An ensemble-based approach that combines machine learning and numerical models to improve forecasts of wave conditions

Abstract

This study investigates near-shore circulation and wave characteristics applied to a case-study site in Monterey Bay, California. We integrate physics-based models to resolve wave conditions (based on inputs from a global wave model, wind data from an operational weather platform, and predictions from a regional flow model) together with a linear machine learning algorithm that combines forecasts from multiple, independent models into a single 'best-estimate' prediction of the true state. The Simulating Waves Nearshore physics-based model is used to compute wind-augmented waves in coastal and inland waters. Wave-condition data reported every 30 minutes were gathered from the National Oceanic and Atmospheric Administrations National Data Buoy Center. These data permit evaluation of the fundamental model performance, training of the machine-learning algorithm, and assessment of the ability of the integrated system to make predictions. Ensembles are developed based on multiple simulations perturbing input data to the model. A learning-aggregation technique uses past observations and past model forecasts to calculate a weight for each model. Finally, we compare the weighted ensemble predictions (forecasts) with measured data to evaluate performance against current state-of-the-art.