Real time runoff forecasting with tank models whose parameter values are identified by Kalman filter

Ichihara, K.; Toyokawa, K.; Sawaguchi, I.; Kawashima, H.

Journal of the Japanese Forestry Society 82(2): 125-131


Accession: 003541707

Download citation:  

Article/Abstract emailed within 1 workday
Payments are secure & encrypted
Powered by Stripe
Powered by PayPal

Tank models whose parameter values are identified by Kalman Filter are proposed and their applicability to forecasting runoff are discussed. Using this system runoff can be forecasted even if flood data parameters are not identified and rainfall can not be measured exactly. Three models are proposed and described: Model I is a series-storage type three-zone model; Models II and III are single-zone models. The locations of outlets and water levels in Models I and II and the coefficient and the location of outlets and water levels in Model III are identified by Kalman Filter. Rainfall events of three watersheds were applied to the three models. When the response time of the rainfall-runoff system was many times longer than the recording interval, the error decreased. Hence, in a watershed, the shorter the sampling period is, the greater the accuracy of forecasting runoff will be. Twenty-seven rainfall events in the watershed where the sampling time was long were applied. The results of Models I and II showed a time lag at the peaks of forecasted runoff for one recording interval with measured runoff. In the results of Model III, the time lag did not appear, but at the peak forecasted runoff was larger than measured, and after the peak forecasted runoff oscillated. Model III exhibited the least error, and thus is recommended as the best model for forecasting runoff.