Water quality and physicochemical conditions drive chlorophyll-a concentrations in two connected subtropical off-stream reservoirs

The assessment of reservoir water quality is vital for preserving ecosystems and ensuring the sustainable use of water resources. Chlorophyll-a (Chl-a) acts as a vital bioindicator, reflecting the dynamics of phytoplankton populations and trophic status of aquatic ecosystems. In this study, we use Generalized Additive Mixed Model (GAMM) models to analyze variations in Chl-a concentration in two connected subtropical off-stream reservoirs (Ren-Yi and Lan-Tan). Water temperature and rainfall are the only two important variables appearing in the optimal GAMM models for both reservoirs. However, Ren-Yi's optimal model additionally includes NH₃, total phosphorus (TP), and water level, suggesting these factors may play a larger role in its nutrient levels and fluctuations. This is supported by the significantly higher chemical oxygen demand (COD), TP, and total nitrogen (TN) levels in Ba-Zhang River, which recharges Ren-Yi Reservoir. Lan-Tan's optimal GAMM model incorporates 'sampling depth' variables due to significant differences between shallow and deep sites. Interestingly, in larger datasets (300 points), 'season' emerges as a crucial variable, highlighting intensified seasonal variations in denser data. Therefore, incorporating 'season' as a nominal variable is essential for accurate modeling. Variance structures of Chl-a vary within Lan-Tan by season and within Ren-Yi by sampling site (RY100) and water temperature (RY300). The optimal GAMM captures this inherent variability by incorporating sampling sites or seasons as random effects. The relationship between dissolved oxygen (DO), COD, and Chl-a concentrations is complex and influenced by multiple factors, including nutrient dynamics, algal activity, water circulation patterns, and local conditions. GAMMs are well-suited to capturing the potentially nonlinear and time-varying nature of these relationships, leading to more accurate modeling.