A novel machine learning algorithm for tracking remotely sensed waves in the surf zone

This paper describes a novel image processing technique that detects wave breaking and tracks waves in the surf zone using machine learning procedures. Using time-space images (timestacks), the algorithm detects white pixel intensity peaks generated by breaking waves, confirms these peaks as true wave breaking events by learning from the data's true colour representation, clusters individual waves, and obtains optimal wave paths. The method was developed and tested using data from four sandy Australian beaches under different incident wave and light conditions. Results are a representation of the position of the wave front through time, i.e., space-time data, which when shown overlaid on the original timestack shows the high degree of accuracy of the method developed here. The utility of the method is demonstrated in two ways: 1) through a comparison between the instantaneous wave speed calculated from the wave paths with the theoretical shallow water wave speed, and 2) by obtaining optical intensities that could be translated into wave roller lengths. The algorithm developed here has the potential to improve understanding numerous nearshore process such as bore propagation and capture in the surf zone, surf zone energy dissipation, surf beat and infragravity waves, and as a direct speed input for depth inversion methods.