This directory collects the methodological notes that accompany every stage of the HF-radar wind inversion workflow. Each subsection distils the rationale, implementation choices, and expected outputs so that researchers and engineers can navigate the pipelines without inspecting the source code. For a broader overview of the objectives and datasets underpinning the project, consult the repository’s main README.
aggregation: Data aggregation procedures covering pivot operations, maintenance metadata attachment, consolidation of radar and auxiliary feeds, and buoy wind height standardisation.analysis: Post-training analysis notes focused on feature-importance studies and comparative diagnostics.geo_utils: Geospatial utilities guiding STAC catalog construction, geodesic annotation, and GeoParquet finalisation.HPO: Hyperparameter optimisation workflow detailing the search strategy, evaluation surrogates, and orchestration of tuning runs.inference: Inference playbooks describing batch scoring jobs, model selection, and artefact handling.partition: Partitioning guides explaining view materialisation, dataset concatenation, and split management.training: Training references outlining baseline learning routines and fine-tuning strategies with anti-forgetting safeguards.