Probabilistic Estimation of Stream Turbidity and Application under Climate Change Scenarios

Streamflow-based rating curves are widely used to estimate turbidity or suspended sediment concentrations in streams. However, such estimates are often inaccurate at the event scale due to inter- and intra-event variability in sediment-streamflow relationships. In this study, we use a quantile regression approach to derive a probabilistic distribution of turbidity predictions for Esopus Creek, a major stream in one of the watersheds that supply drinking water to New York City, using measured daily mean streamflow-turbidity data pairs for 2003 to 2016. Although a single regression curve can underpredict or overpredict the actual observation, quantile regression can estimate a range of possible turbidity values for a given value of streamflow. Regression relationships for various quantiles were applied to streamflows simulated by a watershed model to predict stream turbidity under: (i) the observed historical climate, and (ii) a future climate derived from 20 global climate model (GCM) scenarios. Future scenarios using quantile regression in combination with these GCMs and a stochastic weather generator indicated an increase in the frequency and magnitude of hydrological events that may generate high stream turbidity and cause potential water quality challenges to the water supply. The methods outlined in this study can be used for probabilistic estimation of stream turbidity for operational decisions and can be part of a vulnerability-based method to explore climate impacts on water resources.