Energy Service Demand Projections and CO 2 Reduction Potentials in Rural Households in 31 Chinese Provinces

Until 2012, most of China’s population lived in rural areas with markedly different patterns of household energy consumption from those in Chinese cities. The studies so far done on residential energy use in rural Chinese households have been limited to questionnaire surveys and panel data analyses. Hardly any studies on energy demand in rural areas have considered both the climatic and economic disparities across Chinese regions. In this study we conduct a systematic analysis of the rural Chinese residential sector on a regional basis. We begin by developing a macro-model to estimate energy service demands up to 2050. Next, we apply the AIM(Asia-Pasific Integrated Model)/Enduse model, a bottom-up cost-minimization model with a detailed mitigation technology database, to estimate the mitigation potential of low-carbon technologies in rural China. Our results show that energy service demand in the rural household sector will continue to increase in regions with growing population or income conditions. However, after 2030, the rural residential energy service demand will start to decline in most Chinese regions. The impacts of efficient technologies will vary from one region to the next due to regional climatic and economic disparities. Throughout all of China, the penetration of efficient technologies can reduce CO2 emissions by 20% to 50%. Of the technologies available, efficient lighting, biomass water heaters, and efficient electronics bring the most benefit when implemented in rural households.


Introduction
Most of China's people lived in rural areas until 2011, when the urban population surpassed the rural population for the first time due to the country's rapid urbanization.As of now there are still over 650 million people and 210 million households [1] in the countryside.The types of housing and household energy structures markedly differ between urban and rural areas in China.The differences are so substantial that most of the studies done so far on Chinese energy demand have treated urban and rural households as two different sectors [2][3][4].
In the last century only a few studies looked into the energy service demand of China's rural households.Technology share and energy consumption were rather unclear when compared to the urban households.With China's rapid economic growth, living standards have sharply risen and the gap between the urban and rural economies has steadily narrowed (Figure 1, right).Rural residential service demand is drawing more attention from energy researchers.Several of the first attempts to open the "black box" in the research have relied on questionnaire survey methodologies [4][5][6].While questionnaire surveys cover rural households incompletely, they do provide a representative energy consumption profile of Chinese rural residences in certain areas.
Sustainability 2015, 72015, 7, page-page 2 questionnaire surveys cover rural households incompletely, they do provide a representative energy consumption profile of Chinese rural residences in certain areas.(left, [7]) and per capita residential energy consumption (right, [8]).
The estimation of energy service demand on a national scale requires more systematic methods.Krey et al. [9] used three integrated assessment models to evaluated rural energy use in Asian countries and carried out a comparison study based on their results.Fan et al. [2] and Feng et al. [10] analyzed panel data to examine CO2 emissions and energy use in China's rural households, respectively.All three of these studies were conducted at the country level in spite of the huge disparities in income and climate across regions and provinces.Household energy service demand is often associated with income and expenditure [11][12][13][14].Zhao et al. [4] surveyed household energy consumption in four kinds of Chinese regions and examined the impact of the population scale and per capita income on per household energy consumption.Their method, however, resulted in the same per household energy consumption in cold and warm areas with the same per capita income (e.g., Beijing and Shanghai).This result clearly diverges from the reality, because cold areas have much higher energy consumption due to space heating services in the winter season.Yu et al. [15] proposed a regional approach of estimating building energy demand by dividing China into four climatic areas.However, their study put Beijing, a highly developed region with a much higher disposable income to expend on energy services, in the same climatic area as some of the less developed regions of the west.
China has the third largest territory in the world and an extremely diverse climate.The country is divided into five climatic areas ( [16], Supplementary Figure S1) for designing the built environment: severe cold; cold; hot summer/cold winter; temperate; and hot summer/warm winter.Buildings in cold and severe cold areas are heated by central heating systems, while those in the other three areas are heated mainly by standalone air conditioners.As shown in Figure 2, the heating degree day 18 gradually decreases from north to south in China [17].Economic inequality, meanwhile, has posed a major challenge to Chinese society in recent years.According to The Economist [18], the per capita GDP in some coastal provinces of China is already as high as that in some European countries.However, in some western provinces, the per capita GDP is still at the same level as some of the world's least developed countries (Figure 2, [18]).
Given this complex background, researchers cannot expect to evaluate household service demands in China solely on the basis of socioeconomic indicators or climate characteristics.Hardly any of the studies conducted so far have fully considered the regional disparities in climate or economy.The present study disaggregates China to the provincial level and looks into the patterns of rural residential energy demand in each provincial region.We first develop a macro-model to estimate future outlooks on rural residential energy service demand up to 2050.Then we use the AIM/Enduse model to evaluate the CO2 emission reduction potential in order to analyze the effects  ) and per capita residential energy consumption (right, [8]).
The estimation of energy service demand on a national scale requires more systematic methods.Krey et al. [9] used three integrated assessment models to evaluated rural energy use in Asian countries and carried out a comparison study based on their results.Fan et al. [2] and Feng et al. [10] analyzed panel data to examine CO 2 emissions and energy use in China's rural households, respectively.All three of these studies were conducted at the country level in spite of the huge disparities in income and climate across regions and provinces.Household energy service demand is often associated with income and expenditure [11][12][13][14].Zhao et al. [4] surveyed household energy consumption in four kinds of Chinese regions and examined the impact of the population scale and per capita income on per household energy consumption.Their method, however, resulted in the same per household energy consumption in cold and warm areas with the same per capita income (e.g., Beijing and Shanghai).This result clearly diverges from the reality, because cold areas have much higher energy consumption due to space heating services in the winter season.Yu et al. [15] proposed a regional approach of estimating building energy demand by dividing China into four climatic areas.However, their study put Beijing, a highly developed region with a much higher disposable income to expend on energy services, in the same climatic area as some of the less developed regions of the west.
China has the third largest territory in the world and an extremely diverse climate.The country is divided into five climatic areas ( [16], Supplementary Figure S1) for designing the built environment: severe cold; cold; hot summer/cold winter; temperate; and hot summer/warm winter.Buildings in cold and severe cold areas are heated by central heating systems, while those in the other three areas are heated mainly by standalone air conditioners.As shown in Figure 2, the heating degree day 18 gradually decreases from north to south in China [17].Economic inequality, meanwhile, has posed a major challenge to Chinese society in recent years.According to The Economist [18], the per capita GDP in some coastal provinces of China is already as high as that in some European countries.However, in some western provinces, the per capita GDP is still at the same level as some of the world's least developed countries (Figure 2, [18]).
Given this complex background, researchers cannot expect to evaluate household service demands in China solely on the basis of socioeconomic indicators or climate characteristics.Hardly any of the studies conducted so far have fully considered the regional disparities in climate or economy.The present study disaggregates China to the provincial level and looks into the patterns of rural residential energy demand in each provincial region.We first develop a macro-model to estimate future outlooks on rural residential energy service demand up to 2050.Then we use the AIM/Enduse model to evaluate the CO 2 emission reduction potential in order to analyze the effects of energy-efficient  The remainder of this paper is structured into Sections 2-4.Section 2 describes the methodology for estimating future service demand and evaluating the CO2 emission reduction potential.Section 3 presents the results of our estimations of future service demand, energy consumption, and CO2 emissions at the provincial level.Section 4 discusses the technology selection in detail and presents key conclusions.

Socioeconomic Framework
In this research we use the Shared Socioeconomic Pathways (SSPs), a set of qualitative and quantitative narratives that describe future socioeconomic conditions, to project socioeconomic indicators such as population, GDP, and the urban population share.The SSPs are translated into five storylines based on worlds with various challenges for mitigation and adaption [19,20].Each SSP storyline describes a different socioeconomic circumstance of the future.In this study we use SSP2 3 , a storyline describing a "business-as-usual" world where the typical trends of recent decades continue, with some progress toward achieving development goals.
The original SSP database [21] only provides quantification data at the country level.To conduct our analysis at the provincial level, we disaggregate the original national value to provincial values ( [22], Supplementary File S1).Consistent with SSP2's "business-as-usual" storyline, the difference between the provincial value and national value in a base year is assumed to stay the same up to 2050.The compound population growth rate and urban population share of each province are both disaggregated in this manner.Figure 3 (left) shows the results of a regional rural population projection.With the progress of urbanization in China, the rural population is expected to continue falling.As Krey et al. [9] point out, differences in behavior between urban and rural households may lead to different future emission pathways with high or low urbanization levels.Urbanization is considered a key driver of energy service demand in this study.Meanwhile, the household size is extrapolated up to 2050 based on the past trend (Figure 3, right).Households are larger in rural areas than in cities in China, as the "One-child Policy" is generally waived in the countryside.The projection, however, shows that households will keep shrinking even in rural areas.
The growth rate of per capita GDP is another principal driver of future energy demand, but one that fails to distinguish between urban and rural families.In this study we use per capita income as an economic indicator instead.Specifically, provincial per capita GDP is disaggregated from SSP2 (Figure 4, left) and a linear regression is performed to convert the GDP to rural income (Figure 4, right).The remainder of this paper is structured into Section 2, Section 3 and Section 4. Section 2 describes the methodology for estimating future service demand and evaluating the CO 2 emission reduction potential.Section 3 presents the results of our estimations of future service demand, energy consumption, and CO 2 emissions at the provincial level.Section 4 discusses the technology selection in detail and presents key conclusions.

Socioeconomic Framework
In this research we use the Shared Socioeconomic Pathways (SSPs), a set of qualitative and quantitative narratives that describe future socioeconomic conditions, to project socioeconomic indicators such as population, GDP, and the urban population share.The SSPs are translated into five storylines based on worlds with various challenges for mitigation and adaption [19,20].Each SSP storyline describes a different socioeconomic circumstance of the future.In this study we use SSP2 3 , a storyline describing a "business-as-usual" world where the typical trends of recent decades continue, with some progress toward achieving development goals.
The original SSP database [21] only provides quantification data at the country level.To conduct our analysis at the provincial level, we disaggregate the original national value to provincial values ( [22], Supplementary File S1).Consistent with SSP2's "business-as-usual" storyline, the difference between the provincial value and national value in a base year is assumed to stay the same up to 2050.The compound population growth rate and urban population share of each province are both disaggregated in this manner.Figure 3 (left) shows the results of a regional rural population projection.With the progress of urbanization in China, the rural population is expected to continue falling.As Krey et al. [9] point out, differences in behavior between urban and rural households may lead to different future emission pathways with high or low urbanization levels.Urbanization is considered a key driver of energy service demand in this study.Meanwhile, the household size is extrapolated up to 2050 based on the past trend (Figure 3, right).Households are larger in rural areas than in cities in China, as the "One-child Policy" is generally waived in the countryside.The projection, however, shows that households will keep shrinking even in rural areas.
The growth rate of per capita GDP is another principal driver of future energy demand, but one that fails to distinguish between urban and rural families.In this study we use per capita income as an economic indicator instead.Specifically, provincial per capita GDP is disaggregated from SSP2 (Figure 4, left) and a linear regression is performed to convert the GDP to rural income (Figure 4, right).Besides total population and per capita GDP, the dwelling floor area and ownerships of household electronics are also key indicators in estimating the future energy service demand.To begin determining the dwelling floor area, a regression analysis (Supplementary File S2, Table S5) is performed on panel data of the per household floor area between 2000 and 2012 [7].Next, the regression equation is used to extrapolate the past trend of per household floor area up to 2050.These projections follow the story line of SSP2-"a development pathway that is consistent with typical patterns of historical experience" [19].These data are then synced with data on the total number of households to determine the total floor area.Regional projections of floor area are listed in Supplementary Table S1.The household electronics component of our model consists of four types of electric appliances (white goods and brown goods): televisions, refrigerators, washing machines, and computers.The ownership projections for these goods are listed in Supplementary Table S2.

Energy Service Demand
In our previous research we designed a macro-model [22,23] to estimate the future energy service demand of the urban residential sector.Rural energy service demand in the present study is estimated in the same manner, while the input databases on indicators such as the number of households, floor area, etc., are replaced with rural ones.In urban areas, space cooling service is also  Besides total population and per capita GDP, the dwelling floor area and ownerships of household electronics are also key indicators in estimating the future energy service demand.To begin determining the dwelling floor area, a regression analysis (Supplementary File S2, Table S5) is performed on panel data of the per household floor area between 2000 and 2012 [7].Next, the regression equation is used to extrapolate the past trend of per household floor area up to 2050.These projections follow the story line of SSP2-"a development pathway that is consistent with typical patterns of historical experience" [19].These data are then synced with data on the total number of households to determine the total floor area.Regional projections of floor area are listed in Supplementary Table S1.The household electronics component of our model consists of four types of electric appliances (white goods and brown goods): televisions, refrigerators, washing machines, and computers.The ownership projections for these goods are listed in Supplementary Table S2.

Energy Service Demand
In our previous research we designed a macro-model [22,23] to estimate the future energy service demand of the urban residential sector.Rural energy service demand in the present study is estimated in the same manner, while the input databases on indicators such as the number of households, floor area, etc., are replaced with rural ones.In urban areas, space cooling service is also  Besides total population and per capita GDP, the dwelling floor area and ownerships of household electronics are also key indicators in estimating the future energy service demand.To begin determining the dwelling floor area, a regression analysis (Supplementary File S2, Table S5) is performed on panel data of the per household floor area between 2000 and 2012 [7].Next, the regression equation is used to extrapolate the past trend of per household floor area up to 2050.These projections follow the story line of SSP2-"a development pathway that is consistent with typical patterns of historical experience" [19].These data are then synced with data on the total number of households to determine the total floor area.Regional projections of floor area are listed in Supplementary Table S1.The household electronics component of our model consists of four types of electric appliances (white goods and brown goods): televisions, refrigerators, washing machines, and computers.The ownership projections for these goods are listed in Supplementary Table S2.

Energy Service Demand
In our previous research we designed a macro-model [22,23] to estimate the future energy service demand of the urban residential sector.Rural energy service demand in the present study is estimated in the same manner, while the input databases on indicators such as the number of households, floor area, etc., are replaced with rural ones.In urban areas, space cooling service is also considered as a major residential energy service in China's warm regions [22].However, cooling devices are rarely used in rural families.Although the usage of cooling services is highly influenced by income Sustainability 2015, 7, 15833-15846 level, through the past 20 years there is no evidence suggesting that cooling services have widely penetrated in China's rural households [2,5,6,24].Therefore, in this study, cooling service is excluded and energy service demand in the rural residential sector is categorized into five applications: heating, hot water, cooking, lighting, and electronics.Figure 5 describes the general modeling process and main exogenous inputs of the household applications.
Sustainability 2015, 72015, 7, page-page 5 considered as a major residential energy service in China's warm regions [22].However, cooling devices are rarely used in rural families.Although the usage of cooling services is highly influenced by income level, through the past 20 years there is no evidence suggesting that cooling services have widely penetrated in China's rural households [2,5,6,24].Therefore, in this study, cooling service is excluded and energy service demand in the rural residential sector is categorized into five applications: heating, hot water, cooking, lighting, and electronics.Figure 5 describes the general modeling process and main exogenous inputs of the household applications.The annual space heating service demand is calculated using Equation (1).The annual heating service demand per unit (m 2 ) is taken from Zhang et al. [17].
where : The region : The year , : The annual heating service demand : The annual heating service demand (ktoe/m 2 ) per unit (m 2 ) , : The rural residential floor area The domestic hot water (DHW) service demand for face-washing, cooking, and showering is calculated using Equation (2).The usage of hot water per activity and hot water supply temperature are taken from standard household models [25].The tap water temperature is calculated from the average seasonal temperature. where DHW activity (face-washing, showering, and cooking) : The season (summer, winter, and others) c: The heat capacity of water ((kJ/kg•°C) ρ: The density of water

Domestic hot water supply Other appliances
Calibrated rural annual residential service demand (economic behavior corresponded)

Space heating
Per capita income Annual heating service demand The annual space heating service demand is calculated using Equation (1).The annual heating service demand per unit (m 2 ) is taken from Zhang et al. [17].
HL r,t " HLU r ˆDS r,t where r: The region t: The year HL r,t : The annual heating service demand HLU r : The annual heating service demand (ktoe/m 2 ) per unit (m 2 ) DS r,t : The rural residential floor area The domestic hot water (DHW) service demand for face-washing, cooking, and showering is calculated using Equation (2).The usage of hot water per activity and hot water supply temperature are taken from standard household models [25].The tap water temperature is calculated from the average seasonal temperature.DL r,t " POP r,t ˆÿ s ÿ i cρV i pTS i ´TT r,s q ˆFr,s,i where i: DHW activity (face-washing, showering, and cooking) s: The season (summer, winter, and others) c: The heat capacity of water ((kJ/kg¨˝C) ρ: The density of water DL r,t : The annual DHW service demand (ktoe) V i : The usage of hot water per activity (L) POP r,t : The population TS i : The hot water supply temperature (universal) TT r,s : The average tap water temperature (regional, seasonal) F r,s,i : The activity frequency per season The modeling structure of the other three applications builds on the operating characteristics of each energy-consuming appliance.It can generally be expressed as Equation (3).
OTL r,t " U r ˆAr,t " f pworking power, operation time, etc.q ˆAr,t where OTL r,t : The service demand (ktoe) of lighting, cooking, and electronics U r : The annual service demand per unit (cooking range, etc.) or per m 2 A r,t : The amount of service end users (total electronic units or total floor area) Table 1 summarizes the operating characteristics of lighting, cooking, and the four white/brown goods composing household electronics.Most of the designed working powers and standard operation times are taken from the SCHEDULE program [25].The service demand for cooking stoves is a function of the average household size.As explained in Section 2.2, we extrapolate the numbers of durable consumer goods per household in 2050, with the exception of cooking stoves, which are always set to one unit per household with no future variation.The usage patterns listed in Table 1 describe the general conditions in a standard family of a developed country.For Chinese rural families, the usage patterns need to be adjusted to cope with their lower income level (Section 2.3).

Model Calibration
The frequency of hot water activity, electronics operation time, and other energy usage settings described above more or less represent the standard lifestyle in a developed country.In a less developed region such as rural China, limited financial resources may prevent these settings from being fully applied.Service demand should increase with income and decrease with service price [9,26].In economic terms we can identify a satiation point for energy service comfort: below the satiation point the marginal utility of an energy service is positive; above it, the marginal utility turns negative [27,28].
In the current study we evaluate income-induced service demand in the future by introducing satiation factors to calibrate our model estimates.The satiation factor of each energy service can be described as a logistic function with per capita income as the explanatory variable ( [22]; Equation (4)).When the per capita disposable income reaches the OECD (Organisation for Economic Co-operation and Development) average of 25,596 US$ (year 2011, current prices and purchasing power parity [29]) in the future, the satiation factor is manually set as 100%.A logistic model parameter of energy services sv SF r,sv,t : The satiation factor PCI r,t : Per capita income

SF
In the macro-model from our previous study (Xing et al., 2015 [22]) we determined the satiation factors for certain energy services using the least squares method, as shown in Equation (5).Rural households, however, rely on non-commercial biomass, most of which is collected directly from the fields.The accurate total amount of non-commercial biomass consumption is difficult to tally, and no energy consumption data for the base year (2010) are available in the statistical yearbooks.Without this information, we must assume that the per capita income induces service demand just as it does in urban households and use the same satiation functions to calibrate the model results (Equation ( 6)). Minimizing: where ESD 1 r,sv,t : The calibrated annual energy service demand (rural) ESD estm r,sv,t : The estimated annual energy service demand (rural)

AIM/Enduse
We select effective and efficient technologies for the rural residential sector by estimating the mitigation potential of sustainable policies using the AIM/Enduse model [30,31], a bottom-up cost-minimization model with a detailed mitigation technology database.In the base year (2010) the estimated service demands are disaggregated to the energy usage of existing technologies, while in the target year (2050) service demands are satisfied with a combination of existing technologies and efficient technologies.For each household appliance we set an existing (EXT) technology with a low efficiency and one or more technologies with high efficiencies (NEW and BAT).NEW technology is considered more efficient and BAT (best available technology) is considered the most efficient.The technological energy efficiencies in the base year are average values for widely used electric appliances [32].The technology shares in the base year are taken from empirical studies [33][34][35][36][37]. Table 2 gives an example of the technology levels.A total of 83 existing, new, and best available technologies are prepared for the estimation of emission reduction potentials in the AIM/Enduse model (Supplementary Table S3).The model selects a technology combination for each province based on a least total system cost.The emission factors that are used to estimate CO 2 emission are taken from AIM/Enduse-Global ( [38], Supplementary Table S6).

Scenario Setting
For the model simulation we prepare two scenarios to evaluate the CO 2 emission growth and reduction potential: the Fix (FIX) and Cost-effective selection (CES) scenarios.In FIX the technologies can be regarded as frozen: the technology level remains at the base year (2010) level and no efficient technologies are implemented in the future.FIX is designed to offer a baseline picture of how CO 2 emissions will increase without any efficient technology intervention.In CES, efficient technologies stored in technology databases are allowed to penetrate the future market.Traditional technologies and efficient technologies compete over cost-effectiveness.In other words, CES describes a natural implementation process for efficient technologies.
The model basically allows the choice of any efficient technology to achieve a minimum total system cost.Two constraints take effect, however, during this process.The first constraint addresses the future share of non-commercial energy.Non-commercial energy sources such as firewood or straw currently account for over 80% of the rural residential energy demand [39].Most of the non-commercial energy is collected by household members and has no commercial price, which makes it easy for the model to choose.However, the burning of these non-commercial fuels in indoor environments can cause serious air pollution problems.For this reason, CES does not encourage the use of non-commercial energy or set the share of non-commercial energy beyond the current (2010) level.The second constraint addresses the future share of renewable energy.The applications of renewable energy require expensive appliance costs without incurring costs for the energy supply itself.The minimized total system cost in the model analysis process can lead to a sudden increase in the renewable share within a short period, an outcome unlikely to occur in the real world.Exogenous constraints are therefore added for the renewables under study.In CES, the shares of these renewables will not surpass the IEA's 450 scenario ( [40]; Supplementary Table S4).Table 3 summarizes the definitions of the two scenarios.

Fix (FIX)
Technology efficiency fixed at the 2010 level

Cost-effective selection (CES)
Efficient technologies naturally penetrated in the future market Share of non-commercial energy not to surpass the 2010 level Share of renewable energy not to surpass the IEA's 450 scenario

Energy Service Demands
Figure 6 shows the energy service demand of the whole of China and three Chinese regions with different urbanization levels.The urban and rural populations are disaggregated results from SSP2.The urban service demand is drawn from the results from our previous study [22].Beijing and Tibet represent high and low urbanization cases, respectively.Different factors drive constant increases in the rural service demand in Beijing and Tibet.The share of the rural population in Beijing drops considerably by 2050, but the rural population grows in absolute terms because the total Beijing population doubles.The rural energy service demand thus increases intensively due to increases in Sustainability 2015, 7, 15833-15846 both the population and income level.Tibet currently has the highest rural population share and is expected to keep this rank in 2050.The increase of the rural energy service demand in Tibet probably stems from the slow but steady process of urbanization and rising income.Inner Mongolia represents a middle urbanization case.As income rises and the rural population falls, rural energy service demand in Inner Mongolia grows up to 2030, then declines over the next two decades.The same pattern is observed when aggregating all 31 regions into China's national results.
Sustainability 2015, 72015, 7, page-page 9 a middle urbanization case.As income rises and the rural population falls, rural energy service demand in Inner Mongolia grows up to 2030, then declines over the next two decades.The same pattern is observed when aggregating all 31 regions into China's national results.

Energy Consumption and CO2 Emission
Figure 7 shows the estimated results of energy consumption (top row) and CO2 emissions (bottom row) in China's rural households.We selected four representative regions with different building codes and looked into their future projections.Overall, energy consumption and CO2 emissions both grow drastically in the future.In CES, efficient technologies penetrate the market-in some cases in limited amounts-even without emission taxes or technology subsidies.In the target year (2050), the FIX-compared reduction rate ((FIX-CES)/FIX) of energy consumption appears to be low in cold regions (6% in Heilongjiang, China) and high in warm regions (19% in Guangdong, China).Reductions in the consumption of coal and electricity energy contribute the bulk of the CO2 emission reductions.The FIX-compared reduction rate of CO2 emissions in the target year ranges from 21% in cold regions (Heilongjiang, China) to 54% in warm regions (Guangdong, China).Electricity renewables (photovoltaic panels, etc.) have no observable reduction impact on the results.This tells us that the expensive upfront costs of electricity renewables will dissuade the regions from adopting them unless a carbon tax, technology subsidies, or other financial supports are provided.
Penetration of efficient technologies also has impacts on per capita energy/emission intensities.In Beijing, per capita energy consumption drops from 534.92 ktoe/cap (FIX) to 492.84 ktoe/cap (CES) while per capita CO2 emissions drop from 105.98 tCO2/cap (FIX) to 86.03 tCO2/cap (CES) by 2050.In Guangdong, a warm region, per capita energy/emission intensities are much lower than Beijing's level.Nevertheless, there is a marked reduction of energy/emission intensities in CES.Per capita energy consumption drops from 105.98 ktoe/cap (FIX) to 86.03 ktoe/cap (CES) while per capita CO2 emissions drop from 0.37 tCO2/cap (FIX) to 0.17 tCO2/cap (CES) by 2050.However, compared to the results of the urban residential sector from our previous study (Xing, 2015 [22]), per capita energy/emission intensities in rural households are much lower than in urban households.This

Energy Consumption and CO 2 Emission
Figure 7 shows the estimated results of energy consumption (top row) and CO 2 emissions (bottom row) in China's rural households.We selected four representative regions with different building codes and looked into their future projections.Overall, energy consumption and CO 2 emissions both grow drastically in the future.In CES, efficient technologies penetrate the market-in some cases in limited amounts-even without emission taxes or technology subsidies.In the target year (2050), the FIX-compared reduction rate ((FIX-CES)/FIX) of energy consumption appears to be low in cold regions (6% in Heilongjiang, China) and high in warm regions (19% in Guangdong, China).Reductions in the consumption of coal and electricity energy contribute the bulk of the CO 2 emission reductions.The FIX-compared reduction rate of CO 2 emissions in the target year ranges from 21% in cold regions (Heilongjiang, China) to 54% in warm regions (Guangdong, China).Electricity renewables (photovoltaic panels, etc.) have no observable reduction impact on the results.This tells us that the expensive upfront costs of electricity renewables will dissuade the regions from adopting them unless a carbon tax, technology subsidies, or other financial supports are provided.
Penetration of efficient technologies also has impacts on per capita energy/emission intensities.In Beijing, per capita energy consumption drops from 534.92 ktoe/cap (FIX) to 492.84 ktoe/cap (CES) while per capita CO 2 emissions drop from 105.98 tCO 2 /cap (FIX) to 86.03 tCO 2 /cap (CES) by 2050.In Guangdong, a warm region, per capita energy/emission intensities are much lower than Beijing's level.Nevertheless, there is a marked reduction of energy/emission intensities in CES.Per capita energy consumption drops from 105.98 ktoe/cap (FIX) to 86.03 ktoe/cap (CES) while per capita CO 2 emissions drop from 0.37 tCO 2 /cap (FIX) to 0.17 tCO 2 /cap (CES) by 2050.However, compared to the results of the urban residential sector from our previous study (Xing, 2015 [22]), per capita energy/emission intensities in rural households are much lower than in urban households.This indicates that even with efficient technologies, emission reduction will not be as substantial as that in urban households considering the future growth of the urban population.
Sustainability 2015, 72015, 7, page-page 10 indicates that even with efficient technologies, emission reduction will not be as substantial as that in urban households considering the future growth of the urban population.

Technology Selection
In this study we provide a choice of 83 residential energy service (Table 2) technologies within the AIM/Enduse model to keep the cost as low as possible while attaining CO2 emission reduction.Figure 8 shows the efficient technologies that bring about the greatest CO2 emission reductions in four representative provincial areas.Although the reduction shares of each technology differ among the four regions, the technology selection seems to be unanimous: efficient lighting, biomass water heaters, and efficient electronics.

Technology Selection
In this study we provide a choice of 83 residential energy service (Table 2) technologies within the AIM/Enduse model to keep the cost as low as possible while attaining CO 2 emission reduction.Figure 8 shows the efficient technologies that bring about the greatest CO 2 emission reductions in four representative provincial areas.Although the reduction shares of each technology differ among the four regions, the technology selection seems to be unanimous: efficient lighting, biomass water heaters, and efficient electronics.
The diffusion of the fluorescent lamp contributes over half of the CO 2 emission reduction in all four regions.The fluorescent lamp and LED lamp are respectively considered a "NEW" lighting technology and an optimally efficient "BAT" technology in this category.Nevertheless, the model simulation results suggest a large penetration of the fluorescent lamp with no accompanying penetration of LED.Though less efficient than LED, the fluorescent lamp has the appealing advantage of an affordable price.The fluorescent lamp can thus be seen as a bridge between the traditional incandescent lamp and LED, especially in rural households.Biomass (wood residuals) is considered a carbon-neutral energy source important for CO 2 emission reduction.The biomass water heater ranks as the second most important technology in Heilongjiang and Guangdong, and the third most important in Beijing and Shanghai.
No emission tax or technology subsidy is included in the low-carbon scenario of this study.We examine only the energy savings brought about by efficient technologies.This necessarily limits the selectable technologies.In future studies we will also consider financial instruments when we design our scenarios.We also expect regions to adopt different technology combinations best suited to their respective climates and economies.
Sustainability 2015, 72015, 7, page-page 11 The diffusion of the fluorescent lamp contributes over half of the CO2 emission reduction in all four regions.The fluorescent lamp and LED lamp are respectively considered a "NEW" lighting technology and an optimally efficient "BAT" technology in this category.Nevertheless, the model simulation results suggest a large penetration of the fluorescent lamp with no accompanying penetration of LED.Though less efficient than LED, the fluorescent lamp has the appealing advantage of an affordable price.The fluorescent lamp can thus be seen as a bridge between the traditional incandescent lamp and LED, especially in rural households.Biomass (wood residuals) is considered a carbon-neutral energy source important for CO2 emission reduction.The biomass water heater ranks as the second most important technology in Heilongjiang and Guangdong, and the third most important in Beijing and Shanghai.
No emission tax or technology subsidy is included in the low-carbon scenario of this study.We examine only the energy savings brought about by efficient technologies.This necessarily limits the selectable technologies.In future studies we will also consider financial instruments when we design our scenarios.We also expect regions to adopt different technology combinations best suited to their respective climates and economies.

Conclusions
This study highlights the importance of considering the socio-economic and climate diversity that exist across rural areas in China while estimating the rural energy demand.Future energy demand in rural households is estimated using a well-grounded methodology.The results show that energy service demand in China's rural household sector will continue to increase in regions with growing population or income conditions.In most Chinese regions, the rural service demand will start to decline after 2030.The results show that the decoupling between energy and emission is higher in warm areas compared to cold areas in rural households.Compared to our previous study [22] related to urban household energy demand estimation, emission intensity improvement is lower in rural households in the competitive technologies selection scenario.The results indicate that the technology transition can achieve emission reductions in the rural area in an efficient manner without

Conclusions
This study highlights the importance of considering the socio-economic and climate diversity that exist across rural areas in China while estimating the rural energy demand.Future energy demand in rural households is estimated using a well-grounded methodology.The results show that energy service demand in China's rural household sector will continue to increase in regions with growing population or income conditions.In most Chinese regions, the rural service demand will start to decline after 2030.The results show that the decoupling between energy and emission is higher in warm areas compared to cold areas in rural households.Compared to our previous study [22] related to urban household energy demand estimation, emission intensity improvement is lower in rural households in the competitive technologies selection scenario.The results indicate that the technology transition can achieve emission reductions in the rural area in an efficient manner without any financial

Figure 5 .
Figure 5. Flow chart of the service demand estimation process.

Figure 5 .
Figure 5. Flow chart of the service demand estimation process.

Figure 6 .
Figure 6.Energy service demand of 31 Chinese provinces in SSP2.

Figure 6 .
Figure 6.Energy service demand of 31 Chinese provinces in SSP2.

Figure 7 .
Figure 7. Energy consumption and CO2 emissions of China's rural residential sector.

Figure 7 .
Figure 7. Energy consumption and CO 2 emissions of China's rural residential sector.

Figure 8 .
Figure 8. CO2 emission reduction contributors in four provincial areas.

Figure 8 .
Figure 8. CO 2 emission reduction contributors in four provincial areas.

Table 1 .
Operation conditions of household appliances.
HS: household size (persons per household). sv

Table 2 .
Efficiencies (provided/consumed) of three technology levels.