Environmental Research Letters

LETTER • OPEN ACCESS Related content

- The role of capital costs in decarbonizing
Commitment accounting of CO2 emissions the electricity sector
Lion Hirth and Jan Christoph Steckel

- Assessing carbon lock-in

To cite this article: Steven J Davis and Robert H Socolow 2014 Environ. Res. Lett. 9 084018 Peter Erickson, Sivan Kartha, Michael
Lazarus et al.

- Climate constraints on the carbon intensity

of economic growth
Julie Rozenberg, Steven J Davis, Ulf
View the article online for updates and enhancements. Narloch et al.

Recent citations
- Molecular Modeling Analysis of CO2
Absorption by Glymes
Alberto Gutiérrez et al

- Targeted emission reductions from global

super-polluting power plant units
Dan Tong et al

- Principles to guide investment towards a

stable climate
Richard J. Millar et al

This content was downloaded from IP address 179.218.5.83 on 20/03/2018 at 14:06

 Environmental Research Letters

Environ. Res. Lett. 9 (2014) 084018 (9pp) doi:10.1088/1748-9326/9/8/084018

Commitment accounting of CO2 emissions

Steven J Davis1 and Robert H Socolow2
1
Department of Earth System Science, University of California, Irvine, Croul Hall, Irvine, CA 92697, USA
2
Department of Mechanical and Aerospace Engineering, Princeton University, Princeton, NJ 08544, USA

E-mail: [email protected] and [email protected]

Received 23 June 2014, revised 17 July 2014

Accepted for publication 30 July 2014
Published 26 August 2014

Abstract
The world not only continues to build new coal-fired power plants, but built more new coal
plants in the past decade than in any previous decade. Worldwide, an average of 89 gigawatts per
year (GW yr–1) of new coal generating capacity was added between 2010 and 2012, 23 GW yr–1
more than in the 2000–2009 time period and 56 GW yr–1 more than in the 1990–1999 time
period. Natural gas plants show a similar pattern. Assuming these plants operate for 40 years, the
fossil-fuel burning plants built in 2012 will emit approximately 19 billion tons of CO2 (Gt CO2)
over their lifetimes, versus 14 Gt CO2 actually emitted by all operating fossil fuel power plants in
2012. We find that total committed emissions related to the power sector are growing at a rate of
about 4% per year, and reached 307 (with an estimated uncertainty of 192–439) Gt CO2 in 2012.
These facts are not well known in the energy policy community, where annual emissions receive
far more attention than future emissions related to new capital investments. This paper
demonstrates the potential for ‘commitment accounting’ to inform public policy by quantifying
future emissions implied by current investments.
S Online supplementary data available from stacks.iop.org/ERL/9/084018/mmedia

Keywords: CO2 emissions, committed emissions, carbon lock-in, socioeconomic inertia, power
plants, energy infrastructure

1. Introduction transport and built infrastructures) that limits the rate at which
CO2 emissions can be reduced and climate change avoi-
Each year, governments and firms estimate and report CO2 ded [11–15].
emissions from the burning of fossil fuels, and their efforts to In 2010, Davis et al quantified an important component
slow climate change are measured against these annual of socio-economic inertia by estimating the future emissions
emissions [1, 2]. Ultimately, though, the magnitude of expected from all existing fossil fuel-burning infrastructure
warming we experience will not be determined by emissions worldwide, naming these ‘committed’ emissions [16]. That
in any one year, but by cumulative CO2 emissions [3, 4]. paper provided a single data point (commitments as of 2009),
Thus, climate scientists and energy-economic modelers have but lacked the context perhaps most important to policy-
developed hundreds of plausible scenarios of future emissions makers: how these committed emissions have changed over
[5–7] and used them to identify emissions pathways that time. Here we provide that context. We show that, despite
might achieve climate policy goals [8–10]. Such scenarios are international efforts to reduce CO2 emissions, total remaining
powerful tools for connecting emissions and warming to commitments in the global power sector have not declined in
trajectories of population growth, economic development, and a single year since 1950 and are in fact growing rapidly (by an
energy use. However, these trajectories are constrained by average of 4% per year 2000–2012).
tremendous socio-economic inertia (e.g., existing energy, The annual ‘commitment accounting’ that we demon-
strate here offers policymakers an opportunity to evaluate
historical trends and to quantify the long-term consequences
Content from this work may be used under the terms of the
of current actions in a new way. A focus on committed
Creative Commons Attribution 3.0 licence. Any further
distribution of this work must maintain attribution to the author(s) and the emissions allows decision makers and analysts to improve
title of the work, journal citation and DOI. their representation of socio-economic inertia, carbon lock-in

1748-9326/14/084018+09$33.00 1 © 2014 IOP Publishing Ltd

Environ. Res. Lett. 9 (2014) 084018 S J Davis and R H Socolow

continued to operate after five years, both committed emis-

sions and realized emissions would be increased by one unit
each year until the device was retired. Commitment
accounting assumes specific lifetimes of infrastructure,
acknowledges that these may turn out to be overestimates or
underestimates of actual lifetimes, and thus requires regular
adjustment to reflect the best current information.

2. Materials and methods

Commitment accounting requires four pieces of information

for each CO2-emitting device: (1) the year the device began
operating, (2) the expected operating lifetime of the device,
(3) the annual emissions from the device, and, (4) if the
device is no longer operating, the year the device ceased
operating.
Our analysis of power generators relies on two datasets
for this information: The Platts World Electric Power Plant
database [20] includes data about the year each generator
came online and the year it retired. Version 3 of the Carbon
Monitoring for Action (CARMA) database [21] contains
Figure 1. Schematic of committed emissions accounting. Emissions
that occur over the lifetime of a fossil-fuel-burning electric generator
information about annual power generation and CO2 emis-
are today routinely assigned to the year of emission ((a), blue sions—not for generators but for power plants, and not every
blocks). In this paper, we present an alternate accounting, where the year but only for 2004 and 2009. Approximately 99% of
same emissions are instead recorded as committed emissions in the plants in CARMA have corresponding generators in Platts.
year the generator begins operating (i.e. online year; (a), green We make the arbitrary assumption that CARMA’s
blocks). As time passes, it is possible to track the fraction of these
committed emissions that have been realized and the fraction that
emissions and energy data for 2009 (or, occasionally, 2004)
remain committed, making adjustments if necessary to reflect early are an accurate estimate throughout a plant’s lifetime (despite
or late retirement of the generator (b). We show realized and evidence to the contrary [22]). Until even more comprehen-
remaining commitments as negative values because their sum sive data sources become available, commitment accounting
exactly balances the initial commitment (b). will necessitate such assumptions.

and the effects of political delay [12, 17–19]. We introduce a 2.1. Online year (year generator began operating)
formalism to quantify future emissions associated with any
Of 95 529 fossil-fuel generators included in the Platts data-
new capital investments, and, as well, to consider early
base [20], the online year is available for 79 595, or 83%, of
retirement, retrofit, and plant life extension.
the generators. For an additional 12 275 generators, we assign
The general principle of commitment accounting is
an online year equal to the mean online year of all other
shown schematically in figure 1. Two views are contrasted for
generators for which an online year is provided in the Platts
the case where a new device, when built, is expected to run database (i) that operate in the same country, (ii) that burn the
for five years and to emit one unit of CO2 each year. Today’s same fuel, and (iii) whose generating capacity is within
carbon accounting would report annual emissions of one unit 75 MW of its capacity. Thereby, our analysis creates an
of CO2 in each of the five years of operation. Commitment online year for 91 870 fossil-fuel generators, or 96% of those
accounting instead assigns all five units to the year when the included in the Platts database. The remaining 4% are
device begins to operate (figure 1(a)). We call these antici- ‘planned’ or ‘under construction’ generators that were not yet
pated emissions ‘committed emissions’ or simply ‘commit- in operation when the Platts data were published.
ments.’ Figure 1(b) presents the same device three years after
it begins to operate and shows (below the line) the initial
2.2. Expected lifetime
commitments that have been realized as emissions and those
that remain commitments. We depict realized and remaining The most important assumption of commitment accounting is
commitments as negative numbers to reflect the fact that net the expected lifetime of new generators. The decision to retire
commitments will be zero when fully realized. a generator or power plant is seldom prompted by physical
‘Committed’ does not mean inevitable. If the device in failure of the equipment; rather, the decision is driven by
figure 1 was shut down after operating only four years, its economic considerations of operating costs, replacement
remaining commitment would go to zero, and both committed costs, and revenues [14, 22–24]. The Platts database lists
and realized emissions would be shortened to four units in all 13 026 fossil-fuel generators as ‘retired,’ about 14% of the
subsequent representations. Conversely, if the device total. Of these, the retirement year is available for only half of

2
Environ. Res. Lett. 9 (2014) 084018 S J Davis and R H Socolow

results in a retirement year later than 2012 (i.e. an obviously

incorrect estimate given that the generator in question is
known to be retired), we estimate the retirement date using the
same methodology as in section 2.1 (same country, same fuel,
similar capacity) and add the constraint of similar vintage
(defined as an online year ±5 years). For an assumed gen-
erator lifetime of 40 years, roughly two thirds of the gen-
erators for which no retirement year is stated (4256 out of the
6514 generators) came on line after 1972 and hence required
this adjustment.

2.3. Annual emissions

The CARMA database [21] contains data on electricity gen-

erated and CO2 emitted by each of 68 931 power plants for
the years 2004 and 2009. We allocate the plant-level data of
electricity generation and emissions data to individual gen-
erators in proportion to their nameplate generating capacity.
We use 2009 emissions data unless only 2004 data exist (e.g.,
where a plant retired between 2004 and 2009).
Figure 2. Historical survival of generators by fuel type. Plot shows Although annual emissions usually deteriorate over a
the age at retirement of 5841 generators retired since 1971 by
primary fuel type. Vertical dashed lines indicate the median age at generator’s lifetime because it is run less often as its operating
retirement of generators for each fuel type. Early retirement happens efficiency and economic competitiveness decline [22], we
more often for generators powered by gas or oil than for those make the simplifying assumption that the CO2 emissions
powered by coal: 22% and 26% of the gas and oil generators, produced by each generator in 2009 (or 2004 if there are no
respectively, were retired before they had run for 20 years, but only 2009 data) is a good estimate throughout its period of
6% of coal generators.
operation, and thus we underestimate early emissions from
any older generators that operated only occasionally in 2009.
To the extent that annual emissions of older generators are
them (the retirement year is known for 6447 and not known underestimated, their contributions to committed emissions
for 6514). Moreover, of the 6447 generators with known will also be underestimated. As a check of our method, we
retirement year, another 606 do not state the year they came compare a time series of annual global emissions from our
on line. Rather than use the methodology to estimate online model to a time series of emissions from power plants pro-
date described above, we simply exclude these retired gen- vided by the International Energy Agency (IEA) for the years
erators as well. The data for the remaining 5841 generators 1971–2011 [25] (figures S3). Our estimates are within 42% of
with known online and retirement dates are our primary data the IEA data over the period prior to 1975, within 29% of IEA
source for choosing a reference lifetime. data from 1975–1985, and within 9% of IEA data from
Figure 2 shows these distributions of retirement age for 1985–2011.
generators fired by natural gas, oil and coal. Their median If no emissions data are available from CARMA (e.g.,
ages at retirement are remarkably similar: 37, 35, and 32 because a plant retired before 2004), we estimate emissions
years, respectively. We explored these retirement data in and electricity generation according to the nameplate capacity
detail and found no persuasive evidence of systematic dif- of the generator (from the Platts database) and the mean
ferences or trends in relationships among fuel type, region, capacity factor (MWh electricity generated per MW capacity)
online year, generating capacity, and retirement year (figures and emissions factor (kg CO2 per MWh) of all other gen-
S1 and S2), with the exception of the secondary detail of fuel erators (i) in the same country, (ii) burning the same fuel, and
type, where natural gas- and oil-fired generators were sub- (iii) with similar generating capacity (defined as a nameplate
stantially more likely to be retired before operating 20 years capacity ±75 MW). In 32 cases (0.03%), our method yields
than coal-fired generators (figure 2). We thus chose to sim- carbon intensities that are too low to be credible for the fossil
plify our demonstration of commitment accounting by using a fuel being burned; we use lower limits of 800 kg CO2/MWh
single reference lifetime of 40 years for all generators for coal, 700 kg CO2/MWh for oil, and 490 kg CO2/MWh for
regardless of differences in fuel type, region, online year, or gas. In these cases, we assume the reported electricity pro-
generating capacity and performing a sensitivity analysis duction is correct and estimate CO2 emissions using the
using lifetimes of 20, 30, 40, 50 and 60 years. Unless global average carbon intensity for generators burning that
otherwise noted, the results reported here reflect a 40 year type of fuel. Figures S4 and S5 show that there are no obvious
generator lifetime with a range of uncertainty corresponding relationships between emissions intensity (kg CO2/MWh) and
to generator lifetimes of 30 and 50 years. online year and nameplate capacity, respectively.
When we assume a reference lifetime for a retired plant According to the CARMA database, coal plants built
whose retirement date is not in the Platts database, and this worldwide 2000–2012 on average emit 1090 kg CO2 per

3
Environ. Res. Lett. 9 (2014) 084018 S J Davis and R H Socolow

Figure 3. Commitment accounting of electric generators installed between 1950 and 2012. From the perspective of 2012, contributions to
committed emissions ((a), light green area) grew from approximately 4 Gt CO2 per year in 1960 to roughly 10 Gt CO2 per year between
1970–1995, and then to more than 14 Gt CO2 per year since 2000. Throughout this period, annual commitments have exceeded annual
emissions (blue curve in (a) shows IEA estimates of emissions from the power sector 1971–2011 [25]). Although the commitments made
30–40 years ago have largely been realized (black), a large proportion of commitments made since 1990 remain ((a), darker green area). The
geography of committed emissions (b) has shifted from the US and Europe in 1970s and 1980s to China since 1990, but coal-fired power has
consistently been the dominant source of committed emissions (c). Numbers on the plot indicate total committed emissions and those
remaining committed as of 2012 in Gt CO2.

MWh and have a capacity factor of 44%, weighted by gen-

Table 1. Remaining committed emissions (Gt CO2) from power
erating capacity. Assuming a 40 year lifetime, the average infrastructure varies under different assumptions of generator
committed emissions are 167 Mt CO2 per GW of coal-fired lifetime. ‘Adjustments’ indicate the aggregate revisions in
generating capacity. The corresponding values for natural gas commitments necessary to accommodate generators that operated
are 659 kg CO2 per MWh and 35% capacity factor, leading to fewer or more years than the assumed lifetime over the entire period
83 Mt CO2 of committed emissions per GW of gas-fired 1950–2012.
generating capacity. In other words, roughly, every 6 GW of 20 30 40 50 60
new coal-fired capacity has been committing the world to 1 Gt Country/ Year Year Year Year Year
CO2 of emissions, and so has every 12 GW of gas-fired region lifetime lifetime lifetime lifetime lifetime
capacity.
By fuel
Coal 66.2 128.3 206.1 295.7 390.3
Gas 27.0 53.0 82.3 114.2 147.5
Oil 4.9 10.2 18.2 28.1 38.7
3. Results By region
China 47.7 87.7 129.2 170.8 212.4
3.1. Committed emissions in 2012 India 7.4 14.5 23.1 32.1 41.3
US 8.3 17.5 33.8 57.4 83.8
Figure 3(a) shows commitment accounting (see figure 1) for EU28a 7.7 15.9 26.6 40.2 55.6
the global power sector, which was the source of 40% of Rest of 27.1 56.1 94.2 138.1 184.4
global fossil-fuel emissions in 2011 [25, 26]. The point of world
view is from 2012, a 40 year lifetime is assumed for all Total 98.2 191.7 306.9 438.6 577.5
generators, and all fossil fuel-fired electricity-generating units Adjustments
(‘generators’) that were built globally between 1950 and 2012 Generators 98.3 42.9 12.6 2.6 0.2
are included. Global committed emissions from these gen- retired
erators total 629 (508–761) Gt CO2 (light green area; only the late
Generators 0.6 2.5 7.1 14.6 23.4
central estimate reflecting a 40 year lifetime is shown), of
retired
which 322 Gt CO2 were realized emissions by 2012 (black early
area), and 307 (192–439) Gt CO2 were remaining commit-
a
ments as of 2012 (dark green area). EU28 region includes: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech
The error estimates in parentheses here are for assumed Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary,
Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland,
lifetimes of 30 and 50 years, presented in table 1 (along with
Portugal, Romania, Slovakia, Spain, Sweden, England & Wales, Scotland,
the results of assuming lifetimes of 20 and 60 years). Thus, and Northern Ireland.
the range of estimated remaining commitments in 2012
(192–439 Gt CO2) assigns an uncertainty of 40% to our
central estimate of 307 Gt CO2. This estimate and its uncer-
tainty are generally consistent with the remaining

4
Environ. Res. Lett. 9 (2014) 084018 S J Davis and R H Socolow

commitment from the power sector reported in Davis et al for estimates (40 year lifetime) of ROW commitments represent
2009 (a central estimate of 224 Gt CO2 and an uncertainty 17% and 53% of all remaining committed emissions in 2012
range from 127–336 GtCO2). Davis et al used somewhat from coal generators and oil and gas generators worldwide,
shorter reference lifetimes of 39, 36, and 34 years for coal, respectively (see table S1).
gas, and oil plants, respectively. Its error estimates, like those Of the 307 (192–439) Gt CO2 of remaining committed
in this paper, were based on lifetime corrections of ±10 years emissions as of 2012, figure 3(c) shows that coal-fired gen-
in each case. erators represent 206 (128–296) Gt CO2, or 67% of the total.
Embedded in the calculations of remaining commitments Not shown, China’s remaining commitments are 98% from
in figure 3(a) is the assumption that all generators older than coal (see table S1). Over half of the world’s remaining
40 years in 2012 will be shut down immediately. Table S2 committed emissions from oil- and gas-fired generators are in
shows the consequences of instead allowing all plants to run the ROW region (59 of 100 Gt CO2, 40 year lifetime, see
an additional five to ten years (to 2017 or 2022, respectively), table S1).
for all five assumed lifetimes. For example, assuming a 40 Figure 4 extends figure 3(a) into the future, showing
year lifetime, each five years that overage generators are when the remaining committed emissions in 2012 are
allowed to continue running causes the remaining global expected to occur, for various generator lifetimes.
commitment to increase by 7 Gt CO2 (table S2).
The black dashed line in figure 3(a) indicates the mag-
3.2. Trends in commitments
nitude of adjustments that were made to accommodate gen-
erators that retired either before operating 40 years or after Thus far, we have restricted the analysis to a single year (the
operating more than 40 years (see figures S1 and S2 for most recent year with available data, 2012)—effectively
further analysis of retirements). This line can barely be dis- updating the analysis Davis et al presented for 2009 [16].
tinguished from the top of the light green area because, even Much can be learned, however, from producing the corre-
in aggregate, these adjustments are small: a 7 Gt CO2 sponding analysis for earlier years and examining trends, a
reduction for early retirements and a 13 Gt CO2 addition for line of argument not investigated in Davis et al. In figure 5,
life extension. However, larger adjustments are necessary we show the results of repeating the committed emissions
when assumed reference lifetimes are substantially different analysis for each year from 1950 through 2012 and assuming
from 40 years (see ‘adjustments’ in tables 1 and S2). a 40-year generator lifetime. Figure 5 has six panels: we
As seen in figure 3(a), annual contributions to the total of present the remaining committed emissions by year, the
629 Gt CO2 of committed emissions accumulated as of 2012 change each year, and the dimensionless ratio which is the
rose from approximately 4 Gt CO2 yr–1 in 1960 to ∼9 Gt change each year divided by the actual emissions that year;
CO2 yr–1 between 1970 and 2000, before ramping steeply to we show a decomposition for all three cases both by region of
more than 14 Gt CO2 yr–1 since 2000. In comparison, the the world and by fuel.
International Energy Agency (IEA) estimates for annual The first pair of panels, figures 5(a) and (b), plot total
emissions from the global power sector are consistently less remaining commitments for each year 1950–2012. The 2012
than these annual commitments; the IEA estimates rise nearly endpoint of these plots is the same value, 307 Gt CO2, as the
linearly from 3.7 Gt CO2 yr–1 in 1971 to 13.1 Gt CO2 yr–1 in total remaining emissions for 2012 shown in figure 3. But
2011 [25] (figure 3(a), blue curve) [25]. whereas figure 3 decomposes the committed emissions by the
Figure 3(b) decomposes the global totals from figure 3(a) year in which the generators which contribute to these com-
into world regions, and figure 3(c) does the same for fuels. mitments came on line, figure 5 suppresses the year in which
Among the stories figure 3(b) tells are the expansion of power any particular generator came on line, and instead the time
infrastructure in the US and the EU during the 1960s and axis shows the year for which the total remaining committed
1970s, the ‘race to gas’ in the US between 2000 and 2005, emissions are calculated.
and the rapid construction of generators in China beginning in There has been a monotonic increase in global remaining
the mid-1990s [27, 28]. China’s infrastructure represents 129 commitments over time, from near zero in 1950 to 107 Gt
(89–171) Gt CO2, or 42% (46%–39%) of the total remaining CO2 in 1980 and 307 Gt CO2 in 2012 (data on generators
commitments in 2012 (figure 3(b)). Less anticipated, India’s built prior to 1950 are incomplete). Regional disaggregation
remaining commitments from power-plant emissions in 2012 (figure 5(a)) reveals that prior to about 1985, most remaining
are approximately equal to those of the EU and the US: all commitments were associated with generators operating in the
three are between 23 and 34 Gt CO2 (or between 15–18 and US and Europe. But in the years since, remaining commit-
32–57 Gt CO2 assuming generator lifetimes of 30 and 50 ments increasingly reflect new generators in China, India and
years, respectively). Figure 3(b) also reminds us that a focus the rest of the world, with the US and EU28 accounting for
on the foursome of China, India, the EU and the US can be less than 20% of total remaining commitments in 2012.
overdone: remaining committed emissions from the rest of the Disaggregation by fuel (figure 5(b)) shows that remaining
world (ROW) are 31% (29%–31%) of the global total in commitments have always been dominated by coal-fired
2012. Ten countries represent 56% of remaining committed generators, but the contribution of natural gas-fired generators
emissions from the ROW region in 2012: Japan, South Korea, increased nearly five-fold between 1980 and 2012: from 17 Gt
Russia, Saudi Arabia, Indonesia, Australia, Taiwan, Iran, CO2 in 1980 to 82 Gt CO2 (27% of all remaining commit-
Mexico and Turkey (figure S6). Not shown, the central ments) in 2012.

5
Environ. Res. Lett. 9 (2014) 084018 S J Davis and R H Socolow

Figure 4. Annual committed emissions and future emissions from primary power infrastructure under different assumed generator lifetimes.
In comparison to figure 3, here we show committed emissions (light green) for five different values of the generator lifetime (20, 30, 40, 50
and 60 years). For all five cases, we also project remaining commitments as annual emissions in the future (dashed lines). Realized
commitments do not change under the different lifetime assumptions (black), but remaining commitments grow from 98 Gt CO2 if generators
retire after 20 years to 578 Gt CO2 if the same generators instead operate 60 years. Also shown are the same IEA estimates [25] of actual
global power plant emissions as in figure 3(a) (solid blue line).

Figures 5(c) and (d) show annual changes in remaining through 2012 (figure 5(c)). In comparison, new commitments
commitments (i.e. the derivative with respect to time of the in Europe have been less variable, and remaining committed
curves in figures 5(a) and (b)) by region and fuel, respec- emissions have seldom decreased over the entire period
tively. The rate of increase of global remaining commitments 1950–2012 (figure 5(a) and (c)).
was approximately 6 Gt CO2 per year throughout the 1970s The year-on-year increases in remaining commitments
and 1980s, driven by coal-fired generators (figure 5(d)) being can be compared systematically with actual annual emissions
installed in the US and Europe (figure 5(c)). Construction in by examining their ratio, R, where,
the US and Europe slowed in the late 1980s, and the rate of
increase of global remaining commitments reached a mini- ( Cnew + Clate ) − ( Creal + Cearly )
mum of 2.0 Gt CO2 per year in 1992 (figure 5(c)). Between R= . (1)
( Creal )
1990 and 2012, in spite of a rate of change in remaining
commitments that averaged −0.8 Gt CO2 per year in the US Here Cnew are new commitments in a given year, Clate are
and 0.1 Gt CO2 per year in Europe, the rate of increase of additional commitments in that year related to generators that
remaining global commitments averaged 6.5 Gt CO2 per year. continue to operate after their expected retirement age, Creal
The message of the most recent data is that China is ‘passing are realized commitments in that year (i.e. annual emissions),
the torch’ to other industrializing countries: averaged over the and Cearly are commitments in that year related to generators
most recent three years of our data (2010–2012), the global that are shut down before their expected retirement age. The
rate of increase in committed emissions was 7.7 Gt CO2 yr–1. numerator of equation (1) is graphed already in figures 5(c)
After peaking at an average of 13.7 Gt CO2 yr–1 in and (d). Dividing by new commitments in that year creates a
2005–2006, the rate of increasing commitments in China has useful dimensionless ‘commitment ratio,’ R, which is plotted
fallen to 5.3 Gt CO2 yr–1 and the combined value for India in figures 5(e) and (f). Trends in R say much about evolution
and the rest of the world has risen to 3.0 Gt CO2 yr–1, influ- of the stock of power plants. If no new emitting infrastructure
enced especially by the rapid expansion of generators in is built and all generators retire at the assumed reference year
places like Indonesia, Russia, Saudi Arabia and Iran (see (the case considered by Davis et al [16]), R is sustained at −1,
figure S6 and table S3 for a comprehensive list of remaining as total remaining commitments decrease in step with pre-
CO2 commitments from oil, gas, and coal generators in 233 viously committed emissions being realized. An R of 0 is
individual countries). Note that the boom of natural gas in the obtained for a ‘steady state’ where increases in remaining
US [29] is visible as a roughly five-year surge in both US commitments are exactly offset by realized emissions. An
commitments (figure 5(c)) and gas commitments (figure 5(d)) energy system that is phasing-out fossil-fuel infrastructure
beginning about 1998. Except for this brief spike, remaining and transitioning to carbon-emissions-free energy technolo-
commitments in the US have declined from the late 1980s gies, will maintain an R less than 0 over decades, just as an

6
Environ. Res. Lett. 9 (2014) 084018 S J Davis and R H Socolow

Figure 5. Changes in remaining commitments. Total remaining committed emissions of power generators built worldwide since 1950
increased from near zero in 1950 to 307 Gt CO2 in 2012, driven by coal-fired generators (b) built in the US and Europe during the 1970s and
1980s, and in China, India and the rest of the world beginning about 1985 (a). Annual increases in remaining global commitments hovered
around 6 Gt CO2 per year in the 1970s and 1980s, dipped to a minimum of 2.2 Gt CO2 per year in 1990, then rose sharply between 1994 and
2006, and declined again between 2006 and 2012 (c) and (d). Panels (e) and (f) show the commitment ratio, R, of annual changes in
remaining commitments to annual emissions (here calculated using five-year running averages for both). Since 1994, the ratio shows
recarbonization of the global power sector despite the steady state of commitments in Europe and decarbonization in the US (e).

energy system with an expanding fossil-fuel infrastructure period 1971–2011 (because total remaining commitments
will maintain an R greater than 0. have never decreased). R followed a V-shaped path in those
Figures 5(e) and (f) plot the commitment ratio, R, using a years, decreasing from 1.7 in 1971 to a minimum of 0.4 in
five-year running average for both numerator and denomi- 1994 and then back up to 0.8 in 2011 (black curve in
nator. The global value of R was positive during the entire figure 5(e)). The 2011 value of 0.8 can be calculated, from

7
Environ. Res. Lett. 9 (2014) 084018 S J Davis and R H Socolow

equation (1), as the consequence of five-year-average annual benchmarks might directly inform regulatory decisions gov-
values for Cnew and Creal of 21 Gt CO2 and 12 Gt CO2, erning the permitting of new infrastructure where the objec-
respectively. (Global values of Cnew and Creal are graphed as tive is to reflect the climate implications that will materialize
‘committed’ and ‘annual emissions’ in figure 3(a); values of over many decades.
Cearly and Clate are negligible). We have restricted our analysis of committed emissions
Analysis of the regional data in figure 5(e) reveals that to the power sector, which is probably the easiest sector to
the 20-year average values of the commitment ratio, analyze: although the data requirements are formidable, they
1991–2011, are −0.3 for the US, 0.1 for the EU28, 1.2 for are tractable. In the transportation sector, the analysis will be
India and 3.1 for China (figure 5(e)). Across fuels similar at the level of committed emissions from vehicles:
(figure 5(f)), year-to-year changes in R in most years are lifetimes are shorter and better known, but estimating vehicle-
driven by commitments of coal-fired infrastructure, but miles traveled is a challenge. For the buildings sector, life-
between 2000 and 2005 the R of coal generators leveled off at times are longer and less well known; uncertainties associated
∼0.5 and the R of gas generators spiked upward to more than with retrofitting and replacing the buildings stock will be
2. By 2011, the R of coal was 0.7 and the R of gas was 0.9. large. Commitment accounting can also be applied in con-
Between 2000 and 2011, committed emissions from oil-fired ceptually more difficult territory, such as town planning and
generators fell (few were built), producing negative values in the construction of supporting infrastructure such as roads,
figures 5(d) and (f). power lines, and pipelines [34].
Reducing CO2 emissions will ultimately mean retiring
CO2-emitting infrastructure more quickly than it is built.
4. Discussion and conclusions However, trends have long pointed the other way: fossil
power is expanding globally, and here we have shown that
Commitment accounting of CO2 emissions provides critical even the already substantial committed emissions of the
information about future emissions related to infrastructure power sector are increasing. By revealing the emissions that
that currently exists or might be built. Committed emissions are anticipated decades into the future, commitment
from existing infrastructure can be readily compared to sce- accounting of CO2 emissions may help to integrate analyses
narios of future emissions such as those used by the IPCC. of capital investment, cumulative emissions, and damages
Assuming a 40 year lifetime of generators, our estimates of from climate warming.
committed emissions from the power sector already represent
53%, 41% and 21% of the fossil-fuel CO2 emissions from all
sectors in China, India, and the US, respectively, projected in
RCP2.6 (the low-emissions Representative Concentration Acknowledgements
Pathway [30]) during the period 2013–2050 [31]. The esti-
mated commitments are a smaller percentage of emissions in We thank David Hawkins for long-ago sharing his ideas
other, warmer RCPs, but RCP2.6 is the only RCP pathway about ‘committed emissions’ with R Socolow. We also thank
that is likely to avoid exceeding 2 °C of warming relative to Ken Caldeira, Felix Creutzig, Stéphane Hallegatte, Thomas
the pre-industrial period (i.e., to meet the current international Kreutz, Michael Levi, H Damon Matthews, Massimo Tavoni
climate target) [9, 32, 33]. and two anonymous reviewers for insightful comments and
With reference to what might be built, committed emis- suggestions.
sions can be teased out of energy scenarios, where in most
cases the information is already embedded; once made
explicit, these scenarios will become richer and easier to
References
evaluate. For instance, we estimate that the ‘New Policies’
scenario in the IEA’s World Energy Outlook 2013 entails
new global commitments of roughly 5 Gt CO2 yr–1 of coal [1] World Resources Institute and World Business Council for
generators and 3 Gt CO2 yr–1 of gas generators between 2012 Sustainable Development 2002 GHG Protocol Corporate
Accounting Standard (Washington DC: World Resources
and 2020, while the ‘450 scenario’ would require overall Institute and World Business Council for Sustainable
reductions in the commitments from existing coal-fired gen- Development)
erators (−0.5 Gt CO2 yr–1 between 2012 and 2020, e.g., by [2] United Nations Framework Convention on Climate Change
early retirement or CCS retrofitting) and allow new commit- 2006 Updated UNFCCC Reporting Guidelines on Annual
ments of ∼2 Gt CO2 yr–1 of gas generators over the same time Inventories Following Incorporation of the Provisions of
Decision 14/cp.11 (Geneva: United Nations Framework
period (assuming new generators with the lowest plausible Convention on Climate Change)
carbon intensity given their fuel type) [24]. In comparison, [3] Allen M R et al 2009 Warming caused by cumulative carbon
figure 3(c) shows that gas generators brought online in 2012 emissions towards the trillionth tonne Nature 458 1163
represented ∼5 Gt CO2 of new commitments—roughly twice [4] Matthews H D, Gillett N P, Stott P A and Zickfeld K 2009 The
the level of commitments in the IEA scenarios. Detailed proportionality of global warming to cumulative carbon
emissions Nature 459 829
analysis of this sort might provide policymakers with prac- [5] Nakicenovic N et al 2000 IPCC Special Report on Emissions
tical, annual benchmarks for new emitting infrastructure that Scenarios ed N Nakicenovic and R Swart (Cambridge:
can be built under various mitigation scenarios. In turn, these Cambridge University Press)

8
Environ. Res. Lett. 9 (2014) 084018 S J Davis and R H Socolow

[6] Clarke L et al 2009 International climate policy architectures: [19] Karlsson R 2012 Carbon lock-in, rebound effects and china at
overview of the EMF 22 international scenarios Energy the limits of statism Energy Policy 51 939
Econ. 31 S64 [20] ‘World electric power plant database’ (Platts, 2014)
[7] Krey V and Riahi K 2009 Implications of delayed participation [21] ‘Carma database v. 3.0’ (Center for Global
and technology failure for the feasibility, costs, and Development, 2012)
likelihood of staying below temperature targets–greenhouse [22] Taylor M J and Fuller L C, 1986 Coal-Fired Electric Power
gas mitigation scenarios for the 21st century Energy Econ. Plant Life Extension: An Overview (Oak Ridge, TN: Oak
31 S94 Ridge National Laboratory)
[8] Peters G P et al 2012 The challenge to keep global warming [23] Bernanke B 1983 Irreversibility, uncertainty, and cyclical
below 2 °C Nat. Clim. Change 3 4–6 investment Q. J. Econ. 98 85
[9] Meinshausen M et al 2009 Greenhouse-gas emission targets [24] IEA 2013 World Energy Outlook (Paris: OECD/IEA)
for limiting global warming to 2 °C Nature 485 1158 [25] IEA 2012 Co2 Emissions from Fuel Combustion (Paris:
[10] Rogelj J et al 2011 Emission pathways consistent with a 2 °C OECD/IEA)
global temperature limit Nat. Clim. Change 1 413 [26] Boden T A, Marland G and Andres R J 2014 Global, regional,
[11] Unruh D M and Carillo-Hermosilla J 2006 Globalizing carbon and national fossil-fuel CO2 emissions. (Oak Ridge, TN:
lock-in Energy Policy 34 1185 Carbon Dioxide Information Analysis Center)
[12] Matthews H D and Solomon S 2013 Irreversible does not mean [27] Minx J et al 2011 A ‘carbonizing dragon’: China’s fast
unavoidable Science 340 438 growing CO2 emissions revisited Environ. Sci. Technol.
[13] Wigley T M L, Richels R and Edmonds J A 1996 Economic 45 9144
and environmental choices in the stabilization of [28] Raupach M R et al 2007 Global and regional drivers of
atmospheric CO2 concentrations Nature 379 240 accelerating CO2 emissions Proc. Natl. Acad. Sci. 104
[14] Kramer G J and Haigh M 2009 No quick switch to low-carbon 10288
energy Nature 462 568 [29] Joskow P L 2013 Natural gas: from shortages to abundance in
[15] Friedlingstein P et al 2011 Long-term climate implications of the United States Am. Econ. Rev. 103 338
twenty-first century options for carbon dioxide emission [30] van Vuuren D P et al 2011 The representative concentration
mitigation Nat. Clim. Change 1 457 pathways: An overview Clim. Change 109 5
[16] Davis S J, Caldeira K and Matthews H D 2010 Future Co2 [31] van Vuuren D et al 2007 Stabilizing greenhouse gas
emissions and climate change from existing energy concentrations at low levels: an assessment of reduction
infrastructure Science 329 1330 strategies and costs Clim. Change 81 119–59
[17] Rogelj J, McCollum D L, Reisinger A, Meinshausen M and [32] COP15, Copenhagen accord. (2009)
Riahi K 2013 Probabilistic cost estimates for climate change [33] Alexander L et al 2013 IPCC Working Group I Summary for
mitigation Nature 493 79 Policymakers (Cambridge: Cambridge University Press)
[18] Davis S J, Cao L, Caldeira K and Hoffert M I 2013 Rethinking [34] Guivarch C and Hallegatte S 2011 Existing infrastructure and
wedges Environ. Res. Lett. 8 011001 the 2 °C target Clim. Change 109 801