USE OF MULTIPLE-TIME-STEP INFORMATION IN RAINFALL-RUNOFF MODELING

A methodology is proposed for using information from daily hydrological records in rainfall-runoff models that operate on time bases much shorter than a day. First, the calibration problem is tackled. Parameters of the model runoff generating component or production function are identified on a long series of daily data and then are adjusted effectively to the final subdaily time step. The transfer function component is estimated independently through the FDTF-ERUHDIT method based on limited event-based data on the final time step. The two functions are then brought together to form a new model, the derived model. Subsequently, the latter model is used in an operational context only in flood periods after it has been initialised through the daily model and some of its state variables have been tuned. The methodology for both model calibration and initialisation is evaluated and compared with the direct calibration and use of continuous-time models, Two well-known conceptual rainfall-runoff models (SACRAMENTO and TANK) are tested within a framework that is set up on real-world data from a Creek basin. The methodology proposed proved very efficient for model calibration on the basis of inappropriate data sets that comprise long daily observations and continuous charts for only some flood events. This commonly appears in the planning phase of a flood forecasting system. Also, an effective method for model initialisation is proposed that is useful in real-time flood forecasting.