Air Quality Alerts, Health Impacts, and Adaptation Implications Under Varying Climate Policy

Abstract

Without emission reductions, climate change may increase ozone and PM_2.5 air pollution in the United States; however, we do not know how this will affect air quality alerts that prompt people to stay indoors. Here, we use an integrated modeling framework to find distributions of daily Air Quality Index (AQI) during the smog season at the start, middle, and end-of-century. Considering natural variability, climate change may cause air quality alerts to double (increase by a factor of 2 ± 0.2) by 2100. Days when both ozone and PM_2.5 exceed alert thresholds quadruple (4.3 ± 1.2). More than 100,000,000 (±45,000,000) people experience mean air pollution deemed “Unhealthy for Sensitive Groups”, a growth of 7 (±3) times compared to 2000. If people follow alerts by staying inside, they reduce exposure to outdoor-generated pollutants. Their health benefits are similar whether the alert is caused by ozone or PM_2.5. Senior (age 65+) populations receive much higher benefits per day by adapting (e.g., 95CI for PM_2.5: $4.60 to $147) as young adults (age 18–35; 95CI: $0.15 to $4.22)more than 45 times higher on average. This disproportionate impact requires targeted messaging and guidance, especially as climate-related risks rise.

Introduction

Outdoor air pollution increases morbidity , and premature mortality , in exposed populations. In the United States (US), the Air Quality Index (AQI) reports the “level of concern” associated with different pollutants. AQI is calculated for each pollutant, with the maximum among pollutants reported as the AQI. AQI is categorized with labels ranging from “Good” (0–50) to “Hazardous” (301–500) (Table S1). Alerts may be issued when the AQI exceeds certain thresholds, typically 100 (“Unhealthy for Sensitive Groups”) or 150 (“Unhealthy”). , Ozone and PM_2.5 (particulate matter with an aerodynamic diameter less than 2.5 μm) are responsible for over 99% of AQI alerts.

Climate change is expected to make air pollution worse in the US, , an effect termed the “climate penalty”, but little is known about its effect on AQI alerts. Climate change can increase drought, heat waves, and wildfires, which can contribute to extreme outdoor ozone and PM_2.5 concentrations. ,

Natural variability in the climate system makes it challenging to evaluate the effect of the climate penalty on air quality alerts. Natural, or unforced, annual and multidecadal changes in weather can obscure the effect of climate change and climate policy on air pollution. Filtering out this variability requires multiyear, multiple initial condition ensembles. One such study projected that, even with some emission reductions to combat climate change, the noncompliance zone for areas exceeding the National Ambient Air Quality Standard (NAAQS) for ozone may expand by 8.5 million residents by midcentury. We know of only one study to evaluate these increases in terms of AQI, and it only considered one pollutant. Similarly, it used a multiple initial condition ensemble to isolate the effect of the climate penalty, finding that days with AQI over 100 due to PM_2.5 could quadruple by 2100 without emission reductions. Even under climate policy, there were significant net benefits from adapting to air pollution.

Individual adaptation is recommended by AQI messaging. In a given year, approximately 37% of Americans are aware of poor air quality, of which 57% change their behavior. This implies that approximately 20% of Americans adapt to poor air quality. The most common action in response to poor air quality is spending less time outdoors, , which we term “avoidance adaptation”. Other strategies to reduce exposure to outdoor pollution, though not mentioned with standard AQI guidance, include improving building mechanical system filtration, , wearing masks, and using portable filtration devices.

Limited research examines the exposure reductions achieved by following AQI guidelines. Evidence supporting six recommended behavior changes has been deemed “lacking”, leaving studies to resort to “plausible estimates” of exposure reduction due to a lack of quantitative values. Two recent studies quantified the exposure reductions and health benefits associated with avoidance adaptation. Buonocore et al. (2021) modeled adaptation by removing exposure to the highest waking hour of outdoor concentration in the over-65 population across three US cities. Sparks et al. (2024) incorporated ambient pollution exposure in both indoor and outdoor environments by representing infiltration, but only for residential environments, and only for PM_2.5.

Adaptation in the over 65 (senior) population is especially important because they experience most of the health impacts of air pollution. Both PM_2.5 and ozone exposure have been significantly associated with increased risk of mortality for older adults, even at concentrations below NAAQS standards. Populations are aging, which is expected to amplify effects of the climate penalty. The elderly are part of the “sensitive groups” who are advised to take protective actions at lower pollution levels (AQI > 100) than the general population (AQI > 150). However, seniors may merit different thresholds, communication strategies, and advice. They receive information differently than their juniors, with most (85%) learning of air quality alerts via television, and between 34% and 50% receiving alerts via newspaper. They may also need specialized guidance, as they spend less time outside, on average, with 47% spending no time outside (based on our analysis of American Time Use Survey Data from 2008 to 2023).

In this study, we investigate how AQI mean values and alerts could change through the 21^st century due to the climate penalty. We evaluate modeled hourly concentrations of ozone and fine particulate matter over the contiguous United States during the smog season at the start, middle, and end-of-century. To address natural variability and isolate the effect of the climate penalty, we use a multidecadal initial condition ensemble with constant anthropogenic pollutant concentrations driven by a varying climate. We identify which pollutants are responsible for changes in AQI alerts and which contribute most to health impacts. We improve upon previous adaptation studies by providing simple, quantified estimates of the reduction in exposure to ozone and PM_2.5 of outdoor origin achieved by avoidance adaptation. We compare the benefits of adaptation when responding to AQI alerts driven by ozone or PM_2.5, and identify how much of the future economic health burden can be offset by adaptation. Lastly, we quantify how the benefits of adapting vary with age. This work identifies the size and makeup of populations at increased risk for air quality alerts, and informs targeted thresholds and messaging more closely aligned with individual risk.

Methods

Modeling Future Concentrations

We use future ozone (daily 8 h maximum average) and PM_2.5 (24 h mean) concentration projections based on the Massachusetts Institute of Technology (MIT) Integrated Global System Model (IGSM) − , (Figure S1–S2). The MIT IGSM framework links the Economic Projection and Policy Analysis (EPPA) model, a global computable general equilibrium economic model, to the MIT Earth System Model (MESM,), an earth system model of intermediate complexity. It yields global climate fields in response to a global carbon price. Those fields are linked to the National Center for Atmospheric Research (NCAR) Community Atmosphere Model (CAM) and its atmospheric chemistry component. Resulting pollutant concentrations thus reflect population and economic growth within the EPPA model (Figure S3–S5). Using this self-consistent framework means differences between scenarios are due to the effects of policy, rather than different models or implementations. This framework was previously used to model pollutant concentrations at 2° by 2.5° resolution over the contiguous US under climate change at the start, middle, and end of the 21st century (labeled 2000, 2050, 2100) under three policy scenarios: (1) a reference scenario with no climate policy enacted (REF); (2) a scenario with the global-average temperature increase limited to 2.5 °C above preindustrial (2.5C); and (3) a scenario with the increase limited to 2 °C (2C). − The mean air quality for each period was simulated using three decades and five initializations of the climate model (details in Paltsev et al., 2015). Anthropogenic pollutant emissions were held constant with start-of-century values in all scenarios to isolate the effect of climate change on ambient pollutant concentrations. Biogenic emissions responded to climate change, but wildfires were not modeled explicitly. Modeled ozone concentrations were bias-corrected to address the known high bias in model-based ozone concentrations, using observed ozone concentrations. That bias correction was limited to May first through September 30th, which typically contains the highest ozone levels in the U.S. , Thus, we focus here on alerts in the smog season. Modeled national population-weighted mean PM_2.5 concentrations in the historical period are within 7% of measured values. Projected concentrations of ozone and PM_2.5 agree with similar studies using different models.

Calculating AQI Values

We calculate AQI values for modeled ozone and PM_2.5 data following US Environmental Protection Agency (EPA) methodology. Ozone and PM_2.5 have concentration breakpoints for each AQI category (Table S1). Between each breakpoint, the AQI value for a given pollutant p (I _p ) is calculated based on EPA’s Technical Guidelines for AQI as

$$I_p=\frac{I_{Hi}-I_{Lo}}{B{P}_{Hi}-B{P}_{Lo}}(C_p-B{P}_{Lo})+I_{Lo}$$ 1

where C _p is the concentration of pollutant p (24 h mean concentration in μg/m³ for PM_2.5 and daily 8 h maximum ppb for ozone), BP _Hi is the concentration breakpoint greater than or equal to C _p , BP _Lo is the concentration breakpoint that is less than or equal to C _p , and I _Hi and I _Lo are the AQI values corresponding to BP _Hi and BP _Lo respectively. For example, ozone has an I _Lo of 101 at a BP _Lo of 71 ppb. Thus, a daily 8 h maximum ozone of 71 ppb yields an AQI of 101, triggering an alert for sensitive groups.

Calculating Health Impacts Due to Changes in Air Pollution

We used BenMAP-CE v1.5 to estimate the health effects of changes in exposure to pollution of outdoor origin, following US EPA regulatory analysis. The change in health outcomes (ΔY) is calculated as

$$\mathrm{\Delta }Y=Y_0\times Pop\times (1-\exp (-\beta \times \mathrm{\Delta }{\mathrm{C}}_{\mathrm{p}}))$$ 2

where Y ₀ is the baseline incidence rate of a given outcome, Pop is the affected population, β is the risk coefficient, and ΔC_p is the change in pollutant concentration between two scenarios.

We employed different health impact functions (eq ) to represent morbidities and mortalities, each with their own β value. We used EPA-provided BenMAP-CE compatible configuration files for quantifying and valuing effects of PM and ozone (details in Table S2).

We compared outcomes across age groups with a modified version of eq , in which each parameter reflects an age group. We examined different age ranges within the 2005 population (<18, 18–35, 36–64, and ≥ 65). We used population data from 2005 to represent the start-of-century to align with the base year of our economic model. BenMAP-CE provided population data for this year partitioned by age range, which gave the number of people in each group and grid cell (Pop) based on US Census data. Y₀ varied by population group (including age) for mortality. We used β values appropriate to the age groups (see Table S2). We used two comparisons to find differences in pollutant concentrations (ΔC_p). First, we compared future years to modeled historical years under baseline behavior. Second, we compared adaptation behavior to regular behavior for a given target year. Each of these are explained in the following sections.

Valuing Changes in Health Burden

We valued the change in health outcomes by eq :

$$\mathrm{\Delta }B=\sum _i{C}_i\times {\mathrm{\Delta }Y}_i$$ 3

where ΔB is the change in economic health burden, C _i is the cost associated with health outcome i and ΔY _i is the change in incidences of health outcome i. We specify the cost of a health outcome following EPA Guidelines for Preparing Economic Analyses. We use US EPA’s preferred measure for damages and benefits, namely, willingness to pay (WTP) to reduce the risk of a health outcome. WTP is used to determine the economic cost or benefit of changes in morbidity and premature mortality risk. To value premature mortality, we use the Value of a Statistical Life (VSL), which represents WTP extrapolated to a risk reduction of 1. We use the US EPA’s recommended VSL value of $8.7 million (2015 USD). The second way to quantify C _i is to use the cost-of-illness (COI) of treating an adverse health outcome. COI estimates used are from US EPA. Valuation methods for each morbidity and mortality outcome are presented in Table S2.

When calculating future health impacts, we change concentrations only, keeping the population (Pop), baseline risks (Y₀) and costs (VSL or COI) constant. This is designed to isolate the effects of the climate penalty on concentrations. In practice, future concentrations are generated by future populations in future economies, in which pollutant emissions change alongside greenhouse gas emissions. However, this approach conflates changes in pollution concentrations driven by the climate penalty, population characteristics, and economic growth. For this work, we isolated the effect of atmospheric composition changes on health impacts and valuation.

Quantifying Exposure Reduction from Adaptation

A person’s daily exposure (E _{d p} ) to an ambient air pollutant p is calculated as

$$E_{d_p}=\sum _m{L}_m\times I_m\times {F_{INF}}_{m_p}\times C_p$$ 4

where L _m is the location fraction, or the fraction time spent in microenvironment m, I _m is the ratio of the activity intensity that occurs in microenvironment m compared to the rest of the day, $F_{{INF}_{m_p}}$ is the infiltration factor, or ratio of indoor pollution to outdoor pollution for a given pollutant p in microenvironment m, and C _p is the outdoor concentration of pollutant p. eq is similar to the microenvironment exposure model used in prior studies of total exposure to air pollution. ,

To determine location fraction (L _m ), we split people’s time into four different microenvironments: Residential, Commercial, Vehicle, and Outdoor. We used American Time Use Survey (ATUS) data from 2008 – 2019 to identify the portion of time that people spend in each microenvironment (Location Fraction) in a similar way to Hoehne et al. (2018) (see Table S3–S5). We calculate activity intensity by assigning a metabolic equivalent value (MET) to ATUS data, as in Tudor-Locke et al. (2011) (Table S6). To determine the average activity intensity in each microenvironment (I _m ), we use the ratio of the microenvironment’s average activity intensity value to the daily average activity intensity value.

We determine the equilibrium fraction of ambient particles that penetrates indoors and remains suspended (F_INF) based on previous estimates − (details in Table S7). For PM_2.5, we determined Residential F_INF using the National Renewable Energy Lab (NREL) ResStock air changes per hour (ACH) values as

$$F_{INF}=\frac{P\times ACH}{ACH+\beta }$$ 5

where P is the fraction of PM_2.5 that penetrates the building envelope (0.73, Zhao and Stephens), ACH is the national average air changes per hour based on ResStock data, and β is the settling velocity of PM_2.5. Commercial and vehicle F_INF values were gathered from previous studies. , Outdoor F_INF is 1, as there is no protection from pollution while outside.

For ozone, previous studies use indoor-outdoor ratio (I/O) as a proxy for F_INF because indoor generation is low. , We select values to represent adaptation conditions, when doors and windows should be closed. We derive F_INF values for each microenvironment from Lai et al. (2015), who measured ozone I/O and found a I/O of 0.09 for infiltration, representative of residential and vehicle microenvironments. For commercial, we used their mechanical ventilation value of 0.19. These values are within the ranges reported by other studies.

The values of all parameters on typical nonadaptation and adaptation days are presented in Table . We used national average values for all exposure parameters, primarily given the lack of spatially varying infiltration rates for ozone. However, in sensitivity analysis, we consider the spatially varying differences in location and activity intensity on adaptation effectiveness.

1. Fraction of time spent, activity intensity ratio, and infiltration factors for ozone and PM_2.5 on a typical non-adaptation day, and a typical adaptation day. Values are based on national average from American Time Use Survey data.

Microenvironment	Location Fraction		Activity Intensity Ratio		FINF
Behavior	L m	L m	I m	I m	Ozone	PM2.5
Residential	0.718	0.746	0.923	0.958	0.09	0.45
Commercial	0.197	0.197	1.032	1.059	0.19	0.35
Vehicle	0.057	0.057	1.341	1.286	0.09	0.6
Outside	0.028	0	1.853	N/A	1	1

We defined adaptation as shifting all time in the Outside microenvironment to the Residential microenvironment. This leads to changes in the location fraction and activity intensity ratio values, which lead to a new (reduced) personal daily exposure E _{d p} . E _{d p} is calculated, similar to eq for E _{d p} , as

$$E_{d_p}^{\prime }=\sum _m{L}_m^{\prime }\times I_m^{\prime }\times {F_{INF}}_{m_p}\times C_p$$ 6

Adaptation makes the Location Fraction (L _m ) value for Outside 0 and adds that time to Residential. We kept the activity intensity for the shifted activity the same, such that adaptation does not reduce physical activity, as suggested by the American Thoracic Society. This increased the activity intensity ratio (I _m ) for the Residential microenvironment. The infiltration factors (F_INF ) remained the same between nonadaptation and adaptation days. Location fraction, activity intensity ratio, and infiltration factors under adaptation are shown in Table .

Relating Changes in Exposure to Equivalent Changes in Outdoor Concentration

Health impacts are estimated for differences in outdoor concentrations (ΔC_p). To translate changes in exposure to changes in outdoor concentration used in health impact functions, we first defined the effectiveness of adapting to a given pollutant p as the ratio between E _{d p} and E _{d p} . We defined this value as Ψ _p :

$${\mathrm{\Psi }}_p=\frac{E_{d_p}^{\prime }}{E_{d_p}}$$ 7

Using the location fraction (L _m and L _m ), activity intensity ratio ( I _m and I _m ), and infiltration factors (F_INF,ozone and F_INF,PM2.5 ) in Table , we find Ψ _ozone = 0.30 and Ψ _{PM 2.5} = 0.06. This signals it is much more protective on average, by a factor of 5, to adapt to ozone than PM_2.5.

During the study period, there are n nonadaptation days and a adaptation days, which we index by j (nonadaptation days) and k (adaptation days). We use (1 – Ψ _p ) to convert the outdoor concentration on a given adaptation day k (C _{p k} ) to a concentration that would give the equivalent exposure that one would have on a nonadaptation day (C _{p k} ):

$$C_{p_k}^{\prime }=(1-{\mathrm{\Psi }}_p)\times C_{p_k}$$ 8

Over the course of the study period, we can aggregate the effects of individual adaptations to weight the effect on average exposure, estimated as

$$C_p^{\prime }=\frac{\sum _{j=1}^n{C}_{p_j}+\sum _{k=1}^a{C_p^{\prime }}_k}{n+a}$$ 9

where C _p is the equivalent seasonal average pollutant concentration after adaptation, C _{p j} is the outdoor pollution concentration on nonadaptation days j, C _{p k} is the equivalent outdoor pollution concentration after adaptation, on adaptation days k, n is the number of nonadaptation days in the smog season, and a is the number of adaptation days in the smog season. Values for n and a are calculated based on local concentrations in each grid cell.

We assume that populations adapt when their local AQI is higher than a selected threshold, simulating what we term an AQI alert. To calculate the maximum benefit from adaptation, we assume everybody in a grid cell adapts on each “alert day”, which thus prompts an adaptation day. While full compliance is unlikely, , we use it to test the full potential benefits of adaptation.

We calculate ΔC _p as the difference between C _p and C _p as

$$\mathrm{\Delta }{C}_p=C_p-C_p^{\prime }$$ 10

ΔC _p is used in eq to calculate the health effects of adaptation.

Results

Average AQI values over the summer smog season increase throughout the contiguous US under the REF (no climate policy) scenario (Figure ). At the start-of-century, 79% (68% – 90%) of the population has a mean AQI value within the “Moderate” AQI category (the range reflects the standard deviation over 1981–2010 across five ensemble members). The remaining 16% (6% - 27%) primarily enjoys “Good” air, while very few (5% (0% - 19%)) have mean air quality in the “Unhealthy for Sensitive Groups” category (Figure (d)).

1.

Average Air Quality Index (AQI) over the summer smog season. Results are shown to represent start-of-century (REF 2000) (a), midcentury (REF 2050) (b), and end-of-century (REF 2100) (c) reference concentrations (i.e., without emission reductions). Population is held constant for the year 2005. Subplot (d) shows the number of people experiencing a mean smog season (May 1st – September 30th) AQI in the “Good”, “Moderate”, or “Unhealthy for Sensitive Groups” categories. Error bars reflect the standard deviation derived from 150 annual simulations for each policy scenario and target year (i.e., five initializations of three decades around start, mid, and end of century, including 1981–2010, 2036–2065, and 2086–2115).

At the national scale, these results show good agreement with measured AQI (Figure S6–S7). We compare the distribution of population-weighted AQI values across our ensemble at the start-of-century with that reported by the US EPA using ground-based observations for the years 2000 through 2005. In both cases, more than 80% of days have AQI < 100 and 96% have AQI < 150, respectively. The model shows a slight overprediction in alerts (17% vs 11% of days with AQI > 100; 4% vs 3% of days with AQI > 150), likely due to a known high bias in ozone (Figure S8–S9). Spatially, smoothing due to coarse model resolution means missing some regions with a mean AQI in the Unhealthy for Sensitive Groups category. Based on available data, this includes several counties in California, including some with an AQI just exceeding the threshold of 100 (Figure S10).

By midcentury, under the reference scenario, the climate penalty can expose more people to a mean smog season AQI value in the “Unhealthy for Sensitive Groups” category. The proportion of the population in this category grows to 15% (±12%) – triple that of start-of-century. The proportion of the population with a mean AQI value in the “Good” category slightly decreases to 15%. Consequently, the “Moderate” AQI fraction decreases to 70% (Figure (d)).

In the end-of-century reference scenario (REF 2100), the transition between mean smog season AQI values in the “Moderate” to “Unhealthy for Sensitive Groups” categories continues (Figure (d)). One hundred million (±45,000,000) people would have an average AQI value between May first and September 30th that is “Unhealthy for Sensitive Groups”. This means that approximately a third of the US population would have an average AQI recommending action for children, the elderly, and those with underlying health conditions. There is a factor of 7 increase (±factor of 3) in the number of people with mean AQI values in this category between REF 2000 and REF 2100 (14.3 M vs 99.7M). The percent of the population with mean AQI values in the “Good” category 15% (±9%) and the proportion of people in the “Moderate” category decreased from 79% (±11%) to 49% (±12%). Over the century, the climate penalty increases the likelihood that regions in the Eastern US and California shift from a mean smog-season AQI in the “Moderate” to “Unhealthy for Sensitive Groups” category (Figure (a)­(b)­(c). Very few regions switch between the “Moderate” and “Good” AQI categories.

Episodes of high air pollution during the smog season are projected to affect more people for longer periods of time if emissions are not reduced (Figure ). Figure shows the population-weighted likelihood of an alert under the REF scenario for the start, middle, and end of century. That likelihood is calculated as the proportion of days on which ozone, PM_2.5, or both pollutants exceed their AQI thresholds, over a distribution of 150 annual simulations for each target year (e.g., five initial conditions each of 2036–2065 for 2050). It is population-weighted across the contiguous US with constant year 2005 population.

2.

National population-weighted frequency of air quality alert days for different pollutants and threshold levels from May to September. Frequency is calculated as the proportion of time ozone, PM_2.5, or both pollutants exceed the defined threshold, over the contiguous US, and across 150 annual simulations for each target year (e.g., five initial conditions each of 2036–2065 for 2050). Population is constant at 2005 levels. Upper row plots (a-c) correspond to AQI values reaching the “Unhealthy for Sensitive Groups” category, which is defined as AQI values >100. Lower row plots (d-f) correspond to AQI values in the “Unhealthy” category (AQI values >150). Plots show the distribution of AQI alerts over the smog season for reference scenario concentrations at start-of-century (REF 2000) (a,d), midcentury (REF 2050) (b,e), and end-of-century (REF 2100) (c,f).

Figure (a-c) show alerts for sensitive groups (issued when AQI > 100). At the start of the century, these alerts are primarily driven by ozone, with the frequency of PM_2.5-driven alert days increasing throughout the season and peaking around mid-August. The population-weighted chance of an “Unhealthy for Sensitive Groups” AQI peaks around 30% from approximately July first through mid-August. There is a clear rise in the probability of an alert at the start of the season and a decrease after the peak, though the chance of an “Unhealthy for Sensitive Groups” alert is higher at the end of the smog season than the beginning.

We find that most alerts (77%) are driven by ozone exceedances. Of the remainder, 17% are caused by PM_2.5 and 7% are caused by both pollutants exceeding the AQI threshold. This contribution of ozone is likely overestimated due to a positive bias in modeled ozone (Figure S8); however, the proportion of days with high AQI based on PM_2.5agrees well with observations (Figure S9), which suggests a good estimate of days when both pollutants exceed their thresholds.

The climate penalty increases the odds of an alert for sensitive groups. By end-of-century, the national average chance of an “Unhealthy for Sensitive Groups” alert reaches almost 50% and remains elevated from late-June through late-August. This implies that, without emission reductions, vulnerable people could face nearly even odds of an alert throughout the smog season. Additionally, multipollutant risks rise, with more days in which both ozone and PM_2.5 trigger an alert (“Both” in Figure ).

AQI alert chances for the general population grow by a proportionally larger amount than for the sensitive population. This is shown by comparing the alert probabilities in Figure (a–c) (“Unhealthy”) with those in Figure (d–f) (“Unhealthy for Sensitive Groups”). At start-of-century, these alerts are rare through May and only increase in June, with the probability of an alert peaking at 10% in early August. Under the REF 2050 scenario, (Figure (e)), ozone-driven “Unhealthy” alert days increase slightly, but there is a notable increase in PM_2.5-driven alerts at the height of the smog season, and “both” alerts rise in September. In 2100 (Figure (f)), the population-weighted average chance of an “Unhealthy” AQI day peaks over 20%, which is more than double the peak at the start-of-century period. The peak occurs about 2 weeks later and lasts for a longer portion of the season than at start-of-century.

Climate change may add an average of 28 (±5) alerts days for sensitive groups during the smog season (Figure (b)), reflecting a doubling of these alerts compared to start-of-century (Figure S10). Alert days with both high ozone and PM_2.5 rise from 2 (±1) at start-of-century to 8 (±2), a difference of 6 days and an increase of 300% (Figure S11).

3.

Alerts for sensitive groups (air deemed “Unhealthy for Sensitive Groups”, with AQI > 100) under future concentrations and future health burden for full population (a) Annual mean increases in alerts in end-of-century (“2100”) concentrations compared to start-of-century (“2000”) concentrations in the absence of climate policy. (b) National population-weighted change in number of alerts during the smog season due to the climate penalty. Distribution is across 150 annual simulations for each policy and target year. (c) Decreases in alerts with end-of-century concentrations for the 2C climate policy compared to no policy. (d) Bar charts showing the health-related economic burden for scenario concentrations relative to the start-of-century period (in billions of 2015 USD). Health effects are broken down by responsible pollutant (ozone or PM_2.5). Results are based on projected concentrations only and are not adjusted for population or economic growth from their base year of 2005.

Climate policy that limits global warming to 2.5 or 2 °C reduces the number of alerts for sensitive groups (Figure (b)) by essentially arresting the rise in alerts at midcentury. By end-of-century, the 2.5C policy reduces AQI alerts by 30% (±8%) compared to the REF concentrations. The 2C policy produces very similar outcomes. Both climate policies reduce concurrent alerts from 7 (±2) to 3 (±1).

The rise in “Unhealthy for Sensitive Groups” alerts shows a similar spatial pattern to Figure , with increases in California and Eastern US. The latter sees up to two months of additional alerts per year (Figure (a)). These same areas see the greatest benefit from climate policy (Figure (c)), with some avoiding over 30 additional AQI alert days annually with the 2C concentrations compared to REF.

However, even under climate policy, the climate penalty can cause a rise in air quality alerts and associated health burden. Figure (d) shows the additional health-related economic burden from outdoor air pollution under each scenario compared to start-of-century. While ozone creates more alerts, exposure to PM_2.5 of outdoor origin comprises 94% of the economic burden (Figure (d)).

The two climate policies have similar effects on alert days and economic burden. Though the 2C scenario shows fewer alerts and a lower health burden than 2.5C on average, the ranges of alerts overlap considering natural variability (Figure (b)). We filter out natural variability by using the ensemble mean to estimate health burdens. Nonetheless, uncertainty in health responses and their economic valuation mean that the policies’ burdens overlap with one other and with start-of-century conditions due to uncertainty in health and economic valuation, with a 95th confidence interval (95CI) in the change in burden of -$4.0 to $500 billion (USD 2010) for 2.5C and -$8.0 to $290 billion (USD 2010) for 2C (Table S9).

Avoidance adaptation reduces morbidities and mortalities. Figure shows the mean health-related economic benefit of avoidance adaptation based on 5000 Monte Carlo simulations representing uncertainty in health outcomes and economic valuation (95CI for all scenarios are statistically significant and shown in Table S10). Specifically, it shows the value of the reduced risk of morbidity and mortality achieved by reducing exposure through remaining indoors during alerts (eq ). It assumes that the full population experiencing an alert for “Unhealthy” air reduces their outdoor time to zero (converting it to time in residences). Full compliance is much higher than observed, thus this assumption provides an upper bound on the benefits of adaptation.

4.

Benefits of avoidance adaptation under different climate policy scenarios (billions of 2015 USD) at midcentury and end-of-century. The avoided health burden, or health benefit, is gained by switching outdoor time to residential time during air quality alerts that apply to the full population (AQI > 150, “Unhealthy”). Benefits are shown by the contribution of reducing exposure to ozone (red) and PM_2.5 (blue), respectively. Adaptation to alerts triggered by high ozone levels is shown in (a) and (c). Adaptation to PM_2.5-driven alerts is shown in (b) and (d). Adaptation benefits from avoided morbidity are shown in (a) and (b) and adaptation benefits from avoided mortality are shown in (c) and (d). 95^th confidence intervals shown in Table S10.

When people adapt, they reduce their risk of morbidity primarily by reducing their exposure to ozone (Figure (a) and (b)). This is expected when populations adapt to ozone-driven alerts (Figure (a)), but it also remains true for adaptation to PM_2.5-driven alerts (Figure (b)). This is partially because moving inside is typically more effective at reducing exposure to outdoor ozone (a 30% reduction) than outdoor PM_2.5 (a 6% reduction). There are also more ozone-driven alerts than PM_2.5-driven alerts.

Adapting to all ozone-driven alerts and all PM_2.5-driven alerts avoids a similar amount of premature mortality (Figure (c)–(d)). Climate policy reduces the number of alerts, thereby reducing the potential gains from adapting. Specifically, full compliance would yield around $15 billion (95CI: 2.0, 51) (in 2015 USD) annually at the start of century, compared to $20 billion (95CI: 5.64, 145) in 2050 and up to $45 billion (95CI: 11, 270) in 2100 under the reference case (REF). When adapting to ozone, the mortality-related adaptation benefits are evenly split between reductions in ozone and PM_2.5 exposure, respectively (Figure (c)). When adapting to a PM_2.5-driven alert, however, benefits are dominated (over 80%) by reductions in PM_2.5 exposure.

Moving inside during alerts cannot compensate for rising health risks due to the climate penalty over this century. Potentially rising health benefits of adaptation (Figure ) are a small fraction of the rising health burden of pollution (Figure (d)). Even full compliance with “Unhealthy” air alerts cannot completely offset the climate-driven increase in risks. Under reference conditions, full compliance with “Unhealthy” air alerts would offset only 15% of the climate-driven increase in risks (valued at approximately $600 billion). Under climate policy (with 2C 2100 concentrations), the health burden is reduced, such that full compliance with air quality alerts can potentially avoid 40% of the $100 billion (95CI: 30, 1634) health burden.

Most of the air pollution-related health burden falls on seniors. Figure (a) shows the average proportion of the mortality burden experienced by age group across all scenarios. Seniors comprise 65% of the mortality burden from air pollution, even though they are 12% of the population (Figure (a)). The 36–64 age range accounts for up 29% of the burden and represents 39% of the population. The 18–35 and Under 18 groups only combined for 5.6% of the health burden from air pollution.

5.

Total air pollution related health burden by age (based on the average concentrations across all scenarios examined). (a) Proportion by age group (b) Health benefit per day adapting to an air quality alert for the general population in 2005. Shown for alerts triggered by PM_2.5 (blue) and ozone (red) for each age range, including the mean and 95% confidence interval for uncertainty related to health responses and their valuation.

Adaptation health benefits differ between age groups. Figure (b) shows the health-related economic benefits each age group receives from avoidance adaptation. Benefits are based on average concentrations across scenarios to reflect mean differences between groups. The 95CI reflects uncertainty in health responses and their valuation.

Benefits greatly exceed costs for seniors (Figure (b)). Seniors are outside for 43 min per day, on average. Previous research on the costs of adapting by staying inside provides a mean estimate of $32/h ($10.15-$75.65/h). Scaled to the 43 min of adaptation time, this is an average cost of $23 per adaptation. This is lower than the health-related benefits per adaptation of $80 (95CI: $5, $161) and $74 ($3, $102) for adapting at the “Unhealthy” AQI threshold for PM_2.5 and ozone, respectively. To have positive net benefits from adapting, the average senior would adapt at a mean AQI value of 71 for PM_2.5 based on premature mortality. Considering morbidity or ozone-related benefits would decrease the AQI value at which net benefits are achieved.

Adaptation benefits for seniors are higher from those of the young adult group, even considering the 95CI for both ozone and PM_2.5 (Table S11). On average, they are more than 47 (17–459) and 62 (22–652) times higher than the benefits received from the 18–35 age group for PM_2.5 and ozone, respectively. This is explained primarily by the difference in baseline mortality risk, which rises exponentially with age. These results use national mean activity and housing data. When considering spatially varying time use and activity intensity, the exposure reduction achieved by adapting to PM2.5 (Ψ _{PM 2.5} ) is approximately 0.06 for all age groups, matching the national-average value we used (Figure S11). The exception is for the Under 18 category, for which Ψ _{PM 2.5} is closer to 0.12. This suggests that, while children have the smallest contribution to the total air pollution health burden, they may reap higher benefits from adapting than estimated here, but still far less than those over 36.

Discussion

Changes in Future Air Quality Alerts

Under climate change, we observed a projected increase in days with high ozone and PM_2.5. We found that conditions for air quality alerts become more common in the absence of emission reductions, leading to a doubling (±20%) of alerts for the US population by end-of-century, with 33% (±15%) experiencing air quality “Unhealthy for Sensitive Groups” during smog season, with longer periods of high air pollution affecting more people. The climate penalty could yield two additional months of “Unhealthy for Sensitive Groups” alerts per year in some areas by end-of-century.

Climate policy can reduce the number of alerts per season and the adverse health impacts caused by the pollutants, but policies explored here would not fully offset the effects of climate change. Avoidance adaptation can further reduce health risks. While most alerts are caused by ozone, most of the benefits of adapting come from reducing exposure to PM_2.5 of outdoor origin. Our assumption that the full population complies with alerts will overestimate adaptation benefits, highlighting the need for climate change mitigation.

Implications for AQI Alerts

A single-pollutant index like AQI masks the benefits of adapting when multiple pollutants are elevated. Our findings show that much of the benefit of adapting to ozone is gained through reducing PM_2.5 and vice versa. Accounting for more than one pollutant increases the estimated benefits of air quality alerts substantially. Multipollutant indices, like Canada’s Air Quality Health Index, can offer advantages, especially if days with high ozone and PM_2.5 grow.

Based on net benefits of adaptation, current alerts may under-provide risk protection for sensitive groups. The climate penalty can exacerbate this, increasing days when both PM_2.5 and ozone reach unhealthy levels and dropping the AQI threshold at which seniors can expect net benefits from adapting. More personalized adaptation guidance, calculated and delivered on an individual level, could promote more effective adaptation.

Comparison with Other Studies

Limiting warming to 2.5 and 2 °C at end-of-century reduces the climate penalty’s economic burden by 33% and 41% under 2050 concentrations and 70% and 83% with 2100 concentrations, respectively. This aligns well with a previous study (Saari et al., 2019), which found that climate policy could reduce the climate penalty’s health burden by up to 50% in 2050 and up to 88% in 2100¹⁸. Our findings differ due primarily due to different concentration–response functions and recently published debiased ozone concentrations with improved performance for ozone extremes.

Our quantified adaptation effectiveness is different from values used in previous studies. Our avoidance adaptation PM_2.5 exposure reduction of 6% for remaining indoors is lower than the 35% used by Brook et al. (2024). Our ozone and PM_2.5 reduction values are higher than those used in Buonocore et al. (2021). This partially explains why we found higher benefits per adaptation than they did (up to $14/h). Other explanations include different concentration data, population data, and health impact functions. Our estimates are the first to use detailed time-use and housing data to describe exposure for multiple pollutants across different microenvironments when assessing benefits of adapting to alerts.

Limitations and Future Work

Our work isolates the effect of outdoor air pollution changes under future climate conditions on a static population. Larger, older populations and larger, wealthier economies in the future would cause the health-related economic burden of air pollution to be up to 6 times higher than presently reported (Figure S13), as would the benefits of adaptation (Figure S14). Migration, demographic changes, and other forces could also shift the exposure of populations to alerts, though differences in results when using a 2005 and 2020 population were relatively minor (Figure S15 through S17). To produce many future annual simulations, our study uses a relatively coarse grid resolution. This can lead to spatial smoothing and hinder capturing extreme air pollution. Comparison to historical data showed good agreement with the distribution of AQI across the US, with a slight overestimation of alerts, specifically due to ozone. Using finer resolution would likely result in more frequent alerts in some locations, and less frequent alerts in others. We examine ozone and PM_2.5 over the smog season, which underestimates total annual alerts from all AQI pollutants by approximately 20%. While previous evaluations of our concentrations found that they agreed with both measured air quality and other climate penalty projections, , using a large, multimodel ensemble would reduce model uncertainty. While we model the effects of climate change on biogenic sources of air pollution, specifically simulating the implications for wildfires could increase our resulting AQI alerts. Given our interest in AQI alerts, we focus on exposure to pollution of outdoor origin. Ozone primarily originates outdoors, though activities like smoking and cooking can generate elevated PM_2.5 indoors. If indoor sources increase during adaptation, this will limit the benefits of adapting. Future work should consider this potential feedback and its effect on total exposure to all pollutants input to the AQI. We use health impact functions to relate changes in concentration to health impacts. Differences in study selection, concentration–response function shape, pooling, and threshold concentrations could yield different outcomes. Age-related analysis could be improved by including additional age-specific baseline risks, effect modification by age, and age-varying risk preferences and costs of illness.

Reducing outdoor activity is the most common adaptation measure. However, the benefits of other strategies that reduce exposure to outdoor pollution, such as improving building airtightness, using portable air cleaners, and improving filtration, should also be assessed. Air quality alerts rely on individual awareness and compliance. Other strategies may be more robust to such behavioral factors, including adaptation fatigue. Further, approaches for more individualized guidanceconsidering individual risks, adaptation capacity, and preferencesshould be explored.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.5c12522.

• Additional tables describing methods, including air quality alerts, health impact assessment and economic valuation, and infiltration and supporting results including pollutant concentrations, health-related economic burdens, and sensitivity analysis for exposure reduction (PDF)

R.K.S. and M.S.S. formulated the study. J.D.E. and F.G.M. simulated future pollution based on climate fields from E.M. M.S.S. developed the analysis with input from R.K.S. R.K.S. and M.S.S. wrote the manuscript with input from all authors.

The authors declare no competing financial interest.