Urban and rural energy use and carbon dioxide emissions in Asia

    Urban and rural energy use and carbon dioxide emissions in Asia Volker Krey, Brian C. O’Neill, Bas van Ruijven, Vaibhav Chaturvedi, Vassilis Daioglou, Jiyong Eom, Leiwen Jiang, Yu Nagai, Shonali Pachauri, Xiaolin Ren PII: DOI: Reference:

S0140-9883(12)00090-4 doi: 10.1016/j.eneco.2012.04.013 ENEECO 2296

To appear in:

Energy Economics

Received date: Revised date: Accepted date:

2 August 2011 4 April 2012 23 April 2012

Please cite this article as: Krey, Volker, O’Neill, Brian C., van Ruijven, Bas, Chaturvedi, Vaibhav, Daioglou, Vassilis, Eom, Jiyong, Jiang, Leiwen, Nagai, Yu, Pachauri, Shonali, Ren, Xiaolin, Urban and rural energy use and carbon dioxide emissions in Asia, Energy Economics (2012), doi: 10.1016/j.eneco.2012.04.013

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Urban and rural energy use and carbon dioxide emissions in Asia

PT

Volker Krey1, Brian C. O’Neill,2 Bas van Ruijven2,3, Vaibhav Chaturvedi4, Vassilis Daioglou3,5, Jiyong Eom4, Leiwen Jiang,2 Yu Nagai1, Shonali Pachauri1, Xiaolin Ren2 International Institute for Applied Systems Analysis (IIASA), Schlossplatz 1, 2361 Laxenburg, Austria

2

National Center for Atmospheric Research (NCAR), PO Box 3000, Boulder, CO, 80305, USA.

RI

1

3

SC

PBL - Netherlands Environmental Assessment Agency (PBL), PO Box 1, 3720 BA Bilthoven, The Netherlands

4

NU

Joint Global Change Research Institute, 5825 University Research Court Suite 3500, College Park, MD 20740, USA

5

AC CE P

TE

D

MA

Utrecht University, Copernicus Institute for Sustainable Development, Budapestlaan 6, 3584 CD Utrecht, The Netherlands

ACCEPTED MANUSCRIPT Abstract

MA

NU

SC

RI

PT

The process of urbanization has been shown to be important for economic development, environmental impacts and human wellbeing, particularly in developing countries. In this paper we compare structure, data sources and scenario results of four integrated assessment models that are capable of analyzing different aspects of urbanization. The comparison focuses on residential sector energy use and related CO2 emissions based on a set of urbanization scenarios for China and India. Important insights from this model comparison include that (i) total fossil fuel and industrial CO2 emissions at the regional level are not very sensitive to alternative rates of urbanization and are largely dependent on the linkage between urbanization and economic growth via differentiated labor productivity in urban and rural areas, (ii) alternative urbanization pathways may yield different results for the share of solid fuels in residential energy use, thereby affecting the number of people relying on these fuels and the associated adverse health impacts, and (iii) alternative economic growth scenarios can only be assessed for their welfare implications if urban and rural household are distinguished, even though that distinction does not always strongly affect aggregate outcomes which is often due to two effects that compensate each other in total. It can be concluded that urbanization and heterogeneity of households and consumers are clearly relevant for distributional effects and associated health and social impacts.

TE

We compare four IA models that represent different aspects of urbanization. Total CO2 emissions are not very sensitive to alternative urbanization rates. The linkage between urbanization and economic growth affects total CO2 emissions. Urbanization affects the traditional solid fuel use in the residential sector. Urbanization and income of households are relevant for health and social impacts.

AC CE P

• • • • •

D

Highlights

ACCEPTED MANUSCRIPT

1. Introduction

SC

RI

PT

Differences between urban and rural areas, and the process of urbanization, are important for economic development, environmental pressure and human wellbeing, especially in developing countries. The economic structure and income levels of urban and rural areas are different, household behavior and resource use diverge, and exposure of people to indoor air pollution from traditional fuel use differs.

AC CE P

TE

D

MA

NU

Various approaches exist to analyze the environmental and economic impacts of urban/rural differences and the process of urbanization. A first approach is analysis of household-level historical data on determinants of energy use, yielding insights into the relative importance of economic, social, and technological drivers of changes in energy use that occur in rural and urban households (Barnes et al., 2005; Jiang and O'Neill, 2004; Pachauri, 2007; Pachauri and Jiang, 2008). Another approach is a static analysis of current differences in energy use (and implied emissions) in urban and rural households that is often used in constructing emission inventories (Dodman, 2011; Ramaswami et al., 2008). A third approach is dynamic analysis of potential future developments over time, using dynamic models. This paper focuses on the inclusion of urban/rural differentiation and the process of urbanization in a subset of dynamic models: integrated assessment models (IAMs). In the tradition of integrated assessment modeling, scenarios of future energy use and emissions have been based on dynamics of large (national to continental scale) world regions (van Ruijven et al., 2008). Within these regions, the only further differentiation in models was to macro-economic sectors. In recent years, however, increasing effort has been put into more detailed differentiation and downscaling within regions. One aspect of this differentiation is to distinguish urban and rural areas and income classes of households. The differentiation is mostly applied to residential sector energy use, though in some cases it is a component of a simulation of the full macro-economic system. This overview paper of the urbanization subgroup of the Asian Modeling Exercise (Calvin and et al., 2012), we believe for the first time, systematically compares the development of residential sector energy use, related CO2 emissions and macro-economic implications from several IAMs based on a set of urbanization scenarios for China and India. The comparison includes four models with a differentiation of urban and rural households: MESSAGE-Access (Ekholm et al., 2010; Pachauri et al., in preparation; Riahi et al., forthcoming; van Vliet et al., 2012), IMAGE-REMG (Daioglou et al., 2011; van Ruijven et al., 2011a, b; van Ruijven et al., 2012), GCAM (Chaturvedi et al., submitted; Eom et al., submitted) and iPETS (O'Neill et al., 2010). With such detailed comparison of a limited number of models, it is important to note that these models are initially designed for different purposes, and therefore represent the different processes that are related to urbanization to varying degrees. There are several motivations for differentiating between urban and rural areas and including the process of urbanization in modeling energy use and greenhouse gas emissions. First, there may be aggregate effects of urbanization on national or total energy use and emissions. For example, differences in behavior between urban and rural households may lead to different future emission pathways with high or low urbanization levels. A second motivation for including urbanization is that it may improve the representation of key processes in integrated assessment or macro-economic models. This is especially valid for developing countries, where differences between urban and rural areas are large and urbanization is closely

ACCEPTED MANUSCRIPT linked to different development pathways. In macro-economic models, differences in labor productivity between urban and rural areas and the income effects of urbanization influence urban and rural consumption patterns. A key process is that fuel choices in urban and rural households tend to be rather different.

SC

RI

PT

A third motivation is that it allows deriving disaggregated results for urban and rural households, such as distribution effects across rural and urban households in baseline or policy scenarios. In addition, a broader set of indicators with respect to energy poverty, welfare and health impacts of different policy measures can be calculated. Finally, it allows for the option to link IAM scenarios to vulnerability and adaptive capacity with respect to the impacts of global environmental changes.

MA

NU

To compare the models and their results in-depth, we first present the basic structure of the four IAMs in terms of how they incorporate urbanization (Section 2). Section 3 introduces the study design, discusses the data sources that are used to calibrate the different models and compares socio-economic drivers. The results of future energy use in the central urbanization scenarios are presented in Section 4; Section 5 explores the impact of urbanization by comparing high and low urbanization scenarios between the different models. Finally, Section 6 discusses the findings and limitations of this study and identifies several questions worth further research.

TE

2.1. Overview

D

2. Model structure comparison

AC CE P

The models employed in this scenario comparison vary significantly in their structures, a consequence in part of the different motivations for incorporating urban/rural distinctions by the respective modeling teams. These structural differences are important for understanding the differences in the outcomes of the scenarios, and for interpreting their significance. Urbanization can act, in broad terms, through similar channels across models, but there are also important differences. Broadly speaking (see Figure 1), urbanization can affect energy use and emissions through three channels: direct influences on the preferences of households for energy (or other) goods consumed; influences on income, which indirectly affects the level or composition of consumption; or influences on energy supply infrastructure, in particular electricity access, which also indirectly affects consumption. These consumption effects in turn influence the types and quantities of fuels used in energy production, and resulting emissions. Figure 1: Schematic influence of urbanization in models used in this study. Differences in basic features of model structure already imply differences in the potential for urbanization to act through these channels in particular models. For example, three of the models – GCAM, IMAGE, and MESSAGE1 – are partial equilibrium, taking GDP as an exogenous assumption and from it deriving energy demand, with the energy supply mix determined endogenously. The fourth model, iPETS, is general equilibrium, so that GDP, energy demand, and supply are all endogenous, determined mainly by exogenous assumptions about future rates of labor productivity improvement 1

MESSAGE is coupled to the macro-economic model MACRO (Messner and Schrattenholzer, 2000) which allows estimating general equilibrium effects. However, to date this linkage is not established with the MESSAGE-Access model that has been used in this study.

ACCEPTED MANUSCRIPT and technological change in energy and other inputs to production. These differences alone imply that models in which GDP is exogenous do not have the possibility for urbanization to affect economic growth (and therefore income), ruling out the total income channel as a source of influence.

SC

RI

PT

Similarly, the three partial equilibrium models all include substantial detail in energy infrastructure and supply technologies, while the iPETS model does not. This implies that the iPETS model cannot explicitly represent factors such as changes in access to electricity or other fuels that are typical of the urbanization process. While such an effect could in principle be represented implicitly within the iPETS model, the treatment in the current analysis is limited, making the infrastructure channel potentially important in GCAM, IMAGE, and MESSAGE, but not in iPETS.2

AC CE P

TE

D

MA

NU

In addition, models differ in how urban and rural households are distinguished. For example, in GCAM, IMAGE and MESSAGE, urban and rural households are distinct categories that make separate decisions about consumption. As a consequence, not only can urbanization potentially affect aggregate consumption patterns for a model region, but consumption outcomes for both household types are known. In contrast, iPETS has only a single representative household in each region. Data on urban and rural households are used to calculate parameter values for the representative household that reflect the average values across the whole population. As urbanization proceeds, these parameter values are changed to reflect the changing composition of the population; in other words, the representative household becomes more urban as the underlying distribution of the population does so. As a consequence, while urbanization can affect aggregate outcomes associated with the representative household, outcomes are not available separately for urban and rural households. The IMAGE and MESSAGE models can be used to investigate scenario consequences for, for example, the energy use of low income rural households, a possibility not currently available in iPETS. GCAM also makes a distinction between rural and urban households, however further categorization on the basis of income distribution within rural and urban households is not made. Next we describe in more detail differences across models in the influence of urbanization through these three channels, as determined by model structures.

2.2. Income effects

Urbanization can potentially affect the scale of income, its distribution across households, or both. As discussed above, the iPETS model is the only one in which urbanization has a scale effect on income. Urbanization affects economic growth in this model (and therefore the scale of income) in two ways: through labor supply and through savings rates. Labor supply is exogenous, determined by combining projections of the future mix of household types (urban/rural, age, and household size) with per capita labor supply by household type estimated from current per capita labor income in household survey data (Zigova et al., 2009). Typically, urban households have a larger per capita labor supply than rural households have, and therefore urbanization leads to faster growth in labor supply and GDP (all else equal). The savings rate is endogenous to the model, and is affected by a number of factors, including future prices (since the model is forward looking), the return on capital,

2

The iPETS baseline scenario is tuned to approximate a MESSAGE implementation of the B2 scenario (Riahi et al., 2007), and in this way implicitly captures infrastructure constraints present in MESSAGE. However, changes in urbanization in iPETS do not lead to changes in the implicit representation of infrastructure.

ACCEPTED MANUSCRIPT and future demographic change. Urbanization, by affecting both demand and supply, affects prices and therefore savings rates. Because it is equal to investment, savings ultimately affect GDP growth.

2.3. Direct consumption effects

MA

NU

SC

RI

PT

In GCAM, IMAGE, and MESSAGE, urbanization does not affect economic growth, but in IMAGE and MESSAGE it does affect the distribution of income across households. In these two models, the population is divided not only into rural and urban, but further into different income classes, with the number of people in each category updated over time according to exogenous population and income growth assumptions (see Section 4.4). Although within urban and rural categories the income distribution is unaffected by urbanization assumptions, the aggregate (urban/rural) distribution for a region is influenced by shifts in the population toward urban areas. In GCAM, separate groups of income are not explicitly represented, but average income differs in rural and urban areas and the ever increasing gap between these averages starts to decrease, broadly consistent with the Kuznets’ inverted-U hypothesis (Kuznets, 1955; Qin and Zhou, 2009). As income increases in GCAM, the affordability of various energy services increases which leads to increased penetration of end use technologies for meeting the increased energy service demands both in urban and rural areas.

AC CE P

TE

D

In all models, consumption of different types of goods is influenced by both income and the urban/rural status of households. The way in which this influence is achieved and the types of goods represented, however, differ significantly. In iPETS, urbanization affects consumption preferences. Preferences here are defined as parameters in the household utility function (and therefore in the demand function) that indicate the share of expenditures on each good that would maximize utility if prices remained constant. Expenditure shares are calculated for each household type for the base year from household survey data (Zigova et al., 2009). The preference parameters for the single representative household are then based on the average expenditure shares across urban and rural households within the region. Furthermore, expenditure shares within each household type are a function of income. Thus both income growth and urbanization lead to changes in preferences of the representative household. For example, with urbanization, and with income growth, the preference for consumption of traditional biomass declines (O’Neill et al., 2012). In GCAM, IMAGE, and MESSAGE, demand for useful energy by households is calculated separately for household types defined by urban/rural status, and in IMAGE and MESSAGE further disaggregated by income level. In GCAM, household energy service demand is a function of technical parameters (e.g., building shell efficiency, thermal conductance, etc.) as well as economic factors represented by the ratio of average income to energy service cost within rural and urban households separately. An important characteristic of household energy service demand is satiation, which implies that energy service demand per unit of floor space becomes increasingly less responsive to changes in price and income as income grows. Future expansion of each energy service depends on the preference for the service, as reflected by the base-year demand relative to its affordability. Penetration of technologies within each energy service depends on the base year share of the given technology, capital and operation cost, efficiency of the technology, the fuel cost for providing service with the technology, and, in the case of traditional fuel use, average per capita income to account for increasing time costs associated with collecting traditional biomass.

ACCEPTED MANUSCRIPT

NU

SC

RI

PT

In IMAGE and MESSAGE, demand depends on a number of factors including current energy use patterns from household survey data, assumptions about end use efficiency, and, in the IMAGE model, additional factors such as population density, household size, and temperature. In both models, service demands are satisfied by considering costs of various supply options. The supply mix is determined based on fuel costs, end use technology costs, income dependent consumer discount rates and perceived costs, which capture inconvenience costs as well as other factors such as habits and cultural factors, infrastructure, environmental factors and convenience. The fuel choice formulation is an important difference between the two models: MESSAGE prefers a single fuel per population group and initiates a transition towards this preferable fuel, whereas IMAGE allocates market shares to multiple fuels based on relative cost differences. Thus in both models urbanization can affect both useful energy demand and the manner in which this demand is met. For example, higher income households have higher inconvenience (or perceived) costs for traditional fuels and lower discount rates, which make larger upfront capital investments less of a barrier.

AC CE P

TE

D

MA

In addition to the approach to modeling household consumption, the number and types of goods or services included varies as well. The iPETS model includes all consumption goods, aggregated into six categories: three energy goods (electricity, coal/biomass, other energy), food, transportation, and everything else. GCAM, IMAGE, and MESSAGE include only energy goods or services but with greater disaggregation. IMAGE covers six energy service demands (cooking, space heating, space cooling, appliances, water heating, lighting), GCAM treats the demand for cooking and water heating jointly and also covers the remaining four services of IMAGE, while MESSAGE includes demands for electricity and the services it provides in one aggregated category as well as cooking3. Demand is satisfied with choices across seven fuels in GCAM (coal, traditional biomass, modern biomass, secondary gas, oil, electricity, and district heat4), seven fuels in IMAGE (coal, traditional biomass, kerosene, LPG, natural gas, secondary heat and electricity) and five in MESSAGE (coal, biomass, kerosene, LPG and electricity).

2.4. Infrastructure effects

In GCAM, IMAGE and MESSAGE, infrastructure features of the energy supply system are differentiated between rural and urban areas within model regions. In some of these models urbanization plays a role in driving these differences, in other models urbanization influences energy use outcomes because differences in infrastructure condition consumption choices made by rural vs. urban households. For example in MESSAGE-Access, a cost for expanding the transmission and distribution of electricity is included in rural areas of three developing country regions (South Asia or India+, Other Developing Asia, Sub-Saharan Africa, but not China+) where access is not universal. These costs then affect the use of electricity to meet demand in these areas. Electricity access is also differentiated between rural and urban areas in the IMAGE model, where access is modeled as a function of GDP per capita, urbanization and rural population density. In both IMAGE and MESSAGE urbanization can shift a larger proportion of the population into areas in which access is not limited, which will influence demand for specific fuels. In GCAM, rural and urban supply choices in China+ differ in terms of two fuels: only urban areas have access to district heat, and only rural areas have access to traditional biomass. In contrast, in the iPETS model, and in the India+ region of GCAM, there is no differentiation of supply features between rural and urban areas. 3 4

MESSAGE-Access only covers the India region at present where heating demand is negligible. District heat is only modeled for China, since India has a hot climate with negligible heating demand.

ACCEPTED MANUSCRIPT

IMAGE

GCAM

iPETS

Population

(UN, 2009)

(UN, 2011; World Bank, 2009)

(UN, 2005)

(UN, 2004), Census data

Urbanization

(UN, 2010b)

(CIESIN et al., 2004; UN, 2010b)

(UN, 2007)

GDP

(CIA, 2009; World Bank, 2009)

(UN, 2010a; World Bank, 2009)

(World Bank, 2009)

Economic structure

N/A

N/A

U/R GDP China+

N/A

U/R GDP India+

RI

SC

GTAP7 (Narayanan and Walmsley, 2008)

(World Bank, 2008)

(Government of China, 2010)

N/A

updated from (Grubler et al., 2007)

(World Bank, 2008)

PNNL Assumptions

Total energy use

(IEA, 2010b)

(IEA, 2010b)

(IEA, 2007)

(IEA, 2010b)

U/R energy China+

N/A

(LBL, 2008)

(LBL, 2008)

(NBS, 2004a, b)

U/R energy India+

(NSSO, 2007b)

(NSSO, 2007a)

(NSSO, 2007a)

(NSSO, 2007b) (expenditures)

Population

(UN, 2009)

(UN, 2011)

(UN, 2005)

(UN, 2004)

Urbanization

(Grubler et al., forthcoming; UN, 2010b)

(OECD, 2011)

(Government of China, 2010; UN, 2010b)

(Jiang and O’Neill, in preparation) (India/China), (Grubler et al., 2007; Jiang and O’Neill, 2009) (other regions)

(Riahi et al., forthcoming)

(OECD, 2011)

(Clarke et al., 2008)

(EIA, 2009) (through 2035), IIASA B2 (Riahi et al., 2007)

All residential energy + electrification

All residential energy

Labor, Savings, Consumption

GDP

U/R split

TE

MA

NU

GTAP7 (Narayanan and Walmsley, 2008)

D

Model structure

(UN, 2004), Census data

N/A

Key element of

Variable type

PT

MESSAGE

AC CE P

Scenario projections

Data sources

Table 1: Data sources, model structure and variable types in the models that deal with urbanization and are compared in this study

Cooking fuel

+ electrification

+

N/A

Income distribution

U/R + quantiles

U/R + quantiles

U/R average

Not explicitly modeled

Population

Exogenous

Exogenous

Exogenous

Exogenous

Urbanization

Exogenous

Exogenous

Exogenous*

Exogenous

GDP

Exogenous

Exogenous

Exogenous

Endogenous

Income distribution

Exogenous

Exogenous

Exogenous

Not explicitly modeled

Electrification

Exogenous

Exogenous*

Endogenous

Not explicitly modeled

Floor space

N/A

Exogenous*

Endogenous

N/A

Energy cost

Endogenous

Endogenous

Endogenous

Endogenous

Fuel choice

Endogenous

Endogenous

Endogenous

Endogenous

+

*Linked to GDP/capita; Based on The Government of India (2010), McKinsey Global Institute (2010), and UN Urbanization Prospects (2007)

ACCEPTED MANUSCRIPT 3. Modeling protocol and socio-economic drivers

RI

PT

This section and the following present the input assumptions and results of a partly harmonized model comparison for the China+ and India+ regions. First, Section 3.1 provides an overview of the modeling protocol, and the following Section 3.2 compares the socio-economic drivers of energy use and related CO2 emissions. In this context, we will also give an overview of the different data sources used for calibrating the models (Table 1) and discuss the issue of comparability across models due to differences in definitions.

SC

3.1. Modeling protocol

TE

D

MA

NU

In the result comparison we will concentrate on baseline scenarios that do not include specific policies to deal with climate change mitigation or other energy challenges such as energy security or providing access to modern energy services to the poor. As a result, the differences in the development of urban and rural energy use and its impact on economy-wide energy use and associated CO2 emissions are due to differences in model structure (Section 2) and assumptions about scenario drivers (this Section). The rates of urbanization in the two Asian regions are (exogenously) varied to explore implications for energy use and emissions in the future. In addition to the central urbanization trajectory in China+ and India+, two alternative urbanization pathways are analyzed that reach an urbanization rate in 2050 which is 10%-points higher or lower than the central assumption with a roughly linear increase over time (Figure 3).

AC CE P

Whereas the variation of urbanization rates and the policy setup are harmonized, we do not harmonize other socio-economic variables, such as total population, economic development and even the central urbanization assumptions. However, when correcting for differences in regional definitions, it turns out that population growth and standard urbanization trends are fairly close across the participating models while economic growth shows partly diverging trends. The implication of substantial differences in GDP growth rates are explored in a sensitivity analysis in which two models adjust their trends to be roughly compatible with the other two.

3.2. Socio-economic Drivers

The development trends of total population in both China+ and India+ are very similar across the different models and are largely compatible with the most recent UN medium projection (UN, 2011) (see Figure 2). As all models rely on different revisions of UN population data and projections (see Table 1), with IMAGE and iPETS using additional data from the World Bank and censuses, there is also relatively little variation in the base year data (see online supplementary material, Table A.1). For India+, MESSAGE has a deviating regional definition (South Asia5), and variations are very limited between the other models that specifically distinguish India. Population data for China+ vary over a wider range. Also here, regional definitions play a role, as both IMAGE6 and GCAM7 contain a (different) China+ region. In China+, population peaks around 2030 to 2040 and by 2050 reaches about the same level as today. Population is identical across the alternative urbanization pathways 5

The MESSAGE South Asia (India+) region includes Afghanistan, Bangladesh, Bhutan, India, the Maldives, Nepal, Pakistan and Sri Lanka 6 The IMAGE China+ region includes China, Mongolia, Taiwan, Hong Kong and Macao 7 The GCAM China+ region includes China, Mongolia, Cambodia, DPR Korea and Viet Nam

ACCEPTED MANUSCRIPT for all models. To avoid the different regional definitions distorting the comparison, we will mostly use shares and per capita quantities in the discussion of assumptions and results throughout the remainder of this paper.

PT

Figure 2: Development of total population in China+ (left panel) and India+ (right panel) between 1950 until 2050 in different model projections (historical data from UN (2011)). Note that differences in the base year result from differences in regional definitions.

TE

D

MA

NU

SC

RI

Urbanization levels in the base year are very similar across the different models, for both China+ and India+, largely due to the use of similar data sources, namely, the UN database on population defined by urban/rural residence (UN, 2010b). The definition of urban in the UN data follows definitions used by statistical offices in each country. This practice can lead to two potential inconsistencies. First, changes in the definition of urban within a given country over time can lead to spurious urbanization trends. However, the UN adjusts data within countries, wherever possible, to conform to a single consistent definition. Second, differences in definitions between countries lead to inconsistencies in what is considered urban across the UN dataset. This inconsistency is not corrected for in the UN data. For example, the definition of urban in China includes areas classified as cities or towns, plus nearby villages with certain types of infrastructure present. In contrast, the definition in India includes areas within administrative boundaries plus those that meet a combination of criteria based on total population, population density, and proportion of the labor force in non-agricultural sectors. These differences present an obstacle to drawing conclusions based on cross-country analysis. However in this paper, we focus on differences between urbanization scenarios within each country; thus our main conclusions are not affected by the inconsistencies in the UN data.

AC CE P

The development of urbanization shares over time as shown in Figure 3 follows very similar trends in the central cases across models despite no efforts of harmonizing urbanization. It is worthwhile noting that iPETS, which employs an urbanization scenario developed at NCAR (Jiang and O’Neill, in preparation) rather than a scenario from a literature source, shows a somewhat lower urbanization trend compared to GCAM, IMAGE and MESSAGE, most pronounced in India+ where by 2050 the central case is about 5%-points lower than that of the other models. Accordingly, also the alternative urbanization trajectories are lower compared to the other models. When interpreting results it is important to keep in mind that the alternative MESSAGE urbanization projections do not span the full range of +/-10%-points by 2050 as they are based on an existing set of urbanization scenarios by Grubler et al. (forthcoming). Figure 3: Development of the share of urban population in China+ (left panel) and India+ (right panel) between 1950 until 2050 in different urbanization scenarios (historical data from UN (2010b)). Note that differences in the base year result from differences in regional definitions. The data sources on economic activity, measured in terms of gross domestic product (GDP) in market exchange rate (MER), vary more between the different models, using UN statistics, World Bank data, GTAP and CIA (Table 1). For India+, this leads to a narrow range of base year data in terms of total GDP per capita, but estimations on the urban/rural shares of GDP vary over a wider range. For China+, the total GDP per capita is higher in IMAGE, where the China+ region includes high-income countries like Hong-Kong and Taiwan, and lower in GCAM, whose China+ regions includes lowerincome countries such as Cambodia and Viet Nam. Also for China, the estimation of urban/rural shares in GDP varies widely between the two models that report this indicator (see online supplementary material, Table A.1).

ACCEPTED MANUSCRIPT

RI

PT

In contrast to population growth and urbanization, assumptions about economic development show a much wider variation, even after correcting for differences in population size (Figure 4). While some differences can be explained by differences in region definitions, others are reflections of more fundamental differences in assumptions about the development of labor productivity and capital availability. Combined with the base year differences, these developments result in roughly a factor of 2 difference in India+ between the lowest (iPETS) and the highest (IMAGE) projection in 2050 and an only slightly smaller spread for China+. GCAM and MESSAGE tend to lie between the projections of the two other models with GCAM being very close to iPETS and MESSAGE being closer to IMAGE.

NU

SC

Figure 4: Development of GDP (MER) per capita in China+ (left panel) and India+ (right panel) between 1970 until 2050 in the central urbanization scenarios (historical data from IEA (2010b)). For iPETS also the variation of GDP per capita in the alternative urbanization scenarios are shown. Note that differences in the base year result from differences in regional definitions.

MA

One point that will be important in the interpretation of results which has already been discussed in Section 2 is that iPETS includes an influence of urbanization on total GDP, leading to a +/-7% variation in per-capita GDP in 2050 in both analyzed regions as can be seen in Figure 4.

D

4. Urban and rural residential energy use

AC CE P

TE

This section compares residential energy use, where possible separately for urban and rural population, across models for the central urbanization scenario. In addition, the development of energy use by household income is presented jointly with a sensitivity analysis of economic development to explore the reasons for some of differences observed in the comparison of results.

4.1. Base year data

Base year data sources for energy use are very similar across models. For total energy use, all models rely on IEA data. For the urban/rural distinctions in China+, most models use the LBL China Energy Databook (LBL, 2008). For India+, all models use data from the National Sample Survey Organization (NSSO, 2007a). An important distinction is that the iPETS model uses expenditure data (associated with energy quantities) from these sources, whereas the other models are based on the physical energy use data. Total energy use for India+ is rather similar across the models, just below 7 EJ/yr for the residential sector (except MESSAGE South Asia region) and total energy use by urban/rural areas is also comparable between models. Although the generic picture is comparable (a very high share, >80%, of solids), the ratio of solids vs. liquids/gases varies clearly between models (Table A.1). For rural fuel shares, there is a similar result across model. With respect to urban fuel shares, the role of natural gas is especially different (it should be noted that there is ambiguity between the total sectoral data from the IEA and the urban/rural data from NSSO). For China+, there is more deviation between the base year results of total energy use. It should be noted that IMAGE results are not a direct representation of the data, but the result of calibrating the model to time series data for the period 1971-2007. Total fuel shares match very well across models for China+, as do rural fuel share results. For urban fuel shares, there is more variation, especially on the role of secondary heat.

ACCEPTED MANUSCRIPT 4.2. Residential Energy Use

NU

SC

RI

PT

Figure 5 shows the shares of different fuels in residential final energy use for the central urbanization pathway in the participating models. First of all, there are a number of similarities in the results that bear mentioning. The share of solid fuels in residential energy use, mostly traditional biomass, but in China+ also coal, is declining significantly over time. With the exception of GCAM, the phase-out of traditional biomass is almost complete (