Parametric Inference in Large Water Quality River Systems

Abstract

Environmental models often contain parameters, which are not measurable, yet conceptual descriptions of some physical process. The value of such parameters is often derived by measuring internal state model variables in the system and indirectly tuning/calibrating the value of the parameters so some degree of match is achieved. Bayesian inference is a widely used tool in which the modeller can transfer some prior beliefs about the parameter space, which is updated when additional knowledge on the system is acquired (e.g. more measurements are available). However, the amount of simulations required to perform a formal inference becomes prohibitive when using computationally expensive models. In this work the inference of the hydraulic and dissolved oxygen processes is presented for a large scale integrated catchment model. Two emulator structures were used to accelerate the sampling of the river flow and dissolved oxygen dynamics. Posterior parameter probability distributions were computed using one year of measured data in the river.