This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 182.73.193.34

This content was downloaded on 26/08/2015 at 12:20

Please note that terms and conditions apply.

Cookstoves illustrate the need for a comprehensive carbon market

View the table of contents for this issue, or go to the journal homepage for more

2015 Environ. Res. Lett. 10 084026

(http://iopscience.iop.org/1748-9326/10/8/084026)

Home Search Collections Journals About Contact us My IOPscience

Environ. Res. Lett. 10 (2015) 084026 doi:10.1088/1748-9326/10/8/084026

LETTER

Cookstoves illustrate the need for a comprehensive carbonmarket

Luke Sanford and Jennifer BurneySchool ofGlobal Policy and Strategy, University of California, SanDiego, CA,USA

E-mail: jburney@ucsd.edu

Keywords: cookstoves, carbonmarket, offsets, black carbon, particulatematter, carbonmonoxide, combustion

Supplementarymaterial for this article is available online

AbstractExisting carbon offset protocols for improved cookstoves do not require emissions testing. They arebased only on estimated reductions in the use of non-renewable biomass generated by a given stove,and use simplistic calculations to convert those fuel savings to imputed emissions of carbon dioxide(CO2). Yet recent research has shown that different cookstoves vary tremendously in their combustionquality, and thus in their emissions profiles of bothCO2 and other products of incompletecombustion. Given the high global warming potential of some of these non-CO2 emissions, offsetprotocols that do not account for combustion qualitymay thus not be assigning either appropriateabsolute or relative climate values to different technologies.We use statistical resampling of recentemissions studies to estimate the actual radiative forcing impacts of traditional and improvedcookstoves.We compare the carbon offsets generated by protocols in the four carbonmarkets thatcurrently accept cookstove offsets (CleanDevelopmentMechanism, AmericanCarbonRegistry,VerifiedCarbon Standard, andGold Standard) to a theoretical protocol that also accounts foremissions of carbonaceous aerosols and carbonmonoxide, using appropriate statistical techniques toestimate emissions factor distributions from the literature.We show that current protocolsunderestimate the climate value ofmany improved cookstoves and fail to distinguish between (i.e.,assign equal offset values to) technologies with very different climate impacts.We find that acomprehensive carbon accounting standardwould generate significantly higher offsets for someimproved cookstove classes than those generated by current protocols, andwould createmuch largerseparation between different cookstove classes. Finally, we provide compelling evidence for theinclusion of renewable biomass into current protocols, and propose guidelines for the statistics neededin future emissions tests in order to accurately estimate the climate impact (and thus offsets generatedby) cookstoves and other household energy technologies.

1. Introduction

Roughly one third of the world’s population relies onsolid, unprocessed biomass fuels burned in traditionalstoves to meet daily cooking needs [1]. These fuel andstove combinations typically result in poor combus-tion and significant emission of products of incom-plete combustion (PICs), including aerosol particulatematter (PM) like black carbon (BC) and organiccarbon (OC), and ozone precursors such as carbonmonoxide (CO). These substances contribute directlyto increased morbidity and mortality both for thosenearby and for others exposed when indoor air

pollution is transported outdoors. Notably, air pollu-tion, a significant faction of which comes fromdomestic biomass burning, is now estimated to be thesingle largest environmental risk to humanity [2].

This problem is not new; the literature is repletewith studies across disciplines on the impacts of cook-stove pollution and attempts to solve the problemwithnew technologies (e.g., [3–13]). Yet even with decadesof experience measuring impacts and piloting inter-ventions, widespread adoption of improved cook-stoves has not occurred [14–16]. Studies have shownthat socio-cultural and technical issues contribute tothis low change-over rate, but economics remains the

OPEN ACCESS

RECEIVED

24April 2015

REVISED

13 July 2015

ACCEPTED FOR PUBLICATION

26 July 2015

PUBLISHED

25August 2015

Content from this workmay be used under theterms of theCreativeCommonsAttribution 3.0licence.

Any further distribution ofthis workmustmaintainattribution to theauthor(s) and the title ofthework, journal citationandDOI.

© 2015 IOPPublishing Ltd

main driver [15–18]. The world’s biomass-dependentpopulations are poor, and the cost-benefit analysis ofthe existing suite of technologies does not work intheir favor. On the benefits side, the health effects ofcleaner cooking technologies may be attenuated if, forexample, everyone else in the neighborhood is stillusing highly polluting stoves [19]. That is, commu-nity-based uptake may be necessary for individualhouseholds to reap the maximum health gains. Inaddition, if households have no money to spend onhealth care to begin with, or if they perceive no oppor-tunity cost to gathering free fuels, the benefits ofswitching technologies may not be monetized. On thecost side, more affordable options may not beimproved enough to incentivize behavior change[13, 14, 20]. Finally, absent financing, the cleanesttechnologies remain out of reach for the families thatcouldmost benefit from them [10, 16, 18, 21, 22].

Carbon markets have received increasing atten-tion in recent years as a potential solution to the cook-stove financing problem, and some of the mostimportant carbon markets in the world now includeprotocols for improved cookstoves [23]. In theory,offset revenue could make these improved technolo-gies more affordable for those who need them mostand thus improve health (and other) outcomes formillions of poor families around the world. At present,offset calculations are based on the reduction in CO2

emissions from burning non-renewable biomass in animproved stove versus a traditional, or unimproved,stove. While these protocols are a step in the rightdirection, they miss a key part of every cookstove’simpact: they ignore the fact that biomass cookstovesemit PICs—most importantly, carbonaceous aerosolsand carbon monoxide—as well as carbon dioxide.These PICs have important climate and healthimpacts. BC is the second-or third- most importantclimate forcing agent after carbon dioxide, and PM asa whole from indoor cooking smoke causesmillions ofpremature deaths each year [2, 24, 25]. Carbon mon-oxide is a main precursor to tropospheric ozone, apotent warmer and a compound detrimental to bothhumanhealth and plants [25].

By ignoring PICs and assigning offset values basedon an overly simplified representation of combustion,current protocols thus do not likely assign accurate cli-mate values to different types of cookstoves, either inabsolute or relative terms. At the extreme, it is possiblethat CO2-only protocols could mistakenly assign off-set values to cookstoves that are in fact worse for cli-mate, depending on the abundance and opticalproperties of their PIC emissions. The future climateand human health consequences of CO2-only proto-cols could thus be substantial if markets mistakenlyincentivize uptake of technologies that in fact emitmore PICs.

Here we explore the hypotheses above within thepolicy space of existing protocols. We begin with anoverview of existing protocols. We then examine how

existing carbon markets would value and rank differ-ent classes of cookstoves under the current CO2-basedformulae. We compare results for these offsets calcu-lated (a) using the protocols’ own prescribed emis-sions factors, and (b) emissions factors derived from ameta-analysis of the existing cookstove testing litera-ture. Finally, we contrast these carbon values andrankings with a new holistic protocol that doesaccount for CO2, CO, and BC and OC emissions. Dif-ferent outcomes under the two different sets of ruleswould motivate a more comprehensive carbonaccounting in cookstove (and other) offset protocolsto properly align market forces with climate goals andoverall humanwellbeing.

2. Carbon accounting standards

A growing number of carbon accounting standards(‘carbon markets’) exist around the world for thepurpose of certifying carbon-reducing programs.These markets develop protocols to certify projectsthat demonstrably decrease carbon emissions, andthen sell those reductions as credits (‘offsets’) topolluters who are either seeking to comply withregulations or who are voluntarily participating in acarbon market. In practice, an offset is generated by atechnology or other intervention that creates a mea-surable reduction in emissions compared to statusquo, or an appropriate baseline counterfactual; theamount of the reduction (in units of mass of CO2) isthen converted to a monetary value within theindividualmarket.

Currently four carbon markets include protocolsfor improved cookstove projects: the Clean Develop-ment Mechanism (CDM) [26], the Verified CarbonStandard (VCS) [27], the American Carbon Registry(ACR) [28] and the Gold Standard (GS) [29]. As of theend of 2013, all but one of the current cookstove offsetprojects were registered under either the CDM or GSmethodologies (the lone exception is registered underACR) [23]. Over the next 10 years (the accountingperiod for these protocols), these offset projects areprojected to generate around 10 million tons of CO2

reductions [23], equivalent to about 1% of voluntarycarbon market trading per year [30]. (See the SI for amore in-depth discussion of existing markets and spe-cific protocol formulae, available at stacks.iop.org/erl/10/084026/mmedia).

Three types of tests are generally used to estimatethe climate performance of improved cookstoves.These range in complexity from the relatively simplewater boiling test (WBT) [31] to the more complexcontrolled cooking test (CCT) [32] and kitchen per-formance test (KPT) [33]. The GS accepts only theKPT for emissions estimates; the other three protocolsaccept any of the above tests. (See SI for additionalinformation about the specifics of each test.) Impor-tantly, for the purposes of the offset calculation, no

2

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

actual emissions measurements are required; the onlytest requirement is a measure of the fuel used by a cer-tain stove versus a traditional stove to completethe test.

At present, the offset protocols use emissions fac-tors drawn from Intergovernmental Panel on ClimateChange (IPCC) reports to convert fuel savings toemissions reductions. Two glaring problems exist withthis approach: first, emissions factors derived from theliterature values appear to differ significantly from theIPCC values. Second, and related, no cookstove meth-odology currently accounts for statistical uncertaintyin measurements. Johnson (2010) showed that undercurrent methodologies, for many (even the majority)of projects there are no statistically significant differ-ences between baseline and improved stoves in termsof emissions factors [34] and this is one of the compel-ling problems addressed here. In addition Wang et al(2014) address a key data limitation in that most emis-sions testing papers conduct far too few tests to accu-rately derive emissions factors (and concomitantuncertainties) [35]. These papers point to a key gap inexisting data—the need for statistically-sound emis-sions factor derivations—and a direction for futureresearch; we discuss our approach to this data gap forthe purposes of this analysis in detail below.

3.Methods

3.1. Emissions factor derivationsThe protocols used by the CDM, ACR, and VCS arebased only on the net weight of non-renewablebiomass saved by a given cookstove relative to abaseline technology—typically a three-stone stove,open fire, or the prevailing local version of anunimproved mud hearth. These methodologiesinclude no treatment of uncertainty: the average fuelreduction compared to baseline is simply multipliedby the fraction of fuel that is non-renewable, themarket-proscribed emissions factor to convert fuelmass to CO2mass, and a factor to convert the per-taskoffset to a per-year or per-stove-lifetime offset value.(See the SI for protocol formulae.)

These protocols assume perfect combustion, orthat all carbon in the biomass will be converted toCO2. The GS protocol acknowledges non-CO2 emis-sions by including a fixed emission factor to accountfor CH4 andN2O emissions frombiomass; none of theprotocols include PICs such as BC, OC, or CO. How-ever, because different stoves have different combus-tion efficiencies, they emit more or less of the totalcarbon mass as CO2, with the rest being emitted asPICs [36]. Many studies have pointed out the fact thatBC and other PICs form the largest contributions ofbiomass burning to warming (e.g., [37]), but thisknowledge has not been incorporated into protocols.

To understand the role of uncertainty and the con-tributions of non-greenhouse gas emissions to true

climate impacts of different cookstove technologies,we conducted a meta-analysis of cookstove emissionsstudies and derived emissions factor distributions(means, medians, standard errors) for CO2, CO, OC,and BC. We derived emissions factors for each majorcookstove class represented in the literature—tradi-tional or three-stone stoves, natural draft or rocketstoves, forced draft stoves, gasifier, and charcoalstoves. Natural draft, or rocket stoves, are typicallycylindrical shaped stoves that feature passive struc-tural modifications intended to enhance airflow to thecombustion chamber (and thus combustion effi-ciency); forced draft stoves feature an external fan thatactively drives air into the combustion chamber. Gasi-fier stoves use two-stage combustion inwhich biomassis first burned in the primary combustion chamber(typically in oxygen poor conditions); the CO pro-duced in the primary combustion cycle is then oxi-dized to CO2 in a secondary combustion cycle. Finally,charcoal stoves, as their name suggests, use charcoal astheir primary fuel and are often promoted in peri-urban areas where households cannot easily accessbiomass likewood or crop residues.

While this aggregation blurs distinctions betweenindividual stove models within a class, it does capturemuch of the technology differentiation in theimproved cookstove landscape and helps provide gen-eralized insight into the offset protocols. Moreover,very few studies have sufficient statistical power forappropriate analysis at the individual stove level.Mathematically speaking, a sample size (N) of six is theminimum to know with 95% confidence that thepopulation median value lies between the sampleminimum and sample maximum. More generally,N 20⩽ is considered a small sample. Unfortunatelymost studies use only 3–4 tests per stove, and indivi-dual stoves, testing conditions or even the test con-ducted are not standardized across studies [35]. Theaggregation by stove class therefore provides addedstatistical power by pooling like technologies acrossstudies. Emission factor averages and standard errorsfor each stove class were weighted by number of testsconducted in each study.We similarly derived fuel useefficiency distributions from the same studies for eachstove class to convert emissions and offsets to a per-task basis.

Ourmeta-analysis to derive fuel use and emissionsfactor distributions was complicated by several reali-ties. First, the state of cookstove measurements hasevolved rapidly over time, and formalization of differ-ent testing methods has happened fairly recently.Hence, existing studies include a broad range of lab vs.field tests, conducted with various fuels, etc. Second,studies report results in very different ways—includ-ing, but not limited to, emissions per test, emissionsper unit of energy output, emissions per kg of biomass,andmolar ratios of carbon.We calculated conversionsto standardize all emissions factors to units of g per kgof biomass fuel. Third, very few studies measure the

3

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

emissions of BC (also called elemental carbon or EC)and OC rather than concentrations of total suspendedparticulates (TSP), PM10, or PM2.5. Measurements ofthe optical properties of the emitted PM are criticalbecause BC is a potent warmer, but OC is (moreweakly) cooling. The relative abundances of the twotypes of carbonaceous aerosols is thus a critical factorin the net climate impact of a given cookstove’s emis-sions. The studies that do provide data on the opticalproperties of stove particulate emissions also use dif-ferent methods and report different metrics: somereport the total EC:OC ratio from filter-based mea-surements, while others measure the single-scatteringalbedo (SSA), a measure of the fraction of extinctiondue to scattering (versus absorbing) aerosols.

The initial scope of our project included everystudy that measured cookstove emissions. Many ofthese studies included fuel sources other than wood.Because the only data on BC and OC emissions comesfrom wood and charcoal burning stoves, we only con-sider emissions from wood and charcoal burningstoves (excluding fuels such as dung, crop residue, andothers). Not nearly as many studies measure BC andOC emissions as those that measure CO2 or CO emis-sions, so our BC andOC estimates stem from a smallernumber of tests than estimates for other PICs. The fulllist of included sources can be found in table 1. Dataand methods for harmonizing data across studies canbe found in SI; the final derived emissions factor dis-tributions for CO2, BC, OC, and CO are shown intable 2.

3.2. Resampling calculationsWe created a database of 1000 simulated tests for eachcookstove class using random draws from the derivedemissions factor and fuel use distributions (table 2);this resampling method addresses uncertainty relatedtowithin-class stove differences and non-standardizedtesting conditions. These ‘observations’ for eachcookstove class were then used to conduct threeanalyses. First, we calculated the emissions (in mass ofCO2) for each class using the formulae from existingprotocols (CDM, ACR, andGS; see the SI for formulaeand details).We calculated stove-class emissions using(a) the CO2 emissions factor prescribed by the givenprotocol, and (b) our resampled values from theliterature. We generated an emissions distribution foreach stove type under each protocol, at 10% intervalsof fraction of non-renewable biomass (fNRB) (from0% to 100%). We then subtracted the distribution oftotal emissions for the traditional (baseline) stovefrom each of the improved stove types to get adistribution of total offset in terms of grams of CO2

per cooking task.Second, we calculated offsets by stove class for a

new, modified offset protocol that included BC, OC,and CO emissions resampled from the literature-derived emissions factor distributions, in addition toCO2. Again, we generated an emissions distributionfor each stove type under each protocol, at 10% inter-vals of fNRB (from 0% to 100%), and subtracted theemissions distribution of the traditional stove to getthe net offset. For the modified protocol, however, the

Table 1.Previous studies andmeasurement parameters used to conductmeta-analysis of emissions by cookstove type.With the exceptionsof Roden (2006) [17] Johnson (2007) [53] and Preble (2014) [42] every study conducts three tests per stove.We aggregate data from allstoves of the same class: traditional, or unimproved (T), charcoal (C), natural draft (ND), forced draft (FD), and gasifier.Measured speciesvary by study: TSP= total suspended particles, CO= carbonmonoxide, CO2= carbon dioxide, BC=black carbon,OC=organic carbon.Details of each study and procedures to harmonize data can be found in the SI.

Paper Stove types Tests per stove class Speciesmeasured

Joshi (1989) [3] T, ND 3 TSP, CO

Ballard-Treemer (1996) [4] T, ND 6,17 TSP, CO

Smith (2000) [36] T, ND,C 9, 18, 6 TSP, CO,CO2

Venkataraman (2001) [52] T, ND 4, 12 PM10, CO

Roden (2006) [17] T 13 PM10, CO, BC,OC

Johnson (2008) [53] T, ND 13, 17 CO, CO2, BC

MacCarty (2008) [54] T, ND, FD,G, C 3 each PM10, CO, BC,OC

Jetter (2009) [7] T, ND, FD 3, 6, 3 PM10, PM2.5, CO

MacCarty (2010) [55] T, ND, FD,G, C 3, 66, 15, 9, 12 PM10, CO

Jetter (2012) [56] T, ND, FD,G, C 6, 18, 3, 3, 21 PM2.5, CO,CO2

Preble (2014) [42] T, ND 21, 20 PM2.5, CO,CO2, BC,OC

Table 2. Literature-derivedmean emissions factors (g kg−1 fuel) and normalized fuel use (kg fuel / task,relative to traditional stove) shown, with standard deviations shown in parentheses.

Traditional Charcoal Gasifier Natural draft Forced draft

Emissions factor (T) (C) (G) (ND) (FD)

CO2 1570 (83.7) 2710 (216) 1940 (186) 1582 (115) 1901 (72.2)

CO 57.4 (24.8) 192 (41.0) 29.4 (11.3) 40.0 (6.65) 9.81 (4.25)

BC 1.28 (0.59) 0.28 (0.12) 0.28 (0.10) 1.26 (0.52) 0.060 (0.044)

OC 2.97 (1.63) 1.54 (0.93) 0.82 (0.29) 1.47 (0.11) 0.14 (0.10)

Fuel use 1(0) 0.65 (0.10) 0.67 (0.20) 0.61 (0.13) 0.52 (0.12)

4

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

fNRBwas only applied toCO2 emissions (as CO2 is thespecies ostensibly sequestered annually with renew-able biomass fuels). To convert the different emissionsspecies to CO2 equivalents, we use 100-year globalwarming potentials (GWPs) from the fifth assessmentreport (AR5) of the IPCC [24].

3.3. Treatment of uncertaintyFour uncertainty sources are likely in the estimation ofcarbon offsets from a cookstove project. The first isdue to estimation of the amount of renewable biomassversus non-renewable biomass harvested in the pro-ject area. Because this uncertainty is location-basedand expected to be independent of stove type, we donot include it in our estimations. We run our offsetcalculations for range of different values as a fNRB toprovide bounds on this dimension.

The second source of uncertainty comes from fueluse measures. Current protocols are are based on fueluse reductions (the mass of non-renewable fuel usedrelative to baseline over the reporting period), but thequantities of fuel used necessarily vary depending onlocation, the type of cooking task, the volume of foodor water to be heated, etc. (In all cases, fuel weightsrefer to fuel dried to within test-specificmoisture stan-dards, though both standards and moisture contentdiffer between studies.) To harmonize across tests andstudies, we calculate per-task fuel savings compared totraditional stoves by normalizing fuel use distributionsof improved stoves to the distributions for a tradi-tional stove within the same study. The per-taskmetric ismore intuitive for comparing stovemodels orclasses (versus, for example, to a strict per kg fuel com-parison), an is easily scaleable over the reporting per-iod. We account for fuel use efficiency in allcalculations and note that this means that some stovesthat produce higher quantities of pollutants on a per-kg of fuel basis have a lower emissions footprint on aper-task basis. (It is also worth noting that most exist-ing fuel use data come from studies employing theWBT, which does not account for many of the com-plexities of using a stove in the field, including fuelcharacteristics, thermal power of the stove, accessoriesincluding the pot and (whether or not there is a) hood,how the stove is lit, and what is being cooked, to namea few.)

The third source of uncertainty comes from themeasurements of different emissions components.Emissions measurements have evolved rapidly withadvances in the understanding of aerosol PM and itsimpacts on health and climate. On the health side, thediscovery that smaller particles pose greater threats tohuman health has led to a change in measurementstrategy for air pollution monitoring; on the climateside, information about the chemical composition andoptical properties of PM is necessary to understand itsclimate impacts. Consequently, the majority of the lit-erature about emissions from cookstoves falls under

two categories: one is the literature about healthimpacts of cooking indoors and the other is about cli-mate impacts of burning biofuels. The health litera-ture was the first to examine emissions other thanCO2, but it tends to disaggregate particulates by size(PM2.5 and PM10) rather than by climate forcingpotential (e.g., absorbing versus scattering). Onlymore recently have climate-conscious studies focusedon speciated measurements of particulates; that is,quantifying how much of emitted PM is BC (absorb-ing) versus neutral or scatteringOC [38–41].

The studies included in our analysis (table 1) aredescribed in greater detail in the supporting informa-tion. Currently, the only study that measures the opti-cal properties of particulate emissions with enoughtests to generate statistical confidence only examinedone stove; several multi-stove studies include BC andOCmeasurements, but lack sufficient statistics and donot report uncertainties. In order to bound the var-iance on BC and OC emissions by stove class, wetherefore used the coefficient of variance (COV) ofoverall particulate emissions data to estimate the BCand OC fractions of PM for different stoves. This pro-cess is described inmore detail in the SI.

Because we group many stove models into over-arching stove classes, both efficiency and emissionsuncertainties are higher than they would be for a studythat used a large number of tests on a single stovemodel. We use the results of one recent study [42] todemonstrate that this is especially the case with naturaldraft stoves—by far the largest class in terms of num-ber of different models. This case study on the Berke-ley–Darfur stove is the only existing study featuringboth a large number of tests andmeasurements of BC,OC, andCO.

The fourth source of uncertainty comes from theradiative forcing effects of BC andOC. GWP estimatesfor these particulates vary widely, due to their shortatmospheric lifetimes, their complex effects onweather patterns and snowpack, and regional climaticand source differences [24, 25, 43–45]. The IPCC AR5gives a broad range of possible values for the globalGWP of BC and OC: 900 (range of 100–1700) for BCand −46 (range: −18 to −92) for OC (we use the 100-year time horizon because this is the standard formea-suring GWPs across offset protocols) [24, 46]. We usethe central values for BC and OC for the main portionof our analysis, and a GWP of 2.2 for CO [24].We dis-cuss the implications of GWP uncertainties below andin the SI (figure S4).

4. Results

4.1. Fuel savingsThe cookstove classes considered here are all consid-ered improved because they use less fuel than tradi-tional cookstoves. These average fuel savings acrossstoves classes are shown in table 2. For each study

5

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

(table 1)we normalized the fuel savings of the differentimproved stoves to 1 kg of fuel in a traditional stove toaccount for different tests used across studies (withuncertainties calculated appropriately). The fuel usenumbers in table 2 thus represent the fuel-use-per-task value that would be scaled (i.e., task value * tasksper day * days) to calculate the total offset credit for aproject. Figure 1 shows two sets of boxes for each typeof stove; the left-hand set (‘P’) corresponds to theemissions values (in g CO2 per task) for each stovewhen literature-derived fuel use values are used tocalculate the offset for each protocol (using defaultprotocol CO2 emissions factors). This inclusion ofvariance in fuel savings illustrates the importance ofproper treatment of uncertainty in emissions proto-cols. The assigned emissions values for each stove type,based on fuel use changes relative to 1 kg of biomassburned in a traditional stove, vary by a factor of 2 ormore for each improved stove type and protocol.

4.2. Protocol v. literature-derived emissions factorsWe then compare the value of CO2 emissions usingthe values proscribed by each protocol and the CO2

emissions factors derived from the literature, assum-ing a renewable fraction (fNRB) of 50%. The CDMproscribes a value of 1244 gCO2 kg

−1 fuel. TheGS usesa value of 1747 g kg−1 of fuel, and an additional valueof 455 g kg−1 fuel meant to account for other long-lived greenhouse gas emissions (methane, CH4, andnitrous oxide, or N2O). The ACR and VCS protocolsfix the value at 1792.5 g kg−1 for wood and the ACRsets a separate value of 1747.5 g kg−1 for charcoal.These CO2 emissions estimates for different stovetypes on a per-task basis are illustrated in the right-hand (‘L’) set of boxes infigure 1.

Comparing the protocol and literature-derivedemissions factors in figure 1 shows that the medianvalues proscribed by the protocols are often sig-nificantly different from the values measured in actualtests. The CDMunderestimates the CO2 emissions forevery stove type except the gasifier stove at 95% con-fidence (the distributions do not overlap). The ACRand GS provide good estimates of the amount of CO2

emitted by gasifier, natural draft, and forced draftstoves, but underestimate CO2 emissions from char-coal stoves and underestimate the emissions from tra-ditional stoves. GS values are higher than the otherprotocol values, across stove types, due to the inclu-sion of non-CO2 greenhouse gas emissions in the GSprotocol. The error bars for the traditional stoves aresmaller than for other stove classes because there is nofuel use uncertainty for the traditional stoves (they areall assigned the value of 1 kg biomass). This suggeststhat even when comparing CO2 emissions the currentprotocols are often different from literature values,vary substantially from one another, and are likelysubstantially wrong for Traditional and charcoalstoves.

The 95% confidence intervals (CIs) in figure 1represents the distribution of means of different sam-ples. Conceptually, these intervals are for the ‘averagestove’within a class—something that does not exist inpractice, but is useful concept for writing an emissionsoffset protocol, or differentiating between stove clas-ses. In practice, testing protocols with a large N andprocedures that mimic conditions in the field wouldremove the need for stove class groupings or findingthe emissions profile of an ‘average’ stove of any class.[35] At present, however, the lack of individual stovemodel studies with sufficient statistics necessitates the

Figure 1.Emissions values assigned to different classes of cookstoves by theCleanDevelopmentMechanism (CDM), theAmericanCarbonRegistry (ACR), andGold Standard (GS). The protocol for theVerifiedCarbon Standard (VCS) is almost identical to theACRprotocol and omitted infigures for clarity; see the SI for details. The left three bars for each stove class show the emissions valuewhenthe internal protocol emissions factors are used (‘P’); the right three bars show the calculated emissions valuewhen literature-derivedvalues for emissions factors are used, with resampling as described in themethods section (‘L’). Distributions in the default protocol(‘P’) values arise from resampling of fuel use data, used to convert emissions to a per-task basis. Fuel use efficiency is included bynormalizing fuel use by improved stoves to that of the traditional stove in the same study, including uncertainty where such data exist.Horizontal lines within the boxes show themedian offset value, the boxes give the interquartile range of samplemeans (25th–75thpercentile) andwhiskers give the 95% confidence interval, defined by themean value +/−1.96 times the standard error of each samplemean.

6

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

class grouping. The distributions indicate that thereare stoves in the natural draft, gasifier, and particularlythe charcoal class that may perform worse than tradi-tional stoves under some conditions. See the SI figuresS5 and S6 formore details.

4.3. Emissions offsets by stove classBy subtracting the traditional stove emissions distribu-tion for each protocol from each stove class emissionsdistribution under the same protocol, we are able toderive the emissions offset that would be assigned toeach stove class, on a per-task basis. (This amountwould scale by the number of cooking tasks over theproject time period or cookstove lifetime.) In additionto calculating the offset for the existing protocols, wealso calculate the offset for our proposed (‘new’ or‘modified’) protocol that takes into account literature-derived emissions factors for BC, OC, and CO andtheir GWPs (see the SI for formulae). These results, bystove class, are shown in figure 2, for fNRB = 50%.Most clearly, a protocol that includes emissions ofPICs produces a higher median offset value for eachclass of stove. This is particularly notable for the forceddraft and gasifier stoves where including PICs morethan doubles the median offset, and is statisticallydifferent (at 95% from all existing protocols). Char-coal and natural draft stoves also see increased offsetsizes, with the new lower bound of the 95% CI locatednear the current GS offset estimate. For charcoalstoves, the increased efficiency and decreased BCemissions are partially offset by increased CO andCO2

emissions. For natural draft stoves increased efficiencyis responsible for the majority of the offset. Mostnotable here is that existing protocols fail to differenti-ate between cookstove classes (e.g., the CIs overlap forall classes within each protocol); the comprehensive

carbon accounting protocol creates significant separa-tion in carbon values between stove classes.

4.4. Renewable v. non-renewable biomassExisting offset protocols (aside from the GS) onlyassign offsets based on reductions in non-renewablebiomass. The assumption here is that emissions fromrenewably-harvested biomass, like dung, crop resi-dues, or sustainably-harvested wood, are sequesteredeach year in biomass growth. For example, the growthof corn stalks sequesters CO2 from the atmosphere;combustion of those corn stalks in cookstoves releasesCO2 back into the atmosphere, but there is no netemission or sequestration on an annual basis. Calcu-lated emissions offsets for each protocol and stoveclass, across fNRB are shown in figure 3. The ACR andCDM protocols give no offset credits when 100%renewable biomass is used (panel (c)). The GSprotocol assigns still credits the non-CO2 greenhousegas emissions from renewable biomass (CH4 andN2O), because those compounds—although pre-sented in CO2-equivalent terms in the final calculation—cannot actually be sequestered by biomass growth.

We compare these values across fNRB to our new(modified) protocol that accounts for the emission ofBC, OC andCO. As in the GS protocol, the new proto-col only takes fNRB into account for emissions of CO2,as CO, OC and BC cannot be sequestered by plants.Across stove type, these results show similar patternsto figure 2. (The results from figure 2 are shown in themiddle panel (b) of figure 3.) Again, forced draft stovesstand out as the stove class generating the largest off-sets with statistical certainty; in addition, the modifiedprotocol suggests significantly higher climate offsetvalues for forced draft and gasifier stoves when PICsare included. These results also suggest that there issignificant positive climate impact to be had through

Figure 2.Offset values relative to traditional stoves assigned to each cookstove class by the theAmericanCarbonRegistry (ACR),CleanDevelopmentMechanism (CDM),Gold Standard (GS), and the protocol proposed in this study (New). Values are shown for50%non-renewable biomass. Error bars show 95%confidence interval, defined by themean value+/− 1.96 times the standard errorof each samplemean, derived from resampling literature-derived distributions of emissions factors and fuel use for each stove class.

7

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

improved combustion of renewable biomass across allstoves. Also of note is that while, as a class, naturaldraft stoves outperform charcoal stoves at 100% non-renewable biomass, charcoal stoves generate largeroffsets at 0%non-renewable biomass.

4.5. Analysis of uncertaintyA key result of the above analysis is that the inclusionof BC, OC, and CO emissions raises the calculatedoffset value across stove classes. That is, on averagemost improved stoves reduce emissions of PICsrelative to traditional stoves, and that improvedcombustion, combined with fuel savings, results in anaverage climate benefit that is much larger than whenonly CO2 is considered. However, the literature-derived emissions factors for PICs have much largeruncertainties than the literature-derived emissionsfactors for CO2. The contributions of each pollutant toboth the overall magnitude of the new (modified)offset proposed here, and the uncertainty in that offset,are shown in the top panel of figure 4. Our derivedpoint estimates are in agreement with previous studiesthat have aggregated emissions data from previousstudies, both in that their estimates fall inside of our

95% CI, and in that the result of aggregating dataacross studies is a wide CI [37, 47].

BC is the largest contributor to both offset magni-tude and uncertainty, suggesting that from a carbonmarkets perspective, bettermeasurements of BC emis-sions factors for all stove types (including traditionalstoves) should be a priority for future research. (This isin agreement with previous research on the impact ofresidential combustion, e.g., [37].) However, becausesize of the uncertainty tends to match the size of thetotal offset (the coefficient of variance is similar foreach pollutant, with a few exceptions) there are oppor-tunities for large-n tests on specific stoves to reduce theerrors and generate a larger offset with strong statis-tical certainty.

Our resampling procedures assume that each indi-vidual cookstove test is sampled from the same popu-lation, an assumption that is strained when groupingstoves by class. However, whenwe look at the standarddeviation between the means of different tests, it tendsto be fairly close to the standard deviation within tests.This suggests that much of the error can be attributedto differences in individual stove performance acrosstests, and not large differences across models within a

Figure 3.Offset values (versus traditional stove) assigned to each improved cookstove class by the theAmericanCarbonRegistry(ACR), CleanDevelopmentMechanism (CDM), andGold Standard (GS), for existing protocols using literature-derived values (bluebars) and ourmodified protocol taking into account emissions of black carbon and carbonmonoxidewith literature-derivedemissions factors (green bars). Values shown for (a) 100%non-renewable biomass (inwhich all true CO2 emissions reductions arecredited); (b) 50%non-renewable biomass (inwhich half of all true CO2 emissions reductions are credited; the other half are assumedto be sequestered by the renewable fuel source); and (c) 0%non-renewable biomass (all CO2 emitted in combustion is assumed to besequestered again by the renewable fuel source; offset credit is only given for non-CO2 emissions). In thisfigure the error bars showthe 95% confidence interval, defined by themean value +/− 1.96 times the standard error of each samplemean, derived fromresampling literature-derived distributions of emissions factors and fuel use for each stove class: C= charcoal, FD= forced draft,G= gasifier, ND=natural draft.

8

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

stove class. As a case study, we show the estimated off-sets across protocols (and our modified protocol) forone particular natural draft stove (the Berkeley–Dar-fur stove) in the bottom part of figure 4. For this esti-mate, we draw on the one study with N 20⩾ for anindividual stove model and its traditional comparisonstove [42]. The error bars are much lower for CO2-based emissions due to the higher number of tests.And while the error bars are still very wide when BC,OC and CO are included, the small reduction in stan-dard error (due to higher N) results in a statisticallysignificant offset for the stove at 95% confidence.

Finally, we conducted our analysis using the cen-tral global GWP values for BC and OC from the IPCC

AR5 [24, 46]. However, uncertainty related to BC andOC GWP values could swamp the statistical uncer-tainty related to stove fuel use efficiency and emissionsfactors. To demonstrate the relative importance thesedifferent uncertainties, we simulated the total GWPfor 1 kg of wood fuel combusted in different condi-tions, with results presented in figure 5. Total GWP is afunction of combustion efficiency (the amount of car-bon in the wood fuel that is fully combusted to CO2),the optical properties of the remaining carbon emis-sions (here conceptualized as the EC:OC ratio, thoughthis could also be inferred from SSAmeasurements, asin [40, 48]), and GWP values for BC and OC. The reddots across the three panels—at representative values

Figure 4. (Upper panel) (a) Percentage of the total emissions offset value (CO2e) in the revised protocol attributable to each carbonspecies. (b) Percentage of the total uncertainty in the emissions offset value attributable to uncertainty in emissions of each species.(Lower panel) Current (blue bars) and revised (green bars, as proposed in this study) offset protocol values for the natural draft (ND)stoves studied in Preble et al (2014) [42]. This study conducted farmore tests on each stove than has been the practice (see table 1) andas a result the error bars from statistical resampling of the emissions factor distribution derived from the study result in statisticallysignificant offsets for this natural draft stove, unlike across the other studies (e.g., figure 3). However, the improved protocol whichtakes into account PICs (green bars) still has very large uncertainty, indicating the need for higher numbers of repetitions in cookstoveemissions tests, particularly if PICs are to be included, as this study argues they should.

9

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

of combustion efficiency and EC:OC—illustrate thatGWP uncertainty leads to estimates that vary by morethan two orders ofmagnitude.

5.Discussion

Encouraging widespread uptake of improved cookingtechnologies has been of sustained interest to thepublic health, REDD, climate, and development com-munities for several decades. And in the past few years,attention has coalesced on carbon markets as a meansof financing these technologies, which often remainout-of-reach economically for the households thatwould most benefit from them. This concept, whiletheoretically win–win, remains flawed at present, withimportant consequences. Here, we apply appropriatestatistical techniques to account for the uncertainty infuel savings and emissions factors of differentimproved cookstoves. Then, by calculating the offsetsthat would be generated for different classes of stovesin existing protocols, using these literature-derivedvalues for fuel use and emissions factors, we show thatexisting protocols underestimate the global warmingpotential ofmany types of improved cookstoves.

When we then incorporate PICs into a modifiedprotocol to account for combustion efficiency; we findthat that the simple biomass-to-CO2 calculation at theheart of current protocols further distorts the market.Existing protocols underestimate the climate impactsof the best technologies: accounting for PICs like BC,OC, and CO results in a sizable separation of averageoffsets across stove types, and the average climate ben-efit of improved cookstoves is higher than currentlyaccounted for due to reductions in emissions of BCand CO relative to traditional stoves. In particular,forced draft stoves stand out as the best technology,

generating more than double the climate benefits thatthey are currently assigned by any protocol when PICemissions are included. We also find that inclusion ofPICs implies a need to rethink the renewable v. non-renewable biomass distinction in current protocols.That is, reduction in PIC emissions should be valuedeven when renewable biomass is used. Inclusion ofPIC offset credits for renewable biomass use couldexpand the benefit of carbon financing to many morefamilies currently relying on agricultural residues anddung (considered renewable) as their fuel sources.

We find that uncertainty in emissions factors ofPICsmeans that certain stove classes have large uncer-tainties associated with their calculated offsets, andwhen stoves are considered individually they oftengenerate offsets statistically indistinguishable fromzero. This finding points to the need—acknowledgedelsewhere in the theoretical literature, but not yet thenorm in experimental literature—for more rigorousstatistics in emissions testing. Ideally, 30–50 testswould be conducted on each stove model to suffi-ciently narrow the variance on emissions factors forestimation of offsets. Furthermore, testing needs tomove away from simple laboratory tests and towardsfield testing that emulates the conditions in which astove will actually be used. This includes using morerealistic protocols like the KPT, testing at a range ofthermal powers, using different fuels with varyingmoisture contents, and using different pots and acces-sories for different tasks. Only when this more rigor-ous testing is the norm will offset protocols accuratelyreflect climate benefits and incentivize the besttechnologies.

Finally, a comprehensive carbonmarket that accu-rately values the full radiative forcing impacts of tech-nology changes (beyond just CO2) will also necessitatestandardization of emissions testing protocols to

Figure 5. Simulation illustrating the total global warming potential (GWP) of 1 kg ofwood fuel as a function of combustion efficiency(here represented as the fraction of carbon combusted toCO2) and the ratio of elemental carbon (ECor BC) to organic carbon (OC)of PMemissions. Simulation is shown for three different sets of GWPvalues for EC andOC: the low(GWP 100, GWP 92BC OC= = − ), central (GWP 900, GWP 46BC OC= = − ), and high (GWP 1700, GWP 18BC OC= = − ) valuesfrom the IPCCfifth assessment report [24, 46]. The simulation assumes that 50%of themass of thewood fuel is carbon, and that anycarbon not directly combusted toCO2 is emitted as either elemental or organic carbon PM (we ignore other pathways like CO forsimplicity). The red dots indicate the range of calculated climate impacts across GWPvalues for a combustion efficiency of 0.95(corresponding to theCO2 emissions factor used in theCDMprotocol), and an EC:OC ratio of 0.33 (which lies in the range ofmany ofthe stove classes).

10

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

include particulate and gas-phase measurements.Equally important, a comprehensive carbon marketwill require agreement on GWP values to be used forPICs. Asmentioned above, the GWP ranges for carbo-naceous aerosols are large, reflecting the rapidly evol-ving science of short-lived climate pollutant impacts.

Having accurate and more precise estimates of thetrue climate impacts of improved cookstove technolo-gies is critical for making the carbon market work asintended; to accomplish these goals, inclusion of PICsin carbon accounting standards and a new paradigmfor rigorous testing are needed. More broadly, the lackof uncertainty estimation and lack of fidelity to trueGWP estimates for cookstove emissionsmay be dama-ging the ability of cookstove projects to generate off-set-funded implementation. The collapse of CERprices in the CDMwas a result of violation of the prin-ciple of additionality and an over-estimation of theoffsets generated by projects [49–51]. As the numberof non-credible carbon credits generated outpaceddemand, prices collapsed to less than $1 per ton ofCO2, where they are expected to stay through about2020. Improving estimationmethods by correctly esti-mating uncertainty should prevent a similar collapsein VER prices for cookstove offsets, while ensuringthat offsets take into account all of the climate forcingagents emitted should improve the credibility and sizeof cookstove-generated carbon offsets.

Acknowledgments

This work was supported by a Robertson SummerResearch Fellowship through UC San Diego’s School ofGlobal Policy and Strategy and theQualcommInstitute.

References

[1] World BankWorldDevelopment Indicators (WDIOnline)(http://data.worldbank.org/data-catalog/world-development-indicators)

[2] Lim S S et al 2013 Lancet 380 2224[3] Joshi V, VenkataramanC andAhujaDR 1989Environ.

Manage. 13 763[4] Ballard-TremeerGand JawurekH1996BiomassBioenergy11419[5] EzzatiM,Mbinda BMandKammenDM2000Environ. Sci.

Technol. 34 578[6] Berrueta VM, Edwards RD andMaseraOR 2008Renew.

Energy 33 859[7] Jetter J J andKariher P 2009Biomass Bioenergy 33 294[8] Adkins E, Tyler E,Wang J, Siriri D andModi V 2010Energy

Sustainable Dev. 14 172[9] SmithKR et al 2011 Lancet 378 1717[10] Yu F 2011Environ. Resour. Econ. 50 495[11] Burwen J and LevineD I 2012Energy Sustainable Dev. 16 328[12] Kar A et al 2012Environ. Sci. Technol. 46 2993[13] HannaR,DufloE andGreenstoneM2012Up in smoke: the

influence of household behavior on the long-run impact ofimproved cooking stovesTech. Rep.National Bureau ofEconomic Research (doi:10.2139/ssrn.2039004)

[14] Mobarak AM,Dwivedi P, Bailis R,Hildemann L andMiller G2012Proc. Natl Acad. Sci. USA 109 10815

[15] Lewis J J andPattanayakSK2012Environ.HealthPerspect.120637[16] JeulandMAandPattanayak SK 2012PLoSOne 7 e30338[17] RodenCA, BondTC, Conway S andPinel A BO2006

Environ. Sci. Technol. 40 6750[18] Bailis R, CowanA, Berrueta V andMaseraO 2009WorldDev.

37 1694[19] Rehman IH, AhmedT, Praveen P S, Kar A andRamanathanV

2011Atmos. Chem. Phys. 11 7289[20] JeulandM, Pattanayak SK, SooT and Sheng J 2014 Preference

heterogeneity and adoption of environmental healthimprovements: evidence from cookstove promotionexperiment SRRN (doi:10.2139/ssrn.2490530)

[21] MartinW J, Glass R I, Balbus JM andCollins F S 2011 Science334 180

[22] BhojvaidV, JeulandM,Kar A, Lewis J, Pattanayak S,RamanathanN, RamanathanV andRehman I 2014 Int. J.Environ. Res. Public Health 11 1341

[23] Lee CM,Chandler C, LazarusMand Johnson FX 2013Assessing the climate impacts of cookstove projects: issues inemissions accountingChallenges in Sustainability 1

[24] MyhreG et al 2013Anthropogenic andNatural RadiativeForcing (Cambridge: CambridgeUniversity Press) pp 659–740book section 8 (www.ipcc.ch/pdf/assessment-report/ar5/wg1/WG1AR5_Chapter08_FINAL.pdf)

[25] UNEP 2011 Integrated assessment of black carbon andtropospheric ozone: summary for decisionmakersTech. Rep.UnitedNations Environment Programme andWorldMeteorological Organization (www.unep.org/dewa/Portals/67/pdf/BlackCarbon_SDM.pdf)

[26] UNFCCC 2012AMS II.G.: energy efficiencymeasures inthermal applications of non-renewable biomassTech. Rep.UnitedNations FrameworkConvention onClimate Change,Executive Board of theCleanDevelopmentMechanismBonn,Germany (http://cdm.unfccc.int/methodologies/documentation/meth_booklet.pdf)

[27] VerifiedCarbon Standard 2010 Fuel Switch to renewablebiomass for thermal applicationsTech. Rep.Washington, DC,USAVerifiedCarbon StandardMethodologyVM00XX,version 1.0 (www.v-c-s.org/methodologies/methodology-fuel-switch-renewable-biomass-thermal-applications)

[28] AmericanCarbonRegistry 2011 Switch fromnon- renewablebiomass for thermal applications by the userTech. Rep.AmericanCarbonRegistry Arlington, VA,USA (http://americancarbonregistry.org/carbon-accounting/standards-methodologies/switch-from-non-renewable-biomass-for-thermal-applications)

[29] Gold Standard 2011Technologies and practices to displacedecentralized thermal energy consumptionTech. Rep.TheGold StandardGeneva, Switzerland (www.goldstandard.org/wp-content/uploads/2011/10/GS_110411_TPDDTEC_Methodology.pdf)

[30] Peters-StanleyMandYinD2013Maneuvering theMosaic:State of the Voluntary CarbonMarkets 2013ExecutiveSummary Forest Trends’EcosystemMarketplace andBloombergNewEnergy Finance (www.forest-trends.org/documents/files/doc_3898.pdf)

[31] Bialis R 2014Controlled Cooking Test version 2.0 (Washington,DC:Global Alliance for CleanCookstoves) (http://cleancookstoves.org/binary-data/DOCUMENT/file/000/000/80-1.pdf)

[32] Bailis R, SmithKRandEdwardsR2007Kitchen PerformanceTest version 3.0, kpt-3-0-protocol-english (Washington,D.C:Global Alliance forCleanCookstoves) (http://cleancookstoves.org/binary-data/DOCUMENT/file/000/000/83-1.pdf)

[33] Water BoilingTest2014 version 4.2.3 (Washington,DC:GlobalAlliance forCleanCookstoves) (http://cleancookstoves.org/binary-data/DOCUMENT/file/000/000/399-1.pdf)

[34] JohnsonM, Edwards R andMaseraO2010 Improved stoveprograms need robustmethods to estimate carbon offsetsClimChange 102 641–9

[35] Wang Y, SohnMD,WangY, LaskKM,Kirchstetter TWandGadgil A J 2014Energy Sustainable Dev. 20 21

11

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al

[36] SmithKR,UmaR,KishoreV, LataK, JoshiV,Zhang J,RasmussenR andKhalilM2000Greenhouse gases fromsmall-scale combustiondevices indeveloping countries: phase iiahousehold stoves in IndiaTech. Rep.EPA-600/R-00-052UnitedStates Environmental ProtectionAgency (http://ehsdiv.sph.berkeley.edu/krsmith/publications/00_smith_3.pdf)

[37] BondT,VenkataramanC andMaseraO2004EnergySustainable Dev. 8 20

[38] JacobsonMZ2001Nature 409 695[39] Kirchstetter TW2004 J. Geophys. Res. 109D21208[40] Bahadur R, Praveen P S, XuY andRamanathanV 2012Proc.

Natl Acad. Sci. USA 109 17366[41] Saleh R et al 2014Nat. Geosci. 7 647[42] Preble CV,HadleyOL,Gadgil A J andKirchstetter TW2014

Environ. Sci. Technol. 48 6484[43] BondTCand SunH2005Environ. Sci. Technol. 39 5921[44] RamanathanV andCarmichael GApr 2008Nat. Geosci. 1 221[45] Menon S, KochD, BeigG, Sahu S, Fasullo J andOrlikowskiD

May 2010Atmos. Chem. Phys. 10 4559[46] BondTC, Zarzycki C, FlannerMGandKochDM2011

Atmos. Chem. Phys. 11 1505[47] GrieshopAP,Marshall J D andKandlikarM2011Energy

Policy 39 7530

[48] Cazorla A, Bahadur R, Suski K J, Cahill J F, ChandD,Schmid B, RamanathanV and Prather KA 2013Atmos. Chem.Phys. 13 9337

[49] Kachi A,TänzlerD andSterkW2013The clean developmentmechanismand emergingoffset schemes: options forreconciliation?UFOPLAN-RefNo. 3711 41507EnvironmentalResearch of the FederalMinistry for the Environment,NatureConservation andNuclear Safety,Germany (www.bmub.bund.de/fileadmin/Daten_BMU/Pools/Forschungsdatenbank/fkz_3711_41_507_cdm_offsetting_mechanismen_en_bf.pdf)

[50] Zhang J andWangC2011 J. Environ. Econ.Manage. 62 140[51] Watts D, AlbornozC andWatsonA2015Renew. Sustainable

Energy Rev. 41 1176[52] VenkataramanC andRaoGUM2001Environ. Sci. Technol.

35 2100[53] JohnsonM, Edwards R, Alatorre FrenkC andMaseraO 2008

Atmos. Environ. 42 1206[54] MacCartyN,OgleD, Still D, BondT andRodenC2008Energy

Sustainable Dev. 12 56[55] MacCartyN, Still D andOgleD 2010Energy Sustainable Dev.

14 161[56] Jetter J, Zhao Y, SmithKR, KhanB, Yelverton T,

DeCarlo P andHaysMD2012Environ. Sci. Technol. 46 10827

12

Environ. Res. Lett. 10 (2015) 084026 L Sanford et al