract

The over-proliferation of phytoplankton has been a public concern for the last several decades. To evaluate the importance of different environmental factors on phytoplankton biomass variations with their complex response relationships in the large eutrophic Lake Okeechobee, the nonlinear methods including Generalized Additive Model (GAM) and Random Forest (RF) algorithm were employed. A long-term monitoring dataset of 8 sampling sites ranging from January 1996 to December 2010 was applied to explore the driven factors and response relationships of phytoplankton biomass from the different scales in Lake Okeechobee. Results showed a spatially heterogeneous distribution of phytoplankton biomass there, and the western and southern littoral zones occurred heavier algal blooms than the central pelagic zones. Spearman’s correlation results denoted Chlorophyll a (Chla) was negatively correlated with phosphorus in different scales. In the aspect of temporal variations, phytoplankton biomass appeared a three-peak-two-valley variation trend there, which peaked in 1997, 2003, and 2010, respectively. On the lake-wide scale, the RF model indicated that the inorganic nutrients were the primary predictors of phytoplankton biomass, while underwater light availability factors were followed to play essential roles in the prediction. On the local scale, total nitrogen (TN) was the top predictor in the western near-shore zone, while other zones were nitrate and nitrite (NO23) or phosphate (PO4). GAM results suggested that phytoplankton biomass had positive response to TN increase in both lake-wide and local scales. However, its response to water temperature (WT) appeared spatial heterogeneity. This study provided a new perspective to evaluate the primary predictors of phytoplankton biomass with their response relationships from lake-wide scale and local scale in the large shallow eutrophic lake, and the importance of taking spatial heterogeneity into account for lake water quality management was stressed.