Predictive accuracy determination applied to a linear model phosphorus loading resulting from urban runoff

A case study of the application of an iterative technique for determination of predictive accuracy is presented. The method uses Monte Carlo simulations and split sampling techniques to verify model accuracy. Examination of the ability of a linear parametric runoff loading model's ability to project total phosphorus loadings reveals the sensitivity of the model to calibration procedures. Predictive reliability was found to vary widely as the number of rainfall events considered in the calibration process changed. Predictive reliability was substantially increased by imposing calibration constraints which ensured that a wide distribution of values of the independent variable were presented in the calibration pool.Although the linear model is theoretically weak in its representation of the runoff loading phenomenon, it displays relatively stable predictive capabilities which warrant its consideration for use in management studies. The predictive errors associated with loading projections limit the linear model's value to applications insensitive to errors in the loading projections for individual storm events. This study provides further evidence of the need to consider uncertainties associated with the modelling of water quality phenomena.