An Open Lakehouse for Maritime Mobility Analytics Over AIS Data

Vessels broadcast their positions through the Automatic Identification System (AIS), producing archives of billions of records, often stored in data lakehouses. A lakehouse answers small, local queries by skipping files whose stored minima and maxima cannot satisfy the query predicate. This does not work for trajectories, since a trip’s bounding box encloses mostly open water and overlaps almost every query. We represent each trajectory as a first-class moving-object value cut into short segments on standard Parquet files and Iceberg catalogs, storing each segment’s spatiotemporal extent as six scalar columns, so that a generic catalog prunes them without a new format or index. A second summary, a conservative cell cover, records the grid cells that a segment’s path traverses. From one month of Danish AIS data we build twelve physical layouts that vary partitioning, in-file ordering, and segmentation granularity one at a time, read identically by DuckDB, PostgreSQL, and Apache Spark. Geographic tiling lowers the fraction of the table read per query from 34% to 2.4%, a geometric-mean speedup of 4.9 times, with full recall and under 5% extra storage, whereas in-file ordering alone yields about 1.3 times. On untiled data, the cell cover removes 29% of the candidates before decoding.