Open Collections

Real-time flow forecasting

University of British Columbia

The main objective of this research is to develop techniques for updating deterministic river flow forecasts using feedback of real-time (on-line) flow and snowpack data. To meet this objective, previous updating methods have been reviewed and evaluated and typical error patterns in flow forecasts have been analyzed using standard techniques. In addition, a new criterion based on the coefficient of determination and coefficient of efficiency has been introduced to evaluate systematic errors in flow forecasts. Moreover, lagged linear regression has been suggested as a method for detecting and estimating timing errors. Arising from this initial work, two different updating procedures have been developed. Further work has shown that these two independent procedures can be usefully combined, leading to yet further improvement of forecast. Arising from these methods, two other additional approaches have been formulated, one for correcting timing errors and one for updating snowpack estimation parameters from flow measurements. The first of the updating methods consists of a flow updating model which was developed to update the flow forecasts of the UBC watershed model using the most recent flow measurement. The updating process is achieved using the Kalman filter technique. The performance of the updating model is mainly controlled by the relative values of two parameters of the Kalman filter technique: the measurement variance and the state variance. It is found that the measurement variance is best selected as the square of a percentage of the flow. The updating model has been applied on the Illecillewaet river basin in British Columbia. A significant improvement in flow forecasts has been observed. The second method has been developed to update parameters of an energy budget snowpack model using on-line snowpack measurements. The updating procedure is based on calculating the value of a snowpack parameter that yields a perfect correspondence between measured and calculated snowpacks. The updated value is then used in the snowpack model to enhance its future forecasts with feedback from previous snowpack measurements. The snowmelts generated by the updated snowpack model are then routed to produce flow forecasts. Applying this model on the snowpack measured at Mt. Fidelity in the upper Columbia River Basin in British Columbia showed that both the snowpack forecasts and the flow forecasts generated from these updated snowpack forecasts were greatly improved. Because the above two updating methods operate independently, they can be applied in combination whenever an appropriate measurement is available. The combined use of these two methods to data from the Illecillewaet river basin showed an additional improvement in flow forecasts. As a further development, the snowpack estimation model has been adapted so that a Kalman filter approach can be used to update snowpack estimation parameters from flow measurements. It is finally concluded that flow forecast updating requires the application of several methods, rather than one simple approach, because errors arise from various sources. In addition, updating procedures may prove useful in achieving a better calibration for watershed models.

Text

2011-01-25

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/30815

Civil Engineering

University of British Columbia

Graduate