To be useful for our composite source, data have to meet some minimal requirements. Emissions data have inherent fluctuations due to weather (determining heating requirements), economic activity, and other factors. Not all sources model all these factors equally and therefore exhibit different fluctuations. When combining the sources, we use the year-by-year growth rates from the lower-priority source to extend a higher-priority source (for details see Sect.  and Appendices  and ). To weaken the influence of these fluctuations, we use the trend of several years for the matching instead of a single year. We therefore require that each time series contain at least three data points spread over a period of at least 11 years. Furthermore, we need time series with the detail of sectors and gases listed in Table .

Under the UNFCCC there are several requirements for reporting of greenhouse gas emissions data (see, e.g., ). Both developed (Annex I) and developing (non-Annex I) parties

The term “parties” refers to the countries which have ratified the UNFCCC. Annex I parties refers to those countries listed in Annex I of the Kyoto Protocol (KP) which are the developed countries under the UNFCCC. The definition is now almost two decades old and does not represent the state of economic development any more. The distinction between developed and developing countries is thus subject to constant debate in the UNFCCC meetings.

The EDGAR-HYDE 1.4 “Adjusted Regional Historical Emissions 1890–1990” dataset covers the gases CO2, N2O, and CH4 for the years 1890 to 1995 (). The data are given for 13 regions, some of which are individual countries (USA, Canada, Japan). They are generated from the EDGAR v3.2 dataset () and the “Hundred Year Database for Integrated Environmental Assessments” (HYDE v1.1) (). We use the EDGAR-HYDE dataset to extrapolate country emissions into the past. It has a relatively high sectoral detail, but the sectors differ from the IPCC 1996 definitions, so mapping to IPCC 1996 sectors is necessary. Details are presented in Appendix .

The two gridded datasets included in the generation of the PRIMAP-hist dataset do not contain any emissions data. Instead, they contain data for potential vegetation and simulation data of past existing vegetation. By comparing these, we can determine areas where deforestation has occurred, which we use to downscale the Houghton land cover change emissions data to country level. More information on the use of these datasets is provided in Sect. 4.2.2.

For fossil emissions, our highest priority source is the UNFCCC CRF data as they are both accepted by the countries that report and by other countries because they are reviewed by experts. However, these data are only available for developed-country parties. We use CRF2014 and fill gaps with CRF2013 where necessary. For non-Annex I parties we use data from national communications and national inventory reports with highest priority (UNFCCC2015). For a few developing countries, data from the biennial update reports (BUR2015) are available and fulfill our minimal requirements. These are used to supplement the UNFCCC2015 data. UNFCCC2015 is prioritized over BUR2015 because the latter only contains a few data points for most countries, while the UNFCCC2015 data contain full time series for more countries. These sources of UNFCCC-reported data cover a wide range of gases and sectors. For most countries, CO2, CH4, and N2O are available for all sectors at the level of detail needed for the composite source. Fluorinated gases are only contained for a few countries. For CO2 related to fossil fuel burning, CO2 from flaring, and CO2 emissions from mineral products, we use CDIAC as the next priority source. For CO2 from other sectors and all other gases we use a combination of EDGAR v4.2 FT2010 and EDGAR v4.2 as the next priority source.This combined EDGAR dataset is also used to complement CDIAC data where necessary (e.g., for small countries missing in the CDIAC source). BP2015 data are used to extend the energy CO2 time series until 2014. Where no country-reported data are available, the country-resolved data sources are used as the first sources.

Sources without country-level information, i.e., RCP and EDGAR-HYDE, are used to extrapolate emissions into the past. As EDGAR-HYDE has a higher regional and sectoral resolution it is used as the first priority source for extrapolation of CO2, CH4, and N2O emissions. Emissions from fluorinated gases for years before 1970 are only available from the RCP historical data and only on a global level.

The source prioritization for the individual gases is summarized in Tables , , , and . Details of the source creation methods are available in Sect. .

The first priority source for land use CO2 is FAOSTAT. However, it does not contain information for the period before 1990. EDGAR42 does contain information starting in 1970 but excludes sinks from the calculation of CO2 land use emissions, which is why we exclude EDGAR CO2 land use data from our dataset. The period before 1990 is covered by the Houghton dataset on a regional level, which we downscale using estimates of historical deforestation (see Sect. ).

For CH4 and N2O we use country-reported data, FAOSTAT, and EDGAR data on a per country basis. Regional growth rates from EDGAR-HYDE14 are used to extrapolate the time series.

Details of the source creation are presented in Sect.  and in Tables  and within that section.

The generation of the emissions time series is carried out using the composite source generator (CSG) of the PRIMAP emissions module described in . Data are aggregated on a per country, per gas, and per category level taking into account source prioritization (see Sect. ). The result is one time series for each country, category, and gas. The source creation is organized in the four steps described below.

Source preprocessing

First, each dataset undergoes source-specific preprocessing, e.g., category mapping and country preprocessing, which is explained in Appendix . This is followed by category aggregation: if data are defined on a more detailed level of gases (in the case of HFCs and PFCs) or categories (e.g., categories 4A and 4B), they are aggregated to the resolution described above for all sources individually. The aggregate time series covers the union of all years of the individual gas or sector series. If data are missing for some years in any of the individual gas or subcategory time series, they are interpolated to close gaps and extrapolated to fill missing data at the boundaries before aggregation. After aggregation, the information that a subcategory or gas was missing is lost. If data are missing at the gas and category level we are working with in the PRIMAP-hist dataset, they are not interpolated in preprocessing as they can be filled from other sources.

The largest share of emissions from land use, land use change, and forestry (LULUCF) is in the form of CO2 originating from deforestation.

The IPCC AR5 WG3 states that “fluxes resulting directly from anthropogenic FOLU (forestry and other land use) activity are dominated by CO2 fluxes, primarily emissions due to deforestation, but also uptake due to reforestation/regrowth”. ()

The dataset provides emissions time series for all UNFCCC member states. Some territories are associated with states but have partial independence, while other territories claim independence but are not internationally recognized, or have another special status. We include the emissions from these territories in the country emissions if, and only if, the country includes the emissions when reporting under the UNFCCC. Territories not included in the country reporting are treated independently. However, we cannot provide time series for all such territories. Territories which are uninhabited or have only very few inhabitants, e.g., in a research station, and with no significant emissions are completely excluded from the dataset (Bouvet Island, South Georgia and the South Sandwich Islands). In Table  we show which territories are included in countries, which are treated independently and whether data are available for those territories treated independently. The only territory that is not somehow associated with a single UNFCCC party is Antarctica. It is included in the dataset despite its negligible anthropogenic greenhouse gas emissions.

As a result of the Ukraine crisis, parts of the (former) Ukrainian territory are currently claimed by both Russia and Ukraine. The UN has not recognized any changes to the Ukrainian territory, so we do not make any adjustments to the Ukrainian emissions. There are no country-reported data recent enough to be influenced by the crisis.

We use territorial accounting in this dataset, meaning that emissions that originated from a territory that is now part of country A are always counted as emissions from country A even if the territory belonged to country B in the year the emissions took place. However, we can only be as precise as the datasets we are working with. Unfortunately, many sources are not very precise with respect to the methodology used. CDIAC CO2 and, to a lesser extent, FAO data are somewhat of an exception, where splitting up and merging of countries is made transparent by issuing different country codes. We sum and downscale the data to match the current countries according to the methodology described in Appendix . The CDIAC dataset also tries to account for land exchanges between countries. The CDIAC publication states that “land exchanges between countries were also accommodated, when possible. For example, the emissions from Alsace-Lorraine were included with Germany or France, reflecting which political unit governed these lands at any given time. This maintained the integrity of political entities despite changes in national borders.” This is not reflected in the country codes and thus remains in the final PRIMAP-hist dataset, in contrast to the territorial accounting used in our methodology. We cannot quantify the influence of this accounting discrepancy, because we do not know which regions were affected. However, as the land exchange including large emitters has been small in the recent decades and emissions were relatively low before the recent decades, the influence will likely be small. CRF2014, UNFCCC2015, and BUR2015 data are reported by countries and do not require preprocessing as we use the territorial definitions of the UNFCCC reporting as a basis. For EDGAR data, the rules regarding how emissions are assigned to countries in the case of territorial changes are not clear from the methodology description and we assume that territorial accounting is used.

For some small countries and countries that recently became independent, no emissions data are currently available. In this case we have to construct time series using other countries' emissions data. Emissions data for San Marino and the Vatican are included in Italian emissions data and downscaled using population shares.

GDP data not available.

Globally, production and consumption of fossil fuels is responsible for about two-thirds of current aggregate Kyoto greenhouse gas emissions

In the remainder of this section the term “emissions” refers to aggregate Kyoto GHG emissions.

The contribution of individual gases and gas groups to (global warming potential weighted) economy-wide (IPCC 1996 category 0) emissions is shown in Fig. . It is clearly visible that CO2 is by far the largest contributor, followed by CH4 and N2O, both globally and for individual countries. The contribution of fluorinated gases is, in general, small and negligible for developing countries. Again, China's emissions profile is closer to that of an industrialized country than to other major developing-country emitters. Economy-wide methane emissions are high for countries with a large agricultural sector (India and Brazil). Japan is somewhat of an exception with almost all emissions from CO2.

Uncertainty estimates for the CDIAC dataset of global CO2 emissions from fossil fuels and industry have varied since the first assessment made by , which resulted in an uncertainty range between 6 and 10 % (using a 90 % confidence interval). In a recent publication, a single global fossil fuel CO2 emissions uncertainty of 8.4 % (using a 95 % confidence interval) is offered as a reasonable combination of data (), in an attempt to simplify the different assessments and to make the best of the qualitative and quantitative knowledge developed since the first study of 1984.

Different approaches examine CDIAC global uncertainty as the aggregate of the uncertainties associated with fossil fuel CO2 emissions from individual countries. In a country grouping was introduced that uses seven classes of countries with “similar perceived uncertainty”. calculated uncertainty estimates for these groups, which are presented in Table .

The authors of the EDGAR dataset have stated that it was not feasible to go beyond the uncertainty tables compiled for EDGAR v2.0, where uncertainties are indicated in terms of ranges ranking from small (10 %) to very large (> 100 %) (). However, other institutions, such as UNEP (), estimated an uncertainty range of ±10 % (for a 95 % confidence interval) for total CO2 (including LULUCF). For global emissions of CH4, N2O, and fluorinated gases, uncertainties are estimated to be ±25, ±30, and ±20 %, respectively (using a 95 % confidence interval) ().

FAOSTAT land use emissions estimates are limited to only two carbon pools (above- and belowground biomass) out of six identified by the IPCC guidelines (above- and belowground, dead wood, litter, soil organic carbon, and harvested wood products). Therefore, FAOSTAT estimates greenhouse gas emissions and removals from land use are likely under-estimated ().

provides overall uncertainty estimates of the FAOSTAT database, where global emissions estimates from crop and livestock carry ±30 % uncertainty ranges. Uncertainties in the land use sector are even larger, with a ±50 % range.

Table  gives an overview of available uncertainty estimates for the individual sectors and gases included in the PRIMAP-hist dataset and how we calculated the indicative uncertainty range used in Figs.  and for different sectors and gases.

A different approach at uncertainty estimates is to compare different datasets. If they were completely independent, the distribution of emissions for the same category and gas should represent the uncertainties. This approach also captures uncertainties from different definitions of sectors, which are not included in the uncertainties of individual datasets. Some sources used in the PRIMAP-hist dataset depend on each other or may use common underlying data, so we cannot determine an upper bound on uncertainty but rather a lower bound. Adding independent sources would likely increase uncertainty. In Figs.  and we plot the composite source alongside some of the individual sources and other composite sources for individual gases and sectors at a global level. To compare the lower bound of the inter-source uncertainty to the individual source uncertainty, we also plot an indicative uncertainty range from Table  around the PRIMAP-hist dataset. For most categories and gases, it is apparent that the inter-source uncertainty is lower than the uncertainty estimated for the individual sources. However, as we have some source interdependence, we cannot conclude that the individual uncertainties are overestimated. Additionally, the number of sources is too small to reliably sample the 95 % confidence interval of the individual source uncertainty.

In the following, we investigate discrepancies between sources for total emissions, as well as individual gases and sectors, to analyze whether the discrepancies result from different assumptions and underlying data or lack of data for subsectors or individual gases. The EDGAR-HYDE data have relatively low total Kyoto GHG values. The sector plots show that this is due to low values for industrial processes and land use emissions. The low industrial process emissions can partly be explained by the lack of data for fluorinated gases in the EDGAR-HYDE dataset, but emissions of CH4 and CO2 are also low. Land use CO2 emissions in the EDGAR-HYDE dataset are only about half of the emissions of all other datasets assessed and outside of the sizable uncertainty range applied to the PRIMAP-hist time series. We should note that RCP, MATCH, and PRIMAP-hist include HOUGHTON data in their land use time series and are therefore not independent. The HOUGHTON-based time series are consistent with EDGAR42 and FAO, while the EDGAR-HYDE time series is not similar to any of the time series for more recent emissions.

A further major discrepancy is the RCP CH4 time series, which differs strongly from all other sources. Emissions are significantly higher than in other sources but show a steep decline between 1990 and 2000. No other source used in this analysis shows this effect. RCP CH4 emissions are based on , which uses EDGAR-HYDE but adds information for some sectors missing in EDGAR-HYDE14, namely grassland and forest fire emissions.

International shipping and aviation emissions are also added, but they are not included in this study.

For N2O, MATCH and EDGAR42 economy-wide emissions are lower than the PRIMAP-hist dataset while EDGAR-HYDE14 and RCP are higher. MATCH is based on EDGAR-HYDE growth rates prior to 1990, which explains the very similar pathway profiles and leads to very low emissions before 1970.

Finally, the estimates of emissions of fluorinated gases are higher for EDGAR42 than for our aggregate dataset in the period 2000–2014. This indicates that, for recent years, country reported fluorinated gas emissions are significantly lower than what EDGAR calculates the emissions to be.

Not all discrepancies between sources could be explained, and some are larger than the indicative uncertainty range for an individual source. This indicates that the actual uncertainties of emissions data could be even higher than what is assessed for individual sources.

The creation of this composite dataset implies several decisions on source prioritization, extrapolations, and downscaling options. These questions usually do not have one “correct” solution but rather different options with individual benefits and drawbacks. Different options (e.g., linear or constant extrapolation) have different implications for the calculated emissions, so the decisions introduce an “expert judgment uncertainty” to the final dataset. A further source of uncertainty is the use of regional growth rates for extrapolation. This assumes that all countries within that region shared the same growth rates, which is a simplification. Similarly, downscaling uses simplifications such as constant emissions shares or the use of another source as a proxy. We only use these methods if no individual country data are available and have to accept the uncertainty to fill gaps in data. See also Sect.  below.

The scaling of one source to another also increases the uncertainties associated with the final time series compared to the individual time series. The uncertainty of the final time series due to scaling can be calculated using standard error propagation formulas. For a scaling f, the standard deviation of a scaled time series C=f⋅B would be sc=sf2+sB2, where sB is the standard deviation of the time series B and sf is the standard deviation of the scaling factor, which depends on sB and sA in a manner determined by the exact matching algorithm (A denotes the time series which B is adjusted to).