Assessment of flash flood discharges in Madagascar using vidéos: methodological simplification, sensitivity analysis and feedback

The island of Madagascar experiences an average of two tropical storms each year, bringing torrential rains and destructive flooding. Despite their impact, these floods have been poorly documented until now. However, with the rise of mobile telephony, there is a significant increase in available privately captured videos of flooding rivers. Advances in video analysis techniques, such as Large-Scale Particle Image Velocimetry (LSPIV), enable systematic use of these videos to estimate water velocities and flows. This opens new opportunities for more effective monitoring of exceptional floods.

The application of LSPIV techniques to videos taken by amateurs on their smartphones, however, poses a number of methodological problems, notably linked to the stabilization and rectification (orthorectification) of the shots, requiring the identification, in the images, of landmarks with known coordinates (at least four landmarks). The aim of the study, the results of which will be presented, is to demonstrate that it is possible to produce reliable flow estimates from LSPIV processing of videos, using landmarks located in geographic databases (Google map, Lidar surveys) instead of reference points obtained from time-consuming topographic field surveys.

This study is based on the analysis of twenty videos of controlled flows (hydroelectric power station return channels) and five videos of major floods, including four in Madagascar. In all cases, treatments based on approximate reference points were compared with treatments based on landmarks derived from topographic field surveys. Monte Carlo simulations were carried out to assess the levels of uncertainty in velocity and flow estimates associated with the location errors of these landmarks. The study also explored the usefulness of Lidar data for determining the geometry (cross-sections) of river beds, which is essential for estimating discharges.

The results show that the LSPIV approach, even in a degraded context and with controlled uncertainties, can be used to estimate extreme flood discharges in a robust manner. The proposed simplified methodology paves the way for generalizing the use of flood videos for similar environments, providing discharge estimates along with associated uncertainties, assessed using Monte Carlo simulations.