A probability-based renewal rainfall model for flow forecasting

In real-time flood warning systems, sufficient lead-time is important for people to take suitable actions. Rainfall forecasting is one of the ways commonly used to extend the lead-time for catchments with short response time. However, an accurate forecast of rainfall is still difficult for hydrologists using the present deterministic model. Therefore, a probability-based rainfall forecasting model, based on Markov chain, was proposed in this study. The rainfall can be forecast one to three hours in advance for a specified nonexceeding probability using the transition probability matrix of rainfall state. In this study, the nonexceeding probability, which was hourly updated on the basis of development or decay of rainfall processes, was taken as a dominant variable parameter. The accuracy of rainfall forecasting one to three hours in advance is concluded from the application of this model to four recording rain gauges. A lumped rainfall-runoff forecasting model derived from a transfer function was further applied in unison with this rainfall forecasting model to forecast flows one to four hours in advance. The results of combination of these two models show good performance with agreement between the observed and forecast hydrographs.