REGIONALISATION OF JOINT DISTRIBUTION OF MODEL PARAMETERS: PREDICTION ON UNGAUGED BASINS

摘要

In the case study presented in this work, the calibrated parameters of a Conceptual Rainfall Runoff (CRR) model could not be uniquely identified. Moreover, the covariance and variance of model parameters varied among basins. The uncertainty in the calibrated model parameters have been recognized as a concern for successful model parameter regionalisation. Therefore, regionalisation should include assessment of model parameter uncertainty. For this reason, we proposed a regionalisation scheme which, instead of regionalizing the single best value of model parameters, regionalizes their joint probability distribution. A functional relationship between catchment attributes and characteristic parameters of joint distribution are developed. The uncertainty in model prediction is quantified using multivariate stochastic simulation technique, whereby a large number of model parameter vectors are sampled from their estimated joint distribution that lie within 90% confidence region and fed into the CRR model. Regionalisation of joint distribution addresses the effect of model parameter uncertainty in the result of regionalisation by taking into account the covariance between model parameters which are generally ignored in approximate analytic techniques based on Taylor series expansion. In the case study presented in this paper, the predictive uncertainties quantified from the proposed method closely followed the prediction uncertainties quantified from the calibrated joint probability distribution of model parameters.