Voyage & flight trajectory optimisation (weather routing, fuel, emissions, contrails)

Fuel is the largest voyage cost and the main emissions source for ships and aircraft, and since 2024 shipping pays for emissions in the EU. Optimisation engines combine weather forecasts, ML performance models of the hull or aircraft, and commercial constraints to choose route, speed and altitude per voyage or flight. For flights, the same engine can also steer around contrail-forming regions. Advisory to captains, dispatchers and pilots; the human always decides.

Problem & value

• Today: masters and dispatchers follow routing advice from weather services, noon reports and rules of thumb; ship speed is often dictated by charter terms ("full speed, then wait at anchor"). Airlines use flight-planning systems with wind-optimal routes but mostly ignore non-CO2 effects.
• AI value: (1) more accurate fuel-vs-speed models learned from sensor data instead of sea-trial curves; (2) joint optimisation of route, speed and trim against forecast uncertainty; (3) just-in-time arrival to cut anchoring waiting; (4) pricing carbon and FuelEU penalties into every decision; (5) contrail avoidance for aviation.
• Regulatory pull (EU): shipping entered the EU ETS in 2024 with a phase-in of surrender obligations (40% of 2024, 70% of 2025, 100% of 2026 emissions); FuelEU Maritime applies GHG-intensity limits on energy used from 2025; the IMO CII rates ships annually A–E [1][2][3].
• Benefits: vendors report fuel savings in the single-digit percent range; treat these as vendor claims and measure against your own baseline.

Variants

• Maritime weather routing and voyage optimisation: tankers, bulkers, container ships; pre-voyage route plus en-route re-optimisation.
• Speed and just-in-time arrival: port-call optimisation with berth information (see OPT-037).
• Trim, ballast and hull-performance optimisation: ML on sensor data, hull-cleaning timing.
• Fleet-level compliance optimisation: allocating speed, fuel types and pooling across the fleet to meet CII, FuelEU and ETS exposure.
• Aviation fuel and trajectory optimisation: cost-index and vertical profile, plus contrail avoidance by small altitude or lateral changes [4].
• Wind-assisted ships: routing that accounts for rotor sails or kites.
• Private use: sailing yacht routing (e.g. open tools such as weather routing in OpenCPN or Windy-style apps); no regulatory dimension.

How it works

1. forecasts (NWP, wave, current, ice, humidity) + vessel/aircraft sensors, AIS/ADS-B
2. data cleaning, performance model (ML + physics: resistance, propulsion, fuel burn)
3. forecast ensemble / uncertainty
4. trajectory optimiser (isochrone / dynamic programming / graph search / MIP / RL) objective = fuel + carbon cost + time penalties + safety margins
5. advice to master / dispatcher (route, speed profile, altitude)
6. human decision and execution
7. feedback loop: actual vs. predicted consumption, model retraining

• Performance model: gradient boosting or neural networks fitted to noon reports and high-frequency sensors, corrected for weather, draught and fouling; physics-based models stay as prior and sanity check.
• Optimiser: classical algorithms (isochrone method, Dijkstra/A* on a 3D-4D grid, dynamic programming) remain the backbone; RL and learned surrogates speed up re-planning.
• Contrails: a forecast of ice-supersaturated regions (ECMWF humidity fields plus satellite-based ML calibration) adds a climate-cost term to the flight plan.
• LLM layer: natural-language summaries for masters, charterer reports and what-if questions on top of the solver (see OPT-018). The LLM never computes the route.
• Human-in-the-loop: the master retains final authority under SOLAS and the ISM code; dispatcher and pilot-in-command for flights. Safety constraints (heavy-weather limits, ice, piracy zones, traffic separation schemes) are hard constraints.

Data preparation

Data needed: vessel or aircraft performance curves, noon reports and sensor data (fuel flow, speed, draft, trim), weather and ocean forecasts, port and airspace constraints, charter-party terms and fuel prices. In an SMB shipping company or charter operator this typically sits in Excel noon reports, e-mail, PDF charter parties, the fleet-management system and the ERP; on the flight side in the flight-operations system and AIP/NOTAM feeds.

Typical gaps: manually typed noon reports with errors, sensors with drift, performance curves outdated after hull fouling, contract clauses only in scanned PDFs, and inconsistent vessel or airport naming across systems.

Preparation steps: connect sensor and report streams to one time-series store; clean and validate reports (outlier checks, unit harmonisation); re-estimate performance models from recent data; extract speed and weather clauses from contracts, for example with DOC-014; define a shared vessel and port master list. Weather data is bought as a licensed feed, not prepared.

Expect a medium share of effort, about 25-35 percent. Useful references: PREP-005 for system exports, PREP-011 for turning Excel reports into tidy tables, PREP-006 for scanned contracts and GUIDE-003.

Implementation paths

Which path for whom

• enterprise (A): Shipping lines and airlines get the best value from established weather-routing and flight-planning products with certified weather data and compliance reporting, which are hard to replicate in-house.
• developer (B): Teams building their own planning tools combine commercial weather APIs with optimisation solvers and vessel or aircraft performance models; self-hosting rarely pays off because of weather-data licensing.

Path A — Buy (recommended)

• Maritime: StormGeo s-Suite (Norway, EU-adjacent), ZeroNorth (Denmark, EU), DeepSea (Greece/UK, AI-based routing), Wärtsilä Fleet Operations Solution / Navi-Planner (Finland, EU), Orbital/ Marorka-type performance monitoring, NAPA (Finland, EU), Weathernews (SPOS), Veson Nautical for commercial voyage management.
• Aviation: Lufthansa Systems Lido (Germany, EU), Sabre, Jeppesen, Navblue (Airbus, EU/France), Flightkeys (Austria, EU); contrail services from Google Research/Breakthrough Energy pilots with airlines, and from EU providers via research projects [4].
• Choose A when: fleet below a few hundred vessels or a normal airline; you need class-approved or certified tools and 24/7 support. Check what ship data leaves your fleet and where it is hosted.

Path B — Build on cloud APIs

• Weather data: Copernicus Marine Service and Copernicus Climate Data Store (free, EU), ECMWF open data and AIFS forecasts, NOAA GFS/WaveWatch III; commercial APIs such as Meteomatics (Switzerland) or StormGlass.
• Compute: cloud MIP/CP solvers (Gurobi, CP-SAT) or custom services on EU-region cloud; streaming ingestion of vessel telemetry (MQTT, Kafka).
• Choose B when: you are an operator with an in-house data team needing proprietary objective functions or fleet-wide integration with your own TCE and ETS accounting.

Path C — Self-host (open source)

• Routing and weather: open-source weather routing (OpenCPN weather_routing plugin, GRIB data), libraries such as scikit-learn, XGBoost/LightGBM, PyTorch, OR-Tools, HiGHS; trajectory libraries (e.g. OpenAP for aircraft performance, pycontrails by Breakthrough Energy for contrail modelling [5]).
• Forecast models: open AI weather models (ECMWF AIFS, GraphCast, Pangu-Weather) feed ensembles; see PRED-017.
• Hardware: HW1 for the optimiser; forecast post-processing at HW2/HW3 if you run AI weather models yourself. Ops effort: a small data/OR team.

Path D — Custom training

Worth it for large fleets with high-frequency sensor data: vessel-specific fuel models, hull fouling prediction, RL policies for re-planning. Needs one to several years of voyage and sensor history; data quality (noon report errors, sensor drift) is the main cost.

State of the art (2026)

• ML hull/fuel models replace sea-trial curves; hybrid physics-ML models are standard in leading products. Errors in forecast weather still dominate the savings potential.
• AI weather models (AIFS, GraphCast and successors) provide cheap ensembles, enabling risk-aware routing instead of single deterministic routes.
• Compliance-aware optimisation: ETS, FuelEU and CII are now part of the cost function; pooling and fuel-mix decisions are optimised at fleet level.
• Contrail avoidance has moved from research to airline trials; reported results from trials show that avoidance is feasible but that forecast accuracy and the uncertain climate benefit remain open questions, and avoidance may cost extra fuel [4][5].
• Limits: forecast error, noisy self-reported data, speed constraints in charter parties (split incentives between owner and charterer), safety margins, and rebound effects (slow steaming raises required fleet capacity). Savings claims are rarely independently verified.

Shared resources & combinations

• Same stack (optimization-solver + time-series + tabular-ml): OPT-001 road routing, OPT-005, OPT-008 network design, OPT-037 port operations (berth ETAs feed arrival windows).
• Data: PRED-017 weather forecasting, PRED-007 for energy/fuel price signals.
• Recovery: OPT-020 disruption management for re-planning after delays.
• LLM front-end: OPT-018 natural-language what-if on the optimiser.

Effort, cost & KPIs

• Path A: weeks to a few months for onboarding; SaaS pricing usually per vessel or per flight; the main work is sensor/noon-report data feeds and procedures.
• Path B/C: 6–12 months for a production system with a small OR/ML team and weather-data licences.
• Path D: multi-year programme; justified for large fleets.
• KPIs: fuel per nautical mile or per tonne-mile normalised for weather, CII value and trajectory, ETS and FuelEU cost avoided, waiting time at anchor, forecast error of predicted vs. actual consumption, advice acceptance rate by masters, safety incidents (must not rise).

People & change

Management and the commercial and operations leads decide targets (fuel, punctuality, charter-party obligations). Masters, officers, dispatchers and pilots are the end users; involve them early, as they keep final authority over route and safety decisions. IT and IT security cover vessel connectivity and data flows; suppliers (weather and routing vendors, charterers) are part of the data chain. Legal and compliance check charter-party consequences and CII, ETS and FuelEU reporting.

Works council: depends. If the tool logs individual crew or pilot behaviour or compares performance of named persons, co-determination under §87(1) no. 6 BetrVG is likely; a pure fleet-level route recommender is less critical. Note that seafarers may fall under special rules (Seebetriebsrat), so check early. Inform the works council in good time under §90 BetrVG when work processes change.

Data protection: crew data is limited, so a DPIA is usually not needed unless individual performance is evaluated; involve the DPO in any case. Run the pilot on a few voyages in shadow mode against the existing routing, measure fuel and arrival accuracy, and address the main concern, loss of seamanship, by keeping the recommendation advisory and the captain's override explicit. Plan AI literacy measures for officers and dispatchers. See GUIDE-009, GUIDE-008 and GUIDE-004.

Risks, regulation & compliance

• EU AI Act: advisory routing is generally minimal risk. Systems that are safety components of regulated aviation or maritime products fall under sectoral type-approval, and the AI Act's high-risk rules reach aviation and ship equipment mostly through delegated acts under sectoral law; check the product category. AI literacy (Art. 4) applies.
• GDPR: mostly no personal data; AIS and crew data, voyage-data-recorder and onboard monitoring can relate to identifiable crew (performance monitoring, works council); apply data minimisation and agreements with the vendor.
• Maritime/aviation rules: SOLAS, ISM and the master's overriding authority; EU MRV/ETS reporting needs verified data; FuelEU and CII reporting must be traceable to measured fuel. Cybersecurity (IACS UR E26/E27) for connected ships.
• Risks: over-trust in advice in heavy weather, data manipulation for CII/ETS reporting, commercial disputes over charter performance claims, greenwashing with unverified savings.
• Mitigations: hard safety constraints, master override, audit log of advice and decisions, independent verification of savings.

Getting started

1. Baseline: collect fuel, speed, draught and weather data for 12 months and compute your CII and ETS exposure per vessel.
2. Check data quality of noon reports and sensors; install flow meters where missing.
3. Pilot 2–3 vendors from Path A on a trade lane with A/B comparison against unrouted voyages.
4. Add emissions cost terms and charter-party constraints to the objective.
5. Review results quarterly with masters; retrain performance models.

Sources

1. European Commission — EU Emissions Trading System (incl. maritime) — https://climate.ec.europa.eu/eu-action/carbon-markets/eu-emissions-trading-system-eu-ets_en
2. European Commission — FuelEU Maritime — https://transport.ec.europa.eu/transport-modes/maritime/decarbonising-maritime-transport-fueleu-maritime_en
3. IMO — Carbon Intensity Indicator (CII) — https://www.imo.org/en/MediaCentre/HotTopics/Pages/Reducing-greenhouse-gas-emissions-from-ships.aspx
4. Google Research — Contrails project — https://sites.research.google/gr/contrails/
5. pycontrails (Breakthrough Energy) — https://github.com/contrailcirrus/pycontrails
6. OpenAP aircraft performance model — https://github.com/junzis/openap
7. ECMWF — AIFS machine-learning forecasting — https://www.ecmwf.int/en/forecasts/datasets/aifs-machine-learning-data
8. Copernicus Marine Service — https://marine.copernicus.eu
9. Regulation (EU) 2024/1689 (AI Act) — https://eur-lex.europa.eu/eli/reg/2024/1689/oj