Total Water Level Predictions for a Sandy Beach: Camera-Based Measurements and an Operational Machine-Learning Model

The increasing frequency and impact of coastal inundation due to relative sea-level rise and subsidence pose significant challenges for beach management and conservation. Accurate and timely predictive models that operate in real-time are essential as they provide real-time insights for better predicting and managing these events. Real-time coastal inundation predictions typically rely on a combination of harmonic analysis, numerical model outputs, and sometimes heuristics developed by local National Weather Service offices. Harmonic predictions fail to capture water levels accurately in areas with significant metocean forces, such as our study site at Horace Caldwell Pier in Port Aransas, Texas. Similarly, traditional water level predictions that do not include the impact of wave run-up are insufficient, as this approach measures approximately a meter below the actual total water level observed along the beach. To overcome these limitations, we have developed an innovative operational machine-learning approach to accurately predict total water levels on beaches. This method integrates real-time metocean data and offshore wave measurements, providing a more accurate representation of the coastal inundation forcings. By utilizing this advanced predictive model, we can better prepare for and mitigate the impacts of coastal inundation, thereby enhancing beach management and conservation efforts.

A set of cameras and a computer were installed on Horace Caldwell Pier to collect real-time data and monitor inundation events along the adjacent beach segment. The data collection process involved capturing 30-minute interval imagery of the study area, allowing for continuous monitoring of coastal inundation events. Additionally, topographic surveys were conducted approximately twice a month using conventional surveying methods, establishing a representative set of elevation measurements. This data collection was applied to create digital elevation models (DEMs) that represent the beach morphology. By superimposing the camera imagery on the DEMs, the total water level can be accurately extracted, indicating the maximum landward extent that the water reached on the beach due to the combination of tides, surge, and wave run-up.

The goal of our operational machine learning model is to accurately predict total water levels by utilizing real-time offshore wave inputs and real-time water level data collected from the co-located Port Aransas tide gauge station. The target of the model is the total water level, and the model is trained on a 1-year dataset. Due to the limited size of this dataset, we employed a shallow multi-layer perceptron machine learning architecture. Two versions of the model were developed: an asynchronous high-performance model designed for maximum accuracy, and an operation-ready model suitable for real-time applications. Both models meet the operational requirement set by the National Oceanic and Atmospheric Administration (NOAA), achieving a central frequency of 15 cm with over 90% accuracy for 24- and 48-hour predictions during non-frontal months. The measured total water level time series also provides guidance to coastal planners regarding the application of tidal datums, which may not adequately represent the maximum landward extent of water levels such as the total water level. For instance, the Mean Higher High Water (MHHW) level was observed to be constantly below total water level measurements for this location and is therefore not a good indicator of the actual location of the high water level.

Our findings demonstrate that this machine-learning approach significantly improves the accuracy of total water level predictions, including wave run-up, thereby enhancing the ability to forecast coastal inundation. However, challenges remain in achieving the same level of accuracy during frontal months. Ongoing research will further refine the model, by adding nearshore wave predictions to ensure reliable performance throughout the year. Overall, this study marks a novel advancement in coastal inundation forecasting, providing valuable tools for beach management and conservation efforts.