A Model of Energy Policy Impacts on Pollutant Emissions, Costs,
and Social Benefits Developed for China’s Central Government
Jeffrey Rissman
Energy Innovation: Policy and Technology LLC
98 Battery St. Ste. 202
San Francisco, CA 94111
(415) 799-2169
jeff@energyinnovation.org
In partnership with China’s National Center for Climate Change Strategy and International
Cooperation (NCSC) and China’s Energy Research Institute (ERI)
Abstract
Energy Innovation LLC has worked with China’s central government to develop a System Dynamics
model to assist in selecting policies that will allow China to achieve its emissions reduction goals. The
model simulates years 2013-2030 and covers the Transportation, Electricity Supply, Buildings, and
Industry sectors. It also handles District Heating and Carbon Capture and Sequestration. The effects
of 35 energy policies, as well as increased technical progress through research and development
(R&D), may be investigated in any combination. Numerous outputs are available, including: emissions
of nine different pollutants; cash flow changes for government, industry, and consumers; monetized
social benefits from avoided public health and climate damages; usage of nine fuels as well as
electricity and heat; and the mix of power sources. A Python script can be used to identify optimized
policy packages.
Quantitative results are described in the paper. Some qualitative conclusions: No single policy or
technology is a silver bullet; the greatest emissions reductions at lowest cost are achieved via packages
incorporating many policies that support a diverse set of technologies. It is possible for China to peak
its carbon emissions in the early 2020s while achieving a net reduction in direct monetary outlays.
Keywords
China, pollution, emissions, policy, energy, climate change
Project Background and Motivation
The People’s Republic of China has industrialized rapidly in the last two decades and is now the
world’s largest emitter of greenhouse gasses (GHGs). GHGs are the primary drivers of climate
change, which if unchecked, will have devastating impacts on human societies and on the
environment (Intergovernmental Panel on Climate Change Working Group II, 2014). Many Chinese
cities also suffer from extremely high levels of localized air pollutants, including particulate matter
(PM), nitrogen oxides (NO,), sulfur oxides (SO,), and volatile organic compounds (VOCs). These
levels of pollution are harmful to public health. Chen et al. found that in northern Chinese cities, life
expectancy is 5.5 years lower than in southern cities, due to their use of coal-fired heating in winter
(Chen et al., 2013).
In response to these hazards, the Chinese central government desires to reduce China’s pollutant
emissions. Specifically, they wish to include policies to reduce emissions in China’s forthcoming
13* Five-Year Plan, which will guide the country’s economic development in the years 2016-2020,
along with measures that will enable China to meet its recent, bilateral accord on emissions with
the United States (Nakamura and Mufson, 2014).
A policymaker seeking to reduce emissions faces a dizzying array of policy options that might
advance this goal. Policies may be specific to one sector or type of technology (for instance, light-
duty vehicle fuel economy standards) or might be economy-wide (such as a carbon tax). Sometimes
a market-driven approach, a direct regulatory approach, or a combination of the two can be used to
advance the same goal. For instance, in order to improve the efficiency of home appliances, a
government might offer rebates to buyers of efficient models, might mandate that appliance
manufacturers meet specific energy efficiency standards, or both. In order to navigate this field of
options, policymakers require an objective, quantitative mechanism to determine which policies
will meet their goals and at what cost.
Many studies of energy policy have examined particular policies in isolation. However, it is of
greater value to policymakers to understand the effects of a package of different policies, because
the policies may interact. This can produce results different from the sum of the effects of the
policies when studied individually. For example, a policy that promotes energy efficiency and a
policy that reduces the cost of wind energy, enacted together, are likely to reduce emissions by a
smaller amount than the sum of each of those two policies enacted separately. This is because
some of the electricity demand that was eliminated via the efficiency policy would otherwise have
been supplied by additional zero-emissions wind generation caused by the wind policy. In this
case, the total effects are less than the sum of the individual effects. The opposite is also possible.
For example, a policy that promotes the electrification of light-duty vehicles and a policy that makes
wind cheaper are likely to do more together to reduce emissions than the sum of these policies’
individual effects.
Thanks to the strength of computer models at simulating complex systems, we felt that a
customized computer model would be a crucial tool with which we could assist Chinese
policymakers in evaluating a wide array of different policies. Such a model would need to meet
several requirements: it would need to represent the entire economy and energy system with an
appropriate level of disaggregation, the code would need to be editable by us and sharable with the
Chinese government, and it would need to be possible to represent many policies of diverse types
in this model without unreasonable programming effort. We required outputs that included not
only energy use and emissions, but also economic costs and benefits. Additionally, the model would
need to capture the interactions of policies and other forces in a system whose parameters change
dramatically over the 18-year model run, as China continues to grow and develop. We reviewed
many models and model-creation platforms before determining that no existing model met our
requirements.
Accordingly, we resolved to build a suitable model ourselves. We identified System Dynamics as
the most appropriate intellectual and technical framework for this model, thanks to its focus on
interactions within non-equilibrium systems, the visual presentation of model structure in most
System Dynamics model editors (such as Vensim®), the ability to execute models rapidly (allowing
for real-time experimentation and learning), and the comparative ease of training individuals
without a programming background to use and edit the model.
We partnered with two organizations within the Chinese central government to develop a suitable
model and populate it with data: the National Center for Climate Strategy and International
Cooperation (NCSC) and the Energy Research Institute (ERI). In addition to input from our
partners in China, we have benefitted from the advice of individuals from organizations in the U.S.
with expertise in China’s energy system or energy model development: the Massachusetts Institute
of Technology, Stanford University, Lawrence Berkeley National Laboratory's China Energy Group,
and Climate Interactive. The model has also been reviewed by individuals at a number of
organizations (see the Acknowledgements section below for details), whose comments have helped
to improve the model and expand its capabilities. We call this model the “Policy Solutions model.”
Alongside the model, we have developed several custom tools to assist in obtaining and sharing
output. First is a set of scripts written in the Python programming language. One script allows a
user to specify policies and settings of interest (for instance, various carbon tax rates). The script
will then perform many thousands of runs, combining policy settings in every unique combination,
and log the output to a data file that can be easily imported and manipulated in statistical software.
This script is useful for finding optimal policy packages that meet specific criteria. Other Python
scripts enable the testing of the contribution of each individual policy to a given policy package and
the logging of data from a set of predefined packages.
The other important custom tool is a web application written in Ruby that runs on a server and
provides an internet-accessible, simple, but powerful front-end for selecting policy settings,
running the model, and visualizing and exporting output. This tool will increase the model’s
accessibility to individuals who have limited technical skills or who do not wish to install Vensim
software on their own computers.
Structure and Functionality of the Policy Solutions Model
The Policy Solutions model assesses the effects of 35 energy and environmental policies on a
variety of metrics, including the emissions of nine pollutants; cash flow changes for government,
industry, and consumers; the composition of the electricity generation fleet; the usage of various
fuels; and monetized social benefits from avoided public health impacts and climate damages. The
model is designed to operate at national scale and focuses on four sectors: transportation,
electricity supply, buildings, and industry. The model reports outputs at annual intervals with an
initial year of 2013 and a final year of 2030.
Unlike many energy and economic computer models, the Policy Solutions model does not construct
a future business-as-usual or reference scenario. Instead, it uses a Reference scenario (based on the
results of other scientists’ studies and models) as input data. The Policy Solutions model then
modifies the Reference scenario in response to the policy settings selected by the user. This
approach enables us to take advantage of the good work that has been done in this field, while
providing novel capabilities to analyze policy options that are immediately useful to policymakers
and suggest specific policy actions that could be undertaken.
System Dynamics
There exist a variety of approaches to representing the economy and the energy system ina
computer simulation. The Policy Solutions model is based on a theoretical framework called
“System Dynamics.” As the name suggests, this approach views the processes of energy use and the
economy as an open, ever-changing, non-equilibrium system. This may be contrasted with
approaches such as computable general equilibrium (CGE) models, which regard the economy as an
equilibrium system subject to exogenous shocks, or disaggregated technology-based models, which
focus on the potential efficiency gains or emissions reductions that could be achieved by upgrading
specific types of equipment.
System Dynamics models often include “stocks,” or variables whose value is remembered from
timestep to timestep, and which are affected by “flows” into and out of these variables. The Policy
Solutions model uses stocks for two purposes: tracking quantities that grow or shrink over time
(such as the total solar electricity generation capacity) and tracking differences from the BAU input
data that tend to grow over the course of the model run (for instance, the cumulative differences
caused by enabled policies in the potential fuel consumption of the light-duty vehicle fleet).
' The electricity sector is an exception. Policies in the electricity sector can affect decisions about which types
of power plants to build and how plants are dispatched, so a decision-making framework must be employed.
The decisions made by this framework using Reference input data may result in different outputs from other
models, so in order to ensure our policy case is identical to our Reference case when all of the policies are
disabled, we need to run Reference input data through our decision-making logic to construct a Reference
case.
System Dynamics models often use the output of the previous timestep’s calculations as input for
the following timestep. The Policy Solutions model follows this convention, with quantities such as
the electricity generation fleet, the types and efficiencies of building components, etc. remembered
from one year to the next. Therefore, an efficiency improvement in an early year will result in fuel
savings in all subsequent years, until the improved vehicle/building component/etc. is retired from
service. The Industry sector is handled differently: as the available input data come in the form of
potential reductions in fuel use and process-related emissions by policy, we gradually implement
these reductions (with corresponding implementation costs), rather than recursively tracking a
fleet-wide efficiency. (Due to the diverse forms that input data take in the sectors we model, rarely
does one approach work for all sectors. Accordingly, the Policy Solutions model attempts to use
whichever approach makes the most sense in the context of each specific sector.)
One way in which the Policy Solutions model differs from many System Dynamics models is its
handling of time delays. Many System Dynamics models explicitly implement delays before
compliance with new policies or responses to other changing conditions, reflecting real-world
factors related to human psychology, inertia in business practices and supply chains, etc. (Sterman,
2000, p.409). The Policy Solutions model does not explicitly implement these types of delays.
Policy effects are implemented in one of two ways. Most policy effects are phased in linearly by the
model's end year (2030). For example, if the user selects a carbon tax of $10/ton CO2e, then
halfway through the model run, the carbon tax will be $5/ton CO2e. Human behavior in the year
halfway through the model run will reflect the costs imposed by the $5/ton carbon tax: there is no
delay that would cause people to base their decisions on a $3/ton or a $4/ton carbon tax, the
prevailing rates a few years prior. Some policies are fully implemented in every model year when
they are turned on. For example, a policy requiring improved labels that highlight the energy used
by building components is implemented fully in the first modeled year (2013) and maintained
through every year, because the meaningfulness of implementing one quarter or one half of an
improved labeling policy is questionable. In these cases, people’s behavior reflects the presence of
the new labels in 2013; there is no delay of a year or two for them to notice and begin factoring the
improved labels into their decisions.#
Model Structure
The model's structure can be thought of as occurring along two dimensions: visible structure that
pertains to the equations that define relationships between variables (viewable as a flowchart in
Vensim) and behind-the-scenes structure that consists of arrays and their elements, which contain
data and are acted on by the equations. For example, the transportation sector’s visible structure
consists of policies (such as a fuel economy standard), input data (such as the Reference cargo
4 This model behavior need not be conceptualized as instantaneous compliance: each policy lever in the
model need not refer to the legislative text of the policy, but could instead refer to people’s delayed responses
to the policy.
distance traveled- that is, passenger*miles or freight ton*miles), and calculated values, such as the
quantities of fuels used by the vehicle fleet. The arrays in the transportation sector consist of
vehicle categories (light-duty vehicles (LDVs), heavy-duty vehicles (HDVs), aircraft, rail, and ships),
cargo types (passengers or freight), and fuel types (petroleum gasoline, petroleum diesel,
electricity, etc.). The model generally will perform a separate set of calculations, based on a
separate set of input data, for every combination of array elements. For example, the model will
calculate different fuel economies for passenger HDVs, freight HDVs, passenger aircraft, freight
aircraft, and so fort!
In Vensim, a single dimension of an array is called a “subscript,” an array variable is called a
“subscripted variable,” and the possible values an array dimension may take are called “subscript
elements.” For example, the variable called “Fleet Aggregate Fuel Use[vehicle type, cargo type]” is a
subscripted variable, “vehicle type” and “cargo type” are each subscripts, and the “vehicle type”
subscript has the elements “LDVs,” “HDVs,” “aircraft,” “rail,” and “ships.” Almost every variable in
the Policy Solutions model is subscripted.
The model has four main sectors, plus a few supporting modules and sheets handling other
functions (Figure 1).
i Occasionally, a policy or other structural element of the model will cause a quantity to be shifted from one
combination of array index values to another. For example, the vehicle electrification policy shifts fuel
demand from non-electricity transportation fuels to electricity (with an efficiency adjustment).
1p iE 7 i”
| ccs LH Industry | Buildings Transportation
l |
t
Electricity District Heat
Cash Flow Pollutants
——> Fuel Use ———> Heat Use ——> Cash Flow
Pollutant Emissions
—— R80 Influence
rbon, i
(and associated fuel use)
————> Electricity Use
Figure 1: Diagram of the Policy Solutions model structure
The model's calculation logic begins with the Fuels sheet, where basic properties of all fuels are set
and policies that affect the price of fuels are applied. Information about the fuels is used in the
three “demand sectors”: transportation, buildings, and industry. These sectors calculate their own
emissions from direct fuel use- e.g. fossil fuels burned in vehicles, buildings, and industrial facilities.
These sectors also specify a quantity of electricity or heat (energy carriers supplied by other parts
of the model) required in each year. The electricity sector and district heat module consume fuel to
supply the energy needs of the three demand sectors. All four sectors and the district heat module
produce emissions of each pollutant, which are summed at the end. The same is true for cash flow
impacts, which are calculated separately for particular actors (government, industry, consumers,
and several specific industries). Calculation of changes in spending (for example, on capital
equipment, fuel, and labor), as well as monetized social benefits from avoided public health impacts
and climate damages, are also carried out at this stage.
There are two model components that affect the operation of various sectors. A set of R&D levers
allows the user to specify improvements in fuel economy and decreases in capital cost for
technologies in each of the four sectors and in the carbon capture and sequestration (CCS) module.
The CCS module alters the Industry and Electricity sectors by reducing their CO2 emissions
(representing sequestration), increasing their fuel usage (to power the energy-intensive CCS
process), and affecting their cash flows.
Lastly, there are a number of sections that are not part of the model's calculation flow but serve
other purposes. The “Policy Control Center” and the “R&D Control Center” are pages where the
user can conveniently view and set all of the policy levers. The “Output Variables and Graphs” page
provides certain outputs of interest, converted to more commonly-used units (for example,
converting BTUs of natural gas to trillion cubic feet of natural gas). A “Debugging Assistance” page
provides the means to easily check certain totals that should sum to zero in the absence of bugs.
Available Policies
The policies that the model is able to simulate are listed below. To provide a thorough description
of each of these policies is beyond the scope of this paper, but a very short definition (emphasizing
the way the policy is implemented in the Policy Solutions model) is provided below each policy.
Electricity Sector
1. Renewable Portfolio Standard
This policy requires that a percentage of potential electricity generation come from non-hydro
renewables (wind, solar, and biomass).
2. Additional Growth of Demand Response
This policy increases the capacity for temporally relocating electricity demand, represented
here as a reduction in peak demand and an increase in grid flexibility, without affecting total
demand.
3. Subsidy for Electricity Production
The government pays money to producers of electricity per quantity of electricity generated
and dispatched to the grid. (Set separately for each electricity source.)
4. Early Retirement of Generation Capacity
An amount of electricity generating capacity retires each year in excess of the amount that
retires due to the completion of that capacity’s natural lifetime. (Set separately for each
electricity source.)
5. Lifetime Extension
This policy increases the natural lifetime of electricity generating capacity by a number of
years, thereby reducing retirements during the model run. (Set separately for each electricity
source.)
6. Mandated Capacity Construction Policy (a schedule can be defined by the user)
This policy causes specific quantities of generation capacity to be built in specific years.
7. Additional Growth of Battery Electricity Storage
This policy increases the amount of chemical battery electricity storage available, providing
flexibility that enables more variable renewables to be used on the grid.
8. Use Least-Cost Dispatch (rather than contract-based dispatch, as is done in China today)
Electricity is dispatched from sources in order from least to greatest marginal cost, rather
than guaranteeing certain plants a number of hours they may run to recover their costs.
Industry Sector
9. Reduction in Industrial Production
This policy represents a gradual shift of China’s economy away from manufacturing and
toward services, as well as other targeted measures, such as improving product quality (so
that products, particularly building materials, do not need to be replaced so often) and
shutting down excess industrial capacity that is run despite insufficient demand for the
products. (Set separately for each industry.)
Policies to Reduce Process Emissions
10. Reduction of Vented Non-Methane Byproduct GHGs
This policy requires improvements in production processes or final products that reduce the
release of non-methane, non-CO2 GHGs, such as hydrofluorocarbons (HFCs), to the atmosphere.
11. Methane Destruction (flaring)
This policy requires methane that is currently being vented to instead burned before venting,
converting it mostly to CO2 without adding economic value.
12. Worker Training
Workers are trained to use more efficient processes or to better maintain equipment, which
can reduce process emissions in some cases.
13. Cement Clinker Substitution
Clinker, the main component in cement, is made by breaking down limestone, which releases
large amounts of CO2. This policy requires other materials to be substituted for some of the
clinker, reducing the amount of limestone that must be broken down.
14. Methane Capture
This policy requires methane that is currently being vented or leaked to the atmosphere to
instead be captured. It will ultimately be burned, offsetting the need to burn other methane.
Policies to Reduce Fuel Consumption
15. Early Retirement of Inefficient Facilities
The least efficient industrial facilities of each type are retired and replaced with modern,
highly efficient facilities, with equivalent production capacity.
16. Improved Installation and System Integration
Sometimes efficiency losses are not internal to industrial components like motors or pumps,
but arise because of poor facility design or poor integration of various components. This
policy represents promotion of principles for holistic design, pipe layout, etc. that reduce fuel
use.
17. Waste Heat Recovery and Combined Heat and Power (CHP)
Many industrial facilities generate heat, which is lost to the atmosphere. CHP allows some of
the heat to be used to do useful work, such as creating hot steam to warm a building or turn a
turbine.
18. Replacement of Coal with Other Fuels
This policy requires industrial facilities to purchase new equipment or retool existing coal-
burning equipment to use natural gas or electricity.
19. Industrial Equipment Energy Efficiency Standards
This policy requires industrial equipment to reduce energy use by a percentage relative to the
Reference case. (Set separately for each industry.)
Transportation Sector
20. Fuel Economy Standards
This policy requires new vehicles to reduce their fuel consumption per unit distance that
passengers or tons of freight are transported by a percentage relative to the Reference case.
(Set separately for each vehicle type.)
21. Feebate (for LDVs)
This policy imposes a fee on the sale of inefficient LDVs rebated to buyers of efficient LDVs.
22. Transportation Demand Management, or TDM
This is a package of urban design and pricing policies designed to reduce motor vehicle use,
such as improvements to public transit, construction of walking and biking paths, zoning for
high density along transit corridors, congestion pricing, and parking fees.
23. Vehicle Electrification
This policy causes a percentage of the fleet of specified types of vehicles to be powered by
electricity. (Set separately for each applicable vehicle type and cargo type.)
Buildings Sector
24. Rebate Program for Efficient Building Components
This policy causes utilities to pay a rebate to consumers who buy particularly efficient models
of particular building components. $50-$100 for a clothes washer and $25-$50 for a
dishwasher or refrigerator are typical values. (Set separately for each applicable building
type and component type.)
10
2
2
2
2
2
wa
mS
N
~
so
. Energy Efficiency Standards for Building Components
This policy requires new building components to reduce their energy need (while providing the
same level of services) by a percentage relative to the Reference case. (Set separately for each
component type.)
Improved Appliance Labeling
Improved labels disclose energy use, causing consumers to buy more efficient models and
manufacturers to opt to produce more efficient models.
. Improved Contractor Education and Training (for HVAC and envelope installation)
Improved training allows contractors to construct buildings and install building systems (such
as insulation or low-emissivity windows) with greater skill, preventing thermal leaks and
improving performance.
Building Component Electrification
This policy causes new electricity-using building components to be purchased in lieu of a
percentage of new building components that use a different fuel in the Reference case.
. Accelerated Retrofitting
This policy causes a percentage of buildii 1p s in existing b to be replaced
each year by new components, on top of lifetime-based retirement and replacement. (Set
separately for each component type.)
sa
Cross-Sector
30. Additional Fuel Taxes
3
3
3
B
N
wo
This policy increases the price of a fuel by a specified percentage, with tax revenues going to
the government. (Set separately for each applicable sector and fuel type combination,
including electricity and heat.)
. Carbon Tax
This policy increases the price of fuels according to their carbon content, and it increases the
price of equipment according to its embedded carbon content (carbon that was released in the
course of manufacturing and shipping of the item prior to purchase). (Set separately for each
sector.)
. Phase-Out of Reference Case Subsidies
This policy removes fuel subsidies that exist in the Reference case, including indirect subsidies,
such as those that reduce the cost of drilling for oil or gas.
. Additional Growth of Carbon Capture and Storage (CCS)
This policy increases the amount of CCS used by the electricity supply and industry sectors,
thereby increasing their fuel use and reducing their COz emissions.
11:
34. Use Market-Based Electricity Prices (rather than government-set prices)
Electricity prices are allowed to vary from year to year based on the policy-driven change in
costs for electricity suppliers.
35. Obtaining a Greater Fraction of District Heat from CHP Plants
This policy increases the fraction of heat, an energy carrier like electricity in the model, that is
generated from waste heat or CHP plants and therefore does not require fuel to be burned for
the purpose of generating the heat (as the plant is run to provide electricity in any event).
In addition to the 35 policies listed above (over 200 policies listed above if each subscripted setting
is counted as its own policy), there are 43 R&D policy levers that cause reductions in fuel use or
capital costs for various technologies.
Input Data
The model has significant input data requirements, necessitating the use of a variety of data
sources. Whenever they are available, the model uses data provided by NCSC and ERI. These often
include quantities of specific things, such as the number of miles that passengers are traveling via
different vehicle types or the quantity of fuel used by different industries. Future year projections
come from NCSC and ERI’s other models, such as those based on the Stockholm Environmental
Institute’s “Long range Energy Alternatives Planning System” (LEAP) (Heaps, 2012) and the
International Energy Agency’s “The Integrated MARKAL-EFOM System” (TIMES) (International
Energy Agency Energy Technology Systems Analysis Program, 2015).
When data are not available from NCSC or ERI, the model uses published estimates specific to China
from reputable sources, such as the International Energy Agency (IEA), the U.S. Environmental
Protection Agency, and Lawrence Berkeley National Laboratory's China Energy Group. When no
data specific to China is available at all, the model uses United States data to represent China. This
is most common for coefficients that relate certain (less commonly-studied) policies to their real-
world responses, such as the Percentage Efficiency Improvement due to Contractor Education and
Training (for the installation of heating, ventilation, and air conditioning (HVAC) systems and
building envelope components).
Model Limitations
One model limitation arises because of its reliance on various scientific studies and modeling
results to establish the effects of policies on physical quantities and costs. The studies typically
investigated these relationships under a particular set of real-world conditions. These conditions
cannot reflect all possible sets of policy settings a user might select. Therefore, the relationships
between policies and the quantities they affect might be different in different scenarios. This is not
captured in the Policy Solutions model. Generally, the model’s Reference case is likely to be closest
to the conditions under which the various policies were studied by the creators of the input data.
Therefore, the uncertainty of policy effects is likely smallest when policy levers are set at low
12
values, and uncertainty increases as the policy package includes a greater number of policies and
the settings of those policies become more extreme.
Another limitation of the model is the difficulty of characterizing uncertainty numerically. Almost
all of the input data lacked numerical uncertainty bounds. Even if such bounds had been available,
it would have been difficult to carry them through the model to establish uncertainty bounds on the
final result. As a replacement, the Policy Solutions model supports Monte Carlo analysis, which can
highlight the sensitivity of the model results to changes in any particular input or set of inputs. A
user who lacks confidence in a particular value may run a Monte Carlo simulation, varying the
suspect value within the range that he/she believes is reasonable, to obtain a probability
distribution for any output.
The model generally contains policy levers that imply specific actions (e.g. setting a renewable
portfolio standard, retiring industrial facilities early, etc.) rather than setting targets to be met via
unknown actions (e.g. defining a cap on carbon emissions, a total allowable quantity of energy use
by industry, etc.). The model is designed to predict the outcomes of specific combinations of policy
actions, not to seek an “optimal” set of policy actions to meet a specific target within Vensim.
However, using the Python script developed for use with the model, it is possible to search large
policy design spaces for combinations of settings that optimize particular outputs. For example, if a
user has a maximum allowable carbon emissions in mind, he/she can perform thousands of runs of
the model while varying policies of interest, discard all of the results with carbon emissions in
excess of the cap, and sort the remaining scenarios by another metric of interest (such as change in
capital and fuel expenditures).
Policy Scenarios
While a strength of the Policy Solutions model is the ability to simulate and compare many
thousands of policy packages efficiently, it was necessary to construct a small number of specific
scenarios that could be presented to senior Chinese policymakers. With NCSC and ERI, we
developed three policy packages, as well as two scenarios that represent upper and lower bounds
on emissions:
e A Reference scenario (RS) represents the future if no additional emissions-reducing
policies are enacted. This scenario provides the upper bound on emissions for this study.
e ALow Carbon scenario (LC) was designed by NCSC and ERI to be politically feasible and to
complement their own modeling work on achievable emissions reductions. This scenario
was developed by considering only energy-related emissions (that is, excluding process
emissions from the Industry sector), and it aims to achieve national targets in addition to
emissions reduction, such as greatly increasing the share of natural gas in China’s energy
mix.
13
e An Enhanced Low Carbon scenario (ELC) is similar to the low carbon scenario above, but
with stronger policy settings that achieve peak CO2e emissions in an earlier year.
e AnEnergy h ‘ion LLC ded scenario (EI) was designed with the goals of
reducing CO2e emissions and limiting the number and strength of different policies that
must be enacted. This scenario uses only the ten most effective policies, at settings no
stronger than international best practice, to achieve great emissions reductions while
reducing monetary expenditures in 2030.
e ACO2e-Minimizing scenario (COZeMin) sets each policy to a setting that, in combination
with all of the other policies, minimizes economy-wide CO2e emissions in the last year of
the model run.
The minimum-emissions package was identified by searching through combinations of policy
settings using the Python script. An exhaustive search of the policy design space would be
impossible. (Testing just 3 settings for each of 35 policies would be 3°35 model runs, which at the
rate of 10 runs per second, would take 159 billion years.) However, we were able to optimize
policies in logical chunks, noting which policies were only conditionally effective, and then perform
a second optimization phase in which we freeze the clearly helpful or not-helpful policies at their
final values and vary only the conditionally effective policies.» Finally, we manually test every
policy in proximity to this “tentative” best package, as a double-check in case some policies that are
helpful or harmful in proximity to this package exhibited the opposite behavior in the first-pass
optimization phase. While this procedure does not provide a guarantee that we have found the one
optimal policy package, it is likely close enough to optimal to be within the model’s margin of error.
" Each policy’s maximum allowable setting was bounded by international best practice. Without such
bounds, the concept of a CO2e-minimizing scenario is it as it would be possible to strengthen
policy settings without limit.
’ For example, we might exhaustively search a space in which we test three settings of eight policies (3“8 or
6561 model runs, or about 11 minutes at 10 runs per second). We find that some of these eight policies are
always set to a particular setting in the best-performing runs, and it is likely that the overall optimal package
will include these policies at these settings. We similarly find that some of these eight policies are always
disabled in the best-performing runs, and it is likely they will be similarly disabled in the optimal package.
One or two policies might vary amongst the top-performing runs. We call these “conditionally effective”
policies: whether these policies are worthwhile depends on the settings of other policies around them. Once
we've established the final values for the majority of all policies, we do a run in which we exhaustively test the
settings of the conditionally-effective policies, while the other policies are held constant at their final values.
14
Results and Discussion
Greenhouse Gas Emissions
Figure 2 shows greenhouse gas emissions by year for all of the tested scenarios. The LC and ELC
scenarios were designed by NCSC and ERI only considering energy-related CO emissions (e.g. from
fuel combustion) and did not make use of policies to lower industrial sector process emissions, so
they are graphed in terms of non-process COz. In contrast, the El scenario and the CO2eMin
scenarios were designed to minimize overall COze and to include industrial sector process
emissions. The reference scenario is shown on both graphs.
All of the emissions curves show a visible bend at the year 2020. This occurs because input data for
most variables provided by NCSC and ERI came only in decadal timesteps (2010, 2020, and 2030
values), while the Policy Solutions model uses an annual timestep. Linear interpolation was used to
obtain input data for these variables for the years 2013-2019 and 2021-2029. Since many trends
change between 2010-2020 and 2020-2030, sharp bends are evident. This is an artifact of the
input data format and is not meaningful. We chose not to smooth the curves so as to provide an
accurate view of our outputs, but it may be advantageous to remember that in the real world, the
bend around 2020 would instead be a smoother curve.
CO2 Emissions Excluding Process
rae e CO2e Emissions
Emissions
Million Metric Tons (MMT)
38
88s
ion Metric Tons (MMT)
8
8000
6000 6000
4000 4000
2000
9 0
2D MD od oD oO fh
SYS? SY SY SY SP SM SY PP SN PO Dr PO Dd
PP Pe PH LK HP HK HP F wi Si sy gS Se MS Sr Sr os
—— Reference Scenario (RS) ——Low Carbon Scenario (LC)
~~~ Enhanced Low Carbon Scenario (ELC) ——C02e Minimizing Scenario (CO2eMin)
Energy Innovation Recommended Policy Package (EI)
Figure 2: CO2 and C02e Emissions by Scenario
The COze-minimizing scenario uses nearly all of the 35 policy levers to achieve its deep reductions.
The Energy Innvoation recommended scenario uses only 10 policy levers and is able to achieve
77% of the COzeMin scenario’s reduction in annual CO2ze emissions in 2030. It also achieves this at
15
lower cost (as measured by expenditures on capital equipment, fuel, and labor), because many of
the policies it does not use are relatively expensive per ton COze abated.
Contributions of Specific Policies to Emissions Reductions
The model only reports the combined effects of a package of policies, to capture their interactions.
However, it can be of utility to policymakers to understand the relative contributions of different
policies to emissions reduction. We have developed two methodologies to estimate policy
contributions to emissions reduction; both can be executed in an automated manner via one of the
Python script support tools. The first procedure enables the policies of a given package one-at-a-
time, each time performing a model run and recording the emissions reduction due to that
component policy. The sum of the policies’ individual effects can then be calculated, and each
policy accounts for a particular percentage of that total. We assume the same percentage holds true
when testing the policies in combination.“ This methodology is likely to be of greatest interest to
policymakers who anticipate being able to enact a small handful of energy policies (perhaps due to
limits on time or political capital), so the policies’ individual effects in proximity to the Reference
scenario are of greater relevance than their individual effects in the context of an integrated
package of many policies.
The second procedure starts with all of the policies of a given package enabled, then disables
policies one-at-a-time and records the resulting increases in emissions. From here, the procedure is
similar to that used above. We total these increases, determine the percentage each policy
contributed to the sum of the individual increases, and assume that this is the percentage that the
policy contributes to the abatement achieved by the package as a whole. This methodology is likely
to be most useful for policymakers who anticipate being able to enact the majority of an integrated
policy package, and so the policies’ effects in the context of that package are more relevant than
their effects in proximity to the Reference scenario.
Figure 3 shows the contributions of the ten policies that compose the EI policy package to the
overall reductions achieved by the package. This figure uses the first of the two procedures
detailed above.
“ For example, suppose a particular policy package reduces emissions in 2030 by 80 MMT, and the sum of the
policies’ individual effects is to reduce emissions by 100 MMT. Ifa policy accounts for 10% of the 100 MMT
reduction based on individual testing, we assume it also accounts for 10% of the reduction achieved by the
package when interactions are accounted for (that is, it is responsible for an 8 MMT reduction).
16
China's CO2e Emissions
= Industrial Product Demand Reduction
Industrial Energy Efficiency Standards
Reduced Venting of Non-Methane Gasses
Renewable Portfolio Standard
Feed-In Tariff for Nan-Fassil Electricity
Early Retirement of Coal Power Plants
Carbon Tax
Building Codes and Appliance Efficiency Standards
Building Retrofitting,
Fuel Economy Standards for LOVs and HDVs
Figure 3: Individual policy contributions to
r ded policy p
from the Energy Innovation LLC
Many policies’ results are self-explanatory, but a few policies deserve comment. Industrial product
demand reduction is the single strongest policy, in part because it reduces both fuel use and
process emissions from Industry (the highest-emitting sector in China). Note that insofar as this
policy represents a shift of the Chinese economy from manufacturing to services, it might slightly
increase energy demand in the buildings and appliances sector, and it might slightly decrease
energy demand in the transportation sector. Neither of these secondary effects from the rise of the
service sector is captured in the model.
“Building Codes and Appliance Efficiency Standards” as well as “Fuel Economy Standards for LDVs
and HDVs’ are phased in linearly throughout the model run and only apply to newly-sold building
components or vehicles, so only the items sold in 2030 (a small fraction of the total fleet) comply
with those standards at full stringency. For that reason, the full abatement potential of these
policies is not realized by 2030, making them look less effective than they really are. Figure 4
shows the COze abatement that is achieved by vehicle fuel economy standards on a timeframe that
extends to 2050 rather than 2030. The policy reaches full strength in 2030 and is held constant
thereafter. In 2030, the policy causes less than half of the annual emissions abatement that it will
ultimately achieve, if time is provided for vehicle fleet turnover. The same is true of the building
ae
and appliance efficiency standards policy.
600
orer2ee
ao
_ 500 o
Ee ae
2 ?
= 400 ot
= ?
© ?
2 300 ?
@
=
< 200
Ss
=
100
oO
ALHOHVASHAMTHORAASHAMTH ER BOAAUAMT HOR OAS
SSAA DA AR ANANA NAAN NSOMHOHHHNOTTITSTSISIH
EB 8-8-8 8-8-8 8-0-8 8- 8-8-0 8-8-8 -8-8-8-8-8-5-5-3-0-3- 5-1-1
NANA ANAANANAAN ANNAN ANANANANANANANANAAANANT
Figure 4: COze abatement from vehicle fuel dards. The dards phase in
linearly through 2030 and are held constant thereafter.
Economic Effects of Policy Packages
The Policy Solutions model calculates the change in the amount of money paid by several economic
actors (government, industry, consumers, and several specially-broken-out industries) to each one
of these actors. The sum of all changes in spending and receipts always adds to zero, because every
time money is spent, someone else receives that money. It’s important to note that when seeking to
find the “cost” of a policy package, there are several reasonable, potential cost metrics that can be
defined by summing various combinations of these cash flows. For example, one metric reported
by the model is the total change in outlays, which is the sum of all changes in spending (i.e.
disregarding changes in receipts). However, in this paper, we will use a different metric: the change
in spending on capital equipment, fuel, and labor. This differs from “total change in outlays” in that
it excludes certain cash flows, like payments of subsidies, which are not concrete costs in the same
sense as buying equipment or fuel. Figure 5 shows the change in capital, fuel, and labor expenses
for each scenario in each year of the model run.
18
$250
$200
os Pal
Fa an. ——ELC
$100
=———CO2eMin
ee =
$50
=—— —#
$0
v
-$50
Change in Expenses (billion USD)
-$100
Figure 5: Change in capital, fuel, and labor expenses by scenario
Generally, the scenarios have a greater slope in the early years of the model run, and the slope
begins flattening out or even becomes negative in the early 2020s. This happens because
investments in improved equipment generally are made at a similar rate in all years of the model
run, but fuel savings grow larger over time, as more and more equipment has been replaced.
The policies that have the largest impact on expenses tend to be taxes on commonly-used fuel types
and the carbon tax. This is because the taxes’ effect on reducing fuel consumption (saving money)
is outweighed by increasing the price of the fuel that is still consumed. The LC and ELC cases use
both a carbon tax and a tax on petroleum fuels, which together account for most of the cost. As an
illustration, disabling the carbon tax and the petroleum fuel tax in the ELC package changes the cost
of the package in 2030 from positive $222 billion (an expense) to negative $39 billion (a savings).
The EI package is the only one that reduces costs in 2030. This is in large part because the EI
package forgoes fuel taxes: most reductions in the EI package are driven by standards. However,
the EI package does include a substantial carbon tax (reaching $45/ton COze in 2030), and it is this
policy that accounts for most of the EI package’s costs.
These model results make taxes appear to be expensive, but in the real world, it is possible to
achieve the emissions reduction from fuel taxes or a carbon tax while offsetting the cost with an
equivalent reduction in other taxes, such as income or payroll taxes. This option is beyond the
scope of the Policy Solutions model, but it is worth highlighting, because a revenue-neutral carbon
tax is a policy that is favored by some policymakers (Rosner, 2014). One finding of this work is that
while such a tax would be a powerful way to drive down emissions, it functions best as one element
19
ina package of policies, which collectively can achieve greater emissions reductions and save more
money than the carbon tax alone.
It is possible to attribute costs to particular policies within a package in the same manner as it is
possible to attribute emissions reductions to particular policies (described above). Putting these
things together allows us to construct a policy cost curve, with abatement potential on the X-axis
and cost per ton abated on the Y-axis. Figure 6 is such a curve for the COzeMin scenario, which we
show here because it includes more policies than the other scenarios. Note that a policy cost curve
is simply another way to visualize the component policies within a package; the curve changes
depending on which policies are enabled and what specific settings they are given. There does not
exist a single “correct” curve or policy cost ordering.
Fuel Taxon
lectrcity
Industrial
uel
switching
5,000 Renewable Methane capture \
8 Porto stndaré "and bestton
9 4,000 Building Feed-in Tariff for
oo a a aetoiteg Non ross aon capture
” meustty industn \ lect and Sequestration \
3 Enetey waste neat \Sulting Demand ree cement cher | :
2 2,000 slY ecency \ "codes Response \
2 Dein Sandras RECO f
9,000 MMT/year
carbon Tax
non arly Retirement
ppliance” ints Product N
2,000 sootane noe Demand Plants 1 Cal Power
“ arly fomponent improved Reduction sae
3,000 Standardetiement and Appliance contactor
is sina abte ton Products
Feebate for
[i otal change in Operating Expenditures
i Total change in Capital Investment
Figure 6: Abatement potential and cost per ton abated by policy for the COzeMin scenario
Calculation of Monetized Social Benefits
Change in spending is not the only relevant metric for policymakers. The model also estimates the
monetized values of avoided public health damages and climate impacts. Public health damages are
based solely on mortality (not morbidity). Monetization of these damages is based on figures from
the U.S. EPA (U.S. Environmental Protection Agency, 2015), adjusted upward to reflect China’s
larger population and therefore likely larger human exposure per ton of pollutant emitted, then
adjusted downward to account for differences in median income (as a proxy for a direct adjustment
20
based on Value of a Statistical Life (VSL)vi, because we felt that available VSL figures for China were
unreasonably low). Climate impacts are based on the United States’ social cost of carbon figures,
since China does not have a comparable statistic.
Figure 7 shows the monetized social benefits for each scenario. Avoided climate damages only
account for 9-16% of the total value of the benefits (varying by scenario and by year). The vast
majority of the benefits come from avoided mortality.
$3,500
$3,000
$2,500 a
$2,000 ZO ZO —c02emin
Pa we aia
éxo00 4 —— —Ic
$500
me
9°
Monetized Social Benefits (billion USD)
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
Figure 7: Monetized social benefits by scenario
“i VSL is a means of valuing risk reductions or deaths in monetary terms. It can be estimated from surveys
that ask people how much they would be willing to pay to reduce their risk of dying in the next year. For a
discussion of VSL as used by the U.S. EPA, see the EPA's “Frequently Asked Questions on Mortality Risk
Valuation” at http: ite.epa.gov/EE%5Cepa%5Ceed.nsf/wek MortalityRiskValuation.html
21
The values of social benefits tend to greatly outweigh the direct costs of policy packages. For
example, in 2030, the LC scenario costs $30 billion but generates $1200 billion in social benefits,
and the COzeMin scenario costs $137 billion but generates $3293 billion in social benefits.
However, in one sense, the figures are not comparable, because the costs refer to actual payments
that are made, while nobody pays money to account for the value of social benefits. Instead, people
simply live who otherwise would have died. The main factor that causes the social benefits to be so
large is the size of the monetary value placed on human life for policy analysis purposes.
Effects of Policies on Electricity Generation
We also examine the quantity of electricity provided by each modeled energy source in each
scenario (Figure 8). In the Reference scenario, most new electricity output is from coal-fired
generation. Solar, wind, and biomass also grow and account for 13.7% of output in 2030. In the
Low Carbon scenario, there is very little growth of coal after 2020. Relative to the Reference
scenario, there is significantly more natural gas output, as well as slightly more nuclear, wind, solar
and biomass. Overall output in the LC scenario is also lower than in the Reference scenario by 750
TWh. The Enhanced Low Carbon scenario is the first scenario to exhibit a coal peak, which occurs
around 2020. The ELC scenario includes less coal and natural gas output than the LC scenario, and
more nuclear, hydro, wind, and solar. The EI scenario is the first one to aggressively drive down
coal, slowly through about 2018 and then faster through 2030. Demand is lower in the EI scenario
than the Reference scenario by 2,491 TWh in 2030. Wind, solar, and biomass account for 35.6% of
output in 2030. The CO2eMin scenario is similar in many respects to the EI scenario, but coal is
reduced faster, and there is more growth of hydro and nuclear, two of the more expensive, zero-
carbon power plant types.
22
10000 : 10000
Reference Scenario _ Low Carbon Scenario
i ==
A 5 eee |
8 7
2 2 sooo +
6 6
& =
3 S sc00
H E 20m
a a
1000
8 5g
RERRRE
10000 - . . 10000 F
9000 Low Carbon Scenario <b Energy Innov. LLC Scenario
ie
E = 70
§ 5
S ‘S 6000 +
i 8
@ © sooo |
o 5
° 2 4000
z z
2 2 3000
3 § 20m
a a
3000
°
SERBS
RRRRR
9000 | CO2eMin Scenario
=
| » solar
5 A;
8 = wind
2
6 hydro
zB
= = nuclear
Ky
a
@ natural gas
8 BH coal
Figure 8: Electricity generation (TWh) by energy source in each scenario
Additional Outputs of Interest to Policymakers
The results in this paper are a sampling of some particularly relevant and important results; a
complete overview of all interesting or useful results would be impossible, due to the diversity of
different model outputs that are available, as well as the fact that we only review five specific policy
23
packages in this paper. It is worth briefly mentioning other model results that might be of use to
policymakers or researchers but are not shown numerically here:
e Policymakers might consider outputs other than COze and monetary outlays to be goal or
decision metrics. For instance, if energy security is a primary concern, policy packages can
be designed with an eye toward reducing economy-wide consumption of petroleum fuels
and natural gas, which are mostly imported in China.
e¢ We do not utilize enhanced research and development (R&D) in our policy scenarios, but
the model is equipped with 43 levers that allow R&D-based cost reductions and fuel
efficiency improvements beyond the Reference case for 26 different classes of technologies.
This can enable a user or policymaker to explore which policies would be effective in the
context of different levels of R&D success in different areas. For example, perhaps a solar
subsidy becomes less effective if there is more R&D advancement of the coal and natural gas
technologies, but the Renewable Portfolio Standard policy remains effective irrespective of
the level of coal and natural gas R&D advancement. This would imply that the RPS policy is
more robust against different fossil fuel R&D outcomes than the subsidy policy. It is
important to consider which policies work in a range of R&D environments, because the
future of scientific advancement is not knowable with precision.
e While we used the Python script to identify an approximate minimal emissions policy
package, it can be used to seek a policy package to suit any number of conditions. For
example, one could find the policy package that minimizes monetary outlays while keeping
emissions below a certain, fixed level, thus determining a least-cost method of complying
with a carbon cap.
e Sometimes, itis interesting to see not just what is effective, but what is ineffective, at least
through the 2030 timeframe. For example, the policy that allows for the achievement of
additional CCS potential does not tend to have much effect, because CCS is such anew
technology that its maximum potential by the year 2030 is not very high. Some policies are
conditionally effective, depending on their setting and sometimes the settings of other
policies. One example is a subsidy for electricity production from a particular energy
source: there is a price range where the subsidy alters the model’s decisions about what to
build, but above and below that range, the subsidy makes no difference (because the
subsidized energy source is either far too expensive to build or far too cheap not to build).
Policies that increase the amount of flexibility on the electric grid are effective when the RPS
is high (because they help bring more renewables onto the system), but they have no effect
otherwise (because there is already sufficient flexibility to allow for the integration of as
many renewables as the model wishes to build).
These examples help to illustrate that the model’s greatest utility isn’t in providing the specific
results for the four policy packages discussed in this paper, but to rapidly answer a tremendous
range of questions that policymakers might have about their options for affecting the energy
system.
24
Web Application Model Interface
In order to provide access to key model results and improve the usability of the model for non-technical
users, Energy Innovation has produced a web application that provides a means of interacting with the
model, creating policy packages, and visualizing output in a web browser. The web application was
developed by Todd Fincannon. This web application has particular value because much of the input
data provided by the Chinese government may not be publicly distributed, preventing us from releasing
a functional version of the model for China. The web application provides a means for the public to use
the China version of the model without violating the restriction on sharing of input data. (We are
publicly releasing a United States version of the Policy Solutions model.) Figure 9 is a screenshot of the
web application with several annotations in red.
+ Olct] @/O} hag Slope |O/2
{Energynnoration «| | cozeEmissons Total excusing process emissions) 7
Transporation Re
2 Baking Appin Ability to graph many different outputs
Building Component Electrification ™
Bulking nea éicency Sano 40.0%
Imoroved Labeling
ponent aes RL ~ Policies includedin the model
tesa Econ res
Eloctlcity Supply me
Ios enad econ: 10% ;
Industrial Fuel Switching _*
Cross Soctr :
Ce ae ge AativePoliy Setines
* Increased Retrofitting: 20%
+ Early Retirement of Coal Power Plants: 75000 (NW
Standard: 40.0%
y Production (Nuciar: 125 [Mh] &
i Production (yero: 125 [Wh]
ty Production (Wind): 125 [AAW]
5s
Figure 9: Scr hot of the web application interface for the Policy Solutions model (China
version)
Conclusion
With the right set of policies, China will be able to cut its emissions dramatically and cost-
effectively. A System Dynamics model provides the ideal tool to help policymakers understand the
range of policy options at their disposal and quantitatively estimate their effects. It is our hope that
25
the Policy Solutions model will assist the Chinese central government in selecting policies that will
achieve their emissions reduction targets. But more than that, we hope that seeing that large
emissions cuts are possible with reasonable policy options will inspire China to set more aggressive
targets. Strong emissions cuts will pay for themselves through fuel savings, public health benefits,
and reduced damages from climate change. Climate change may be the most serious problem we
presently face, but with dedication and smart policy choices, China can be a leader in meeting this
challenge.
Acknowledgements
This work was made possible through the contributions and advice of individuals at the following
organizations:
e China’s National Center for Climate Change Strategy and International Cooperation
e China’s Energy Research Institute
e Massachusetts Institute of Technology
e Stanford University
e Lawrence Berkeley National Laboratory's China Energy Group
e Climate Interactive
Also, we wish to thank model reviewers from Climate Interactive, Argonne National Laboratory,
Lawrence Berkeley National Laboratory, the National Renewable Energy Laboratory, and Stanford
University. Note that having served as a model reviewer does not imply endorsement of the model
or its findings.
References
Chen Y, Evenstein A, Greenstone M, Li H. 2013. Evidence on the Impact of Sustained Exposure to
Air Pollution on Life Expectancy from China’s Huai River Policy. Proceedings of the National
Academy of Sciences 110(32): 12936-12941.
Heaps CG. 2012. Long-Range Energy Alternatives Planning (LEAP) System. Stockholm
Environment Institute, Somerville, MA. Retrieved Feb 23, 2015 from
http://www.energycommunity.org.
Intergovernmental Panel on Climate Change Working Group II. 2014. Climate Change 2014:
Impacts, Adaptation, and Vulnerability, Part A: Global and Sectoral Aspects. Retrieved Feb 23, 2015,
from http://www. ipcc.ch/report/arS/wg2/.
International Energy Agency Energy Technology Systems Analysis Program. 2015. TIMES (The
Integrated MARKAL-EFOM System) Model. Retrieved May 28, 2015, from http://www.iea-
etsap.org/web/Times.asp.
26
Nakamura D, Mufson S. 2014. China, U.S. Agree to Limit Greenhouse Gases. The Washington Post.
Retrieved Feb 23, 2015 from http://www.washingtonpost.com/business/economy /china-us-
agree-to-limit-greenhouse-gases/2014/11/11/9c768504-69e6-11e4-9fb4-
a622dae742a2_story.html.
Rosner H. 2014. Q&A: The Conservative Case for a Carbon Tax. National Geographic News.
Retrieved Sept 22, 2014, from http://news.nationalgeographic.com/news/2014/09/140922-
carbon-tax-climate-change-conservatives-environment-science/.
Sterman, JD. 2000. Business Dynamics: Systems Thinking and Modeling for a Complex World.
Irwin/McGraw-Hill, Boston.
U.S. Environmental Protection Agency. 2015. Response Surface Model (RSM)-Based Benefit Per
Ton Estimates. Retrieved Feb 26, 2015, from http://www2.epa.gov/benmap/response-surface-
model-rsm-based-benefit-ton-estimates.
27
