Using stochastic state-space modelling to specify and reduce hydrograph error

A Bayesian post-processor is used to generate a representation of the likely hydrograph forecast flow error distribution using raingauge and radar input to a stochastic catchment model and its deterministic equivalent. A hydrograph ensemble is so constructed. Experiments are analysed using the model applied to the River Croal in north-west England. It is found that for rainfall input to the model having errors less than 3mm h-1, corresponding to about a 15% error in peak flow, the stochastic model outperforms the deterministic model. The range of hydrographs associated with the different model simulations and the measured hydrographs are compared. The significant improvement possible using a stochastic approach is demonstrated for a specific case study, although the mean hydrograph derived using the stochastic model has an error range associated with it