Environmental and welfare gains via urban transport policy portfolios across 120 cities

City-level policies are increasingly recognized as key components of strategies to reduce transport greenhouse gas emissions. However, at a global scale, their total efficiencies, costs and practical feasibility remain unclear. Here we use a spatially explicit monocentric urban economic model, systematically calibrated on 120 cities worldwide, to analyse the impact of four representative policies aimed at mitigating transportation greenhouse gas emissions, also accounting for their economic welfare impacts and health co-benefits. Applying these policies in all cities, we find that total transportation greenhouse gas emissions can be reduced by 31% in 15 years, compared with the baseline scenario. However, the consequences of the same policies vary widely between cities, with specific effects depending on the policy considered, income level, population growth rate, spatial organization and existing public transport supply. Impacts on transport emissions span from high to almost zero, and consequences in terms of welfare can either be positive or negative. Applying welfare-increasing policy portfolios captures most of the emission reductions: overall, they reduce emissions by 22% in 15 years. Our results highlight that there is no one-size-fits-all policy. However, with context-specific strategies, large emission reductions can globally be achieved while improving welfare. Although key to reducing transport greenhouse gas emissions, not much is known about city-level policies globally. With a spatially explicit monocentric urban economic model, this study analyses the impact of four representative policies to mitigate transport greenhouse gas emissions across 120 cities worldwide.

Urban action could substantially help to close the gap between nationally determined contributions and the reductions in emissions needed to keep the world within +1.5 °C of warming [1][2][3] .This is especially true for urban transportation, which accounts for about 8% of total emissions 4,5 .However, so far, the actual potential of such local policies to reduce emissions on a global scale remains largely unknown 6 .Correspondingly, nationally determined contributions largely neglect city policies, despite urban transport emissions being a critical factor in mitigating climate change 7 .These local policies are also crucial for wider sustainability goals.Decarbonizing urban transport can indeed bring considerable benefits for a large array of issues such as cleaner air, noise and road accident reduction, or better health due to the shift to active transportation modes 8 .
Global assessments of the environmental impacts of urban transport policies have been carried out either using descriptive approaches focusing on current situations and comparing cities, or using aggregated models aiming at simulating the potential of future policies.Comparing current emissions in 274 cities worldwide and using threshold regressions, a 2015 study estimated that adequate urban planning policies could reduce emissions by about 25% in 2050 compared with a business-as-usual (BAU) scenario 9 .Using a scenario-based approach and an aggregated model, a 2019 study estimated a decrease of 21% in urban transport emissions by 2050 compared with a BAU scenario, via a reduction in travel demand and a shift to electric and more efficient vehicles 37 .Using six representative urban archetypes, a 2017 report estimated a 22% decrease via transit-oriented development, new infrastructures for mass transit, walking, cycling, next-generation vehicles and commercial freight optimization 10 .Global assessments of transport policies, including urban and non-urban transport, have also been carried out using Article https://doi.org/10.1038/s41893-023-01138-0

Aggregated impacts of the four policies
Our analysis demonstrates that policies are effective but affect each city differently (Fig. 1).Indeed, emission reductions typically stand between 21% and 32% (25th and 75th quantile), with a median reduction of 26%.The combined four policies could lead to a reduction in annual urban transport emissions of 31% compared with the baseline scenario over the sample by 2035.This emission reduction is slightly higher than existing global assessments in the literature, but of the same order of magnitude.A 2015 review of local scenarios of low-carbon urban transport strategies estimated a potential global emission reduction of 20-50% in 2050 compared with baseline scenarios 36 .Using machine learning to meta-analyse thousands of case studies of climate change mitigation in cities, a 2020 article estimated a possible decrease of 28% in 2050 through travel demand management, fuel shift and intelligent transportation systems (and potentially more, but with no clear quantification, with pan-city expansions of public transportation systems and more efficient vehicles) 32 .The two aggregated studies mentioned at the beginning of this paper estimated, also in 2050, a possible decrease of 21% and 22% compared with a baseline scenario 10,37 .Our study suggests that mitigation can happen earlier than modelled in other studies.A key reason for this difference is that we model spatially explicit policies, thus broadening the portfolio of options and improving the resolution of effects.
However, urban-scale policies are difficult to take into account in such studies because of the complexity of capturing the spatial heterogeneity, inside and within cities, of the travel demand and mode choices of households [19][20][21][22] .The local characteristics of cities, especially their urban forms, substantially impact their transportation emissions and the potential efficiency of possible policies 9,23 .For example, a high population density and the coexistence of spatially distinct job centres when cities are large enough are associated with lower emissions per capita 24 .A street-level analysis reveals that households' distances to the city centre and subcentres are a key predictor of urban transport greenhouse gas (GHG) emissions 21 .Hence, possible mitigation strategies for urban transportation depend on precise city characteristics, which are difficult to consider globally 25 .For instance, in sprawled cities, promoting electric vehicles may be more efficient than investing in mass rapid transit, while the contrary holds in dense cities 26,27 .Another difficulty comes from the fact that there is an interplay between transport policies and the real estate market.Transport policies impact housing prices, with large consequences on households' welfare and, indirectly, on long-term changes in transport demand 12,28 .
At the local scale, such mechanisms can be taken into account using city models, such as land-use transport interaction models simulating transport and land planning policies in cities 29 .Examples are numerous, with rich literature analysing case studies in various cities [30][31][32][33] .However, generalization is difficult as this field is highly fragmented, with a large diversity of methods, frameworks and indicators, and limited reference to previous works, which does not allow for accumulating knowledge or doing comparisons 32,33 .What is missing is a scalable approach that provides an assessment for a large number of cities while taking urban idiosyncrasies into account.
Here we use a spatially explicit land-use transport interaction model to systematically assess and compare, on a collection of 120 cities worldwide, the consequences of four urban transport policies on public finance, transportation emission reduction and housing affordability, as well as health benefits due to variations in air pollution, noise, car accidents and exercise through active transportation modes.The cities cover all continents except Africa, due to data availability, and count in total 525 million inhabitants, or about 20% of the total population living in cities larger than 300,000 inhabitants (see Supplementary Section C for the city selection process).
The model combines a transport mode choice model with a residential location choice model derived from the monocentric standard urban economics framework 34,35 .It simulates, in each city, the residential and transportation choices of households as a function of detailed city characteristics, such as the location of jobs, transportation costs and the local land-use policies.The model is calibrated for each city individually, with parameters structurally estimated using databases of population densities, transport times and rent levels within each city (Methods and Supplementary Section A).Thus, our model enables simulation of city-level prospective scenarios that downscale global techno-economic scenarios 35 .
The four policies that we analyse (Table 1 and Supplementary Section E) are simplified representations of four broad types of city-level transport policy: a local fuel tax targeting car use; investments in 'cleaner' transportation modes with the development of a bus rapid transit (BRT) network; a restrictive land-use regulations policy promoting urban density, in particular near public transport; and a 'fuel efficiency' policy that makes the use of low-emission vehicles mandatory.These policies are simple enough to be applied to a large sample of cities, but remain representative of existing policies.We simulate their impact in 2035 in terms of transportation GHG emissions and social welfare, and compare it with a BAU scenario in which we assume the continuation of current trends with no additional city-level policy (Supplementary Section D).

Restrictive land-use regulations
Same assumptions as in the BAU scenario, except that new construction beyond areas already built in 2020 is forbidden unless inhabitants have access to mass public transport.To accommodate the growing population, cities have to become denser.The associated increase in housing prices can negatively impact inhabitants' welfare 28 .
Fuel efficiency Same assumptions as in the BAU scenario, except that the fuel consumption of new cars decreases by 3.7% per year instead of 1%, in line with the International Energy Agency's 2 °C scenario 77 .

Article
https://doi.org/10.1038/s41893-023-01138-0public investment required by the construction of the BRT system, the increased fuel cost for the fuel tax or the increased housing prices, in particular for the restrictive land-use regulations.However, this burden is counterbalanced by health benefits through decreases in air pollution, car accidents and noise, together with an increase in active mode uses 38 .Finally, households' disposable income increases in the fuel tax scenario as tax revenues are redistributed.The fuel tax revenues depend on distances travelled as well as on the fuel consumption of private cars: therefore, our model accounts for the fact that fuel tax revenues decrease if vehicles are becoming more fuel-efficient or if the share of electric vehicles increases.We find that, when expressed in monetary terms, benefits do not seem to fully compensate for the financial losses, leading to an average decrease in welfare by 3.3% (Fig. 1).This decrease occurs in almost all cities: the median variation in welfare is −3% and only 11 cities in our sample experience a positive (but moderate) welfare increase.Our computation of welfare suffers from some limitations (see Methods section Outputs of the model and Supplementary Section A.5 for details about the welfare analysis).For instance, we account for congestion through transportation costs and, in particular, through the opportunity cost of time, as travel time data have been measured during rush hour (Supplementary Section B).However, as we do not explicitly model congestion, our welfare impacts do not account for the variations in congestion due to the policies.In Supplementary Section J, we present a version of the model in which congestion is, although very simply, explicitly modelled: it shows that, in the main version of the model, our estimates of the welfare impacts of the BRT, the fuel tax and the urban growth boundary policy are conservative, though we probably overestimate the positive welfare impacts of the fuel efficiency policy because it increases car use, by rebound effect, thereby causing traffic congestion.

Policies' effectiveness depends on city characteristics
The four policies' impacts on transportation emissions and welfare are heterogeneous (Figs. 1 and 2).Depending on the city, emissions reduction ranges from very high to almost zero, and the BRT, the fuel tax and the fuel efficiency policy have a positive welfare impact in some cities and a negative one in others.
To better understand this heterogeneity, we try to explain policies' impacts (welfare variations and emissions variation) by individual city characteristics and by city archetypes, using linear regressions and principal components, respectively (Supplementary Section F).
The BRT increases welfare and largely mitigates emissions in highly populated, low-income and rapidly growing cities with little public transport, in line with Fig. 3 displaying a large mitigation potential of the BRT in South America and a positive welfare impact in South America and Europe.This finding is consistent with the fact that this policy is largely developed in this world region 39 .By contrast, we find that the BRT has a low impact on emissions and a low or negative impact on welfare in small and high-income cities.
Public transportation availability strongly determines the emissions and welfare impacts of the fuel tax.Indeed, it allows commuters to change transport mode in response to the tax, which enables them to avoid the full cost of the tax while largely mitigating emissions.This result is consistent with the previous findings that the price elasticity of GHG emissions is twice as high in the short run if public transport

Article
https://doi.org/10.1038/s41893-023-01138-0options exist 40 .Accounting for other city characteristics, the fuel tax has the largest impact on emissions mitigation and the largest positive impact on welfare in large and dense cities, while it has a smaller impact on emissions and a negative welfare impact in big and sprawled or small cities with little public transport.Geographically, the fuel tax has the largest emissions mitigation potential in Europe and South America (Fig. 3), where cities are generally dense and/or have a developed public transport network.Income also matters, with poor cities being potentially more harmed.Yet, one limitation of our study is that we do not explicitly model income inequalities, and thus we cannot determine whether the fuel tax is regressive or progressive.Existing studies find that carbon pricing might be progressive in the transportation sector in low-income countries, as poor households are less likely to have a car and thus to be affected 41 .Geographically, the fuel tax has a positive welfare impact in North America, Oceania and most of Europe, but can be harmful in South America, Eastern Europe and Asia (Fig. 3).The restrictive land-use regulations policy largely mitigates emissions in compact cities with a high modal shift potential (Supplementary Table 8) or in large, poor and growing cities (Supplementary Table 10).Consistently, Fig. 3 shows that the restrictive land-use regulations policy largely mitigates emissions in South America, parts of China and parts of Europe.However, this policy also has a large negative welfare impact in growing cities, but that can be mitigated by the availability of public transport (Supplementary Table 7).
The fuel efficiency policy is more efficient at mitigating emissions and increasing welfare in small, high-income or big, poor and sprawled cities with little public transportation.It is less beneficial in compact cities, with good public transportation and low private vehicles modal shares.Consistently, Fig. 3 shows that the fuel efficiency policy has a similar impact in most cities, except in Europe where emissions mitigation and welfare impacts are lower.
However, the characteristics we list are insufficient to fully explain policies' impacts, as almost 50% of the variations remain unexplained (R 2 values of the regressions are between 0.24 and 0.54; Supplementary Tables 7 and 8).Moreover, for a given policy, the characteristics significantly determining its impact on emissions mitigation are generally not the same as those influencing its welfare impacts.These results highlight the utility of a spatial model explicitly accounting for cities' spatial characteristics and their interplay to capture mitigation policies' impacts.

Tailored welfare-increasing policy portfolios
As policies' impacts are heterogeneous between cities, alternative policy portfolios designed in a context-adequate way may have higher efficiencies and fewer negative side-effects than a policy portfolio consisting of the same policies for all cities.Here, we simulate a scenario in which we implement, in each city, the policy mix that maximizes emissions mitigation with the constraints that it has to increase welfare.This way, we account for the fact that some policies can decrease welfare when implemented alone, while increasing it when combined with other policies.In particular, the restrictive land-use regulations policy alone always reduces inhabitants' welfare, but may increase it when combined with other policies (Supplementary Table 13).
In such a scenario, the minimum decrease in urban transport emissions is 15.2% (the median is 18.0% and the interquartile range is  Therefore, by designing policies adapted to each city's characteristics, it appears possible to systematically improve welfare while reducing emissions by more than two-thirds of the initial figure.These results are a priori underestimated, as the policy portfolio simulated in this paper is not optimal.The magnitude of emissions reduction could potentially be increased while keeping welfare variation positive by tailoring, in greater details, the policies to each city's characteristics, using, for instance, different tax levels, or designing tailored restrictive land-use regulations policies.

Discussion
The main message of this study is that the current increase in available urban data allows us to model, although simply, the consequences of local policies in a large set of cities, explicitly accounting for their spatial characteristics.This enables us to downscale global scenarios such as those produced by IAMs at the city scale and to quantitatively assess the consequences of local strategies involving land-use planning or local transport infrastructure provision under such scenarios.Moreover, such spatially explicit modelling also enables us to capture the impact of these strategies on households' expenses related to housing and transport, and on several side-effects of the policies, especially health co-benefits.
In line with existing studies, we find that urban forms and cities' spatial characteristics impact the mitigation policies' efficiency in a complex way, with no direct one-to-one mapping 9,23,42 .Even within the same continent or country, differences can be large, and city models can help to capture this heterogeneity.In line with the literature, we also find that the positive side-effects of urban transport policies can be high, especially regarding the financial cost of these policies 38 .It appears possible to reduce emissions in a welfare-increasing way in each city while keeping most of the global emission reductions.However, a context-adequate policy portfolio is required with strategies tailored to each city.
There are many limitations to the present study.Although more and more city-level data become available, data availability still heavily constrains our modelling and scenarios.We could not include in our analysis any African cities, and there is a strong geographical bias in favour of developed countries, a common weakness of the literature on cities 30 .The current increase in available spatialized urban data, either from direct sources or predicted using machine learning approaches, should allow for expanding our analyses to larger and more representative city samples in the near future 43,44 .Furthermore, our model is simple and did not consider, for instance, any mechanism relative to endogenous job locations or description of income inequalities inside cities.Recently, models capturing these dimensions have been proposed in the literature and could be used to reproduce our analysis in the future, when adequate data about the location of jobs and income groups within cities become available 45 .We also ignored cities characterized by high levels of informal settlements.Indeed, modelling such cities is still a research challenge, as is the identification of low-carbon and sustainable mobility policies in this context 46 .
Using more sophisticated models and additional data may enable the analysis of important policies that we could not assess with our framework.The promotion of mixed land use, for instance, or the Article https://doi.org/10.1038/s41893-023-01138-0development of bicycle lane networks could not be evaluated here.We also could not capture the inequalities created by the policies, something that would require data about where richer and poorer inhabitants live within each city.The welfare variations and health co-benefits that we simulate therefore give an indication of the average effect of the policies, but should not be considered as a direct indication of the political feasibility of their implementation.With the current and continuous increase in available socioeconomic, land-use and transport data on cities, however, progress on these issues may occur in the coming years.

Urban modelling
For each city individually, we run a spatially explicit urban model based on the non-equilibrium dynamic urban model (NEDUM) 28,35 .This model combines a simple transport allocation model with a land-use model based on the monocentric standard urban model (SUM) of urban economics, or Alonso-Muth-Mills model [47][48][49] , with inertia in city evolution.It allows spatially explicit modelling of residential and transportation choices of households as a function of employment centre locations, transportation costs and land-use constraints, which enables the analysis of housing, transport or land-use policies.
The monocentric SUM is an old model but remains empirically relevant.It has been shown to successfully capture the spatial patterns of population density and housing prices in several cities across the world 34 .For instance, in Berlin, employment accessibility is a determinant of urban land prices, and the evolution of public transport supply explains the urban sprawl 50,51 .The recent rise of available urban data has allowed testing of some predictions of the SUM in a large number of cities, for instance, on urban sprawl in 329 US cities 52 , on urban sprawl and urban fragmentation in 282 European cities 53 , and on population density and land use in 300 European cities 54 .A recent research paper based on the same sample of cities as the present paper, in particular, has shown that the monocentric SUM is capable of capturing the inner structure of these cities, both in developed and developing countries 55 .In addition, the monocentric SUM requires a limited amount of data, so it can be applied to a large number of cities and suits our approach, and relies on a limited number of hypotheses and mechanisms, so its outputs are easily interpretable.It is also grounded in microeconomic theory, offering a robust framework for analysis between location and transport, contrary to the gravity/spatial interaction approach, for instance 56 .Models based on the SUM have often been used to analyse mitigation and transport policies: to cite a few recent examples, the construction of the BRT system in Bogotá 57 , London's congestion charge 58 or an urban growth boundary in Cape Town 59 .
One concern may be that, as cities are growing, they might become less monocentric, threatening the validity of the monocentric SUM.For instance, while the monocentric city model adequately represents land values in Chicago at the beginning of the twentieth century, this is no longer the case at the end of the twentieth century: because of the development of a new airport, which became an important employment centre, the city can no longer be considered as monocentric 60 .Still, recent research papers investigating large samples of cities in the USA 61 and Europe 54 show that, in the vast majority of them, the city centre continues to play a predominant role.For instance, in Berlin, even though transportation costs have decreased and the city has sprawled, the city remains roughly monocentric 62 and distance to the main centre remains the strongest predictor of transportation patterns 21 .In this paper, we carefully selected our sample of cities to ensure the validity of the monocentricity assumption (Supplementary Section C).This requirement was, however, a limiting factor in the number of cities that we could analyse and, in the future, the availability of global detailed data on job locations inside cities may enable this limit to be overcome.
Our model is fully described in Supplementary Section A and summarized below.In a first step, we assume that, within a city, households trade-off between transportation costs to employment centres and rents per unit of dwelling, resulting in rents decreasing when transportation costs increase.In this paper, we assume that households select the transport mode with the lowest generalized transportation costs, choosing between private cars, public transport and walking.
In a second step, private developers build the amount of housing that maximizes their profit, accounting for households' bid-rents and for land-use constraints.Under standard hypotheses on the construction function, this results in the construction of capital-intensive buildings near employment centres, where bid-rents per unit of housing are higher.Here, we use a dynamic version of the model, assuming housing depreciation and inertia in housing construction.
As a result, our urban model allows us to estimate population density, housing supply, transportation choices and rents as a function of employment centres' location, transportation infrastructures and land-use constraints.

Outputs of the model
Our model allows us to estimate two main outputs at the city level: GHG transportation emissions and inhabitants' welfare.The computation of these outputs is detailed in Supplementary Section A.5 and described below.
Transportation emissions are derived from the transportation demand of each mode and the GHG intensity of each transportation mode, assuming that the level of emissions per unit of distance of public transport is fixed and exogenous, and that the level of emissions of private cars per unit of distance depends on their fuel consumption.Furthermore, we assume that the private vehicles fleet is homogeneous in fuel consumption in the base year; then, the fuel consumption of new cars decreases each year at a rate that depends on the scenario, and we assume a lifespan of private cars of 15 years 63 .
Total social welfare is measured as the sum of individual utilities, derived in our framework from the consumption of housing and composite goods, and from health co-benefits related to transportation.Consumption of housing and composite goods is constrained by the level of income net of the generalized transportation cost, meaning that an increase in transportation cost or travel time will, in turn, reduce welfare.While the consumption of housing and composite goods is a standard output of urban economics models, directly resulting from households' utility maximization, we also include four health co-benefits in our analysis: exposure to noise, air pollution and car accidents, which negatively impact welfare, and the positive health impact of active transportation modes.
We compute the monetary equivalent of the impacts of air pollution, noise and car accidents, assuming that they linearly depend on the demand of transportation from private cars, as well as on the fuel consumption of private cars for air pollution.For health improvements through active transportation modes, we adapt the Health Economic Assessment Tool (HEAT) model of the World Health Organization 64 , assuming that walking or cycling to work brings reduced mortality, which translates into a monetary gain through the value of statistical life.In the HEAT model, values of statistical life vary by country and are based on a comprehensive review published by the Organisation for Economic Co-operation and Development 65 .We include these co-benefits in the utility function as described in Supplementary Section A.2.

Data sources and parameter calibration
We individually calibrate the model's city-specific parameters on each city of the sample using spatially explicit data on population densities, rents, transportation costs, land use and dwelling sizes for the 120 cities for 2015 (Supplementary Section A.6). Owing to data availability, our model is monocentric: in each city, we consider only the city centre as the main employment centre.We carefully selected our sample of cities to ensure the validity of the monocentricity assumption (see Article https://doi.org/10.1038/s41893-023-01138-0Supplementary Section C for a description of the sample selection process; see Methods section Urban modelling and Supplementary Section A.1 for a more detailed discussion of the monocentricity assumption).
We use the dataset from ref. 66, which provides spatially explicit data on population densities, rents, dwelling sizes, land use and transportation costs for 191 cities on five continents at 1 km resolution.Population density and land cover are from the Global Human Settlement Population Grid (GHS-POP) 67 and the European Space Agency Climate Change Initiative (CCI) 68 databases, respectively, while transportation and real estate data have been obtained from the Google Maps Application Programming Interface (API) and the web scraping of real estate websites.To our knowledge, this dataset is the first including spatialized data on real estate and transportation in a large sample of cities covering both developed and developing countries and allowing an integrated analysis of density, real estate, transportation and land use.In addition, we use city-level data on city characteristics, including incomes, the fuel consumption of private vehicles, fuel costs and agricultural rents.In particular, the fuel consumption of private vehicles in the base year is given 69 .All data sources are in Supplementary Section B.
We tried to assess the model's ability to reproduce urban structures: results can be found in Supplementary Section A.7 and are summarized below.First, within each city, we compared simulated densities and rents with density and real estate data (Supplementary Figs. 2 and  3, and Supplementary Table 1).For densities, the fit between the model and the data is generally good, with a minimum correlation coefficient of 0.31, a median of 0.63 and a maximum of 0.89.For rents, the fit is good for most cities (median of the correlation coefficients of 0.46, maximum of 0.84), but is low for others (the correlation coefficient is below 0.32 for 25% of the cities).The fit between the model and density data is the best for Europe and South America, and the lowest for North America, with heterogeneity within continents as well.A recent study 55 highlights city characteristics that might explain the poor fit of the SUM in some cities and world regions, including polycentricity, informal housing and local amenities.
Then, at the city level, we compared simulated modal shares and transportation emission levels with existing data.For modal shares, we compared the model's outputs with three existing databases: Deloitte data 70 , CDP data (https://data.cdp.net) and European Platform on Mobility Management data (EPOMM; https://epomm.eu;Supplementary Table 2).Among the 76 cities that are in common between our sample and at least one of these three databases, the Pearson coefficient of correlation is 0.36 (P value <0.01) for private cars, 0.63 (P value <0.01) for public transport and 0.05 for active modes (P value 0.609).An explanation for the poor fit on active modes is that, as external databases often have narrower definitions of urban boundaries, usually limited to administrative boundaries, they tend to overestimate the modal share of walking and cycling compared with our model.Another explanation is that the external databases come from the aggregation of many sources (self-reported data by cities, governments, non-governmental organizations, expert judgments and so on), threatening the validity of the comparisons between cities.
Regarding transportation emissions, we compared the model's outputs with external databases (Supplementary Table 3): Moran et al. 71 , Nangini et al. 72 and Kona et al. 73 .We find a Pearson correlation coefficient of 0.

Scenarios
The baseline scenario is a BAU scenario, assuming the continuation of current trends with no additional city-level mitigation policies.It uses the income per capita growth scenarios from the global IAM IMACLIM-R 74 .In particular, it uses the baseline scenario based on the 'middle of the road' Shared Socioeconomic Pathway 2 (SSP2), which is quite standard in the IAM community and corresponds to a central scenario.
For population, it uses the population growth scenarios of the United Nations 75 , which provide population growth projections from 2015 to 2035.Compared with the SSPs, the United Nations projections have the advantage of being available at the city level for cities of more than 300,000 inhabitants.In addition, population growth projections of the United Nations are broadly consistent with SSP2 at the global level in terms of population size for the first half of the twenty-first century 76 .Still, the SSP2 relies on underlying hypotheses in terms of age, sex and education that lead to differences in fertility rates, and in turn, in population growth scenarios compared with the United Nations scenarios: for instance, the SSP2 assumes lower fertility rates for Africa than the United Nations, leading to population size differences in the long run.
Finally, we assume that there is no change in public transport infrastructures and that the fuel consumption of new private cars decreases by 1.0% per year, in line with current trends for light-duty vehicles 77 , assuming a vehicle lifespan of 15 years 63 .More details about our BAU scenario are in Supplementary Section D.
We also run our four policy scenarios, designed to be representative of a wide spectrum of potential urban policies, yet simple enough to be generically applied to our sample of cities.We assume the same trends for population and income as in the BAU scenario, while transportation infrastructures and land-use constraints are impacted by the policies.Details about the policy scenarios are in Supplementary Section E.

Robustness checks
In Supplementary Section I, we carry out a robustness check of the results of this paper: we simulate an alternative version of each type of policy and check that the results remain qualitatively the same.
More precisely, we assume, starting in 2020, that the fuel tax increases fuel prices by 10% instead of 30%, that the fuel efficiency policy reduces the fuel consumption of new vehicles by 2% each year instead of 3.7% and that the restrictive land-use regulations policy prevents new constructions in areas with a density below 400 inhabitants per square kilometre in 2020, unless inhabitants will use public transport or active modes.For the BRT, instead of using OpenStreet-Map street network data, we assume more simply that two new public transport lines, north-south and east-west, are opened in each city.
Results are qualitatively the same as with the main policies' specifications.The combined four policies allow mitigation of transport emissions by 12.1% in 15 years, with large heterogeneity in policies' efficiencies between cities.However, city-specific policy portfolios that maximize emissions mitigation while increasing welfare allow for keeping most of the emissions mitigation (−11.6%) while increasing average welfare by 0.9%.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested
A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g.means) or other basic estimates (e.g.regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g.confidence intervals) For null hypothesis testing, the test statistic (e.g.F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable.
For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g.Cohen's d, Pearson's r), indicating how they were calculated Our web collection on statistics for biologists contains articles on many of the points above.

Software and code
Policy information about availability of computer code Data collection Our main dataset is based on the following data paper: Q. Lepetit, V. Viguié, C. Liotta (2023)."A gridded dataset on densities, real estate prices, transport, and land use inside 192 worldwide urban areas", Data in Brief, Volume 47, 108962, ISSN 2352-3409, https://doi.org/10.1016/j.dib.2023.108962.Data from this data paper are available here: https://zenodo.org/record/7086267.Data collection for this data paper has been made using R 3.4 and Python 3.7, and code are available here: https://github.com/CIRED/gridded_dataset_192_cities.

Data analysis
The analysis has been done using Python 3.9.Codes are available on github: https://github.com/CIRED/policy_portfolios.For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers.We strongly encourage code deposition in a community repository (e.Note that full information on the approval of the study protocol must also be provided in the manuscript.

Field-specific reporting
Please select the one below that is the best fit for your research.If you are not sure, read the appropriate sections before making your selection.

Life sciences
Behavioural

Fig. 1 |
Fig. 1 | Impact of the four policies on annual transport emissions and average welfare in the 120 cities in 2035, compared with the BAU scenario.Each dot represents a city.The boxplots represent the first quartile, the median and

Fig. 2 |
Fig. 2 | Decomposition of welfare variations between different drivers in the 120 cities.The figure represents the variation in 2035 compared with the BAU scenario.The boxplots represent the first quartile, the median and the third quartile among cities, and the whiskers correspond to 1.5 times interquartile

Fig. 3 |
Fig. 3 | Effect of the four policies on urban transport emissions and on welfare (including direct and indirect financial cost of the policies and health cobenefits) in 2035, compared with the BAU scenario.See also Supplementary Figs.11 and 12.
Corresponding author(s): Charlotte LIOTTA Last updated by author(s): 03/04/2023 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish.This form provides structure for consistency and transparency in reporting.For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Table 1 | Policies analysed (see Supplementary Section E for a detailed description) Policy name Description
45 (P value <0.05) with Nangini et al., 0.32 (P value <0.1) with Kona et al. and 0.53 (P value <0.01) with Moran et al.However, two elements question the relevance of the comparison.First, Moran et al. and Nangini et al. report only total emissions, whereas our estimates are for transport emissions.Second, the external data themselves are not consistent: comparing the databases on the cities they have in common, we find a correlation coefficient of 0.41 (P value of 0.000) between Nangini et al.'s and Moran et al.'s data, a correlation coefficient of 0.16 (P value of 0.573) between Nangini et al.'s and Kona et al.'s data, and a correlation coefficient of 0.04 (P value of 0.843) between Moran et al.'s and Kona et al.'s data.Indeed, the methodologies of the three databases differ: Nangini et al. and Kona et al. use a bottom-up approach, with cities reporting their emissions, whereas Moran et al. use a top-down approach, downscaling national or subnational emissions at the city scale; Nangini et al. report scope 1 emissions, Kona et al. report direct transport emissions and Moran et al. report scope 3 emissions.
& social sciences Ecological, evolutionary & environmental sciencesFor a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdfnatureportfolio| reporting summaryWe require information from authors about some types of materials, experimental systems and methods used in many studies.Here, indicate whether each material, system or method listed is relevant to your study.If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.