Future energy demand and supply is usually provided by integrated energy system models. Several energy system models, such as GCAM (Clarke et al., 2008, 2018; Collins et al., 2015; http://www.globalchange.umd.edu/gcam/, last access: 25 March 2020), the Integrated Model to Assess the Global Environment (IMAGE; Alcamo et al., 1996; Braspenning Radu et al., 2016; https://models.pbl.nl/image/index.php/Welcome_to_IMAGE_3.0_Documentation, last access: 25 March 2020), and the Model of Energy Supply Systems And the General Environmental Impact (MESSAGE; Miketa et al., 2006; Zhang et al., 2019b; http://pure.iiasa.ac.at/id/eprint/1542/, last access: 25 March 2020), are widely used by researchers and policymakers.

GCAM is a global partial equilibrium model with 32 energy–economy regions representing the behaviour and interactions among five systems: the energy system, water, agriculture and land use, economy, and climate. GCAM is stewarded by the Joint Global Change Research Institute (JGCRI) (GCAM, 2019, http://jgcri.github.io/gcam-doc/index.html, last access: 25 March 2020), and more detailed documentation on GCAM can be found at http://www.globalchange.umd.edu/models/gcam/ (last access: 25 March 2020). Over time, GCAM has been increasingly used in climate (Zhou et al., 2013; Fawcett et al., 2015; Calvin et al., 2019; Sinha et al., 2019), energy (Belete et al., 2019; Silva Herran et al., 2019; Wang et al., 2019), land use (Dong et al., 2018; Turner et al., 2018; Vittorio et al., 2018), and modelling studies. In addition, GCAM also provides a number of scenarios and assessments for various organizations and reports, such as the Energy Modeling Forum (EMF), the U.S. Climate Change Technology Program, the U.S. Climate Change Science Program (Edmonds and Reilly, 1982, 1983; Edmonds et al., 1984), and the IPCC assessment reports (Reilly et al., 1987; Pachauri et al., 2014; Calvin et al., 2017). Most energy system models, including GCAM, take China as a whole section in simulations and fail to reflect the differences in regional or provincial energy and socio-economic developments. To better explore the provincial evolution and development in China, JGCRI developed and expanded GCAM to include greater spatial detail in China's provinces, and this model is referred to as GCAM-China (Yu et al., 2019). In GCAM-China, the 31 provinces are operated as explicit regions within the global GCAM model. Energy transformation and end-use demand processes are simulated at provincial levels. Therefore, we used GCAM-China (version 4.3) to project future energy demand and supply in China under different economic-energy scenarios at the provincial level.

To better illustrate the emission characteristics, emission models, such as the MEIC model, always have much more elaborate emission sources and fuel type categories than energy models. Here, we downscaled and disaggregated the GCAM-China outputs and matched them to the MEIC framework (Table S1). The 227 fundamental emission categories in the MEIC model are composed of intercombinations of seven major sectors (including power, heating, industry, residential, transportation, solvent use, and agriculture) and various fuels and productions (Table S1). Different technologies would further divide these emissions into 745 detailed sources; however, the evolution of technology distributions is simulated in the DPEC. Therefore, the interaction data system mainly conducts sector mapping and fuel mapping to match the GCAM-China outputs to the 227 MEIC categories. Except for the heating sector, all the linkage works were operated at the provincial level.

From the sector perspective, the energy and economic-related outputs from the GCAM-China or GCAM model can be divided into four parts: resource production (primary energy), energy transformation (electricity, heat, refining, and other energy transformation), final energy use (buildings, industry, and transportation), and socio-economics (population and gross domestic product, GDP). The power and heating sector in DPEC were linked from the energy transformation part in GCAM-China. The power sectors of the two models have basically no gaps and can be directly matched. Due to the unavailability of provincial energy information in the heating sector in GCAM-China, we first matched the national heat outputs from the GCAM-China energy consumption sector to DPEC and then downscaled to the provincial level with district heat outputs. When downscaling, the heating industrial (to offer thermal energy) in the DPEC is derived from industry district heat in the industry sector, and residential heating (refers to centralized heating) in the DPEC is obtained from the commercial and residential urban district heat in the building sector. The residential, industry combustion, and transportation sectors in DPEC are matched with the final energy-use parts in GCAM-China, which are building, industry, and transportation. Similar to the power sector, the cement industry and transportation sources could also be seamlessly connected in the two models. Residential in DPEC is taken from the building sector, including cooking and heating. Boilers and kilns are two major industrial combustion sources in DPEC. Activity rates of industrial boilers and cement kilns in DPEC are provided by industry final energy use and cement energy consumption in GCAM-China, respectively. For other industrial kilns (brick and lime), the activity rates are estimated by their base-year data and the future cement energy-use curve. Future iron and steel manufacturing would be simultaneously affected by energy transformation and socio-economic development. Due to the lack of iron and steel projections in GCAM-China (version 4.3), we estimated future iron and steel productions with projected GDP using the elastic coefficient method (Cao et al., 2016) and fixed furnace fractions (electric, coal-fired, gas-fired, and other-fuel-fired) of newly built capacities to maintain the same energy structure as that of the whole industry sector. The activity rates of non-energy-related sectors in the DPEC, including industrial processes, solvent use, and agriculture, were mostly driven by socio-economic outputs from GCAM-China, which were specifically described in Sect. 2.3.

In terms of fuel mapping, the fuel types in GCAM-China include coal, liquids, gas, biomass, solar resources, nuclear, wind resources, geothermal, and hydro-energy, while the MEIC model partitions each fossil fuel or biofuel in a more detailed manner (Table S1). A fuel type ratio database is established based on each specific fuel structure of coal, liquids, gas, and biomass in the MEIC model, and the energy outputs of GCAM-China are then distributed by this base-year proportion to show the detailed fuel use in DPEC. The standard coal equivalent is adopted with the different units of different fuel types. This reaggregation process of fuel types indicates that the absolute amount and relative proportion of coal, liquids, gas and biomass will evolve under the driver of the energy model, while the substructure inside each major fuel type will remain the same as the 2015 levels in MEIC.

Eliminating discrepancies in the base year between MEIC and GCAM-China models is pivotal to project future emissions, which can maintain the consistency with energy outputs to the best extent. In this study, 2015 is chosen as the base year, the historical energy and activity rates in 2015 used in MEIC are obtained from China Energy Statistical Yearbook (National Bureau of Statistics (NBS), 2016), and the 2015 information in the GCAM-China model is projected, as GCAM is calibrated in 2010 using the historical datasets from the International Energy Agency (IEA, 2011). The deviation ranges of major fuel types in the base year vary from 6 % to 13 %, and these discrepancies are mainly caused by different statistical methods and raw data sources (Hong et al., 2017). These balances should also evolve with the projected future trends instead of remaining unchanged. Thus, we harmonized the GCAM-China energy outputs (which have first been reorganized and downscaled to DPEC fuel type categories) by multiplying the base-year balance ratio (MEIC base-year energy values divided by GCAM-China base-year energy values), rather than add or subtract these balances.

Considering the coal-dominated structure in the power sector and a unit-based power plant database during 1990–2015 developed in the MEIC (Liu et al., 2015; Tong et al., 2018a, b), a unit-based emission projection model for coal-fired power plants was developed in our previous study to assess the evolution of the coal-fired power unit fleet and associated emissions (Tong et al., 2018a). This model was designed to simulate power plant fleet turnover by tracking the lifespan of each power generation unit, which can be used in various future policy analyses and emission estimates for coal-fired power plants. In this work, the total coal power generation and coal consumption are directly obtained from GCAM-China (Table S1). We integrated this already-built projection model into the DPEC to project future emissions from coal-fired power plants over China through the year 2050.

Other fuel types combusted in China's power plants mainly include natural gas and biomass, and their future energy consumption is obtained from GCAM-China (Table S1). A technology-based emission projection model is developed for other-fuel-fired power plants due to limited historical emission information. The emission factors are estimated by projecting the effects of different combustion technologies and end-of-pipe control technologies in the target years (i.e. 2020, 2030, and 2050) according to the environmental policies (Xing et al., 2011; Tian et al., 2013), and the effects in the other years of the future are obtained through linear interpolation.

District heating is usually supplied by conventional heat plants or combined heat and power (CHP) plants (Rezaie et al., 2012) for industrial and residential purposes. CHP systems are more thermally efficient than producing process heat alone (Lasseter et al., 2004). Therefore, the government promotes the use of CHP plants. In this work, energy consumption in heat plants by fuel type is obtained from GCAM-China (Table S1), and we developed a power technology-based model to project the technology evolution of heat plants. First, we split energy consumption for CHP plants and conventional heat plants according their thermal efficiencies and heat supply policies. Then, we assumed that CHP plants share the same combustion technology and control technology distributions as power plants under corresponding emission scenarios. For conventional heat plants, we simply adopted a similar model that was developed for other-fuel-fired power plants.

The coal used in the industry sector is commonly combusted in coal-fired boilers or kilns (i.e. cement, lime, and brick kilns). Coal consumption in boilers is estimated as total industrial coal consumption from GCAM-China minus estimated coal consumption in kilns (Table S1). Future cement coal consumption is obtained directly from GCAM-China. By assuming a simultaneous demand change among these industries of non-metal building materials and similar improvement in energy efficiencies (i.e. energy consumed per unit product), the projections of coal consumption in lime and brick kilns are therefore based on the future trends in cement coal use (Table S1). Thus, the coal consumed in industrial boilers is derived.

A technology-based turnover emission projection model for coal-fired industrial boilers is developed (Fig. S1). The historical information of coal-fired industrial boilers is obtained from the MEE (unpublished data, hereafter referred to as the MEE-boiler database), which includes unit-level boiler capacity, combustion technology, and end-of-pipe control devices. Given that the MEE-boiler database is incomplete, instead of developing a boiler-based emission projection model, we aggregated all industrial boilers into 16 categories based on boiler size (< 7, 7–14, 14–24.5, and ≥ 24.5 t h-1) and combustion technology (pulverized boiler, circulating fluidized bed, auto-grate, and hand-feed grate) to generate size and technology distributions, which were then used to simulate the turnover of the industrial boiler fleet. As shown in Fig. S1, the model is designed to simulate the operating industrial boiler turnover driven by the demand and retirement policy, assuming that small boilers and outdated combustion technologies are retired early. In a given year, the model first estimates the supply capability of in-fleet boilers after retirement and then estimates the supply gap under the total coal consumption and fills the gap using new coal-fired industrial boilers. Then, we modelled the changes in emission factors by boiler size and combustion technology by considering the evolution of end-of-pipe control technologies under different emission control policies. Note that we separately considered the newly built and old industrial boilers because there are usually different requirements or emission limits for newly built and old industrial boilers.

Emissions from cement, lime, and brick kilns and other associated industrial processes are estimated within the same projection model and are detailed in the industrial non-combustion sector (Sect. 2.3.3).

The energy consumption by other fuel combustion is directly obtained from GCAM-China (Table S1). The changes in emission factors are estimated by projecting the effects of different combustion technologies and end-of-pipe control measures in the target years (i.e. 2020, 2030, and 2050) according to the environmental policies (Xing et al., 2011), and then these changes are linearly interpolated to the other years in the future.

As the world's largest steel production area, China is reported to contribute almost half of global raw steel production (USGS, 2016). The iron and steel industry involves a series of closely linked processing steps, including preparation of raw materials, iron-making, steel-making, and finishing processes (Wang et al., 2016). First, the productions of sinter, iron, and steel were driven by GDP with a resilience factor law, similar to the cement projections in GCAM-China (Table S3). For the coke industry, most of the coke is used in the iron-making process; therefore, we projected the coke production based on the change trend in iron production (Table S3). The climate and energy policies would change the capacity structures and electric furnace proportions of these subsectors.

Here, a technology-based turnover model for the iron and steel industry is built. Emissions from each process (sinter, iron, and steel) are independently projected in the model (Fig. S1). The historical unit-based information of each process is also obtained from MEE (unpublished data, hereafter referred to as the MEE-steel database), which includes the unit-based and process-based operational status (when the unit was commissioned and decommissioned), capacity, production, technology type, control devices, and corresponding removal efficiencies. We first aggregated the facilities of each process (sinter, iron, and steel productions) into 32 categories based on years under operation (< 20, 20–40, 40–60, and ≥ 60 years), capacity size (< 0.2, 0.2–1, 1–3, > 3 million tonnes) and current technology (electric furnace and nonelectric furnace). Similar to the model for coal-fired industrial boilers, this model starts by simulating the turnover of steelwork plant fleets. In a given year, the model first estimates the production capability of in-fleet facilities after implementing retirement policies by assuming the early retirement of small and old facilities with outdated technologies and then filling the gap with newly built facilities under the total predicted activity rates. Through fixing the furnace fractions (electric, coal-fired, gas-fired, and other-fuel-fired) of newly built capacities, the ultimately fused capacities shared the same energy structure with the whole industry sector of a given year. Finally, we model the changes in emission factors by considering the evolution of end-of-pipe control technologies for existing and newly built facilities under different environmental regulations. For coke plants, we assumed that all coke is produced in machinery ovens (Huo et al., 2012), and we simply estimated the effects of advanced control measures according to assumed emission standards due to the unavailability of plant-level data.

China is the world's largest cement producer and consumer (Lei et al., 2011b). To project future emissions from the cement industry, a kiln-based turnover model is built, which is similar to the model built for coal-fired power plants (Tong et al., 2018a). We began with a kiln-based emission inventory for the 1990–2015 period, which provides historical clinker kiln-level technology and emission information including capacity, operating year, production technology, annual production of clinker and cement, and end-of-pipe control devices and corresponding removal efficiencies (Lei et al., 2011b).

A schematic of the model for the cement industry is shown in Fig. S2. The total cement demand is obtained directly from GCAM-China (Table S1). Beginning with the estimated clinker capacity demand, we simulate the year-to-year dynamics of clinker production structure turnover by considering the retirement of outdated kilns (e.g. small, old, or inefficient capacity) and construction of new kilns. The retirement rate of old kilns is driven by the future demand and forced elimination policy of outdated production capacity. A function for ordering the retirement of individual kilns is developed at the provincial level by considering the production technology, age, and designed capacity with descending priority, which is similar to the function created for coal-fired power plants (Tong et al., 2018a). After considering the retired kilns, for a given year, the model then estimates the capacity gap after evaluating the clinker production capacity of in-fleet kilns and fills the gap using newly built kilns. We finally model the changes in kiln-based emission factors by considering the evolution of the end-of-pipe control technologies. To determine the order of end-of-pipe technology upgrades for each individual kiln, we set up an evolution function of end-of-pipe technology at the provincial level by considering the production technology, designed capacity, and age of each kiln (Tong et al., 2018a).

Except for coke, iron, and steel plants, as well as cement plants, the emission sources of all other metal products, nonferrous metals, non-metal building materials, and other industrial products from the MEIC are grouped into “other metals and non-metals”. Specifically, other metal products and nonferrous metals include foundry products, aluminium, copper, zinc, alumina, and other nonferrous metals. Non-metal building materials include glass (flat glass and glass products), lime, and brick. Other industrial products mainly include products from the food and drink industry (i.e. bread, cake, biscuits, sugar, beer, wine, and spirits) and the textile industry (i.e. wool, silk, cloth, and synthetic fibres).

The future productions of the above-mentioned industrial products are projected in different ways due to unavailability from the GCAM-China model. The productions of other metal products and nonferrous metals are projected by building the regression models, which are used to describe the relationships among steel production (or GDP) and production of each product based on relevant statistical data from 1990 to 2015 (Table S3). For non-metal building materials, the future productions of flat glass and glass products are estimated based on the annual growth rate of the new building area (Table S3). Lime and brick productions are forecasted by applying the future trends in cement production (Table S1), which are consistent with their coal use projections. To project the productions of the food and drink industry and textile industry products, we built a series of regression models linking per capita GDP with their historical productions (Table S3).

Then, we estimated the changes in emission factors for each production process. In addition to production processes in the food and drink industry and textile industry, we assumed outdated production technologies in other metal industry and non-metal building material industry have similar retirement rates as those from the iron and steel industry and cement industry, respectively. The evolution of different control measures is estimated according to emission standards. For products from the food and drink industry and textile industry, only volatile organic compound (VOC) emissions are considered to be emitted. Because there is no specific production technology provided by the MEIC, we only considered the future evolution of VOC control measures. Here, we simply projected the effects of advanced devices designed to reduce VOCs in the targeted years (i.e. 2020, 2030, and 2050) according to the assumed environmental regulations.

The petrochemical industry is considered to be the key VOC-related industry, and VOC emissions from 34 types of petrochemical products are estimated in the MEIC inventory. Here, we grouped these products into five subsectors: oil and gas production, distribution, and refining; fertilizer production; solvent production; synthetic materials; and other chemical products.

Specifically, oil and gas production, distribution, and refining include crude oil production, crude oil handling, oil refining, natural gas production, natural gas distribution, oil depots (gasoline and diesel), and oil stations (gasoline and diesel). The activity rates of these industrial processes are projected based on the change rates of energy demands for corresponding fuel types obtained from GCAM-China (Table S1). Fertilizer production includes the production of urea, ammonium bicarbonate, other nitrate fertilizers (i.e. sodium nitrate and calcium nitrate), and NPK (i.e. nitrogen, phosphorus, and potassium) fertilizer. Fertilizers are commonly used in the agriculture sector, and production is determined by fertilizer demand of national crop yield. Therefore, we assumed that the change in fertilizer production is consistent with fertilizer consumption, which is estimated in the agriculture sector (Sect. 2.3.7). Similarly, solvent production is also determined by the market demand, which includes varnish paint, architectural paint, printing ink, and glue production. We projected the production based on the change rates in corresponding solvent use (Sect. 2.3.6).

Synthetic materials mainly include polyvinyl chloride (PVC) products, polystyrene, ethylene, low-density polyethylene (LDPE), high-density polyethylene (HDPE), styrene, polystyrene, vinyl chloride, PVC, propylene, and polypropylene. Each synthetic material is projected by developing the regression models, and they are used to describe the relationship between national GDP and national production based on historical statistical data (Table S3). Other chemical products, including carbon black, sulfuric acid, synthetic ammonia by coal, pulp, asphalt production, rubber, and tires, are projected using either the change trends in other related products or the regression models linked to GDP (Zhang et al., 2018; Table S3).

For the above-mentioned products, we assumed that the emission factors are only affected by the end-of-pipe control measures due to no specific production technology provided by the MEIC, which is driven by the related environmental policies and emission standards.

Total energy consumed by fuel type in rural and urban areas in the residential sector is separately provided by the GCAM-China building sector (Table S1). The residential sector includes two distinct types of coal combustion equipment for different uses (boilers for heating and stoves for cooking and heating), and their emission factors are quite different (Zhang et al., 2007; Peng et al., 2019). The final residential energy-use split for each usage (i.e. residential heating, cooking, and hot water) is provided directly by GCAM-China (Table S1), which is driven by population, building area, and energy service intensity. The split ratio of two combustion technologies (boiler or stove) is exogenous and evolves with specific clean air policies.

A technology-based projection model for the residential sector is developed (Fig. S3). We projected the year-to-year dynamics of coal combustion technologies (boiler or stove) by assuming that coal stoves are used in individual houses for decentralized heating, cooking, and hot water supply, while coal-fired boilers are used for heating in large buildings in urban areas (Zhang et al., 2007). Finally, we projected the effects of clean coal use, advanced coal stoves and boilers, and end-of-pipe control technologies for coal-fired boilers in the target years (i.e. 2020, 2030, and 2050) under different environmental regulation assumptions and then estimated the effects in the other years of the future through linear interpolation.

For fuel types other than coal, due to limited historical information obtained from MEIC, the effects of advanced combustion technologies and control measures are estimated according to their promotion rates based on the assumed environmental policies.

There are nine vehicle types contained in the MEIC, including four types of passenger vehicles (heavy-duty buses, HDBs; medium-duty buses, MDBs; light-duty buses, LDBs; and minibuses, MBs) and four types of trucks (heavy-duty trucks, HDTs; medium-duty trucks, MDTs; light-duty trucks, LDTs; and mini-trucks, MTs) as well as motorcycles (MCs). Additionally, passenger vehicles and trucks are further subdivided based on four fuel types, including gasoline, diesel, natural gas, and electricity. The provincial-level on-road transportation energy consumption by fuel type is obtained directly from GCAM-China (Table S1).

A vehicle fleet turnover model is developed at the provincial level to simulate future energy consumption and emissions for each vehicle type by tracking the lifespan of each vehicle. As shown in Fig. S4, the model is built to include the vehicle fleet turnover simulation and the evolution of emission factors. For a given year, the model first estimates newly registered vehicles using a back-calculation method based on total on-road energy consumption and historical vehicle registration data (Zheng et al., 2015). Then, the model estimates the number of vehicles that survive (called “in-fleet vehicles”) using historical and estimated vehicle registration data. Therefore, we derived the future's vehicle fleet and corresponding energy consumption for each vehicle type. Finally, we modelled the changes in emission factors of in-fleet vehicles, which are estimated by the product of the base emission factors and deterioration correction factor (Zheng et al., 2014). Unabated emission factors of in-fleet vehicles are determined by their registration year and the implementation year of vehicle emission standards (Fig. S5). To avoid double counting in the emission estimation, our model estimated tank-to-well emissions, which means that vehicles using electricity have zero emissions in on-road transportation (Huo et al., 2015).

Off-road transportation includes agriculture machinery, construction machinery, low-speed trucks, three-wheelers, locomotives, and inland waterways in the MEIC inventory. Because the total energy consumption for off-road transportation in GCAM-China is blended in the industrial sector, we estimated these values exogenously and subtracted them from the GCAM-China industry energy outputs. We assumed that the proportions of on-road and off-road total energy consumptions in the future are the same as the average historical rates during 2010–2015 (the historical rates varied from 0.192 to 0.203, and we used an average rate of 0.198). On the other hand, the electrification ratio of off-road energy was assumed to be similar to that of the on-road sector. Finally, the total energy consumption and structure for off-road transportation are estimated using the historical on-road and off-road split rate, the projected on-road energy consumption, and structures.

At the provincial level, future energy consumption of each off-road type is first estimated based on the annual average growth rate during 2010–2015. Under the constraint of total off-road energy consumption, in a given year, we distributed total energy consumption to each off-road type according their energy consumption shares (Table S1). The changes in emission factors are evaluated according to the reduction proportions caused by the upgrade in emission standards (Fig. S5). Here, we assumed that the proportion reduction in emission factors between adjacent emission standards in the future is the same as the mean proportion reduction estimated with the published emission standards.

Solvent use is identified as one of the major VOC emission sources, which refers to the applications of products containing solvents. Solvents include paints, adhesives, inks, textile coating, pesticides, industrial and domestic cleaning agents, and so on (Wei et al., 2014). In this work, 17 emission sources from solvent use in the MEIC model are further classified into paint use, printing use, pharmaceutical production, vehicle treatment, wood production, pesticide use, and household solvent use.

Paint use includes the paint applied to architecture, vehicles, wood, and other industrial infrastructure (Li et al., 2014, M. Li et al., 2019). The activity rates of different types of paint use are projected by building various regression models (Klimont et al., 2002; Wei et al., 2011). Specifically, architecture interior wall coating and other architecture paint use, as well as paint use for decorative wood and wood furniture, are projected based on the annual growth rate of newly built areas (Table S3). The amount of new car varnish paint and vehicle refurnish paint use is forecasted by developing the regression model linked with the newly registered vehicles and total vehicles, respectively (Table S3). The activity rates of other industry paint use are projected according to the annual growth rate of the above paint use (Table S3).

For solvent use other than paint, printing use (including printing ink and printing cleaning-gasoline solvent) and solvent use for wood production and pharmaceutical production are also forecasted by developing the regression models, and they are used to describe the relationship between national GDP and national amounts based on relevant statistical data from 1990 to 2015 (Table S3). Similar to vehicle paint use, the regression model of passenger vehicle treatment (for dewax or reseal) is also linked with newly registered vehicles (Table S3). Household solvent use here includes domestic solvent, dry clean C2Cl4 usage, and glue use, which we estimate through linkage with per capita GDP (Table S3).

Then, the changes in emission factors for various types of solvent use are estimated through the substitution rates of environmentally friendly products (including low-VOC and zero-VOC products) and the effective rates of recovery technologies (e.g. carbon adsorption, incineration, and membrane vapour separations) (Belaissaoui et al., 2016) according to the assumed emission standards and environmental policies.

The agricultural sector is distinguished as the main emission source for NH3 emissions (> 90 % of total NH3 emissions), including fertilizer use and livestock (Kang et al., 2016). The livestock category includes dairy cattle, other cattle, horses, donkeys, mules, pigs, goats, sheep, and broiler chickens in the MEIC emission inventory. For each type of livestock, we projected the future number of livestock by building the regression model, which is used to describe the relationship between national population and national annual amount based on relevant statistical data from 1990 to 2015 (Table S3). The projected population is obtained from GCAM-China. Then, changes in emission factors are evaluated by assessing the proportions of intensive farming systems (Xu et al., 2017).

Similar to the classification of fertilizer production, the regression model is developed to forecast the total fertilizer application, which is used to describe the relationship between national crop yield and total consumption of fertilizer (Table S3). The future national crop yield is estimated using the product of per capita crop yield and population (Ray et al., 2013). Then, four types of fertilizer use are estimated through multiplying the total fertilizer use by their shares in 2015 (Table S3). The changes in emission factors are modelled by evaluating the different promotion levels of slow-release fertilizer application.

The total power generation in China has significantly increased by 132.6 % during 2005–2015, which is primarily driven by population growth, industrialization and urbanization (NBS, 2006 and 2016; Tong et al., 2018a). However, up to 71 % of China's power generation was coal-fired in 2015. Given the dominant role of coal-fired power generation, China's government has promoted the development of clean energy in the power sector. Meanwhile, China has also undertaken great efforts to improve the efficiency of coal-fired power units by retiring small and inefficient coal-fired units and building large and high-efficiency units. The optimization of the generation unit fleet mix caused a significant decrease in the coal consumption rate by 91.3 gce kW h-1 (where gce is grammes of coal equivalent), which represents 22.4 % of the energy efficiency improvement achieved during 1990–2015 (Liu et al., 2015; Tong et al., 2018a), and the coal consumption rate decreased to 315.4 gce kW h-1.

The energy scenarios adopted in this study reflect different evolution of energy structure and power unit fleet in the power sector. Low radiative forcing targets represent aggressive low-carbon energy transformation required in the future, as well as advanced carbon removal technologies (e.g. carbon capture and storage, CCS). Under the SSP1-26 scenario, it is projected that the share of coal-fired electricity will decrease to 14.1 % in 2050 (including 8.3 % of the coal-fired with CCS electricity share) compared to 77.6 % in 2050 under the SSP5-85 scenario. Accordingly, the rapid reduction in coal-fired electricity implies the early retirement of the currently operating coal-fired power capacity in our turnover model. According to estimates, 65 % and 45 % of current coal-fired power capacity needs to be retired early (lifetimes < 40 years) fitting the power structure under the SS1-26 and SSP2-45 scenarios, respectively. The retirement rate for built power units derived from each energy scenario is used to simulate the evolution of the power unit fleet.

Meanwhile, five energy scenarios also reflect various efforts on future energy efficiency improvements. It is projected that the net coal consumption rate decreases of 12.7 %, 12.1 %, 11.3 %, 11.9 %, and 10.4 % during 2015–2050 will be achieved under the SSP1-26, SSP2-45, SSP3-70, SSP4-60, and SSP5-85 scenarios, respectively. The achievement of energy efficiency improvement relies not only on the optimization of the future power unit fleet but also on the application of advanced technologies for newly built units (e.g. ultra-supercritical technology). The penetration rates of advanced technologies are estimated and integrated into our projection model.

During 2005–2015, the energy consumption of China's industrial sector greatly increased by 35.7 % (from 1879.1 to 2922.8 million tce – tonnes of coal equivalent), which is driven by rapid industrial development and ever-increasing demand of energy-intensive products (NBS, 2007 and 2016). In contrast, the energy intensity per unit GDP in the industry sector rapidly decreased during the same period. In recent years, the Chinese government has greatly adjusted the industrial structure by phasing out outdated industrial technologies and capacities, especially in key industries (e.g. coal-fired boilers, steel and iron plants, and cement plants; Zhang et al., 2019a). Meanwhile, China aims to save energy through industrial energy transformation and energy efficiency improvement.

For industrial boilers, China first proposed eliminating coal-fired boilers with capacities smaller than 7 MW by the end of 2017 in the Action Plan. Energy is saved through both the replacement of large high-efficiency coal-fired boilers and switching to other clean-energy-fired boilers. The evolution of boiler fleet turnover under different energy scenarios was fully simulated in our projection model. Similar to the SSP1-26 scenario, the coal used in industrial boilers decreases rapidly, with 54.3 % of coal saved in 2050 compared to 2015. Accordingly, the reduction in coal use would reduce the capacity demand of coal-fired boilers and drive the early retirement (typical lifespan ∼ 20 years) of coal-fired industrial boilers, and it is estimated that the entire coal-fired boiler capacity (smaller than 24.5 MW) should be retired by 2050 and 75 % of large coal-fired boiler capacity (larger than 45.5 MW) would be built in our projection model.

Similarly, to reduce energy intensity, the outmoded production technologies would be replaced with more energy-efficient ones in the steel and iron industry and the cement industry. The facility fleet turnover is simulated under various retirement policies created from corresponding energy scenarios. As a result, under the SSP1-26 scenario, to restrict the development of energy-intensive heavy industry, the total production of steel and cement is projected to decrease by 55.3 % and 93.9 % during 2015–2050, respectively.

Residential energy consumption in China has steadily increased in the past few decades, driven by increases in total population and building areas (Wang et al., 2014). The total energy consumption in the residential sector increased by 44.9 % from 2005 to 2015, with a 4.2 % annual average growth rate (NBS, 2007, 2016). In recent years, the Chinese government has promoted a series of energy-saving measures to fight against air pollution from the residential sector, including the use of clean energy and clean use of coal (Shen et al., 2019). On the one hand, coal cleaning technologies (e.g. coal washing) are in widespread use, and the coal-washing rate is required to be up to 65 % by the end of 2015 according to the energy development of the 12th FYP. On the other hand, China aimed to switch residential coal to other types of clean energy (e.g. natural gas, electricity, or renewable energy). For instance, by the end of 2017, energy consumption in 6 million households in China (4.8 million households in BTH and surrounding regions) switched from coal to electricity and natural gas (Zhang et al., 2019a).

As estimated in GCAM-China and processed through our energy module, we estimated that approximately 63.0 and 27.3 million households nationwide would switch from coal to electricity and natural gas by 2050 under the SSP1-26 and SSP2-45 scenarios, respectively. Accordingly, 181.3 and 73.9 million tonnes of coal energy are saved by 2050 under the SSP1-26 and SSP2-45 scenarios, respectively, compared to the SSP5-85 scenario. Coal washing can not only substantially reduce air pollution emissions by lowering the ash and sulfur contents in coal but also improve the thermal efficiency. The shares of washed coal during 2015–2050 are estimated according to current energy-saving policies and assumptions of future energy-saving policies under different energy scenarios because GCAM-China cannot reflect these measures due to no specific coal classification. We assumed that a measure of coal washing would continue to be carried out in the future under the SSP1-26 and SSP2-45 scenarios to fit other energy-saving measures under low radiative forcing targets. We estimated that 85 % and 55 % of coal-washing rates would be achieved by 2050 under the SSP1-26 and SSP2-45 scenarios, respectively.

Attributed to the dramatic rise in the total number of vehicles, the energy consumption in China's transportation sector grew 104.9 % in total during 2005–2015 (NBS, 2007 and 2016). Energy consumed in the transportation sector can be saved by improving fuel efficiency and promoting electric vehicles (Wang et al., 2014). China has implemented fuel-efficiency standards for light-duty vehicles since 2004, and an updated standard issued in 2011 requires that the efficiency for passenger cars is up to 14.3 km L-1 by 2015. Meanwhile, China also promotes the development of electric vehicles, and it is reported that the total of electric vehicles reached > 3.1 million by 2019 (The Ministry of Public Security, 2020).

Energy scenarios from GCAM-China are adopted to estimate future vehicle fleet turnover. It is projected that the total vehicle population would be up to 2.64, 2.29, 2.06, 2.34, and 3.25 billion by 2050 under the SSP1-26, SSP2-45, SSP3-70, SSP4-60, and SSP5-85 scenarios, respectively. We found that the share of electric vehicles is as low as 13.8 %, even under the SSP1-26 scenario, which underestimates the future development of electric vehicles in China according China's 13th FYP development planning of electric vehicles. For consistency, we followed the energy structure and related assumptions from GCAM-China in our emission projections. The improvement in fuel efficiencies for each vehicle type reflected in GCAM-China is also estimated though energy consumption and projected vehicle kilometres travelled (VKTs), which has been integrated into our fleet turnover model. As a result, there is a slight but consistent improvement in the average fuel economy. Under the SSP1-26 scenario, fuel efficiency increases from 4.8 km L-1 in 2015 to 5.3 km L-1 in 2050 for heavy-duty diesel vehicles, from 3.8 km L-1 in 2015 to 4.6 km L-1 in 2050 for heavy-duty gasoline vehicles, and from 14.3 km L-1 in 2015 to 25 km L-1 in 2030 for light-duty gasoline vehicles and then remains steady.

Figure 3 shows the policy evolution under each emission scenario in the power sector. Under the BAU scenario, we assumed that all the coal-fired power plants would follow the emission limits of the standard GB 13223-2011 until 2050. The emission limits for SO2, NOx, and particulates are 100, 100 (200 for existing), and 30 mg m-3, respectively. Under the ECP scenario, China pledged an “ultra-low” emission standard in December 2015, the emission standard is strengthened to 30 for SO2, 50 for NOx, and 10 mg m-3 for particulates, and all the retrofits are achieved nationwide by 2020 and earlier in key regions. Under the BHE scenario, additional recommended BAT values (limits of 20 for SO2, 30 for NOx, and 5 mg m-3 for particulates) from the developed countries are considered after 2030 to achieve the 2050 air quality target. The policy evolution of other thermal power plants is similar to that of the coal-fired power plants except the policy effective year. Like under the ECP scenario, the additional special limits were enforced before the ultra-low emission standard between 2018 and 2024.

The industrial sector includes various subsectors, as described above. Figure 4 shows the evolution of emission control policies in different subsectors under three emission scenarios, as we can see, emissions from all the main industries are regulated through national emission standards, except the key VOC-related industries (i.e. the petrochemical industry), by 2015. All the regions would follow these current emission standards until 2050, and there are no specific regulations for key VOC-related industries under the BAU scenario. Furthermore, under the ECP scenario, ultra-low emission transformation is assumed to be completed in all industries by the end of 2030 except the key VOC-related industries. We also assumed the key VOC-related industries would reach low emission levels through current mature VOC-removal technologies (Fig. 4). More precisely, based on the ultra-low emission standards, all industries would achieve the BAT recommended values by the end of 2050 under the BHE scenario.

Different emission scenarios reflect to what extent the emission standards are strengthened; here, we comprehensively considered the evolution differentiations among subsectors and regions. First, we assumed that the completion year for each standard or policy in key control zones is a few years earlier than other regions in China according to the promulgated policies. For instance, a policy on ultra-low emission transformation in the iron and steel industry is implemented in 2019, which requires the completion of retrofits using the ultra-low emission technique in key regions (i.e. the BTH and Fenwei Plain, and the YRD region) by the end of 2025. Therefore, we assumed that all the ultra-low emission retrofits would be finished in other regions by the end of 2030. Second, subsectoral differentiation within the same policy is considered. Taking the ultra-low emission standard as an example, we assumed that the ultra-low emission standard would eventually be achieved in all industries. The ultra-low emission standard was first raised in coal-fired power plants, which was implemented in 2016 and would be completed nationwide by the end of 2020 as planned. Hereafter, the ultra-low emission standard for the iron and steel plants was issued in 2019, which required the retrofits to be completed to at least 80 % capacity nationwide by the end of 2025. Following the coal-fired power plants and iron and steel plants, we projected an ultra-low emission standard for cement plants that would then be proposed during 2020–2025, and retrofits would be accomplished by 2030.

There is no specific regulation in the residential sector before 2015; therefore, we assumed that emissions from the residential sector are not regulated under the BAU scenario (Fig. S6). Under the ECP scenario, clean coal and advanced stoves have been promoted to reduce emissions in recent environmental policies. We assumed continual upgrades for stoves and coal washing to reach relatively low emission levels through 2030. While under the BHE scenario, for the long-term air quality target, we supplemented the enhanced controls through innovations of stoves and residential coal stoves until 2050.

Emission reductions from the transportation sector are mainly achieved through fleet turnover in recent years, which means that old vehicles are being replaced by newer, cleaner models subjected to tougher emission standards (Zheng et al., 2018). Therefore, upgrading emission standards plays a vital role in reducing emissions. We modelled the evolution of emission standards for light-duty gasoline vehicles, and heavy-duty gasoline vehicles, light-duty diesel vehicles, heavy-duty diesel vehicles for on-road transportation and off-road transportation (Fig. S5). Under the BAU scenario, we assume that all the registered vehicles comply with the emission standards issued before 2017 and through 2050 with no more stringent emission standards. Therefore, China V emission standards for all on-road vehicles except heavy-duty gasoline vehicles (China IV) are implemented under this scenario. The China III emission standard for off-road transport is implemented. To reduce emissions, further implementation of China VI emission standards for all on-road vehicles and China V for off-road transport is assumed under the ECP scenario. Under the most stringent scenario (BHE scenario), “assumed China VII” emission standards for all on-road vehicles and China VI emission standards for off-road transport would be gradually implemented during 2030–2050.

Similar to the VOC-related industries, there are currently no specific regulations for controlling VOC emissions from solvent use. Therefore, we assumed that no effective regulations are implemented under the BAU scenario. Under the ECP scenario, to reach low emission levels of VOCs, we further improved the water-soluble solvent use and installed widespread VOC control facilities in the coating and painting industry. Note that emissions decrease to relatively low levels in the key control zones earlier than in the other regions. Under the BHE scenario, to maximally reduce VOC emissions, we considered the innovations of solvent use and VOC control facilities in the last 5 years before 2050 (2045–2050) according to the best-available technologies from developed countries (European Commission, 2016).

Agriculture is one of the least-controlled emission sources in recent years (Zheng et al., 2018). We assumed enhancement of NH3 controls under all emission scenarios except the BAU scenario. Before 2020, we promoted the use of organic fertilizer and resource utilization of poultry excrement and straw. During 2020–2030, we further reduced emissions through the enhancement of intensive cultivation and grazing and the promotion of slow-release fertilizer under the ECP scenario (Pan et al., 2016; Ju et al., 2019). Under the BHE scenario, except for the early achievement of relatively low emission levels nationwide, the innovation of cultivation and grazing is further considered during 2045–2050.