ABSTRACT
Vicens-Miquel, M.; Tissot, P.E.; Colburn, K.F.A.; Williams, D.D.; Starek, M.J.; Pilartes-Congo, J.; Kastl, M.; Stephenson, S., and Medrano, F.A., 0000. Machine-learning predictions for total water levels on a sandy beach.
Coastal inundation can significantly impact beach management and conservation efforts, particularly as the frequency of such events increases due to relative sea-level rise. Therefore, improved, timely models are needed to predict the potential degree of coastal inundation accurately. Existing numerical models and observational platforms were investigated, and their capabilities and limitations were highlighted. Despite advancements, accurate coastal inundation forecasting remains a challenge. This is primarily because tide gauge water-level measurements are approximately a meter below the actual total water level, which includes wave run-up. To address this issue, a machine-learning approach is proposed to predict total water levels on a beach, incorporating local and regional metocean data and wave run-up along the central Texas Gulf Coast. This unique approach involves installing cameras in the study area to record images and videos every 30 minutes. Additionally, beach-profile surveys were conducted bimonthly to measure the elevation across the berm using conventional surveying methods. Beach-profile surveys facilitated the creation of digital elevation models, which, combined with imagery, allowed for extracting the wet/dry shoreline elevation. This elevation indicates the most landward point reached by water on the beach, combining tides, surge, and wave run-up, defined in this paper as the total water level. Combining this technique with tide gauge and wave buoy data facilitated the creation of a 1-year dataset that served as input for a multilayer perceptron model. Two versions are presented: a high-performance and an operation-ready model. Both satisfy the National Oceanic and Atmospheric Administration criterion that the central frequency of 15 cm exceeds 90% for both 24- and 48-hour predictions during nonfrontal months. The success of this novel machine-learning application for predicting total water levels, including wave run-up, is presented; however, model performance during the frontal months will require additional improvements.
