A hybrid model for flow forecasting on large catchments

It has been shown (M. Garrick and coworkers) that on catchments whose flow variation is highly seasonal, a simple seasonal forecast for any date d, as the mean of the observed flows on that date in previous years (i.e. in the period of calibration) may be better, in the least-squares sense, than a forecast obtained from even a good rainfall-discharge model. In this paper, an attempt is made to combine the concept of rainfall-discharge modelling with that of purely seasonal prediction in order to obtain a better forecast than can be obtained from either method separately. The hypothesis, that the rainfall and discharge functions of time expressed as departure from the seasonal means are linearly related, is tested on some of the data of the W.M.O. Intercomparison, and it is shown that significantly better results are obtained with this hybrid model than are obtained with the seasonal predictor alone or with an elaborate rainfall-discharge model. It is suggested that the models of this kind might be useful on large catchments exhibiting marked seasonal behaviour.