Date of Award

2013

Document Type

Thesis

First Advisor

Jha, Dr. Manoj Kumar

Abstract

A suite of regression models were tested for the construction of rating curves and constituent load estimation for 17 water quality parameters monitored at 16 stations regularly since 1999 by the City of Greensboro in North Carolina. Best models were selected based on the statistical evaluation within the framework of the LOAD ESTimator (LOADEST) model. The constituent prediction varied from the “true load” by –6% to 16% for Nitrate; -14% to +12% for Nitrite; -6% to 0% for Total Dissolved Solids (TDS); -2% to 9% for Total Kjeldahl Nitrogen (TKN); -22% to 9% for Total Phosphorus (TP); and -51% to 23% for Total Suspended Solids (TSS). There was a systematic bias towards under-prediction for TDS, TP and TSS whereas nitrate and TKN were over predicted and none for Nitrite. The predicted loads were compared with five interpolation methods (M1, M2, M3, M4 and M5) in the following pattern: for nitrate, TDS and TSS, load estimated by M3, M4 and M5 > LOADEST > M1 and M2; for nitrite, TKN and TP: LOADEST > M3, M4 and M5 > M1 and M2. Multivariate analyses used cluster analysis (CA), factor analysis (FA) and principal component analysis (PCA) on all parameters at all stations. CA grouped the water quality station into four spatially similar clusters. PCA/FA was applied on the entire dataset of entire watershed and spatially similar stations. Combination of FA/PCA and CA reduced the size of the dataset by 71% and represented the 64% of the total variance.

Share

COinS