MARIS-Forecast: a multi-region AIS trajectory dataset with temporally aligned maritime context for vessel trajectory forecasting

MARIS-Forecast is a multi-region ship-trajectory forecasting benchmark that unifies four maritime jurisdictions (Denmark / DMA, United States / NOAA, Greece / Piraeus, Norway / Kystverket) and two prediction horizons (10-min Track A and 60-min Track B) into a single, identically-shaped release. Every sample carries an OpenStreetMap raster, a signed-distance field, a 3-km social-neighbour context, a unified ship-class label, and — new in v2.1 — 15 temporally-aligned maritime-context columns (ERA5 weather, ECMWF WAM sea state, OSM port/fairway/TSS features). Highlights 330,000 Track A windows + 106,857 Track B windows across four jurisdictions. Full inline OSM-temporal-consistency flag with historical-OSM rebuild for Piraeus 2019. Anchor-time weather (10-m wind, 2-m temperature, MSL pressure, cloud), sea state (significant wave height, direction, period, swell), port proximity (nearest port distance + name), TSS / fairway proximity — all derived from CC-BY / public-domain sources. Cross-domain evaluation splits (in-domain / zero-shot transfer / combined / LOSO) documented in the companion paper (Ma, in prep., Scientific Data). Nine pre-trained baselines (DMA Track A) shipped for reproducibility. Layout: mirrors the Hugging Face release (track_a_short-term_Cross-domain_Datasets/, track_b_medium-term_Cross-domain_Datasets/) plus documentation, licence, citation. See the accompanying README.md for a Quickstart, the DATA_CARD.md for a Gebru-style datasheet, and CHANGELOG.md for provenance of every derived column. Licensing (composite). Different files carry different licences. The Danish (DMA) trajectory data are Danish public-sector information under the PSI Act (act no. 596/2005; free reuse, no warranty, no re-identification — not Creative Commons); NOAA is U.S. public domain; Piraeus is CC BY 4.0; Norway is NLOD 2.0. All OpenStreetMap-derived rasters, signed-distance fields, vector objects and OSM-derived columns are ODbL with share-alike; the weather and sea-state columns derive from ERA5/ECMWF (Copernicus) via Open-Meteo. See the top-level LICENSE and NOTICE.md for the per-file mapping.