Effect of Health-related Uncertainty and Natural Variability on Health

Subscriber access provided by United Arab Emirates University | Libraries Deanship

Policy Analysis

Effect of Health-related Uncertainty and Natural Variability on Health Impacts and Co-Benefits of Climate Policy Rebecca Karina Saari, Yufei Mei, Erwan Monier, and Fernando Garcia Menendez Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.8b05094 • Publication Date (Web): 09 Jan 2019 Downloaded from http://pubs.acs.org on January 10, 2019

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 36

1 2 3

Environmental Science & Technology

Effect of Health-related Uncertainty and Natural Variability on Health Impacts and Co-Benefits of Climate Policy

4 5

Rebecca K. Saari*†, Yufei Mei†, Erwan Monier§⸸, Fernando Garcia-Menendez‡

6

† Civil

7

Waterloo, ON, CANADA, N2L 3G1

8

§Joint

9

Technology, Cambridge, MA, USA, 02139

and Environmental Engineering, University of Waterloo, 200 University Ave W.,

Program on the Science and Policy of Global Change, Massachusetts Institute of

10

⸸ now

11

CA, USA, 95616

12

‡

13

University, Raleigh, NC, USA, 27695

at: Department of Land, Air and Water Resources, University of California, Davis, Davis,

Department of Civil, Construction and Environmental Engineering, North Carolina State

14 15

*Corresponding author:

16

Rebecca K. Saari

17

University of Waterloo

18

200 University Ave W.

19

E2-2324

20

Waterloo, ON N2L 3G1

21

Office phone: (+1) 519-888-4567 x. 30362

22

General fax: (+1) 519-888-4380

23

[email protected]

ACS Paragon Plus Environment


24

ABSTRACT

25

Climate policy can mitigate health risks attributed to intensifying air pollution under climate

26

change. However, few studies quantify risks of illness and death, examine their contribution to

27

climate policy benefits, or assess their robustness in light of natural climate variability. We

28

employ an integrated modelling framework of the economy, climate, air quality, and human

29

health to quantify the effect of natural variability on US air pollution impacts under future

30

climate and two global policies (2ºC and 2.5ºC stabilization scenarios) using 150-year ensemble

31

simulations for each scenario in 2050 and 2100. Climate change yields annual premature deaths

32

related to fine particulate matter and ozone (95CI: 25,000-120,000), heart attacks (900-9,400),

33

and lost work days (3.6M-4.9M) in 2100. It raises air pollution health risks by 20%, while

34

policies avert these outcomes by 40-50% in 2050 and 70-88% in 2100. Natural variability

35

introduces “climate noise”, yielding some annual estimates with negative co-benefits, and others

36

that reach 100% of annual policy costs. This “noise” is three times the magnitude of uncertainty

37

(95CI) in health and economic responses in 2050. Averaging five annual simulations reduces this

38

factor to two, which is still substantially larger than health-related uncertainty. This study

39

quantifies the potential for inaccuracy in climate impacts projected using too few annual

40

simulations.

41

2


Page 2 of 36

Page 3 of 36


42

INTRODUCTION

43

Climate change and air pollution are public health challenges linked in ways that affect policy

44

outcomes. Climate change increases health risks due to air pollution1–3. This ‘climate penalty’4

45

on air quality affects adaptation, vulnerability, and preparedness5,6. Mitigating climate change

46

could yield health-related air quality co-benefits that offset policy costs7–12, though most studies

47

focus on health co-benefits due to reducing co-emitted pollutants only, and do not consider the

48

effect of climate change. This is partly due to the fact that climate-related improvements in air

49

quality (‘climate co-benefits’) are considered small compared to those of co-emitted pollutants,

50

and partly due to the added computational complexity of simulating climate change, which is

51

obscured by natural (or internal) variability. Most studies that do simulate the effect of climate

52

change on pollutant concentrations address natural variability by averaging five years or less of

53

simulations, though there is evidence that more years may be needed 13. No studies have yet

54

considered the adequacy of this typical approach for health impact analysis. The well-known

55

uncertainty in relationships between concentrations and health responses may preclude the need

56

for more precise concentrations, or may, by comparison, reveal the significance of uncertainty

57

introduced by natural variability. Here, we quantify air pollution-related health risks from

58

climate change, and isolate the effect of natural variability on climate co-benefits of stringent

59

global policies that avert those risks. By comparison with health-related uncertainty, we assess

60

the public health and policy relevance of addressing natural variability, and test the adequacy of

61

current practice to address this source of uncertainty in projections of climate impacts.

62 63

Climate change can affect air pollutant concentrations, including ozone and fine particulate

64

matter (PM2.5), through multiple mechanisms, such as enhancing pollutant formation at higher

3



65

temperatures or via reduced atmospheric ventilation.14 Climate policy improves air quality by

66

directly reducing the effect of climate change on air pollution (the ‘climate co-benefit’15), and by

67

reducing co-emitted air pollutants. Many studies estimate health co-benefits of climate policy

68

related to air pollution16,17; however, most use constant meteorological inputs, focusing only on

69

co-emitted pollutants. Co-benefits from co-emitted pollutants are likely larger than climate co-

70

benefits15. However, climate co-benefits may increase over time, as pollutant emissions are

71

reduced through air quality measures, and as the climate responds to greenhouse gas reductions.

72

Garcia-Menendez et al.18 found climate co-benefits of $8-42/tCO2e in 2050 rose to $45-207/tCO2

73

by 210018. Still, few studies include climate co-benefits7,19–23, and few isolate this effect15,18,24.

74

Those that do isolate climate co-benefits find significant annual impacts, up to thousands of

75

deaths avoided and trillions of dollars in benefits offsetting up to one quarter of annual policy

76

costs15,18,23.

77 78

Despite their significance25, human health co-benefits lack traction in climate policy analysis16.

79

This is due, in part, to a variety of methods25 and uncertainties26 that limit general conclusions17.

80

Previous studies assessed uncertainties, including emissions scenarios27, demographics28,29,

81

model selection24,28 and resolution30–33, exposures34,35, and health-related uncertainty. Health-

82

related uncertainties include the shape of the concentration-response function (CRF)36,37 (relation

83

of air pollution exposure to health risk), thresholds24, baseline incidence rates8,24, confounding,

84

and effect modification38,39. In United States (US) air quality policy analysis, the most commonly

85

quantified health-related uncertainty40 is the confidence interval in the CRF associated with

86

individual studies, such as those for PM2.5-related mortality from the American Cancer Society

87

study41 and the Harvard Six Cities42. Capturing uncertainty related to the health impacts of

4


Page 4 of 36

Page 5 of 36


88

climate change or climate policy, therefore, requires an assessment of multiple endpoints

89

(mortality and morbidity) using multiple CRFs.

90 91

While recent reviews1–3 present estimates of projected health burdens of air pollution due to

92

climate change, policymakers need a better understanding of future mortality and morbidity

93

risks, and associated uncertainties. Many US studies address only ozone-related mortality23,28,43–

94

46

95

globally49. There is some intermodel agreement that PM2.5 increases in many locations under

96

climate change50, yielding higher PM2.5-related US mortality by 210024. Existing studies of the

97

entire US include mortalities in 205028,45,18,27,51,52 and 210024 due to PM2.5 and ozone; however,

98

only one27 includes morbidities, such as heart attacks and hospitalizations, needed for public

99

health preparedness3 and uncertainty assessment.

and morbidity23,47. However, PM2.5 is a leading cause of disease burden in the US48 and

100 101

Quantifying uncertainties helps to assess their policy relevance, and to inform more consistent

102

methods. For example, previous studies found the choice of climate change-air quality modeling

103

system yielded the greatest uncertainty in ozone-related mortality under climate change,

104

compared to population projections and concentration-response functions28. Different climate

105

change-air quality modeling systems yield deaths due to climate change that differ by 3,000

106

deaths for ozone in 205028 and 28,000 deaths for PM2.5 in 210024. Other work27 suggests that

107

uncertainty in climate projections may have a comparable effect on health impacts. One study27

108

estimated the effect of uncertainty in future climate projections on ozone- and PM2.5-related

109

deaths, using the 0.5th and 99.5th percentiles of a probabilistic distribution of meteorological

110

variables derived from the MIT Integrated Global System Model (MIT IGSM). The difference in

5



111

estimates across these variables was 2,600 ozone-related deaths and 9,300 PM2.5-related deaths

112

in the US in 2050 under RCP8.5. While this suggests uncertain climate projections may be as

113

significant as inter-model differences for future public health, the effect of natural variability on

114

climate co-benefits has not been quantified.

115 116

Natural climate variability can introduce “noise”53 into climate projections that obscures

117

estimates of the future health burden of air pollution. Natural variability is the unforced

118

fluctuation resulting from the chaotic nature of the climate system53, including non-linear

119

interactions, feedbacks, and varying response times among climate system components54.

120

Consequently, in one year the reference simulation might be warm and dry, and the policy

121

simulation cool and wet, yielding differences due to natural variations in meteorology that are

122

incorrectly attributed to policy13. This yields large uncertainty in climate projections, including

123

projections of the climate penalty14, which can be evaluated using multiple initial condition

124

ensembles of climate simulations.55

125 126

To date, however, the implications of natural variability for health impacts of air pollution

127

remain unclear56. This is primarily due to computational intensity14 of many required simulations

128

of climate, air quality, and human health responses across multiple pollutants and health

129

outcomes. Unlike uncertainty in CRFs, the effect of natural variability can be filtered out with

130

sufficient simulations.13 Most (e.g., 29 of 41 studies on ozone13) air quality projections average

131

less than five years of simulations13. Recent findings suggest more simulations may be needed –

132

perhaps more than ten-year averages – to achieve sufficient precision in future pollutant

133

concentrations13,57,58. Since this finding exceeds current practice for health impact assessment,

6


Page 6 of 36

Page 7 of 36


134

potentially introducing further computational burden, the significance of natural variability

135

should be assessed in the context of other significant and well-quantified sources of uncertainty,

136

such as in CRFs.

137 138

Here, we employ a large multi-decadal, multiple initial condition ensemble to assess the effect of

139

natural variability and uncertain human health responses on US mortality, morbidity, and

140

economic impacts of future climate and two global climate policies. We use constant pollutant

141

emissions to isolate the effect of climate change on PM2.5 and ozone. We employ self-consistent,

142

coupled models of the economy, climate, air quality, and human health at midcentury and end-

143

of-century. Our scenarios include a reference case (REF) and two policies – Policy 4.5 (P4.5)

144

and Policy 3.7 (P3.7) – designed to stabilize global temperature rise at 2.5°C and 2.0ºC from pre-

145

industrial, respectively, by 210059. We apply 30-year averages across five climate model

146

initializations, a total of 150 annual simulations per policy and target year, to filter out natural

147

variability and estimate the future health burden of air pollution due to climate change. We

148

quantify the influence of natural variability and health-related uncertainty on climate co-benefits

149

and policy selection. We compare their relative influence versus policy cost. Finally, we assess

150

the effect of the common practice of using five-year averages to address natural variability.

151 152

METHODS

153

Integrated Modeling Approach

154

We examined two global climate policies described elsewhere59 using an internally consistent

155

modeling framework, developed in Garcia-Menendez et al.18, depicted in Table 1. It employs the

156

MIT Integrated Global System Model (IGSM)60,61, an integrated assessment model that couples

7



Page 8 of 36

157

the MIT Economic Projection & Policy Analysis (EPPA)62,63 to the model of the world economy

158

to the MIT Earth System Model (MESM)64 of intermediate complexity. EPPA is a multi-region

159

multi-sector computable general equilibrium model that projects economic activity and

160

associated emissions of greenhouse gases, aerosols, and other climate-relevant species under

161

policy constraints by determining prices that balance supply and demand across five-year

162

periods, 25 sectors and 16 world regions62,63. We use the MIT IGSM-CAM framework65, which

163

links the IGSM to the National Center for Atmospheric Research (NCAR) Community

164

Atmosphere Model (CAM) to provide three-dimensional climate variables used to simulate

165

global atmospheric chemistry with the Community Atmosphere Model with Chemistry (CAM-

166

Chem)66 at a 1.9° × 2.5° horizontal resolution, an approach balancing accuracy and

167

computational efficiency needed for uncertainty analysis.

168

Table 1: Experimental Design to Isolate the Effect of Natural Variability on Climate Co-Benefits Framework (EPPA)

Variables Constant Anthropogenic pollutant emissions

Simulations Scenarios Reference, Policy 4.5, Policy 3.7

Output Impacts due to fine particulate matter and ozone exposure (multiple CRFs)

Climate

Population age/spatial distribution

Years of Interest 2000, 2050, 2100

Varying Population growth

Annual simulations 30-year periods: 1986-2015; 20362065; 2086-2115

- All-cause mortality - Morbidity including: - acute myocardial infarction, - hospital admissions (respiratory, cardiovascular, emergency), - respiratory symptoms (upper respiratory symptoms, asthma exacerbation, acute bronchitis), - lost productivity (work loss days, school loss days, minor restricted activity days) - Economic valuation of benefits

Policy

(MESMCAM)

Economic growth Air Quality (CAM-Chem)

Baseline mortality incidence rates

5 initializations

GHG emissions

169 170 171 172 173

Health &

Climatic conditions

Valuation

Pollution concentrations

150 annual simulations per scenario and year of interest

(BenMAP)

EPPA =MIT Economic Projection & Policy Analysis; MESM = MIT Earth System Model; CAMChem = Community Atmosphere Model with Chemistry; BenMAP = environmental Benefits Mapping and Analysis Program; CRF = concentration-response function

8


Page 9 of 36


174

CAM-Chem predictions of ground-level ozone and PM2.5 concentrations have been evaluated66–

175

68.

176

inventory66. Health and economic impacts of ozone and PM2.5 are estimated with the

177

environmental Benefits Mapping and Analysis Program–Community Edition (BenMAP-CE

178

v1.0814)69,70. MIT IGSM-CAM-Chem-BenMAP framework is used here to estimate global

179

policy costs, US health outcomes, and economic impacts across three scenarios.

Emissions are based on the Precursors of Ozone and their Effects in the Troposphere (POET)

180 181

Health Outcomes and Valuation of Climate-Induced Air Quality Impacts

182

We used CRFs included in Table S1. The selection and pooling of CRFS and valuations

183

followed regulatory analyses by the US Environmental Protection Agency (EPA)71,72, to estimate

184

health outcomes as:

185

∆𝑂𝑢𝑡𝑐𝑜𝑚𝑒𝑠 = 𝑦0 × 𝑃𝑜𝑝 × (1 ― exp ( ―𝛽 × ∆𝑥))

186

Where y0 is the baseline cause-specific incidence rate, Pop is population size, β is the risk

187

coefficient for the health end point of interest, and Δx is the change in pollutant concentration

188

between two scenarios. We applied no lower concentration threshold. While some CRFs had

189

different functional forms, all depended on the change between two scenarios in daily mean

190

PM2.5 or May-through-September daily maximum 8-hour average ozone. We used two CRFs for

191

for PM2.5 (Krewski et al.41 and Lepeule et al.42) and ozone (Zanobetti and Schwartz73 and Smith

192

et al.74) to estimate all-cause mortality. Morbidity endpoints included nonfatal myocardial

193

infarction, hospital admissions, and other symptoms (Table 1). Incidences valued per Table S1.

194 195

Reference and Policy Scenarios

9



196

Three scenarios of economic activity, greenhouse gas emissions and climate change were

197

developed for US EPA’s Climate Change Impacts and Risk Analysis (CIRA) project75, and used

198

to assess impacts across US sectors76, presented elsewhere18,59,77 and the SI. REF has

199

unconstrained emissions, CO2 reaching 830 ppm, and global mean surface temperature

200

increasing by 6°C in 2100. Policies implement a global tax on carbon emissions to stabilize total

201

radiative forcing at 4.5 W/m2 (for P4.5) or 3.7 W/m2 (for P3.7) by 210077. They avoid 8 and 9

202

billion tons of US CO2 emissions in 2100 under P4.5 and P3.7, respectively, and limit CO2 levels

203

to 500 ppm (P4.5) and 460 ppm (P3.7) and temperature rise to 2.5°C (P4.5) and 2.0ºC (P3.7)

204

from preindustrial.

205 206

To assess the public health burden, we isolate climate-induced changes in ozone and PM2.5 at the

207

middle and end of the 21st century relative to their start-of-century levels (details in Garcia-

208

Menendez et al.18). We consider only climate co-benefits, and not co-emissions co-benefits. To

209

do this, we isolate climate-induced effects with constant year 2000 anthropogenic pollutant

210

emissions, and constant natural emissions. The response of biogenic emissions to temperature is

211

simulated, but other effects of climate change on natural sources, including dust and wildfires,

212

are not modeled. PM2.5 changes include sulfate, black carbon, organic aerosol and ammonium

213

nitrate particles18,78. Under REF, US population-weighted annual 8-hr-max ozone increases by

214

0.8 ± 0.3 and +3.2 ± 0.3 ppbv and population-weighted annual PM2.5 increases by 0.5 ± 0.1

215

μg/m3 and 1.5 ± 0.1 μg/m3 in 2050 and 2100, respectively (see SI). Drivers of ozone change

216

include greater stagnation, enhanced photochemical formation, and higher emissions of biogenic

217

precursors. Climate-induced effects on PM2.5 vary by component, with meteorological variations

10


Page 10 of 36

Page 11 of 36


218

affecting oxidation, gas-particle partitioning, and atmospheric ventilation. Policies reduce

219

penalties by over 80% for ozone and 70% for PM2.5 in 2100.

220 221

Start-of-century population, demographics, and baseline incidence rates were derived from

222

BenMAP-CE. Future population was projected (growing 50% by 2050 and 75% by 2100 relative

223

to 2005)59. Baseline mortality incidence rates were projected as in Garcia-Menendez et al.18,

224

based on cardiopulmonary mortality for PM2.5 (12% increase in 2050, -8% in 2100, relative to

225

2008) and respiratory mortality for ozone (24% in 2050, 64% in 2100). Migration and

226

demographic shifts were not modeled. Valuations were projected using GDP per capita growth

227

(80-100% by 2050 and 240-250% by 2100, relative to 2005). We applied a 0.4 income elasticity

228

(central estimate in BenMAP-CE79), assuming most benefits are due to reduced mortality risk.

229 230

Isolating Uncertainty from Natural Variability and Health Responses

231

We characterized interannual variability with 30-year simulations covering start-of-century

232

(1981-2010), mid-century (2036-2065), and end-of-century (2085-2115). We captured

233

multidecadal variability by modeling each 30-year period with five perturbations of conditions in

234

the MIT IGSM-CAM65. To simulate average conditions in 2050 and 2100, we detrended

235

concentrations within each future 30-year period using least squares linear regression at each

236

grid cell. We applied GDP, baseline mortality incidence, and population values in the year 2050

237

or 2100.

238 239

To filter out natural variability, we calculated ensemble mean concentrations yielding 150-year

240

averages for each scenario and pollutant as input to BenMAP. To quantify the effect of annual

11



241

natural variability on climate co-benefits, we compared each annual simulation under policy to

242

REF. In total, we simulated health impacts for 1200 years (two pollutants, two policies, two 30-

243

year periods, and five initializations). To assess common practice, we averaged five years of

244

impacts to partially filter out natural variability, which we do around each year within the

245

respective 30-year periods (details in SI).

246 247

Health and economic uncertainty is the 95% confidence interval (95CI) obtained from 5,000

248

Monte Carlo simulations of pooled CRFs and economic impacts. The choice of the CRF is

249

assessed using two CRFs for PM2.5-related mortality. We likely underestimate this uncertainty

250

because other CRFs are unchanged, other sources of uncertainty are not quantified (e.g.,

251

valuation methods18, shape of CRF 80, and there remains significant epistemic uncertainty (see

252

SI).

253 254

RESULTS AND DISCUSSION

255

Future Health Burden of Climate Change and Health Benefits of Reducing Climate Penalty

256

Compared to start of the century, climate change increases US air pollution-related risks,

257

yielding annual premature deaths (95CI: 40,000 to 120,000 using Lepeule et al.42), non-fatal

258

heart attacks (940 to 9,400), and lost work days (3.6M to 4.9M) in 2100 (Figure 1 and Table S2).

259

Figure 1 compares future conditions to start-of-century, including climate change and projected

260

population. Natural variability was filtered out using the 30-year average of annual atmospheric

261

chemistry simulations across the five model initializations (i.e., 150 years averaged) for each

262

scenario. The effect of climate change is not always significant at 95CI for each health endpoint

12


Page 12 of 36

Page 13 of 36


263

(details in SI). Global policies consistent with 2 and 2.5°C warming targets reduce these risks

264

across all endpoints in Figure 1 by 40-60% in 2050 and 70-88% in 2100.

265

266 267 268 269 270 271 272 273 274 275 276 277

Figure 1. Change in US premature mortality and morbidity at end-of-century relative to start-ofcentury related to (a) PM2.5 and (b) ozone under the REF, P4.5, and P3.7. The ranges are 95th confidence intervals due to health-related uncertainty, after natural variability is filtered out using ensemble-mean concentrations. Lepeule = estimated with Lepeule et al.42. Krewski = estimated with Krewski et al.41. Z &S = mortality estimated with Zanobetti and Schwartz73. Smith = mortality estimated with Smith et al.74. ERA = Emergency Room visits for asthma; RHA = respiratory hospital admissions; CHA = cardiovascular hospital admissions; NFMI = non-fatal myocardial infarction. WLD = work loss days; AE = Asthma exacerbation; URS = upper respiratory symptoms; AB = Acute bronchitis; MRAD = minor restricted activity days; SLD = school-loss days.

278

Policies avoid thousands of annual premature deaths and illnesses in the US. Table 2 compares

279

policies to the reference case in 2050 and 2100, respectively, for the same future populations. By

280

midcentury, P4.5 prevents 11,000 (5,000 to 17,000) premature deaths (using Lepeule et al.42),

281

470 (130 to 1,300) non-fatal heart attacks, and 570,000 (480,000 to 660,000) lost work days

282

annually. The more ambitious P3.7 reduces these risks by an additional 15% at midcentury for

283

PM2.5-related endpoints. By end-of-century, 59,000 (29,000 to 89,000) premature deaths are

284

avoided annually (using Lepeule et al.42), also 2,500 (910 to 6,700) heart attacks and 3 million

13



285

(2.6 to 3.5 million) lost workdays under P4.5. P3.7 avoids an additional 40% of PM2.5-related

286

outcomes.

287 288

Controlling natural variability yields consistent estimates of relative and absolute values of

289

policies. Mean climate co-benefits of both policies are positive, offsetting a fraction of costs

290

increasing from 2-6% in 2050 to 6-16% in 2100. The more stringent P3.7 has higher costs and

291

higher mean co-benefits18. Annually, it yields $10 billion more than P4.5 in 2050 and $30 billion

292

more in 2100 using Krewski et al.41 (or, using Lepeule et al.42: $30 billion more in 2050; $100

293

billion more in 2100).

294

14


Page 14 of 36

Page 15 of 36

295


Table 2: Future Health Impacts Avoided by Policy in 2050 and 2100. Health Outcomes Avoided Annually (95% confidence interval)

Endpoint Group

P4.5 2050

P3.7 2100

2050

2100

59,000* (29,000, 89,000)

13,000* (5,800, 20,000)

68,000* (33,000, 100,000)

29,000* (18,000, 39,000) 3,000,000* (2,600,000, 3,500,000)

6,200* (3,800, 8,500) 680,000* (570,000, 800,000)

32,000* (21,000, 44,000) 3,500,000* (3,000,000 4,000,000)

860* (270, 1,600) 130,000* (15,000, 240,000)

4,700* (1,800, 8,700) 680,000* (120,000, 1,200,000)

1,000* (320, 2,000) 150,000* (15,000, 290,000)

5,500* (2,100, 10,000) 790,000* (140,000, 1,400,000)

470* (130, 1,300) 2,500,000* (550,000, 4,400,000) 890,000 (-22,000, 2,500,000)

2,500* (910, 6,700) 17,000,000* (5,800,000, 28,000,000) 6,400,000* (1,200,000, 17,000,000)

570* (150, 1,600) 2,400,000* (470,000, 4,300,000) 860,000 (-52,000, 2,400,000)

3,000* (1,100, 7,800) 16,000,000* (4,600,000, 26,000,000) 5,700,000* (560,000, 15,000,000)

Adult Mortality Lepeule et al.

11,000* (2012) (5,000, 17,000)

Krewski et al.

5,200* (2009) (3,200, 7,300) 570,000* (480,000, Work Loss Days 660,000) Cardiovascular Hospital Admissions Upper Respiratory Systems Acute Myocardial Infarction, nonfatal Minor RestrictedActivity Days School- Loss Days

296

* Significant at p < 0.05. (School-loss days in 2050 are significant at p < 0.07)

297 298 299 300 301

Annual avoided premature mortality and morbidity in 2050 and 2100 under P4.5 and P3.7 relative to REF due to reducing the climate penalty on ozone and PM2.5. The ranges are 95th confidence intervals due to health-related uncertainty, after natural variability is filtered out by using ensemble-mean concentrations. Mortality estimates include those related to ozone as estimated by taking the median of Zanobetti and Schwartz73 and Smith et al.74

302

Uncertainty in Climate Co-Benefits from Improved Air Quality under Two Global Policies

303

When natural variability is not filtered out, it obscures relative and absolute climate co-benefits.

304

Figure 2 depicts a multi-trillion dollar range of climate co-benefits in the US under two global

305

policies and constant pollutant emissions. The spread represents uncertainty due to natural 15



306

variability, health impacts, and valuation. Some individual estimates (e.g., negative estimates) do

307

not reflect true policy impacts. Climate co-benefits in Figure 2 span from -4 to 3.5 trillion USD

308

per year around 2050, to -6 to 13 trillion USD annually around 2100. Policies cost $9-11 trillion

309

annually in 210059 (all USD in year 2000 currency). Thus, when climate co-benefits are obscured

310

by these two factors, they can seem comparable to policy costs, appearing to offset over 100% of

311

annual climate policy costs in some simulations at the end of the century (further details in SI).

312

313 314 315 316 317 318 319 320 321 322

Figure 2. Annual health co-benefits from reducing the climate penalty on US air quality by policies P4.5 (red) and P3.7 (blue) for a 30 year period around (a) 2050 (b) 2100. The shaded regions indicate the 95th confidence intervals given uncertainty in health impacts and economic benefits, while the dots represent the mean co-benefits for each annual simulation. There are five initializations per policy per year. The variation between dots of the same color represents the effect of natural variability. Co-benefits include reduced mortality and morbidity for PM2.5 and ozone. Adult mortality from PM2.5 estimated using Lepeule et al.42 (results using Krewski et al.41 in SI). Currency is year 2000 USD.

323

Natural variability and uncertain health responses have comparable effects on climate co-

324

benefits, especially at mid-century. Figure 3 shows the fraction of US annual climate policy costs

325

offset by climate co-benefits in 2050 and 2100. Health and economic uncertainty is estimated by

16


Page 16 of 36

Page 17 of 36


326

pooling multiple health and economic studies for various health outcomes (Table S1). For those

327

estimates, natural variability is filtered out using ensemble mean (across 150 years) pollutant

328

concentrations in 2050 and 2100. Uncertainty due to CRF selection is represented using CRFs

329

for PM2.5–related mortality in Krewski et al.41 and Lepeule et al.42. Uncertainty due to natural

330

variability is represented by the variation in mean health and economic estimates across 150

331

years. Remaining uncertainty across five initializations is shown after applying the common

332

practice of using five-year averages health and economic impacts (details in SI).

333 334 335 336 337 338 339 340 341

Figure 3. Effect of uncertainties on percent of policy costs offset by reducing health risks from climate penalty in (a) 2050 and (b) 2100. Range depicts 95th confidence interval. In “Health and Economic Uncertainty”, natural variability is removed with the entire relevant ensemble (five initializations over 30 years). “Natural Variability” shows the variability across the ensemble (five initializations over 30 years). Reflecting typical practice, five year averages are then used to partially filter out natural variability. Benefits include reduced mortality and morbidity for PM2.5 and ozone. Adult mortality from PM2.5 estimated using either Lepeule et al.42 or Krewski et al.41.

342

In estimating climate co-benefits at midcentury, natural variability has a larger impact than

343

health-related uncertainty. Compared to the 95CIs associated with health and economic

344

uncertainty, those representing natural variability are three to four times larger (Figure 3(a)).

345

Using Lepeule et al.42 instead of Krewski et al.41 increases the size of the 95CI, but this type of 17



346

health-related uncertainty is still less than the effect of natural variability. Averaging five years’

347

of simulations to estimate co-benefits reduces this uncertainty, but it remains commensurate with

348

health-related uncertainty. The resulting 95CI is 1.1 to 1.6 times that of health and economic

349

uncertainty for a given CRF, and 0.8 times the spread across CRFs. The 95CI is -$54 to 80/tCO2e

350

due to natural variability alone (using Krewski et al.41), -$15 to 43/tCO2e due to natural

351

variability after applying five-year averaging (using Krewski et al.41), and -$1 to 75/tCO2e due to

352

health-related uncertainty (across health and economic uncertainty and two CRFs).

353 354

By end-of-century, CRF selection carries greater weight than natural variability (Figure 3(b)).

355

As the climate responds to greenhouse gas mitigation measures, the signal of policy becomes

356

stronger compared to the noise of natural variability, and health-related uncertainty begins to

357

dominate. Using a different CRF (Lepeule et al.42 instead of Krewski et al.41) doubles the mean

358

and more than doubles the size of the 95CI for policy cost offset. Conversely, natural variability

359

remains equal in its 95CI to that of health and economic uncertainty. Averaging improves this,

360

with five year averages shrinking uncertainty due to natural variability to about one third (33-

361

36%) of health and economic uncertainty.

362 363

Spurious negative impacts appear in our ensemble. This occurs primarily when an annual

364

simulation of the climate under a policy leads to higher pollutant formation than the reference,

365

i.e., “noise” from natural variability overwhelms policy benefits13. At midcentury, in Figure 3(a),

366

a small (

Effect of Health-related Uncertainty and Natural Variability on Health

Recommend Documents