I Ec

Assessing the Mortality Burden of
Air Pollution in Lima-Callao

Final Report | December 14, 2020

prepared for:

U.S. Environmental Protection Agency
prepared by:

Industrial Economics, Incorporated
2067 Massachusetts Avenue
Cambridge, MA 02140
617/354-00074

INDUSTRIAL ECONOMICS, INCORPORATED


-------
lEc

TABLE OF CONTENTS

ACKNOWLEDGEMENTS

CHAPTER 1 | INTRODUCTION

1.1	Background 1-1

1.2	Analytic Overview 1-2

1.2.1	Research Objectives 1-2

1.2.2	Analytic Steps 1-3

1.3	Report Organization 1-3

CHAPTER 2 | SCENARIO DEVELOPMENT

CHAPTER 3 | EMISSIONS ESTIMATION

3.1	Baseline Emissions Inventory 3-1

3.2	Accounting for Non-Compliance 3-3

3.3	Emissions Modeling Results 3-4

CHAPTER 4 | AIR QUALITY DATA AND MODELING

4.1	Air Quality Surfaces 4-1

4.1.1	Monitor Surfaces 4-1

4.1.2	Satellite Surfaces 4-3

4.1.3	Summary of Air Quality Surfaces 4-3

4.2	Air Quality Modeling 4-5

CHAPTER 5 | MORTALITY BURDEN ESTIMATION AND VALUATION

5.1	Overview of Approach 5-1

5.2	Data Inputs 5-2

5.2.1	Population 5-2

5.2.2	Baseline Mortality Incidence 5-2

5.2.3	Health Impacts Functions 5-2

5.2.4	Valuation 5-4

CHAPTER 6 | RESULTS

6.1	Total PM2 5 Attributable Mortality Burden 6-1

6.2	Transport Attributable PM2 5 Mortality Burden 6-2

6.3	PM2 5 Mortality Burden from Non-Compliant Vehicles 6-3

INDUSTRIAL ECONOMICS, INCORPORATED

i


-------
lEc

CHAPTER 7 | SUMMARY OF FINDINGS

7.1	Summary of Findings 7-1

7.2	Uncertainties 7-2

7.3	Next Steps 7-3

REFERENCES
APPENDICES

Appendix A | Supplemental Emissions Estimation Results
Appendix B | Satellite Measurements and Processing
Appendix C | Health Impact Estimation
Appendix D | District-Level Results

INDUSTRIAL ECONOMICS, INCORPORATED

ii


-------
lEc

ACKNOWLEDGEMENTS

This report was prepared by Industrial Economics, Incorporated (IEc) for the U.S.
Environmental Protection Agency (USEPA) Office of Air Quality Planning and
Standards (OAQPS), under USEPA contract EP-D-14-032 as part of EPA's Megacities
Partnership program in Lima, Peru. Consistent with the Megacities Partnership approach
to air quality management, this project was a collaborative effort between USEPA, IEc,
and local partners at the Peruvian Ministry of Environment (MINAM). In particular, IEc
acknowledges critical support from Luis Antonio Ibanez Guerrero, Katyuska Baija
Paredes, and Guisselle Castillo Coila of MINAM, who provided key data sets, valuable
local institutional knowledge, and helpful advice. We also thank USEPA staff including
the Lima Megacities Project Manager Paul Almodovar as well as a Richard Baldauf, Ali
Kamal, and Ken Davidson for their input on the analysis and helpful feedback on earlier
drafts of this report.

INDUSTRIAL ECONOMICS, INCORPORATED

iii


-------
lEc

CHAPTER 1 | INTRODUCTION

The United States Environmental Protection Agency (USEPA) and the Ministry of the
Environment of Peru (MINAM) are collaborating under the USEPA Megacities
Partnership to:

•	Strengthen air quality management in the Lima-Callao region through policy
development, community outreach, and stakeholder engagement;

•	Support air quality monitoring initiatives; and

•	Build technical capacity in Lima-Callao for scientific and economic analyses and
communication planning in support of air quality management plan (AQMP)
development.

This report presents results from assessments of the current overall mortality burden
attributable to concentrations of fine particulate matter (PM2.5) in the region. In addition
to estimating total PM2 5-attributable health burden with respect to premature deaths, we
employ emissions inventories and reduced-form air quality modeling techniques to
analyze the burden attributable to concentrations of PM2.5 associated with emissions
from on-road motor vehicles. We further evaluate the burden of a subset of these
vehicles—those out of compliance with currently established emissions limits—and
highlight the potential benefits associated with increased enforcement and expanded
vehicle inspections and maintenance (I&M) programs.

1.1 BACKGROUND

The Lima-Callao metropolitan region is home to approximately 10 million people, nearly
one-third of Peru's total population. This large and growing population is exposed to
significant air pollutant concentrations due to emissions from sources such as motor
vehicles. These exposures can be exacerbated by Lima-Callao"s meteorological
conditions, primarily the Humboldt ocean current and the Andes Mountains to the east.
The Humboldt ocean current carries cold water north from the tip of South America,
which lowers atmospheric temperatures and prevents the formation of rain clouds (Thiel
et al. 2007). The Humboldt ocean current is also responsible for persistent fog in Lima-
Callao. The fog, combined with the obstruction of warmer and more humid air masses
from the Amazon, by the Andes, results in frequent air inversions in Lima-Callao. These
air inversions trap ambient pollutants at the surface level, causing pollutants to
accumulate rather than disperse via coastal winds. As a result, Lima-Callao is ranked
among the most polluted cities in Latin America by the World Health Organization
(WHO, 2016).

INDUSTRIAL ECONOMICS, INCORPORATED

1-1


-------
Transportation sources are responsible for much of the region's air pollution. Despite
comprising one-third of the country's population, Lima-Callao is home to roughly two-
thirds of Peru's vehicle fleet. In addition to the size of the vehicle fleet, its age plays a
significant role in resulting air pollution (MINAM, 2018). According to surveys of
registered vehicles from 2016-2019, approximately 34 percent of the on-road vehicles in
Lima-Callao are greater than 15 years old. These older vehicles tend to have poor fuel
economy and lack the emissions controls required of new vehicles. Based on emissions
tests, MINAM partners note that many vehicles are non-compliant with national
emissions standards; however, empirical estimates of non-compliance rates are currently
limited.

To address emissions from the transportation sector, the Government of Peru has passed
several laws and regulations concerning emissions standards and vehicle inspection
requirements. For example, in 2008, Peru established the National System of Technical
Vehicle Inspection, which is responsible for inspecting and testing vehicles for safety and
compliance with emissions standards. More recently, Peru adopted the Euro 4 vehicle
emissions standards for all new vehicles and is considering implementing Euro 6
standards. Because of the slow rate of turnover in the vehicle fleet—as shown by the
prevalence of old vehicles in circulation—additional measures may be needed to address
the sector's emissions. In this report, we provide insight into the potential magnitude of
these on-road vehicle emissions and associated adverse health outcomes.

1.2 ANALYTIC OVERVIEW

In this section, we summarize our analytic approach to estimating the mortality burden
associated with PM2.5 concentrations in Lima-Callao. We first define our research
objectives and then outline the analytic steps we follow in the remainder of the report.

1.2.1 RESEARCH OBJECTIVES

In close consultation with MINAM and USEPA, we developed three research objectives
addressed in this report. First, we aim to quantify and value the premature deaths
associated with overall ambient PM2.5 concentrations in Lima-Callao. Second, we aim to
quantify and value the premature deaths associated with emissions from Lima-Callao's
on-road vehicles, to better understand how vehicles contribute to the overall burden of
premature deaths. Third, we aim to quantify and value the premature deaths associated
with emissions from non-compliant vehicles in the Lima-Callao on-road vehicle fleet, to
help MINAM understand the potential health gains from focusing on the non-compliance
issue. For each research objective, we consider annual impacts using data that best
characterize recent conditions for air quality, population, baseline health, and other
relevant data. In Chapter 7, we highlight additional areas of future research that could
complement this report. We hope that the analytical framework IEc applied in this
analysis will serve as a useful guide for addressing these research topics, including
estimating the benefits of specific transportation emissions control measures.

INDUSTRIAL ECONOMICS, INCORPORATED

1-2


-------
1.2.2 ANALYTIC STEPS

Our methodology is comprised of five key steps:

•	Step 1: Scenario development. Define the research objectives and the spatial and
temporal scale of our analyses. Specify conditions under a ""busincss-as-usual" or
"baseline" scenario and under a "regulatory" scenario in which the proposed
regulation is implemented. Based on these scenario definitions, develop baseline
and regulatory air quality surfaces.

•	Step 2: Emissions estimation. Develop or obtain emissions inventories for the
transportation sector. Develop estimate of the fraction of non-compliant vehicles.
Estimate the excess emissions associated with non-compliant vehicles in the
Lima-Callao fleet. (Note: this step is not needed for assessing the total PM2 5-
attributable mortality burden.)

•	Step 3: Air quality modeling. Obtain and process air quality data, such as
satellite-based estimates and data from air quality monitors to characterize
baseline conditions in Lima-Callao. Use air quality modeling methods to estimate
the impact of vehicle emissions on ambient PM2 5 concentrations.

•	Step 4: Health impact estimation. Quantify premature deaths associated with
PM2 5 concentrations using BenMAP-CE and relevant datasets, including
population, air quality, baseline mortality incidence, and concentration-response
relationships from the epidemiological literature.

•	Step 5: Valuation. Apply economic valuation estimates to quantified mortality
values to characterize the PM2 5-attributable mortality burden in monetary terms.

These steps are described in greater detail throughout the report.

1.3 REPORT ORGANIZATION

The remainder of this report is organized as follows:

•	In Chapter 2, we briefly summarize our scenario development efforts, including
defining baseline and regulatory air quality conditions needed for estimating
PM2 5-attributable mortality burden.

•	In Chapter 3 we detail our methods for quantifying emissions from the on-road
vehicle fleet, including accounting for empirically-derived estimates of non-
compliance with emissions standards.

•	In Chapter 4, we summarize available air quality data in Lima-Callao and methods
for estimating changes in air quality stemming from emissions changes.

•	In Chapter 5, we describe our methods for conducting health impact estimation
and valuation using USEPA's BenMAP-CE tool, including summaries of key data
inputs such as population, baseline mortality incidence, health impact functions,
and valuation estimates.

INDUSTRIAL ECONOMICS, INCORPORATED

1-3


-------
lEc

•	In Chapter 6, we present the results of our health burden analyses, including an
all-PM2 5 mortality estimate, a transportation-attributable PM2.5 mortality estimate,
and an estimate associated with excess PM2.5 mortality resulting from non-
compliant vehicle emissions.

•	In Chapter 7, we discuss the findings of this research and notable data and
methodological limitations. We then provide recommendations for next steps to
build upon this collaborative research effort.

•	In Appendices A-D, we provide supplemental methods discussions and results
beyond the primary estimates provided in the main text.

INDUSTRIAL ECONOMICS, INCORPORATED

1-4


-------
lEc

CHAPTER 2 | SCENARIO DEVELOPMENT

In this chapter, we define the scenarios assessed in the remainder of the report to estimate
the number of premature deaths attributable to PM2.5 concentrations in Lima-Callao. As
described in Chapter 1, a key element of scenario development involves defining
temporal and geographic scope. First, we focus on estimating recent mortality burden
associated with ambient PM2.5. Therefore, we employ datasets that best characterize
conditions in recent years, including air quality, population, and baseline mortality
incidence.1 Second, we consider emissions, air quality, and associated health impacts in
the Lima-Callao region. We work to employ spatially resolved datasets and report results
at fine geographic resolutions (e.g., districts) where data allow.2 Pollutant emissions, air
quality, and health impacts outside of Lima-Callao are not considered in our analysis.

For regulatory benefit-cost analysis, we typically define business-as-usual and regulatory
scenarios. In a forward-looking analysis of a proposed regulation, the business-as-usual
scenario reflects conditions as they are now (or are expected to be in the future) without
the proposed emissions control measures in place. The regulatory scenario reflects
expected conditions now or in the future if the proposed regulation is implemented. In the
context of burden analyses—the focus of this report—we similarly define baseline and
control scenarios. The baseline scenario reflects observed, recent PM2 5 concentrations in
the region. The control scenarios are hypothetical representations of what recent PM2.5
concentrations would be absent contributions from some or all emissions sources. While
our baseline scenario is the same across our three analyses, the control scenario differs for
each run. These scenarios are summarized in Exhibit 2-1. Importantly, air quality is the
only data input that varies across baseline and control scenarios.

Our estimates do not account for any effects of the COVID-19 pandemic. To the extent that the virus has affected air
quality, population, and baseline death rates in Lima-Callao, these impacts are not quantified in our analysis.

In Peru, administrative divisions are geographically resolved, from largest to smallest, into regions, provinces, and districts.
Lima and Callao are the names of both a province and a district within a province. This analysis encompasses the Lima and
Callao provinces, which are comparable to US states. The districts analyzed are comparable to US counties and range in size
from 1 to 2,800 km2.

INDUSTRIAL ECONOMICS, INCORPORATED

2-1


-------
lEc

EXHIBIT 2-1. DEFINING AIR QUALITY SCENARIOS

ANALYSIS

BASELINE AIR QUALITY

CONTROL AIR QUALITY

Total PM2.5 burden

Recent characterization of
observed PM2.5 concentrations

PM2.5 concentrations set to 0
Mg/m3

Transportation sector
PM2.5 burden

Observed PM2.5 concentrations
minus transportation sector
contributions

Non-compliant vehicles
PM2.5 burden

Observed PM2.5 concentrations
minus contributions from non-
compliant vehicles'

By comparing the estimated health impacts between baseline and control air quality
conditions, we can attribute the mortality burden to various sources. For example, by
comparing recent PM2 5 concentrations with a hypothetical scenario where we reduce
PM2.5 concentrations by the transportation sector's contribution, we can quantify the
mortality burden associated with the sector as a whole. In the following chapters, we
explain how we estimate the contributions of transportation sources—and non-compliant
vehicle emissions alone—to ambient PM2.5 concentrations.

We do not remove all PM2.5-relevant emissions associated with non-compliant vehicles. Rather, we only assess the emissions
in excess of comparable vehicles compliant with vehicle emissions standards.

INDUSTRIAL ECONOMICS, INCORPORATED	2-2


-------
lEc

CHAPTER 3 | EMISSIONS ESTIMATION

In this chapter, we explain our data sources and methods for characterizing emissions
from the transportation sector in Lima-Callao. While MINAM possesses a comprehensive
emissions inventory for on-road vehicles, the inventory assumes that vehicles emit at
fixed emissions rates representing "compliant" rates for a given vehicle class (e.g., bus,
passenger vehicle), fuel type (e.g., gasoline, diesel), and emissions class (e.g., Euro 2,
Euro 6). Yet, inspection data from the Urban Transport Management Office (UTMO) of
the Municipality of Lima indicates a significant fraction of the vehicle fleet is out of
compliance. This may be due to several factors including the potential tampering of
emission control devices, the age of the vehicle fleet, and driving cycles in Lima-Callao.
Therefore, the inventory likely underestimates emissions from the vehicle fleet. Although
the true rate of non-compliance is not known, it is expected to be significant based on
recent data collected by the UTMO. We used these estimates to update the emissions
inventory to account for observed rates of non-compliant emissions in the region so that it
better reflects the true rate of emissions.

3.1 BASELINE EMISSIONS INVENTORY

The calculations in this section detail IEc's implementation of the MINAM transportation
sector emissions inventory model. This model generates annual emissions estimates
based on the size and composition of the region's on-road vehicle fleet. The model
summarizes emissions in tons per year for seven pollutants: PM2.5, NOx, CO, total
hydrocarbon, black carbon, SO2 and CO2. Estimates were derived for each combination
of vehicle type, emissions class, and fuel type (e.g., Euro 2 diesel automobiles).
Additional detail on these vehicle characteristics is summarized in Exhibit 3-1.

EXHIBIT 3-1. TRANSPORTATION EMISSIONS INVENTORY DATA ELEMENTS

VARIABLE

VALUES

Fuel type

Diesel, high octane gasoline, low octane gasoline, liquified
petroleum, natural gas

Emissions class

Pre-Euro, Euro 2, Euro 3, Euro 4

Vehicle type

Automobile, station wagon, pick-up truck, rural truck, panel truck,
omnibus, heavy-duty truck, tow truck, motorcycle

INDUSTRIAL ECONOMICS, INCORPORATED

3-1


-------
lEc

To generate emissions estimates for each vehicle group, the model utilizes information on
vehicle counts, average distance driven per year, and emissions factors to calculate annual
emissions estimates. First, we multiplied vehicle category counts by the fraction of
vehicles with each fuel type to estimate vehicle counts by fuel type and category.4,5 Next,
we multiplied vehicle counts by average annual distance for each fuel type and vehicle
category to estimate total annual distance driven for all vehicles by fuel type and
category.6 We then multiplied the total annual distances by the percentage of vehicles in
each Euro level, specific to fuel type and vehicle category, to determine total annual
distance by fuel type, vehicle category and Euro emissions level.7 Finally, we applied
emissions factors to total distance estimates. Emissions factors reflect the rate of
emissions per unit of distance traveled (e.g., 0.05 grams of PM2 5 emitted per kilometer
traveled). These steps result in 180 emissions estimates (all combinations of nine vehicle
categories, five fuel types, and four Euro levels). Example calculations with hypothetical
values are illustrated in Exhibit 3-2.

EXHIBIT 3-2. ILLUSTRATIVE EMISSIONS ESTIMATION CALCULATIONS

STEP

EXAMPLE

1

100,000 automobiles * 30% diesel fuel use
= 30,000 diesel automobiles

2

30,000 diesel automobiles * 10,000 km avg. distance
= 300,000,000 km traveled by diesel automobiles

3

300,000,000 km * 10% Euro 4
= 30,000,000 km traveled by Euro 4 diesel automobiles

4

30,000,000 km * 0.05 g PM2.5/km
= 1,500,000 g = 1.5 tons PM2.5 emitted annually by Euro 4 diesel automobiles

4

Estimates of vehicle category counts are derived from a sum of the Peruvian National Statistical System (INEI) vehicle
registry from 2011 to 2016. See

https://www.inei.qob.pe/media/MenuRecursivo/publicaciones digitales/Est/Lib1483/cap20/ind20.htm

5

Estimates of the share of vehicles with each fuel type are from the 2012 Ml NAM National Inventory of Greenhouse Gases.
See http://infocarbono.minam.gob.pe/annios-inventarios-nacionales-gei/ingei-2012/

6

Average annual distances are provided by the Climate Change Planning Project. See http://planccperu.org/wp-
content/uploads/2016/05/informe final.pdf. These data are supplemented by information on total distance per year for
diesel and gas vehicles within the automobile category in Lima from Swisscontact.

7

The portion of the Lima vehicle fleet within each Euro-level are provided by a Nationally Appropriate Mitigation Action
(NAAAA) Support Project report. See http://www.transferproiect.org/proiects/transfer-partner-countries/peru/. Vehicles
15 years or older are classified as emitting at the Pre-Euro level, vehicles between the ages of 15 and 12 years are classified
as Euro 2, vehicles between the ages of 11 and one years are classified as Euro 3, and vehicles less than 1 year old are
classified as Euro 4.

INDUSTRIAL ECONOMICS, INCORPORATED

3-2


-------
lEc

In the example above, Euro 4 diesel automobiles are estimated to emit 1.5 tons of PM2.5
annually. These estimates would be compiled with the emissions from the 179 other
vehicle type, fuel type, emissions class combinations to yield total transportation sector
emissions in the region. As noted above, the inventory implicitly assumed perfect
compliance in its application of emissions factors - vehicles cannot emit above (or
below) the rates established by each Euro class. In the section below, we explain how we
adapted these values to account for non-compliance in the vehicle fleet.

3.2 ACCOUNTING FOR NON-COMPLIANCE

Empirical evidence demonstrates that many on-road vehicles in Lima-Callao do not
comply with emissions limits established by MINAM. In a 2017 analysis of 2,625
vehicles in Lima-Callao conducted by the UTMO, roughly half of vehicles tested as out
of compliance with emissions limits.8 Compliance rates, depicted in Exhibit 3-3, varied
by vehicle fuel type.

EXHIBIT 3-3. COMPLIANCE RATES BY FUEL TYPE

100%

80%

60%

40%

20%

0%

Diesel	Gasoline	GLP	GNV

¦ Pass aFail

Notes: GLP = liquified petroleum, GNV = natural gas, n = sample size.

For diesel vehicles, compliance is determined using an opacity standard and for gasoline, natural gas and liquefied
petroleum compliance is based on combined CO, CO2, and HC standards.

INDUSTRIAL ECONOMICS, INCORPORATED

3-3


-------
lEc

Compliance rates were found to range from 40 percent for liquified petroleum vehicles to
66 percent for gasoline vehicles. We used these data to adjust the emissions inventory
estimates to account for emissions rates that likely exceed those quantified in the model's
emissions factors. For each fuel type, we divided total annual distances (by vehicle
category, fuel type and emissions class) into compliant and non-compliant designations.
For compliant vehicles, we applied the conventional emissions factors associated with the
vehicle category, fuel type, and emissions class. Non-compliant vehicles, however, were
assumed to emit at higher rates (i.e., more pollution per kilometer driven). Given
uncertainty in the true emissions rates of non-compliant vehicles, we estimated a lower-
and upper-bound emissions estimate using alternative assumptions:9

•	Lower bound emissions estimate: Non-compliant vehicles were assumed to emit
at one emissions standard older than previously assigned. For example, a non-
compliant Euro 4 vehicle was assumed to emit at a rate consistent with Euro 3
emissions factors.

•	Upper bound emissions estimate: All non-compliant vehicles were assumed to
emit at the Pre-Euro level.

Notably, IEc only adjusted emissions for PM2.5 and NOx, i.e., the precursors of ambient
PM2.5 to account for non-compliance. We were unable to adjust for SO2 emissions as the
data were not broken out by Euro-level. However, it is reasonable to assume that
compliant and non-compliant vehicles emit comparable SO2 per kilometer traveled (for a
given fuel type and vehicle type) as SO2 emissions are a function of distance traveled and
sulfur content in fuel—not emissions class.

We summed the non-compliant and compliant emissions for each vehicle category and
Euro-level to determine annual emission estimates of PM2.5 and NOx for each fuel type.
Finally, IEc summed across fuel type to determine estimates of total annual PM2 5 and
NOx emissions.

3.3 EMISSIONS MODELING RESULTS

The results of emissions inventory model are summarized in Exhibit 3-4.

In estimating non-compliance rates, we assumed a random sampling of vehicles and accuracy of the inspection testing
method by the UTMO. We explored our assumption that non-compliance corresponded to older Euro standards of emissions
using the following sensitivity analysis. IEc assessed the distributions of opacity inspection measurements for diesel vehicles
and the distributions of CO measurements for gasoline, natural gas and liquified petroleum vehicles. The percent increase
in opacity for compliant to non-compliant diesel vehicles was comparable to the percent increase in PM2.5 emissions for
changing from the Euro 4-2 level to the Pre-Euro level. Therefore, IEc determined that accounting for non-compliant
vehicles in the inventory model by scaling diesel emissions factors by opacity would likely not change the result
significantly. Additionally, the percent increase in CO for compliant to non-compliant gasoline, natural gas and liquified
petroleum vehicles was orders of magnitude higher than the percent increase in PM2.5 emissions for changing from the Euro
4-2 level to the Pre-Euro level. IEc determined that scaling gasoline, natural gas and liquified petroleum emissions factors
by CO measurements to account for non-compliance would not be appropriate.

INDUSTRIAL ECONOMICS, INCORPORATED

3-4


-------
lEc

EXHIBIT 3-4. 2018 VEHICLE EMISSIONS ESTIMATES FOR PM2 5, NOx, AND S02

SCENARIO

TOTAL VEHICLE EMISSIONS (TONS/YEAR)

PM2.5

NOx

S02

Full Compliance

4,092

164,038

21,908

Non-

Compliance
Adjusted

Upper
Bound

5,176

203,564

21,908

Lower
Bound

4,297

175,097

21,908

Implementing the MINAM emissions model assuming perfect compliance results in
estimates of PM2.5 emissions of 4,092 tons per year, NOx emissions of 164,038 tons per
year, and SO2 emissions of 21,908 tons per year. We found that accounting for non-
compliance resulted in a 5 to 26 percent increase in PM2.5 emissions and a 6 to 24 percent
increase in NOx emissions (Exhibits 3-5 and 3-6). SO2 emissions were not adjusted and
therefore do not vary between the full compliance and non-compliance adjusted
scenarios. Appendix A provides greater detail on these estimates, including the share of
emissions by vehicle fuel type.

EXHIBIT 3-5. ESTIMATED PRIMARY PM25 EMISSIONS, 2018

6,000









5,000













Emissions (tons/year)

J\J JJJ

0 0 0
000
000

























































1,000





























0















Full Compliance

Upper Bound

Lower Bound

INDUSTRIAL ECONOMICS, INCORPORATED

3-5


-------
lEc

EXHIBIT 3-6. ESTIMATED N0X EMISSIONS, 2018

250,000

200,000

<5

150,000

C

o

c

¦£ 100,000

50,000

Full Compliance

Upper Bound

Lower Bound

INDUSTRIAL ECONOMICS, INCORPORATED

3-6


-------
CHAPTER 4 | AIR QUALITY DATA AND MODELING

In this chapter, we describe our approach to characterizing recent PM2.5 concentrations in
Lima-Callao. First, we describe available air quality monitor data and satellite-derived
PM data in the region, as well as our approaches to fuse these data sources. Second, we
outline our approach to estimating the contribution of transportation sources to ambient
PM2.5 concentrations based on the emissions data described in the previous chapter.

4.1 AIR QUALITY SURFACES

Several data sources provide estimates of ambient PM2.5 concentrations in Lima-Callao.
We describe these air quality surfaces below and weigh the relative benefits and
limitations of each data source.

4.1.1 MONITOR SURFACES

Monitor data in Lima-Callao were provided by MINAM covering a range of pollutants
(PM2.5, PM10, SO2, NO2, O3, and CO) and temporal scales. Pollutant concentrations were
summarized at hourly, daily, monthly, and annual timesteps since 2000.10 The monitors
analyzed are owned and operated by two separate agencies: MINAM and the National
Meteorology and Hydrology Service of Peru (SENAMHI).

To best depict the air quality in the Lima-Callao metropolitan area, including "hot spots"
of concern to MINAM (e.g., localized high pollutant concentrations in Callao), we used
hourly PM2.5 measures from 10 monitors in 2019 and supplemented these values with
daily PM2.5 measures from two active samplers in the Callao district.1112 We combined
the measures into a single dataset by converting hourly PM2.5 concentrations to daily
averages for each 24-hour period.13

Exhibit 4-1 maps these air quality monitor stations in Lima-Callao. Since monitor data
provides concentrations at a fixed location, we used the Voronoi Neighborhood
Averaging (VNA) method in BenMAP-CE to interpolate PM2 5 concentrations at a 1km x

Some pollutants and metrics are only available in select years.

11

Daily monitoring data in Callao are only available for approximately one week of each month in 2019 from the
Environmental Assessment and Enforcement Agency (OEFA).

12

Additional monitors provide data in earlier years (e.g., 27 in 2016), but these data risk reflecting older air quality levels
and distributions.

13

After converting the hourly data to daily averages, we excluded eight outlier daily values for three monitors.

INDUSTRIAL ECONOMICS, INCORPORATED

4-1


-------
lEc

lkm grid (not pictured). Hie VNA method calculates an inverse-distance weighted

14

average for each grid cell from the monitors surrounding the grid s center.

EXHIBIT4-1. AVAILABLE AIR QUALITY MONITORING STATIONS IN LIMA-CALLAO (2019)

: arabayllo

Rjfcntf Pi**jrn
~ ~

RicatJo P atma

Lurigancho

Lnnoqarho *
RijTjnrt X

Linj/> . r

_lkiVicToria

* Anita

~ ~

Santiago de
Surc6

'San ' V,lla
J(ian dp Matja

Mir a floret *
Triunfo

Villa El
Salvador

~ Monitor Locations
Province

Callao
Lima

See Appendix B in the BenMAP-CE user manual for a detailed discussion of VNA methods:

https://www.epa.gov/sites/product!on/files/2015-04/documents/benmap-ce user manual march 2015.pdf

INDUSTRIAL ECONOMICS, INCORPORATED

4-2


-------
4.1.2 SATELLITE SURFACES

In areas lacking monitor coverage, satellite data can be useful for estimating ambient
surface concentrations of PM2.5. In areas with robust monitoring networks, those data are
likely to best represent ground-level ambient concentrations (assuming appropriate
quality control procedures are followed); however, satellite data still play an important
role by filling gaps between monitored locations and providing information on the spatial
distribution of PM2.5 at finer resolutions. Lima-Callao has an established network of
monitors; however, coverage is more limited in outer districts. Therefore, we leveraged
two satellite-based estimates for the Lima-Callao metropolitan area: an estimated surface
from van Donkelaar et al. (2016) and an estimated surface from Shaddick et al. (2017).
The van Donkelaar surfaces provide annual estimates of PM2 5 at 0.01° resolution (1km x
lkm) for 1998-2016 and the Shaddick surfaces provide annual estimates of PM2.5 at 0.1°
(10km x 10km) resolution for 2014 and 2016.

The van Donkelaar et al. (2016) and Shaddick et al. (2017) surfaces combined
information from satellites, model simulations and ground-level monitors. Their methods
are explained in further detail in Appendix B. Notably, both surfaces incorporate monitor
data from the WHO Global Ambient Air Quality Database, which contains only one
monitor with a directly measured estimate for PM2 5 for Lima-Callao.15 As such, while the
surfaces may provide insight into the spatial distribution of air pollution, the magnitude of
PM2.5 concentrations may not accurately reflect real-world conditions (as measured by
monitors).

Therefore, IEc performed additional local calibration of the van Donkelaar and Shaddick
surfaces for 2016 using data from twelve SENAMHI and MINSA monitoring stations.
Satellite surface calibration was broken into four steps: (1) calculating annual PM2.5
averages at the monitor locations, (2) calculating the ratio between monitor and satellite
annual PM2.5 averages, (3) spatially interpolating the ratios to create a calibration surface,
and (4) multiplying the calibration surface against the satellite surface to create a locally
calibrated air quality surface. These steps, implemented in ArcMap version 10.4.1 using
the Spatial Analyst package, are explained in greater detail in Appendix B.

4.1.3 SUMMARY OF AIR QUALITY SURFACES

Exhibit 4-2 displays the three final surfaces used to assess the mortality burden of
ambient PM2.5 in Lima-Callao. Exhibit 4-2 (a) shows 2019 monitor data interpolated to a
lkm x lkm grid. The 2019 monitor surface had an average daily PM2.5 concentration of
33.0 (ig/m3, with a minimum and maximum observed concentration of 16.2 (ig/m3 and
46.9 (ig/m3, respectively. Exhibit 4-2 (b) is the 10km x 10km 2016 Shaddick model
surface, locally calibrated using monitor data (see Appendix B). The Shaddick model
surface had an annual average PM2.5 concentration of 21.4 (ig/m3, with a minimum and

World Health Organization. WHO Global Ambient Air Quality Database (Update 2018); WHO: Geneva, 2018:
https://www.who.int/airpollution /data/cities/en/

INDUSTRIAL ECONOMICS, INCORPORATED

4-3


-------
lEc

EXHIBIT 4-2. AIR QUALITY SURFACES

a) 2019 Monitors

16

Lima-Callao Districts
PM2.5 Concentration (ng/m3)

< 23.6
23.7 - 28.6
28.7-32.1
32.2 - 34.0
34.1 - 35.9
36.0 - 39.2
¦ > 39.3

b) 2016 Shaddick Model
(locally calibrated)



Lurigancho
Chadaeayo

Lima _ -
tfi-VKtona

Juan Jo Ma"a
Miralaes



c) 2016 Van Donkelaar Model
(locally calibrated)

If#

Luiiumidw



There are numerous ways one could interpolate between monitors to develop an air quality surface. Exhibit 4-2a was developed using the BenMAP-CE default interpolation
procedure, Voronoi Neighborhood Averaging (VNA). The VNA interpolation method may or may not accurately reflect ground-level conditions in Lima-Callao.

INDUSTRIAL ECONOMICS, INCORPORATED

4-4


-------
lEc

maximum observed concentration of 11.8 (ig/m3 and 36.4 (ig/m3, respectively. Lastly,
Exhibit 4-2 (c) is the 1km x 1km van Donkelaar model surface, calibrated using monitor
data (see Appendix B). The van Donkelaar model surface had an annual average PM2.5
concentration of 21.6 (ig/m3, with a minimum and maximum observed concentration of
10.5 (ig/m3and 53.5 (ig/m3, respectively.

4.2 AIR QUALITY MODELING

In the previous section, we presented data sources and methods used to characterize
recent concentrations of PM2.5 in the Lima-Callao region. The resulting air quality
surfaces are used to assess the mortality burden associated with all sources of PM2.5. To
assess the transport-attributable mortality burden, we needed a means of quantifying the
effect of transportation emissions on ambient PM2 5 concentrations. We employed a
reduced-form air quality modeling technique employed in past studies commissioned by
MINAM: emissions concentration factors (FECs, from the Spanish factor emisidn-
concentracion). The FECs used in this report were developed for vehicle emissions in the
Valparaiso region of Chile.17 While FECs exist for Lima, they are not specific to the
transportation sector. After consultation with MINAM, we elected to use the Valparaiso
FECs, as this coastal and mountainous region may be similar to the Lima-Callao
Metropolitan area in important meteorological and topographical respects. FECs are
modeled by the following equation:

6Cf\_1 Ef

FEC' [sEy ~ Cf

where FECf is the emission-concentration factor in zone /' for year t in tons/((.ig/ni3). Cf is
the ambient concentration of PM2 5 in zone /' for year t in (ig/m3, and Ef is pollutant
emissions in zone /' for year t in tons. Pollutant-specific FECs for vehicle emissions are
shown in Exhibit 4-3.

EXHIBIT 4-3. POLLUTANT-SPECIFIC EMISSION-CONCENTRATION FACTORS (FEC)

POLLUTANT

FEC

(TON/YEAR PER Mg/m3)

PM2.5

1,148.106

S02

15,220.700

NOx

18,867.925

17

GreenLab, 2011: https://silo.tips/download/estudio-co-beneficios-de-la-mitigacion-de-gei

INDUSTRIAL ECONOMICS, INCORPORATED

4-5


-------
lEc

We estimated the total contribution to ambient concentrations from primary PM2 5 from
NOx, SO2 and PM2 5 vehicle emissions according to the following equation:

°p = (:FEC~) * Ep'

where C is the contribution to ambient PM2.5 concentrations of pollutant p, FECP is the
emission-concentration factor for pollutant p, and Ep are the primary emissions for
pollutant p. We estimated contributions to ambient PM2.5 concentrations for the full
compliance, non-compliance upper bound, and non-compliance lower bound scenarios.
Finally, we isolated the contribution of non-compliant emissions by subtracting the full
compliance total emissions from the non-compliant vehicle emissions, such that:

Cnc Cnc adjusted. ^full

where C„c are non-compliant emissions, C„c adjusted are total vehicle emissions adjusted for
non-compliance, and Cf„n are total vehicle emissions assuming full compliance. This
estimated non-compliant emissions rather than the total emissions of non-compliant
vehicles. Specifically, a non-compliant vehicle has a portion of emissions that are
compliant (i.e. would have still been emitted if they met standards) and a portion of
emissions that are in exceedance of the standard. To address the mortality burden of non-
compliance, we determined the contribution of only the excess emissions. Results of FEC
calculations are shown in Exhibit 4-4.

EXHIBIT 4-4. TRANSPORTATION AND NON-COMPLIANT VEHICLE EMISSION CONTRIBUTIONS TO
AMBIENT PM2.5 CONCENTRATIONS

SCENARIO

CONTRIBUTION OF VEHICLE EMISSIONS TO AMBIENT
PM2.5 CONCENTRATIONS (Mg/m3)

CONTRIBUTION
OF NON-
COMPLIANT
EMISSIONS

PM2.5

NOx

S02

TOTAL

Full Compliance

3.56

8.69

1.44

13.70

N/A

Non-

Compliance
Adjusted

Upper
Bound

4.51

10.79

1.44

16.74

3.04

Lower
Bound

3.74

9.28

1.44

14.46

0.76

Exhibit 4-5 shows the relative contributions of the transport sector (compliant and non-
compliant emissions) and other sources to ambient concentrations of PM2.5. We estimated
the contribution of other sources, both anthropogenic and non-anthropogenic, by

INDUSTRIAL ECONOMICS, INCORPORATED

4-6


-------
lEc

subtracting the total vehicle contribution to ambient PM2 5 concentrations from the 2016
annual average of ambient PM2.5 concentrations in Lima-Callao18.

The annual average PM2.5 concentration for 2016 was 25.0 (ig/m3. Vehicle emissions
contribute between 67 and 58 percent (upper and lower bounds) of ambient PM2.5
concentrations and non-compliant emissions contribute between 12 and 3 percent (upper
and lower bounds).

EXHIBIT 4-5. CONTRIBUTIONS OF TRANSPORT SECTOR AND OTHER SOURCES TO 2016 AVERAGE
PM2 5 CONCENTRATIONS

25.0

20.0

15.0

10.0

5.0

0.0

0.8







3.0















13.7













13.7





10.6



8.3



Lower	Upper

I Other Sources ¦ Compliant Emissions ¦ Non-compliant Emissions

We note that the PM2.5 contributions displayed in Exhibit 4-5 reflect mean effects in the
region. While this analysis evaluates transportation emissions impacts and contributions
to air pollution at the regional and district level, research has shown that air pollution can
be significantly higher within a short distance of large roadways and other transportation
facilities, especially within the first 150-300 meters, compared with district-level air
pollution concentrations (Karner et al., 2010). Individuals living, working and going to
school within this short distance of roadways have increased risks for adverse health
effects, including premature mortality (Health Effects Institute 2010). As a result, relying
on district-level air quality analyses may underestimate the impacts of transportation
sources on the overall mortality burden to the population of Lima-Callao.

The annual average was estimated in BenMAP-CE using 2016 monitor data and is population- and spatially-weighted. The
other sources category is expected to include both anthropogenic and non-anthropogenic sources of emissions.

INDUSTRIAL ECONOMICS, INCORPORATED

4-7


-------
lEc

CHAPTER 5 | MORTALITY BURDEN ESTIMATION AND VALUATION

In this chapter, we detail our methods for assessing the mortality burden associated with
PM2.5 concentrations in Lima-Callao. In total, we performed three health benefits
analyses:

1.	The total mortality burden of current PM2.5 concentrations in Lima-Callao;

2.	The contribution of transportation emissions to the total mortality burden; and

3.	The contribution of non-compliant vehicle emissions to the total mortality
burden.

For these analyses, we used the USEPA's Environmental Benefits Mapping and Analysis
Program - Community Edition (BenMAP-CE) version 1.5.2.0, an open-source program
that quantifies and values the adverse health effects associated with changes in pollutant
concentrations. The remainder of this chapter provides an overview of our approach,
including our data sources for key inputs such as population, baseline incidence rates, and
concentration-response functions from the epidemiological literature. Finally, we provide
an overview of our valuation approach.

5.1 OVERVIEW OF APPROACH

We used BenMAP-CE to estimate the impact of PM2.5 concentrations on premature
mortality by assessing the difference in the risk of those endpoints under the baseline and
control scenarios presented in Chapter 2. BenMAP-CE relies on health impact functions
to quantify the change in incidence of adverse health impacts stemming from changes in
ambient pollutant concentrations:

Ay = y0 ¦ (l —	¦ Pop

where Ay is the change in the incidence of the adverse health effect, y0 is the baseline
incidence rate for the health effect, beta (/?) is a coefficient derived from a relative risk
(RR) estimate associated with a change in exposure (i.e., pollutant concentration) as
expressed in concentration-response functions, APM is the change in concentrations of

19

fine particulate matter, and Pop is the exposed population.

19 Based upon the functional form of the underlying concentration-response function, the functional form of the health
impact function may differ. A PM may also be replaced by concentrations of other pollutants (e.g., ozone) or conditions
(e.g., temperature).

INDUSTRIAL ECONOMICS, INCORPORATED

5-1


-------
5.2

DATA INPUTS

We drew upon multiple data sources to parameterize and implement the generic health
impact function presented above. These data sources are described below.

5.2.1	POPULATION

MINAM provided district-level population data from the Peruvian National Statistical
System (INEI) for the period 2005 to 2015, as well as national projected population data
from 1950 through 2070 in five-year increments. Both population datasets include age
stratification into five-year age bins. To account for population growth since 2015 (the
most recent year with district-level population estimates), we projected the 2015 district-
level population to the year 2020 by applying age-specific national growth rates. Finally,
we formatted these data for use in BenMAP-CE.

5.2.2	BASELINE MORTALITY INCIDENCE

To characterize baseline rates of death, we processed data from the Peruvian Ministry of
Health (MINSA) National Center for Epidemiology, Prevention and Control of Disease
(CNEPCE). These data include counts for a range of mortality and morbidity endpoints
from 1986 through 2016. Available mortality data are reported by district, gender, and
five-year age increments. For this analysis, we focused on the mortality incidence for the
following endpoints: Ischemic Heart Disease (IHD), Acute Lower Respiratory Infection
(LRI), Chronic-Obstructive Pulmonary Disease (COPD), Lung Cancer, Cerebrovascular
Disease, and Natural Causes (hereafter referred to as Non-Communicable Diseases
(NCD) plus LRI). These endpoints were selected to match the health impact functions
described in the following section.

Prior to use in BenMAP-CE, we converted mortality count data to incidence rates (cases
per person per year). First, we formatted the mortality data to align with the level of
aggregation in the population dataset (year, district, endpoint, and age group). We then
divided the counts by the district- and age-specific population for corresponding years.
To minimize variability across years, we estimated incidence rates for a five-year period
(2011 to 2015). In some cases, we aggregated incidence rates from various causes to align
with the endpoint definitions in the health impact functions (described below). For
example, Tapia et al. (2020) reflects respiratory and circulatory mortality.

5.2.3	HEALTH IMPACT FUNCTIONS

As described above, health impact functions provide the quantitative framework to
estimate changes in health outcomes resulting from changes in pollutant concentrations,
incorporating data on population and baseline incidence. These functions are derived
from concentration-response relationships published in epidemiological research, which
provide insight into the strength of a pollutant's effect on health. For example, a study
may suggest that for every 10 (ig/m3 change in PM2 5, we can expect baseline mortality
incidence to change by 6 percent. Exhibit 5-1 summarizes our selected health impact
functions for assessing mortality burden.

INDUSTRIAL ECONOMICS, INCORPORATED

5-2


-------
lEc

EXHIBIT 5-1. BENMAP-CE HEALTH IMPACT FUNCTIONS

AUTHOR

MORTALITY ENDPOINT GROUP

AGES

Burnett et al. (2018)

Non-communicable diseases plus lower respiratory
infection (NCD + LRI)

25-99

Cerebrovascular disease

25-99

Chronic obstructive pulmonary disease (COPD)

25-99

Ischemic heart disease (IHD)

25-99

Lung cancer

25-99

Lower respiratory infection (LRI)

25-99

Tapia et al. (2020)

All respiratory and circulatory

0-99

Two epidemiological studies provide the concentration-response relationships
summarized in Exhibit 5-1. First, we utilized the Global Exposure Mortality Model
(GEMM) health impact functions pre-loaded into BenMAP-CE. The GEMM is a family
of functions developed by Burnett et al. (2018) to estimate the global burden of disease
attributable to PM2.5 exposure over the entire global exposure range. The GEMM consists
of risk functions for six mortality endpoints: NCD + LRI, Cerebrovascular Disease,
COPD, IHD, Lung Cancer, and LRI.2" Notably, the GEMM is a meta-analytic function
developed based on high-quality PM2.5 studies conducted globally. Additionally, the
functions are non-linear. That is, the strength of the effect of PM2.5 on premature deaths
depends upon the observed PM2.5 concentrations. In general, the functions suggest that
the marginal effect of PM2.5 lessens at higher concentrations.

Second, we utilized Tapia et al. (2020) estimates of PM2 5-attributable respiratory and
circulatory mortality.21 While these estimates are specific to Lima, Peru, the study

While age-specific GEMM functions are presented in Burnett et al. (2018), we leverage the all-ages (25-99) functions to
capture population-wide effects. Similarly, Burnett et al. (2018) provide estimates with and without a Chinese male cohort
included in their meta-analysis. We leverage the GEMM functions with the Chinese male cohort because these estimates
are, in part, informed by higher PM2.5 exposure levels experienced by the Chinese cohort. These higher concentrations may
be relevant to air quality conditions in Lima-Callao.

21

We identified the Tapia et al. (2020) study by conducting a literature review for PM2.5 epidemiological studies local to
Lima-Callao, Peru, or South America using a broad keyword search in Google Scholar and PubMed. We identified 17
potentially relevant papers and abstracts, then narrowed this list to three potential candidates for health impact
calculation in BenMAP-CE: Hansel et al. (2018), Tapia et al. (2019) and Tapia et al. (2020). The three studies were selected
due to their relevant respiratory and cardiovascular endpoints. Hansel et al. (2018) provided functions that relate PM2.5
exposure to asthma morbidity, including uncontrolled asthma, adverse asthma-related quality of life, health care utilization
and missed school days. Tapia et al. (2019) provided functions that relate PM2.5 exposure with cardiorespiratory emergency
room visits and Tapia et al. (2020) provided functions that relate PM2.5 exposure with cardiorespiratory mortality. We did

INDUSTRIAL ECONOMICS, INCORPORATED

5-3


-------
lEc

assesses mortality associated with short-term PM2 5 exposures. As such, we expect the
study to understate the total impacts of PM2.5 due to strong empirical evidence that much
of the pollutant's effect is associated with its long-term exposure.

5.2.4 VALUATION

We valued mortality using a value per statistical life (VSL) estimate from Robinson et al.
(2018). The VSL represents individuals" willingness to pay (WTP) for incremental
reductions in their annual risk of death. While we are unaware of any WTP studies
conducted in Peru, the Robinson et al. provides methods and results for transferring VSL
values to countries without primary estimates. The authors synthesize available estimates
in other countries and transfer these values to Peru, among many other countries, by
accounting for differences in per capita income, a key factor influencing WTP and VSL.22
The resulting value, $1.21 million, is applied to premature deaths to reflect the welfare
loss associated with PM2 5-attributable mortality. As we note in the subsequent chapters,
this estimate is associated with some uncertainty, and alternative valuation methodologies
may result in different estimates. For example, MINAM's analysis of the Euro 6
standards included a range of estimates from $0.14 to $1.6 million (2017$).23

not apply the other functions identified as they are for morbidity endpoints and this report focuses solely on the mortality
health burden.

22	.

Robinson et al. (2018) extrapolate a Peru-specific VSL from the OECD VSL base ($3 million) using the ratio of gross national

income per capita between Peru and OECD countries.

23	.

MINAM included a value of $0,138 million derived by "Seminario de Marzi (2017)", which employed a human capital

approach for valuing averted deaths. This approach values mortality effects by considering the labor productivity of
individuals and is frequently thought to underestimate individual's true willingness to pay to reduce their risk of death.

INDUSTRIAL ECONOMICS, INCORPORATED

5-4


-------
lEc

CHAPTER 6 | RESULTS

In this chapter, we present the results of our mortality burden analyses using the methods
and data sources described in previous chapters. First, we present the quantified and
valued mortality impacts associated with total PM2.5 concentrations in Lima-Callao.
Second, we present results specific to the transportation sector. Finally, we estimate the
burden associated with excess emissions from non-compliant vehicles in the region. In
Appendix D, we present these results stratified by Lima-Callao District.

6.1 TOTAL PM2. 5 ATTRIBUTABLE MORTALITY BURDEN

We estimate that ambient PM2.5 emissions in Lima-Callao result in over 10,000 deaths
annually. These results, summarized in Exhibit 6-1, vary depending on the selected health
impact function and baseline air quality surface. We highlight the GEMM NCD & LRI
results as our preferred estimates. This HIF captures a broader range of air pollution
attributable deaths relative to the five cause-specific (5 COD, i.e., cause-of-death)

GEMM estimates and the local Tapia et al. (2020) results. Further, the Tapia et al. study
only accounts for short-term exposure to PM2.5 and ignores the substantial long-term
mortality impacts of air pollution. Results are largely stable across air quality surfaces,
with the greatest burden resulting from the 2019 monitor data (12,016 deaths) relative to
the Shaddick and van Donkelaar surfaces (10,556 and 10,838, respectively). For the
remainder of this report, we present the Shaddick surface and GEMM NCD & LRI results
as our primary estimates. We highlight the Shaddick surface results because Shaddick
contains more recent data and was generated using a WHO model which builds upon
earlier van Donkelaar methods. The model estimates the spatially varying relationship
between ground measurements of PM2 5 and factors from the various air quality models
(see Appendix B for details).

EXHIBIT 6-1. ESTIMATED PM2 5 ATTRIBUTABLE MORTALITY BURDEN



PREMATURE DEATHS







VAN

CAUSE OF MORTALITY

MONITORS

SHADDICK

DONKELAAR

GEMM: NCD + LRI

12,016

10,556

10,838

GEMM: 5 COD

7,425

6,517

6,514

Lower respiratory infection

4,022

3,538

3,531

Ischemic heart disease

1,486

1,321

1,338

INDUSTRIAL ECONOMICS, INCORPORATED

6-1


-------
lEc

CAUSE OF MORTALITY

PREMATURE DEATHS

MONITORS

SHADDICK

VAN
DONKELAAR

Cerebrovascular disease

987

843

841

Lung Cancer

590

524

508

COPD

339

292

296

Tapia: Respiratory & circulatory (short-term
exposure)

1,486

1,245

1,272

Total PM2.5 attributable deaths, estimated at 10,556, represent the annual toll of air
pollution in the region. To the extent that air quality, population, and baseline rates of
death are relatively comparable over time, we expect that these adverse impacts are likely
to occur each year. The costs associated with these deaths amount to $12.7 billion
annually (2015$).24 Notably, these costs are irrespective of source: both anthropogenic
(e.g., industry, transportation) and non-anthropogenic (e.g., sea salt, crustal dust) sources
contribute to the total ambient PM2.5 concentrations in the region. In the following
sections, we present the burden associated with transportation sources.

6.2 TRANSPORT ATTRIBUTABLE PM2.5 MORTALITY BURDEN

Emissions from on-road vehicles in Lima-Callao result in 5,150 to 6,200 premature
deaths each year. These results are summarized in Exhibit 6-2 along with the associated
economic costs. It is important to note that the contribution of vehicle emissions to
ambient PM2.5 assumes there is some level of non-compliance in meeting emissions
standards (i.e. neither estimate represents 100 percent compliance with emissions
standards). Additionally, the variance in vehicle emissions contribution is solely due to
the variance in non-compliant vehicle emissions.

EXHIBIT 6-2. ESTIMATED PM2 5 ATTRIBUTABLE ANNUAL MORTALITY BURDEN, TRANSPORTATION
SECTOR

VEHICLE EMISSIONS
CONTRIBUTION

ANNUAL PM2.5 ATTRIBUTABLE
DEATHS (SHADDICK)

ANNUAL ECONOMIC COSTS
(2015$, BILLIONS)

Lower bound

5,150

$6.2

Upper bound

6,200

$7.5

24

Using the range of VSL estimates in MINAM's Euro 6 analysis, the total monetized mortality burden in Peru may range from
$1.5 billion (VSL = $0.14 million) to $16.9 billion (VSL = $1.61 million).

INDUSTRIAL ECONOMICS, INCORPORATED

6-2


-------
lEc

The estimated mortality burden for transportation sources amounts to over half of the
total PM2.5 mortality burden, reflecting the sector's outsized influence on air pollution in
the region. The range in mortality estimates reflects uncertainty in the exact emissions
rates of non-compliant vehicles. As summarized in Exhibit 6-2, the economic costs
associated with transportation-attributable PM2.5 concentrations are $6.2 to $7.5 billion
annually. As we describe in Appendix C, these estimates may understate the burden of
transportation sources due to the non-linearity of the GEMM function. Additionally, as
noted in Chapter 4, these results may not fully capture local-scale effects, including
exposures to significant greater PM2.5 concentrations on or near roadways.

6.3 PM2.5 MORTALITY BURDEN FROM NON-COMPLIANT VEHICLES

As discussed in Chapter 3, MINAM emissions measurements suggest roughly half of all
on-road vehicles in Lima-Callao are out of compliance with emissions standards. We
adjust emissions inventories accordingly and estimate the resulting air quality and health
impacts. In total, we find that emissions from non-compliant vehicles in excess of
emissions standards are responsible for 248 to 991 deaths annually. These results are
presented in Exhibit 6-3. Excess emissions (i.e., above and beyond compliant emission
levels) from non-compliant vehicles account for roughly 5 to 16 percent of the
transportation mortality burden (comparing Exhibits 6-2 and 6-3).

EXHIBIT 6-3. ESTIMATED PM25 ATTRIBUTABLE ANNUAL MORTALITY BURDEN, NON-COMPLIANT
EMISSIONS

VEHICLE EMISSIONS
CONTRIBUTION

ANNUAL PM2.5 ATTRIBUTABLE
DEATHS (SHADDICK)

ANNUAL ECONOMIC COSTS
(2015$, MILLIONS)

Lower bound

248

$300

Upper bound

991

$1,200

As noted above, the range in mortality estimates reflects uncertainty in the exact
emissions rates of non-compliant vehicles. The associated economic costs for non-
compliant vehicle emissions amounts to $0.3 to 1.2 billion annually. Policymakers may
interpret these estimates as the potential annual benefits that may be achieved through
regulatory measures that address the entirety of non-compliant emissions. That is,
achieving perfect compliance with vehicle emissions standards would be expected to
avoid 248 to 991 deaths annually, resulting in annual benefits of $0.3 to 1.2 billion. While
perfect compliance may be infeasible, the mortality burden associated with non-
compliant vehicle emissions is sizeable—any policy that materially improves compliance
will produce significant public health benefits in Lima-Callao.

INDUSTRIAL ECONOMICS, INCORPORATED

6-3


-------
CHAPTER 7 | DISCUSSION AND NEXT STEPS

In the previous chapters, we described our analytic approach and findings in detail. In this
chapter, we summarize our findings and discuss their implications. We also suggest
possible next steps for researchers and MINAM staff

7.1	SUMMARY OF FINDINGS

Our mortality burden analyses provide evidence that PM2 5 concentrations in the Lima-
Callao region represent a substantial public health concern. Overall, ambient
concentrations typically range from 11.8 to 36.4 (ig/m3 on an annual basis, with some
"hot spots" likely to experience markedly higher pollution concentrations, particularly at
shorter time and distance scales. These values exceed the WHO annual guideline of 10
(.ig/nr\ In total, we estimate that over 10,000 deaths each year result from PM2 5 exposure
in Lima-Callao. The economic costs of this loss amount to $12.8 billion USD annually.

The transportation sector represents a major contributor to ambient PM2 5—and premature
mortality—in the region. We expand upon the overall PM2 5 burden analysis by more
closely evaluating the mortality burden associated with PM2 5 concentrations originating
from on-road vehicles. We estimate that 14.5 to 16.7 (ig/m3 of ambient PM2 5
concentrations (58 to 67 percent) result from on-road transportation emissions in the
region. These emissions result in 5,150 to 6,200 premature deaths annually, equating to
$6.2 to $7.5 billion in economic costs.

Regulatory interventions in Lima-Callao may lessen the mortality burden associated with
transportation emissions. We highlight the role that non-compliant vehicles play in
regional emissions by adjusting available emissions inventories using recent MINAM
estimates of non-compliance rates (roughly 3 to 12 percent) in the region. We find that 1
to 3 (ig/m3 in the region may be explained by emissions from this subset of vehicles. The
wide range in estimates stems from uncertainty in quantifying emissions from these
vehicles. Emissions from non-compliant vehicles in excess of federal vehicle emissions
standards result in 250 to 990 premature deaths annually, equating to $0.3 to $1.2 billion
in economic costs. These costs may be reduced through regulatory measures, such as
increased enforcement or enhanced I&M programs.

7.2	UNCERTAINTIES

The results presented in this report are accompanied by numerous sources of uncertainty,
of which the net effect on our estimates is ambiguous. We attempt to catalogue the major
sources of uncertainty in Exhibit 7-1.

INDUSTRIAL ECONOMICS, INCORPORATED

7-1


-------
lEc

EXHIBIT 7-1. KEY UNCERTAINTIES WITH BURDEN ANALYSES

POTENTIAL LIMITATION / SOURCE
OF ERROR

DIRECTION OF POTENTIAL
BIAS FOR ESTIMATED BURDEN

Health impacts associated with
other pollutants are not
quantified.

Underestimate. Epidemiological evidence supports a causal
relationship between ozone exposure and mortality and
morbidity effects. Quantifying and valuing these outcomes
would increase the overall burden of air pollution in the
region.

Morbidity effects are not
quantified.

Underestimate. Epidemiological evidence supports a causal
relationship between PM2.5 exposure and numerous non-
fatal respiratory and cardiovascular effects. Quantifying
and valuing these outcomes would increase the overall
burden.

Mortality burden for near-road
populations.

Underestimate. Epidemiological evidence supports
additional health effects to populations living very close to
large roadways. Quantifying and valuing these outcomes
would increase the overall burden of air pollution in the
region.

Prevalence of non-compliant
vehicles.

Unable to determine based on current information. Non-
compliance rates are calculated based on a limited sample
(n = 2,625) but may be lower or higher in the entire
vehicle fleet.

Emissions rates for non-
compliant vehicles.

Unable to determine based on current information. Non-
compliance is determined based on tests that do not
measure for NOx or PM2.5. Empirical estimates of the
effects on these pollutants are not available. We provide
two potential assumptions on the emissions from non-
compliant vehicles; however, the true emissions may fall
above or below these bounds.

Air quality modeling approach.

Unable to determine based on current information.
Alternative air quality models may better characterize the
magnitude and spatial distribution of PM2.5 concentrations
stemming from one ton of precursor emissions.

Concentration-response
relationship between PM2.5 and
mortality.

Unable to determine based on current information. The
GEMM function compiles results from many epidemiological
studies, some of which find weaker or stronger PM2.5-
induced effects. We also do not have an estimate of long-
term mortality impacts based on a locally conducted
study; the only local study of mortality impacts only
assessed those associated with short-term exposures,
which will underestimate longer-term mortality impacts.

No cessation lag used for
premature mortality.

Overestimate. If there is a time lag between PM2.5 changes
and premature mortality, then benefits occurring in the
future should be discounted.

Valuation of mortality benefits.

Unable to determine based on current information. No
primary studies have been conducted in Peru to estimate
WTP for mortality risk reductions. MINAM's Euro 6 analysis
includes VSL estimates smaller and larger than the
estimate employed in this analysis.

INDUSTRIAL ECONOMICS, INCORPORATED

7-2


-------
lEc

Addressing specific uncertainties, where possible, is an important next step MINAM staff
and academic researchers. While the uncertainties presented in Exhibit 7-1 may be
deemed acceptable for the types of analyses summarized in this report, addressing one or
more of these limitations may be warranted for regulatory benefits analyses. We present
our recommended topics for future research in the following section.

7.3 NEXT STEPS

As discussed above, our results are accompanied by data and methodological limitations,
some of which may be addressed in future analyses by MINAM and academic
researchers. We recommend several areas of focus for building upon the methods
presented in this report:

•	Pursue more advanced air quality modeling, such as leveraging photochemical air
quality models previously employed in Lima-Callao (see Sanchez-Ccoyllo et al.
2016);

•	Quantify and value morbidity effects (e.g., onset asthma and exacerbations,
respiratory and cardiovascular hospitalizations and emergency room visits);

•	Consider the impacts of transportation-attributable ozone formation and exposure;

•	More closely evaluate near roadway exposures (i.e., health impacts resulting from
exposure to elevated PM2 5 and other pollutant concentrations near busy, polluted
roads and highways).

In addition to addressing the methodological limitations outlined above, this report
presents a framework that may be expanded to answer related research questions. First,
the burden analyses conducted thus far may be adapted to assess the benefits of specific
regulatory measures, such as increased enforcement for vehicle emissions standards and
enhanced vehicle I&M programs. Second, the burden analyses may be expanded to assess
the PM2 5-attributable mortality burden stemming from other sources in Lima-Callao. For
example, the results could be stratified further to highlight the mortality burden
associated with specific sources within the transportation sector, such as buses or trucks.
Additionally, this analysis could be expanded to assess the mortality burden attributed to
PM2 5 generated through energy production or chemicals manufacturing, which are two
prominent industries within Lima-Callao (MINAM, 2018). Such results could also be
expressed at the vehicle level. Understanding the average mortality burden resulting from
one non-compliant vehicle, for example, may serve as a useful guide for policymakers in
(1) identifying vehicle types for targeted emissions controls and (2) assessing whether
emissions control costs would be justified based on the societal costs associated with each
vehicle. In addition, evaluations could be made for implementing PM2.5 mitigation
strategies to reduce concentrations at the community-level, such as street cleaning, low
emission zones, and roadside barriers as part of expanding this work to assess near-road
exposures. Additionally, this report presents a framework that could be expanded to
include other sectors. Pending emissions data availability and compatibility with air

INDUSTRIAL ECONOMICS, INCORPORATED

7-3


-------
lEc

quality models, this framework could be applied to other sources, such as industrial point
sources.

Finally, we hope this report may serve as a conduit for continued technical capacity
building for health benefits analysis. Similar analyses conducted under other Megacities
efforts have been accompanied by workshops focused on the BenMAP-CE tool, best
practices for conducting air pollution benefits analyses, and policy synthesis for
advancing local air quality management efforts. We understand that MINAM is currently
conducting a parallel analysis using the AirQ+ software. Comparing results and methods
would serve to bolster the numbers presented in this report and to improve MINAM
capabilities with each tool. Further, engaging additional MINAM staff and relevant
stakeholders (e.g., academics, industry experts, municipalities) may serve to (1)
disseminate the results more broadly and improve the usefulness of the report, and (2)
enhance this study by incorporating alternative data sources and methods recommended
by stakeholders.

INDUSTRIAL ECONOMICS, INCORPORATED

7-4


-------
lEc

REFERENCES

Burnett, R., Chen, H., Szyszkowicz, M., Fann, N., Hubbell, B., et al. 2018. Global

estimates of mortality associated with long-term exposure to outdoor fine particulate
matter. Proceedings of the National Academy of Sciences. Vol. 115 (38): 9592-
9597.

Hansel, N.N., Romero, K.M., Pollard, S.L., Bose, S., Psoter, K.J., J. Underhill, L.,
Johnson, C., Williams, D.A., Curriero, F.C., Breysse, P. and Koehler, K. 2019.
Ambient air pollution and variation in multiple domains of asthma morbidity among
Peruvian children. Annals of the American Thoracic Society, 76(3), pp.348-355.

Health Effects Institute. 2010. Traffic-Related Air Pollution: A Critical Review of the
Literature on Emissions, Exposure, and Health Effects.
https://www.healtheffects.org/svstem/files/SR17TrafficReview.pdf

Karner, A., Eisinger, D., and Niemeier, D. 2010. Near-Roadway Air Quality:

Synthesizing the Findings from Real-World Data. Environmental Science and
Technolology, 44(14): 5334-44.

MINAM. 2018. Avance: Plan de Accion para el Mejoramiento de la Calidad del Aire de
Lima-Callao; Version Preliminar: Diagnostico de la Gestion de la Calidad
Ambiental del Aire de Lima v Callao.

Robinson, L.A., Hammitt, J.K., and O'Keefe, L. 2018. Valuing Mortality Risk Reductions
in Global Benefit-Cost Analysis. Guidelines for Benefit-Cost Analysis Project
Working Paper No. 7.

Sanchez-Ccoyllo, O.R., C.G. Ordonez-Aquino, AG. Munoz, A. Llacza, M.F. Andrade,
Y. Liu, W. Reategui-Romero, and G. Brasseur. 2018. Modeling Study of the
Particulate Matter in Lima with the WRF-Chem Model: Case Study of April 2016.
International Journal of Applied Engineering Research, 13(11): 10129-10141.

Shaddick, G., Thomas, M. L., Green, A., Brauer, M., van Donkelaar, A., Burnett, R.,
Chang, H.H., Cohen, A., Van Dingenen, R., Dora, C. and Gumy, S. 2018. Data
integration model for air quality: a hierarchical approach to the global estimation of
exposures to ambient air pollution. Journal of the Royal Statistical Society: Series C
(AppliedStatistics), 67(1), 231-253.

Tapia, V., Steenland, K., Sarnat, S. E., Vu, B., Liu, Y., Sanchez-Ccoyllo, O., Vasquez,
V., and Gonzales, G. F. 2019. Time-series analysis of ambient PM 2.5 and
cardiorespiratory emergency room visits in Lima, Peru during 2010-2016. Journal
of Exposure Science & Environmental Epidemiology, 1-9.

Tapia, V., Steenland, K., Vu, B., Liu, Y., Vasquez, V., and Gonzales, G.F. 2020. PM2 5
exposure on daily cardio-respiratory mortality in Lima, Peru, from 2010 to 2016.
Environmental Health, 19(63). https://doi.org/10.! 186/sl2940-020-00618-6

INDUSTRIAL ECONOMICS, INCORPORATED

R-1


-------
lEc

Thiel, M., E.C. Macaya, E. Acuna, W.E. Arntz, H. Bastias, K. Brokordt, et al. 2007. The
Humboldt Current System of Northern and Central Chile. Oceanography and Marine
Biology 15: 195-344.

https://www.researchgate.net/publication/210779714 The Humboldt Current Svste
m of Northern and Central Chile

Van Donkelaar, A., Martin, R. V., Brauer, M., Hsu, N. C., Kahn, R. A., Levy, R. C.,
Lyapustin, A., Sayer, A.M. and Winker, D. M. 2016. Global estimates of fine
particulate matter using a combined geophysical-statistical method with information
from satellites, models, and monitors. Environmental science & technology, 50(7),
3762-3772.

INDUSTRIAL ECONOMICS, INCORPORATED

R-2


-------
lEc

APPENDIX A | SUPPLEMENTAL EMISSIONS ESTIMATION RESULTS

In this section, we provide greater detail on estimated pollutant emissions from the
transportation sector. In Chapter 3, we summarize emissions estimates by pollutant and
compliance assumptions. Below, we further stratify these estimates to show the relative
contributions by vehicle fuel type.

For PM2.5 and NOx, diesel emissions comprise the majority of total annual emissions
(Exhibits A-l, A-2). We determined that high diesel emissions are not caused by a greater
share of diesel vehicles in the fleet. Diesel vehicles make up less than 20 percent of the
fleet, whereas gasoline vehicles made up greater than 70 percent. Instead, diesel vehicles
make up large share (more than 85 percent) of high emitting vehicle types such as pick-
up trucks, omnibuses, trucks, and tow trucks. In contrast, gasoline emissions are the
largest contributor to SO2 emissions, as seen in Exhibit A-3.

Notably, SO2 emissions are assumed to vary with sulfur content in fuels and total fuel
consumption. Emissions controls, and thus compliance status, are assumed to not affect
SO2 emissions in our model.

EXHIBIT A-1. 2018 PM2.5 EMISSIONS BY FUEL TYPE

6,000

5,000

I. 4,000

c
o

3,000

c
o

.t2 2,000

1,000

14

12

5 6 6

Diesel	Gas	GNV	GLP

Full Compliance ¦ Upper Bound ¦ Lower Bound

Notes: GLP = liquified petroleum, GNV = natural gas.

INDUSTRIAL ECONOMICS, INCORPORATED

A-1


-------
lEc

EXHIBIT A-2. 2018 NOx EMISSIONS BY FUEL TYPE

200,000
¥ 150,000 1L

(D	'

>

oo
c

~ 100,000

00

C

O
"oo
—

E 50,000

168,070

153,728

10,813

24,808

13'126 5 453
aii _l_ /

4,008 J 5,082

Diesel	Gas	GNV

Full Compliance ¦ Upper Bound ¦ Lower Bound

Notes: GLP = liquified petroleum, GNV = natural gas.
EXHIBIT A-3. 2018 S02 EMISSIONS BY FUEL TYPE

25,000





Emissions (tons/year)

nj

o yi o

o o o
o o o
o o o





















5,000













0





I I I	1

Gas

Diesel GLP GNV
Fuel Type

Notes: GLP = liquified petroleum, GNV = natural gas.

INDUSTRIAL ECONOMICS, INCORPORATED	A-2


-------
lEc

APPENDIX B | SATELLITE MEASUREMENTS AND PROCESSING

In Chapter 4, we summarize available data sources characterizing air pollution in Lima-
Callao, including both monitor and satellite data. In this Appendix, we provide greater
detail on the satellite datasets and our methods for ""ground-truthing" these datasets to
more closely reflect monitor concentrations.

Van Donkelaar et al. (2016) combine information from satellites, model simulations and
monitors. Satellites provide global measurements of aerosol optical depth (AOD). The
van Donkelaar surfaces combine AOD retrievals from the NASA Moderate Resolution
Imaging Spectroradiometer, Multi-angle Imaging SpectroRadiometer and the Sea-
Viewing Wide Field-of-View Sensor. Next, model simulations from the GEOS-Chem
chemical transport model are used to convert total column AOD to near-surface PM2 5
concentrations. Finally, ground-based monitor measurements from the WHO Global
Ambient Air Quality Database25 are used with a GWRto predict and adjust for residual
PM2 5 bias in each grid cell from the initial satellite derived values.

The Shaddick et al. (2017) surfaces are the result of the Data Integration Model for Air
Quality (DIMAQ) developed by the WHO Data Integration Task Force. This model
integrates monitor measurements from the WHO Global Ambient Air Quality Database,
satellite remote sensing, population estimates, topography, and measures of specific
contributors of air pollution from chemical transport models. The same methods as van
Donkelaar et al. (2016) are used to combine AOD from multiple satellites with GEOS-
Chem chemical transport model simulations to produce estimates of near-surface PM2 5 at
0.1° resolution. DIMAQ goes beyond the methods of van Donkelaar et al. (2016) by
using a Bayesian model to estimate the spatially varying relationship between ground
measurements of PM2 5 and factors from the GEOS-Chem, TM5, and TM5-FASST
chemical models that estimate air quality.

Our methods for locally calibrating the satellite surfaces are broken into four steps: (1)
calculate annual PM2 5 averages at the monitor locations, (2) calculate the ratio between
2019 monitor and 2016 satellite annual PM2 5 averages, (3) spatially interpolate the ratios
to create a calibration surface, and (4) multiply the calibration surface against the satellite
surface to create a locally calibrated air quality surface.

First, we determine the annual average PM2 5 concentration measured by monitors at each
location. A 2016 PM2 5 annual average is available for the ten stations in the SENAMHI

World Health Organization. WHO Global Ambient Air Quality Database (Update 2018); WHO: Geneva, 2018:
https://www.who.int/airpollution /data/cities/en/

INDUSTRIAL ECONOMICS, INCORPORATED

B-1


-------
lEc

ground monitoring network. Station information is available in Exhibit 1. Additionally,
we use an adjusted 2019 PM2.5 annual average for two MINSA operated stations, "CA-
VMP-r and ""CA-VMP-2" as these new stations show a hotspot, or area of higher
concentrations, in Callao which is not captured by the SENAMHI monitor network. To
include as many monitoring locations as possible, we adjust the 2019 PM2.5 annual
average at the Callao stations to 2016 concentrations using a ratio of 2016 to 2019 values.
For numerator of the ratio, we use a distance-weighted average of the 2016 annual
average from the two closest SENAMHI monitors. For the denominator of the ratio, we
use the grid cell value at the Callao monitor locations of a 2019 annual average surface
created using data from the ten SENAMHI monitors.

We then determine the annual average satellite surface PM2 5 concentration at each
location by creating a one kilometer buffer around each station and calculating an area-
weighted average within the buffer zone.

Second, we calculate a calibration factor for each station, which is equal to the monitor
annual average divided by the satellite annual average for each station. A calibration
factor greater than one adjusts satellite data upwards and a factor less than one adjusts
satellite data downwards. Third, we interpolate the calibration factors across the domain
using a Kriging function to create a calibration surface (Exhibit B-l). Finally, we
multiply the calibration surface against the original satellite surface to create a locally
calibrated final surface.

INDUSTRIAL ECONOMICS, INCORPORATED

B-2


-------
lEc

EXHIBIT B-1. SHADDICK AND VAN DONKELAAR CALIBRATION SURFACES

b) 2016 Van Donkelaar Calibration a) 2016 Shaddick Calibration

Calibration Factor

| 0.60 - 0.75
0.75 - 0.89
~ 0.89 - 1.02
^ 1.02-1.13
| 1.13 - 1.39
	j Lima-Callao Districts

INDUSTRIAL ECONOMICS, INCORPORATED

B-3


-------
lEc

EXHIBIT B-2. SHADDICK AND VAN DONKELAAR AIR QUALITY SURFACES PRE- AND POST-
CALIBRATION

a) 2016 Shaddick Model

~ Lima-Callao Districts
PM2.5 Concentration (pg/m3)

10.15-16.99
17.00-24.01
24.02-31.30
H 31.31 -43.01
¦I 43.02 - 59.57

b) 2016 Shaddick Model
truthed

d) 20

Van Donkelaar Model
truthed)

INDUSTRIAL ECONOMICS, INCORPORATED

B-4


-------
lEc

APPENDIX C | HEALTH IMPACT ESTIMATION

In this Appendix we discuss additional health impact results, as well as the implication of
using a non-linear mortality function. This Appendix will include discussion of the
GEMM functions, alternative transport contribution results, alternative non-compliant
vehicle contribution results, and the health endpoints not discussed in Chapter 6.

GEMM NON-LINEARITY

As discussed in Section 5.2.3, we utilized six of the 83 non-linear GEMM functions pre-
loaded into BenMAP-CE. It is important to note that these GEMM functions are non-
linear with a decreasing marginal relationship between PM2 5 concentration and the
mortality hazard ratio (Exhibit C-l). Because we modeled mortality burden analyses by
removing the transportation sector's contribution from the high end of PM2.5
concentrations (i.e.. "rolling back" baseline values), we may understate mortality burden
due to the lower mortality response per unit change in PiVl > 5 at these higher
concentrations.

EXHIBIT C-1. BURNETT ET AL. (2018) FIGURE S6

Non Communicable Diseases
+ Lower Respiratory Infections

Ischemic Heart Disease

PM? s - ng/m3

PMai-nQ/m3

Chronic Obstructive
Pulmonary Disease

Lung Cancer

20 40 I
PM2 5 • ng/m3

40

PM2 5 - ng/m3

INDUSTRIAL ECONOMICS, INCORPORATED

C-1


-------
lEc

s a sensitivity analysis, we calculated the contribution of transport emissions and non-
compliant vehicle emissions to the total burden of PM2.5 in Lima-Callao by calculating the
mortality burden as a percentage of total mortality burden based on the sector's
contributions to ambient PM2.5 concentrations. That is, if the sector was responsible for
50 percent of ambient concentrations, we would apportion 50 percent of the total Lima-
Callao mortality burden to this sector. Exhibits C-2 and C-3 compare the BenMAP-CE
GEMM function results against a direct proportional analysis for the transport and non-
compliant vehicle burden contributions. Overall, the GEMM functions may
underestimate the contribution of the transport sector and non-compliant vehicles to the
total burden of PM2 5.

EXHIBIT C-2. TRANSPORT ATTRIBUTABLE PM25 MORTALITY BURDEN

CAUSE OF MORTALITY

GEMM FUNCTION

DIRECT PROPORTIONAL
CONTRIBUTION

LOWER

UPPER

LOWER

UPPER

GEMM: NCD + LRI

5,150

6,200

6,097

7,057

GEAAM: 5 COD

4,221

4,855

3,764

4,357

Lower respiratory
infection

2,522

2,855

2,044

2,365

Ischemic heart disease

680

814

763

883

Cerebrovascular disease

534

616

487

563

Lung Cancer

315

368

302

350

COPD

171

201

168

195

INDUSTRIAL ECONOMICS, INCORPORATED

C-2


-------
lEc

EXHIBIT C-3. NON-COMPLIANT VEHICLE ATTRIBUTABLE PM2.5 MORTALITY BURDEN

CAUSE OF MORTALITY

GEMM FUNCTION

DIRECT PROPORTIONAL
CONTRIBUTION

LOWER

UPPER

LOWER

UPPER

GEAAM: NCD + LRI

248

991

322

1,281

GEAAM: 5 COD

223

890

199

791

Lower respiratory
infection

135

541

108

429

Ischemic heart disease

33

133

40

160

Cerebrovascular disease

30

116

26

102

Lung Cancer

16

65

16

64

COPD

9

34

9

35

INDUSTRIAL ECONOMICS, INCORPORATED

C-3


-------
lEc

APPENDIX D | DISTRICT-LEVEL RESULTS

In this Appendix, we highlight the variability in mortality burden results across districts.
The tables highlight the importance of the geographic resolution of the selected air
quality model when viewing district level results. We compare our primary results (10km
x 10km grid, Shaddick) with a finer scale surface (1km x 1km grid, van Donkelaar) and
present an average effect across these models. Exhibits D-l through D-5 provide the
district level NCD + LRI mortality results for:

•	Total mortality burden attributed to ambient PM2 5: Exhibit C-l

•	Transport sector mortality burden: Exhibits C-2 (lower bound) and C-3 (upper
bound)

•	Mortality burden of non-compliant emissions: Exhibits C-4 (lower bound) and C-
5 (upper bound)

Overall, while the regional-level differences in mortality burden are negligible across
surfaces, we observe notable differences at the district level. Such results may motivate
air quality controls be focused in regions with significant mortality burden.

INDUSTRIAL ECONOMICS, INCORPORATED

D-1


-------
lEc

EXHIBIT D-1. DISTRICT-LEVEL TOTAL PM2 5 MORTALITY BURDEN

DISTRICT NAME

DISTRICT
POPULATION
(AGES 25-99)

SHADDICK

VAN DONKELAAR

AVERAGE

DEATHS

DEATHS

DEATHS

RATE PER
100,000

Overall

6,815,428

10,556

10,838

10,697

157

San Juan De Lurigancho

695,326

1,054

1,009

1,031

148

Comas

361,764

624

707

665

184

San Martin De Porres

488,004

606

654

630

129

Lima

207,527

538

646

592

285

Callao

304,221

535

503

519

171

Villa Maria Del Triunfo

294,591

504

397

451

153

Ate

394,943

370

512

441

112

Santiago De Surco

272,592

354

376

365

134

Carabayllo

189,698

472

247

360

190

Lurigancho

139,146

484

215

349

251

Villa El Salvador

291,540

256

370

313

107

Chorrillos

224,392

350

263

307

137

La Molina

132,567

382

228

305

230

Los Olivos

254,864

277

320

299

117

San Juan De Miraflores

272,025

211

381

296

109

Rimac

120,902

302

285

294

243

Ventanilla

271,339

315

254

285

105

La Victoria

125,163

225

314

270

216

Puente Piedra

210,135

242

289

266

127

Independencia

149,822

253

272

263

175

San Miguel

108,804

217

196

207

190

El Agustino

125,281

165

241

203

162

Miraflores

74,718

127

229

178

238

San Isidro

49,229

197

145

171

348

San Borja

94,488

130

190

160

170

Jesus Maria

61,365

108

174

141

230

Magdalena Vieja

63,935

116

157

136

213

Santa Anita

149,168

91

177

134

90

Brena

60,003

80

144

112

187

Cieneguilla

30,285

166

23

94

311

Magdalena Del Mar

45,329

84

100

92

203

Surquillo

72,492

61

116

89

122

Lince

42,557

69

108

88

208

Bellavista

59,552

66

97

81

137

Lurin

52,254

101

48

75

143

INDUSTRIAL ECONOMICS, INCORPORATED

D-2


-------
lEc





SHADDICK

VAN DONKELAAR

AVERAGE



DISTRICT





















POPULATION







RATE PER

DISTRICT NAME

(AGES 25-99)

DEATHS

DEATHS

DEATHS

100,000

Pachacamac

76,225

94

51

73

96

San Luis

42,772

48

75

61

143

La Perla

49,511

33

79

56

113

Chaclacayo

31,232

33

78

55

176

Ancon

27,202

69

38

53

196

Barranco

24,956

38

57

48

191

Carmen De La Legua
Reynoso

31,190

48

37

42

135

Santa Rosa

12,157

25

8

17

138

La Punta

3,131

8

8

8

249

Punta Hermosa

5,230

9

2

6

110

Pucusana

10,443

5

4

5

46

Mi Peru*

-

6

2

4

NA

Punta Negra

5,400

5

3

4

68

San Bartolo

4,845

2

3

3

54

Santa Maria Del Mar

1,113

1

1

1

109

'The Mi Peru district in Callao had zero population in our dataset. Because the Shaddick and van Donkelaar surfaces
overlap the population grid (districts), BenMAP apportions incidence results from air quality grid cells into the
overlapping districts (including Mi Peru).

INDUSTRIAL ECONOMICS, INCORPORATED

D-3


-------
lEc

EXHIBIT D-2. DISTRICT-LEVEL PM2 5 MORTALITY BURDEN, TRANSPORTATION SECTOR (LOWER
BOUND)

DISTRICT NAME

DISTRICT
POPULATION
(AGES 25-99)

SHADDICK

VAN DONKELAAR

AVERAGE

DEATHS

DEATHS

DEATHS

RATE PER
100,000

Overall

6,815,428

5,150

5,195

5,173 | 76

San Juan De Lurigancho

695,326

509

476

492 | 71

Comas

361,764

299

333

316 | 87

San Martin De Porres

488,004

296

314

305 | 63

Lima

207,527

265

318

292 | 141

Callao

304,221

287

248

267 | 88

Villa Maria Del Triunfo

294,591

242

188

215 | 73

Ate

394,943

178

242

210 | 53

Santiago De Surco

272,592

171

177

174 | 64

Carabayllo

189,698

227

117

172 | 91

Lurigancho

139,146

233

102

168 | 120

Villa El Salvador

291,540

123

173

148 | 51

Chorrillos

224,392

170

123

147 | 65

La Molina

132,567

183

107

145 | 109

Los Olivos

254,864

134

152

143 | 56

Rimac

120,902

149

135

142 | 117

San Juan De Miraflores

272,025

101

179

140 | 51

Ventanilla

271,339

153

120

136 | 50

La Victoria

125,163

111

153

132 | 105

Puente Piedra

210,135

116

136

126 | 60

Independencia

149,822

122

128

125 | 83

San Miguel

108,804

108

100

104 96

El Agustino

125,281

80

113

97 | 77

Miraflores

74,718

62

110

86 | 115

San Isidro

49,229

97

73

85 | 172

San Borja

94,488

63

91

77 | 82

Jesus Maria

61,365

53

88

71 | 115

Magdalena Vieja

63,935

57

80

68 | 107

Santa Anita

149,168

44

83

63 | 43

Brena

60,003

39

72

56 | 93

Magdalena Del Mar

45,329

41

51

46 | 102

Cieneguilla

30,285

80

11

46 | 151

Lince

42,557

34

54

44 | 104

Surquillo

72,492

30

56

43 | 59

Bellavista

59,552

35

49

42 | 70

INDUSTRIAL ECONOMICS, INCORPORATED

D-4


-------
lEc

DISTRICT NAME

DISTRICT
POPULATION
(AGES 25-99)

SHADDICK

VAN DONKELAAR

AVERAGE

DEATHS

DEATHS

DEATHS

RATE PER
100,000

Lurin

52,254

49

24

37

70

Pachacamac

76,225

46

26

36

47

San Luis

42,772

23

36

29

69

La Perla

49,511

18

40

29

59

Chaclacayo

31,232

16

37

26

84

Ancon

27,202

34

18

26

95

Barranco

24,956

19

27

23

91

Carmen De La Legua
Reynoso

31,190

24

18

21

67

Santa Rosa

12,157

12

4

8

66

La Punta

3,131

4

4

4

134

Punta Hermosa

5,230

5

1

3

56

Pucusana

10,443

3

2

3

24

Mi Peru

-

3

1

2

NA

Punta Negra

5,400

2

1

2

36

San Bartolo

4,845

1

2

1

29

Santa Maria Del Mar

1,113

1

1

1

57

'The Mi Peru district in Callao had zero population in our dataset. Because the Shaddick and van Donkelaar surfaces
overlap the population grid (districts), BenMAP apportions incidence results from air quality grid cells into the
overlapping districts (including Mi Peru).

INDUSTRIAL ECONOMICS, INCORPORATED

D-5


-------
lEc

EXHIBIT D-3. DISTRICT-LEVEL PM2 5 MORTALITY BURDEN, TRANSPORTATION SECTOR (UPPER BOUND)

DISTRICT NAME

DISTRICT
POPULATION
(AGES 25-99)

SHADDICK

VAN DONKELAAR

AVERAGE

DEATHS

DEATHS

DEATHS

RATE PER
100,000

Overall

6,815,428

6,200

6,238

6,219 91

San Juan De Lurigancho

695,326

607

564

585 84

Comas

361,764

356

394

375 104

San Martin De Porres

488,004

358

381

369 76

Lima

207,527

323

387

355 171

Callao

304,221

355

302

329 108

Villa Maria Del Triunfo

294,591

290

224

257 87

Ate

394,943

211

287

249 63

Santiago De Surco

272,592

205

212

208 76

Carabayllo

189,698

270

140

205 108

Lurigancho

139,146

277

122

200 143

Villa El Salvador

291,540

147

206

177 61

Chorrillos

224,392

206

147

176 79

La Molina

132,567

218

127

173 130

Rimac

120,902

181

162

171 | 67

Los Olivos

254,864

160

183

171 | 142

San Juan De Miraflores

272,025

122

213

167 61

Ventanilla

271,339

184

142

163 60

La Victoria

125,163

135

186

160 128

Puente Piedra

210,135

138

161

150 71

Independencia

149,822

146

152

149 100

San Miguel

108,804

132

123

128 117

El Agustino

125,281

96

134

115 92

Miraflores

74,718

75

134

104 140

San Isidro

49,229

118

89

103 210

San Borja

94,488

76

110

93 98

Jesus Maria

61,365

65

108

86 141

Magdalena Vieja

63,935

69

98

84 131

Santa Anita

149,168

52

99

75 50

Brena

60,003

48

88

68 113

Magdalena Del Mar

45,329

50

63

57 125

Cieneguilla

30,285

95

13

54 180

Lince

42,557

41

67

54 127

Surquillo

72,492

36

68

52 72

Bellavista

59,552

43

60

51 | 86

Lurin

52,254

59

29

44 84

INDUSTRIAL ECONOMICS, INCORPORATED

D-6


-------
lEc





SHADDICK

VAN DONKELAAR

AVERAGE



DISTRICT





















POPULATION







RATE PER

DISTRICT NAME

(AGES 25-99)

DEATHS

DEATHS

DEATHS

100,000

Pachacamac

76,225

55

32

43

57

La Perla

49,511

23

49

36

85

San Luis

42,772

28

43

35

72

Chaclacayo

31,232

19

44

31

101

Ancon

27,202

41

22

31

114

Barranco

24,956

23

32

27

109

Carmen De La Legua
Reynoso

31,190

29

22

26

82

Santa Rosa

12,157

14

5

10

79

La Punta

3,131

6

5

5

166

Punta Hermosa

5,230

6

2

4

68

Pucusana

10,443

3

3

3

30

Mi Peru

-

4

1

2

NA

Punta Negra

5,400

3

2

2

44

San Bartolo

4,845

1

2

2

36

Santa Maria Del Mar

1,113

1

1

1

71

'The Mi Peru district in Callao had zero population in our dataset. Because the Shaddick and van Donkelaar surfaces
overlap the population grid (districts), BenMAP apportions incidence results from air quality grid cells into the
overlapping districts (including Mi Peru).

INDUSTRIAL ECONOMICS, INCORPORATED

D-7


-------
lEc

EXHIBIT D-4. DISTRICT-LEVEL PM2 5 MORTALITY BURDEN, NON-COMPLIANT EMISSIONS (LOWER
BOUND)

DISTRICT NAME

DISTRICT
POPULATION
(AGES 25-99)

SHADDICK

VAN DONKELAAR

AVERAGE

DEATHS

DEATHS

DEATHS

RATE PER
100,000

Overall

6,815,428

248

250

249 | 4

San Juan De Lurigancho

695,326

25

24

25 | 4

Comas

361,764

15

17

16 | 4

San Martin De Porres

488,004

14

15

14 | 3

Lima

207,527

12

15

13 | 6

Callao

304,221

13

11

12 | 4

Ate

394,943

9

12

10 | 4

Villa Maria Del Triunfo

294,591

12

9

10 | 3

Carabayllo

189,698

11

6

9 I 3

Santiago De Surco

272,592

8

9

8 | 4

Lurigancho

139,146

12

5

8 | 6

Villa El Salvador

291,540

6

9

7 I 3

La Molina

132,567

9

5

7 I 3

Chorrillos

224,392

8

6

7 I 5

Los Olivos

254,864

7

7

7 I 3

San Juan De Miraflores

272,025

5

9

7 I 6

Rimac

120,902

7

7

7 I 2

Ventanilla

271,339

7

6

7 I 2

Puente Piedra

210,135

6

7

6 | 5

Independencia

149,822

6

6

6 | 3

La Victoria

125,163

5

7

6 | 4

El Agustino

125,281

4

6

5 | 4

San Miguel

108,804

5

4

5 | 4

Miraflores

74,718

3

5

4 | 5

San Isidro

49,229

5

3

4 | 8

San Borja

94,488

3

4

4 | 4

Jesus Maria

61,365

2

4

3 | 5

Santa Anita

149,168

2

4

3 | 5

Magdalena Vieja

63,935

3

4

3 I 2

Brena

60,003

2

3

3 I 4

Cieneguilla

30,285

4

1

2 I 5

Magdalena Del Mar

45,329

2

2

2 I 7

Surquillo

72,492

1

3

2 I 5

Lince

42,557

2

2

2 I 3

Bellavista

59,552

2

2

2 I 3

INDUSTRIAL ECONOMICS, INCORPORATED

D-8


-------
lEc





SHADDICK

VAN DONKELAAR

AVERAGE



DISTRICT





















POPULATION







RATE PER

DISTRICT NAME

(AGES 25-99)

DEATHS

DEATHS

DEATHS

100,000

Lurin

52,254

2

1

2

3

Pachacamac

76,225

2

1

2

2

San Luis

42,772

1

2

1

3

La Perla

49,511

1

2

1

3

Chaclacayo

31,232

1

2

1

4

Ancon

27,202

2

1

1

5

Barranco

24,956

1

1

1

4

Carmen De La Legua
Reynoso

31,190

1

1

1

3

Santa Rosa

12,157

1

0

0

3

La Punta

3,131

0

0

0

6

Punta Hermosa

5,230

0

0

0

3

Pucusana

10,443

0

0

0

1

Mi Peru

-

0

0

0

NA

Punta Negra

5,400

0

0

0

2

San Bartolo

4,845

0

0

0

1

Santa Maria Del Mar

1,113

0

0

0

3

'The Mi Peru district in Callao had zero population in our dataset. Because the Shaddick and van Donkelaar surfaces
overlap the population grid (districts), BenMAP apportions incidence results from air quality grid cells into the
overlapping districts (including Mi Peru).

INDUSTRIAL ECONOMICS, INCORPORATED

D-9


-------
lEc

EXHIBIT D-5. DISTRICT-LEVEL PM2 5 MORTALITY BURDEN, NON-COMPLIANT EMISSIONS (UPPER
BOUND)

DISTRICT NAME

DISTRICT
POPULATION
(AGES 25-99)

SHADDICK

VAN DONKELAAR

AVERAGE

DEATHS

DEATHS

DEATHS

RATE PER
100,000

Overall

6,815,428

991

1,003

997 | 15

San Juan De Lurigancho

695,326

101

95

98 | 14

Comas

361,764

59

67

63 | 17

San Martin De Porres

488,004

56

59

58 | 12

Lima

207,527

49

59

54 | 26

Callao

304,221

51

46

48 | 16

Villa Maria Del Triunfo

294,591

47

37

42 | 14

Ate

394,943

36

48

42 | 11

Carabayllo

189,698

45

23

34 | 12

Santiago De Surco

272,592

33

35

34 | 18

Lurigancho

139,146

46

20

33 | 24

Villa El Salvador

291,540

24

34

29 | 10

La Molina

132,567

36

21

29 | 13

Chorrillos

224,392

32

24

28 | 21

Los Olivos

254,864

26

30

28 | 11

San Juan De Miraflores

272,025

20

36

28 | 23

Rimac

120,902

28

26

27 | 10

Ventanilla

271,339

29

24

27 | 10

Puente Piedra

210,135

23

28

25 | 20

Independencia

149,822

24

25

25 | 12

La Victoria

125,163

21

28

25 | 16

El Agustino

125,281

15

23

19 | 18

San Miguel

108,804

20

18

19 | 15

Miraflores

74,718

12

21

16 | 22

San Isidro

49,229

18

13

16 | 32

San Borja

94,488

12

17

15 | 16

Jesus Maria

61,365

10

16

13 | 21

Santa Anita

149,168

9

17

13 | 20

Magdalena Vieja

63,935

11

14

12 | 8

Brena

60,003

7

13

10 | 17

Cieneguilla

30,285

16

2

9 | 20

Magdalena Del Mar

45,329

8

9

8 | 28

Lince

42,557

6

10

8 | 19

Surquillo

72,492

6

10

8 | 11

Bellavista

59,552

6

9

8 | 13

INDUSTRIAL ECONOMICS, INCORPORATED

D-10


-------
lEc





SHADDICK

VAN DONKELAAR

AVERAGE



DISTRICT





















POPULATION







RATE PER

DISTRICT NAME

(AGES 25-99)

DEATHS

DEATHS

DEATHS

100,000

Lurin

52,254

9

4

7

13

Pachacamac

76,225

9

5

7

9

San Luis

42,772

4

7

6

13

La Perla

49,511

3

7

5

10

Chaclacayo

31,232

3

7

5

16

Ancon

27,202

6

3

5

18

Barranco

24,956

3

5

4

18

Carmen De La Legua
Reynoso

31,190

4

3

4

12

Santa Rosa

12,157

2

1

2

13

La Punta

3,131

1

1

1

24

Punta Hermosa

5,230

1

0

1

10

Pucusana

10,443

0

0

0

4

Mi Peru

-

1

0

0

NA

Punta Negra

5,400

0

0

0

6

San Bartolo

4,845

0

0

0

5

Santa Maria Del Mar

1,113

0

0

0

10

'The Mi Peru district in Callao had zero population in our dataset. Because the Shaddick and van Donkelaar
surfaces overlap the population grid (districts), BenMAP apportions incidence results from air quality grid cells into
the overlapping districts (including Mi Peru).

INDUSTRIAL ECONOMICS, INCORPORATED

D-11


-------