Investigating the correlation between optically identified rip channels in time exposure imagery and in situ observations, including bathymetry and flow

Abstract:

Previous research has relied on time exposure (timex) images created from video camera data to indicate the alongshore position of rip channels. In these images, it was assumed that the offshore directed flow of the rip current appears as a dark, shore-normal feature disrupting bright, shore-parallel pixels indicative of breaking waves. Maryan et al. 2019 evaluated the ability of multiple machine learning algorithms to identify these features, and found that a Viola–Jones cascade with an Ada-Boost meta-classifier was able to identify 98% of the shore-normal dark features seen in the images. In this study, field measurements of flow and bathymetry in and around a rip channel were used to evaluate how often these dark, shore-normal features were spatially correlated to offshore directed flow. A confusion matrix between the flow measurements and machine-identified rip current regions shows that although the algorithm is able to consistently identify visual features, these regions are not always indicative of strong (0.23ms^-1) offshore-directed flow. Recall for the image-based classification was only 0.68, with a precision of 0.32 and an f-measure of 0.43. This study provides a quantitative evaluation of how often dark, shore-normal features in timex imagery are indicative of rip current flow. Additionally, it provides a physics-based measure of the accuracy of the rip detection algorithm of Maryan et al. 2019. While the algorithm excels at identifying the visual signature with which it is trained, these results show that dark, shore-normal features in the timex images are not always rip currents, and that rip currents are not always visible as such. Further research is required to determine the wind, wave, and other surf characteristics which cause these dark features to appear in images despite insufficient or absent rip current flow, as well as the conditions under which flow occurs without the visual feature.