Stochastic transfer function time-series modeling of suspended sediment concentrations as a function of streamflow

The conventional sediment rating curve approach results in simple linear regression models of log-transformed suspended sediment values, expressed as a function of log-transformed stream discharge. Such data, however, are frequently stochastic in nature and a time-series approach may be more appropriate. Box/Jenkins time-series analysis is applied to two sets of 1982 data for Duck and Arrow creeks near Creston, British Columbia, to illustrate this alternative approach. First order transfer function time-series models are generated that represent significant improvements over the simple regression models. The Box/Jenkins procedure is also applied to a 1981 data set for Arrow Creek to produce a first order model with a coefficient of determination of 0.722 comared to 0.418 for the simple regression model.