kawihss/ais-inland-vessel-trajectories

AIS Inland Vessel Trajectories — Germany 2025

Preprocessed AIS trajectory dataset for short-term motion prediction of inland vessels, derived from German measurement stations (Kiel, Bremerhaven, Wedel) over the full year 2025.

Produced as part of the bachelor's thesis "Short-Term Trajectory Prediction for Inland Vessels using AIS Data" at RWTH Aachen University (Chair of AI Methodology and Institute of Automatic Control).

Data Description

Each parquet file contains sliding-window trajectory segments formatted as input–prediction pairs (10 context steps + 10 prediction steps at 30-second intervals, i.e. 5-minute context and 5-minute horizon).

Filename convention: {split}_{context}_{station}_{date}.parquet

• —split: train or val
• —context: geographic context label — river, channel, harbour, lock
• —station: AIS measurement site — kiel, wedel, bremerhaven

Key features per row:

• —Relative UTM displacements (dx, dy) — Z-score normalized
• —Speed over ground and course over ground (circular-encoded)
• —Rate of Turn (analytically derived)
• —Geographic context label (harbour, river, channel, lock)
• —Vessel split assignment (train/val, stratified at vessel level to prevent leakage)

Pipeline

Raw data → extraction → cleaning & outlier removal → 30-second resampling → sliding-window segmentation → stratified vessel-level split → Z-score normalization → Parquet export.

Full preprocessing code and evaluation framework:

• —motion-models (RWTH GitLab)
• —GitHub mirror

Source Data

Raw AIS XML feeds from GovData, collected at three German inland waterway measurement sites (Kiel, Bremerhaven, Wedel) throughout 2025.

Citation

If you use this dataset, please cite the underlying thesis (RWTH Aachen University, 2026).