Calibrated stochastic simulation of river pollution control strategies

Abstract

To account for both the major physical phenomena governing pollution transport in rivers and the uncertainty of parameters that influence that transport, a physically-based stochastic water quality model, supported by empirical stochastic models, is developed. Empirical stochastic models, linking a contemporaneous autoregressive moving average model (CARMA) with a modified general Monte Carlo simulation (GMCS) model, are derived from historical data to represent the statistical characteristics of selected river system parameters and state variables, including spacetime correlation structure. These models are used to generate possible realizations of parameter values used as input to a modified version of the physically-based stochastic mass-transport model, QUAL2EU. The QUAL2EU model is applied using Monte-Carlo simulation to generate distributions of possible water quality concentrations at locations and times of interest along a river. The model is calibrated by using an optimization algorithm to adjust the statistics of selected stochastic parameters to minimize the difference between the statistics of water quality state variables predicted by the QUAL2EU model and the corresponding statistics generated by an empirical stochastic model based on historical measurements of the state variables at designated locations along the river. Model validation (testing) is conducted over a four-year period by comparing the statistics of measured water-quality constituents at several locations along the river generated by a stochastic empirical model, as well as the values as directly measured, to corresponding statistics of model predictions over the period. The models are applied to a 190-km segment of the South Platte River in Colorado to simulate the diminished quality of river flows available for irrigation diversion due to return flows from economic activities. Distributions and statistics of flow rates, total dissolved solids (TDS), dissolved nitrogen (N) species, and fecal coliform concentrations are predicted along the river. To estimate the performance of alternative engineering interventions, eleven different pollution control strategies are considered, alone and in combination. Statistics of water quality constituent concentrations, and the associated probabilities of violation of recommended criteria, along with the spatial and temporal uniformity of these probabilities, are estimated. Over a 25-year horizon, for example, combined pollution control measures are predicted to achieve average reductions from 27 to 36% and from 31 to 39% in mean concentrations of TDS and N, respectively, for the 11 irrigation diversions located along the 53-km reach just downstream of Denver. Uncertainty in predicted concentrations, as indicated by the coefficients of variation, is found to be considerable. An average decrease of 18 percentage points in the probability of violation for TDS is predicted, while the estimated average decrease for N and fecal coliform are only 6 to 8 percentage points. Major conclusions are drawn and recommendations for further study are presented.