Global and regional drivers of power plant CO2 emissions over the last three decades

a Department of Earth System Science, Ministry of Education Key Laboratory for Earth System Modeling, Institute for Global Change Studies, Tsinghua University, Beijing 100084, China b State Key Joint Laboratory of Environment Simulation and Pollution Control, School of Environment, Tsinghua University, Beijing, 100084, China c State Environmental Protection Key Laboratory of Quality Control in Environmental Monitoring, China National Environmental Monitoring Centro, Beijing 100012, China d Institute of Environment and Ecology, Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, 518055, China e Center of Air Quality Simulation and System Analysis, Chinese Academy of Environmental Planning, Beijing 100012, China f Department of Environmental Engineering, School of Energy and Environmental Engineering, University of Science and Technology Beijing, Beijing 100083, China


Introduction
Carbon dioxide (CO2) emissions from fossil fuel burning are recognized as one of major causes of the global temperature increase of approximately 1°C since the beginning of the industrial era [1][2][3]. As the largest source of anthropogenic CO2 emissions, the power sector accounted for 37% of global total anthropogenic CO2 emissions in 2019 compared to 30% in 1990, and it plays an increasingly critical role in global carbon commitment and climate change mitigation [4][5][6]. Global fossil fuel power generation grew from 7,609 TWh in 1990 to 17,642 TWh in 2019, at an annual average rate of 3.0%, driven by population growth and economic development [7]. Global CO2 emissions from the power sector thus have increased rapidly with the growth in demand for electricity generation in recent decades, and higher growth rates of electricity generation and power plant CO2 emissions have been observed in many developing countries [8][9][10]. However, the growth of power generation is likely to continue with the increase in electrification and the substitution of direct fuel consumption in end-use sectors with electricity [11][12][13][14][15]. The rapid decoupling of global power generation demand from its CO2 emissions is a necessary step in the coming decades to achieve the Paris Agreement of limiting the temperature increase to well below 2 ℃ above pre-industrial levels and pursuing 1.5 ℃ [16][17][18].
Over the past few decades, both developed and developing countries have made efforts to reduce CO2 emissions from the power sector [19][20][21]. A large number of national climate and energy policies have been implemented to reduce CO2 emissions [22][23][24][25][26], playing a vital role in tackling climate change [27,28]. For example, on the one hand, climate and energy policies of improving the thermal efficiency of fossil fuel power plants have been proven effective in reducing CO2 emissions of power plants in developed countries [29][30][31][32].
On the other hand, studies have also proven that fuel switching from coal to natural gas has also helped decrease CO2 emissions from the power sector [33][34][35]. Although these studies have partly analyzed regional and national drivers of power plant CO2 emissions, systematically assessing both global and regional drivers of power plant CO2 emissions are restricted by global uniform high-resolution power plants emissions database. Further, comprehensively understanding the global and regional drivers of power plant emissions could reveal how climate and energy policies can help decarbonize global power sector and support the future exploration of deep mitigation. 4 progress made by existing databases, it is valuable to develop a long time-series global power plant database to provide uniform and consistent data supporting the analysis of global and regional drivers of power plant CO2 emissions, as well as other scientific researches and the investigation of more innovative research topics.
To fill the gap in the development of time-series high-resolution power plant CO2 emissions database and identification of global and regional driving forces in power unit structure and emissions, we first construct an extended version of Global Power plant Emissions Database (named GPED v1.1), which is based on the integration of different available global and regional power plant databases. In our previous work [40,46], we developed a unit-based coal-fired power plant database for China which is named China coal-fired Power Plant Emissions Database (CPED). CPED represents more accurate information for coal power units over China than other global databases [40], but it is not integrated in GPED because GPED is a publically available database while information in CPED is not publically available due to restriction from the original data owners.
Instead, emissions in CPED were used to override GPED over China to support the analysis presented in this study. The combined database is in a significant position that can be used to comprehensively recognize global and regional driving forces of power plant CO2 emissions over the last three decades. Specifically, we assess multi-scale spatial and temporal changes in generating capacities, fuel types, unit sizes, age structure and CO2 emissions, as well as global and regional drivers of capacity evolution and CO2 emission changes from 1990 to 2019. We identify five factors contributing to emissions changes: power generation demand, fossil fuel share, fuel mix, energy efficiency and emission intensity, and highlight the future best opportunities in climate mitigation for the power sector. The new-built GPED v1.1 is developed on the basis of the extension of our previously developed GPED v1.0 [45], which encompasses more than 100,000 power units that burn coal, oil, natural gas, biomass, or other fuels operating during 1990-2019 worldwide. The basic information of power units, containing plant name, unit capacity, fuel type, starting year of operation, the year of decommissioning, and geophysical location, are completely derived in this database. The GPED v1.1 is developed by complying, combining, and harmonizing the available data related to power-generating units burning coal, natural gas, oil, biomass or other fuels. The diagram of the construction of the GPED v1.1 is presented in Fig. S1.
We begin by using multi versions of the WEPP databases [47] to compile unit-based information of generators in service and out of service during the period of 1990-2019, which provide information on the physical address, specific fuel type, installed capacity, status, starting year of operation, and retirement year of global power-generating units. Next, another database of Global Energy Monitor database (GEM) [48] is integrated to fill the missing unit information of physical address, installed capacity, status, and starting year of operation by mapping with the WEPP databases. Further, the GPED v1.1 combines and harmonizes the more comprehensive and reliable data contained in the national databases of the United States and India according to the data availability: the Emissions & Generation Resource Integrated Database (eGRID) for the United States [49] and the Indian Coal-fired Power Plants Database (ICPD) for India [50,51]. The eGRID is a comprehensive source of data from EPA's Clean Air Markets Division on the environmental characteristics of almost all electric power generated in the United States [49], including unit-level basic information and plant-level operation information (i.e. power generation) and CO2 emissions for multi years. It is noted that unit-level operation and emission information based on the plant-level information from the eGRID only in the year of 2010 are carefully derived previously [45]. The ICPD only includes generator-level operation information for Indian coal-fired power units in the year of 2010 [50,51]. Under the comprehensive consideration of dataset consistency, unit-level information availability and integration difficulty among different datasets, available derived unit-level data in the year 2010 for both the United States and India are integrated during the development of the GPED v1.1 currently. In summary, the GPED v1.1 presents an integration of the best available unit-level data we think, which is developed on basis of various global and regional power plant database, including global datasets of the WEPP database and the GEM database, regional datasets covering main countries of the United States and India.
For the geographical locations of power plants, which are unavailable from the global WEPP dataset. We first obtain the exact latitudes and longitudes for the power plants existing in our database of previous version [45]. For the remaining plants with a total capacity >10 MW, we geolocate them by searching data from the GEM database and Google Earth. The locations of the remaining and smaller plants are collected by directly mapping the physical addresses contained in the WEPP database to Google Maps following our previous study [43].

Estimates of CO2 emissions
Unit information related to the estimates of CO2 emissions from above-mentioned global and regional datasets is also integrated. Where available, we directly adopt unit-based estimates of CO2 emissions from where k, i, j, and m represent country, generating unit, fuel type, and year, respectively; E represents unitbased emissions (kg), A represents specific fuel consumption per unit (kg for solid-or liquid-fired units and m 3 for gas-fired units), and EF is the emission factor (g/kg for solid-or liquid-fired units and g/m 3 for gasfired units).
Because detailed activity data (i.e. unit-level power generation and fuel consumption) for each generating unit are not available, we thus estimate unit-based activity data from country-level power generation and fuel consumption. Unit-level fuel consumption is a function of power generation and fuel consumption per unit power generation, and power generation is again determined by the installed capacity and annual operating hours. But of these, only installed capacity data are readily available, we therefore first estimate unit-level power generation from country-level power generation. Country-level power generation data from 1990-2018 are obtained from the International Energy Agency (IEA) [7] and extended to 2019 by using the BP Statistical Review of World Energy [52]. To estimate unit-level annual power generation, unit-level information of annual operating hours (i.e. capacity factor) is first collected from regional databases of the eGRID and ICPD, and we then apply to other years due to data availability [45]. For the rest of power units not contained in those regional databases, the annual operating hours of power units burning the same fuel (65 fuel types) are consistent at the country level according to the simplifying assumption of our previous study [45]. Therefore, we calculate unit-level power generation using equation (2).
where i, k, j, and m represent the generating unit, country, fuel type, and year, respectively; G represents power generation; and represents the operating status according to the online year and retirement year of individual generating units. If the generator is operating, = 1; otherwise, = 0. C is the installed capacity of power units, and T is the annual operating hours.
Unit-level fuel consumption is further estimated starting from the country-level fuel consumption. As described above, country-level fuel consumption data from 1990-2018 are also derived from world energy statistics published by the IEA [7]. Based on energy consumption in 2018, we apply the growth rate from 2018-2019 by fuel type (coal, natural gas, and oil) and by country according to the BP Statistical Review of World Energy [52] to estimate 2019 energy consumption. Fuel consumption per unit power generated is inversely related to electric efficiency [43]. Instead, we directly adopt unit-based electric efficiency information from existing databases, which is applied for units for the whole period 1990-2019 as the electric efficiency is mainly related to electricity technology [45]. When detailed fuel consumption information (i.e. the electric efficiency) for remaining power units are not available from existing databases, we estimate electric efficiency by using the functions developed in our previous study [45] according to the nonlinear relationship between installed capacity and the electric efficiency of different fuel types. Here, therefore, unitlevel activity rates where unavailable are finally estimated from country-level fuel consumption according to equation (3) below.
where i, k, j, and m represent the generating unit, country, fuel type, and year, respectively; A represents country-level fuel consumption (kg for solid-or liquid-fired units and m 3 for gas-fired units); and represents the operating status according to the online year and retirement year of individual generating units. If the generator is operating, = 1; otherwise, = 0. C is the installed capacity of power units, T is annual operating hours, and e is electric efficiency.

CO2 emission factors are quantified according to guidelines from the Intergovernmental Panel on Climate
Change (IPCC) [53] using equation (4).
where j, k, and m represent fuel type, country and operating year, respectively; 2 represents the CO2 emission factor in g/kg; represents the carbon content in kg-C/GJ; represents the carbon oxidation factor; 44/12 is the molecular weight ratio of CO2 to carbon; and is the heating value in kJ/g for solid and liquid fuels and kJ/m 3 for gaseous fuels. In this study, the carbon oxidation factor is assumed to be 1, and carbon contents data are obtained from the IPCC guidelines [53]. The heat value of each fuel type and country is from the IEA [7].
In summary, CO2 emissions and related information of technology, activity data, operation situation, emission factor employed in the emission estimates for each individual unit in the GPED v1.1 database covering the period of 1990-2019 are derived in various ways, which are again combined with the unit-level emissions information contained in the CPED to decompose and characterize global and regional drivers of CO2 emissions.

CPED
The CPED includes detailed basic power plant information on the unit capacity, boiler type, operation and phasing-out procedures and geographical locations, as well as emission information on the activity data, operation situation, emission factors and CO2 emissions of China's individual coal-fired units covering the whole period of 1990-2019 [40,46,54], which consists of more than 9,000 coal-fired electric-generating units.
In detail, instead of modeling the related parameters of the activity rates, the annual coal use and power generation for each unit are directly available in the CPED, which can accurately reflect the differences of capacity factors and electric efficiencies among units. Again, annual CO2 emission factors are estimated by using the national heating values of coal, which characterized the annual changes of coal quality. Unit-level CO2 emissions are therefore estimated in a more accurate way. Unfortunately, information in CPED was not able to incorporated in to GPED due to the restriction of data sharing. The detailed unit-level information in CPED are then used to override information in GPED over China for this study, to represent the best knowledge of spatial and temporal evolutions of China's power unis and their emissions.

Uncertainty assessments
Uncertainty analysis is an important part of accuracy assessments of emissions inventories. Uncertainties in inventory can be caused by the incomplete information of fossil fuel consumption data, emission factors and other parameters. A comprehensive analysis of uncertainties in emissions is conducted at the national and unit levels using a Monte Carlo approach [55][56][57]. Monte Carlo simulations are employed to propagate the uncertainties induced by both fossil fuel consumption and emission factors to provide the uncertainty ranges for emission estimates. For uncertainties in national emissions, we first assume probability distributions for both fossil fuel consumption and emission factors. Then, random sampling of both the activity data and emission factors is conducted 10,000 times, generating 10,000 estimations of CO2 emissions. The uncertainty range in this study is estimated by the lower and upper bounds of 95% confidential intervals around the central estimate of emissions [58]. The probability distributions and coefficients of variation (CVs, equal to one standard deviation divided by the mean) of the parameters are obtained from previous studies [36,39,40,45].
From the perspective of unit-level emission estimates, uncertainties associated with input parameters may vary over time and by country due to the different accuracies of information from global and national databases.
Countries without available national databases could have higher uncertainties than countries with higherquality data sources. Following the method used in our previous study [40,45], we randomly select one large coal-fired unit (≥300 MW) from nine key regions to demonstrate that the emission uncertainties differ among regions. The uncertainties can be considered larger for a coal power-generating unit operating in 1990 than for one operating in 2019 because the accuracy of unit-level information improved over time. We quantify the emission uncertainties of the selected coal power units for 1990 and 2019 to demonstrate the changes in uncertainties over time. We assume that both the unit-level energy consumption and emission factors follow a normal distribution with the CVs, as discussed above.

Decomposition of emission drivers
Decomposition analysis methods have been widely used to quantify the contribution of socioeconomic drivers to changes in environmental pressures [59][60][61][62]. The two most popular decomposition approaches are index decomposition analysis (IDA) and structural decomposition analysis (SDA). Compared to SDA, which is based on input-output tables [63,64], IDA is more suitable for time-series energy and emission studies [65,66]. Among IDA methodologies, the logarithmic mean Divisia index (LMDI) has been shown by past studies to be favorable because of its path independence, consistency in aggregation, and ability to handle zero values [67][68][69]. In this study, we choose LMDI to identify how each driving factor contributes to the changes in CO2 emissions. The drivers are classified as power generation demand, fossil fuel share, fuel mix, generation efficiency and emission intensity, as shown in equation (5).
where n and i represent region and fuel type, respectively; nine regions are included in this study (see Fig. S2).
Fuel type includes coal, oil, natural gas, biomass and others. E represents CO2 emissions, G represents power generation, Q represents power generation from fossil fuels, and A represents energy consumption.
Hence, the arithmetic change in total emissions from year t+1 to year t (∆ ) is decomposed as follows:

Evolution of technologies
The capacity of global fossil fuel and biomass-fired power plants experienced a substantial increase during the past three decades. Figure 1

CO2 emission trends
Here, we conduct a multi-scale analysis of the changes in the characteristics of CO2 emissions for global power plants. Section 3.2.1 presents global and regional CO2 emission trends. Sections 3.2.2 and 3.2.3 present the evolutions of age-based CO2 emissions, and identify those low-efficiency units based on unit-level quantification, respectively.

Low-efficiency power units
High-resolution emission information can help us identify the detailed evolutions for each unit. Figure   5 shows the relationship between power generation and annual CO2 emissions from coal-fired units in China, India, Europe and the United States, highlighting the evolutions of low-efficiency units between 1990 and 2019, which we define as those units whose emission intensity (i.e. tons CO2 per MWh) is more than 90th percentile greater than the average emission intensity in 1990 in the same region. Across all regions, a large fraction of total CO2 emissions was produced by a disproportionately small fraction of total power generation.
For instance, 4.3% and 0.6% of total power generation from coal-fired units in the United States and Europe

Drivers of CO2 emissions
The characteristics of age-based CO2 emission distributions and unit-level emission intensities had revealed the different driving forces in changes of CO2 emission trends at the global and regional scale to some extent, and we further conducted a systematical decomposition of global and regional drivers. Figure 6 shows the effects of power generation demand, energy efficiency, fuel mix and fossil fuel share on CO2 emissions during 1990-2019, as well as the regional contribution to the changes in the emissions. Overall, We further compare the regional drivers of power plant CO2 emissions in the 1990-2000 (Fig. 6b)

Uncertainty and comparison
Uncertainty both at the country and unit level is quantified in this study. The gray area in Fig uncertainty ranges than those in Europe due to the availability of unit-level data (e.g., -11.4%-11.9% for the selected coal power unit (≥300 MW) in China compared to -22.5%-24.2% in Europe). The unit-level uncertainty ranges in China, India and the United States were smaller than the global average uncertainty ranges due to the application of regional databases, whereas other regions corresponded to higher uncertainties because some key parameters (e.g., activities and efficiency) were derived from extrapolations and assumptions.
In addition to the uncertainties of fossil fuel consumption and emission factors considered in the Monte

Conclusion and discussion
The capacity of global fossil-fuel-and biomass-fired power plants experienced a substantial increase, driven by the growing demand for power generation during the past three decades. CO2 emissions increased from 7.5 Gt in 1990 to 13 coal to natural gas helps to achieve short-term emission reductions [33][34][35]. However, for the long-term future, in most scenarios, accomplishing a global transition to energy systems with net-zero emissions may require a large share of renewable electricity (i.e., solar and wind resources) [70][71][72]. Developing countries could switch directly from coal power to renewables rather than using natural gas or other fossil fuel given the rapid decrease in renewable electricity costs and the pressure of rapid transitions [73][74][75]. The expansion of renewables will thus likely represent an increasingly significant factor driving future emission reductions in the power sector, which could be captured and evaluated using our data-driven assessment in future. In summary, our combined databases could further contribute to applications related to climate change mitigation, facilitate multiple research perspectives for global environmental issues and policy making, and enhance our abilities to track emission mitigation progress toward sustainable power systems and support effective strategies for future emission mitigation.

Data availability
For the company name, plant name, plant location, number of power generating units, CO2 emissions at the unit-level contained in the GPED is available at: http://gidmodel.org.cn/dataset-gped. Other information at unit-level is obtained from commercial database and not publically available. For the database CPED, unitlevel information is not publically available due to restriction from data providers.            The shaded ranges illustrate the uncertainty range of the 95% CI calculated in this study.