Tools and Approaches for Evaluating Uncertainty in Streamflow Predictions in Ungauged UK Catchments

A main limitation of using conceptual models for predicting flow in ungauged catchments is the errors in identified relationships between calibrated conceptual parameters and known (or estimated) catchment descriptors. It is hypothesised here that these errors may be reduced if the modeller does not explicitly identify relationships, but applies all feasible models within a Bayesian averaging scheme. This maintains the information about parameter inter-dependencies obtained as part of local calibration, and also provides a strong basis for integrating various sources of uncertainty into the predicted average flow and associated confidence intervals. A case study of UK catchments provides encouraging results.