RIVER FLOW FORECASTING .3. APPLICATIONS OF LINEAR TECHNIQUES IN MODELING RAINFALL-RUNOFF TRANSFORMATIONS

The Simple Linear Model (SLM) and the Linear Perturbation Model (LPM) are applied to 14 catchments chosen from various geographical locations in the world. These models are applied in non-parametric form as well as under the constraints of the gamma function impulse response. Parameter optimization is carried out by the method of ordinary least squares and by Rosenbrock's search technique for the non-parametric and parametric versions, respectively. The loss of generality associated with the constraint of the gamma function shape is found to be insignificant. For large catchments which exhibit seasonal behaviour, good results can be obtained with the Linear Perturbation Model which involves the assumption of a linear relationship between departures from seasonal expectations in the input and output series. For use in real-time forecasting, an autoregressive updating component obtained from the examination of the residuals is used.