Comparison of a deterministic and a statistical model for predicting streamflow recession curves

Prediction of the master baseflow recession curve at an ungauged site is an important and difficult problem in water resource management. Two approaches, one deterministic and one statistical, are developed and their relative performance as predictors is compared to find which is better to apply at sites where streamflow is supplied from bed and bank storage. Both models are used to calibrate a given inverse square recession formula. The basis of the calibration is estimation of the time for the median outflow rate to halve. The deterministic model expresses this drainage time as net storm rainfall inflow, less quick flow, divided by the average exfiltration rate from channel bed and bank storage. The statistical model uses six basin characteristics to calculate the relative degree of similarity between an ungauged basin and each one of a set of reference basins. Similarity is expressed as a weight and these are combined with measured drainage times to form a pooled estimate for the ungauged basin. Data from ten gauged basins in the greywacke geologic terrain of Canterbury, New Zealand are used to calibrate and test the models. These basins have no significant lakes or springs and master curves were constructed using recessions occurring in the January to March period. Reliable predictions for practical purposes can be made in the specified terrain using the statistical model in its present form. In comparison, improved definitions and measurement of some key variables, and definition of an exfiltration function, are necessary before the deterministic model may be usefully employed. While further development and testing of the models is needed, particularly in other geologic terrains, results indicate that both types have the potential to supply reasonably accurate predictions of master recession curves in basins where little is known about water storage behaviour.