Modeling Long-term Dissolved Organic Carbon Patterns Using Environmental Variables     

Abstract

Organic matter (OM) in rivers is an important food source to sustain aquatic ecosystem health. However, in surface water supply systems where chlorination is often used for disinfection, OM is also a precursor for the carcinogenic and mutagenic disinfection byproducts (DBPs) such as trihalomethanes (THMs) and haloacetic acids (HAAs). Effective management of OM in rivers to maintain both aquatic ecosystem functions and high water-supply quality requires better understanding of the OM transport patterns, where dissolved organic carbon (DOC) can be used as a surrogate measurement of OM. Analysis of DOC data on a watershed scale to estimate fluxes and to determine long-term trends remains challenging, largely due to the spatial and temporal variations in DOC, and low sampling frequency. To help improve the understanding of DOC sources and export processes, we compared long-term temporal patterns in six watersheds in the New York City (NYC) Water Supply System, which supplies drinking water daily to over 8.5 million people in NYC and one million people in the upstate counties. Firstly, we compared six empirical water quality models for DOC prediction. The models include flow-based linear regression (LM), dynamic linear models (DLMs), LOAD ESTimator model (LOADEST), Weighted Regressions on Time, Discharge, and Season (WRTDS), multiple linear regression (MLR), and general additive models (GAMs). Given the differences in dominant land-use and hydrological conditions in the study watersheds, we found that GAMs produced the most robust results. Secondly, we used GAMs with multiple predictor variables to predict long-term daily DOC concentrations in the six study watersheds, which allowed better trend analysis and flux estimates than using the routine grab-sample data with inconsistent sampling frequencies. Lastly, we compared the relationships between temporal patterns in DOC and watershed features to investigate the regional differences, focusing on the watershed mechanistic processes associated with DOC by parsing out the climate signals from the historical trends. The results show that hydrology plays a larger role on DOC temporal patterns in some watersheds whereas nutrient associated production processes are more important in others. The study presents a better performing approach than the solely hydrology driven models and can inform targeted monitoring strategies for DOC management in water-supply source waters.