Datasets:
EnvShip-Voyage — Data Card (voyage_v1)
Task. Long-term vessel trajectory imputation (primary) and prediction (secondary) from AIS. Region. DMA (Danish waters), bbox {'lat_min': 53.5, 'lat_max': 59.0, 'lon_min': 7.5, 'lon_max': 14.0}. Ship classes. ['cargo', 'tanker']. Sampling. native 1 min; flagship view 5 min. Windows. [12, 6] h, stride 6 h. Masks. continuous curriculum + MCAR + real gaps (frozen bank, seed 1220). Splits. vessel-disjoint MMSI-hash [0.7, 0.1, 0.2] + temporal pretrain pool. Provenance. DMA AIS (['2026-01', '2026-02', '2026-03', '2026-04', '2026-05', '2026-06']); reuse of EnvShip stages 01-04; see PLAN + RESEARCH_NOTES.
Summary statistics
{
"n_samples": 179131,
"splits": {
"train": 118292,
"test": 40903,
"val": 19936
},
"window_len_h": {
"6": 100278,
"12": 78853
},
"difficulty": {
"medium": 82255,
"easy": 63205,
"hard": 33671
},
"ship_class": {
"cargo": 121340,
"tanker": 57791
},
"n_vessels": 5804
}
Comparison
| KDD'24 imputation | EnvShip (short-term) | EnvShip-Voyage | |
|---|---|---|---|
| horizon | 12 h | 10-60 min | 12 h / 6 h |
| sampling | 5 min | 20 s | 1 min native / 5 min flagship |
| task | imputation | prediction | imputation + prediction |
| masks | cont 6/8/10 h | - | cont 1-10 h + MCAR + real gaps |
| environment | none | anchor raster | streaming basemap + fairway network |
| region prior | none | coastline | fairway / traffic network (first-class) |
Intended use / limits
Research benchmark for maritime trajectory imputation and long-horizon prediction. Not for navigation or vessel re-identification. MMSI is pseudonymised. Derived from public AIS + OSM + ETOPO 2022 bathymetry.