How much CO2 does a container ship emit per kilometre? Numbers by vessel class

A modern Ultra-Large Container Vessel running a 19,000-TEU box stack emits around 0.6 g CO₂e per laden tonne-kilometre on a well-to-wake basis. A 3,000-TEU feeder grinding up a regional coast emits about 12 g. Two orders of magnitude between the smallest and largest ships moving boxes today. On most lanes, vessel class matters more than fuel choice.

I get a version of this question every couple of weeks — from procurement teams trying to validate a carrier's emissions number, from a journalist wanting a clean comparison, from a shipper who just got an EU ETS surcharge invoice and wants to know what they were actually paying for. The honest answer is "it depends, and the dependence is bigger than people think." Below is the table I share when I want to give a real answer instead of a polite hedge.

Per-class numbers, laden, WTW

These are GLEC Framework v3.2 default factors for container ships, cross-checked against the IMO's Fourth Greenhouse Gas Study (2020) fleet averages. The right-hand column is the absolute per-kilometre figure assuming the vessel is sailing at typical laden utilisation — roughly 70% of nominal TEU capacity filled and average box weight of 10 tonnes, which is the working assumption most operators use when they don't have a primary data feed.

Two things worth flagging before someone misreads the table. First, the per-kilometre absolute number goes up as ships get bigger, even though the per-tonne-kilometre factor goes down. A ULCV burns more fuel per hour than a feeder — substantially more — it just spreads that fuel across so much more cargo that the unit emission drops. Second, the WTW factor includes upstream fuel production (the well-to-tank share), which for very-low-sulphur fuel oil adds about 18% on top of the combustion figure. If you see a number that's roughly four-fifths of these, it's a tank-to-wake figure and not the one you want for Scope 3 or CSRD reporting.

Why bigger ships emit less per tonne-kilometre

The physics underneath this is unromantic and worth saying out loud. A ship's hull resistance scales with displacement and wetted surface area, not linearly with cargo mass. Double the size of the ship and you don't double the resistance — you add maybe 60-70% to it, which means the fuel burn per tonne of cargo drops. Naval architects have been chasing this curve for forty years, which is why container ships keep getting larger even as the port infrastructure complains.

The other piece is the hotel load. A vessel running its engines, generators, accommodation, reefer plugs, and bridge electronics burns a more-or-less fixed amount of fuel for those auxiliary systems regardless of how many boxes are on deck. Spread that fixed cost over 3,000 TEU and it hurts. Spread it over 19,000 TEU and it almost disappears into the rounding. This is the same dynamic that makes a half-full passenger flight worse per passenger than a packed one, applied to twenty thousand metal boxes.

What actually moves the number

The table is a starting point. Four variables can pull the real-world figure up or down by more than the difference between two adjacent vessel classes, and any honest discussion of container ship emissions has to name them.

Speed. Fuel burn scales with roughly the cube of speed. A vessel that drops from 22 knots to 17 knots cuts daily fuel consumption by about half. The trade-off is voyage time and asset utilisation, which is why slow-steaming is a service-network decision, not a single-ship one. But if you're benchmarking a specific voyage, the speed it ran at is usually the biggest single explanatory variable. Maersk's published slow-steaming reductions on the AE6 and AE7 services are the cleanest public data on this, and they consistently show 30-50% per-voyage fuel savings against the pre-2008 schedule.

Fuel. The default factors above assume VLSFO (very-low-sulphur fuel oil). Switch to LNG and the well-to-wake number looks better — combustion CO₂ drops about 25%, but methane slip on a four-stroke dual-fuel engine pushes the WTW back up. The ICCT's 2023 methane-slip work landed on a roughly 15% WTW reduction for LNG against VLSFO, not the 25% you'd get from CO₂ alone. Methanol from a green-hydrogen pathway saves about 7% WTW once you account for the upstream production emissions of the methanol itself; the same engine running grey methanol is roughly a wash with VLSFO. The point is that "we run LNG, we're 25% better" is not a number any serious analyst should accept without a methane-slip assumption attached.

Empty mileage and load factor. GLEC defaults assume a 25% adjustment for empty containers and backhaul imbalance. On a Far East-Europe headhaul, the eastbound leg often runs at 50-60% TEU utilisation against a 90% westbound, and the network factor is the blend of both. If your carrier provides Tier 2 corrections from CII-verified voyage data, that adjustment can move the per-tonne-km figure 15-20% in either direction. It is almost always worth asking for.

Hull condition. Biofouling on a vessel that hasn't been to drydock for 18 months can add 15-20% to fuel consumption, and the operator has every commercial incentive to delay drydocking. There is no public dataset on this. You can sometimes infer it from CII trajectory if you have access to a fleet's per-vessel records, but for most practical purposes it sits inside the noise.

Worked example: Shanghai to Rotterdam

A 20-tonne box loaded on an 8,000+ TEU vessel sails from Shanghai to Rotterdam via the Suez Canal. Routed distance is 19,600 km. Pick the Post-Panamax factor at 5.5 g CO₂e per tonne-km. Plug it in:

19,600 × 20 × 5.5 ÷ 1,000 = 2.16 tonnes CO₂e for that box. Use the slightly older 7.5 g Panamax factor if the carrier deployed a smaller ship, and the same shipment comes out at 2.94 tonnes. Now repeat with the feeder figure of 12 g, the way the math would land if you put that same box on a 3,000-TEU regional ship for the whole way: 19,600 × 20 × 12 ÷ 1,000 = 4.70 tonnes. Same cargo, same route, same distance, more than double the emissions. The vessel class is the single biggest lever before you touch speed, fuel, or anything else.

The EcoFreight calculator shows this directly — you can pick a TEU class and watch the per-shipment figure move in real time. If you want to see the underlying factor tables and the WTW/TTW/WTT breakdown, the methodology page has the same data laid out.

Where these numbers come from

The per-class factors above are the GLEC Framework v3.2 defaults for container ships, originally derived from the IMO's Fourth Greenhouse Gas Study (2020) fleet inventory and updated using Smart Freight Centre's verified carrier submissions. The per-kilometre absolute figures are calculated from those factors using the laden utilisation assumption I described — 70% of nominal TEU, 10-tonne average box weight, which is what GLEC documents as its working assumption when no primary data is provided. The IMO study itself is the most thorough public dataset on fleet-level container ship emissions; the Carbon Intensity Indicator data MEPC publishes annually is a finer-grained, per-ship version of the same exercise.

For the LNG and methanol comparisons, the ICCT methane-slip work is the source I'd trust over any operator press release. Their measured slip rates for low-pressure dual-fuel four-stroke engines are roughly 3.5% of fuel mass at low load; for two-stroke high-pressure engines it's closer to 0.2%. That difference is the entire reason LNG numbers vary so wildly across reports.

One opinion I'm not going to hedge

I do not believe the per-tonne-kilometre numbers from any operator that does not name the trade lane and the vessel class. The variance between an empty headhaul on a Sub-Panamax and a fully-loaded Neo-Panamax on the same nominal trade is more than 3x. If the press release says "we cut emissions 20%", ask which lane. If they can't tell you, the 20% is a marketing number, not an emissions number. There is no polite way to put this.

The gap I'm not pretending to solve

The IMO's CII rating system technically assigns A through E grades to individual vessels every year, and in principle those grades should let a charterer pick the cleanest ship for a specific voyage. In practice the public CII data lags by 12-18 months — the 2024 ratings became broadly accessible in mid-2025, and 2025's won't be public until late 2026. Real-time decisions about which ship to put your cargo on still rely on AIS feeds, bunker delivery notes, and direct carrier disclosures. The CII regulation does the right thing, eventually. It just doesn't do it in time for the booking conversation, and there is no public mechanism I know of that closes that lag.

Sources and notes

Per-class factors and the laden utilisation assumption are from the Smart Freight Centre GLEC Framework v3.2 (2023), container ship table. Fleet-level cross-check from the IMO Fourth Greenhouse Gas Study 2020. LNG methane-slip figures are from the ICCT's working papers on low-pressure dual-fuel engines (Comer et al., 2023). EU ETS Maritime allowance coverage from Article 3gd of the amended EU ETS Directive. Carbon Intensity Indicator methodology from IMO MEPC 78. Per-kilometre absolute figures recomputed from factor × assumed laden cargo mass; underlying spreadsheet available on request from the EcoFreight team.

To calculate for your lane, pick the vessel class and let the calculator do the per-leg arithmetic; the GLEC v3.2 factor table is documented on the methodology page, and the per-revision record sits in the methodology changelog. The same vessel-class breakdown ships on every response from the freight emissions API. For a worked example of what happens when these ships have to take the long way around, see the Strait of Hormuz piece.