TR(BR)-131 : Real time flow forecasting

Abstract

The computer based models of rainfall runoff process are useful tools in short term management of water resources systems. The better management of these systems is possible by using short-term forecasts of streamflows. The conceptual and statistical models represent two classes of models which are frequently applied for catchment modelling and real-time forecasting. It is possible to have significant improvements in forecasts if these two types of models are used conjunctively. Further improvements in the forecasts are possible by incorporating some error updating algorithm.

A hybrid model was developed to overcome the shortcomings of the individual approaches. This model consists of a conceptual model and a statistical model. The conceptual model was used to simulate the catchment response using the input rainfall and potential evaporation data. The error sequence consisting of the difference between observed and simulated response was modelled using an ARMA model. Further, the forecasts were updated using the recent measurements by following the Kalman filter formulation which is used to combine the results of the models with noisy measurement of the discharge to obtain an improved forecast.

The data of a few Indian catchments were used for testing the methodology. The data used are not ideal to test a particular methodology since they may contain various types of errors. However, they represent the scenarios of a typical a real-life application. The results obtained are encouraging. In general, good forecasts for the lead times of the order of several hours were obtained. The filter statistics were also within the acceptable limits.