Research areas
Six classes of maritime evidence problem
These pages map questions, possible data, methods, and evaluation targets. They do not claim that every capability is implemented or validated in the public demonstrator.
Maritime Domain Evidence
What can open sources establish beyond cooperative vessel reporting?
For
Researchers, public institutions, analysts, and maritime operators
Potentially relevant data
The problem
Public AIS is partial and cooperative. Satellite imagery, registries, weather, and reporting can add context, but they have different coverage, timing, licensing, and error modes. A responsible method must show which source supports each conclusion.
Methods to test
- Compare cooperative AIS observations with bounded SAR and optical samples
- Record detection, association, coverage, timing, and uncertainty separately
- Preserve gaps, disagreement, and alternative explanations
- Require human review and cited evidence before escalating a lead
Evaluation targets
- Measured detection and association performance on a defined sample
- Visible source coverage, freshness, and missing-data states
- A reproducible evidence record for each conclusion
Identity & Sanctions Evidence
Which vessel-risk indicators remain defensible when identities and behaviour change?
For
Compliance researchers, insurers, public institutions, and maritime analysts
Potentially relevant data
The problem
A list match is only one form of evidence. AIS gaps, identity changes, ownership records, transfers, and imagery can be relevant, but none proves misconduct by itself. The question is how to combine them without turning indicators into unsupported allegations.
Methods to test
- Resolve IMO, MMSI, flag, name, owner, and registry records with dates
- Separate official designations from behavioural indicators and model inference
- Corroborate material claims across independent source types where possible
- Record false-positive risk, missing evidence, and the point at which the method abstains
Evaluation targets
- Traceable identity and source history
- Calibrated indicator performance on labelled cases
- Clear boundaries between designation, concern, and insufficient evidence
Emissions & Regulatory Evidence
How should activity estimates, reported emissions, and policy rules be reconciled?
For
Researchers, ship operators, policy teams, and sustainability analysts
Potentially relevant data
The problem
AIS-derived emissions are estimates, verified reporting has its own scope, and regulatory regimes use different boundaries and definitions. Combining them without a source and calculation record can create false precision.
Methods to test
- State the applicable rule, geography, period, vessel scope, and reporting boundary
- Estimate activity with documented vessel and fuel assumptions
- Compare estimates with available verified or reported observations
- Publish uncertainty ranges, exclusions, and sensitivity to key assumptions
Evaluation targets
- Reproducible calculations for a bounded vessel or voyage sample
- A visible reconciliation between estimated and reported values
- Sensitivity ranges that show which assumptions drive the conclusion
Trade-Flow Inference
Which conclusions survive gaps, latency, cargo uncertainty, and aggregation?
For
Economic researchers, policy analysts, and maritime market observers
Potentially relevant data
The problem
Port calls and vessel movement can illuminate trade, but cargo, utilisation, destination, ownership, and timing are not always observed. A useful flow estimate must distinguish direct observation from classification and inference.
Methods to test
- Define the unit of analysis, time window, geographic boundary, and observation coverage
- Keep vessel movement, cargo classification, utilisation estimate, and economic interpretation separate
- Test sensitivity to missing observations, delayed feeds, and alternative classifications
- Compare aggregate results with dated trade and port statistics
Evaluation targets
- A reproducible flow estimate with an explicit coverage denominator
- Error bounds for latency, classification, and missing observations
- A statement of which economic conclusions the sample cannot support
Voyage & Port Modelling
When can open observations improve an arrival estimate?
For
Researchers, ports, terminals, carriers, and cargo owners
Potentially relevant data
The problem
An ETA model can look accurate on an easy sample and fail under congestion, weather, route change, anchorage, or incomplete history. The relevant question is performance against a defined baseline across conditions.
Methods to test
- Define arrival, port-call, anchorage, and prediction-horizon labels
- Compare simple baselines with vessel-history and metocean features
- Evaluate by route, vessel class, horizon, and disruption condition
- Report calibration, error distribution, exclusions, and failure cases
Evaluation targets
- Performance against a named baseline on a bounded dataset
- Error distributions rather than one headline accuracy measure
- Documented conditions in which the model should not be used
Maritime AI Claims
How should a maritime-AI capability claim be tested?
For
Researchers, boards, investors, public institutions, and technical reviewers
Potentially relevant data
The problem
A polished maritime-AI demonstration can hide purchased data, narrow test conditions, leakage, weak ground truth, or dependence on one provider. Technical and strategic review should reconstruct the claim and identify what evidence would confirm or disconfirm it.
Methods to test
- State the claimed prediction, decision use, baseline, and acceptable error
- Trace data rights, coverage, labelling, transformations, model evaluation, and provider dependencies
- Test performance across time, geography, vessel class, and difficult cases
- Separate technical performance from market, regulatory, operational, and investment conclusions
Evaluation targets
- A claim-and-evidence register
- A reproducible evaluation design and dependency map
- Clear disconfirming tests and questions that remain unanswered
Further reading
Pressure-test a maritime evidence question.
MarineAware welcomes research critique, methods discussions, seminar invitations, and bounded collaboration proposals.