To compile information on each individual power plant, including its name and exact geographical location, several public and commercial datasets were combined (Table 1). For the European database, the data sources used are the European Pollutant and Transfer Register database (E-PRTR_v18; EEA, 2020), the Large Combustion Plants database (LCP_v5.2; EEA, 2019), the Platts World Electric Power Plant dataset (WEPP Europe, September 2015, Platts, 2015) and the integrated Industrial Reporting Database (IRD_v7; EEA, 2022). For the non-European database, the datasets considered included the Global Coal Plant Tracker (GCPTv2021_01; GEM, 2021a), the Global Gas Plant Tracker (GGPTv2021_02; GEM, 2021b), the Global Power Plant Database (GPPDv1.3.0; Global Energy Observatory et al., 2021), the IndustryAbout database (IndustryAbout, 2021), the Emissions and Generation Resource Integrated Database (eGRIDv2018; US EPA, 2020), the Chinese Ministry of Ecology and Environment's domestic waste incineration power plant database (MIEE, 2022), the Tai biomass power plant database (DEDE, 2020), the Geocomunes Mexican power plant database (Geocomunes, 2020), the Taiwanese waste-to-energy plant database (Taiwan EPA, 2014), the electrical Japan power station database (Electrical Japan, 2022), the Argentinian renewable power plant database (MINEM, 2022) and the UNFCCC Clean Development Mechanism database (UNFCCC CMD, 2022). For both the European and non-European databases, substantial effort was put into identifying missing and incorrect facility geographical locations. Coordinates were checked or searched manually using Google Maps or other websites and added to the dataset as follows.

For Europe, the reported coordinates were consistently checked and corrected for the top 100 facilities (in terms of 2017 CO2 emissions). Furthermore, all coordinates that did not fall within the correct country borders or that were inconsistent between the reported dataset versions were manually checked and corrected. In addition, many other coordinates (likely about 400) were checked during the process of linking up facilities between datasets, identifying fuel types and looking at the resulting emission maps. In total, all the checks resulted in 360 plants with corrected coordinates, including about 75 of the top 100 plants.

Each of the emission values in the European power plant dataset is allocated to one of five fuel types (i.e. biomass, coal, oil, natural gas or waste). Three methods were used to allocate the fuel type.

Link to the LCP dataset: as LCP reporting includes the reporting of fuel input (but not for waste), this could be used to allocate emissions to different fuels when there was a link between E-PRTR and LCP facilities. Still, as only one emission value is reported, in the case of a multi-fuel plant (e.g. co-combustion of biomass in a coal-fired power plant), a proxy emission value for each fuel type was estimated using country- and fuel-specific emission factors from the Greenhouse Gas–Air Pollution Interactions and Synergies (GAINS) model at the International Institute for Applied Systems Analysis (IIASA) (Amann et al., 2011; Klimont et al., 2017). The ratio between the proxy emission values was then used to allocate the actual emission values to specific fuel types.

Country-dependent (state-dependent for the USA) and fuel-dependent monthly, weekly and hourly temporal profiles were constructed for all the power plants (i.e. European and non-European datasets) using the electricity production statistics summarised in Table 5. For countries where electricity generation statistics are not disaggregated by fuel type, we assumed the same temporal distribution for all types of power plants. For countries with no information on electricity generation or with information only available at e.g. the monthly scale but not at the hourly scale, averaged profiles from countries belonging to the same world region were used. The definition of world regions was taken from the EDGARv5 emission inventory (Crippa et al., 2018). The resulting profiles were assigned to each facility as a function of the country and fuel type information.

Figure 5 illustrates, on the one hand, the countries for which specific monthly, weekly and hourly profiles were constructed based on the statistics compiled and, on the other hand, the resulting share of total CO2 emissions for which specific monthly, weekly and hourly profiles were available. For the monthly profiles, the database constructed covers a total of 96 countries plus 42 USA states, which translates to more than 90 % of the total CO2 emissions from the power sector. For weekly and hourly profiles, the coverage in terms of total CO2 emissions is much lower (approximately 46 % and 36 %, respectively), partially because no information on electricity production at the daily and hourly levels was available for China. For this country, we assumed that the weekly cycle of emissions follows the pattern obtained for India, which shows no significant difference between weekdays and weekends. This assumption is in line with the results found by Wu et al. (2022) in which weekly profiles for Chinese power plants were constructed using measured emissions derived from continuous emission monitoring systems.

Hourly effective emission heights at the facility level were simulated by combining 2018 global hourly gridded meteorological information (i.e. air temperature at stack height, wind speed at stack height, surface temperature, boundary-layer height, friction velocity and Obukhov length) simulated by the Multiscale Online Nonhydrostatic AtmospheRe CHemistry model (MONARCH) at 0.3×0.3∘ (Badia et al., 2017) with facility-level stack parameter information (i.e. height, diameter, exit velocity and exit temperature). Information on stack parameters was obtained from the following sources.

The point source database of electric generation units (PTEGUs), obtained from the US EPA emission modelling platform (US EPA, 2021), which reports plant-level stack parameter information for US power plants.

In some European coal-fired power plants built in recent years, which must be equipped with a flue gas cleaning system, the cooling tower also takes on the function of the chimney. The original chimneys were dismantled, and now emissions are released through the cooling towers, which have different stack conditions. For Germany, we identified the list of power plants with cooling towers used as chimneys and the associated stack heights through Wikipedia (2022d), and we completed the information with the stack diameter, exit temperature and exit velocity reported by Brunner et al. (2019). This level of detail is not considered in facilities from other countries due to a lack of information.

Fuel-dependent and CO2-emission-weighted average stack parameters were calculated using the PTEGU dataset and assigned to all those facilities for which no specific information was found. For waste-to-energy power plants, we considered the stack parameters reported by Pregger and Friedrich (2009) as the PTEGU dataset does not include this type of facility. Table 6 summarises the stack parameters proposed per fuel type and the associated number of units considered to calculate the values.

Figure 6 illustrates, on the one hand, the facilities assigned specific (red circles) or emission-weighted averaged (white circles) stack height information and, on the other hand, the share of the total CO2 emissions from the power sector assigned specific stack parameter information. In terms of emission coverage, only 28 % of the total CO2 emissions from the power sector are assigned specific stack height values. This share significantly varies across world regions. In the USA, South Africa and India the share is between 75 % and 90 %, while in central Europe it is around 50 %. In many Asian, African and South American regions the share is below 5 %. The coverage of total CO2 emissions for stack diameter, exit velocity and temperature is even lower than for the stack height parameter (i.e. approximately 15 % globally in all the cases), the differences between regions being equally heterogeneous. These results indicate the current lack of stack parameter information.

The plume rise calculations at the hourly and facility level were performed using the High-Elective Resolution Modelling Emission System version 3 (HERMESv3) bottom-up emission system (Guevara et al., 2020), which includes plume rise formulas as described by Gordon et al. (2018). The HERMESv3 system was used to break down facility-level annual emissions into hourly resolution using of the temporal profiles described in Sect. 2.4 and to estimate hourly effective emission heights per plant considering the meteorological information provided by the nearest grid cell of MONARCH. Hourly plume top and plume bottom values per facility (htoph,f, hboth,f) were derived from the estimated effective emission heights following the expressions reported by Bieser et al. (2011) (Eqs. 1 and 2): 1htoph,f=hsf+1.5⋅Δh(h,f),2hboth,f=hsf+0.5⋅Δh(h,f), where hsf is the stack height of the facility f and Δh(h,f) is the modelled effective emission height for the facility f and hour h.

Figures 7 and 8 show the plant-level CO2 and NOx annual emissions as reported by the resulting global point source database. Results are distinguished by fuel type. It is observed that coal-fired power plants (red circles) are the main contributors to the total CO2 emissions, the top emitters being in China, India, the USA, Australia, South Africa, central Europe and Indonesia. CO2 emissions from natural gas power plants (blue circles) are dominant in Russia and some countries from the Middle East (e.g. Saudi Arabia and Iran). For NOx, the main contributors are also coal-fired power plants, but several oil-fired power plants (black circles) gain importance when compared to their contributions to the CO2 emission map, especially in the Middle East (i.e. Iran and Saudi Arabia), Indonesia, Venezuela and some countries in northern Africa. In China, India, the USA, Australia, South Africa and central Europe, NOx emissions are mainly dominated by coal-fired power plants. For both pollutants it is observed that the number of high emitters in Africa and South America is rather scarce, expect for South Africa and some countries in northern Africa as well as Venezuela. This is related to the fact that in both regions the electricity production is mainly dominated by renewable sources (e.g. hydro, solar) (IEA, 2021a). Linked to this aspect, it is interesting to see the large number of biomass power plants in Brazil (brown circles), as this fuel represents the second largest energy source in the country, just behind hydropower. A significant number of waste-to-energy plants (green circles) is reported in Japan and China, the two countries with the largest installed incineration capacity (Lu et al., 2017).

Tables 7 and 8 list the top 15 CO2- and NOx-emitting power plants worldwide and in EU-27 + UK.

At the global level, the Belchatów (Poland), Taean (South Korea), Taichung (Taiwan), Dangjin (South Korea) and Datang Tuoketuo (China) power plants are the top five CO2 emitters. These five facilities are also the five largest coal-fired power stations in the world (with installed capacities between 6700 and 5300 MW). All top 15 CO2 emitters are coal-fired power plants, except for Surgutskaya GRES-2 (Russia), which is the largest combined-cycle natural-gas-fired power station of Russia (8865 MW) and supplies energy to nearly 40 % of the population. Most of the top 15 CO2 emitters are in Asian countries, including South Korea (3), China (2), Taiwan (2), Malaysia (2), India (1) and Kazakhstan (1), while the rest are in Europe: Germany (2), Poland (1) and Russia (1). Seven out of the top 10 emitters identified in this work are also listed in the 2018 top 10 CO2-polluting power plants reported by Grant et al. (2021). In EU-27 + UK, it is observed that most of the 15 top CO2 emitters are in Germany (6) and Poland (3). Similarly to what is observed at the global scale, 14 out of the 15 facilities are coal-fired power plants, the remaining worst polluter being the Drax biomass power station, the largest power plant in the UK (3906 MW) that is also capable of co-firing petroleum coke. The largest emitter in EU-27 + UK (Belchatów, Poland) reports almost 5 times more CO2 emissions than the 15th facility (As Pontes, Spain).

For NOx, the list of top emitters mainly consists of coal-fired power plants (14 out of 15). Seven of these plants appear in both the CO2 and NOx top 15 emitter lists, including Surgutskaya GRES-2 (Russia), Taean (South Korea), Dangjin (South Korea), Manjung (Malaysia), Yeongheun (South Korea), Ekibastuz-1 (Kazakhstan) and Vindhyachal (India). Concerning the other top 15 emitters, 6 of them are located in South Africa and 2 in India. At the EU-27 + UK level, Belchatów is again the largest emitter. Four out of the top five emitters are in Germany, all of them being coal-fired power plants. There are also four Spanish facilities, three of them being oil-fired internal combustion engines located in the Canary Islands. The other non-coal facilities that complete the European top 15 list are Drax (UK) and Atherinolakkos (Greece), the latter also being operated with diesel engine units. Additional information on the total emissions obtained at the country level is provided in Sect. 3.2.

The estimated annual emissions were compared against other independent plant- and country-level inventories. The following sub-sections present and discuss the results.

Figure 15 illustrates with an example the plant-level hourly CO2 emission estimates (kg h-1) obtained when combining the information on total annual emissions with the temporal profiles reported in the resulting catalogue. We used the HERMESv3 emission system (Guevara et al., 2020) to combine the total annual emissions per facility (Sect. 2.3) with the corresponding country- and fuel-dependent profiles (Sect. 2.4) and to derive hourly emissions for the year 2018. To distribute the annual emissions to hourly emissions per facility, the following relationship is used in HERMESv3 (Eq. 1). 3Ep,t=Ep⋅Mp,m12⋅Wp,d⋅nd,m∑d=17Wp,d⋅nd,m⋅Hp,h24 Ep,t are the hourly emissions (kg h-1) for power plant p and date t (e.g. 27 November 2018, 17:00 UTC). Ep are the original annual emissions (kg h-1) for power plant p. Mp,m is the monthly weight factor [0–12] for power plant p and month of the year m. Wp,d is the weekly weight factor [0–7] for power plant p and day of the week d [1 Monday–7 Sunday]. nd,m is the number of day of the week d in month m, and Hp,h is the hourly weight factor [0–24] for power plant p and hour of the day h.

The results shown in Fig. 15 correspond to four coal-fired power plants: As Pontes (Spain), Belchatów (Poland), Jänschwalde (Germany) and Matimba (South Africa). The Matimba power plant is the facility that presents the flattest distribution, the results indicating that it is a base load power source. On the other hand, emissions from Belchatów, Jänschwalde and As Pontes present a clear seasonality, with emissions peaking during February, coinciding with a European cold spell that caused below-average temperatures in most European countries (C3S, 2018) and, in the case of As Pontes, also during summer, when energy demand increased due to the use of air conditioning systems. A weekend effect is also clearly observed for all the facilities, with emissions significantly dropping during Saturday and Sunday when compared to the weekdays.

Country-dependent monthly, weekly and hourly profiles were constructed using as a basis the estimated plant-level hourly emissions. The resulting emissions were aggregated at the country level and were normalised to derive the corresponding temporal profiles. Results were compared against the temporal profiles reported by Denier van der Gon et al. (2011) for the power industry sector (hereinafter referred to as the TNO profiles), which are widely used in the modelling community to quantify observation-based emission estimates (e.g. Kuhlmann et al., 2021). Figure 16 shows examples of monthly, weekly and hourly profiles constructed for the power sector for selected countries and comparison against the TNO profiles.

At the monthly level, large variations are observed between countries. Profiles for the United Arab Emirates (ARE) and Kuwait (KWT) present a clear peak during summer, coinciding with the intensive use of air conditioning systems. In the case of USA Pennsylvania (USA-PA), we identify two types of peaks, one related to space cooling needs during July and August and another one linked to space heating needs during January and December. In Germany (DEU) and Poland (POL), we also distinguish the peaks during wintertime, while the increase in emissions during summer is much lower than the previous cases as these countries are at higher latitudes where summers are not too hot. The seasonalities in India (IND), China (CHN), South Africa (ZAF) and Australia (AUS) are much flatter. The TNO profiles were designed for Europe, and the mismatch for countries with different climatic regimes such as the United Arab Emirates and Kuwait is to be expected. Nevertheless, we can see that all the profiles differ significantly with the TNO profile, which reports a V-shaped seasonality, with emissions peaking during wintertime, presenting their lowest value during summer and therefore not capturing the peak related to space cooling needs, which is also relevant in Europe.

Concerning the weekly variability, profiles constructed for the European countries (i.e. Germany and Poland) are in line with the TNO profile, showing a strong weekend effect, with emissions being reduced by more than 20 % between weekdays and Sundays. On the other hand, profiles estimated for USA Pennsylvania, South Africa and Australia are much flatter (5 to 10 % differences between weekdays and weekends), while India shows almost no differences between weekdays and weekends.

Finally, constructed hourly profiles are quite consistent between countries, all of them showing a rather flat variation, with emissions being slightly larger (10 %–15 %) during daytime (between 07:00 and 20:00 LST). Similarly to what we see for the monthly profiles, large inconsistencies are observed between the constructed profiles and the TNO profiles, the latter showing a much larger variation between emission levels during nighttime and daytime and not reproducing the afternoon peak reported by the constructed profiles in most of the countries.

The country-level monthly profiles derived in this work were also compared against the EDGAR temporal profiles (Crippa et al., 2020). We normalised the 2018 EDGARv7 CO2 monthly emissions reported per country for the energy sector, which are calculated using the temporal distribution profiles described in Crippa et al. (2020) and then compared against our profiles. Figure 17 shows the correlation values obtained between monthly profiles per country as well as the comparisons between monthly profiles for selected countries. Correlations are large in several of the top 20 emitting countries (e.g. China, 0.75; Australia, 0.95; Russia, 0.92; Japan, 0.78; Mexico, 0.87; South Korea, 0.80; Taiwan, 0.89; Turkey, 0.95; Kazakhstan, 0.89). Low or even negative correlations are observed in some of the top emitters, including India (r=-0.19), South Africa (r=0.16) and the United States (r=0.54). Nevertheless, when looking at the comparisons between monthly profiles reported in these three countries, we observe that both EDGARv7 and the present work suggest a very similar seasonality, with India and South Africa presenting a rather flat distribution and the United States showing two peaks in winter and summer, coinciding with the increase in electricity demand for space heating and cooling purposes, respectively. Brazil, Peru and Kuwait are among the countries presenting the largest negative correlation values (between -0.77 and -0.45). In the three cases, the seasonality reported by EDGARv7 and this work are completely opposite. While we suggest that emissions from the energy sector peak between June and September, coinciding with summer in Kuwait and southern winter in Brazil and Peru, EDGARv7 indicates that the largest emission levels occur between December and February (southern summer in Brazil and Peru). For these three countries, our profiles were constructed from national electricity generation statistics (see Table 5), while in the case of EDGARv7 they were indirectly estimated from regional averages computed using country-specific profiles belonging to the same world region. Most of the other countries for which a negative correlation exists are in Africa and Southeast Asia, where the information on electricity statistics to derive monthly profiles is rather scarce.

The temporal profiles constructed in the present work are country- and fuel-dependent but are not facility-dependent. Large differences between the emission temporal distribution of plants belonging to the same country may occur, e.g. base load versus peak load, or if they are used for electricity only or electricity and heat. Figure 18 illustrates these differences by comparing the monthly, weekly and hourly profiles constructed in the present work for German coal-fired power plants (black solid lines) against profiles estimated for individual facilities making use of plant-level electricity generation statistics provided by ENTSO-E (2021). The comparison includes the top 20 producing coal-fired German power plants, which together supplied more than 75 % of the national electricity from burning coal in 2018. Profiles from power plants are represented with different colours and sizes that indicate their annual electricity production (the thicker and darker the line is, the more the electricity that is supplied). Results indicate a significant heterogeneity between profiles across plants. As expected, the more electricity a power plant produces, the more continuously it supplies electrical energy throughout the year and, subsequently, the flatter its associated monthly, weekly and hourly profiles are. By contrast, power plants producing less energy tend to show large variations between e.g. weekdays and weekends or between daytime and nighttime hours, as their behaviour is more linked to demand response. The discrepancies between our profiles and the plant-level profiles tend to be lower when looking at the top generating facilities. Consequently, our profiles better represent the temporal behaviour of those facilities emitting more emissions and that subsequently are easier to detect by satellite observations. However, important differences are still observed when comparing the profiles from the catalogue with the ones derived from the largest generating plant (i.e. Neurath), with differences of up to 18 %, 10 % and 8 % for the monthly, weekly and hourly weight factors, respectively. The largest discrepancies occur with the monthly distributions as they are influenced by many factors besides the typical changes in electricity consumption (e.g. more demand during weekdays than during weekends), including economic variables, meteorological conditions and electricity trade.

Figure 19 shows an example of the daily bottom (blue) and top (red) plume values (m) at the Matimba (South Africa) and Bełchatów (Poland) coal-fired power plants estimated by the HERMESv3 model for the year 2018. Dashed lines indicate the stack height of each facility (i.e. 250 m for Matimba and 300 m for Bełchatów). Large month-to-month and day-to-day variations are observed for both the bottom and top plume heights at the two facilities, which are related to changes in the meteorological parameters and atmospheric stability driving the plume rise calculations, mainly the air temperature at the stack height and the boundary-layer height (Guevara et al., 2020). The bottom plume heights are, on average, 41 % (Bełchatów) and 70 % (Matimba) higher than the corresponding physical stack heights, while the top plume height values are on average 124 % (Bełchatów) and 206 % (Matimba) higher.

For each facility, the estimated CO2 hourly emissions were first uniformly allocated across 16 vertical layers (from 0 m up to 1500 m with breaks every 100 m and above 1500 m) considering the modelled hourly plume top and bottom values, and they were then summarised to the annual level and finally normalised to 1 to derive annual and emission-weighted vertical profiles. Figure 20 shows the emission-weighted average annual vertical profiles computed for the As Pontes (Spain), Belchatów (Poland), Jänschwalde (Germany) and Matimba (South Africa) coal-fired power plants. The estimated profiles are compared against the vertical distribution proposed by TNO in the Copernicus CAMS-REG inventory for the public electricity and heat production sector (Kuenen et al., 2022). Jänschwalde is the power plant with the largest share of emissions occurring in lower layers (i.e. 78 % of the total emissions allocated between 100 and 300 m). This is due to the fact that emissions from this facility are released through the cooling towers, which have a height of only 120 m. On the other hand, As Pontes is the facility with the largest share of emissions allocated between 400 and 600 m (76 %), as it is the power plant with the highest chimney in Europe (365.5 m). Belchatów and Matimba present rather similar vertical distribution profiles, partially because both facilities have stacks of similar heights (300 and 250 m, respectively). Matimba is the power plant allocating the largest share of emissions across the top layers (8 % of total emissions above 1000 m). This is related to the larger exit velocity of the gases when compared to e.g. As Pontes (i.e. 26 versus 21 m s-1, almost 25 % larger) and to differences in the local climatological conditions. The TNO profile distributes most of the emissions (85 %) between 200 and 500 m, the shares reported at higher altitudes (i.e. between 500 and 800 m) being considerably lower (8.5 %) than the ones computed for As Pontes, Belchatów and Matimba (between 32 % and 51 %). This discrepancy is in line with the fact that the TNO profile is derived from the work by Bieser et al. (2011), which assumed an average stack height of 159 m for calculating the plume rise in large power plants (>50 MW), a value significantly lower than the stack heights of the three power plants included in the comparison (365.5, 250 and 300 m). The results suggest that the TNO vertical profile may not be representative of power plants with high chimneys.

The current point source catalogue represents an effort to improve the spatial (horizontal and vertical) and temporal characterisation of emissions from CO2 and co-emitted species derived from power plants to be used for modelling efforts. Future work should focus on overcoming the limitations currently identified (see Sect. 5) and extending the temporal coverage to more recent years in order to capture, on the one hand, the impact of the decarbonisation efforts that are occurring in several countries and regions such as EU-27, the UK or the USA and, on the other hand, the large uptick in commissioning of new coal power plants that is happening in China (https://www.carbonbrief.org/mapped-worlds-coal-power-plants/, last access: January 2024). In parallel, other large CO2-emitting industries that are detected by satellite instruments, including cement and steel-and-iron plants, should be added in future versions of the global point source database.

The current catalogue does not report prediction intervals or standard errors of the estimated emissions for each plant. Hence, the uncertainty information is unknown. The comparison against independent inventories indicates good agreement for CO2 but also important discrepancies for NOx and SOx, highlighting that the co-emitted species estimates and their uncertainty deserve more attention in future research. This is in line with the fact that the estimation of NOx and SOx emissions is typically much more complex than for CO2, as there are more elements that influence the emission rates, such as combustion conditions, combustion technologies or air pollution control levels implemented in the facilities. Including information on the emission abatement technologies implemented in each plant could help define more detailed (i.e. technology-level) emission ratios for the estimation of co-emitted species. Unfortunately, none of the other plant-level emission inventories used for comparison includes information on co-emitted species, and therefore plant-level emission comparisons were limited to CO2. In this sense, performing intercomparisons against existing satellite-derived point source catalogues (e.g. Beirle et al., 2023; Fioletov et al., 2023) could also help to better constrain bottom-up emissions from co-emitted species.

The present work revealed that information on stack parameters is currently limited not only in developing countries but also in developed regions such as EU-27. Efforts should be made to compile this information from individual national environmental permits and to centralise it in a European database, at least for the large point sources considered under the European Industrial Emissions Directive (2010/75/EU). Furthermore, flagging of the power plants with channel emissions through cooling towers should be assessed to better represent the vertical distribution of these emissions.

Finally, future works will include performing study applications that show the impact of using this emission catalogue on modelling results.