At the EU28 level, GHG emission reporting is mandatory for all countries and is done under the consistent framework of UNFCCC. Every year in May all EU parties report to the convention their National Inventory Report (NIR) and provide data using the standardized common reporting format (CRF) tables. The NIRs contain detailed descriptive and numerical information on all emission sources and the CRF tables contain all GHG emissions and removals, implied EFs, and AD for the whole time series from 1990 to 2 years before the submission year (https://unfccc.int/process-and-meetings/transparency-and-reporting/reporting-and-review-under-the-convention/greenhouse-gas-inventories-annex-i-parties/national-inventory-submissions-2018, last access: February 2020). It is important to note that the 2006 IPCC GLs used for this process do not provide methodologies for the calculation of CH4 emissions and CH4 and N2O removals from agricultural soils and field burning of agricultural residues. Parties that have estimated such emissions should provide, in the NIR, additional information (AD and EF) used to derive these estimates and include a reference to the section of the NIR in the documentation box of the corresponding sectoral background data tables.

Further in this section, we present estimates of CH4 and N2O agriculture fluxes during the period from 1990 up to the last available year reported by each of the data sources. The detailed values for the last available year are shown in Appendix A, Table A2.

In recent assessments of the global CH4 budget (Saunois et al., 2019), wetlands CH4 emissions from top-down and bottom-up estimates for the period 2008–2017 are statistically consistent and average 178 Tg CH4 yr-1 (range 155–200) and 149 Tg CH4 yr-1 (range 102–182), respectively (Saunois et al., 2019).

In the EU28, natural emissions of CH4 are represented by wetlands which are not yet fully accounted for and reported under NGHGIs, their emissions reporting being only recommended under the 2013 IPCC Wetlands Supplement (IPCC, 2014) complement to 2006 IPCC GL. According UNFCCC NGHGI (2019), between 2008 and 2017, the natural CH4 emissions in the EU28 reported under LULUCF (CRF Table 4(II) accessible for each EU28 country

https://unfccc.int/process-and-meetings/transparency-and-reporting/reporting-and-review-under-the-convention/greenhouse-gas-inventories-annex-i-parties/submissions/national-inventory-submissions-2018, last access: February 2020

Wetlands are sinks for CO2 and sources of CH4. Their net GHG emissions therefore depend on the relative sign and magnitude of the land–atmosphere exchange of these two major GHGs. Undisturbed wetlands are thought to have a large carbon sequestration potential because near-water-logged conditions reduce or inhibit microbial respiration, but CH4 production may partially or completely counteract carbon uptake (Petrescu et al., 2015). The net GHG balance of natural wetlands is thus uncertain. Natural emissions of CH4, in particular wetlands and inland waters and their net GHG balance, are the most important source of uncertainty in the methane budget (Saunois et al., 2019), due to the GWP100 of CH4 and the generally opposite directions of CO2 and CH4 fluxes.

Under the new EU LULUCF Regulation article 7 (footnote 7), the accounting of natural wetland emissions will become mandatory from 2026 onwards; i.e., the reported numbers will be compared to numbers already reported under category 4(II) wetlands between 2005 and 2009, and the net difference will count towards reaching the EU climate targets.

Since CH4 emissions are highly variable in time and space as a function of climate and disturbances, it makes EF-based methods impractical and national budget estimates difficult, making it challenging to accurately estimate CH4 emissions in NGHGIs. There is also a risk of double counting with emissions from inland waters as discussed, e.g., by Saunois et al. (2019) for the global CH4 budget. The sum of all natural sources of CH4 as inferred by different models may be too large by about 30 % compared to the constraint provided by global inversions. The spread of wetland emissions from process-based wetland emission models used in the global CH4 budget (Poulter et al., 2017) forced by the same variable flooded area dataset is 30 % (80 Tg CH4 yr-1) globally (given their estimated emissions of 177–284 Tg CH4 yr-1 using bottom-up modeling approaches), up to 70 % for the EU28 calculated based on the model-to-model variability and even larger at a national scale. In the absence of any better information, we used in this study the results of these ensemble models (see Appendix B) to provide a first estimate of this source.

According to Poulter et al. (2017), between 2005 and 2017, the total wetland CH4 emissions in the EU28 averaged 3 Tg CH4 with an uncertainty (1σ spread) of 70 %, with seven countries having the highest emissions (Fig. 9). Finland, Italy, Sweden, UK, France, Greece and Germany accounted for 75 % of total EU28 wetland CH4 emissions. For the same period, UNFCCC NGHGI 2019 reports an average of 10.34 kt CH4 (0.01 Tg CH4), a highly underestimated value compared to the modeled results, due to nonreporting and accounting under NGHGIs.

Given this current gap between modeled and NGHGI reported data on CH4 emission from wetlands in the EU28, we stress the need of investing in better modeling methodologies for emission calculation and verification. Out of all EU28 countries, for the purpose of reporting, only Finland developed its own biogeochemical CH4 model to provide to NGHGIs a very detailed list of estimates for all CH4 subactivities.

The forestry and other land uses, referred to here as the LULUCF section, include CO2 emissions and removals from forests (including soils and harvested wood products) and soil organic carbon (SOC) changes from grasslands and croplands. A comprehensive assessment of the overall carbon stocks and fluxes of forests would need to be complemented by the analysis of climate change impacts on forest productivity and composition (Lindner et al., 2015). Several studies analyzed the European forest carbon budget from different perspectives and over several time periods using GHG budgets from fluxes, inventories and inversions (Luyssaert et al., 2012), flux towers (Valentini et al., 2000), forest inventories (Liski et al., 2000; Nabuurs et al., 2018; Pilli et al., 2017), and IPCC GLs (Federici et al., 2015).

Achieving the well-below-2 ∘C temperature goal of the PA requires, among others, negative emission technologies, low-carbon energy technologies and forest-based mitigation approaches (Grassi et al., 2018a; Nabuurs et al., 2017). Currently, the EU28 forests act as a sink, and forest management will continue to be the main driver affecting the productivity of European forests for the next decades (Koehl et al., 2010). Forest management, however, can enhance (Schlamadinger and Marland, 1996) or weaken (Searchinger et al., 2018) this sink. Furthermore, forest management not only influences the sink strength, it also changes forest composition and structure, which affects the exchange of energy with the atmosphere (Naudts et al., 2016) and therefore the potential of mitigating climate change (Luyssaert et al., 2018; Grassi et al., 2019).

We compared net CO2 emissions and removals from the LULUCF sector reported by UNFCCC NGHGI (2018) to those included in FAOSTAT and to the carbon balance here termed as the net biome production (NBP) from different models (Table 4). Categories presented in this study are forest land, cropland and grassland. We present separately the results from forest land and land use, because some models (e.g., CBM and EFISCEN) use a different definition of forest land than the DGVMs ensemble TRENDY (Sitch et al., 2008; Le Quéré et al., 2009) or bookkeeping models (Houghton and Nassikas, 2017; Hansis et al., 2015).

To better illustrate differences between estimates we exemplify how four of the data sources interpret and calculate the NBP:

UNFCCC NBP definition depends on the method used by each country.

Cropland and grassland (CL and GL) (in UNFCCC NGHGI, 2018, IPCC sector 4B and 4C, respectively) include net CO2 emissions/removals from soil organic carbon (SOC) under the remaining and conversion categories. Similar to FL, fluxes include effects from both environmental changes and from land management and land use change. FAOSTAT GHG emissions in the domain cropland and grassland are currently limited to the CO2 emissions from cropland/grassland organic soils associated with carbon losses from drained histosols under cropland/grassland. This can be one of the reasons for differences between estimates reported by the two sources (Fig. 11).

The cropland definition in IPCC includes cropping systems, and agroforestry systems where vegetation falls below the threshold used for the forest land category, consistent with the selection of national definitions (IPCC glossary). According to EUROSTAT, the term “crop” within cropland covers a very broad range of cultivated plants. In 2015 more than one-fifth (22 %) of the EU28's area was covered by cropland (EUROSTAT, available at https://ec.europa.eu/eurostat/statistics-explained/index.php/Land_cover_statistics, last access: January 2020). Denmark (51 %) and Hungary (44 %) had the highest proportion of their area covered by cropland in 2015. For the vast majority of the EU member states (MS), cropland accounted for between 15 % and 35 % of the total area, with this share falling to 10 %–15 % in Latvia, Estonia and Portugal, while the lowest proportions were registered in Slovenia (9 %), Finland (6 %), Ireland (6 %) and Sweden (4 %). In absolute terms, France, Germany, Spain and Poland had the biggest areas of cropland in 2015.

Grassland definition in IPCC includes rangelands and pasture land that is not considered cropland, as well as systems with vegetation that fall below the threshold used in the forest land category. This category also includes all grassland from wild lands to recreational areas as well as agricultural and silvopastoral systems, subdivided into managed and unmanaged, consistent with national definitions. Grasslands tend to be concentrated in regions with less favorable conditions for growing crops or where forests have been cut down. Some of these are found in northern Europe (e.g., Finland and Sweden), while others are in the far south, i.e., the south of Spain.

In 2015 just above one-fifth of the EU28's area (21 %) was covered by grassland. There is a broad range across EU member states, with Ireland having 56 % of its total land area as grassland and Finland and Sweden less than 6 % of the land (EUROSTAT, https://ec.europa.eu/eurostat/statistics-explained/index.php/Land_cover_statistics, last access: January 2020).

Figure 11 shows that in the EU28 croplands and grasslands are CO2 sources to the atmosphere in the UNFCCC NGHGI (2018) and FAOSTAT databases. Cropland CO2 emissions are rather stable with time and are in good agreement between FAOSTAT and UNFCCC, except in 1990. Grassland emissions reported by countries to UNFCCC are higher than the FAOSTAT and show an abrupt increase in 2016 compared to the previous years. The high estimates of grassland emissions in 2016 UNFCCC NGHGI submissions are explained by increased emissions in Austria, Denmark and Croatia; Sweden changed from being a sink in 2015 to being a very high source in 2016, and Hungary and Greece reported a lower sink. Ireland was the only country which reported a higher sink in 2016 compared to 2015.

Climate change and climate effects on soil temperature and moisture are key drivers in the 21st century increase in soil decomposition and decrease in the soil carbon stock (Smith et al., 2005). Avoiding soil carbon losses or restoring stocks requires practices that increase C input in excess of losses from erosion and decomposition, such as diminished grazing intensity for grasslands, higher return of residues or reduced tillage for croplands, and manure additions for both. Further change in land use and management will also affect the soil carbon stock of European cropland and grasslands (Smith et al., 2005).

Land-related carbon emissions can also be estimated by global models such as DGVMs (here we used the TRENDY v6 ensemble) and two bookkeeping models (BLUE and H&N). In this section we compare these global model results with data from FAOSTAT and UNFCCC NGHGI (2018). There is significant uncertainty in the underlying datasets of land use changes, the coverage of different land use change practices and the calculation of carbon fluxes. In addition, marked differences in definitions must also be considered to compare independent estimates. Bookkeeping models give net emissions from land use change, including immediate emissions during land conversion, legacy emissions from slash and soil carbon after land use change, regrowth of secondary forest after abandonment, and emissions from harvested wood products when they decay. DGVMs estimate net land use emission as the difference between a run with and a run without land use change, and their estimate includes the loss of additional sink capacity, that is, the sink that favors the environmental changes (e.g., CO2 fertilization). This sink created over forest land in the simulation without land use change is “lost” in the simulation with land use change because agricultural land lacks the woody material and thus has a higher carbon turnover (Gasser et al., 2013; Pongratz et al., 2014). This different definition from bookkeeping models historically implies higher carbon emissions from DGVMs, even if all postconversion carbon stock changes were the same in DGVMs and bookkeeping models.

The key difference between DGVMs and bookkeeping models, on the one hand, and FAO and UNFCCC methodologies, on the other, is that the latter are based on the managed land proxy (Grassi et al., 2018a) (Fig. 12).

Land fluxes can be differentiated into three processes (IPCC, 2010): (1) direct anthropogenic effects (land use and land use change, e.g., harvest, other management, deforestation), (2) indirect anthropogenic effects (e.g., changes induced by human-induced climate change, including CO2 fertilization and nitrogen deposition changes), and (3) natural effects (i.e., that would happen without human-caused climate change, such as natural disturbances).

Models and GHGIs capture these effects in a different way:

Biogeochemical models. Bookkeeping approaches only estimate direct anthropogenic effects. DGVMs also consider fluxes linked to indirect effects and natural processes. In the GCB 2018 (Le Quéré et al., 2018b) and GCB 2019 (Friedlingstein et al., 2019), the fluxes associated with the direct anthropogenic effects are estimated with bookkeeping models and DGVMs, while the remaining land sinks (including all indirect and natural effects) are estimated by DGVMs.

The difference between biogeochemical models and NGHGIs of around 4–5 Gt CO2 yr-1 globally is to a large part attributable to the accounting of indirect effects on greater-area managed land by NGHGIs compared to models (Grassi et al., 2018a; IPCC, 2019b). The differences at the EU28 level are smaller, because most forest land is considered managed by both models and NGHGIs.

Independent estimates of the land-related flux for the EU28 are presented in Fig. 13. The data behind the three main estimates, bookkeeping models, NGHGIs and FAOSTAT represent the total net land use emissions/removal from forests, including conversions to and from one category to another. Next to them, we plotted each of the net land use change fluxes (in grey; difference of simulation with and without land use change) from eight of the TRENDYv6 DGVMs used in the GCB 2017 (Le Quéré et al., 2018a) with their mean, as they mostly simulate the indirect and natural sink considered unmanaged. FAOSTAT includes emissions from peatland drainage and fires and from biomass fires (not considered herein). It does not include however other carbon stock changes in cropland and grassland. We additionally excluded from the UNFCCC estimate the categories wetlands remaining wetlands and settlements remaining settlements, as well as biomass burning and drainage and transitions between nonforest lands.

The UNFCCC NGHGI (2018) and H&N's estimates are similar because the managed areas for the EU28 are similar in both estimates (Grassi et al., 2018a). Differences between the two bookkeeping models, BLUE and H&N, relate to the different input data applied by each of the models and differences in biome types. The input used by H&N is based directly on FAOSTAT agricultural and wood harvest data and FRA forest area changes, while BLUE uses LUH2 (Hurtt et al., 2011, 2020). LUH2 is based on HYDE3.2 (Klein Goldewijk et al., 2017a, b), which provides annual, 0.5∘, fractional data on cropland and pasture based on FAOSTAT but overlays subgrid-scale transitions between all land use types and wood harvesting. H&N allocates pasture expansion preferentially on natural grasslands, while all available vegetation types of a grid cell are assigned proportionally to agricultural expansion in BLUE. Carbon densities and regrowth and decay curves are structurally similar but differ in detail.

The EU28 has a very small area of unmanaged land and this denotes that most of the LULUCF emissions in the EU28 are from direct effects in the forestry sector (including agricultural expansion/abandonment). According to FAOSTAT and UNFCCC NGHGIs, the net forest conversion is relatively small in the EU, so the simulations include mostly managed net area.

DGVMs differ strongly in their estimate of the net land use change flux due to different comprehensiveness of including land use practices such as wood harvesting, shifting cultivation, or fire management (Le Quéré et al., 2018a); different land use change datasets (HYDE3.2 or LUH2) and their implementation; and general model differences of how photosynthesis, respiration, and natural disturbances are simulated. Most striking in comparison to the other, more empirical, approaches is the large interannual variability, related to the climate dependency of vegetation processes. Though DGVMs are conceptually similar to NGHGIs in simulating all indirect and direct fluxes on a given area, differencing of the simulations with and without land use change leaves only the land-use-related effects to be attributed to the net land use change flux (see Fig. 12). DGVMs are thus closer to the bookkeeping definition of LULUCF emissions, apart from differing assumptions on environmental changes (constant in bookkeeping, historical in TRENDY) and the loss of additional sink capacity included in DGVMs.

At the European level the largest inconsistencies between estimates from AFOLU emission sources/sinks were found to be mainly caused by the use of different methodologies, including use of different AD and/or tier level. When looking at final emission estimates, inconsistencies in methodology and tier application in calculating emissions give as much as 10 %–20 % variation across estimates (e.g., CH4 from agriculture). Higher tiers require more detailed AD for calculating emissions/removals from AFOLU sectors.

Within the UNFCCC practice, for agriculture, each country uses its own country-specific method which considers specific national circumstances (as long as they are in accordance with the 2006 IPCC GLs) as well as IPCC default values, which are usually more conservative. The EU GHG inventory underlies the assumption that the individual use of national country-specific methods leads to more accurate GHG estimates than the implementation of a single EU-wide approach (UNFCCC, 2018). The tier level a country applies depends on the national circumstances, which explains the variability of uncertainties among the sector itself as well as among EU countries. For example, inventory estimates of N2O emissions have very large uncertainties (> 100 %) owing to the heterogeneity of sources and uncertainty in emission factors for the main N2O sources, in particular agriculture. Since agricultural soil and manure management emissions vary strongly from site to site depending on, e.g., soil properties and background emissions, management, and meteorology, it is extremely challenging to determine accurate mean emission factors (JRC report, https://ec.europa.eu/jrc/en/publication/eur-scientific-and-technical-research-reports/atmospheric-monitoring-and-inverse-modelling-verification-greenhouse-gas-inventories, last access: February 2020). Winiwarter et al. (2018) stated that, under current technologies, agricultural emissions have a large potential for abatement, and, in the short term, reductions of N2O emissions must rely on the adoption of existing technologies. Currently available technology could reduce global N2O emissions by about 26 % below the baseline projection in 2030 (Winiwarter et al., 2018). The most applicable pathways to enhance emission reductions are the refinements of existing options (use of fertilizers), increasing the efficiency of measures (N use efficiency) and changing human diets (lower consumption of animal protein). Oenema et al. (2013) estimated a total reduction potential for N2O emissions from agriculture including human diet changes of up to 60 % in 2050, adding about half to the reductions available from technical measures alone (41 % reductions). For CH4, according to Höglund-Isaksson et al. (2012) and the scenario work based on the GAINS model, mitigation opportunities in agriculture are found limited and often costly both from social and private cost perspectives.

Concerning the IPCC calculation of CH4 emissions from enteric fermentation, depending on the type of animal, the situation within the EU28 varies from country to country. For cattle (IPCC sector 3.A.1) emissions are calculated with very sophisticated methods, with only Cyprus using partially Tier 1. For the enteric fermentation of sheep (3.A.2), the situation is more diverse, with 13 countries using Tier 1 methods and 15 using higher tiers (including those with higher emissions). For other cattle (3.A.4), only three countries (Romania, France and Portugal) are using higher tiers, with all the others combining different methods. CH4 and N2O emissions from manure management (3.B.1 and 3.B.2) are even more mixed, with Germany, Denmark, Finland, France, Croatia and Romania using exclusively higher tiers in both categories. For the calculation of emissions from soils, the share of high tiers is very low; only Denmark and Sweden use solely higher tiers in indirect N2O emissions from agricultural soils (3.D.2), while there are no countries using only high tiers in direct N2O emissions (3.D.1) but only some combining high with low tier methods (UNFCCC, 2018b). All these differences in calculating emissions produce evidently higher uncertainties in the results. For the UNFCCC, throughout the variability of the analyzed NGHGIs, it turned out that N2O emissions from manure management and direct and indirect emissions together with CH4 emissions from rice cultivation have the largest uncertainties. When we aggregated UNFCCC uncertainties at the country level (using the methodology described in Appendix C), we also noticed the fact that not all countries report subsectoral uncertainties (e.g., Greece for grazing) and some countries (Sweden, Poland, Croatia and the Czech Republic) had no uncertainty analysis performed for all subactivities due to lack of data (e.g., confidential data).

There is as well the need to define a common methodology for overall uncertainty calculation while checking for consistency in the way uncertainties are calculated for different data sources and the way data are aggregated for different sectors. We noticed that for agricultural N2O emissions the split in subactivities is not always consistent with IPCC sectors, and this leaves room for differences when aggregating the results (Table 3).

For the LULUCF sector, methods for the estimation of GHGs and CO2 fluxes still differ among countries and land use categories. Within the UNFCCC practice, strict good practice guidance is prescribed, but there are still small differences between countries as each considers specific national circumstances (as long as they are in accordance with the 2006 IPCC GLs), as well as IPCC default values. When we analyze the estimates from multiple sources (inventories and models) we observe that published estimates contain two main sources of uncertainties: (a) differences due to input data and structural/parametric uncertainty of models (Houghton et al., 2012) and (b) differences in definition (Pongratz et al., 2014; Grassi et al., 2018b). These differences result from choices in the simulation setup and are partly predetermined (for b) in particular) by the type of model used – bookkeeping models, DGVMs, or inventory based – and whether fluxes are attributed to LULUCF emissions due to the cause or place of occurrence (indirect fluxes on managed land included in NGHGIs and FAOSTAT). Differences in definitions and methodology calculation of estimates across model types are crucial and may lead to model-to-model variability. Depending on the degree of independence between assumptions, variability can become a reliable proxy for structural uncertainty when more accurate estimates are lacking (Solazzo et al., 2018). In Fig. 13 the variability between the mean of the DGVMs ranges between 44 % in 1996 and 186 % in 2016 (distance between interquartile range and median across models for each year).

In general the definition of NBP denotes the net gain or loss of carbon from a region. NBP is equal to the net ecosystem production (NEP) minus the carbon lost due to a disturbance (e.g., forest fire, harvest) taking into account as well the net C balance of harvested products (described by the 2006 IPCC GLs) and C emitted by inland waters. In the context of land use change, the GCB 2017 (Le Quéré et al., 2018a) highlighted harvest as one of the main uncertainties. As an example, according Nabuurs et al. (2018) the uncertainty affecting all studies is that EU harvesting levels are rather uncertain. According to the FRA report 2015 (FRA, 2015), most European countries have a solid forest inventory, but there is still large uncertainty over harvesting levels. For many countries forest statistics from FAO have shortcomings such as very large differences between reported periods, data corrected in later versions and unreported (harvest) removals (Nabuurs et al., 2018).

Checking collective progress towards meeting the goals of the PA will be done by the PA's global stocktake. At present, there is a discrepancy of about 4–5 Gt CO2 yr-1 in global anthropogenic net land use emissions (Grassi et al., 2018a; IPCCC, 2019b) between DGVMs reflected in IPCC assessment reports and aggregated national UNFCCC GHG inventories. Grassi et al. (2018a) shows that about 3.2 Gt CO2 yr-1 can be explained by conceptual differences in anthropogenic forest sink estimation, related to the representation of environmental change impacts and the areas considered managed. In order to limit the temperature increase to 1.5 ∘C and keep it well below 2 ∘C, as set by the PA, net-zero CO2 emissions at the global level need to be achieved around 2050 and neutrality for all other GHGs somewhat later in the century. At this point, any remaining GHG emissions in certain sectors need to be compensated for by absorption in other sectors, with a specific role for the land use sector, agriculture and forests (European Commission Report, 2018).

It is important to distinguish between reporting and accounting in the GHG inventory context, as not all reported emissions account towards emission reduction efforts (Grassi et al., 2018b). Reporting refers to the inclusion of estimates of anthropogenic GHG fluxes in NIRs, following the methodological guidance provided by the IPCC. The NIR should, in principle, aim to reflect “what the atmosphere sees” (Peters et al., 2009) in managed lands, within the limits given by the method used and the data available. In the context of mitigation targets (e.g., the PA), accounting refers to the comparison of emissions and removals with the target and quantifies progress toward the target. For the LULUCF sector, specific accounting rules are used to filter reported flux estimates with the aim to better quantify the results of mitigation actions (Grassi et al., 2018b). The UNFCCC reporting principles allocate emissions to the territorial location (national boundaries) at the time that they occur (Peters et al., 2009).

The different definitions and concepts used by the global models and inventory communities mean that the land fluxes cannot necessarily be consistently compared. The framework developed by Grassi et al. (2018a) and shown in Fig. 12 can be generalized to make a more direct comparison as applied to EU28 (Fig. 14). Figure 14 disaggregates managed forest land into components that are reported in the UNFCCC CRFs: converted land (e.g., land changing from cropland to forest land) and the remaining land (e.g., forest land remaining forest land) are split into land that is “production” (remaining forestry, RF) or land that is used for ecological or social functions (other Ro), based on the definitions of managed land. Unmanaged land (sink, S) cannot have direct human-induced effects.

Overall, our results suggest that most of the LULUCF emissions in the EU28 are from direct effects in the managed forest sector, including age-legacy effects (forest expansion and regrowth after WWII), with small net emissions from land conversion as they are largely compensated for by deforestation (from CRFs). With appropriate data and models, it is theoretically possible to expand and enumerate the estimates more accurately.