Natuna loitering dataset

Problem

Loitering — a vessel lingering without an evident purpose — is a useful signal and a hard one to share. Everyone re-derives it from raw tracks, so no two analyses agree on what counts or where it happened.

Approach

The dataset fixes loitering events to an H3 spatial index and a time bucket, so a claim about “the shelf, last spring” resolves to specific cells and intervals anyone can query. It ships as Parquet, readable directly from DuckDB without a server.

• Detect loitering from cleaned tracks with a documented threshold.
• Index each event by H3 cell and time bucket.
• Publish as Parquet with a schema and a query cookbook.

Outcome

A record others can query off the Natuna shelf and cite precisely — the same cell, the same window, the same answer.