Performance of image-based velocimetry (LSPIV) applied to flash-flood discharge measurements in Mediterranean rivers

Flash-floods that occur in Mediterranean regions result in significant casualties and economic impacts. Remote image-based techniques such as Large-Scale Particle Image Velocimetry (LSPIV) offer an opportunity to improve the accuracy of flow rate measurements during such events, by measuring the surface flow velocities. During recent floods of the Ardèche river, LSPIV performance tests were conducted at the Sauze–Saint Martin gauging station without adding tracers. The rating curve is well documented, with gauged discharge ranging from 4.8 m3 s−1 to 2700 m3 s−1, i.e., mean velocity from 0.02 m s−1 to 2.9 m s−1. Mobile LSPIV measurements were carried out using a telescopic mast with a remotely-controlled platform equipped with a video camera. Also, LSPIV measurements were performed using the images recorded by a fixed camera. A specific attention was paid to the hydraulic assumptions made for computing the river discharge from the LSPIV surface velocity measurements. Simple solutions for interpolating and extrapolating missing or poor-quality velocity measurements, especially in the image far-field, were applied. Theoretical considerations on the depth-average velocity to surface velocity ratio (or velocity coefficient) variability supported the analysis of velocity profiles established from available gauging datasets, from which a velocity coefficient value of 0.90 (standard deviation 0.05) was derived. For a discharge of 300 m3 s−1, LSPIV velocities throughout the river cross-section were found to be in good agreement (±10%) with concurrent measurements by Doppler profiler (ADCP). For discharges ranging from 300 to 2500 m3 s−1, LSPIV discharges usually were in acceptable agreement (<20%) with the rating curve. Detrimental image conditions or flow unsteadiness during the image sampling period led to larger deviations ranging 30–80%. The compared performances of the fixed and mobile LSPIV systems evidenced that for LSPIV stations, sampling images in isolated series (or bursts) is a better strategy than in pairs evenly distributed in time.