Investigation of Abiotic Predictors for Phytoplankton Blooms in a Typical Reservoir Tributary Using Statistical Machine Learning Methods

Under favorable environmental conditions, excessive nutrients in waters can stimulate extremely rapid growth of microscopic algae or phytoplankton, forming algal blooms. Since the operation of reservoirs, serious phytoplankton blooms have been observed in the weakly flushed tributaries. The water impounding of the Three Gorges Reservoir (TGR) of china decreased the flow velocity and the clearing capability in the Yangtze River. As a result of the backwater blocking effect, this has caused many water quality problems in its tributaries. Understanding the causality and dynamics of phytoplankton blooms in the tributary water is therefore important for the holistic management of the important reservoir. We develop a management strategy to assess the relative importance of abiotic predictors of phytoplankton blooms by using the Pearson correlation analysis method, the Random Forests and Decision Tree machine learning algorithm. To reveal abiotic predictors of phytoplankton blooms, four-year data set comprising daily observations of chlorophyll-a concentrations, nutrient levels, physical factors (e.g. temperature, light, mixing, and pH) was analyzed in a bloom-impacted tributary of the TGR, Xiangxi Bay. The results show that mixing depth and the ratio of euphotic depth to mixing depth, but not nutrient levels, are important predictors to promote the onset of phytoplankton blooms in the seasonal thermal stratification reservoir tributaries. Although a strong relationship hasn’t found between vertical average temperature and chlorophyll-a concentrations, significant water temperature stratification cause density current that be an important determinant of mixing depth in Xiangxi Bay. The temperature still plays an indispensable role for phytoplankton blooms. Based on these findings, water level fluctuations in a reservoir to reduce thermal stratification in its tributaries can provide a potential mean for controlling phytoplankton blooms in the short term. Our results also suggested that nonlinear relationships and critical thresholds between chlorophyll-a concentrations and abiotic predictors can be used to quantitatively predict phytoplankton blooms in Xiangxi Bay. This study can help us to advance our understanding of the mechanisms that affect phytoplankton blooms and the factors that can be managed to mitigate blooms in the reservoir tributary systems.