Bayesian Model Calibration Using Surrogate Streamflow in Ungauged Catchments

We present a novel approach for modeling streamflow in ungauged catchments. Because of their widespread availability and global coverage, remotely sensed data provide an attractive alternative to supplement the absence of streamflow data in hydrological model calibration. One observable signal holds particular appeal; the satellite-derived calibration ratio-measurement (C/M ratio) has been widely studied as a direct measurement of streamflow because of its physical relationship to streamflow and demonstrated correlation with in situ streamflow. This study identifies the challenges in calibrating a hydrological model using a satellite-derived C/M ratio, presenting a rationale designed to account for the limitations that these data pose. A new Bayesian calibration approach is developed that uses the surrogate streamflow derived from the C/M ratio in place of direct streamflow observations. We assess and demonstrate our approach for three Australian Hydrologic Reference Stations, which can be considered free from anthropogenic effects, with distinct attributes. The results indicate the competency of the proposed approach, showing model performance with 0.54 similar to 0.78 Nash-Sutcliffe efficiency values, with the uncertainties in the model calibration quantified via Markov Chain Monte Carlo sampling. Overall, our study finds the new model calibration promising for predictions in ungauged basins (PUB), as the global data coverage of satellite data and the suitability of the approach suggest significant improvements over traditional approaches to PUB.