Each maritime sensor sees a different shadow of the same vessel. AIS gives an identity but no presence guarantee. SAR gives a hard hull detection but no identity. RF geolocation gives a position estimate but no class. Optical confirms class but only in good light. Multi-source fusion is the layer that takes these heterogeneous, non-aligned, partially-overlapping detections and produces a single unified track per vessel — with an identity, a confidence score, and a behaviour history.
This is increasingly where competitive advantage in maritime intelligence lives. Once the underlying sensors become commodity (and they are), the differentiation is in the fusion: can the system correctly merge the AIS broadcast at 14:02 with the SAR detection at 14:11 and the RF emission at 14:23 into one vessel, with high confidence, while keeping a thousand other near-by detections separate? The leading providers (Windward, Spire's Maritime AI, Kpler, Lloyd's List Intelligence, EMSA's IMS) compete heavily on fusion quality. The technique stack draws from track-association theory developed for air-defence radar fusion, modern probabilistic graph models, and increasingly LLM-based reasoning over unstructured records (port logs, ownership filings, news).