Wave runup observations are key data for coastal management, as they help validate predictive models for the inundation frequencies and erosion rates. Efforts to develop automated algorithms that effectively identify the instantaneous water line from video imagery have led to a plurality of methods. However, under dissipative conditions, the presence of a seepage face often hinders proper extraction and requires time-intensive data quality control or manual digitization. In this study, we explore traditional color contrast (CC) preprocessing techniques and develop a novel method that incorporates a measure for texture roughness — specifically, local entropy — alongside saturation. The CC-model showed good agreement with the manually digitized water line (0.12 m root-mean-square-
error (RMSE) and correlation coefficient (r) of 0.94), and with its runup statistics (0.08 m
RMSE, and r of 0.97 for the 2% runup exceedance, R_2%). The timing aspect of the method also
showed good agreement (3.88 s RMSE, and r of 0.70 for T_m−1,0). Concurrently, a convolutional
neural network (CNN) informed by CC-preprocessed images was cross-validated using nine manually labeled video time series, each lasting 1 hour and 30 minutes. The CNN model demonstrated good agreement during cross-validation with manually labeled time series (0.10 m RMSE and r of 0.96 for the full-time series, and 0.09 m RMSE and r of 0.97 for R_2%). The temporal dimension of the CNN estimate was also satisfactory (3.51 s RMSE, and r of 0.79 for Tm_−1,0). The observed R_2% values showed the best agreement with the formula for extremely dissipative conditions from Stockdon et al. (2006), with RMSE-values lower than 0.13 m and r-values that exceeded 0.70 for all three methods. When applied to other datasets, the CNN method occasionally failed to accurately capture the water line due to specific characteristics of the new timestack images. These results validate our ML method as a viable proof of concept and challenge us to enhance its adaptability and accuracy across varied environmental conditions. Despite these limitations, the CNN method can be effectively implemented for long-term runup analysis. Additionally, the CC method is anticipated
to be applicable across similar beaches along the northern Gulf of Mexico for long-term extreme value analysis and wave-by-wave analysis. Both methods demonstrate potential in reducing the time required to extract the instantaneous runup from video imagery under dissipative conditions and enhance real-time monitoring, enabling better predictive modeling of coastal processes.