Improving current forecast by Leveraging Measured Data and numerical models via LiESNs

Abstract

Forecasting metocean conditions is essential for applications such as navigation and maritime operations. In this work, Leaky-integrator Echo State Networks (LiESN) are investigated to combine irregular time series of measurements to numerically modeled currents, producing a forecast of currents for a port entrance channel. The evaluated method can assimilate data from irregular time series, automatically addressing missing data. Different settings that convey scenarios with data unavailability before the forecast are evaluated using the Index of Agreement (IOA) and Mean Absolute Error (MAE) as the performance metrics. Results show that while a numerical model does not improve accuracy, it improves the system’s robustness in the case of missing data.