Non-Linear Modeling of Dissolved Oxygen Variations in Cahaba River, Al

The Cahaba River watershed, located in northern Alabama, has experienced many environmental problems due to the rapid urbanization of the surrounding areas. The ecosystem of the river could be compromised by unnatural environmental occurrences, most notably large scale algal bloom. This project set out to determining the portions of the river experiencing these problems and their severity.

The method of determination was nonlinear regression modeling of continuous, time series data for six stations along the river. Specific conductance, turbidity, pH, and temperature were the input parameters for the model and dissolved oxygen concentration was the output. Data was chosen to ?train? the model based on the saturated dissolved oxygen levels; portions of the data at or below saturation were of greatest interest. Coefficients for the nonlinear model were generated using the nonlinear regression function, nlinfit, of the MATLAB programming software. The difference between the predicted value of the model and the actual value of dissolved oxygen was used to determine the magnitude of the error term of the model, essentially the variation in the dissolved oxygen levels not accounted for by the input parameters of the model at times of normal dissolved oxygen saturation levels. Data exhibiting error terms of magnitude four or greater, variation that is dangerous for most marine animals, indicates unnatural variations in the dissolved oxygen levels due to external influences such as algal blooms.

The analyzed data indicates unnatural variations in the dissolved oxygen levels for site #1, the furthest down stream. Sites #5 and #8 have signs of unnatural variation but the results are less conclusive than that for site #1. Further analysis using the technique of multivariate partial least squares could be done to look more closely at the contribution of the individual inputs to yield more conclusive results on the variations in dissolved oxygen levels.