Site-specific post-processing of spectral wave forecast by learning from buoy measurements: Machine-learning pipeline

This repository accompanies a manuscript submitted to JGR: Machine Learning and Computation. It contains the code, configuration files, and retained diagnostic outputs used to evaluate a site-specific machine-learning post-processing method for ECMWF spectral wave forecasts on the Australian North West Shelf. Operational numerical wave forecasts provide essential information for maritime operations, but local systematic errors remain. At the deep-water Browse Basin study site, the ECMWF forecast can exhibit structured swell errors, including a premature swell-arrival bias. This work frames forecast correction as a supervised learning problem: the model learns from historical ECMWF directional wave spectra, contextual variables, and local buoy measurements to produce a buoy-consistent one-dimensional energy spectrum over a five-day forecast horizon.