Huang, Haiyan (2016-08). Improving Regression Analysis to Predict Aquatic Life Potential in the Streams of Austin, Texas. Master's Thesis.
The aim for this study was to test if the City of Austin (COA) would gain more accurate estimations on Aquatic Life Potential (AQP) by regression equations with common watersheds characteristics, rather than using one general equation from Glick et al. (2010) for all twenty-four monitoring sites. Therefore, four objectives including grouping monitoring sites according to shared characteristics, creating regression models to predict AQP in each group, selecting the best regression model and comparing the best equation with the regression from Glick et al. (2010) were established. The monitoring sites were divided into five groups based on five hydrological characteristics. The multivariate regression analysis and simple linear regression analysis were performed in each group to form five sets of regression equations. The R^2 (Coefficient of Determination), standard error, the distribution of the monitoring sites, the goodness-of fit statistics of NSE (Nash-Sutcliffe efficiency index) and RMSE (Root Mean Square Error) were used to compare for the best equation. The R^2, NSE and RMSE were also used to compare the best regression with the regression equation from Glick et al. (2010). The results indicated that the best predicting equations were from the Impervious Cover group with R^2 larger than or equal to 0.68 and RMSE less than or equal to 11.99. That is much better than the regression equation from Glick et al. (2010) which has a slightly higher R^2 of 0.70, but much larger RMSE of 15.84. Therefore, the hypothesis that grouping monitoring sites based on common characteristics would result in better predictive models for AQL than one equation for all watersheds in the City of Austin was approved.