Public health air pollution impacts of pathway options to meet the 2050 UK Climate Change Act target: a modelling study

Article history

The research reported in this issue of the journal was funded by the PHR programme as project number 11/3005/13. The contractual start date was in April 2013. The final report began editorial review in January 2017 and was accepted for publication in June 2017. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The PHR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Melissa C Lott reports grants from BHP Billiton outside the submitted work. Mike Holland reports personal fees from the NHS during the conduct of the study.

Copyright statement

© Queen’s Printer and Controller of HMSO 2018. This work was produced by Williams et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2018 Queen’s Printer and Controller of HMSO

Aims of the project

It is clear that there has been little peer-reviewed research on the detail of the impact on British public health of climate change policies. Where work has been done, it is clear that the potential benefits to public health from reduced air pollution exposure are large and could represent the largest such benefits since the Clean Air Act of 195634 eliminated the notorious ‘smogs’ of the 1950s and 1960s. Equally, however, it is clear that if the wrong choices are made, these improvements could be forgone and the situation could potentially worsen. There is therefore a need for more detailed assessments than those currently available. Moreover, no study has yet attempted to identify an optimal scenario or pathway to achieve the 2050 climate change target while at the same time maximising the benefits for public health from air pollution exposures.

The overall objectives of the project are to quantify the air pollution impact on public health in GB from various pathways to achieve the 2050 target in the CCA of a reduction of 80% in CO₂-equivalent emissions relative to 1990, and to identify one or more pathways which will maximise the benefits to public health while achieving the 2050 target. The objectives in summary as originally proposed, taken verbatim from the original proposal, are as follows.

• To develop a series of policy pathways to achieve the CCA 2050 target, spanning a maximum health benefit scenario at one extreme and a scenario that rejects the 2050 target at the other. In between these we will identify up to three scenarios which deliver increasingly large benefits for public health.

• On the basis of these scenarios, to produce air pollution emissions estimates for the UK and Europe for 2050 in grid form across the modelling domain.

• To evaluate the air quality model for the current year and to establish the contribution to concentrations of PM, NO₂ and O₃ from sources within the UK and in Europe. These pollutants are the ones for which concentration–response functions (CRFs) are available to allow health impact assessments.

• To forecast air quality across the UK, including detailed modelling in urban areas for the future year and for the chosen policy scenarios.

• Calculate exposures at 9 × 9 km resolution for the UK and at 20 × 20 m resolution in major cities. Further calculate personal exposures using the hybrid time–activity model.

• To calculate the impact of the alternative future policies on mortality (and morbidity) using the life table approach. The monetary implications of the air quality impacts will also be provided.

• An assessment of the impact of the climate scenarios on health inequalities in the future years will also be undertaken in the UK.

• Optimal strategies for maximising the public health benefits of achieving the CCA target for 2050 will be produced.

During the course of the project some of these objectives had to be modified, principally to provide a more advanced form of modelling and projection than was foreseen at the beginning of the work. This adjustment of the work packages also resulted in an increasing amount of time taken for the work and a consequent extension of the deadline for producing a final report. Nonetheless, we believe that what we have now produced represents the most sophisticated modelling of air quality concentrations and health impact assessment ever carried out in GB. A flow diagram of the work is given in Figure 1.

FIGURE 1.

Flow chart of stages in the project. CMAQ, Community Multiscale Air Quality; UKTM, The UK Integrated MARKAL–EFOM System.

We had originally intended to produce our own future scenarios using the DECC ‘calculator’, a ‘do-it-yourself’ spreadsheet tool which allows the user to produce energy scenarios which meet the CCA target for CO₂-equivalent reductions. However, during the course of the project, after over a year had passed (during which time we were developing the air quality model), we made contact with the energy group at UCL, which, it transpired, had developed and was running the energy system model for the UK (UKTM), which is used by DECC (now subsumed within the Department of Business, Energy & Industrial Strategy). This presented an opportunity to produce far more robust and credible scenarios than we had previously envisaged and so we began to collaborate with the team at UCL, now co-author of the report. This, among other things, resulted in a certain delay to the project, but provided a higher quality of output. Furthermore, linking the UKTM model to an air quality model of the sophistication of the Community Multiscale Air Quality (CMAQ) model (the air quality model we used and which is described in Chapter 4) is a major step forward in modelling the impact of climate policies. This took a considerable amount of effort, but now for the first time in GB we have produced a system linking energy modelling with calculations of air pollution concentrations as opposed to the relatively easier step of linking to air pollution emissions, as several earlier studies had done.

The difference is considerable. The relative emissions from different sectors or scenarios are not necessarily directly proportional to their relative contributions to ambient pollution levels and hence population exposures and health impact. A simple example can illustrate this. A power station in a rural area with a chimney > 150 m in height will have a much smaller impact on health than a fleet of road vehicles emitting in total the same amount of pollution in an urban area. Previous analyses of air pollution impact arising from climate policies in the UK have calculated only emissions; for the first time we have produced calculations of concentrations airborne particles and direct public health impact. We have concentrated our analyses on GB rather than the UK because (1) the ultimate aim of the work was to calculate health impact from air pollution exposures and the relevant data on mortality rates were not as readily available at the required spatial resolution across the whole UK; and (2) configuring the air quality model to include Northern Ireland would have lengthened the project unduly.

Another aspect that has been somewhat time-consuming has been the development of the air quality model. We had hoped to model some cities at a fine spatial resolution, but as this is at the forefront of modelling research in the UK we were unsure of how practicable it would be. It turned out that we could run a number of major cities in GB at 20-m grid resolution using the supercomputer facility operated by the Natural Environment Research Council (NERC). We spent some time configuring the model to run on this system. The large benefits to the project came at the cost of time delays (users have to queue for run time on the NERC system). However, we have now produced the most spatially detailed concentrations and health impact ever calculated in GB.

The calculation of the impact of air pollution on health is based on numerical relationships between air pollution concentrations and the extent of adverse health outcomes such as mortality or morbidity. The relationships and indications of causality linking concentrations of PM_2.5 were well established at the commencement of the project, although they were confirmed and re-emphasised by a review carried out by the World Health Organization (WHO) in the early phase of the project. However, during the course of the project it became clear that evidence was building that NO₂ would need to be included in the health impact assessment, as evidence has accumulated for adverse effects of NO₂ independent of those of PM_2.5. Recommendations for the quantification of these impacts had been tentatively given by WHO, but during the project the advisory committee of the Department of Health and Social Care [Committee on the Medical Effects of Air Pollutants (COMEAP)] undertook to produce a report recommending an updated quantification of the impact on health from NO₂. Although some interim statements were published by COMEAP, the final agreed recommendation had not appeared before the current project had been completed; we have therefore used the interim recommendations. The inclusion of NO₂ in the health impact calculations thus represents a departure from the original protocol, but incorporates the most recent evidence on the adverse effects of NO₂ on health. There is some uncertainty over the size of the NO₂ contribution to the CRF compared with PM_2.5 and other pollutants. Owing to these uncertainties we have not suggested adding the impact for the different pollutants, nor do we consider that the NO₂ benefits cancel out the adverse health impact shown for PM_2.5, for which the health evidence is stronger.

The other pollutant for which health effect evidence is strongest is O₃, and we have quantified the impact of both long- and short-term exposures to this pollutant.

We have therefore calculated the health impact from exposures to the pollutants PM_2.5, NO₂ and O₃.

Following the calculations of health impact, we undertook an analysis of the different exposures to air pollution in populations of differing socioeconomic deprivation and ethnicity, using the Carstairs index of deprivation together with GB-wide maps of modelled air pollution concentrations.

We had originally hoped to calculate health impact using the time–activity model of exposure which we have developed at King’s College London, in addition to the normal procedure of using ‘static’ measures of exposure from fixed-point monitoring data or from modelled grid square concentrations. We had hoped that exposure–response relationships for health outcomes might become available from research elsewhere for these new type of exposures, but in the event these have not been forthcoming so we have not been able to apply this potentially more advanced exposure assessment in this project. Nonetheless, the model promises to provide a valuable advance in health impact calculations when more detail exposure–response relationships emerge.

Scenario development

The scenarios used in this research were developed to illustrate the possible pathways towards a decarbonised energy system in the UK through a technoeconomic lens. In this research project, the scenario outputs were then used to produce a detailed air pollution inventory and air quality analysis to help in better understanding the air quality implications of potential decarbonisation pathways.

A set of four scenarios was developed for this research project in order to assess the impact on air quality and public health in GB of possible future energy systems, as shown in Table 1 and Figure 3. Two of these scenarios (‘baseline’ and ‘reference’) were types of baseline or ‘business as usual’ scenarios, in which the 2050 target in the CCA was not achieved. The other two scenarios [low greenhouse gas (LGHG) and nuclear replacement option (NRPO)] achieved the 2050 decarbonisation target laid out in the CCA – including interim carbon budget targets – via two slightly different pathways. The difference between these two pathways lay in the assumptions made with regard to the construction of new nuclear power plants.

FIGURE 3.

Primary energy consumption in 2050 (PJ) by scenario.

With regard to decarbonisation, the baseline and reference scenarios did not include the UK’s 2050 decarbonisation goal or interim carbon budget targets. With the baseline scenario, no carbon price was included. Conversely, the reference scenario included a £30 per tonne carbon price that was linearly phased in from 2015 to 2030 and then held constant to 2050 in order to simulate a central case in which the system moves away from the most carbon-intensive technologies (e.g. coal in the electricity sector), but long-term decarbonisation goals are not achieved.

In the remaining two scenarios (i.e. LGHG and NRPO), the energy system was required to meet existing UK decarbonisation targets for a total reduction in greenhouse gas emissions of 80% by 2050 compared with 1990 levels as stated in the CCA. Furthermore, these scenarios required that all interim targets through the Fourth Carbon Budget be met. It is noted here that, in late July 2016 (i.e. during the course of this research project), the UK government set a Fifth Carbon Budget of 1725 million tonnes of CO₂ equivalents for the 2028–32 budgetary period, in agreement with recommendations from the CCC. The reduction trajectory used in this analysis is broadly consistent with this more recently agreed Fifth Carbon Budget. 42

Within these four scenarios, a key sensitivity was explored relating to the assumptions within UKTM for the nuclear power sector. The electric power sector has been responsible for most of the energy sector’s decarbonisation to date, and its future trajectory is of particular interest because of this role as the ‘first mover’. Furthermore, specifically in the context of this research, it was particularly important to understand the sensitivities surrounding nuclear power plant build as nuclear energy is an air pollution emissions-free source of power, whereas the alternatives (e.g. biomass, natural gas, coal with or without carbon capture and storage) are not. Therefore, its inclusion in (or exclusion from) the future energy technology pathway could have a potentially significant impact on the air pollution levels within the UK.

All of these scenarios examined the period to 2050 for the UK using demand drivers that relied on official population and economic growth projections and energy efficiency expectations. Primary energy consumption in 2050 by fuel type is displayed in Figure 3 for all scenarios. Also included in Figure 3 is the 2011 primary energy consumption for the UK for the sake of comparison.

Each of these scenarios was analysed using the CMAQ air quality model discussed elsewhere in this report for the year 2050 (i.e. the year the decarbonisation target is met) (see Chapter 4). Furthermore, as the UKTM output showed that domestic wood burning peaked in 2035 in the LGHG scenario, an additional air pollution emissions inventory and model run was undertaken for this interim year to understand its air quality and public health impact.

In carrying out the air quality and health impact modelling, as described in Chapters 5 and 6 of this report, we confined the analysis to GB rather than the whole of the UK, as noted in Chapter 1, even though the energy use and fuel consumption results from UKTM are presented for the UK in this section. Because of the uncertainty surrounding a future carbon price, and the fact that the emissions were not significantly different from those in the baseline scenario, we chose to restrict the air pollution concentration and health impact modelling to the baseline scenario. The wood burning increase and peak in 2035 is common to both the LGHG and the NRPO scenarios, and the overall emissions in these two scenarios are very similar, so we modelled air quality and health impact in 2035 for the NRPO scenario along with the baseline scenario for comparison. Together with the modelling of a scenario that held ‘current’ (2011) concentrations constant through to 2050, we therefore analysed a total of four scenarios in 2050 with further analyses of the baseline and NRPO scenarios in 2035.

Current emissions in 2011: an improved inventory

The emissions used to model current year (2011) concentrations were derived from ECLIPSE v5a for non-UK emissions, whereas the NAEI50 and the London Atmospheric Emissions Inventory,51 and, for GB on-road traffic, used sources from the King’s College London UK emissions model. 52 ECLIPSE is a collaborative project funded under the European Union (EU) Seventh Framework Programme [(ENV.2011.1.1.2-2) climate forcing of non-United Nations Framework Convention on Climate Change gases, aerosols and black carbon]. The emission data are available at www.iiasa.ac.at/web/home/research/researchPrograms/air/Global_emissions.html (accessed 6 February 2018).

The emissions were summarised into 11 Selected Nomenclature for sources of Air Pollution Prototype (SNAP) sectors (Table 2) and were processed into gridded chemical species using speciation and temporal profiles developed in the US–EU project, Air Quality Model Evaluation International Initiative (URL: http://aqmeii-eu.wikidot.com/; accessed 6 February 2018). An additional two SNAP sectors, domestic wood burning and cooking emissions, were added, as well as SNAP 7 (road transport), intermediate volatile organic compounds (IVOCs) emissions. The temporal profiles for cooking were derived from Ots et al. 53 Biogenic emissions were estimated using the biogenic emission inventory system (version 3) model in Sparse Matrix Operator Kernel Emissions (SMOKE) v2.6. This calculation accounts for U.S. Geological Survey land cover data (URL: http://landcover.usgs.gov/; accessed 6 February 2018), high-resolution MODerate resolution Imaging Spectroradiometer (MODIS)-based land use54 and meteorological data from the weather research and forecasting (WRF) model.

The input emissions data were provided at a range of spatial scales, from 50-km grids across non-UK Europe (ECLIPSE), down to 1-km grids in GB and traffic counts on individual roads. The link between each emissions source and the CMAQ model domains was determined by reprojecting all of the emissions data into the Lambert conformal conic projection and then intersecting each emissions database and model domain in QGIS v2.8.2-Wein.

Large industrial plant emissions from ≈5000 sources in GB were taken from the NAEI database and included major power stations, iron and steel smelting and oil refineries. Details of stack heights, stack diameters, volume flow rates and release temperatures were used by SMOKE to account for the effects of plume rise. Total emissions of PM₁₀, PM_2.5 and NO_x are shown in Figure 4.

FIGURE 4.

Emissions of PM₁₀, PM_2.5 and NO_x in the base year (2011). (a) Emissions of PM₁₀; (b) emissions of PM_2.5; and (c) emissions of NO_X.

Linking the UK Integrated MARKAL–EFOM System output to the air quality model (CMAQ)

We then built on the current year emission inventory to produce analogous inventories for the future scenarios produced by the UKTM model. The UKTM model yields energy and fuel use by a large number of user sectors, whereas the CMAQ model uses as input emissions split by just 13 SNAP sectors (see Glossary). To limit the risk of losing information available from UKTM, we converted the UKTM output via an intermediate step using the large number of nomenclature for reporting (NFR) sectors (see Glossary) and then aggregating the resulting NFR emissions back up to the 13 SNAP sectors used by CMAQ.

The soft linking of UKTM to the UK NAEI and King’s College London road traffic, IVOC and biomass (wood burning) emissions inventories allowed future emissions to be calculated separately for each emissions sector and for each pollutant in the following way:

where i = separate energy/emissions sectors, j = 2035 or 2050 and k = NO_X, CO, PM₁₀, PM_2.5 and SO₂.

To do this, the emissions sources within the NAEI have been linked to the relevant UKTM output. This was done by assigning each of the 370 NFR codes and associated fuel types to a SNAP code (1–1311) and one of the 129 UKTM energy sectors and fuel type predictions. This resulted in a total of 570 separate source categories, which have been scaled using UKTM to obtain 2035 and 2050 UK emissions totals for NO_X, CO, PM₁₀, PM_2.5 and SO₂ (note that these emission totals are UK totals rather than GB totals). Note also that hydrochloric acid was not forecast forwards and remained at the 2011 level for all scenarios and ammonia (NH₃), VOC and IVOC emissions were estimated separately as described below.

After calculating future emissions by NFR sector, the total emissions were split by year and pollutant and summed, giving a ratio of emissions between the base year (2011) and future years (2035/50) for each pollutant and SNAP sector combination. The factor was then applied to the NAEI 2011 1 × 1 km emissions in GB, scaling them uniformly across GB and using the same scaling factors for GB point source emissions provided separately by the NAEI.

As a final step, the scaled 1 × 1 km emissions were converted into a format for each air pollution model run: a GB-wide 50 × 50 km emissions file, a GB 10 × 10 km emissions file and a 2 × 2 GB cities emissions file. Each of these files gives emissions (tonnes/annum) for eight pollutants and 13 SNAP sectors across regular GB spatial grids and in a Lambert conic conformal projection used by the CMAQ model.

For the emissions associated with road transport (SNAP 7), the UKTM estimates billion vehicle kilometres (Bvehkm) for each of seven vehicle types: diesel, electric, petrol, hybrid electric, hydrogen, liquefied petroleum gas and compressed natural gas. The estimates of Bvehkm have been used, in conjunction with the King’s College London’s bottom-up road transport emissions inventory,52 based on, inter alia, the King’s College London’s remote sensing measurements of vehicle emissions,63 to calculate new vehicle kilometres in each of the future years. As the increase in GB vehicle kilometres is large and the capacity of city roads is unlikely to allow such significant expansion, a two-step approach has been taken, whereby increases in larger cities such as London, Manchester and Leeds, etc., are assumed to be smaller than those elsewhere. The increase in cities was limited to those predicted in London and summarised in Table 3. Elsewhere in GB, the increase in vehicles was greater than in urban areas to maintain the increased GB Bvehkm predicted by UKTM (see Table 3). The increase in vehicle kilometres in urban areas was associated with a decrease in speed, with an elasticity of 1 meaning that a 10% increase in vehicle kilometres results in a 10% decrease in speed. Finally, vehicle stock estimates of the proportion of Euro 5/V and 6/VI plateau in the GB model from 2035 onwards, because there are, at present, no plans to go beyond Euro 6/VI vehicles. However, the type of vehicles changed markedly during this period, predicted by UKTM. A complete list of the vehicle kilometres forecasts from the UKTM model is available from the authors.

Emissions of two pollutants, NH₃ and VOCs, are not well predicted using the UKTM, as fuel use from sources emitting these pollutants is a small part of the sources’ characteristics. The majority of the emission of these pollutants arises from non-combustion processes such as solvent evaporation, in the case of VOCs, and animal excretions and fertiliser use in the case of NH₃. Therefore, as an addition to the UKTM estimates, we used NH₃ and VOC emissions projections between 2013 and 2050 from the GAINS model – ECLIPSE version 5a, and described in the Thematic Strategy on Air Pollution report #16,64 to scale the existing NAEI emissions for NH₃ and VOC. Version 5a was chosen as the forecast both is recent and includes all of the emissions reductions agreed under the member states National Emissions Ceiling Directive (NECD).

Future diesel IVOC, cooking and domestic wood burning emissions inventories were also forecast differently to those of the other NAEI sources. Diesel IVOCs, as the name suggests, are diesel vehicle-related VOCs with a higher carbon number (C₁₀+). These higher carbon number VOCs are not routinely measured or considered within current vehicle emissions regulations and we have therefore assumed that vehicle manufacturers are not controlling them effectively. Hence, any future emissions estimate has been scaled to total diesel vehicle kilometres provided by the UKTM model.

A specific domestic wood burning emissions inventory has been developed for GB and is described in Domestic wood burning and cooking emissions in Great Britain. Future wood burning emissions are forecast using the UKTM model energy use estimates in combination with a change in PM₁₀ and PM_2.5 emissions factors, which changed from 8.2 g/kg and 7.7 g/kg, respectively, to 6.6 g/kg and 6.2 g/kg, respectively, a drop overall of approximately 20%. The latter emissions factor change is based on a weighted average emissions factor combining existing fireplaces, new stoves and highly efficient palletised wood burners. Finally, cooking emissions were scaled by the growth in population and taken from the Office for National Statistics (ONS) (URL: www.ons.gov.uk/ons/rel/pop-estimate/population-estimates-for-uk--england-and-wales--scotland-and-northern-ireland/mid-2014/sty---overview-of-the-uk-population.html; accessed 6 February 2018).

In the future scenarios we assumed that the following large power plants were closed as a result of their becoming uneconomic following the need to meet new emission limits: Kingsnorth, Ironbridge, Grain, Ferrybridge C, Tilbury B, Didcot A, Littlebrook D, Fawley, Cockenzie and Kingsnorth oil treatment plants.

Great Britain pollutant emissions in 2011 and future years

For clarity, and because these are the pollutants for which the health effects evidence is strongest, in this section we discuss only the emissions of PM₁₀, PM_2.5 and NO_x; emissions of other pollutants are available on request but are omitted here.

Emissions of PM₁₀, PM_2.5 and NO_x in the 2011 ‘base’ year, the baseline scenario in 2035 and 2050, the NRPO for 2035 and 2050 and the LGHG scenario for 2050 are shown in Table 4 as GB totals and in Table 5 for NO_x and PM_2.5 disaggregated by SNAP sector.

Turning to the baseline scenario, the fundamental premise was that there would be no further carbon mitigation beyond those measures already in place. It should be noted that the CCA target was not included as a constraint and, consequently, the 80% emission reduction target in the CCA is not achieved: the overall reduction in CO₂ equivalents in 2050 is 39%. Nuclear power generation is phased out and there is no carbon price to influence energy options or emissions. The output from UKTM in 2050 shows a decrease in primary energy consumption from 9355 PJ in 2010 to 7091 PJ in 2050. A significant part of this decrease is because of the decline in nuclear power generation from 562 PJ in 2010 to 73 PJ in 2050. Biomass (wood) use shows a modest increase from 288 PJ to 307 PJ, whereas natural gas use decreases only by a small amount, from 3748 PJ to 3515 PJ, and oil use shows a larger decrease, from 3349 PJ to 2653 PJ.

The scenario includes large increases in energy use for combined heat and power (CHP) in the industrial and residential sectors, fuelled by biomass and natural gas. In the industrial and residential sectors, the use of biomass for CHP increases from 8.17 PJ and zero, respectively, in 2010 to 148.17 PJ and 21.76 PJ in 2050. Natural gas consumption for CHP in the industrial sector shows a modest increase from 143.9 PJ in 2010 to 157.76 PJ in 2050, whereas the residential sector shows a large relative increase from 3.5 PJ in 2010 to 77.69 PJ in 2050. These changes contribute to the fact that NO_x emissions in 2050 are higher in the baseline scenario than in the other scenarios.

The LGHG scenario, by contrast, shows a large increase in biomass consumption in 2035 (1089 PJ, compared with 289 PJ in 2010), before decreasing in 2050 to 932 PJ, still about three times higher than in the base year. This is the main reason for the large increase in PM_2.5 and PM₁₀ emissions in 2035 shown in Tables 4 and 5. By 2050 biomass use is still above that of the base year, but the increase in PM emissions from this source is now offset by reductions in other sectors, notably transport, as the majority of the road vehicle fleet is projected to be electric. However, the electrification of the vehicle fleet will reduce only exhaust emissions of PM₁₀ and PM_2.5; non-exhaust emissions from tyre, brake and clutch wear will not fall unless action is taken to reduce emissions from this source. It is possible that technologies such as regenerative braking on electric vehicles may act to mitigate these non-exhaust emissions, but there is little evidence at present to allow us to quantify this, so we have assumed that emission factors stay at present levels out to 2050. In the LGHG scenario, vehicle activity is projected to increase substantially from 2010 to 2050, by roughly 50% for cars and heavy goods vehicles and by roughly a factor of 2 for vans. With constant non-exhaust emission factors this represents a substantial relative increase in non-exhaust emissions and hence concentrations from this source. The health implications of this are discussed in Chapter 5. The LGHG scenario has the largest contribution of nuclear power to the energy mix in 2050, at 2454 PJ, but the air pollutant emission benefits of this are offset by the relatively large contribution from biomass.

The nuclear replacement scenario seeks to limit nuclear power station build in that existing nuclear stations will be replaced but no additional new ones will be built. The contribution of nuclear power to the energy mix in 2050 is therefore much less than in the LGHG and reference scenarios and is similar to the contribution in PJ currently. The biomass contribution is also relatively high and similar to that in the LGHG scenario, and both the LGHG and the nuclear replacement scenarios show a large peak in biomass use in 2035, at 1089 PJ and 1126 PJ, respectively. This suggests that both scenarios will have potentially detrimental effects on air quality and public health, and this will be explored in Chapters 5 and 6 of this report.

Non-Great Britain emissions

As well as GB emissions, we need to take into account the contribution of sources in the rest of Europe. For emissions of pollutants from these countries we have used data from the ECLIPSE project. We used version 5a, which dates from July 2015 and incorporates current agreed legislation (the ‘CLE’ scenario) on air quality and climate change in the EU, currently agreed together with emission projections carried out by IIASA following discussions with the individual member states of the European Union during the revision of the NECD. Emission totals for non-UK Europe are shown in Table 6.

As it is impossible to predict, with any degree of confidence, future emission reduction agreements between countries in the EU and beyond, we made no attempt to model any more ambitious scenarios for which projections were made in the ECLIPSE project. The non-UK emission projections may therefore be pessimistic; the modelling undertaken in this project was largely carried out while the NECD was being negotiated and, at the time of writing, the final position had still to be formally agreed. The revised NECD [discussed in Linking the UK Integrated MARKAL–EFOM System output to the air quality model (CMAQ)] and further measures in countries following the United Nations Framework Convention on Climate Change Paris Agreement on climate change may well lead to further reductions than it has been possible to include here, so air quality in GB may improve more than we have currently projected. This could be the subject of further research using the models developed here.

These non-UK emissions will have the biggest influence on GB concentrations of PM_2.5, PM₁₀ and O₃. For NO₂, GB emissions will be dominant and the policies and measures enacted in response to air quality legislation and the measures put in place to meet the CCA targets will largely determine concentrations of NO₂.

Of the precursors of PM_2.5, the reductions in NO_x are the largest (≈29%), with SO₂ reductions of ≈16% by 2050. The other important precursor, NH₃, is projected to increase by a small amount. This continues the poor performance regarding NH₃ reductions in Europe, which have already lagged significantly behind those of other pollutants over the past few decades. Emissions of VOCs will be important for peak concentrations of O₃ in photochemical episodes, and total European emissions are projected to decrease by ≈22% out to 2050.

However, the non-UK emissions in Table 6 include emissions from Eastern Europe, countries some distance from GB with relatively small influence on GB concentrations of PM_2.5 and O₃. For PM_2.5 and O₃, emissions of NO_x, SO₂, VOCs and NH₃ in GB’s nearest neighbours (France, Germany, Belgium, the Netherlands and Poland) will be the most important, as will those from shipping,65 and these are shown in Table 7.

Several features are apparent from Table 7. First, the reductions in PM_2.5 and O₃ precursors are greater in GB’s near neighbours than in the whole of Europe (including Eastern Europe). Second, although shipping is projected to make significant reductions in emissions out to 2050, this source remains large and by 2050 is projected to be larger than the land-based emissions from GB’s neighbours. The third point of importance is the very small projected reduction in NH₃ emissions, which will limit the reductions in PM_2.5 concentrations in GB in future years. As noted elsewhere in this report, the lack of large reductions in NH₃ emissions has represented a major challenge to most European countries over the past few decades. The recently agreed NECD, however, does suggest that further action should be forthcoming: the agreement on NH₃ is for a 19% reduction by 2030 on a 2005 base for the 28 EU member countries as a whole and reductions of 13–29% for GB’s near neighbours. This figure is not directly comparable to the figures in Table 7 and, as the NECD was only very recently agreed, time has not permitted a full analysis. However, the commitments in the new directive suggest that the NH₃ emission reductions used here may be pessimistic. The corresponding reductions in NO_x, SO₂ and primary PM_2.5 range from 39% to 69%, from 53% to 77% and from 39% to 58%, respectively. Further work would be required to compare these commitments with the projections we have used based on the ECLIPSE project. As noted above, there should be a degree of consistency between the two as the ECLIPSE project involved detailed discussions with EU member states, but clearly the two will not necessarily match.

The CMAQ volatile basis set scheme

Modelling the concentrations and composition of atmospheric particles is a complex task and the science is continuallly developing. An important part of improving the model’s predictive capability for PM_2.5 and PM₁₀ was the addition of a number of emissions sources for the UK and Europe, including diesel IVOCs and organic aerosol (OA) from cooking, based on aerosol mass spectrometry measurements, and a large increase in domestic wood-burning emissions. Another important improvement in the chemical mechanism in CMAQ was the use of the VBS scheme84,85 to account for the chemical ageing and volatility of organic gases and aerosols. OAs comprise POA emitted from sources such as motor vehicles and SOA formed in the atmosphere from gas phase precursors. Briefly, a proportion of POA evaporates on dilution to ambient conditions and then photo-oxidises to give SOA. 86

Gas-to-particle partitioning of POA emissions depends on its volatility distribution and has been traditionally assumed to be non-volatile. However, Robinson et al. 86 found that additional diesel IVOCs were emitted, contributing between 1.5 and 3 times the POA emission. The VBS scheme aims to represent this more realistic OA behaviour by representing emissions in volatility bins, ranging from those that are non-volatile at ambient conditions to semivolatile organic compounds (SVOCs) (i.e. that partition between particle and gas phase, and IVOCs that start in the gas phase). The VBS also includes oxidation or ageing processes, which, starting from SVOC and IVOC, produce compounds of lower vapour pressure and generate substantial amounts of SOA over time scales from hours to days. Use of VBS increases the amount of regional SOA in urban areas, a different split between POA and SOA, although not necessarily higher OA levels than assuming non-volatile POA. 86

CMAQ-Urban outputs

The CMAQ model was run for GB at 10 km, and in major cities at 2-km and 20-m, resolution with all runs nested within the larger European domain. The CMAQ predictions of hourly air pollution at 2 × 2 km were downscaled to a fixed 20 × 20 m grid using the bilinear interpolation method described by Shaoa et al. 91 The CMAQ hourly concentrations are smoothly varying in space for each hour, and this limits the error introduced through the interpolation step and allows a direct match with the ADMS model outputs. The ADMS model predicts concentrations of primary pollutants, which are then added to the CMAQ background concentrations and for NO, NO₂, and O₃ close to roads, using a simple chemistry scheme applied for 2011/12. 82

The coupling of the local-scale model with the outputs of CMAQ at a 2-km scale results in a double-counting error as a result of local road emissions being counted in the CMAQ background concentrations. This has been resolved at model run time in the following way. Using the contribution to air pollution from each road separately, the equivalent road emissions are mixed into the 2 × 2 km CMAQ grid to estimate the double counting concentration for each hour. This is repeated for each road in turn as the model run is under way and allows a correction of the background CMAQ concentration prior to the inclusion of the local road contribution and the application of the chemistry scheme.

We ran CMAQ at 2-km resolution in urban areas because the finer the grid size the better the model is at identifying pollution ‘hot spots’ which are important for health impact calculations. As discussed in Chapter 5, although we report concentrations at 20 m resolution in the major British cities, we were unable to use these results for health impact calculations as the relevant health data are not available at such fine spatial resolution.

CMAQ-Urban model evaluation in 2011/12

The purpose of model evaluation is to establish the suitability of the CMAQ-Urban model for predicting present-day concentrations of NO₂, PM_2.5, PM₁₀ and O₃, without the need for bias correction using measurements. This is important to establish as models that require correction through assimilation with ambient data have the potential to introduce bias into any future predictions and are therefore less well suited for policy development.

There have been two major development areas required in establishing suitable model performance for predicting PM_2.5, PM₁₀, O₃ and NO₂: (1) the adoption of more sophisticated modelling of OAs, including the generation of secondary species, using the now well-established VBS scheme in CMAQ; and (2) the development of emissions not currently available in GB, including diesel-generated IVOCs, organic cooking emissions (based on aerosol mass spectrometer measurements) and an update of the GB biomass burning emissions inventory (based on the most recent surveys of wood use and emissions factors from the UK and Europe). Furthermore, as NO₂ is a problem pollutant close to roads in GB, and NO_X and NO₂ emissions are important for predicting O₃, improvements have been made to vehicle emissions factors using UK remote sensing data within the King’s College London UK NO_X and NO₂ traffic emissions model. Finally, it is expected that non-exhaust PM emissions are likely to dominate over exhaust emissions in the future and so the King’s College London non-exhaust emissions factors have also been used in the traffic emissions model.

The evaluation presented here includes an assessment against measurements in 2012 using the CMAQ model results at a 10 × 10 km spatial scale. This scale is commensurate with the PM_2.5/PM₁₀ and O₃ outputs from the future climate scenarios considered. However, as NO₂ and non-exhaust PM₁₀ is a much more localised and traffic-related problem, a further evaluation has been made against ambient measurements using the CMAQ-Urban model at a 20-m spatial scale, including at roadside and kerbside sites. This level of spatial resolution represents the most detailed modelling of GB and its major urban areas ever carried out.

Performance of CMAQ-Urban for annual means

In this section we discuss the comparison between 10-km annual average predictions of PM_2.5, PM₁₀ and O₃ against GB ambient measurements. We use the 10-km resolution here as the concentrations of these pollutants show less spatial variability than NO₂. We report the 20-m urban resolution model results for NO₂ in Chapter 6. To understand the annual performance of the CMAQ-Urban model predictions, an evaluation process has been undertaken against fixed-site measurements from the Department of the Environment, Food & Rural Affairs’ Automatic Urban and Rural Network and from the London Air Quality Network, as shown in Figure 6. This has been based on our experience of model evaluation from both phases of the UK model intercomparison exercise. 92 The focus of the analysis has been on the comparison between 10-km predictions at rural, suburban and urban background sites, as the spatial scale of pollutants close to sources, such as at roadside and kerbside sites, renders them unsuitable for comparison, with a separate comparison (NO₂ and PM₁₀) close to roads.

FIGURE 6.

Automatic Urban and Rural Network and London Air Quality Network measurement sites (green circles) used in the model evaluations; 10-km grid emissions (coloured), over which model concentrations are predicted; and the road network (black lines) used for the CMAQ-Urban model.

The results in Figures 7 and 8 show that the model predicts background concentrations well across GB and, in concert with the performance statistics in Table 8, has very little overall bias (normalised mean bias –2%) and a good correlation coefficient of r = 0.8. Concentrations at some monitoring sites are overpredicted, such as rural Rochester and Glazebury and suburban London Eltham, whereas others such as urban background Fort William/Aberdeen, Plymouth and Edinburgh are underpredicted. It is important not to overinterpret these data as the model predicts averages over 10-km grid squares compared with measurements at a single point and this leads to the possibility of increased variance in the site-by-site comparison. The small NO₂ bias is despite a –23% underprediction of NO_X which is associated with the underprediction of emissions from road transport (this has been well documented in the recent literature63). The consequence of this is that O₃ is overpredicted and in this case by the same proportion (+23%), although not in absolute terms (–8 µg/m³ for NO_X and +10 µg/m³ for O₃). However, as NO_X, NO₂ and O₃ are chemically linked, this suggests that the model’s chemistry is working well. The correlation coefficient of the model for O₃ (r = 0.82) also points to good model performance. Other important determinants of the O₃ in the CMAQ model include model boundary conditions. In this run we have used MOZART-4, which provides good O₃ performance, but still contributes to model uncertainty and positive bias.

FIGURE 7.

Scatterplots of 2012 annual average modelled vs. measured NO₂ concentrations (µg/m³) at GB sites.

FIGURE 8.

Scatterplots of 2012 annual average modelled vs. measured O₃ concentrations (µg/m³) at GB sites.

The model also performs well for PM_2.5 (Figure 9), with good agreement across a range of concentrations from 5 to 17 µg/m³. Correlation coefficients are lower than for NO₂ and O₃, at r = 0.72, with a small bias of –9% (which again represents good model performance). Some sites are underpredicted, however, including Aberdeen, Leeds and Sheffield centre and Eastbourne, and, although care is needed not to overinterpret the results, all are at urban background locations and may reflect the need for more detailed spatial modelling to correctly represent local sources. At first sight, PM₁₀ is more problematic as there is a reduction in the value of r (0.31), albeit still with only a small positive bias overall of 3% (Figure 10). Four sites in particular are grossly overpredicted (including a rural site); these are Newport, Hull Freetown, Rochester Stoke and Portsmouth. As Rochester also features in the PM_2.5 plot and is underpredicted, this suggests an overprediction of coarse PM (PM₁₀–PM_2.5) and, as all four of the sites are coastal, points to the overpredictions of sea salt. Analysis of the contribution of sea salt at the four locations shows that Rochester is highest of all sites at 13.7 µg/m³, Portsmouth is third highest (13.3 µg/m³), Hull is fifth highest (11.2 µg/m³) and Newport is 14th (8.3 µg/m³). The influence of sea salt on the model’s performance is reinforced by removing the modelled sea salt estimate from the model forecast and a typical value of 2.5 µg/m³ from the observation, and reanalysing the results at the 31 PM₁₀ sites (see PM₁₀ SS in Table 8). The scatter reduces (not shown) and the grossly overpredicted sites fall much closer to the 1 : 1 line, which results in an r-value of 0.63 and a bias of –10%, suggesting that predictions of anthropogenic PM₁₀ sources are well represented in the model. Finally, the fact that approximately 23 of the British Automatic Urban and Rural Network sites are close to the coast, yet only four showed significant influence from high predictions of sea salt, suggests that this is a local coastal phenomenon and of less importance for the majority of GB. On closer inspection, the four sites are a number of kilometres from the coast but within a 10-km square that has a significant ‘surf zone’ within it. Again, more spatially detailed modelling around the British coasts alongside coastal PM measurements would better establish the representation of this important source of PM₁₀ in the CMAQ model.

FIGURE 9.

Scatterplots of 2012 annual average modelled vs. measured PM_2.5 concentrations (µg/m³) at GB sites.

FIGURE 10.

Scatterplots of 2012 annual average modelled vs. measured PM₁₀ concentrations (µg/m³) at GB sites.

Comparison between 20-m annual average calculations of NO₂ against Great Britain ambient measurements

Although dominated by sites in London, we have evaluated the CMAQ-Urban model at 20-m resolution against observations at 76 sites throughout GB in 2011; the results are summarised in Table 9 and Figures 11 and 12. In common with predictions at the background sites described above, a degree of underestimation of NO_X within the emissions inventories and especially from road transport leads to underprediction of both NO_X and NO₂, although the NO₂ mean bias is small at –5.2 µg/m³ or –11%. For both NO₂ and NO_X the correlation coefficient shows the model to be working well close to road sources in urban areas giving values of 0.81 and 0.75, respectively.

FIGURE 11.

Scatterplots of 2011 annual average modelled at 20 m vs. measured NO₂ concentrations (µg/m³) at GB site.

FIGURE 12.

Scatterplots of 2011 annual average modelled at 20 m vs. measured PM₁₀ (µg/m³) at GB sites.

The scatterplot in Figure 11 demonstrates the close agreement between the model and observations at lower rural concentrations and an increase in divergence at higher concentrations, in which road traffic is a more important emissions source and the lack of NO_X within the inventory is more influential.

Comparison of 73 sites from rural through to kerbside locations shows that the prediction of PM₁₀ concentrations has a consistent bias of –5.2 µg/m³ or –20%, suggesting that a spatially consistent source of PM₁₀ was missing from the model. As there is a consistent bias, the correlation coefficient between model predictions and observations is in reasonably good agreement, with an r-value of 0.8.

Hourly time series of air pollution data in Great Britain in 2012

It is also important for the model to capture the seasonal changes in air pollution concentrations as well as high air pollution ‘episodes’ during the year. To demonstrate CMAQ-Urban’s capability, we have plotted the hourly NO₂ and O₃ concentrations from the 10-km model (‘modelled’ time series in Figures 13 and 14) and measurements (‘observed’ time series in Figures 13 and 14) averaged across all GB sites, together with the composite frequency distribution of hourly values.

FIGURE 13.

Time series of 2012 hourly average modelled vs. measured (observed) NO₂ concentrations (µg/m³) at GB sites. Mod, modelled; obs, observed.

FIGURE 14.

Time series of 2012 hourly average modelled vs. measured (observed) O₃ concentrations (µg/m³) at GB sites. Mod, modelled; obs, observed.

The seasonal variation of average model and measured NO₂ concentrations show typical maxima during autumn and winter months and a spring/summer minimum. Most of the high episodes of NO₂ are present in both time series, although they sometimes differ in magnitude (e.g. the modelled episodes in October/November are larger in magnitude than in the observed trace). Furthermore, the variability (peak to trough) of NO₂ concentrations in summer is higher in the model than in the measurements. Overall, however, the hourly model results perform very well against the observations at the GB sites.

Similarly, the average modelled and measured hourly O₃ concentrations show a similar seasonal profile, with a spring maximum at the end of April and autumn/winter minimum (see Figure 14), albeit with far more variability in concentrations between the largest and smallest values during this period. It is also notable that peak O₃ concentrations in early January are almost as high as during the peak O₃ season of April to June. Overall, the minimum–maximum range in concentrations is similar, although the modelled mean values are approximately 10 µg/m³ higher than from observations. This may also be a result of the underestimation of NO_x emissions. The modelled frequency distribution is slightly skewed to higher concentrations, whereas the observed distribution is skewed towards lower concentrations, suggesting that the model overestimates lower O₃ concentrations because titration of O₃ with NO is reduced.

Finally, the measured and modelled average hourly time series of both PM_2.5 and PM₁₀ (Figures 15 and 16) show very good agreement throughout the year with less of a seasonal variation evident than for NO₂ and O₃, but a summer minimum and autumn and winter maximum, by virtue of the difference in frequency and magnitude of air pollution episodes during these periods. The mean difference in PM_2.5 and PM₁₀ is also small (approximately 1–2 µg/m³).

FIGURE 15.

Time series of 2012 hourly average modelled vs. measured (observed) PM_2.5 concentrations (µg/m³) at GB sites. Mod, modelled; obs, observed.

FIGURE 16.

Time series of 2012 hourly average modelled vs. measured (observed) PM₁₀ concentrations (µg/m³) at GB sites. Mod, modelled; obs, observed.

Conclusions

Model evaluation at rural, suburban and urban background sites in 2012 and from sites at locations from rural to kerbside in 2011 has demonstrated that the CMAQ-Urban model is capable of producing robust predictions of air pollution and, importantly, that this can be undertaken without the need for a ‘calibration’ of the model against measurements. This is important as models that have been ‘calibrated’ risk introducing bias into future predictions and under different policy scenarios as they assume that the same correction factor applies both now and into the future. Here we have made a determined effort to develop the model by incorporating instead new methods for predicting PM_2.5/10 using the VBS scheme and a comprehensive set of new or revised emissions from diesel car IVOCs, cooking and biomass burning, as well as improved NO_X and NO₂ from road transport. Despite making good progress with PM predictions, which have hitherto been underpredicted, and often by a large margin, there remains room for improvement in some areas, including, but not limited to, NO_X emissions from road traffic, the seasonal representation of sources in the model and the suitability of chemical boundary conditions. Nonetheless, the model is capable of producing robust projections of future air quality on which to base assessments of future policy scenarios.

Air quality: nitrogen dioxide

Absolute values of the annual mean concentrations of NO₂ in the various scenarios are shown as box-and-whisker plots in Figure 17, and the changes relative to the base year are shown as maps in Figure 18.

FIGURE 17.

Box-and-whisker plots for annual mean GB NO₂ concentrations (10-km resolution).

FIGURE 18.

Annual mean NO₂ concentrations in the various scenarios. (a) 2011; (b) 2035 baseline; (c) 2050 baseline; (d) 2035 NRPO; (e) 2050 NRPO; and (f) 2050 LGHG.

Almost all the scenarios show significant decreases in annual mean NO₂ concentrations, largely as a result of the electrification of large parts of the road transport sector. The one exception is the baseline scenario. Here, there was no obligation to meet the CCA target, nor was there any carbon price imposed in the model. This resulted in a higher consumption of natural gas (notably in the industrial sector) and oil products (in the transport sector) in 2050 (compared with the LGHG scenario), leading to a much smaller reduction in NO_X concentrations in 2050 than in the other scenarios. The decreases in all scenarios are largely a consequence of the electrification of the vehicle fleet: the LGHG and the NRPO scenarios have very similar levels of car fleet electrification, but the NRPO scenario has more gas consumption in stationary sources leading to smaller reductions in NO₂ overall compared with the LGHG scenario. The LGHG scenario shows the largest decrease in NO₂ concentrations in 2050, with the highest level of nuclear energy and the smallest use of oil and gas in meeting the CCA target, although by 2050 the reductions in the NRPO scenario are similar. Annual mean concentrations of NO₂ in these scenarios are projected to decrease by more than ≈50% in the highest 25% of 10-km grid squares in GB. The scenario, which does not meet the CCA target, shows the smallest decreases in NO₂. These are important conclusions given the current interest in NO₂ levels and the legal limits in the legislation.

The hourly results for all GB sites for NO₂ are shown in Figure 19. The effect of the electrification in the road transport sector is clearly apparent in the frequency distributions.

FIGURE 19.

Hourly NO₂ concentrations at all GB sites in the different scenarios.

Even the baseline scenario, which does not meet the CCA target, shows a large decrease in hourly peak concentrations, but the NRPO and LGHG scenarios show even larger reductions.

Figures 20 and 21 show the time series for the base year 2011 and the NRPO showing even more clearly the reduction in hourly concentrations of NO₂ in the future as the road transport fleet becomes electrified.

FIGURE 20.

Hourly concentrations of NO₂ across all GB sites in the base year and NRPO scenarios.

FIGURE 21.

Hourly concentrations of NO₂ at the London North Kensington site in the base year (2011) and the NRPO scenario.

Air quality: PM_2.5

Similar plots are shown for annual mean PM_2.5 in Figures 22 and 23.

FIGURE 22.

Box-and-whisker plots of annual PM_2.5 concentrations in the different scenarios.

FIGURE 23.

Annual mean PM_2.5 concentrations in the various scenarios. (a) 2011; (b) 2035 baseline; (c) 2050 baseline; (d) 2035 NRPO; (e) 2050 NRPO; and (f) 2050 LGHG.

It is clear from these plots that, overall, total PM_2.5 concentrations are projected to decrease relative to 2011 in all scenarios. However, this overall behaviour masks potentially important issues. The reductions in total PM_2.5 concentrations are influenced strongly by the secondary aerosol components of ammonium sulphate and nitrate, which are projected to decrease significantly out to 2050 largely because of reductions in NO_x and SO₂ emissions arising from GB and the rest of Europe. These reductions outweigh the changes in primary PM_2.5 emissions. Reductions in traffic exhaust primary PM_2.5 emissions are projected to occur as a result of increasing use of diesel particle filters in Euro 6 and IV vehicles in the earlier years and the increasing electrification of the road transport fleet further out to 2050. However, within this broad behaviour, projections from UKTM show increased use of biomass use at a similar scale in both the LGHG and the NRPO scenarios. This use is projected to peak in 2035 in the NRPO scenario with a consumption of 1126 PJ compared with only 292 PJ in 2010. By 2050 the biomass use has only decreased slightly from the 2035 peak to 932 PJ in the NRPO scenario. This means that, on this basis, although the median total PM_2.5 concentrations across the country are projected to decrease in 2035, in some areas, notably London, South Wales and the North East, total PM_2.5 concentrations are projected to increase. The same behaviour is likely to be true for the LGHG scenario, although the plots for this scenario are not shown here. The baseline scenario does not show this increase in biomass use, but this is because the scenario does not have the CCA target as a constraint, nor does it have a carbon price included. By 2050, total PM_2.5 levels are projected to decrease in all scenarios with smaller reductions in absolute levels (but not necessarily percentage reductions) occurring in Scotland, where existing levels are overall lower than in the rest of the country.

The increase in primary PM_2.5 is more clearly demonstrated in Figures 24 and 25, in which we have plotted the sum of elemental carbon (EC) and OA, as this sum is a good approximation of the primary PM_2.5 concentration.

FIGURE 24.

Box-and-whisker plots for annual mean EC + OA.

FIGURE 25.

Annual mean EC + OA in the various scenarios. (a) 2011; (b) 2035 baseline; (c) 2050 baseline; (d) 2050 LGHG; (e) 2035 NRPO; and (f) 2050 NRPO.

It is clear from these plots that the NRPO scenario, with the large increase in biomass use, shows a large relative increase in EC and OA in the period out to 2035; the LGHG scenario will show the same behaviour. Within the health effects community there is, at present, no consensus on the relative toxicity of the various components of PM_2.5 (or of PM₁₀) so the precise health implications of this increase are unclear. However, there is a body of evidence linking black carbon (essentially the same component as EC + OA) with adverse health effects (and these emissions are likely to contain carcinogens). 93 As noted above, the biomass use peaks in 2035 but decreases further by only a relatively small amount by 2050. By that year, the median level of EC and OA in the LGHG and NRPO scenarios has decreased by a small amount, but the outliers at high concentration remain as high as they were in 2011.

Overall, therefore, although both scenarios that meet the CCA target show decreases in total annual mean PM_2.5 levels by 2050, both scenarios have an increase in biomass use, which prevents a more accelerated reduction in total PM_2.5 and which actually leads to an increase in concentrations of primary PM_2.5 components in 2035 with a relatively small reduction in 2050 compared with 2011 levels.

Annual mean PM_2.5 concentrations in the LGHG and NRPO scenarios are projected to fall by around 40% in the top 25% of grid squares, but by only ≈25% in the highest areas. However, the concentrations of EC and OA are projected to increase in 2035 in the NRPO and LGHG scenarios by around 30–60% in the more polluted grid squares. By 2050, in those two scenarios, levels are only slightly smaller than 2011 values and in the highest grid square very similar to 2011 concentrations. If this amount of primary PM_2.5 were to be removed, by avoiding the use of biomass as indicated in the NRPO and LGHG scenarios, the total PM_2.5 concentrations could fall even further than projected, potentially down by ≈50% in the highest areas compared with a ≈25% reduction with the increased biomass use.

Air quality: PM₁₀

Similar plots for PM₁₀ are shown in Figures 26 and 27.

FIGURE 26.

Box-and-whisker plots for annual mean PM₁₀ concentrations in the various scenarios.

FIGURE 27.

Annual mean PM₁₀ concentrations in the various scenarios. (a) 2011; (b) 2035 baseline; (c) 2050 baseline; (d) 2035 NRPO; (e) 2050 NRPO; and (f) 2050 LGHG.

The comments regarding biomass use in Air quality: PM_2.5 also apply to PM₁₀. Moreover, an additional component of PM₁₀ emissions is important here, namely non-exhaust emissions from brake, tyre and clutch wear and resuspended road dust. Emissions from this source are uncertain and further research is needed to make more accurate estimates. However, based on current knowledge, these emissions already make up the largest part of emissions of primary PM₁₀ from the road transport sector. In 2014, non-exhaust emissions from road transport in the UK were 73% of the total road transport PM₁₀ emissions. By 2050 the PM₁₀ emission factors from non-exhaust sources are assumed to be the same as they are now, as discussed above. On this basis, the total road transport PM₁₀ emissions in the NRPO scenario are projected to increase by about 15% from 2011 to 2050 as a result of the projected increase in vehicle kilometres driven over that period. This increase in non-exhaust emissions by 2050 will also affect PM_2.5 concentrations, although to a lesser extent than PM₁₀ (in 2050 the road transport emissions of PM_2.5 are about 27% of those of PM₁₀).

The consequence of this is that, despite the reduction in secondary PM₁₀ concentrations resulting from the decreases in NO_x and SO₂ emissions in GB and the rest of Europe, PM₁₀ concentrations are projected to increase in 2035 in the NRPO (and LGHG) scenario in many areas of GB, despite the median concentration showing a decrease. By 2050 there is projected to be a small decrease in much of England but increases in western areas, Wales and Scotland, partly influenced by sea salt contributions in coastal areas.

The projected increase in non-exhaust emissions is clearly of potential concern as regards compliance with standards and guidelines. It is also of concern regarding adverse effects on public health. Research in this area is still relatively new, but there are already grounds for concern over the toxicity of non-exhaust emissions, as outlined in a recent review. 94

Air quality: ozone

Ozone is unique in the set of pollutants discussed in this report, in that different metrics behave differently in response to mitigation measures, because the different metrics are determined by different processes, unlike a less reactive pollutant, concentrations of which are generally proportional to emissions. In the case of O₃, peak hourly concentrations are determined by the well-known photochemical ‘smog’ reactions involving emissions of NO_x and VOCs on a regional scale. This overall process is strongly non-linear, but, in many of the more polluted areas in GB and Europe, peak O₃ concentrations have up until now been more responsive to reductions in VOC emissions. Longer-term average concentrations, however, tend to behave differently, mainly because the photochemical episodes which determine the peak O₃ levels occur on only a small number of occasions, at least at more temperate northern European latitudes such as GB. In the case of these longer-term annual or summer-time averages, the O₃ concentration is governed by the photostationary state equations involving NO, NO₂ and O₃. These essentially ‘titrate’ O₃ via reaction with NO:

The NO₂ photolyses in sunlight to reform NO plus an oxygen atom, which re-forms O₃ by reacting with an oxygen molecule:

These reactions take place very quickly (in minutes) during sunlight and thus interchange O₃ and NO₂ very quickly, such that the sum O₃ + NO₂, defined as oxidant (O_x), is conserved. The important point for long-term average O₃ concentrations is that as NO_x concentrations decrease, the titration reaction removes less O₃ and average O₃ levels increase.

The health impact of O₃ is discussed in more detail in the next section of the report, but here we describe the behaviour of two health-relevant concentration metrics in response to the emission reductions in the future scenarios.

The two metrics refer to long- and short-term exposures and are based on 8-hour averages, an averaging time originally determined by chamber exposure studies as providing the most relevant averaging period. The long-term exposure metric was published by WHO in its project Health Risks of Air Pollution in Europe (HRAPIE)95 and refers to the April–September average of the daily maximum 8-hour mean above 35 p.p.b. (equivalent to 70 µg/m³). The short-term metric we use here derives from the recommendation of COMEAP in its report on the quantification of the health effects of O₃ 96 and is the annual average of the daily maximum 8-hour mean with no threshold or cut-off point, unlike the long-term metric. This point is important in determining the response of the metric to emission reductions.

The plots of the long-term exposure metric are shown in Figures 28 and 29 for all scenarios for GB, modelled at 10-km resolution, and for London, modelled at 2-km resolution, respectively.

FIGURE 28.

Box-and-whisker plots of April–September mean of daily maximum 8-hour O₃ concentrations over GB in the various scenarios (the long-term O₃ exposure metric).

FIGURE 29.

The long-term O₃ exposure metric for London in the various scenarios.

Before discussing these results it is instructive to compare the results for the short-term metric shown in Figures 30 and 31.

FIGURE 30.

Box-and-whisker plots of the annual mean of the daily maximum 8-hour O₃ concentration for GB (the short-term O₃ metric) in the various scenarios.

FIGURE 31.

The short-term O₃ exposure metric for London.

There are three main processes contributing to O₃ levels in the UK. The global tropospheric background provides a baseline of around about 70 µg/m³ everywhere in GB. Superimposed on this level are summer-time O₃ ‘smog’ episodes when hourly peaks can reach higher levels, typically up to ≈200 µg/m³ in the worst episodes. The third main mechanism is the so-called ‘titration’, referred to above, whereby fresh emissions of NO_x can destroy O₃. Combinations of these three mechanisms allow interpretation of the results in Figures 28–31. It is worth noting here that the global background O₃ concentration in our modelling increased slightly in future years. This level will be dependent on global emissions of methane and of NO_x and, in the scenario we have used from the ECLIPSE projections, the tropospheric baseline is projected to increase, which is the main influence on the increase of O₃ annual means in GB in our model. Although we have used one set of projections for these emissions, the scope of this research does not allow us to investigate the sensitivities of GB long-term mean O₃ concentrations to variations in emissions of O₃ precursors in different global scenarios.

It is clear from Figures 28 and 29 that the long-term O₃ exposure metric decreases in response to emission reductions in all scenarios across GB, including in the urban areas shown. This metric is calculated over the summer months, so the reductions are probably because of the reduction in VOC emissions, leading to less photochemical production of O₃.

The short-term metric, however, applies over the whole year, which includes the winter months, when NO_x concentrations are highest and when photochemical production from VOCs is virtually absent. This metric increases, unlike the metric for long-term exposure. The reason is that the short-term metric covers the winter months as well as the summer months and, most importantly, incorporates no threshold, so the average includes the very low O₃ concentrations. These concentrations increase as NO_x emissions are decreased in the future years and this mechanism clearly dominates, leading to an increase in the metric, suggesting that health-relevant short-term exposures will increase. The net effect of these two competing responses to emission reductions is not immediately clear and will be discussed in the next section of the report dealing with health impact (see Air quality: oxidant).

Hourly concentrations of O₃ in the various scenarios are shown in Figures 32 and 33. The effect of the scenarios on peak hourly O₃ concentrations is relatively small, although each scenario shows a maximum hourly mean higher than the maximum in the base year of 2011, with values increasing from ≈138 to ≈170–174 µg/m³. The main difference is shown by an inspection of the frequency distributions which shows clearly the increased proportion of lower O₃ concentrations in the future scenarios caused by the reduction on NO_x emissions. This is also the reason why the mean O₃ concentration increases in future scenarios.

FIGURE 32.

Hourly O₃ concentrations over all GB sites in the various scenarios.

FIGURE 33.

Hourly O₃ concentrations across all GB sites in the base year (2011) and the LGHG scenario.

The role of the various mechanisms in determining hourly O₃ concentrations is very clearly shown in Figure 33. Here, the increase in O₃ concentrations in the winter months is very clear, caused by the reduction in NO_x emissions reducing the extent of O₃ ‘titration’. The reduction in the summer months is probably the result of the reduction in VOC emissions in GB and the rest of Europe leading to a reduction in photochemical formation of O₃.

Air quality: oxidant

It was noted in the previous section (see Air quality: ozone) that in the absence of photochemical episodes, and, hence, in terms of longer-term average concentrations, the sum of O₃ plus NO₂ is conserved as the two pollutants interchange rapidly in sunlight, and, as NO_x concentrations reduce, O₃ concentrations increase to approach the rural or tropospheric background. As well as atmospheric chemistry reasons for investigating O₃ and NO₂ together, there are also health reasons, as both are O_x and have the potential to cause harmful effects in humans. Literature has begun to emerge linking O_x with adverse health effects,97 so it is pertinent to investigate the effect of the energy scenarios here on future concentrations of O_x in GB.

Figures 34 and 35 show the box-and-whisker plots and the annual average O_x concentrations in the various scenarios, from which it can be seen that O_x levels change very little in the different scenarios, with an indication of a slight increase in future years compared with 2011 values. The oxidising power of the atmosphere in the future scenarios will depend on the balance between O₃ and NO₂ at any given point in space or time, as the two pollutants have different redox potentials. 97 In future, as NO₂ concentrations decrease and annual average O₃ and O_x concentrations are projected to increase, the proportions of O₃ and NO₂ making up O_x will shift in favour of O₃. Ozone has a higher redox potential than NO₂,97 so the oxidising power of the atmosphere, in terms of long-term average exposures, is projected to increase.

FIGURE 34.

Box-and-whisker plots of O_x concentrations in the various scenarios.

FIGURE 35.

Annual mean O_x concentrations in the various scenarios. (a) 2011; (b) 2035 baseline; (c) 2050 baseline; (d) 2035 NRPO; (e) 2050 NRPO; and (f) 2050 LGHG.

Introductory remarks, health outcomes, concentration–response functions and methodology

It is now well established that adverse health effects, including mortality, are statistically associated with outdoor ambient concentrations of air pollutants. Moreover, toxicological studies of potential mechanisms of damage have added to the statistical evidence, such that many organisations (e.g. US Environmental Protection Agency, WHO, COMEAP) consider the evidence strong enough to infer a causal relationship between the adverse health effects and the air pollution concentrations. 98 The evidence is strongest for PM_2.5, and the impact on mortality in GB of PM_2.5 concentrations in 2008 were estimated by COMEAP as approximately 29,000 deaths and typically a loss of 340,000 life-years; alternatively, COMEAP suggested that this could be expressed as a 6-month loss of life expectancy from birth. 99

In the Review of the Health Aspects of Air Pollution project,93 WHO systematically examined peer-reviewed articles on air pollution and health and concluded that the evidence for the adverse effects of PM_2.5 had strengthened. Significantly, this review also concluded that the evidence had now strengthened for adverse effects of NO₂ independent of those from PM mass, chiefly from the use of two pollutant models in epidemiological studies. Previous assessments had concluded that it was not possible to separate the effects of PM (or other particle metrics such as ultrafine particles) and NO₂, as they are highly correlated in most urban areas. The review further discussed the emerging evidence for adverse effects of long-term exposures to O₃.

This section uses this epidemiological evidence to estimate the health impact of the changes in air pollutant concentrations predicted by the policy scenarios discussed in Chapter 6.

Impact of long-term exposures

Impact of PM_2.5 (per se, perhaps with contributions from other pollutants)

The results from the life table calculations are shown in Table 13, for anthropogenic PM_2.5 assuming no net migration, with projected new births (2011–2154) in four scenarios, compared with life-years lived with baseline mortality rates (incorporating mortality improvements over time), with a RR of 1.06 per 10 µg/m³ of anthropogenic PM_2.5 and lags from the US EPA.

TABLE 13 - Life-years lost across the GB population for anthropogenic PM2.5

Scenario	Year	GB population-weighted mean concentrations (µg/m3)	Total life-years lost compared with baseline mortality rates, whole population with new births, 2011–2154
			Male	Female	Total
Concentration does not reduce from 2011 levels	2011	9.31	26,636,321	23,505,243	50,141,565
	2035	9.31
	2050	9.31
Baseline (current policies)	2011	9.31	17,148,400	15,154,017	32,302,416
	2035	4.74
	2050	5.77
NRPO	2011	9.31	17,727,599	15,697,023	33,424,622
	2035	7.46
	2050	5.36
LGHG	2011	9.31	17,380,115	15,394,103	32,774,218
	2035	7.46
	2050	5.20

Concentrations were modelled (10-km grid rural/2-km grid urban) at 2011, 2035 and 2050, but also interpolated for the intervening years. Concentrations shown are local authority population-weighted mean concentrations by sex and age (≥ 30 years), averaged across GB. Population-weighted mean concentrations by sex and age (≥ 30 years) by local authority were actually used in the calculations. The RR (hazard ratio) has a plausibility interval of 1.01–1.12 (i.e. from roughly one-sixth to double the numbers given).

The life-years lost give a large number because the life-years (one person living for 1 year) is summed over the whole population and over 143 years. For context, the total life-years lived with baseline mortality rates is around 10.8 billion, so these losses of life-years involve about 0.5% of total life-years lived.

If 2011 concentrations of anthropogenic PM_2.5 remained unchanged for 143 years, around 50.1 million life-years would be lost across the population over that period. This improves to around 32.3–33.4 million life-years lost with the example policies examined here.

Table 14 shows the differences between the three policy scenarios and the scenario with particulate levels constant at 2011 levels, and Table 14 shows the differences between the two extended climate change policy scenarios and the current policy scenario. The figures are illustrated in Figures 36 and 37.

TABLE 14 - Life-years saved across the GB population compared with 2011 anthropogenic PM2.5 concentrations remaining unchanged

Scenario	Year	Change in GB population-weighted mean anthropogenic PM2.5 concentrations compared with 2011 (µg/m3)	Total life-years saved compared with 2011 concentrations maintained, whole population with new births, 2011–2154
			Male	Female	Total
Base case (current policies)	2011	0	9,487,921	8,351,226	17,839,149
	2035	–4.57
	2050	–3.54
NRPO	2011	0	8,908,722	7,808,220	16,716,943
	2035	–1.85
	2050	–3.95
LGHG	2011	0	9,256,206	8,111,140	17,367,347
	2035	–1.85
	2050	–4.11

FIGURE 36.

Cumulative life-years lost for four anthropogenic PM_2.5 scenarios across the GB population (no migration), with projected new births, compared with life-years lived with baseline mortality rates (incorporating mortality improvements over time), 2011–2154. RR 1.06 per 10 µg/m³, US EPA lag.

FIGURE 37.

Life-years saved across the GB population (no migration), with projected new births, 2011–2154, for anthropogenic PM_2.5 in three greenhouse gas reduction scenarios, compared with 2011 anthropogenic PM_2.5 concentrations remaining unchanged, RR 1.06 per 10 µg/m³, US EPA lag.

Although the life-years gained are broadly similar, it can be seen that it is actually the baseline (current policies) scenario that gains the most life-years (17.8 million). Although PM_2.5 concentrations are lower for the NRPO and LGHG scenarios than the base case in 2050, the concentrations do not actually reduce as much in these scenarios in 2035 compared with the baseline scenario. This is because of the increase in biomass emissions discussed above.

Having illustrated that life-years are gained for the base case, Table 15 shows the comparison of the NRPO and LGHG scenarios more clearly. Figure 37 also shows the development of the gains/losses over time.

TABLE 15 - Life-years saved across the GB population for anthropogenic PM2.5 concentrations in two alternative greenhouse gas reduction scenarios compared with the baseline scenario

Scenario	Year	Change in GB population-weighted mean concentrations compared with 2011 (µg/m3)	Total life-years saved compared with the baseline (current policies) scenario whole population with new births, 2011–2154
			Male	Female	Total
NRPO	2011	0	–579,199	–543,006	–1,122,206
	2035	2.72
	2050	–0.41
LGHG	2011	0	–231,715	–240,086	–471,802
	2035	2.72
	2050	–0.57

Table 15 shows that both scenarios are actually worse than the baseline case, with current policies showing 1.1 million and 470,000 life-years lost (negative life-years gained) for NRPO and LGHG, respectively, compared with the base case. Figure 36 shows that the cumulative life-years lost in the base case (current policies) accumulate more slowly than either the constant 2011 concentration results or in the NRPO and LGHG scenarios, although the last two get closer to each other for the later years. Figure 37 also shows this. It is worth remembering that there is a delay before the full benefits of concentration reductions are achieved. This is not just attributable to a lag between exposure and effect, but also because the greatest gains occur when mortality rates are highest (i.e. in the elderly). Thus, although by 2050 the concentration reduction is a little bit greater for the NRPO and LGHG scenarios and this difference in concentration is maintained unchanged after 2050, it is another 30–40 years before the cumulative life-years gained start to ‘catch up’ with the base-case scenario. Similarly, when comparing the NRPO and LGHG scenarios with each other, their concentrations start to diverge in 2035, but it is not until 2070 or so before the slopes start to diverge noticeably from each other. When the concentration differences first occur, the effect will be strongly apparent only in the elderly portion of the population but, over time, more and more of the original population become elderly and thus affected by the maintained change. This is why long time periods are needed for the analysis. These effects can be more clearly illustrated by plotting the life-years lost per year during the period of analysis relative to the baseline scenario. This quantity is shown in Figure 38 and shows the effect of the increase in biomass use in the NRPO and LGHG scenarios compared with the baseline scenario. The large increase in life-years lost per year peaks in the 2040s before becoming ‘life-years gained’ well into the period beyond the CCA target on this analysis, around 2070.

FIGURE 38.

Life-years lost per year from long-term exposure to anthropogenic PM_2.5 relative to the baseline scenario.

Total life-years across the population is the most appropriate metric for cost–benefit analysis of policies, as it captures effects in the entire population. However, it is a difficult type of metric to communicate as it is difficult to judge what is a ‘small’ answer or a ‘large’ answer. Life expectancy from birth is a more familiar concept for the general public, although it captures effects only on those born on a particular date. Results for life expectancy from birth are shown in Table 16.

TABLE 16 - Loss of life expectancy from birth in 2011 for anthropogenic PM2.5 for four scenarios compared with life expectancy with baseline mortality rates

Scenario	Year	GB population-weighted mean concentrations (µg/m3)	Loss of life expectancy from birth compared with baseline mortality rates, 2011 birth cohort followed for 105 years (weeks)
			Male	Female
Concentration does not reduce from 2011 levels	2011	9.31	29.5	25.8
	2035	9.31
	2050	9.31
Base case (current policies)	2011	9.31	18.5	16.2
	2035	4.74
	2050	5.77
NRPO	2011	9.31	17.6	15.3
	2035	7.46
	2050	5.36
LGHG	2011	9.31	17.1	14.9
	2035	7.46
	2050	5.20

This shows that loss of life expectancy from birth would be about 26–30 weeks (6–7 months) if 2011 PM_2.5 concentrations were unchanged, but improves by about 9–11 weeks for the base case and by 10–12 weeks for the NRPO and LGHG scenarios. This is only a small improvement over the base case. Considering the confidence intervals in the CRFs, the results varied from one-sixth to almost twice these values when using the plausibility interval from a hazard ratio of 1.01 to one of 1.12.

Modelling scale can affect modelled concentrations and, hence, health impact. However, as PM_2.5 is fairly spatially homogeneous as a result of regional contributions of PM_2.5, changes in modelling scale had only a small effect when tested on the NRPO scenario. GB population-weighted mean concentrations in 2035 were 7.33 µg/m³ for 10-km grid-modelled concentrations in both rural and urban areas and 7.44 µg/m³ for 10-km rural/20-m urban modelling, compared with 7.46 µg/m³ for the 10-km grid rural/2-km grid urban modelling used for all the scenarios. The equivalent figures for 2050 were 5.26, 5.36 and 5.36 µg/m³. Unsurprisingly, these differences reduced the answers in terms of life-years lost by only 3% for the 10-km rural and 10-km urban modelling and by 0.1% for the 10-km rural/20-m urban modelling. This was based on concentrations averaged up to ward level, which will average out some of the exposure contrasts, perhaps explaining the relatively small impact of finer-scale, 20-m, modelling (see Table 10).

Although the results discussed in this section refer to PM_2.5, and it is likely that PM_2.5 itself is a significant contributor to the effects, it is possible that the effects of other traffic pollutants are being reflected to some extent.

Impact of nitrogen dioxide (and possibly other traffic or combustion pollutants)

The results for NO₂ are given in Table 17. The exact pollutants responsible for the associations in reality are unclear, although it is thought likely that NO₂ itself makes some contribution. 100 References to effects of changes in NO₂ concentrations in the following text should be read with this in mind. The results are given as a range from using a cut-off (counterfactual) point of 5 µg/m³, to assuming effects can be extrapolated down to zero.

TABLE 17 - Life-years lost across the GB population for NO2 in four scenarios

Scenario	Year	NO2 GB population-weighted mean concentrations (µg/m3)	Total life-years lost compared with baseline mortality rates, whole population with new births, 2011–2154
			Male	Female	Total
Concentration does not reduce from 2011 levels	2011	9.53–14.4	9,439,172 to 13,322,689	8,243,553 to 11,689,378	17,682,726 to 25,012,067
	2035	9.53–14.4
	2050	9.53–14.4
Base case (current policies)	2011	9.53–14.4	6,015,864 to 9,856,596	5,260,688 to 8,667,812	11,276,553 to 18,524,408
	2035	2.44–6.98
	2050	6.43–11.25
NRPO	2011	9.53–14.4	3,408,301 to 7,145,468	2,976,271 to 6,291,577	6,384,572 to 13,437,045
	2035	2.82–7.37
	2050	2.55–7.08
LGHG	2011	9.53–14.4	2,370,507 to 6,019,928	2,088,795 to 5,326,466	4,459,303 to 11,346,394
	2035	2.82–7.37
	2050	1.33–5.65

The calculations use a RR (hazard ratio) of 1.017 per µg/m³ of NO₂, with a confidence interval of 1.007 to 1.027 (i.e. from roughly 40% to 1.5 times the numbers given). The effects are unlikely to represent effects of NO₂ alone, as this pollutant is closely correlated with other traffic/combustion pollutants.

As expected, the results assuming no effect < 5 µg/m³ are smaller. In addition, the lower the population-weighted mean concentration gets, the more proportional difference the cut-off point makes to the results. This is because a greater number of areas are < 5 µg/m³ and set to zero in the calculations. The ‘minus 5’ population-weighted mean varies from 66% (for 2011) to 24% (LGHG, 2050) of the zero cut-off point result. The life-years lost results range from 71% (2011 constant) to 39% (LGHG) of the zero cut-off point result. (The rank order for the life-year results follows that of the population-weighted mean concentrations, but the proportions do not match exactly.) There are probably a variety of reasons for this: the population-weighted means vary over time; the relationship between change in population-weighted mean and change in hazard rate is not linear; and the life table calculations take account of some wards within a local authority, but not others, falling < 5 µg/m³ over time, including the intervening years between 2011, 2035 and 2050.

In contrast to the results using the PM_2.5 coefficient, there is more difference between the scenarios ranging from 4.5 million life-years lost (LGHG, 5 µg/m³ cut-off point) to 25 million life-years lost (2011 constant, no cut-off point), or within the same cut-off point assumption (11–25 million life-years lost). Overall, life-years lost are lower than for PM_2.5, the smaller size of the coefficient compensating for the slightly larger NO₂ concentration in 2011.

Compared with no change in 2011 concentrations (Table 18), the reduction in population-weighted mean concentrations is now much closer between the zero and 5 µg/m³ cut-off points, although as the reductions get larger (and more areas fall < 5 µg/m³), there is more of a difference (the extent of the reduction is ‘blunted’ by the cut-off point at 5 µg/m³).

TABLE 18 - Life-years saved across the GB population (no migration) for NO2 in three greenhouse gas reduction scenarios, compared with 2011 NO2 concentrations remaining unchanged

Scenario	Year	Difference in NO2 GB population-weighted mean concentrations (µg/m3)	Total life-years saved compared with 2011 concentrations maintained, whole population with new births, 2011–2154
			Male	Female	Total
Baseline (current policies)	2011	0	3,423,308 to 3,466,093	2,982,865 to 3,021,566	6,406,173 to 6,487,659
	2035	–7.09 to –7.42
	2050	–3.1 to –3.15
NRPO	2011	0	6,030,871 to 6,177,221	5,267,282 to 5,397,801	11,298,154 to 11,575,022
	2035	–6.71 to –7.03
	2050	–6.98 to –7.32
LGHG	2011	0	7,068,665 to 7,302,761	6,154,758 to 6,362,912	13,223,423 to 13,665,673
	2035	–6.71 to –7.03
	2050	–8.2 to –8.75

The life-years gained vary by a factor of 2, with the LGHG scenario giving considerably more benefits (13.2–13.6 million life-years saved) than the base case (6.4–6.5 million life-years saved). The gains are smaller than for PM_2.5, despite the larger concentration changes, as a result of the smaller coefficient.

Comparing the NRPO and LGHG scenarios with the base case in Table 19, the population-weighted mean concentrations are actually very slightly higher for the NRPO and LGHG scenarios than in the base case in 2035, but by 2050 these two scenarios lead to substantially lower concentrations. Compared with PM_2.5, concentrations of which are a large amount higher in 2035 and a small amount lower in 2050, the net result for NO₂ is for gains in life-years. The gains for the LGHG scenario are greater, at 6.8–7.2 million life-years, than for the NRPO scenario (4.9–5.1 million life-years).

TABLE 19 - Life-years saved across the GB population for NO2 concentrations in two alternative greenhouse gas reduction scenarios compared with the baseline scenario

Scenario	Year	Difference in NO2 GB population-weighted mean concentrations (µg/m3)	Total life-years saved compared with the base case, whole population with new births, 2011–2154
			Male	Female	Total
NRPO	2011	0	2,607,563 to 2,711,128	2,284,417 to 2,376,235	4,891,981 to 5,087,363
	2035	0.38 to 0.39
	2050	–3.88 to –4.17
LGHG	2011	0	3,645,357 to 3,836,668	3,171,893 to 3,341,346	6,817,250 to 7,178,014
	2035	0.38 to 0.39
	2050	–5.1 to –5.6

Figure 39 shows that, with the sharper concentration changes from 2035 to 2050, the impact of the scenarios diverge more obviously than for PM_2.5 from about 2050. The pattern is similar for the 5 µg/m³ cut-off point, at a smaller level of life-years lost (Figure 40). Figure 41 shows the comparison with 2011 concentrations remaining unchanged from 2011 and Figure 42 plots the cumulative life-years lost compared with baseline mortality rates with and without a cut-off point on the same graph.

FIGURE 39.

Cumulative life-years lost for four NO₂ scenarios across the GB population (no migration), with projected new births, compared with life-years lived with baseline mortality rates (incorporating mortality improvements over time) 2011–2154. RR 1.017 per 10 µµg/m³, US EPA lag, no cut-off point.

FIGURE 40.

Cumulative life-years lost for four NO₂ scenarios across the GB population (no migration), with projected new births, compared with life-years lived with baseline mortality rates (incorporating mortality improvements over time) 2011–2154. RR 1.017 per 10 µg/m³, US EPA lag, 5 µg/m³ cut-off point.

FIGURE 41.

Life-years saved across the GB population (no migration), with projected new births, 2011–2154, for NO₂ in three greenhouse gas reduction scenarios, compared with 2011 NO₂ concentrations remaining unchanged, RR 1.017 per 10 µg/m³, US EPA lag, no cut-off point.

FIGURE 42.

Cumulative life-years lost across the GB population (no migration), with projected new births, 2011–2154, for NO₂ in an example greenhouse gas reduction scenario (LGHG), compared with life-years lived with baseline mortality rates (incorporating mortality improvements over time), RR 1.017 per 10 µg/m³, US EPA lag, with and without cut-off point of 5 µg/m³.

Loss of life expectancy from birth is shown in Table 20.

TABLE 20 - Loss of life expectancy from birth in 2011 for four scenarios compared with life expectancy with baseline mortality rates

Scenario	Year	NO2 GB population-weighted mean concentrations (µg/m3)	Loss of life expectancy from birth compared with baseline mortality rates, 2011 birth cohort followed for 105 years (weeks)
			Male	Female
Concentration does not reduce from 2011 levels	2011	9.53–14.4	10.7 to 15	9.3 to 13.1
	2035	9.53–14.4
	2050	9.53–14.4
Base case (current policies)	2011	9.53–14.4	7.1 to 11.3	6.2 to 9.9
	2035	2.44–6.98
	2050	6.43–11.25
NRPO	2011	9.53–14.4	3.2 to 7.3	2.8 to 6.4
	2035	2.82–7.37
	2050	2.55–7.08
LGHG	2011	9.53–14.4	1.7 to 5.7	1.4 to 4.9
	2035	2.82–7.37
	2050	1.33–5.65

This shows that loss of life expectancy from birth would be about 9–15 weeks (2–3.5 months) if 2011 NO₂ concentrations were unchanged, but improves by about 3–4 weeks for the base case, by 7–8 weeks for the NRPO scenario and by 8–9 weeks for the LGHG scenario. The loss of life expectancy from birth would be very small with respect to NO₂ (and/or pollutants with which it is closely correlated) after implementation of the NRPO or LGHG scenarios compared with the base case. Results varied from 41% of the central estimate to 1.6 times the relevant values with or without a cut-off point when using the confidence interval from a hazard ratio of 1.007 to one of 1.027.

The plot of life-years lost per year is shown in Figure 43 and the difference between this plot and that for PM_2.5 is very clear. Apart from a very small loss in life-years at the beginning of the period, before the larger emission reduction measures take effect, life-years are saved in both scenarios compared with the baseline, with a relatively small difference with and without a cut-off point.

FIGURE 43.

Life-years lost per year for NO₂, relative to the baseline scenario.

The sensitivity on modelling scale for the NRPO scenario showed GB population-weighted mean concentrations of NO₂ (no cut-off point) in 2035 were 6.93 µg/m³ for 10-km grid modelled concentrations in both rural and urban areas and 7.37 µg/m³ for 10-km rural/20-m urban modelling, compared with 7.37 µg/m³ for the 10-km rural/20-km urban modelling used for all the scenarios. The equivalent figures for 2050 were 6.72, 7.07 and 7.08 µg/m³. The differences are slightly greater than for PM_2.5, as would be expected for a more spatially homogeneous pollutant. These differences reduced the life-years lost by 8% for the 10-km rural and 10-km urban modelling and increased the life-years lost by 0.3% for the 10-km rural/20-m urban modelling. For NO₂, omitting more detailed 2-km modelling of urban areas underestimates the impact, but moving up to 20-m modelling from 2 km makes less of a difference. Again, averaging up to ward level probably averages out some exposure contrasts from the 20-m modelling (see Table 10). The life-years lost results for NO₂ with a cut-off point (2011 baseline) were 12% lower for the 10 km/10 km modelling and 3% higher for the 10 km/20 m modelling. The respective values for the NRPO scenario were 17% lower and 1% higher. As expected, these are larger differences than for NO₂ without a cut-off point, as areas close to 5 µg/m³ may be above or below this cut-off point according to modelling scale.

Impact of ozone

In considering the long-term impact of O₃, we need to be aware of the uncertainties surrounding the epidemiological evidence. The WHO’s HRAPIE study95 recommended a CRF as set out in Table 11, but it was graded as B (a measure of confidence), compared with, for example, the CRF for PM_2.5 which was graded A*. We therefore consider this quantification as a sensitivity calculation and will discuss the results in less detail than for PM_2.5 and NO₂.

The results are shown in Table 21.

TABLE 21 - Life-years lost across the GB population from exposure to the April–September average of the daily 8-hour maximum O3 concentration above 35 p.p.b. (70 µg/m3), population-weighted mean

Scenario	O3 GB population-weighted mean concentrations (µg/m3)	Total life-years lost compared with baseline mortality rates, whole population with new births, 2011–2154
		Male	Female	Total
Concentration does not reduce from 2011 levels	34.7	3,146,765	2,692,044	5,838,808
Baseline (current policies)	27.7	2,536,606	2,176,301	4,712,907
NRPO	28.7	2,655,877	2,273,953	4,929,830
LGHG	28.7	2,669,961	2,284,088	4,954,049

The same results are shown graphically in Figure 44.

FIGURE 44.

Cumulative life-years lost associated with long-term O₃ exposure compared with baseline mortality rates.

The overall impact is of a similar order to that estimated for NO₂ with a cut-off point of 5 µg/m³ and the baseline (no further measures) scenario shows fewer life-years lost than the two CCA-compliant scenarios. The reason for this is probably because of the lower NO_x emissions in the two CCA-compliant scenarios, LGHG and NRPO, in which the effect of removing the ‘titration’ effect of NO and O₃ is having an effect on the higher O₃ concentrations.

The loss of life expectancy from birth in 2011 if O₃ concentrations remain unchanged was estimated at around 3 weeks, with a very slight improvement for the scenarios to 2–3 weeks. This is an average across all those born in 2011, and, particularly as the O₃ coefficient only relates to respiratory mortality, it is more likely that there is a greater loss of life expectancy for some, while others are unaffected.

The impact on health of long-term O₃ exposure, albeit uncertain, in all three future scenarios is significantly smaller than that of PM_2.5, and that of NO₂ with no cut-off point, by factors of around 6 and around 3–4, respectively.

The question of the existence of a threshold for O₃ effects is extremely important. In considering the impact of short-term exposures to O₃, COMEAP96 did not recommend qualifying the effects of long-term exposures, but recommended a short-term CRF for all-cause mortality and recommended using no threshold (the COMEAP approach was to meta-analyse studies of all-cause mortality, whereas WHO HRAPIE95 work selected a particular study that showed effects on respiratory mortality). This is a significant statement as noted by Heal et al. ,130 as an approximate calculation shows. Taking the COMEAP CRF of 0.34% increase in mortality per 10 µg/m³ increase in the daily maximum 8-hour mean concentration and assuming linearity, the mortality outcome is determined by the annual mean concentration, which is of the order of 100 µg/m³ (see Table 10). Assuming a UK population of 65 million and an annual mortality base rate of 1%, one estimates the number of premature deaths from short-term exposure to O₃ as ≈22,000, a figure broadly comparable to the figure of ≈29,000 premature deaths estimated from long-term exposure to PM_2.5. 59 Heal et al. ,130 in a study of 2003 O₃ concentrations, derived a figure of ≈11,500 DBF using an earlier CRF recommended by WHO. We intend to do more investigation of the effects of changes in short-term exposure to pollutants as a result of climate change policies in further development of this work.

Background

Environmental inequality, the concept that more vulnerable individuals, communities and subpopulations are more likely to be exposed to higher levels of environmental pollution, is well documented, particularly in relation to air pollution. 132,133 Environmental inequality implies disadvantages in subpopulations and certain communities because increased levels of both pollution exposure and socioeconomic deprivation may lead to impaired health. 134 Social gradients in health are well established in England, and it has been estimated that between 1.3 and 2.5 million years of life are lost because of health inequalities. 135 Socially and economically disadvantaged people may experience increased susceptibility to the negative air pollution-related health effects, ranging from conditions such as respiratory irritation and cardiovascular disease to premature death, as a result of higher underlying baseline disease rates in deprived communities. 136 Some studies have shown this effect modification on morbidity and mortality risks, in particular for urban areas, where individuals of high social class are not as affected by the negative health effects of air pollution as are individuals of lower social classes. 137 The relationships between the geographical distribution of vulnerable communities and air pollution are, however, more complex than often implied. 138 They vary considerably by geographical area,139 by environmental pollutant140–142 and over time. 143

Understanding how air pollution exposure differs between subpopulations is important for environmental policy and public health, not only to reduce the average risk across the population but, also, to safeguard that no population subgroup such as deprived or ethnic minorities is more burdened than others. 144

Here, we aim to investigate small area associations of air pollution exposures with socioeconomic deprivation and ethnicity and explore the underlying geographical relationships in environmental inequality. We analyse variations in exposure to air pollution by subpopulation across GB and explore if differences in observed national patterns vary by country and government region. This allows us to identify areas at highest risk of environmental inequality for each of the air pollution scenarios.

Data and methods

Study area

The small area unit for this study is the ward 2011 level. Wards are one of the dissemination units for which output statistics from the 2011 census are reported for England and Wales by the ONS and for Scotland by the National Records of Scotland. There are 8941 wards in GB, with an average population of ≈3500. To analyse geographical differences across GB we stratified our analysis by country and, within England, by region, using the 2011 ONS region boundaries (Figure 45).

FIGURE 45.

Study area.

Population characteristics

We analysed routinely available population characteristics at the ward 2011 level focusing on socioeconomic status and ethnicity.

We used the Carstairs 2011 index as the area-level socioeconomic indicator. The Carstairs index is a composite measure of deprivation which is commonly used in epidemiological and health analysis. 145,146 The index consists of census variables which are available and consistent across the three countries of England, Scotland and Wales. The Carstairs index includes the following four variables:

1. unemployment – defined as unemployed males aged 16–74 years as a proportion of all economically active males aged 16–74 years

2. car ownership – defined as households which do not own a car as a proportion of all private households

3. overcrowding – defined as households with more than one person per room as a proportion of all households

4. low social class – defined as persons in households with household reference person aged 16–74 years in social class D or E as a proportion of all persons in households.

We calculated the Carstairs 2011 index following the methodology described by Carstairs and Morris147 using 2011 census data at the ward 2011 level. To compute the Carstairs deprivation score, the four input variables are combined by summing the standardised input variables. A higher Carstairs score thereby implies a greater degree of deprivation. To compare the difference in the relative ranking of all wards 2011 level, we categorised the Carstairs index into five equal groups (the Carstairs quintiles). Figure 46 shows the national distribution of Carstairs quintiles. We assume the same deprivation patterns observed in 2011 for the 2035 and 2050 scenarios, as relative deprivation patterns such as deprivation quintiles are fairly stable over time. 148

FIGURE 46.

Carstairs index using data from the 2011 census, standardised across all wards in GB (left) and ethnicity defined as % non-white based on 2011 census (right).

We categorised the wards (2011) according to their ethnic composition as white or non-white (see Figure 46). Using ethnicity information from the 2011 census, we defined white as white British, white Irish and white other. All other ethnic groups, including mixed ethnicity, were defined as non-white. We used a 16% cut-off point to differentiate between the two categories which reflects twice the national average of 8% non-white across all wards (2011) in GB. We analysed differences in air pollution levels by ethnicity for 2011 but not for the 2035 and 2050 scenarios, as ethnic composition of wards varies over time and no ethnicity projection information is available to conduct such analysis.

To analyse correlations of each air pollution variable with Carstairs index and ethnicity we used Pearson’s correlations, scatterplots and box-and-whisker plots. All statistical analyses were performed with open-source software R version 3.3.2 and R Studio (The R Foundation for Statistical Computing, Vienna, Austria).

Results

Descriptive statistics for annual mean NO₂, Dmax NO₂, O₃, PM₁₀ and PM_2.5 concentrations at the ward 2011 level across GB are presented in Appendix 2. Concentrations were approximately normally distributed for all pollutants (histograms on request).

Associations of air pollution exposure and deprivation

Figure 47 shows the relationship between selected air pollutants (NO₂, PM_2.5 and O₃) and Carstairs deprivation scores, and the Pearson’s correlations between the variables for all wards (2011) in GB. Scatterplots for the other regions are available on request from the authors.

FIGURE 47.

Scatterplot showing relationship between Carstairs 2011 and pollutants (a–f) NO₂; (g–l) PM_2.5; (m–r) O₃ (the long-term exposure metric); and (s–x) O₃ no threshold (the short-term exposure metric), at ward 2011 and under different scenarios. Green indicates linear regression line and Pearson’s correlations are shown in blue.*Statistical significance p < 0.001.

Nitrogen dioxide and PM_2.5 concentrations are weakly correlated with the Carstairs score in 2011 (NO₂, r = 0.309; PM_2.5, r = 0.139), indicating that wards with a higher deprivation score have higher air pollution concentrations (although the spread across wards is very wide for both pollutants). For both pollutants, baseline scenarios have slightly weaker correlations than NRPO scenarios in 2035 and 2050. In contrast, in the LGHG scenario, the NO₂ correlation with deprivation is the weakest of all the scenarios (r = 0.203) while for the PM_2.5 correlation with deprivation is the highest of all 2050 scenarios (r = 0.213). All correlations for NO₂ and PM_2.5 are statistically significant with α = 0.001.

For O₃, there is a weak negative correlation with Carstairs scores in 2011 using either O₃ metric (with threshold, r = –0.340; without threshold, r = –0.303). For either of the scenarios, there were no, or very weak, correlations, which ranged from –0.173 to 0.015 for the O₃ metric with threshold of 70 µg/m³ and from –0.223 to –0.094 for the O₃ metric without a threshold.

Table 24 shows the mean air pollution levels for the highest compared with the lowest Carstairs quintile for GB and the regions London, Wales and Scotland (results for other regions are available on request from the authors). NO₂ patterns by Carstairs quintile suggest that as air pollution levels decrease, so does the ratio of mean concentrations in the most deprived fifth of wards to those in the least deprived fifth of wards (2011) across GB. In London, this is true for the baseline and LGHG scenarios, whereas the ratio for the NRPO scenario increases. In Wales, however, the ratio of mean NO₂ concentrations in the most deprived fifth of wards to those in the least deprived fifth of wards (2011) increases for all scenarios as NO₂ levels decrease, and is highest for the LGHG scenario in 2050. PM_2.5 patterns by Carstairs quintile show a similar trend for GB, Scotland, Wales and London. As PM_2.5 levels drop, the ratio of the most deprived to the least deprived fifth of wards (2011) increases. This increase is particularly strong for the NRPO scenarios in all countries. In 2011, O₃ levels in London were higher in the least deprived quintile than in the most deprived quintile. For all scenarios, this difference disappears and levels are almost the same in 2050. For other countries and GB, differences between the most and least deprived fifth of wards are minimal, with marginally higher exposure in the least deprived fifth of wards, except in Wales, where exposures are slightly higher in the most deprived fifth of wards.

TABLE 24 - Mean air pollution concentrations (µg/m3) for the most compared with the least deprived fifth of wards

Region	Pollutant (µg/m3)
	NO2			PM2.5			O3 with threshold			O3 without threshold
	Most deprived	Least deprived	NO2 ratio	Most deprived	Least deprived	PM2.5 ratio	Most deprived	Least deprived	O3 ratio	Most deprived	Least deprived	O3 ratio
GB
2011	15.9	11.6	1.37	9.9	9.4	1.05	33.0	35.8	0.92	93.8	96.9	0.97
2035 baseline	7.3	5.7	1.28	5.0	4.7	1.06	28.2	28.3	0.99	100.1	100.9	0.99
2035 NRPO	7.9	5.8	1.36	8.1	7.0	1.16	27.9	28.2	0.99	99.6	100.8	0.99
2050 baseline	11.8	9.0	1.31	6.6	6.1	1.08	25.9	27.0	0.96	96.8	98.7	0.98
2050 LGHG	5.9	4.8	1.23	5.6	5.0	1.12	28.0	27.8	1.00	100.7	101.1	1.00
2050 NRPO	7.7	5.6	1.38	5.8	5.2	1.12	27.7	27.9	0.99	99.6	100.7	0.99
London
2011	26.8	19.5	1.37	12.9	11.5	1.12	30.3	35.6	0.85	87.4	93.8	0.93
2035 baseline	11.0	8.8	1.25	6.9	5.9	1.17	28.9	30.1	0.96	98.7	100.9	0.98
2035 NRPO	13.8	9.5	1.45	12.6	10.4	1.21	27.8	29.7	0.94	96.4	100.1	0.96
2050 baseline	15.0	12.2	1.23	8.6	7.4	1.16	26.9	28.5	0.94	95.7	98.4	0.97
2050 LGHG	8.9	7.2	1.24	8.3	6.9	1.20	29.0	29.8	0.97	99.9	101.5	0.98
2050 NRPO	13.2	9.1	1.45	8.4	7.0	1.20	27.8	29.5	0.94	96.8	100.2	0.97
Wales
2011	8.6	7.8	1.10	8.1	7.9	1.03	35.4	35.2	1.01	98.4	98.4	1.00
2035 baseline	4.0	3.5	1.14	3.7	3.6	1.03	29.1	28.5	1.02	103.2	103.0	1.00
2035 NRPO	4.0	3.4	1.18	6.2	5.5	1.13	29.1	28.5	1.02	103.2	102.9	1.00
2050 baseline	6.1	5.4	1.13	4.9	4.8	1.02	27.7	27.4	1.01	101.6	101.6	1.00
2050 LGHG	3.8	3.0	1.27	4.2	3.9	1.08	28.5	28.1	1.01	103.0	102.8	1.00
2050 NRPO	4.3	3.5	1.23	4.3	4.0	1.08	28.6	28.2	1.01	102.8	102.7	1.00
Scotland
2011	7.8	5.0	1.56	5.2	4.4	1.18	33.1	35.3	0.94	97.4	101.2	0.96
2035 baseline	3.9	2.7	1.44	2.9	2.5	1.16	27.0	28.1	0.96	100.7	102.3	0.98
2035 NRPO	4.1	2.9	1.41	3.9	3.1	1.26	26.8	27.9	0.96	100.4	102.1	0.98
2050 baseline	6.4	4.2	1.52	3.9	3.3	1.18	26.2	27.9	0.94	99.0	101.5	0.98
2050 LGHG	3.1	2.3	1.35	3.0	2.6	1.15	26.6	27.4	0.97	101.0	102.2	0.99
2050 NRPO	3.9	2.8	1.39	3.1	2.6	1.19	26.8	27.7	0.97	100.5	102.1	0.98

Figures 48–51 show the distribution of ward-level NO₂ concentrations by deprivation quintile, for GB, London, Wales and Scotland, respectively. NO₂ is the pollutant with the highest difference in concentrations; results for other pollutants and regions are available on request from the authors.

FIGURE 48.

Box-and-whisker plots showing distribution of ward 2011-level NO₂ concentrations by Carstairs 2011 quintile at 2011 baseline and under different scenarios in GB. (a) 2011; (b) 2035 baseline; (c) 2035 NRPO; (d) 2050 LGHG; (e) 2050 baseline; and (f) 2050 NRPO.

FIGURE 49.

Box-and-whisker plots showing distribution of ward 2011-level NO₂ concentrations by Carstairs 2011 quintile at 2011 baseline and under different scenarios in London. (a) 2011; (b) 2035 baseline; (c) 2035 NRPO; (d) 2050 LGHG; (e) 2050 baseline; and (f) 2050 NRPO.

FIGURE 50.

Box-and-whisker plots showing distribution of ward 2011-level NO₂ concentrations by Carstairs 2011 quintile at 2011 baseline and under different scenarios in Wales. (a) 2011; (b) 2035 baseline; (c) 2035 NRPO; (d) 2050 NRPO; (e) 2050 LGHG; and (f) 2050 baseline.

FIGURE 51.

Box-and-whisker plots showing distribution of ward 2011-level NO₂ concentrations by Carstairs 2011 quintile at 2011 baseline and under different scenarios in Scotland. (a) 2011; (b) 2035 baseline; (c) 2035 NRPO; (d) 2050 LGHG; (e) 2050 baseline; and (f) 2050 NRPO. NA, not applicable.

In GB, we observed stable mean NO₂ concentrations for the first to fourth quintiles, but higher mean NO₂ concentrations for the fifth quintile. In London, NO₂ levels steadily increase from lowest to highest deprivation quintile. In Wales and Scotland, we observed more U- and J-shaped distributions, respectively, of NO₂ concentrations with lowest concentrations in the middle deprivation quintiles. In all regions, however, wards in the lowest deprivation quintiles had the highest mean NO₂ concentrations.

Discussion and conclusion

We analysed small area associations of air pollution exposures with ethnicity and socioeconomic deprivation for current and future projection scenarios. We observed differences in air pollution levels by subpopulations for all analysed pollutants and for each geographical area. Differences in exposure were most marked for NO₂, as shown in Table 23 for ethnicity and Table 24 for socioeconomic deprivation. This is unsurprising as NO₂ is predominantly a primary pollutant (i.e. its concentrations are less dependent on slow chemical reactions in the atmosphere and are hence highest near to sources, which, for virtually all areas of GB, are roads). Wards with higher proportions of non-white residence and higher deprivation are expected to be closer to roads and, therefore, have higher NO₂ levels. Relative differences (i.e. ratios) in NO₂ between white and non-white populations were much larger than the differences between the most and least deprived populations in GB and Wales in 2011, but slightly smaller in London. Relative differences between most and least deprived populations in 2011 were highest in Scotland, closely followed by London; relative differences in Wales were the smallest.

Ozone and PM_2.5 show smaller differences. A significant fraction of PM_2.5 in GB is made up of secondary aerosol formed over long periods (up to several days) from emissions in other parts of GB and in the rest of Europe. Spatial patterns of PM_2.5 annual averages are therefore more homogeneous than for NO₂. The relative differences between subpopulations are consequently smaller than for NO₂.

Ozone is a purely secondary pollutant with a fairly homogeneous spatial distribution across the country, but with the added factor that it is destroyed by chemical reactions with fresh NO_x emissions. Annual average O₃ concentrations are, therefore, smaller in areas where NO₂ is higher and levels are, hence, lower in areas closest to roads and in cities. This effect across the ward level is averaged out to some extent, so the differences between subpopulations are relatively small for all geographical areas and there is essentially no difference in annual mean O₃ exposures across the country at the ward level.

The future scenarios all reduce the absolute values of the concentrations, with the exception of the NRPO scenario in 2035. This scenario includes a large increase in biomass, discussed above, and this is probably the reason why the concentrations are no lower than those in the baseline scenario. The LGHG scenario also has a similarly large increase in biomass and, although we did not model the 2035 LGHG scenario, the exposures for PM_2.5 would be expected to be very similar to those of the NRPO scenario in 2035, as they are in 2050.

There is some indication that the future scenarios for NO₂ reduce the differences between most and least deprived populations as measured by the ratios (apart from the high biomass 2035 NRPO and probably the 2035 LGHG scenarios). This is less clear for PM_2.5, and there is even some indication that, in this case, the ratios increase.

For O₃, all future scenarios show almost identical exposures across the indices for most and least deprived populations.

The plots in Figures 48–51 show that, despite the large decreases in NO₂ concentrations, there are still differences in exposures across the deprivation quintiles. This is because, in the future scenarios, NO₂ concentrations are still highest near roads despite the large reductions in road traffic emissions, as discussed in Chapter 6. The increase in exposure across the deprivation quintiles is steady in London, whereas in Scotland the lower quintiles are broadly similar in exposure with the highest quintile showing a marked increase. In Wales, the NO₂ levels by quintiles show more of a U-shaped behaviour, with the least deprived classes having slightly higher exposures, although the most deprived wards show the highest exposures in all scenarios.

Overall, therefore, we conclude that all future scenarios reduce the absolute levels of pollution exposure in all deprivation quintiles across GB, except in those scenarios where there is a large increase in biomass burning. Differences in exposure between the most and least deprived populations remain in all future scenarios, most clearly for NO₂, but are somewhat reduced. From the environmental inequality perspective, there appears to be little difference in NO₂ exposure differentials between most and least deprived groups when comparing the baseline scenario, which does not meet the CCA target, and the NRPO scenario, which meets the target but with limited nuclear build. The NRPO scenario leads to a higher NO_x emission than the LGHG scenario, which meets the CCA target with no constraint on nuclear build. In this latter scenario, for London at least, the differences in exposure between the most and least deprived populations look to be smaller than in the other two scenarios. The absolute differences for the LGHG scenario are 1.7 µg/m³, baseline 2.8 µg/m³ and NRPO 4.1 µg/m³.

Methods

Results

Under the baseline scenario, the overall magnitude of benefits equates to an annual cost of between £24B and £28B, depending on whether or not a threshold is adopted for assessment of effects quantified against NO₂ exposure. Of this, roughly two-thirds of the impact is attributable to quantification against PM_2.5 exposure (Table 26). To put these figures into some context, they are equivalent to 1.5% of current UK gross domestic product (GDP) under the baseline scenario. Of course, personal utility, as measured using the WTP paradigm for health, is not captured by GDP: this should be seen as a deficiency in the extent to which GDP describes wealth, rather than an indication that these monetised benefits are in some sense inferior or not truly a part of the national economy. 152

Total annual monetised impact for each scenario is shown in Figure 52. It should be stressed here that combining the impact for PM_2.5 and NO₂ is performed here for illustrative purposes and is likely to represent an upper bound to the valuation. The issues around the role of NO₂ in affecting health vis-à-vis PM_2.5 have been discussed in Chapter 8, and the possible overlap in effects from the two pollutants has been partially accounted for in the CRFs, on the basis of the most recent advice available from COMEAP at the time of this study.

FIGURE 52.

Total monetised impact, £M/year, under each scenario for PM_2.5 and NO₂ (no threshold) combined (this is illustrative, see Results).

The figure first demonstrates the extent to which impact in England dominates the assessment. This domination is largely a consequence of the difference in population relative to Scotland and Wales, although higher pollutant levels, particularly in London and south-east England, also play a role.

The figure also demonstrates the decline in impact relative to the situation had 2011 conditions persisted. Most of the forecast benefit is attributable to the baseline rather than the NRPO or LGHG scenarios.

Table 27 breaks down the benefits for the LGHG scenario relative to the baseline. Table 27 shows negative benefit (increased impact) for assessment relative to PM_2.5, through increased use of biomass for combustion. Benefits expressed against NO₂ exposure outweigh these effects related to PM_2.5 sufficient to generate an overall benefit for the LGHG scenario relative to the baseline of > £3B/year, equivalent to 0.2% of GDP, although this may be an overestimate. There is little difference between benefits regardless of whether a threshold is applied to NO₂ exposure, indicating that by far the greater share of exposure reduction under the scenario occurs in locations that exceed threshold.

Discussion

The results demonstrate that the economic impact of air pollution on health is substantial, equivalent to around 1.5% of GDP (accepting that GDP is an imperfect measure of ‘wealth’, as it considers only the ‘productive’ economy and ignores goods such as health and well-being). These results are broadly consistent with those generated in other assessments of air quality impact.

The results across different scenarios, and the comparison with 2011 conditions, show that the policy scenarios make a smaller difference to the benefits of improving air quality than policies already accounted for in the baseline scenario (as shown by the difference between the benefits of baseline relative to 2011, and the NRPO and LGHG scenarios relative to the baseline). Given that certain of the measures in place in the baseline scenario are designed specifically to improve air quality, this is not surprising, as it leaves reduced scope for future improvements. Given that results highlight that there remains a substantial level of impact beyond NRPO and LGHG, it is appropriate to ask whether or not the climate policies could be better targeted on air quality: the most efficient approach for society will be derived from analysis that takes all impact of climate policies, including the cobenefits and trade-offs of climate actions, as well as direct climate-related benefits.

There are several methodological issues that affect the magnitude of economic estimates. The following examples provide illustration of issues that affect the results in different directions:

• The valuations used in the UK are based on a limited literature and ignore the international literature (e.g. the OECD, 2012 meta-analysis152). Inclusion of studies from elsewhere could increase the valuations significantly. There is very little evidence to support the adoption of values lower than those used in the UK for the UK population.

• The 2% uplift is based on an assumed growth in WTP in line with growth in incomes. This in turn implies an elasticity of 1. In relation to mortality valuation, OECD152 recommends an elasticity of 0.8, which would reduce estimated values in the future compared with the current assumption. OECD also recommends use of an even lower elasticity, 0.4, in line with the recommendations of US EPA.

Overall, it is considered here that there is more bias towards potential underestimation of economic impact than to overestimation, given the size of the difference between mortality valuations accepted by UK government and the wider valuation literature. Although different assumptions would affect the magnitude of results, they would not affect the relationship between the scenarios. They would, however, obviously affect the economic appraisal of abatement measures.

Scientific conclusions

Limitations of the research

Although we have presented an advanced and detailed modelling study of the air pollution impact on health from climate policies, there are inevitable limitations to the work. We used the complex UKTM energy model as this represents a much more detailed method of generating energy scenarios than our original proposal. Because of this we were able to run only a limited number of scenarios. A wider range of pathways to the CCA would have potentially quantified a larger degree of health improvements in future years. The complexity of the UKTM model and the system we have built requires significant computer resources so that it is impracticable to undertake a range of sensitivity analyses around the economic parameters and energy and transport futures in the UKTM model.

Air quality modelling is always limited by several factors, the most important of which is the accuracy of the emissions inventory input. We improved the existing inventories using the most up-to-date information, but there are inevitably limitations to this knowledge. Equally, our understanding of the mechanisms of particle formation is developing continuously and, although we have used the best available chemical/physical mechanism of particle formation, there are still uncertainties involved here.

The health impact calculations are limited by the uncertainty in the numerical coefficients relating health outcomes to air pollutant concentrations, although to a degree we have allowed for this via the confidence intervals incorporated in the epidemiological studies. An important limitation is the extent to which the science supports an independent effect of NO₂ compared with PM_2.5 and the degree of overlap between the two pollutants in the association with adverse health outcomes. The review of the evidence by COMEAP, due to be published near the time of writing, had not appeared as this report was finalised.

Although improving the modelling scale down to 20 m in urban areas is an important advance in picking up exposure contrasts (particularly close to roads), the health impact methodology used is not, at this stage, able to take full advantage of this. In order to be able to use routinely available statistics on population and mortality by age group, concentrations were averaged up to ward level. Depending on how small-scale variations in population and mortality line up with variations in pollutant concentrations (particularly NO₂), results could differ if finer-scale inputs were used for population and mortality as well as concentration.

Uncertainties

The project’s contribution to advances in knowledge

This project has, for the first time in the UK, delivered a sophisticated tool to enable the explicit calculation of public health impact arising from future energy strategies in GB using a state-of-the-art air quality model with an energy systems model used to inform government policy on greenhouse gas mitigation. This represents a major improvement over previous approaches to the impact of greenhouse gas emissions. Our work has established a method of calculating public health impact of air pollution resulting from climate policies through the full ‘impact pathway’ approach rather than the cruder, more approximate, ‘damage cost’ approach. The latter approach has been used to date by government in the UK to appraise climate policies; it relies on simply assigning a monetary value to a unit of air pollution emission. Consequently, there is no explicit calculation of pollution concentrations in the air, or of the impact on mortality and morbidity. Our system now allows that to be done in a linked system beginning with the economic model of the British energy system, through a sophisticated air quality model to a detailed life table model for calculating impact on health, on exposures in socioeconomic classes and for calculating economic impact. Moreover, the system we have developed allows this impact to be calculated at the finest spatial resolution yet achieved in GB, in which we model the rural areas at 10 km and major urban centres at 2 km or as fine as 20 m.

The science of air quality and of PM has continued to evolve during the life of the project. During the study it was necessary for us to incorporate emerging research on the sources of PM from cooking sources that were not previously included in emission inventories in GB. We have also built on King’s College London’s expertise in understanding the contribution of wood (biomass) burning to air quality to improve the inventory of emissions from this source. We were also able to enhance our model to treat the major British cities at a spatial resolution of 20 m, where previously we had been able to do this only for London. We have evaluated this improved air quality model and shown it to behave well against accepted criteria.

During the course of the project, we formed a collaboration with the Institute for Sustainable Resources at UCL to allow us to link the UKTM energy systems model with our air quality model and our health impact capability. This formed a major advance and the link now establishes a credible system to assess the impact of energy futures and climate policies in GB.

Discussion

There are several important policy messages which arise from this project. The CCA target, in principle, offers a great opportunity to make very large reductions in air pollution emissions as the UK energy system is decarbonised. However, the PM_2.5 emissions from the large increases in residential and CHP biomass use and the increase in non-exhaust PM emissions from transport in the two CCA-compliant scenarios we have studie mean that PM_2.5 and PM₁₀ concentrations do not fall as much as they would otherwise have done without the biomass use. This increase in biomass use has resulted in the finding that the CCA-compliant scenarios result in more life-years lost than the baseline scenario which incorporates no further climate action beyond that already in place.

Solutions to improve air quality impact on health could include measures to discourage the use of biomass in small installations, or to increase the stringency of the emission limits in the Ecodesign Directive. The related study by Lott et al. ,33 using the damage cost approach, carried out a calculation including damage costs for biomass use to attempt to account for the public health impact. This succeeded in reducing the use of biomass in the hypothetical calculation. In reality, measures to discourage biomass use would probably be best delivered through revisions to the renewable heat incentive. In terms of improving air quality and minimising the impact on public health, wood burning, if it were to be used at all, would be best deployed in large, efficient power stations rather than small-scale domestic or CHP use. Fiscal measures in the renewable heat incentive to encourage this shift would have benefits to public health without necessarily compromising the achievement of the CCA target.

Non-exhaust emissions of PM₁₀, and to a lesser extent PM_2.5, are projected to increase significantly by 2050 as traffic activity increases. The precise agents in tyre and brake wear and resuspended dust responsible for the potential toxicity of these emissions are, as yet, unclear, so reformulation of these products would need to await more clarification from toxicological research. However, in the meantime, the obvious solution to ameliorate potential impact from all emissions from road transport here would be to discourage traffic use, particularly in urban centres.

On the positive side, electrification of the road transport fleet results in large reductions in the potential adverse impact on health from NO₂ and potential compliance with legal standards. This will also have benefits for PM concentrations and will, to a limited degree, offset the impact of any continued increase in biomass use.

The use of economic appraisal provides a mechanism for assessing the efficacy of measures for further action, permitting direct comparison of costs and benefits of measures, and enabling collation of a variety of different types of effect. As shown above, economic damage associated with air pollution in the UK is substantial and will remain so over the period covered by the scenarios considered here. It is noted that the UK approaches to valuation appear highly conservative compared with assumptions followed in the wider international literature.

Further research with the linked UKTM and our air quality model, CMAQ-Urban, could investigate other possible scenarios which achieve the CCA target and which could minimise the problem with residential and CHP biomass use and the impact of non-exhaust road transport emissions.

Recommendations for future research

The system we have developed links together a sophisticated energy system model – used by government in the UK – with a detailed chemical–transport model for air quality, health impact calculations and assessments of exposures and impacts in different socioeconomic classes. It also allows the monetary valuation of this impact on health. Because of the complexity of the system we have been able to run only a small number of scenarios and, although we have demonstrated some significant issues for future climate change mitigation measures, there is still scope to address optimal pathways for attaining the CCA target by minimising the impact on air quality and public health.

The work has shown that trends in different fractions of the atmospheric particle mix may be different in future. Primary particles (containing known carcinogens) may increase, whereas secondary particles may decrease. This highlights the importance of studies to elucidate the differential toxicity of different particle fractions.

The work has shown that, with increased penetration of ultra-low and zero-emissions road vehicles, concentrations of NO₂ will decrease by large amounts. The precise role of NO₂, compared with that of PM_2.5 and other pollutants, in affecting human health is still uncertain. More clarity is needed here before any health benefits from reductions in NO₂ can be confidently quantified.

The effects of long-term exposure to air pollution on mortality generally dominate cost–benefit analysis, but a full investigation of the health impact would involve quantifying the potential effects on a wider range of health outcomes. In addition, further sensitivity analyses on the data inputs and assumptions regarding CRFs (e.g. effect modification) would be helpful. There is a need to explore how using population and mortality inputs at a finer geographical scale affects the result. More meta-analyses of epidemiological studies on PM_2.5 and NO₂ will also be useful.

Contributions of authors

Martin L Williams (Professor of Air Quality Research, King’s College London) was responsible for the concept and design of the study, advice and direction of scenario building and modelling, and overall production of the report.

Sean Beevers (Senior Lecturer, Air Quality Modelling, King’s College London) was responsible for the management and direction of air quality modelling, transforming UKTM output to give air pollutant emissions and contributing to report writing.

Nutthida Kitwiroon (Senior Air Pollution Scientist) was responsible for setting up CMAQ air quality model runs and running the model, plotting and analysing the air quality model results.

David Dajnak (Principal Air Quality Scientist, King’s College London) was responsible for health impact assessment and life table calculations.

Heather Walton (Senior Lecturer, King’s College London) was responsible for the direction of, and contributing to, the health impact assessment, life table calculations and contributing to report writing.

Melissa C Lott (Doctoral Researcher, UCL) was responsible for UKTM scenario design and production.

Steve Pye (UCL) was responsible for UKTM scenario design and production.

Daniela Fecht (Imperial College London) was responsible for the social deprivation analysis.

Mireille B Toledano (Imperial College London) provided advice on the design of the social deprivation analysis.

Mike Holland (Economic Consultant, Ecometrics, Research and Consulting) provided advice on economic impact.

Publication

Williams ML, Lott MC, Kitwiroon N, Dajnak D, Walton H, Holland M, et al. The Lancet Countdown on health benefits from the UK Climate Change Act: a modelling study for Great Britain. Lancet Planet Health 2018;2:e202–13.

Data sharing statement

Data used, or referred to, in this report may be obtained by contacting the corresponding author (Professor Martin Williams, King’s College London).

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the PHR programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the PHR programme or the Department of Health and Social Care.

Spatial scale

The spatial scale of the air pollution modelling can have an important effect on the estimation of health impact. The larger the grid squares in the model, the more the pollutant concentrations are averaged out over areas where emissions may be low, or even absent, so the modelled concentrations are lower than is the case for finer grid sizes. This effect can be particularly marked for single point sources, such as industrial plants, or for roads. The effect of the latter can be seen in the results at different scales for NO₂ (which derives mainly from traffic in UK cities) in Table 10. Although we can now model at a spatial resolution as low as 20 m, the spatial resolution used in estimating health impact is usually limited by the spatial resolution of the health outcome and mortality data. Here we used mortality data at GB local authority level, which are obviously much larger than 20 m, hence our use of 2-km resolution in the major urban areas for the life table calculations.

Long-term exposure to PM_2.5 and all-cause mortality: concentration–response function, population subgroup, lag from exposure to effect and counterfactual

The COMEAP recommended CRF for long-term exposure to PM_2.5 and all-cause mortality. 99 The hazard ratio was 1.06. A plausibility interval, derived from expert elicitation of wider sources of uncertainty, was also presented. The recommendation was to be applied to anthropogenic PM_2.5 (i.e. non-anthropogenic PM_2.5 concentrations were used as the counterfactual), and to the population aged ≥ 30 years. A lag between a change in concentration and the resulting effect on mortality was recommended, based on an expert elicitation by the US EPA. 98

The HRAPIE report95 central estimate was the same.

Long-term exposure to NO₂ and all-cause mortality: concentration–response function, population subgroup, lag from exposure to effect and counterfactual

The report on the impact of NO₂ on health is due to be published in 2017, but at the time of writing the only material on which we could base our calculations was contained in the interim COMEAP statement and minutes of the meeting on 24 February 2016. 100 There are issues to be resolved on the quantification, as noted above, and also the question of the additivity of the NO₂ and PM_2.5 effect estimates. Further work will possibly be necessary when the COMEAP conclusions are published.

Long-term exposure to PM₁₀ and chronic bronchitis (not included for future sensitivity analysis work)

The overall conclusion of a report by COMEAP158 did not support an association between PM₁₀ on chronic bronchitis; the CRF was provided for sensitivity analysis only. This was because associations were demonstrated in some studies. The CRF was in terms of prevalence of chronic bronchitis rather than incidence as in HRAPIE. 95 The studies showed reversibility of reports of the disease, suggesting elements of exacerbation rather than incidence of disease. It was noted that the definition of the outcome in epidemiological study associations is imprecise and only provides a minimum indication of severity and duration. It is distinct from a clinical definition of chronic bronchitis in which the information used to make a diagnosis may be both more extensive and less standardised. It therefore cannot be assumed that the associations necessarily represent incidence of serious disease.

The WHO’s recommendation on chronic bronchitis was based on one of the studies included in the COMEAP review above.

Future work will also look at other issues such as long-term exposure to NO₂ and asthma prevalence.