A model parameter identification technique based on the Bayesian inference is presented and applied to a long-term application of a distributed rainfall-runoff model. Initial model parameters are given from a probability distribution of values, which is called as "prior" distribution. This prior distribution is then updated by incorporating available observed data, producing a "posterior" probability distribution of model parameter values and the technique of update is repeated. Seven model parameters of the distributed runoff model, which are directly or indirectly related to generate surface and subsurface runoffs, are identified using this technique when the model is applied to the Daido River watershed of Shiga prefecture, Japan. The identified model parameters are capable to represent the watershed characteristics for application of distributed model. Most of the model parameters, population variances are found to be low to medium, which supports the stability of identified model parameters. However, model parameters in some cells show relatively high variance which expresses considerable uncertainty. The estimated hydrograph using the sequentially updated model parameters are well suited with the observed hydrograph. The identified model parameters can produce a good model efficiency criterion R^2 of 84.82%. The proposed model parameter identification technique could be a tool for understanding the spatial properties of the model parameters of a distributed rainfall-runoff model.
View full abstract