The Diet Target That Cuts Emissions in Rich Cities and Raises Them in Poor Ones

There is a number in this study that is true of the world and true of almost no city in it. Ask a global food model what would happen if every urban population on Earth switched to the reference diet drawn up by the EAT-Lancet Commission, and the answer comes back small: total dietary footprints fall by something between 5% and 15%. As a headline, that reads like a disappointment. A diet designed around planetary limits, adopted everywhere at once, barely moves the total. As an average, it is something else. It is what remains after two large movements in opposite directions have partly cancelled each other out.

The model is the work of Hongyi Cai, Shruti Jain and Oliver Taherzadeh at the Institute of Environmental Sciences at Leiden University, with Daniel Moran of NILU in Norway. Their paper, "The dietary footprints and transition priorities of over 12,000 urban centers", was published in Environmental Research Letters on July 31 under a Horizon Europe award, following a formal peer review of the team's earlier analysis.

What they built is a database. The method, in the authors' words, combines "demographic, group-specific diet, and supply-chain environmental footprint data to estimate the dietary footprints in 12,228 urban centers across 159 countries." Each urban centre gets a population split into demographic groups; each group gets a diet; each food item in that diet carries an environmental cost per unit, traced back through the supply chain that delivers it. Multiply and sum, 12,228 times over, and you have MATILDA-City, which the team describes as a globally harmonised base for benchmarking, ranking and monitoring urban dietary footprints.

One word in that sentence carries most of the weight, and it is "estimate." Nobody weighed what 12,228 urban centres actually ate. The baseline footprints are consumption-based accounts, assembled from statistics about diets and supply chains rather than from anything observed city by city. Every EAT-Lancet figure then sits one layer further out again: it is a counterfactual, a scenario in which a diet change that has not happened is applied to a baseline nobody surveyed directly. Read the numbers as what a well-specified model says, not as what a city emitted.

With that established, the central result is worth quoting as the authors wrote it. Adoption of the EAT-Lancet 2.0 diet across all studied centres delivers a modest net reduction in total footprints of 5% to 15%. "However," they continue, "this aggregate outcome masks substantial heterogeneity: high-income centers achieve considerable reductions through lower meat and dairy consumption, while lower-income centers presently consuming below EAT-Lancet recommended levels experience footprint increases, partially offsetting global gains." The word doing the work there is "net." The global figure is a subtraction, and the two things being subtracted point in opposite directions.

The range itself is easy to misread. The paper spells out what it spans: total footprints fall "by 5–15%, depending on impact category." That spread is not a confidence interval around a single answer. It is the gap between what the scenario does to greenhouse gases and what it does to the other pressures the model tracks, land and water among them.

Why would a poorer city's footprint rise? Because the reference diet is not only a ceiling on meat and dairy. It is also a floor, setting recommended intakes of food groups such as vegetables, fruit, legumes and nuts that many urban populations currently fall short of. Moving those diets onto the reference means eating more, and more food carries a larger footprint. The increase is what nutritional improvement looks like when it is expressed in greenhouse gases. It describes a gap being closed. It is not a case against closing it.

This is not a quirk of one model. Six years ago a separate group led by Richard Semba reported the same two-directional structure at country scale, in a paper in Nature Food that traced food from cradle to farm gate and accounted for international trade. A worldwide shift to the planetary health diet, they estimated, would lower agricultural greenhouse gas emissions globally and in 101 countries. In primarily low- and middle-income countries, it would raise them by 12% to 283%. Different authors, different data, a different spatial unit, and the same split. The new study finds it again inside city boundaries.

One more figure in the abstract invites a wrong reading, and the authors themselves supply the correction. The study looks at three transnational city networks: C40 Cities, Eurocities and the Milan Urban Food Policy Pact. Within their member centres, the same dietary switch delivers reductions of 22% to 50% in greenhouse gas emissions, several times larger than the global figure. The reason sits in the same sentence of the abstract: member centres are "predominantly high-income." Assemble a group of mostly wealthy cities, apply a scenario whose cuts fall hardest on wealthy diets, and the group posts a large cut. Nothing in that result tests whether the networks' policies work. It measures who joined them.

The paper's policy conclusion is the authors' own, and should be read that way. Complementary supply-side interventions, they write, including decarbonisation of agricultural production, processing and distribution, offer additional reductions, "indicating that dietary and supply-side levers must be deployed jointly." That is an inference the team draws from its own scenarios, not something the footprint accounting measures on its own.

Which leaves the database as the paper's most durable output. A global average of 5% to 15% tells a city government nothing about itself; it is the residue of thousands of local movements, most of them larger than the residue and many of them pointing the other way. What MATILDA-City offers instead is a single row per urban centre, built by the same accounting method for all 12,228 of them, so that one entry can be set against another. Whether any particular entry is right is a question the database now makes it possible to ask. A year ago, asking it would have meant building the database first.