Uncertainty analysis of moving-boat ADCP discharge measurements

The moving-boat Acoustic Doppler Current Profiler (ADCP) gauging method has been extensively used to measure the discharge of rivers and canals. To assess the quality of the data produced for all gauging situations, it is necessary to conduct an uncertainty analysis. It indicates to the operator the sources of error which contribute the most to the final uncertainty, in order to optimize the measurement process. To quantify measurement uncertainties in hydrometry, two approaches can be considered: the analytical propagation method of uncertainties, which makes it possible to "calculate" the uncertainty of a measurement result, and the inter-laboratory experiment method, which makes it possible to “measure” the uncertainty of a measurement method in the conditions of the experiment. The evaluation of the uncertainty is however a difficult task because of the complexity of the ADCP data workflow and the lack of discharge reference in rivers.

The OURSIN method has been developed according to the GUM framework. The uncertainty of discharges extrapolated in unmeasured areas is estimated from sensitivity calculations. The uncertainty due to transect-to-transect flow variability is computed from the coefficient of variation of single-transect discharges. This coefficient of variation is estimated through Bayesian inference to supplement the limited observational information with prior information. To evaluate and validate the OURSIN method, two inter-laboratory experiments were conducted in the Rhône River at Génissiat (2010) and the Taurion River at Chauvan (2016). These experiments, consisting of 634 and 574 ADCP gaugings respectively, cover different measuring conditions in terms of sites and flow ranges (from approximately 14 to 440 m3/s).

For each experiment, uncertainty results using the OURSIN method were compared to the empirical uncertainty estimates obtained using the inter-laboratory method, ranging from 4% to 13%, at the 95% probability level. The comparison shows that the OURSIN uncertainty estimates are similar to the empirical uncertainty results. The decomposition of uncertainty sources is relevant to the conditions observed on site. The OURSIN uncertainty calculation is now implemented in the QRevInt quality assurance software. This allows a wider use of this method while continuing the research on the quantification of the most complex uncertainty components.