Reducing the operator effect in LSPIV image-based discharge measurements

The operator effect is a prominent error source in image-based velocimetry methods. The LSPIV method is known to be sensitive to the parameters and choices of the user, as shown in the literature and emphasized by the results of a video gauging intercomparison, the Video Globe Challenge 2020 (VGC2020) (Le Coz et al., 2021). The intercomparison was carried out during the COVID-19 lockdown of spring 2020 and involved 15 to 23 participants using the LSPIV method among other techniques on 8 videos representative of the diversity of river gauging conditions and imaging viewpoints. Each video came with a discharge reference and associated uncertainty. An in depth investigation of the intercomparison results has been carried out to identify the most sensitive parameter(s) for each video and also to review the common setting mistakes (cf. Bodart et al., 2022). The investigation highlighted the strong impact of the image temporal sampling (extraction framerate) and of the velocity filtering on the discharge errors. The ortho-rectification and the surface coefficient were also found to be impacting in given cases. Based on these observations, several assistance tools and automated filters are proposed to reduce the operator effect. They are evaluated on the intercomparison dataset. The assistance tools use available information (e.g. transect data) or basic user inputs (e.g. manual spotting of some velocities) to determine the optimal extraction framerate, grid points and searching area (SA) for LSPIV computation. The sequence of automated filters is built for the specific context of discharge measurement: spatial coherency of the velocities in a local neighborhood and temporal coherency of the velocities computed at a point. These velocity filters are systematic and do not require any input from the user. The application of the assistance tools and automated filters to the intercomparison dataset leads to a significant improvement of the results. On the eight videos, the mean interquartile range of the percent error initially at 17% is reduced to 2% and the mean median of the percent error initially at -9% is reduced to 0.6% with the assistance tools and filters. The results are encouraging and can be implemented in software tools for the operational deployment of the LSPIV method for discharge measurement.Le Coz, J., Hauet, A., and Despax, A. (2021). The Video Globe Challenge 2020, a video streamgauging race during the Covid-19 lockdown, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-2116, https://doi.org/10.5194/egusphere-egu21-2116, 2021Bodart, G., Le Coz, J., Jodeau, M., and Hauet, A.: Quantifying the operator effect in LSPIV image-based velocity and discharge measurements, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4457, https://doi.org/10.5194/egusphere-egu22-4457, 2022.