Physical Factors Control Phytoplankton Production and Nitrogen Fixation in Eight Texas Reservoirs

Abstract

We compared regression tree analyses and multiple linear regression models to explore the relative importance of physical factors, land use, and water quality in predicting phytoplankton production and N2 fixation potentials at 85 locations along riverine to lacustrine gradients within eight southern reservoirs. The regression tree model (r2 = 0.73) revealed that differences in phytoplankton production were primarily a function of water depth. The highest rates of production (mg C m−3 h−1) occurred at shallow sites (<0.9 m), where rates were also related to total phosphorus (TP) levels. At deeper sites, production rates were higher at sites with relative drainage area (RDA, ratio of drainage area to water surface area) below 45, potentially due to longer hydraulic residence times. In contrast, multiple linear regression selected TP, RDA, dissolved phosphorus, and percent developed land as significant model variables (r2 = 0.63). The regression tree model (r2 = 0.67) revealed that N2 fixation potentials (mg N m−3 h−1) were substantially higher at sites with relatively smaller drainage areas (RDA < 45). Within this subgroup, fixation rates were additionally related to TP values (threshold = 41 μg l−1). The multiple linear regression model (r2 = 0.67) also selected RDA as the primary predictor of N2 fixation. Regression tree models suggest that nutrient controls (phosphorus) were subordinate to physical factors such as depth and RDA. We concluded that regression tree analysis was well suited to revealing nonlinear trends in data (for example, depth), but yielded large uncertainty estimates when applied to linear data (for example, phosphorus).