Rainfall-runoff model calibration using informal likelihood measures within a Markov chain Monte Carlo sampling scheme

This paper considers the calibration of a distributed rainfall-runoff model in a catchment where heterogeneous geology leads to a difficult and high-dimensional calibration problem and where the response surface has multiple optima and strong parameter interactions. These characteristics render the problem unsuitable for solution by uniform Monte Carlo sampling and require a more targeted sampling strategy. MCMC methods, using the SCEM-UA algorithm, are trialed using both formal and informal likelihood measures. Each method is assessed in its success at predicting the catchment flow response and capturing the total uncertainty associated with this prediction. The comparison is made at both the catchment outlet and at internal catchment locations with distinct geological characteristics. Informal likelihoods are found to provide a more complete exploration of the behavioral regions of the response space and hence more accurate estimation of total uncertainty. Last, we demonstrate how information gained from the investigation of the response space, in conjunction with qualitative knowledge of system behavior, can be used to constrain the Markov chain trajectory.