These scripts were used to build and audit the MARIS-Forecast release. They are
provided as reference implementations to document exactly how the dataset
was produced. Input locations are configurable through environment variables
(DMA_ROOT, NOAA_ROOT, NORWAY_ROOT, MARIS_ROOT, PYTHON_BIN) with generic
data/<region> defaults; because the original build also depended on cached
OSM/reanalysis inputs and a specific software environment, these scripts are
meant to be read and adapted rather than run unmodified end-to-end.
The primary supported, portable entry points for using the dataset are the
maris_forecastpackage and theexamples/script. The pipeline below documents construction.
| Script | Role |
|---|---|
stage_17_osm_temporal_consistency.py |
Per-sample OSM-temporal-consistency check: projects every trajectory point onto the per-sample shore SDF and records inland depth / inland-point count / consecutive-inland run, then the four flags. |
pbf_to_tile_cache.py |
Rebuilds OSM tiles from a Geofabrik .pbf extract (historical-OSM rebuild for the 2019 Piraeus subset; failed-tile refill for DMA/Norway). |
apply_unified_taxonomy.py |
Maps raw ship types to the canonical eight-way taxonomy (see ../maris_forecast/taxonomy.py). |
enrich_norway_static.py |
VesselFinder static-information enrichment for the Norwegian feed. |
build_meteo_features.py |
Attaches the 15 anchor-time weather / sea-state / port / traffic scalars (ERA5, ECMWF WAM, OSM). |
stage_10b_track_b_filter.py |
Track-B duration prefilter (segments long enough for 30 + 60 min windows). |
publish_dual_subset.py |
Emits the all-sample and OSM-consistent filtered views with side-car flags. |
verify_v2_artifacts.py |
Checks array shapes, CSV↔tensor row alignment and flag-column presence across all splits. |
aggregate_stats.py |
Per-partition summary statistics. |
reproduce_all.sh · post_rebuild_refresh.sh |
Orchestration wrappers (paths must be edited for your environment). |
A full end-to-end rebuild requires the upstream AIS snapshots, cached OSM/meteo inputs and a matching software environment, some of which are pinned to the original build machine. The scripts here document the procedure and the exact thresholds/parameters used; they are not a one-command turnkey rebuild.