Energy Policy 84 (2015) 1121

Contents lists available at ScienceDirect

Energy Policy
journal homepage: www.elsevier.com/locate/enpol

Impacts of climate change on power sector NOx emissions: A long-run

analysis of the US mid-atlantic region
Yihsu Chen a,n, Benjamin F. Hobbs b, J. Hugh Ellis b, Christian Crowley c, Frederick Joutz d
a

Sierra Nevada Research Institute, School of Engineering and School of Social Sciences, Humanities and Arts, University of California, Merced, CA, USA
Department of Geography and Environmental Engineering, Johns Hopkins University, Baltimore, MD, USA
c
Ofce of Policy Analysis, US Department of Interior, Washington, DC, USA
d
Department of Economics, George Washington University, Washington, DC, USA
b

H I G H L I G H T S

We develop a framework to study the impact of climate-induced changes on electricity sector.

It could affect spatial and temporal distribution of pollution emissions in the long run.
Under a worse-case assumption, signicant emissions during peak demand hours could occur.
It could possibly worsen regional air quality, even if seasonal emissions are constant under cap.
A separate cap or tax can be applied to extreme weather conditions to avoidworsening air quality.

art ic l e i nf o

a b s t r a c t

Article history:
Received 6 October 2014
Received in revised form
7 April 2015
Accepted 12 April 2015

We propose a framework for analyzing the long-run effects of climate change on the spatial and temporal
distribution of nitrogen oxide (NOx) emissions from the power sector. Elevated ground-level temperatures could increase electricity demand during the ozone season, altering the generation mixes and
ultimately changing emissions. A sequence of load forecasting, supply investment and operation, and
facility siting models is used to project spatial and temporal distributions of NOx emissions. Under a
worse-case scenario with no renewable additions or other interventions, the results indicate that even if
total NOx is limited by cap-and-trade policies, climate-warming-induced changes in the timing of pollution emissions can be signicant, especially under warmer or high-load conditions. This suggests that a
continued reliance on fossil-fuel together with a temperature sensitivity of generation efciency and
peak electricity demands increases the likelihood that emissions will be greater during the warm days
when ozone episodes also occur. The paper advances the integrated assessment by identifying ways at
which climate-change-derived energy demand can impact generation mixture, operations and local air
pollution. The downscaled emissions can be used in regional air quality models such as the Community
Multiscale Air Quality (CMAQ) to project changes in tropospheric ozone due to climate change.
Published by Elsevier Ltd.

Keywords:
Climate change
Air pollution
Electric sector
NOx
Ground-level ozone
Mid-atlantic region

1. Introduction
Climate change-induced increases in surface temperatures
could inuence human health in several ways, for instance by
increased heat-related deaths during heat waves (Donaldson et al.,
2001; Kosatsky, 2005), or by the spread of certain diseases (e.g.,
malaria and dengue fever) (Colwell, 1996; Tanser et al., 2003).
n

Corresponding author.
E-mail addresses: ychen26@ucmerced.edu (Y. Chen),
bhobbs@jhu.edu (B.F. Hobbs), hugh.ellis@jhu.edu (J. Hugh Ellis),
christian_crowley@ios.doi.gov (C. Crowley), bmark@gwu.edu (F. Joutz).
http://dx.doi.org/10.1016/j.enpol.2015.04.013
0301-4215/Published by Elsevier Ltd.

Warmer air temperature could also enhance the formation of

tropospheric ozone (O3) (due to increased temperature-dependent
reaction rates) and other pollutants by changing the amounts and
the distributions of anthropogenic and biogenic emissions, as well
as mixing heights and winds which affect pollutant transport
(Sillman and Samson, 1995; Constable et al., 1999).
Formation of ozone involves oxidation of organic compounds
by NOx in the presence of sunlight. It is a highly nonlinear process
that in part depends on the ratio of NOx and VOC (volatile organic
compounds) concentrations, which can vary greatly over time and
space (Seinfeld and Pandis, 1997). Tropospheric ozone is subject to
the National Ambient Air Quality Standards (NAAQS) under the US

12

Y. Chen et al. / Energy Policy 84 (2015) 1121

Clean Air Act (Bell et al., 2006), and epidemiological studies have
established increases in mortality from short-term ozone exposure
(Bell et al., 2004) A number of air pollution modeling studies also
indicate that climate change is likely to affect ground-level ozone
formation (Jacob and Winner, 2009; USEPA, 2009; Weaver et al.,
2009; Woo et al., 2009). In the US, the primary sources of VOCs are
biogenic sources with important anthropogenic contributions
(USEPA, 2006), while NOx is mostly generated by combustion
processes from mobile and stationary sources.
Several integrated assessments have considered how air pollution might be affected by climate change (Anderson et al., 2001;
Donaldson et al., 2001; Racherla and Adams, 2006; Bell et al.,
2007; Civerolo et al., 2008). These studies are usually based on
IPCC (Intergovernmental Panel on Climate Change) or other commonly used emissions scenarios, e.g., IPCC Special Report Emissions Scenario A2 (Nakicenovic and Swart, 2000). One advantage
of using such scenarios is that the results can be compared and
possibly generalized across different studies. However, these scenarios only provide information on annual emissions, and by denition they will show zero change when emissions are capped on
an annual basis, as they are under US law for utility sources of SO2
and NOx in most of the country. Thus, with that approach, they
also cannot address shifts in locations and timing of emissions
from particular economic sectors (e.g., increases in summer electricity demand in response to warming climate), which can be
critical to ground-level ozone formation. As a result, these annual
scenarios lack the spatial and temporal granularity necessary for
use in fate and transport models, and interactions of climate
change with particular pollution control policies, such as NOx caps,
cannot be analyzed. The purpose of this paper is to develop and
demonstrate an integrated framework that allows examination of
the effects of climate change on the spatial and temporal distribution of NOx (and other pollutants in general) from the power
sector.
Climate change could affect power systems in several ways. It
will alter the level and timing of electricity demands, as well as the
efciency of electricity generating units (EGUs) (e.g., heat rate and
available generating capacity) (Amato et al., 2005). In the short
run, with a given capital stock of EGUs, the result will be changes
in their operations and emissions. In the long run, the mix of
various plant types will adjust in response to fuel and emissions
allowances prices as well as climate-induced changes in the intraannual distribution of electricity demands. Thus, in order to understand the effects of climate change on ground-level ozone
formation, impacts upon spatial and temporal distributions of EGU
NOx emissions must be considered.
This paper is motivated by two research questions concerning
power sector NOx emissions and other air pollutant in general.
First, how might long run (mid-21st century) spatial and temporal
distributions of power plant NOx emissions in the US mid-Atlantic
region shift as a result of climate change? Second, how might inter-year variability of climate impact electricity usage and NOx
emissions, which in turn could potentially impact the frequency of
summertime ozone episodes? This is in contrast to most energy
models used in other climate change impact analyses that assume
a typical or average year (Amato et al., 2005; Hadley et al.,
2006). Due to nonlinear relationships among temperatures,
emissions, and their impacts, the average impact on air quality
over a number of years may be quite different (and possibly
higher) than the impact on air quality in an average year.
The framework we developed relies upon a sequence of power
sector load forecasting and supply models to address these questions. These models predict locations of new generation capacity,
and temporal and spatial distributions of air emissions from power
sector. Generating technologies considered in the analysis include
scrubbed coal (steam), integrated coal-gasication combined cycle

(IGCC), combined cycle units, combustion turbines, and nuclear

power plants. Coal-red plants are assumed to install various
types of pollution control equipment, namely ue gas desulfurization, selective catalytic reduction, and electrostatic precipitators. Renewable and other technologies (e.g., fuel cell and
clean coal technology with carbon capture and sequestration) are
not included in the analysis for two reasons. First, we are interested in the worst-case scenario in which fossil-fueled units remain the dominant technology. Second, renewable siting is less
predictable, as it depends on the availability of resources as well as
local and state policies that are designed to promote their deployment. Of course, there are various scenarios concerning policy
and technology changes could unfold over the next two decades
that are not considered by this analysis. Possible policy, technological and economic changes could certainly interact with temperature change in a way that alters the assumptions. However,
what we demonstrate in the paper is a method that could be used
to explore the effects of climate change on the spatial and temporal distribution of emissions from power sector under alternative assumptions.
To model inter-year variability, we use 14 years of independently simulated ground-level temperatures from the GISS
(Goddard Institute of Space Sciences) GCM, where the years 1991
1998 (1990s hereafter) represent normal climate conditions, and
years 20502055 (2050s) represent a warmer climate (CCSR,
2006). Under the worse-case scenario, e.g., no renewable addition
or without other regulatory intervention to suppress outputs from
high emitting eets, the results suggest that even if total seasonal
NOx emissions are unchanged due to the presence of cap-andtrade policies, changes in their spatial and temporal distribution
imply that the severity of ozone episodes could become worse.
Our analysis focuses on the US mid-Atlantic region1 for three
reasons.2 First is its nonattainment status for ambient ozone
standards. Second, a synthesis of results from studies conducted
by more than twenty institutions concluded that the US Northeast
and the Mid-Atlantic were the regions with the highest likelihood
of adverse ozone air quality impacts from climate change (Weaver
et al., 2009). Third, the mid-Atlantic regional power system has
been deregulated over the last two decade, with the objective of
decreasing costs and making generation decisions (and thus pollution emissions) more responsive to market conditions.
Fig. 1 displays the study region, consisting of 15 demand and
generation subregions (corresponding to individual utilities) and
18 transmission corridors. Our analysis includes the high voltage
(i.e., 500 kV) network, taking into account the effect of regionallevel transmission constraints upon the spatial distribution of
generation and emissions. We split several utilities so that transmission congestion within each can be represented, including ME
(Metropolitan Edison), JC (Jersey Power and Light), PPL (PPL
Electric Utilities) and BGE (Baltimore Gas and Electric). Meanwhile,
part of BGE together with PEPCO (Potomac Electric Power) is represented the BGEPEP. Since the former East Central Area Reliability Council (ECAR) (ECAR, 2006) is located upwind of the midAtlantic region and its EGUs emit signicant amounts of NOx, we
also include its eastern portion in the analysis. Because transmission constraints are less of a problem within ECAR (PJM, 2003),
1
Conventionally, the region often includes New York, New Jersey, Pennsylvania, Delaware, Maryland, Washington D.C., Virginia, and West Virginia.
2
In European context, ground-level ozone among other pollutants have also
been a concern (Orru et al., 2013). Policies such as National Emission Ceilings (NEC)
under the Directive of the European Commission (Commission, 2014a, 2014b) and
the Gothenburg Protocol (Europe, 2014) as well as TSAP (Thematic Strategy on Air
Pollution) scheme (Commission, 2014a, 2014b) have been implemented to reduce
pollutants responsible for acidication, eutrophication and ground-level ozone
pollution.

Y. Chen et al. / Energy Policy 84 (2015) 1121

13

Fig. 1. Transmission network of study region(PJM ; PowerWorld, 2005).

this analysis represents that region as a single subregion. Congestion between the mid-Atlantic and ECAR regions will be captured by limited capacity in the two main corridors connecting
them.

2. Material and methods

Linear programs (LPs) are a common tool for simulating operations and capacity expansion decisions for the power sector

(Hobbs, 1995). Examples include IPM (USEPA, 2005c) and MARKAL

(ETSAP, 2006). LP solutions are equivalent to a competitive market
equilibrium subject to price-insensitive demand. An advantage of
using LPs is the existence of efcient solution algorithms. LPs are
the primary modeling tool in this analysis. Several are used in
succession to downscale regional projections of electricity demand
into temporal and spatial distributions of NOx emissions for the
1990s and 2050s climate scenarios.
Fig. 2 is a ow chart summarizing the six steps of the analysis.
Each step uses one or more models, summarized below. Details on

Step 1: Construct average electricity demand distribution curve for year 2025 under
1990s and 2050s climate scenarios using NEMS demand module and coolingdegree day scenarios, and short run load forecasting model
Regional Load distribution as function of average climate
Step 2: Estimate the amount of new generating technologies at the regional level using
transmissionunconstrained least-cost capacity model and average load
patterns under each climate scenario
Generation capacity by type & region
Step 3: Allocate new capacity to utility (subregional) level using transmissionconstrained least-cost capacity expansion model
Generation capacity by zone
Step 4: Assess county-level probability of siting power plants using empirical logit
models
Generation siting probabilities by county
Step 5: Site new capacity using mixed integer optimization, consistent with empirical
model
Generation capacity by zone
Step 6: Generate hourly NOx emissions using short run least-cost dispatch models and
year-by-year electricity demands & meteorology
Emissions by hour and plant for particular year:
1990s climate scenario and 2050s climate scenario
Fig. 2. Flow chart of the analysis procedure.

14

Y. Chen et al. / Energy Policy 84 (2015) 1121

the model formulations can be found elsewhere (Chen, 2006).

First Step. The rst step is the construction of average annual
load duration curves3 (LDCs) for 1990s (19911998) and 2050s
(20502055)4 climates, adjusting for general changes in load
shape due to economic activities and climate conditions using the
electricity demand modules of the 2025 National Energy Modeling
System (NEMS5) (EIA, 2014). Two substeps are involved in Step 1.
The rst substep captures hour-to-hour variability using meteorological output from the MM5 (derived from GISS GCM) to
drive a set of short run electric load forecasting models (Crowley,
2006) (one for each major utility in the region) in order to produce
14-years of hourly electricity demand uctuations. The outcome is
the hourly demand for each of the 14 years. The short run electric
load forecasting models (Crowley, 2006) consist of dynamic statistical relationships that account for time of day, recent hourly
temperatures, and weekend/weekday/holiday effects on relative
load. Consequently, these models enable us to represent hour-tohour load variations and their temporal relationship with the
meteorological conditions that are crucial to ozone formation.
Each year's hourly load data are then grouped into 27 time blocks
Dty [MW] consistent with NEMS groupings.6 However, the short
run models only represent short-term load variations in response
to weather, and not long-term adjustments in energy-using capital
stock.
The second substep of Step 1 addresses such stock adjustments.
NEMS electricity demand modules for those sectors represent
long-term responses to temperatures, accounting for changes in
the efciency of HVAC equipment in response to price. Therefore,
we have rescaled loads in each block in substep 1 so that their
averages are consistent with NEMS year 2025 simulations of residential and commercial loads by region and block. The output
from NEMS's 2025 runs simulates two specic conditions: a 150 Fday and a 0 F-day increase in CDDs (cooling degree days/yr), representing by Lt(0) and Lt(150), respectively, where t indicates the
t-th block. While the NEMS simulation representing the 1990s
climate is based on NEMS's assumption of approximately 150 Fday CDDs for the major cities in the US Mid-Atlantic region, our
analysis on the GISS's 2050s output suggests that their 2050s CDD
values correspond to 414 F-day higher than 1990s. Therefore, we
linearly extrapolate electric load under a X F temperature increase
based upon Lt(0) and Lt(150) by the following equation:

Lt (X) = (1 X /150) Lt (0) + (X /150) Lt (150)

(1)

t, y = Lt (414)/Dty foryin2050s[unitless]

The parameters in (2) and (3) for the 1990s and 2050s climates are then applied to scale each of the individual hours demand within the block to generate the hourly MW loads Loadhy for
each hour h and climate y that account for end-use technology
change as a result of temperature. The resulting Loadhy is further
grouped into 27 blocks Lty for analysis in Step 2, below, consistent
with NEMS's denition of electricity demand blocks.
Admittedly, there is an inconsistency in our use of 2025 electricity loads under a 2050 climate, but 2025 loads are the last simulated in the NEMS runs that were available to us. An advantage
of using the 2025 NEMS results is that they allow us to base
generation technology and cost assumptions upon the NEMS database, permitting construction of a consistent set of load, emission and generation assumptions for a scenario year well into the
future. Scenarios based on assumptions other than NEMS could be
applied and might yield different LDCs. However, NEMS is recalibrated by the US Energy Information Agency annually to incorporate new technological and economic information. NEMS is
by far the most comprehensive energy model for US energy sector,
and is commonly used to examine the impact of environmental
and energy policies (DOE, 2000) (Morris et al., 2004) Thus, NEMS
assumptions reect the best available information for building
future US energy scenarios.
Second Step. The second step in Fig. 2 estimates the 2025
overall generation capacity mix for the two climate scenarios using
a LP model (4)(11). (For simplicity, we ignore subscript/superscript associated with scenarios.)

LCj Uj

min

xht , z jt , Uj

t , y = Lt (0)/ Dty for y in 1990s[unitless]

(2)

Bt NCj z jt

j, t

Bt Ch xht
h, t

(4)

Subject to: xht Xh , h, t

(5)

z jt (1 F jN ) Uj , j, t

(6)

Bt xht

8760Xh CFh , h

8760(1 F jN ) Uj CF jN , j

Bt zjt
t

Bt xht

Bt zjt

Xh

Uj

(8)

(9)

Lpeak R
(10)

Bt Eh xht +

h H OTC, t

(7)

Bt Lt , t

3
An annual LDC ranks hourly electricity demand (the y-axis) in descending
order against cumulative hours (1-8760 hours on the x-axis).
4
Ideally, we would like to have an identical sample size for each decade, but
2055 is the last year that was available at the time we did the analysis.
5
NEMS is an energyeconomy model developed and maintained by the US
Energy Information Administration (EIA) to provide predictions of US energy
conditions 25 years into future. With consideration of macroeconomics (e.g., GDP),
nancial and global energy markets, NEMS is used by the EIA to study the impact of
energy, economic, environmental, and supply security of alternative energy policies
and assumptions concerning energy markets and economic growth. EIA publishes
the outcomes of each years projections in its Annual Energy Outlook (AEO). More
information can be found in NEMS website (EIA, 2014). The NEMS runs we used
assumed a GDP (gross domestic product) growth rate of roughly 2.5% per year
between 2006 and 2025 (DOE, 2009).
6
In particular, there are (a) three load groups, depending on the time of a given
day: midday (9:0016:00), morning/evening (6:008:00 and 17:0024:00) and
night (1:005:00); (b) three seasons: summer (JuneSept), Winter (DecMarch)
and Spring/Fall (AprilMay and OctNov); and (c) three levels (or segments) of load
in each load group: highest 1%, next highest 33% and the lowest 66%.

The respective changes for each hour h belonging to block t

(h) can then be expressed as follows:

(3)

j J OTC , t

Bt E jN z jt NOx
(11)

xht , z jt , Uj 0
The objective function (4) is to minimize the total annualized
cost for 2025, which consists of three terms: annualized construction costs (LCj, $/MW/yr) for new generators, their xed and
variable operations and maintenance (O&M) costs (NCj, $/MWh)
(EIA, 2012a), and O&M costs (Ch, $/MWh) for existing generators.
Coefcient Bt is number of hours in each block t. Indices h and j
denote plant types for existing and new generation, respectively. Lt
is the MW load in block t, obtained by summing over the average
load of those years within each scenario in each subregion in the
network. The LP's decision variables include output levels (xjt and
zjt, in MW) and new capacity (Ujt, in MW) by types. Candidate
plant types j include steam (pulverized coal), advanced nuclear for

Y. Chen et al. / Energy Policy 84 (2015) 1121

baseload, integrated gasication combined cycle, conventional

combined-cycle, advanced combined-cycle (cycling), conventional
combustion turbines, advanced combustion turbines (peaking),
respectively (for more detailed description on technologies, please
refer to NEMS Electricity Market Module (EIA, 2006), Table 38, p.
73).7 The constraints include a NOx emissions cap during the
ozone season (May 1September 30) (11); energy balances (supplydemand in each time block, which differs between the 1990s
and 2050s scenarios) (9); capacity limits on generation by plant
type (accounting for both existing and new capacity) (5)(6); capacity reserve margin requirements (11); and capacity factor
constraints that limit the hours each plant can operate (7)(8).
Generation data assumptions include the following. The primary data source for existing generators is the NEMS data base
(EIA, 2010) collected through EIA Form 860 (EIA, 2012b). It contains plants in place in 2000, comprising 1453 EGUs, of which 731
were located within PJM, and the remaining were in ECAR. That
capacity is derated using forced and maintenance outage rates
(NERC, 2006) to account for differences in reliability of various
generator types, while older plants were retired. The total derated
existing capacity was 90,564 MW. An ination-free interest rate of
13% is used to annualize capital costs for new plants, consistent
with the EPA's IPM model (USEPA, 2005c). (All costs in this study
are expressed as 2004 dollars.) Each plant's variable cost equals
the sum of fuel cost, SO2 permit cost, and non-fuel variable O&M
expenses. All fuel costs are exogenous and depend on plant location and type. The heat rate and capacity of plants are adjusted in
the warming (2050s) scenario to reect a degradation of thermal
efciency. In particular, heat rates are adjusted based on singlecycle Carnot efciency,8 while capacities are derated in the warmer (2050s) case based on our analysis of experienced ratios of
summer to winter generating capacity and ambient temperature
between these two seasons in the mid-Atlantic region.
Turning to assumed pollution rules, as the US Clean Air Act's
SO2 trading program is national in scope, we treat SO2 allowance
prices as an exogenous component of O&M costs in (4), assuming
750 $/Ton. In contrast to SO2, under US law, NOx trading is more
regional in nature, and so the model explicitly caps NOx emissions
in the region. Under those rules, there are two NOx caps: seasonal
and annual. We assume that only the ozone season cap is binding.
However, since some states are only partially contained within our
study region, the availability of NOx allowances to our market is
adjusted downwards based on the proportion of the state's generation capacity considered (USEPA, 2005a). The number of NOx
allowances assumed available is about 130,000 t. Three sources for
emissions rates include IPM (USEPA, 2005c), the Emission, Generation Resource Integrated Database (eGRID) (USEPA, 2005b), and
USEPA Continuous Emission Monitoring Data (USEPA, 2005d).
7
Of course, additional technologies such as renewables, storage or even demand response programs can be considered in the analysis. However, we limit their
role for at least three reasons. First, we are interested in worst-case emission
scenarios, while these technologies could reduce NOx emissions during peak-demand hours. Second, consideration of the intermittency of solar and wind generation would complicate our already complex analyses. Third, uncertainty remains
concerning the future penetration and performance of these technologies. Given
that our emphasis is upon demonstration of the methodology, we limited the
generation investment and dispatch analysis to the conventional technologies
mentioned in this paragraph. However, the general framework we developed can
readily consider oth.er technologies.
8
In particular, we assume that the difference between the generator operating
temperatures and the heat sink temperatures will shrink by the amount of
warming in the 2050s scenario Based on this assumption, plant heat rates are then
adjusted based on the average differences of temperature between 1990s and
2050s series. In our analysis, for each 1 F increase in the ambient temperature,
generating capacity is adjusted downward on average by 0.4%, and efciency (i.e.,
heat rate) is worsened by 0.6 and 0.7% for combined-cycle and combustion turbine
EGUs, respectively.

15

Third Step. The third step allocates the new capacity estimated
in Step 2 to subregions using a LP-based transmission-constrained
model for simulating capacity investment, considering the effects
of regional fuel cost variations on siting decisions.9 To identify the
location of each generator and assign it to a subregion in the
network, we used information from USEPA eGRID and other
sources. This is similar to the model in Step 2 with an objective to
minimize the total cost (levelized capital cost for new capacity and
production cost) but also explicitly models transmission ows
using a linearized DC in (12). Essentially, the ow in the network
has to follow Kirchoff's Voltage and Current Laws (Schweppe et al.,
1988), i.e., linearized direct-current (DC) ow,10 and satises
thermal limits of transmission lines. The term PTDFBkn is the PTDF
matrix, Tk Bis an upper bound for transmission interface k, and the
variable yit represents the net injection/withdraw in subregion i.
The main source for transmission network data was PowerWorld
(PowerWorld, 2005).

PTDFki yit
i

Tk, t

(12)

With the explicit consideration of transmission constraints, we

augmented the decision variables with additional subscripts to
account for new capacity by subregion. However, that spatial
disaggregation is insufcient for air quality simulations. Therefore,
in Steps 4 and 5, we allocate generating capacity by county.
Fourth Step. The fourth step estimates siting probabilities for
counties within the subregions using empirical logit models (one
for each of three generation technologies11) based on siting decisions during 19952004.12 During that time, the generating capacity for the continental US increased by 40%, from 686 to 962
GW. The distribution of additional generating capacity over that
time period by technology in terms of GW capacity (number of
generating units) for coal-red steam, combined-cycle gas units,
combustion turbines, and other types is 126.6 (  59), 158.7
(1300), 84.8 ( 1094), and  94.2 GW (150), respectively.
(Negative numbers indicate decreases in either capacity or number
of units.) Ideally, this analysis would construct a dataset with repeated observations per year for each county: a panel dataset for
10 years with 3193 counties. However, several independent variables would likely be time-invariant or would be difcult to determine for each year. Thus, we pool the 10 years of data and
conduct a pure cross-sectional analysis.
The empirical logit models estimate how various factors affect
actual siting decisions, and assume that the relationships that
governed siting in the past will also apply to future siting choices.
The equations, one for each generation technology j, are as follows:
9
The reason to go through the transmission-unconstrained model rst in Step
2 is that the combined siting/transmission and generation mix models are too
large, and since NEMS chooses mix within a region without considering transmission, we take an approach consistent with NEMS.
10
The method is relatively standard practice in the power engineering community. We therefore refer to interested readers to additional reference (Hobbs
et al., 2008).
11
The seven technologies we considered initially in the Step 2 are then
grouped into three technologies, i.e., coal, combined cycled and combustion turbines, partially due to the data limitation in siting probabilities. Nuclear plants are
omitted because they are not economically feasible as predicted by the Step 2.
12
We relied here upon two sets of assumptions in this step. First, power plant
facility location can be modeled and determined by the factors we consider in the
logit models. Second, the siting probability in the future will be consistent with
historical trends in 19952004. Of course, public attitudes toward energy projects,
the regulatory environment, and market conditions are likely to change and may
impact the siting decisions in ways that we cannot capture by our analysis. However, we argue that this is a not unreasonable approach since it is based on historical data, and will lead to more plausible outcomes compared to other, more
heuristic approaches that are less consistent with actual siting behavior such as
assigning capacity in proportion to population.

16

Y. Chen et al. / Energy Policy 84 (2015) 1121

Pr(ynsij = 1|x) =

znsij {0, 1, 2... }

exp(F (x))
1 + exp(F (x))

(13)

F (x) = 0j + 0ij +

1j extnsi
j=1

2jp statusnsip

+ 3j dennsi

p=1

+ 4j degis + 5j popnsi + 6j incomensi

(14)

The dependent variable (Pr) is the siting probability (between

0 and 1) for a county, where ynsij 1 indicates that a plant of type j
is sited in county n. The model independently estimates three
types (subscript j) of generating technology: coal for base-load,
combined cycle for cycling capacity, and combustion turbine for
peaking capacity. The independent variables (x) include presence
of existing power generators, ozone attainment status, population
density, state utility deregulation status, county population, and
median income. Ozone attainment status is represented by two
indicator variables for three categories: attained, marginal (or
moderate) and severe, and is obtained from USEPA sources (USEPA, 2008). The restructuring status and proposed new generators
data are from the Energy Information Agency (EIA, 2008; EIA,
2012a, 2012b). Demographic data (e.g., county population and
median income data) are obtained from the US Census Bureau (US
Census Bureau, 2012). We use a random-intercept model (also
called mixed-effect or hierarchical model), given that siting decisions can be modeled as two levels: a number of counties are
nested within a state. The random intercept formulation allows for
the information from other states to be used by a given state
(Goldstein, 2002). For instance, even if a county has no generator
situated, it may still have a positive siting probability when other
counties with similar values of x have generating facilities within
their territory. The term 0ij, is state-specic (subscript i) unobserved random effect for technology j. We assume a underlying
distribution: 0ij  N(0, s20). Thus, the term 0j in Eq. (1) gives an
overall US unobserved intercept for technology j. The resulting
predicted probabilities by county and technology are used in Step
5 to create county-level siting scenarios for new capacity.
Fifth Step. Step 5 sites capacity to allocates capacity to the
county level using a mixed-integer nonlinear program (15)(18)
(MINLP). This MINLP minimizes (15), which is the sum of squared
deviations of sited EGUs from an ideal distribution that is proportional to the siting probabilities predicted by Step 4s empirical
models. Decisions variables are integer variables znsij denoting the
number of facilities with the j-th technology assigned to county n,
lnsij is expected capacity that
state s, and subregion i. The term w
would be assigned to county n if assigned in proportion to the
probability (18). Thus, wnsij is the assigned capacity dened by an
integer variable znsij. Counties with higher probabilities will
therefore obtain more capacity, and integer variables znsij are used
to ensure that sited EGUs are of realistic size. We assume unit
capacities of 600, 400 and 230 MW for coal, combined-cycle and
combustion turbine units, respectively (EIA, 2012a). The term Nij in
(17) is the largest integer of the ratio unj/Ij.

lnsij wnsij )2
min (w
znsij

s, n

Subjectto:

wnsij = znsij Ij

(16)

wnsij = Ni , i
(17)

s, n N (i)

lnsij =
w

(15)

pnsij uij
s, n pnsij

(18)

To account for the fact that there is no transmission representation within ECAR, the new capacity in that region is rst
assigned to each state in proportion to existing capacity. Finally,
since the total capacity of a given type is unlikely to be a multiple
of the assumed unit size, leftover capacity is assigned to the
county within subregion i that has the highest siting probability
for technology j. New plants are assumed to be located at each
county's centroid.
Sixth Step. The sixth (and nal) step generates year-specic
hourly pollutant emissions by EGU using a LP-based short run
least-cost dispatch model, based upon the generation mix and
locations obtained from the earlier models and demand patterns
consistent with the meteorology in each year. In other words, the
model was run fourteen times, once for each year, and is subject to
the demand condition in that year. The model formulation differs
from the capacity model (1) in several ways. First, the model is a
transmission-constrained operations model, and so only considers
short-run costs (fuel, allowance costs, and other variable O&M),
subject to xed capacity of the existing and siting generators. Next,
the simulation period is just the ozone season from May 1 to
September 30, a total of 3672 h. The LDC is approximated using 20
load blocks, with the number of hours in each block ranging from
25 to 300. Lastly, the model is dispatched separately against each
year's summertime LDC, instead of the decadal averages used in
Steps 2 and 3. Thus, the model can be used to assess the variability
in the intra-annual distribution of NOx emissions due to varying
meteorology.

3. Results and discussions

We review the results of each step, starting with the annual
variability of temperature and long-run load from Step 1.13 Fig. 3
plots their duration curves for the entire study region over ozone
season: summertime temperature (left) and load (right) duration
curves for study region (the US mid-Atlantic and partial ECAR) for
1990s and 2050s climates, which show the number of hours that
each respective quantity exceeds the value given at y-axis. Not
only does the 2050s series (warming climate) lie above normal
climate group (1990s climate), but the former series also has a
greater interannual variability for both load (left) and temperature
(right). The load variability is greatest during high-demand hours
(left end of the x-axis). This is likely to yield signicant year-toyear variation in average ozone levels as well as numbers of severe
ozone episodes, because hotter, higher load years will also be more
likely to have conditions favorable to ozone formation.14 More
importantly, if environmental damage as a function of emissions is
convex, consideration of emissions only from an average load year
may understate the ozone impact. That is, the average ozone
concentration (or average days of ozone NAAQS exceedences per
13
We would like to emphasize that our main focus of the paper is on the
framework we developed, and the numeric results we discussed in this section are
based on a situation at which no renewable addition and no other regulatory policies
that were introduced to suppressed output from coal and other high emitting eets.
14
This assumes that the frequency of severe pollution is a nonlinear function of
weather conditions. Then Jensen's inequality can be invoked: the expected value of
a convex function E(f(X)) is at least equal to, and generally greater than f(E(X)), the
function evaluated at the expected value of the input. In this context, X is annual
weather conditions, and f(X) is an air quality index such as frequency of ozone
NAAQS exceedences. For instance, in a cool year, there may be two exceedences,
while an average year might have ve exceedences. But a hot year might have 23
exceedences. If those three possibilities are equally probable, then E(f(X))
(2 5 23)/3 10 4f(E(X)) 5. It is a well accepted principle of risk analysis that if
a system is nonlinear, a full distribution of inputs (such as meteorology) should be
considered, not just average or typical conditions.

Y. Chen et al. / Energy Policy 84 (2015) 1121

17

Fig. 3. Summertime temperature (left) and load (right) duration curves for study region (mid-Atlantic and partial ECAR) for 1990s and 2050s climates.

year) might be less for an average load year than when calculated
over a sample of years reecting year-to-year temperature variations. In other words, if a system is nonlinear, full distributions of
inputs (such as meteorology) should be considered, not just
average conditions.
The long-run capacity expansion in Step 2 is based on average
load distributions within each decade (1990s and 2050s). The 20block summertime LDCs t to the data in Fig. 3 are combined with
NEMS 2025 non-summer blocks to form a 30-block system in the
LDC. Table 1 summarizes the results using the transmission-unconstrained model in Step 2. The overall estimated additional capacity is greater than the average peak demand in Fig. 4 due to the
model's inclusion of a reserve requirement (7.5%, consistent with
NEMS assumptions (EIA, 2012a). The capacity mix reects not only
the load increase in 2025 relative to today, but also changes in the
load prole. The increased peak load resulting from the 2050s
climate means that about 20,000 MW (about 10%) more capacity is
required compared to the 1990s climate run.
Results from Step 2 (Table 1) are then used by the transmissionconstrained model for allocating capacity to subregions. Compared
with the 1990s capacity allocation, more combustion turbines in
2050s are projected to be allocated to a few subregions: PECO,
PPL1, PPL2, BGEPEP. The concentration of turbines in those subregions is due to their proximity to load centers. However, given
the low NOx emission rates of new turbines, their impact on local
air quality is expected to be negligible. Most capacity is allocated
to the ECAR region, given its proximity to fue1 sources. However,
the fact that transmission lines from ECAR to the mid-Atlantic are
congested more than 35% of the time limits the additional capacity
that can be allocated to ECAR. Finally, because of cheaper coal in
PPL3 compared to Maryland, most coal plants in the mid-Atlantic
are assigned to PPL3 to meet load in PECO and BGE2.
The empirical logit siting models (Step 4) together with the
MINLP EGU siting model (Step 5) are the means we use to allocate
each subregion's new capacity to its counties. (The estimated logit
models can be found in the on-line appendix.) The outcome represents the spatial distribution of capacity that is consistent with
both historical trends and the market conditions simulated by the
LPs of Steps 2 and 3.
Figs. 45 present the EGU siting results for the 1990s and 2050s
climates, respectively. The symbol in Fig. 5 indicates the locations of the additional combustion turbines under 2050s climate.
For each point shown, there is at least one new generator situated
at that location. In both scenarios, the new capacity is primarily
allocated to ECAR because of its less expensive fuel. Incremental

Table 1
Summary of results from LP capacity mix models of mid-Atlantic and partial ECAR
for 1990s and 2050s climates (2025 generation and load conditions considered)
1990s climate 2050s climate
Average Temperature, Ozone Season [K]
Average Peak Load [MW]
Existing Capacity [MW]
New Coal-red Steam [MW]
New Gas-Fired Combined Cycle [MW]
New Combustion Turbines [MW]
NOx price [$/ton] (2004 dollars)
Average NOx emissions during top 25 hours
[tons]

295
178,105
90,564
30,394
78,439
0
16,947
119

293
196,030
90,564
35,022
86,602
7,727
17,872
127

combustion turbine siting for the 2050s compared to the 1990s is

also indicated in Figs. 45, respectively. The additional combustion
turbines under the 2050s warming climate conditions are located
near the load centers in order to supply peaking demand during
hot summertime. This spatial allocation of new capacity is required to project the spatial distribution of emissions in the next
step.
The nal step of the analysis is to generate hourly NOx emissions for the ozone season using the transmission-constrained
operations model considering both existing and new capacity. The
3672 h in this season are clustered into 20 periods with similar
load levels. Fig. 6 plots the average NOx duration curves for the
1990s and 2050s scenarios for our study region. The thick solid
line in the plot are the average of NOx emissions duration curves
over 8 years of the 1990s, while the thick dashed sold line portrays
the 6 years of the 2050s. The plot suggests that for the top 750 h,
NOx emissions for 2050s are greater than 1990s case by a margin
of 38 t per hour, with the highest difference occurring for the top
25 h (8 t/hr, or 7%) (Table 1). For hours 1500 to 3672 the emissions
proles are nearly identical. Since total NOx emissions are capped
for each year, the area under the average NOx duration curves will
equal the allowances cap. This explains why the 1990s curves emit
more NOx during the hours 7501500.
To construct the condence intervals in Fig. 6, we boot strapped
the NOx emissions 1000 times, assuming that, for every block, each
of the 8 years (for 1990s) and the 6 years (for 2050s) is equally
likely to occur. The uncertainty is represented by the thin dashed
lines delineating 95% condence intervals based upon the 1000
samples of 8 years and 6 years data for two series, respectively.
Although Fig. 6 indicates signicant variation in peak emissions
from year to year as gauged by the 95% CI (marked by the

18

Y. Chen et al. / Energy Policy 84 (2015) 1121

Fig. 4. County level allocation of new capacity under 1990s climate scenarios.

horizontal lines), the between-year variation in NOx emissions

under allowances banking could be even higher than simulated.15,
16
This is because we assumed no banking of allowances from one
year to the next. But in reality, a rm may emit more in a hot
summer if it can use banked allowances from previous years. On

the other hand, if demands are low, then a rm may choose to

emit less and bank surplus allowances.
Overall, the variability of hourly emissions is larger during high
emissions periods. For our study region, the standard deviation of
the highest-emission 25 h is approximately equal to 10 and 15 t/h

Fig. 5. County level allocation of new capacity under 2050s climate scenarios.

Y. Chen et al. / Energy Policy 84 (2015) 1121

1990s
1990s 95% CI

150

Hourly NOx Emissions [tons]

Hourly NOx Emissions [tons]

150

100

50

19

2050s
2050s 95% CI

100

50

0
0

1000

2000

3000

Hours [hours]

1000

2000

3000

Hours [hours]

Fig. 6. Average summer NOx duration curves for the US mid-Atlantic and ECAR for 1990s (left) and 2050s (right) climate conditions.

(8 and 12%) for 1990s and 2050s, respectively (Fig. 6), which is
substantially larger than in the other hours. Thus, climate change
increases emissions of NOx during warm periods (when peak demands occur), and thus likely increases the frequency of severe
ozone episodes. Of course, air quality simulations would be necessary to verify this conjecture.
The variability associated with NOx duration curves is mainly
the consequence of demand uctuations due to meteorology, not
because of EGU efciency degradation as a result of warm weather.
In our analysis, for each 1 F increase in the ambient temperature,
generating capacity is adjusted downward on average by 0.4%, and
efciency (i.e., heat rate) is worsened by 0.6 and 0.7% for combined-cycle and combustion turbine EGUs, respectively. However,
if these changes are not considered, the distribution of emissions
is only slightly altered, in part due to the seasonal cap.
We also examined subregional emissions. A one-tailed t-test
shows that the mean peak-period (highest 25 h) seasonal NOx
emissions for the 2050s is signicantly (p o5%) greater than for
the 1990s climate for the ME1, ME2, PPL1, BGEPEP, PL2, JC2, PN,
and mid-Atlantic subregions. But this is not true for the study
region as a whole, due to the cap. This suggests that climate
change's effects on emissions differ by location.
Finally, we examine the relationship between hourly NOx
emissions and corresponding ambient temperatures. We created a
detailed load duration curve that includes 173 blocks: 153 single
hour blocks, each representing the 2 p.m. load of a single day, and
20 blocks that aggregate the remaining hours. This allows us to
15
In calculating CIs, we assume that the data within a given block are independent from their counterpart from other scenarios. We acknowledge that they
could be positively correlated due to similar load levels as a result from block denition. In that case, we then under estimate the CIs.
16
Of course, one alternative way to making the comparison is by scaling the
historical emission (1990s) by load growth or other parameters to reect the future
scenarios. However, as historical NOx emissions are likely to also be impacted by
other factors not consider in the analysis, this approach would make 1990s emissions less comparable with the 2050s emissions, which were generated by our
analysis.

study the relationship between temperature and NOx emissions

during that particular hour. Fig. 7 displays the scatter plots of
ambient temperature versus hourly NOx emissions for ECAR. The
different symbols in Fig. 7 represent two selected years from two
series: 1995 () and 2055 (  ). A trend line based upon linear
regression is also plotted for the 1990s (solid) and 2050s scenarios
(dashed). Each point corresponds to a pair of reported NOx hourly
emission (Y-axis) and ground-level temperature (X-axis) in the
ECAR region for a specic year among 14 years, which tends to be
near or at the time of peak power demand. In other words, there
are a total of 2422 data points (173  14) in Fig. 7. The gure
shows that the warming climate generally increases hourly NOx
emissions as well as temperatures during this hour, as the 2050s
106

1990s
2050s
105

104

103

102

101

285

290

295

Temperature [k]

Fig. 7. Scatter plots of the ECAR hourly NOx emissions at each 2:00 p.m. versus
ambient temperature during summertime based on re-running Step 6s dispatch
model.

20

Y. Chen et al. / Energy Policy 84 (2015) 1121

points are somewhat shifted up and to the right of the 1990s

points.
The correlation between ambient temperature and hourly NOx
emissions is around 0.7. This suggests that under warming
weather, hourly NOx emissions could be considerably higher as a
consequence of increased electricity demand, even if annual
emissions are unaffectedby designdue to the seasonal cap. Yet
the temperature-NOx relationship is subject to a number of limitations, including the omission of dynamic constraints upon
generator operation in the model of Step 6, such as min-run levels
or ramp rate limits. Because these constraints are omitted, the
model may mis-predict the generators' actual output level in
particular hours. Since NOx emission rates could be highly nonlinear for some units such as combustion turbines, the hourly NOx
emissions might be miscalculated as well.
This study is subject to several limitations. First, the characteristics of future technologies, the exact location of new emissions sources, and the nature of future pollution laws is highly
uncertain. For instance, the location of each county is represented
by its geometric centroid. Emissions from new generators associated with that county are assumed to occur at that geographic
point. Thus, this approach may over-concentrate air pollution
emissions locally in subsequent fate and transport modeling. Siting and generation mix decisions by utilities could also be inuenced by other factors that we did not consider in this analysis.
Second, we assume that power plants cannot bank allowances
between years, although banking is permissible in reality. Thus,
our results might overestimate the impact of climate change in the
peaking periods. One way to explore the variation of NOx emissions under banking scenarios is to adjust emissions in each year
so that the marginal cost (permit price) of emissions is the same in
each year, and the average annual emissions meet the cap.
Third, as examples of regulatory change, the Regional Greenhouse Gas Initiative (RGGI) or recently announced Clean Power
Plan by USEPA are not considered in our analysis. They could affect
our conclusions because, in the absence of federal CO2 limits, RGGI
would encourage power plants located in the upwind ECAR (nonRGGI) states to increase their output in the short run or to build
more coal-red plants in the long-run. The consequence would
then be to lower emissions in RGGI states but to increase them in
ECAR (Ruth et al., 2008; Chen, 2009). Our framework can be used
to quantify the impact of RGGI or other CO2 regulatory scenarios
by imposing a CO2 price in the appropriate regions.
Fourth, we limit our attention to study a worst-case scenario
since our intention is to (1) demonstrate how the proposed
methodology, relies on a series of optimization-based and empirical siting models, can be used to construct plausible future
scenarios considering climate change and (2) to illustrate the fact
that climate change could have a signicant impact on air quality
especially during hot summertime and/or high demand hours in
mid-Atlantic region through its impact on generating capacity and
technology choices. However, the approach can be modied to
incorporate changes in other considerations that might impact
pollution emissions, such as fuel prices, demand or population
growth, technology advances, and changes in energy and environmental policies. The downscaled emissions can be used in
fate and transport models such as the Community Multiscale Air
Quality (CMAQ) to project changes in tropospheric ozone due to
climate change.
Finally, we are fully aware of the fact that anthropogenic NOx
emissions are also closely related to other pollutants, such as ne
particular matter, carbon monoxide, and VOCs, all of which are
also emitted by power plants and contribute to air pollution. Admittedly, the paper is a partial analysis, which addresses the impact of NOx emissions directly from the power sector. However,
the paper is the rst attempt, to the best of our knowledge, to

integrate a series of models, i.e., capacity expansion, operation,

short and long-run load forecast, empirical facility siting models,
to study climate change-induced impact on NOx emissions.

4. Conclusions and Policy Implications

This paper proposed a framework that is capable of examining
the long-run effects of climate change on the spatial and temporal
distribution of NOx emissions by the power sector. The framework
is based on a series of optimization-based market simulation
models that represent future power plant investment and operating decisions. While the study is subject to some limitations as
alluded to earlier, the framework we developed in this paper is
transparent and economic sound. It is also exible, allowing
changes in environmental and energy regulation, technological
advances, market conditions and other factors to be incorporated
in the analyses. Given the worse-case scenarios we considered for
the US mid-Atlantic and ECAR regions, the results show that climate-induced changes in the quantity and temporal distribution of
electricity demand will substantially modify the mix of generation
capacity and distribution of pollution emissions in the long-run,
even in the presence of a seasonal emissions cap. It suggests that
signicantly higher emissions during peak demand hours will
occur, possibly worsening regional air quality.
While national air emissions are reported to have declined over
the past several years (USEPA, 2012), the more frequent occurrences of extreme air quality episodes in some regions pose a
signicant threat to public health (Lin et al., 2001). Given a set of
worse-case assumptions, the analysis in this paper shows that
higher emissions during peak demand hours could contribute to
increases in this frequency in the future. Thus, in addition to a
current seasonal cap system, a separate cap that limits emissions
in those hours, a pollution tax that deters outputs from polluting
sources or other policies that are directed at prevention of worsening air quality under forecast extreme weather should be
considered regulatory agencies in order to avoid extreme air
quality episodes (Marnay, 1993; Wang et al., 2005; Sun et al.,
2012). Any of these instruments would provide economic incentives for the power sector to opt for cleaner generation eets,
which in turn, would lower the pollution emissions under extreme
weather conditions.

Acknowledgments
The authors would like to thank John Cymbalski (EIA) and
Frank Morra (Booz-Allen-Hamilton) for assistance with the NEMS
model. This work was supported by USEPA STAR Grants
R929731010 and RD83183601 and by NSF Grants ECS 0080577 and
0224817. Any errors are our responsibility.

References
Amato, A.D., Ruth, M., Kirshen, P., Horwitz, J., 2005. Regional energy demand responses to climate change: methodology and application to the commonwealth
of Massachusetts. Climatic Change 71, 175201.
Anderson, H.R., Derwent, R.G., Stedman, J., 2001. Air pollution and climate change.
Health effects of climate change in the UK.. In: McMichael, A.J., Kovats, R.S.
(Eds.), UK Department of Health, London, UK, pp. 193217.
Bell, M.L., Goldberg, R., Hogrefe, C., Kinney, P.L., Knowlton, K., Lynn, B., Rosenthal, J.,
Rosenzweig, C., Patz, J.A., 2007. Climate change, ambient ozone, and health in
50 US cities. Clim. Change 82 (12), 6176.
Bell, M.L., McDermott, A., Zeger, S.L., Samet, J.M., Dominici, F., 2004. Ozone and
short-term mortality in 95 US urban communities, 19872000. J. Am. Med.
Assoc. 292 (19), 23722378.
Bell, M.L., Peng, R.D., Dominici, F., 2006. The exposure-response curve for ozone and

Y. Chen et al. / Energy Policy 84 (2015) 1121

risk of mortality and the adequacy of current ozone regulations. Environ.

Health Perspect. 114 (4), 532536.
US Census Bureau. 2012. Jan. 2006. Poverty and median incomes estimates. From
http://www.census.gov/hyousing/saipe/estmod03/est03All.xls.
CCSR. Center for climate systems research, 2006. From http://www.ccsr.columbia.
edu/.
Chen, Y., 2006. Analyzing interaction of electricity market and environmental policy
using equilibrium models. The Johns Hopkins University, Ph.D. Thesis).
Chen, Y., 2009. Does a regional greenhouse gas policy make sense? A case study of
carbon leakage and emissions spillover. Energy Econ. 31 (5), 667675.
Civerolo, K.L., Hogrefe, C., Lynn, B., Rosenzweig, C., Goldberg, R., Rosenthal, J.,
Knowlton, K., Kinney, P.L., 2008. Simulated effects of climate change on summertime nitrogen deposition in the eastern US. Atmos. Environ. 42 (9),
20742082.
Colwell, R.R., 1996. Global climate and infectious disease: the cholera paradigm.
Science 274, 20252031.
Commission, E. 2014a. Review of the EU Air policy. Retrieved from: http://ec.
europa.eu/environment/air/review_air_policy.htm (Retrieved 2014).
Commission, E. 2014b. Thematic Strategy on Air Pollution. Retrieved from: http://
europa.eu/legislation_summaries/environment/air_pollution/l28159_en.htm
(Retrieved 10.12.2014).
Constable, J.V.H., Guenther, A.B., Schimel, D.S., Monson, R.K., 1999. Modelling
changes in VOC emission in response to climate change in the continental
United States. Global Change Biol. 5, 791806.
Crowley, C., 2006. Modeling in Natural Resource Economics: Exploring Three
Techniques. The George Washington University, Ph.D. Thesis.
DOE. 2000. Analysis of strategies for reducing multiple emissions from power
plants: sulfur dioxide, nitrogen oxides, and carbon dioxide. Retrieved from:
http://www.eia.doe.gov/oiaf/servicerpt/powerplants/index.html (Retrieved
24.02.2009).
DOE. 2009. Assumptions to the Annual Energy Outlook 2008: with projection for
2030. Retrieved from http://www.eia.doe.gov/oiaf/aeo/assumption/pdf/0554
(2008).pdf (Retrieved 24.02.2009).
Donaldson, G., Kovats, R.S., Keatinge, W.R., McMichael, A.J., 2001. Heat- and coldrelated mortality and morbidity and climate change. Health Effects of Climate
Change in the UK., A. J. McMichael and R. S. Kovats. Department of Health,
London, UK, UK, pp. 193217.
ECAR. 2006, June 12, 2004. East central area reliability council. Retrieved from:
http://www.ecar.org/.
EIA. 2006. Assumptions to the Annual Energy Outlook 2006. Retrieved from http://
www.eia.doe.gov/oiaf/aeo/assumption/pdf/electricity.pdf (Retrieved
20.05.2006).
EIA. 2008. US state electricity deregulation status. Retrieved from: http://www.eia.
doe.gov/cneaf/electricity/page/restructure.html.
EIA. 2010. Existing electric generating units in the United States. Retrieved from:
http://www.eia.doe.gov/cneaf/electricity/epa/epat2p4.html.
EIA. 2012a. Assumptions to the annual energy outlook 2006: electricity market
module. Retrieved from: http://www.eia.doe.gov/oiaf/aeo/assumption/pdf/
electricity.pdf.
EIA. 2012b. Form EIA-860 database annual electric generator report. Retrieved
from: http://www.eia.doe.gov/cneaf/electricity/page/eia860.html.
EIA. 2014. The national energy modeling system: an overview. Retrieved from:
http://www.eia.gov/oiaf/aeo/overview/ (Retrieved 08.12.2014).
ETSAP. 2006. MARKAL. Retrieved from: http://www.etsap.org/Tools/MARKAL.htm
(Retrieved 12.06.2006).
Europe, U. N. E. C. f. 2014. Gothenbug Protocol: Guidance documents and other
methodological materials for the implementation of the 1999 Protocol to Abate
Acidication, Eutrophication and Ground-level Ozone. Retrieved from: http://
www.unece.org/environmental-policy/treaties/air-pollution/guidance-docu
ments-and-other-methodological-materials/gothenburg-protocol.html (Retrieved 10.12.2014).
Goldstein, H., 2002. Multilevel statistical method. Oxford University Press.
Hadley, S.W., Erickson, D.J., Hernandez, J.L., Broniak, C.T., Blasing, T.J., 2006. Responses of energy use to climate change: a climate modeling study. Geophys.
Res. Lett. 33, L17703.
Hobbs, B.F., 1995. Models for integrated resource planning by electric utilities, invited review. Eur. J. Oper. Res. 83 (1), 120.
Hobbs, B.F., Drayton, G., Fisher, E.B., Lise, W., 2008. Improved transmission representations in oligopolistic market models: quadratic losses, phase shifters,
and DC lines. IEEE Trans. Power Syst. 23 (3), 10181029.
Jacob, D.J., Winner, Darrell A., 2009. Effect of climate change on air quality. Atmos.
Environ. 43 (1), 5163.
Kosatsky, T., 2005. The 2003 European heat waves. Euro Surveill. 10 (7), 148149.
Lin, C.-Y.C., Jacob, D.J., Jacob, A.M., 2001. Trends in exceedances of the ozone air

21

quality standard in the continental United States, 19801998. Atmos. Environ.

35, 32173228.
Marnay, C., 1993. Intermittent electrical dispatch penalties for air quality improvement. University of California, Ph.D.
Morris, S.C., Goldstein, G.A., Ftjemalos, V.M., 2004. Nems and markal-macro Models
for energy-environmental-economic analysis: a comparison of the electricity
and carbon reduction projections. Environ. Model. Assess. 7, 207216.
Nakicenovic, N., Swart, R., 2000. Special report on emissions scenarios, Cambridge,
UK, Intergovernmental Panel on Climate Change (IPCC). Cambridge University
Press.
NERC. 2006. Generating availability reports and historical availability statistics.
Retrieved from http://www.nerc.com/ lez/gar.html (Retrieved 13.12.2005).
Orru, H., Andersson, C., Ebi, K.L., Langner, J., Astrom, C., Forsberg, B., 2013. Impact of
climate change on ozone-related mortality and morbidity in Europe. Eur. Respir. J. 41 (2), 285294.
PJM 2003. 2002 state of the market. Norristown, PA, PJM Interconnection, Market
Monitoring Units: pp. 101126.
PowerWorld. 2005. Power world corporation: the visual approach to analyzing
power systems. Retrieved from: http://www.powerworld.com/.
Racherla, P.N., Adams, P.J., 2006. Sensitivity of global tropospheric ozone and ne
particulate matter concentrations to climate change. J. Geophys. Res. Atmos. 111
(D24).
Ruth, M., Gabriel, S.A., Palmer, K.L., Burtraw, D., Paul, A., Chen, Y., Hobbs, B.F., Irani,
D., Michael, J., Ross, K.M., Conklin, R., Miller, J., 2008. Economic and energy
impacts from participation in the regional greenhouse gas initiative: a case
study of the State of Maryland. Energy Policy 36 (6), 22792289.
Schweppe, F.C., Carmanis, M.C., Tabors, R.D., Bhon, R.E., 1988. Spot pricing of electricity. Kluwer.
Seinfeld, J.H., Pandis, S.N. 1997. Atmospheric chemistry and physics: from air pollution to climate change, Willey interscience.
Sillman, S., Samson, P.J., 1995. Impact of temperature on oxidant photochemistry in
urban, rural and remote environments. J. Geophys. Res. 100, 14971508.
Sun, L., Webster, M., McGaughey, G., McDonald-Buller, E.C., Thompson, T., Prinn, R.,
Ellerman, A.D., Allen, D.T., 2012. Flexible NOx abatement from power plants in
the Eastern United States. Environ. Sci. Technol. 46 (10), 56075615.
Tanser, F.C., Sharp, B., Sueur, D. l, 2003. Potential effect of climate change on malaria
transmission in Afriaca. Lancet 362, 17921798.
USEPA. 2005a. Clean Air Interstates Rules. Retrieved from: http://www.epa.gov/
cair/ (Retrieved 13.12.2005).
USEPA. 2005b. Emissions and generation resource integrated database (eGRID).
Retrieved from: http://www.epa.gov/cleanenergy/egrid/index.htm (Retrieved
13.12.2005).
USEPA. 2005c. EPA modeling applications using the integrated planning model.
Retrieved from: http://www.epa.gov/airmarkets/epa-ipm/ (Retrieved
10.12.2005).
USEPA 2005d. Technology transfer network emissions measurement center.
USEPA. 2006. Biogenic emissions characterization. Retrieved from: http://www.
epa.gov/appcdwww/apb/biogenic.htm.
USEPA. 2008, March 10, 2006. 8-Hour ozone nonattainment areas. Retrieved from:
http://www.epa.gov/oar/oaqps/greenbk/o8index.html.
USEPA 2009. Assessment of the impacts of global change on regional U.S. air
quality: a synthesis of climate change impacts on ground-level ozone.
USEPA. 2012. National emissions inventory (NEI) air pollutant emissions trends
data. Retrieved from: http://www.epa.gov/ttn/chief/trends/index.html.
Wang, L., Allen, D.T., McDonald-Buller, E.C., 2005. Air quality modeling of interpollutant trading for ozone precursors in an urban area. J. Air Waste Manag.
Assoc. 55 (10), 15431557.
Weaver, C.P., Cooter, E., Gilliam, R., Gilliland, A., Grambsch, A., Grano, D., Hemming,
B., Hunt, S.W., Nolte, C., Winner, D.A., Liang, X.Z., Zhu, J., Caughey, M., Kunkel, K.,
Lin, J.T., Tao, Z., Williams, A., Wuebbles, D.J., Adams, P.J., Dawson, J.P., Amar, P.,
He, S., Avise, J., Chen, J., Cohen, R.C., Goldstein, A.H., Harley, R.A., Steiner, A.L.,
Tonse, S., Guenther, A., Lamarque, J.F., Wiedinmyer, C., Gustafson, W.I., Leung, L.
R., Hogrefe, C., Huang, H.C., Jacob, D.J., Mickley, L.J., Wu, S., Kinney, P.L., Lamb, B.,
Larkin, N.K., McKenzie, D., Liao, K.J., Manomaiphiboon, K., Russell, A.G., Tagaris,
E., Lynn, B.H., Mass, C., Salath, E., ONeill, S.M., Pandis, S.N., Racherla, P.N., Rosenzweig, C., Woo, J.H., 2009. A preliminary synthesis of modeled climate
change impacts on U.S. Regional ozone concentrations. Bull. Am. Meteorol. Soc.
90 (12), 18431863.
Woo, J.H., He, E., Fau-Tagaris, S., Tagaris, K.J., Fau-Liao, E., Fau-Manomaiphiboon, K.
Liao Kj, Manomaiphiboon, P., Fau-Amar, K., Amar, A.G., Fau-Russell, P., Russell,
A.G., 2009. Development of North American emission inventories for air quality
modeling under climate change. J. Air Waste Manag. Assoc. 58 (11), 11831194.