1 Introduction

Atmospheric concentrations of greenhouse gases have increased substantially since pre-industrial times. The rise in CO2 is mainly associated with fossil fuel burning and contributions from the land sector (IPCC 2022), commonly referred to as ‘agriculture, forestry, and other land uses’ (AFOLU). AFOLU emissions result from changes in the natural fluxes of CO2 and other non-CO2 gases, largely methane (CH4), and nitrous oxide (N2O), through anthropogenic changes in land use/cover (e.g., conversion of forest lands and grasslands to cropland and pasture, afforestation), and management of croplands, forests, and wetlands. AFOLU activities lead to both sources and sinks of CO2, mainly from forestry and land-use change, and to non-CO2 emissions mainly from agriculture, e.g., CH4 from livestock and rice cultivation, N2O from manure storage, and cropland soils and biomass burning (Tubiello et al. 2013, 2014). Globally, the AFOLU sector was responsible for 13 GtCO2 eq. year−1 in 2019, which was about 22% of net anthropogenic greenhouse gas (GHG) emissions, with approximately half from agriculture and the other half from forestry and other land use activities (FOLU) (Pathak et al. 2022).

AFOLU emissions for different GHG gases (CO2 and non-CO2) are available through the Food and Agriculture Organization (FAO) Statistical database (Tubiello et al. 2013, 2014). In addition, global and national CO2 emissions for the FOLU sector are also provided by bookkeeping models, i.e., BLUE, Houghton and Nassikas, and OSCAR models (Gasser et al. 2020; Hansis et al. 2015; Houghton et al. 2017), and the estimates have recently been updated for the Global Carbon Budget 2022 (Friedlingstein et al. 2022). Agricultural CO2 and non-CO2 emissions are available through the FAO database, Emissions Database for Global Atmospheric Research (EDGAR) (Crippa et al. 2019; Crippa et al. 2020), and the US Environmental Protection Agency (EPA) databases (US EPA 2019). Past comparative analyses showed reasonably good agreement among these data sources in the total gross emissions at the aggregated level for the tropical regions, but significant disagreements when disaggregating the emissions by sources, continents, and gases, particularly for the forest sector, with fire leading the differences (Roman-Cuesta, Herold, et al., 2016). In addition, emissions from the tropical regions have remained consistently higher compared to global estimates. For instance, for the FOLU sector, Roman-Cuesta et al. (2016a) reported emissions of ca. 6 GtCO2 eq. year−1 for the tropics (2000–2005), with this value being higher than the global AFOLU emission estimates of ca. 5.0 GtCO2 eq. year−1 in the Intergovernmental Panel for Climate Change (IPCC) AR5 synthesis report over the same period (Smith et al. 2014).

Besides these global datasets, article 14 of the Paris Agreement includes an enhanced transparency framework (ETF) to periodically track and review the progress of countries towards achieving their Nationally Determined Contributions (NDCs), and a periodic global stocktake process for assessing the countries’ collective progress towards achieving the long-term goals of the agreement (UNFCCC 2016). All parties to the United Nations Framework Convention on Climate Change (UNFCCC) are required to document the anthropogenic emissions and removals of different sectors and report these through the national greenhouse gases inventories (NGHGIs). Recently, a dataset including compilation of individual NGHGIs for the land use, land-use change and forestry (LULUCF) sector has become available (Grassi et al. 2022b). Previous studies revealed conceptual inconsistencies between the scientific modelling assessments and countries in estimating the net anthropogenic land related CO2 emission, with a gap of about 4 to 6.7 GtCO2 year−1 globally between the aggregated NGHIs and global model estimates (Grassi et al. 2018, 2021, 2023). The large differences are mostly due to different approaches in estimating the forest sink. While these two studies have made significant progress in reconciling the differences between global models and NGHGIs, their focus was on forest-related emissions and sinks at the global level. In a recent study, differences between the NGHGIs and FAOSTAT have also been assessed at the global and regional level (Grassi et al. 2022b), and the discrepancy between global models and NGHGIs has been resolved at the country-level as well (Schwingshackl et al. 2022). However, there’s still a need for a thorough comparative analysis of emissions in tropical regions that integrates all the latest available datasets. With all UNFCCC parties expected to adopt a harmonized reporting by 2024 under the Enhanced Transparency Framework (UNFCCC Secretariat 2020), such an analysis would not only be useful in identifying and understanding the causes of differences among the datasets but would also offer information on how countries can improve their reporting using the available datasets.

In this study, we present a comparative analysis of AFOLU emission estimates for the tropics for the period 1990–2018 using seven data sets: NGHGIs, FAOSTAT, the BLUE, OSCAR and Houghton and Nassikas booking models (here after updated H&N2017), EPA, and EDGAR. Although the approach for our analysis is in some ways like that used in the comparative study by Roman-Cuesta et al. (2016b), our study goes a step further by including new databases (e.g., NGHGIs), more updated global data sources, longer analysis periods, and disaggregated analysis for FOLU emissions. The objectives of the study are (i) to provide a comprehensive and updated analyses on the global and tropical regions aggregated and disaggregated AFOLU emissions from different data sets (i.e., at continental level and activity), (ii) assess the trend of the emissions, and (iii) understand the reasons for the differences between the datasets.

2 Methods

2.1 Study scope

Our study comprises a detailed analysis of global AFOLU datasets and estimates, with a focus on three tropical regions: Latin America (LAM), sub-Saharan Africa, and South and Southeast Asia. We chose to analyse the emissions for these regions because of the rapid land use change in the tropics and the increased reporting requirements on tropical countries under the Paris Agreement. In addition, the tropics have the largest emissions uncertainty and gaps, hence the need to understand the causes of differences between the available AFOLU data sources (Romijn et al. 2012, 2015). Such detailed comparisons already exist for other regions like Europe (Maria et al. 2020). We analysed the annual average emissions over three different time periods (1990–1999, 2000–2010, 2011–2018), which allowed for an assessment of how the reported emissions have changed over time. We focused on the period 1990–2018, because it was the period where emissions were available in most of the datasets. The average emissions over a given period was calculated by summing the annual emissions and dividing by the total number of years in that period. Our analysis includes net FOLU emissions, gross emissions from deforestation, net emissions or sinks from other forestland activities (e.g., harvest), and other land uses. Net land use emissions consider the emissions by the sources and the removals by sinks in a final emission balance where the removals are discounted from the emissions. Gross assessments include the contributions of different sources and sinks to the net land use emissions. A key advantage of the gross assessments compared to net AFOLU emissions is that they offer direct information on the sinks and sources that can facilitate the development of more targeted measures and policies to enhance and promote mitigation (Deb Richter and Houghton 2011; Roman-Cuesta et al. 2016a), and provide a better indication of the climate benefits of forests.

2.2 AFOLU data sources

2.2.1 FAOSTAT

FAOSTAT covers agriculture, forestry, and other land uses and their associated emissions of CO2, CH4, and N2O (Federici et al. 2015; Tubiello et al. 2013, 2021) . FAOSTAT database includes emission estimates for 242 countries annually for the reference period 1961–2018 (agriculture) and 1990–2019 (FOLU), based on national activity submitted by countries and further compiled by FAO (Food and Agriculture Organization of the United Nations). Estimates are computed following Tier 1 methods and default factors of the 2006 IPCC Guidelines for National Greenhouse Gas Inventories. FAOSTAT includes estimates of carbon emissions from net forest conversion, biomass fires, organic soils in cropland and grasslands, peatland drainage, and emissions from other forestland activities (Table 1). These estimates are based on national statistics. Forest carbon stock changes (both emissions and removals) have recently been updated based on the national-level FAO Forest Resources Assessment 2020 (FAO 2020a, 2020b).

Table 1 A summary of the characteristics of the four forestry and land use (FOLU) activity data sets analysed in the present study
Full size table

2.2.2 Bookkeeping models

We used the three bookkeeping models that were part of the 2022 Global Carbon Budget: the OSCAR model, the bookkeeping of land use emissions (BLUE), and the Houghton and Nassikas model (updated H&N2017) (Gasser et al. 2020; Hansis et al. 2015; Houghton et al. 2017). The models are based on the original bookkeeping approach of Houghton (2003) that keeps track of the carbon stored in the vegetation and soils before and after a land-use change. The models use literature-based response curves to describe the decay of vegetation and soil carbon, including transfer of product pools of different lifetimes, as well carbon uptake due to regrowth. Bookkeeping models estimate emissions of CO2 from deforestation, shifting cultivation, wood harvest, woody debris decay, biomass burning (deforestation and peatland fires), and soil organic matter (SOM) from cultivated soils.

The three bookkeeping models differ with respect to the land-use change data used to drive the model. The updated H&N2017 bases its estimates on the Forest Resource Assessment (FRA) of the FAO, which provides statistics on forest change and management at intervals of 5 years and is currently updated until 2020 (FAO 2020a). Changes in land use other than forests are based on annual national changes in cropland and pasture areas reported by FAO (FAOSTAT, 2021). BLUE uses the harmonized land-use change data LUH2-GCB2022 covering the entire 850–2021 period, which is an update to the previously released LUH2 v2h dataset (Hurtt et al. 2020). OSCAR was run with both LUH2-GCB2022 and FAO/FRA (as used in the updated H&N2017), where the drivers of the latter were linearly extrapolated to 2021 using their 2015–2020 trends. BLUE and updated H&N2017 models exclude land ecosystems’ transient response to changes in climate, atmospheric CO2, and other environmental factors, and the carbon densities are based on contemporary data from literature and inventory data. Since carbon densities remain fixed over time, the additional sink capacity that ecosystems provide in response to CO2 fertilization and some other environmental changes is not captured by the models (Pongratz et al. 2014). Although the OSCAR model does account for the transient response associated with environmental changes, the emissions included here are only those associated with land-use change. Emissions from peat burning and peat drainage are added from external datasets (see Appendix C2.1 in Friedlingstein et al. 2022). We obtained the net FOLU emissions for the three bookkeeping models and annual gross fluxes of CO2 (i.e., both sinks and sources) resulting deforestation, forestland, and other land use transitions for the three bookkeeping models. For the three models, annual emissions were available for 218 countries. The FOLU emissions used here are based on the 2022 Global Carbon budget (Friedlingstein et al. 2022).

2.2.3 National Greenhouse Gases Inventories (NGHGIs) for the LULUCF sector

National Greenhouse Gas Inventories (NGHGIs) presented here include estimates of LULUCF CO2 emissions and sinks submitted to the UNFCCC. Figure 1 illustrates the conceptual differences between the NGHGIs, the bookkeeping models, and FAOSTAT in terms of the processes accounted for and the area extent. The data were compiled from various submissions to UNFCCC, prioritizing the most recent and considering the completeness of the information (Grassi et al. 2022a; Grassi et al. 2022b). For non-Annex 1 countries, National Communications (NCs), Biennial Update reports (BURs), and nationally determined contributions were used in compiling the data. Only one submission was used in compiling the data national GHGIs. While NCs are typically submitted every 4 years, BURs provide an update of the information presented in NCs, typically every 2 years.

Fig. 1
figure 1

A conceptual diagram indicating the key differences between the National Greenhouse Gases Inventories (NGHIs), the bookkeeping models and the FAOSTAT in their Forestry and Land Use (FOLU) emissions. Bookkeeping models and NGHIs provide gross emissions of deforestation, while FAOSTAT provides net emissions from forests, where emissions from deforestation are discounted with the gains from afforestation. Bookkeeping models consider as anthropogenic only direct human-induced fluxes on forestland, while NGHGIs and FAOSTAT account for both the human-induced fluxes and the indirect effects resting from environmental changes, e.g., the CO2 fertilization and nitrogen deposition. In addition, the extent of areas differs with NGHGIs also including unmanaged lands. On cropland and grasslands, bookkeeping models only include fluxes from the land use conversion, FAOSTAT include only the fluxes from organic soils, while the NGHIs account for the fluxes from both mineral and organic soils

Full size image

The time of reporting varies for the different non-Annex 1 counties. To ensure a complete time series for the period 2000–2020, gaps were filled using standard statistical methods, with the aim to maintain the levels and trends of the underlying reported raw data (see Tables 4 and 5 of the online dataset in Grassi et al. 2022a, showing the original and gap-filled time series, respectively). However, most of the non-Annex 1 counties with the biggest emissions, including Brazil, Indonesia, and Democratic republic of Congo, reported relatively complete time series. The gap-filled data includes all land uses (i.e., forestland, cropland, grassland, wetlands, settlements, and other land) and covers 195 countries. Almost all Annex 1 countries report all land uses, while many non-Annex 1 countries report only on forestland and deforestation. Here, we used the NGHGI data compiled in Grassi et al. (2022b), which include a split of the net LULUCF flux into the categories of forest land, deforestation, organic soils, and other land use fluxes.

2.2.4 EDGAR

The Emissions Database for Global Atmospheric Research version 5 (EDGARv5.0) provides emissions of three main greenhouse gases (CO2, CH4, and N2O) per sector at 0.1 deg × 0.1 deg grid maps and country level (Crippa et al. 2019, 2020). The EDGAR database for this version covers the energy, industry, transport, waste management, and agriculture (including agricultural soils, agricultural waste burning, enteric fermentation, manure management) sectors, mainly quantified using the IPCC 2006 guidelines for emissions estimations. However, emissions of large-scale biomass burning in savannahs, forest fires, and sources and sinks from LULUCF are excluded in this version. EDGAR emission estimates are mostly calculated using activity data from FAO statistics. However, N2O emission from fertilizer use and CO2 from urea fertilisation are estimated using both the International Fertilizer Association (IFA) and FAO statistics data. Furthermore, CH4 emissions from enteric fermentation for dairy and non-dairy cattle are calculated using Tier 2 emission factors. We downloaded the EDGARv5.0 (January 2020) emissions over the period 1990–2018 on an annual basis, at the country level (Table 1) (Crippa et al. 2019).

2.2.5 The US Environmental Protection Agency (EPA)

The EPA dataset contains global non-CO2 projected emissions for the period 1990–2050 for the agriculture (livestock, cropland, rice cultivation, and biomass burning), energy, industrial processes, and the waste sectors. EPA provides future projections of non-CO2 GHGs using a combination of country-reported inventory data and EPA-estimated calculations consistent with the 2006 Intergovernmental Panel on Climate Change (IPCC) Guidelines for national GHGIs. The EPA data provides information that can be used to understand national contributions of GHG emissions, historical progress on reductions, and mitigation opportunities. The EPA data does not include estimates of FOLU activities. We downloaded annual agricultural emissions offered at a 1-year interval at country level and disaggregated by gas and emission sources from the recently updated data that was used in the 2015–2050 projections (Table 1) (US EPA 2019).

2.3 Estimating net and gross emissions and sinks

We analysed the data using two major categories: (1) forestry and other land use activities (FOLU) and (2) agriculture. We quantified the net FOLU and agriculture emissions by summing the emissions and sinks for all the activities in the respective category under Table 1. We compared the net FOLU, and agriculture emissions estimated from the different data sources at the global level and within the three continents in the tropical regions. To assess the trend of the emissions in the tropical regions, we calculated the average values for three different time periods (1990–1999, 2000–2010, and 2011–2018). Disaggregated values from the NGHGIs were available in four time slices (2000–2005, 2006–2010, 2011–2015, and 2016–2020). For comparability with other data sets, the 2016–2020 emissions were included in the 2011–2018 period.

FOLU emissions from all the considered data sets only included CO2 emissions and excluded non-CO2 greenhouse gases. To gain an in-depth understanding of the emissions patterns and identify the possible causes of differences among the data sources, the FOLU emissions were further disaggregated into three categories: deforestation, forestland, and other land uses (emissions and sources). Emissions from organic soils were included in other land use emissions. Emissions for forest fires were included in the forestland category. Any other non-forest-related activities were categorized as other land-use emissions. FAOSTAT do not provide gross deforestation emissions fluxes, but instead include the net forest conversion emissions (i.e., from forest area losses minus area gains). For comparison purposes with the other data sources, the net forest conversion emissions for FAOSTAT were included in the deforestation category. The disaggregation of FOLU emissions into deforestation, forestland, and other land use was chosen for comparability of other data sets to in the FOLU sector recently published national GHGIs (Grassi et al. 2022b).

Emissions from agriculture were separated into four categories based on the different sources, i.e., croplands, livestock, rice cultivation, and agricultural fires. CH4 and N2O emissions were converted to CO2eq.

3 Results

3.1 FOLU emissions

3.1.1 Global and regional net emissions

Globally, all the data sources show that the net FOLU activities contributed to CO2 emissions over the periods 1990 to 1999 (Table 2; Fig. 2). However, there was disagreement between NGHGIs and the other data sources on the emissions for the period between 2000 to 2010 and 2011–2018, with the inventories indicating a net sink of −1.32 and −2.03 GtCO2 year−1 over the two periods, while the other data sources indicated a source (Table 2). Over the period 2011–2018, the updated H&N2017, BLUE, and OSCAR bookkeeping models estimated a net global source of 3.18, 5.16, and 5.90 GO2 year−1, while the emissions from FAOSTAT were 0.26 GtCO2 year−1. The total emissions for the tropics over the same period were 5.47, 5.22, 4.28, 3.21, and 1.17 GtCO2 year−1 (for BLUE, OSCAR, updated H&N2017, FAOSTAT, and NGHGIs). All the other data sources indicate a decline in the net global FOLU emissions over time.

Table 2 The total net FOLU emissions in GtCO2 year−1 for the tropical and other regions for the various data sources. The tropical regions include Latin America, sub-Saharan Africa, and South and Southeast Asia
Full size table
Fig. 2
figure 2

Global and tropical regions net forestry and other land use (FOLU) emissions for the five data sources. BLUE, OSCAR, updated H&N2017, FAOSTAT, and NHGIs include emissions due to all forestry and land use activities, while the reported emissions in NGHGIs vary depending on the countries

Full size image

Regional analyses showed that the largest contributions for the global FOLU emissions stemmed from the tropical regions of South and Southeast Asia, sub-Saharan Africa, and Latin America (Fig. 2; Table 2). Three of the data sources (updated H&N2017, FAOSTAT, and NGHGIs) indicate that the large emissions in the tropics were offset by the forest sink in the non-tropical regions, leading to lower net emissions globally. However, the BLUE and OSCAR models generally estimated a source of CO2 also for the non-tropical regions (Table 2). The updated H&N2017 model, FAOSTAT, and NGHGIs indicate a decline in the net tropical emissions over the three considered period, while the BLUE and OSCAR models indicate a decline in the emissions over the two recent periods (2000–2010 and 2011–2018. Consistent with the observed global patterns, the net emission estimates from the NGHGIs were smaller than those from the other data sources in the three tropical regions (Fig. 2). Overall, there was generally a decline in the emissions over Latin America and an increase over sub-Saharan Africa, but the different datasets indicated different trends over South and Southeast Asia.

Further analyses showed substantial differences between NGHGIs and other data sources at the sub-regional levels for the tropical regions. Within West and Southern Africa, the NGHGIs indicated a small sink in CO2, while the other four data sets indicated a source over the regions (Fig. S1). The bookkeeping models generally estimated an increase in the emissions over the East Africa region between 1990–1999 and 2000–2010 and a decrease in 2011–2018, while FAOSTAT and NGHGIs indicated no change. An increase in the emissions over Central Africa was evident in all the data sources. As expected, South America is the main contributor of emissions in Latin America, with almost all the data sources estimating a decline in the emissions over the period 2011–2018 (Fig. S1). Emissions within Central America were relatively small. In addition, the Southeast Asia sub-region was the main source of emissions within South and Southeast Asia region. NGHGIs indicated a sink in over the South Asia and Central America.

3.1.2 Disaggregated emissions and sinks

Global gross deforestation emissions (i.e., forest losses only) estimates from the three bookkeeping models (updated H&N2017, OSCAR, and BLUE) were higher than the estimates from NGHGIs and FAOSTAT (Fig. 3a). The NGHGIs and FAOSTAT indicated a decline in the global deforestation emissions, while the updated H&N2017 estimated no change in the emissions. The BLUE model indicated an increase in the emissions in the first two considered periods and decline over the period 2011–2018, while the OSCAR model indicated a small increase in the emissions over the same period. A similar trend was also observed in the total deforestation emissions for the tropical region (Fig. 3b). Overall, the global deforestation emissions over the recent period (2011–2018) were 6.90, 8.06, 5.23, 4.10, and 3.17, and 5.64 GtCO2 year−1 for BLUE, OSCAR, updated H&N2017, NGHGIs, and FAOSTAT. The total tropical deforestation emissions were 5.88, 7.16, 5.48, 3.74, and 2.96 GtCO2 year−1 (BLUE, OSCAR, updated H&N2017, NGHGIs, and FAOSTAT), which amounts to about 90 to 96% of the global deforestation emissions (Fig. 3).

Fig. 3
figure 3

a Global and b tropical region disaggregated emissions and sinks from deforestation, other forestland activities (e.g., harvest, reforestation, and fires), and other land use activities. Deforestation emissions for the three bookkeeping models (updated H&N2017, OSCAR, and BLUE) and NGHGIs represent the emissions associated forest losses only, while the emissions for FAOSAT represent the net emissions from forest area losses discounted with the sink from forest area gains. The sum of the three components is equivalent to the total emissions represented in Figure 2

Full size image

Regional analyses indicated a decline in the deforestation emissions over Latin America and a slight increase over sub-Saharan Africa and South and Southeast Asia (Fig. 4). Within sub-Saharan Africa, the three data sources generally agreed on the trend of deforestation emissions for the different sub-regions (Fig. S2): An increase in the emissions was evident in Central Africa, while there was a decline in the emissions over Western Africa and very small changes over Eastern and Southern Africa. As expected, Central Africa had the largest deforestation emissions, because of the high forest cover and deforestation rate mainly in the Congo Basin. Similarly, deforestation emissions in Latin America were mainly from South America region, with the Central America emissions being substantially smaller (Fig. S3). The Southeast Asia sub-region contributed to nearly all the emissions in the South and Southeast Asia region (Fig. S4).

Fig. 4
figure 4

Disaggregated emissions and sinks from deforestation, other forestland activities (e.g., harvest, reforestation, and fires), and other land use activities for Latin America, sub-Saharan Africa, and South and Southeast Asia. Deforestation emissions for the three book keeping models (BLUE, OSCAR, and updated H&N2017) and NGHGIs represent the emissions associated forest losses only, while the emissions for FAOSAT represent the net emissions from forest area losses discounted with the sink from forest area gains. The sum of the continental emissions is equivalent to the total tropical regional emissions represented in Fig. 3b

Full size image

Other forestland activities (e.g., afforestation, reforestation, regrowth, and harvest) contributed to a global net sink in CO2 across all the data sets over the considered periods (Fig. 3a). However, the sink in the NGHGIs was larger than that in FAOSTAT and the bookkeeping models estimates. The global forestland sink was −2.86, −4.45, −2.96, −6.44, and −3.89 GtCO2 year−1 (for BLUE, OSCAR, updated H&N2017, NGHGIs, and FAOSTAT, respectively) over the period 2011–2018. The total net forest sinks for the tropical regions were −1.84, −3.08, −2.09, −3.00, and −0.53 GtCO2 year−1 (for BLUE, OSCAR, updated H&N2017, NGHGIs, and FAOSTAT respectively). Hence, the tropical region contributed to about 13 to 47% of the global forest sink. Overall, an increase in the global forest sink was evident across all the datasets, but there were no notable changes in the tropics (Fig. 3).

All the data sources indicated a CO2 sink in the forest over Latin America and South and Southeast Asia (Fig. 4). However, the NGHIs estimated a larger sink than the bookkeeping models and FAOSTAT. The data sets differed in sub-Saharan Africa with the NGHGIs and the three bookkeeping models estimating a sink in CO2 while FAOSTAT indicated a source. In terms of trend, the three data sources indicated a small increase in the forest sink over Latin America and a relatively small decline over sub-Saharan Africa and South and Southeast Asia (Fig. 4). The forestland sink was evident in the NGHGIs, and bookkeeping model estimates across all the sub-regions within sub-Saharan Africa (Fig. S2), with the largest sink being in Central Africa and Western Africa and the smallest over Southern Africa. FAOSTAT mostly showed a small source of CO2 across the regions. As expected, the forest sink was larger over South America than Central America, with fewer differences between the data sources over South America (Fig. S3). The NGHGIs and the bookkeeping models estimated a much larger forest sink over Southeast Asia compared to FAOSTAT (Fig. S4).

Other land use activities (e.g., non-forest related land use conversions, biomass burning, and peatlands drainage) contributed to emissions, with the total emissions being much smaller than that of deforestation across all the considered regions (Fig. 3). The total net global emissions from other land use activities were 1.87, 1.55, 0.92, 0.33, and 0.98 GtCO2 year−1 (for BLUE, OSCAR, updated H&N2017, NGHGIs, and FAOSTAT, respectively) over the period 2011–2018. The tropical regions contributed largely to these emissions with values 1.56, 1.13, 0.88, 0.45, and 0.77 GtCO2 year−1 (for BLUE, OSCAR, updated H&N2017, NGHGIs, and FAOSTAT) over the same period. For FAOSTAT, emissions from other land use activities were only from the organic soils (Table 3). At the global and tropical region level, emissions from organic soils were generally higher in FAOSTAT than in the NGHGIs and in the average data estimates used in the bookkeeping models. All the data sets estimated higher emissions for the organic soils in the tropics compared to other regions. Regional analyses indicated no major changes in the total emissions from other land use activities over the considered time period. However, at the sub-regional level an increase in emissions over Western Africa and Eastern Africa and a decline of Central Africa were evident among most of the data sources (Fig. 4, S2, S3 and S4).

Table 3 Comparions of emission estimates for organic soils for the tropical and non-tropical regions. The three book keeping models use the same source of dataset for organic soils emissions in calculating the total FOLU emissions
Full size table

3.2 Agricultural emissions

3.2.1 Global and regional net emissions

Global values showed a good agreement in emissions from agriculture among the three data sources (Table 4; Fig. 5). All the sources indicated an increase in the agricultural emissions over time, with values of 5.89, 6.08, and 6.27 GtCO2 eq. year−1 (for FAOSTAT, EDGAR, and EPA, respectively) for the recent period (2011–2018). Global agricultural emissions were generally higher than the FOLU emissions (comparing total emissions in Tables 2 and 4). Analysis of the contribution of the different GHGs generally indicates higher methane emissions compared to carbon dioxide and nitrous oxide over all the considered time periods (Table 4).

Table 4 Contribution of CO2, CH4, and N2O to the global and tropical regions agricultural emissions from FAOSTAT, EDGAR and EPA
Full size table
Fig. 5
figure 5

Global and tropical regions net agricultural emissions (GtCO2 eq. yr−1) estimates from EDGAR, EPA and FAOSTAT

Full size image

Regional analyses indicate that the emissions were generally higher in South and Southeast Asia followed by Latin America and sub-Saharan Africa for EDGAR and FOASTAT (Fig. 5). The total emissions over the tropical regions were 3.08, 3.69, and 3.70 GtCO2 eq. year−1 for the period 2011–2018. Overall, the regions contributed to about 50–60% of the global agricultural emissions. All data sources indicate a slight increase in the emissions in all the three tropical regions. Sub-regional analyses showed a good agreement among the data sources in Latin America and South and Southeast Asia (Fig. S5). In contrast to the FOLU emissions, agricultural emissions were generally higher in Southeast Asia than in South Asia across the three data sources. The agricultural emissions from Latin America were mainly from South America. The total EPA emissions for the four sub-regions in sub-Saharan Africa were consistently higher than the estimates from EDGAR and FAOSTAT, with differences of up to 0.3 GtCO2 eq. year−1 over Central Africa (Fig. S5).

3.2.2 Disaggregated agricultural emissions.

Disaggregated values show a good agreement on the global and tropical region emissions from livestock, cropland, and rice cultivation, while there was a discrepancy in the emissions from agricultural fires (Fig. 6). All the data sources indicate that the livestock sector is the major contributor to agricultural emissions globally followed by croplands, rice cultivation and biomass burning from agricultural activities. Over the period 2011–2018, the global emissions from livestock were 2.52, 2.41, and 2.56 GtCO2 eq. year−1 for EDGAR, EPA, and FAOSTAT, respectively (Fig. 6), with the total emissions for the three tropical regions being 1.49, 1.32, and 1.43 GtCO2 eq. year−1. All the data sources show an increase in the global emissions from cropland cultivation and livestock, while there were no notable changes in the rice cultivation emissions. EDGAR and EPA had higher agricultural fire emissions compared to FAOSTAT globally and for the tropics (Fig 6).

Fig. 6
figure 6

(a) Global and (b) Tropical region disaggregated gross emissions from different agricultural activities

Full size image

Regional analyses showed that the largest livestock emissions were from Latin America followed by South and Southeast Asia, with lower values in sub-Saharan Africa (Fig. 7). The increase in the cropland emissions was evident across the three tropical regions, while an increase in the livestock emissions was more evident in sub-Saharan Africa and Latin America. Emissions from rice cultivation remained constant over time, with the South and Southeast Asia region being the major contributor. Although there was generally good agreement on the global emissions caused by agricultural fires among the EPA and EDGAR datasets, regional analyses show substantial differences in the emissions from sub-Saharan Africa with EPA estimates being much higher than EDGAR (Fig. 7). Further analyses show that this difference is mainly due to the emissions in Central Africa, with EPA estimating large emissions from agricultural fires compared to EDGAR and FAOSTAT. This also explains why the total agricultural emission estimates for EPA were larger than the other two data sources in Central Africa.

Fig. 7
figure 7

Disaggregated gross emissions from different agricultural activities for the three tropical regions

Full size image

4 Discussion

4.1 Forest-related emissions

Causes of different FOLU estimates in the available global data sources have been assessed in detail in previous studies and can be broadly categorized into two: (1) differences in the system boundaries and (2) different estimates from similar contributing processes. For bookkeeping models and GHGIs, the differences in the system boundaries mainly result from varying spatial areas of managed lands with bookkeeping models having a broader definition of these lands (Grassi et al. 2018). In addition, there are conceptual differences in accounting for the direct and indirect human-induced effects of land-use change: NGHGIs include the indirect effects of CO2 fertilization, N deposition, and changes in climatic variables (e.g., temperature and precipitation), while bookkeeping models only account for the direct effects of land-use change, harvest, and management. Differences between FAOSTAT and NGHGIs can be attributed to differences in estimates of the forest land area, incomplete accounting of all the carbon pools in FAOSTAT, and different methodologies in accounting for sink in some of the non-Annex 1 countries (Grassi et al. 2022b). Here, we assessed the differences in these three data sources through analysing both the net emissions and disaggregated emissions for the different FOLU activities.

The estimated difference in the net emissions for the tropics between the NGHGIs and bookkeeping models is about 3.82 GtCO2 year−1 in the 2011–2018 period, assuming an average of 4.99 GtCO2 year−1 in the three booking models estimates (5.47, 5.22, and 4.28 GtCO2 year−1 in BLUE, OSCAR, and updated H&N2017, respectively - Table 2). The generally higher emissions in BLUE compared to the updated H&N2017 model are largely due to differences in carbon densities between natural or primary vegetation and secondary vegetation, and higher allocation of cleared and harvested material to fast turnover pools in BLUE compared to HN2017 (Bastos et al. 2021). The disaggregated emission data show that the differences in the emissions for the tropics between NGHGIs and the H&N2017 model are mainly due to the varying estimates of the forestland sink, which stems from the differences in the systems boundaries explained above. Although the results show a good agreement in the forest sink estimate between NGHGIs and FAOSTAT in Latin America, a difference of about 2 GtCO2 year−1 exists in sub-Saharan Africa and South and Southeast Asia. The differences in the sink between FAOSTAT and NGHGIs are mainly related to the different estimates of the forestland areas and carbon pools. In particular, Grassi et al. (2022a) found that five countries in the tropics (i.e., Malaysia, Philippines, Central African Republic, Mali, and Namibia) report an implausible forest sink due to the inaccurate implementation of the IPCC methodology. In addition, there is a large uncertainty in the carbon stock changes over time in the two data sources. Schwingshackl et al. (2022) investigated differences in NGHGIs and bookkeeping model estimates that remained even after the sink induced by environmental changes on managed forest land was accounted for in detail for several countries, including Brazil, Indonesia, and the DR Congo. Various shortcomings are discussed, such as the land-use data underlying the bookkeeping model estimates likely underestimating forest cover changes in Brazil or the NGHGIs capturing only parts of the substantial forest degradation in Indonesia and the DR Congo.

The disaggregated data show a difference of up to 2 GtCO2 year−1 the deforestation emissions, with the largest difference mainly stemming from the tropical regions of sub-Saharan Africa and South and Southeast Asia. This difference can be mainly attributed to the differences in the accounting in the considered data sources. FAOSTAT provides net forest conversion fluxes where the net loss emissions are discounted against the gains from forest area increase (Tubiello et al. 2021), while the NGHGIs and bookkeeping models provide gross deforestation fluxes. This explains why the deforestation fluxes in FAOSTAT were lower than the bookkeeping model and NGHGIs estimates. For the NGHGIs, some countries (e.g., India) do not report gross deforestation emissions, but instead provide net fluxes (Grassi et al. 2023). Excluding the gains from forest recovery and regrowth would increase the estimated deforestation emissions and provide a better indication of the absolute mitigation potential in forests. Although the bookkeeping models estimate higher emissions than NGHGIs, a recent study has shown that deforestation emission in the models are much lower when using high resolution data (Ganzenmüller et al. 2022). Overall the gross deforestation emission values of 2.96 to 5.43 GtCO2 eq. year−1 in the three data sources are within the range of the gross forest emission values of 5.3 ± 2.4 GtCO2 eq. year−1 estimated using spatially observed data combined with IPCC factors (Harris et al. 2021).

4.2 Other land use–related emissions

Emissions from other land use activities (e.g., shifting cultivation, non-forest related land use conversions, biomass burning, and peatland drainage) in the tropical regions are relatively small compared to those from deforestation. The net emissions for these activities were larger in the bookkeeping models compared to FAOSTAT and NGHGIs. The higher deforestation emissions mainly stem from shifting cultivation, particularly for the tropical regions. Shifting cultivation has a relatively large importance, and a strong sensitivity has been shown towards the choice of underlying land-use dataset and spatial resolution (Ganzenmüller et al. 2022). Excluding emissions from organic soils in the other land use emission estimates in the NGHGIs indicates that croplands, grasslands, and wetlands contribute to a sink of CO2 in the tropical regions. However, most non-Annex 1 countries do not account for all emissions from other land use activities in their national reporting (Grassi et al. 2021; Grassi et al. 2018), which may explain why these are nearly negligible in sub-Saharan Africa and Latin America. A lack of reporting of these values may lead to uncertainties in the global stocktake especially when comparing NGHGIs to other data sources. While emissions from these activities are currently small and do not exhibit a clear trend (Figs. 2 and 3), current national and global efforts to curb deforestation are likely to shift the trends in the land-use and land-use change activities. Increased activities in grassland areas and poor management of cropland areas are also likely to increase emissions from other land use; thus, making them equally relevant for the global stocktake.

Overall, our regional analyses provided insights on the trend of the three tropical regions. Kondo et al. (2022) found no clear trend in the net FOLU emissions for Southeast Asia with the two bookkeeping models indicating differing trends — the BLUE model indicated an increase while the HN2017 model a decrease. The additional analyses with FAOSTAT and NGHGIs indicate no clear trend in the emissions for the same region. A slight decline in the emissions over Latin America and an increase over sub-Saharan Africa is evident in all the data sources. This is in line with a recent study which shows that the upward trend in emissions over Brazil has likely been overestimated due to artifacts in earlier land-use datasets (Rosan et al. 2021). A recent study covering the same period in this study showed that there has been a global increase in the net AFOLU emissions (Lamb et al. 2021). The analyses presented here show that for the tropics, this increase is mainly from agricultural emissions and an increase in FOLU emissions particularly over sub-Saharan Africa. Although the analyses show a decline in the deforestation emissions over Latin America over the period 1990–2018, which is mainly because of government initiatives (Nepstad et al. 2014), this trend is likely to have reversed subsequently, as the 2019 and 2020 season had the highest deforestation rates in the Amazon region since 2008 (Silva Junior et al. 2021).

4.3 Agriculture-related emissions

Total global agricultural emissions for the considered period were more than the emissions from FOLU activities, with fewer discrepancies between the data sources for the emissions from livestock, cropland, and rice cultivation. The high level of agreement on the net emissions for the agricultural sector is because the considered data sources generally rely on the same activity data from FAO statistics. However, there are some small differences in the estimates between the data sources. Agricultural emission estimates from FAOSTAT solely rely on the IPCC Tier 1 methodology (FAO 2022). While most of the estimates in EDGAR are also based on Tier 1 methodology, CH4 emission factors for enteric fermentation of dairy and non-dairy cattle have been updated including the IPCC 2006 Tier 2 methodology (Crippa et al. 2019). In addition, EDGAR uses data from both FAO statistics and International Fertilizer Association (IFA) in calculating N2O and CO2 emissions from fertilizer and urea use (Janssens-maenhout et al. 2019). This explains observed differences in the emission estimates between FAOSTAT and EDGAR. Quantifying agricultural emissions has clearly received less attention than forestry and other land use emissions, and the hidden assumption in estimates based largely on Tier 1 factors is that there is not a lot of spatial or temporal variability in these emissions. This is an argument for a greater diversity of approaches to better estimate the spatial and temporal variation of these emissions.

The results show that emissions from biomass burning in agricultural areas are highly uncertain, with FAOSTAT estimating small emissions compared to the other two data sources. The higher emissions in EPA compared to EDGAR, particularly for sub-Saharan Africa, is mainly because EPA estimates include emissions from savannah burning for agricultural expansions and these are not accounted for in EDGAR (Crippa et al. 2019). A lack of detailed and comprehensive reporting of national emissions from agricultural activities like that of FOLU limits a comparative analysis of these emissions. Future comparative analysis could benefit from improved agricultural reporting, and this would help in understanding the limitations of existing agricultural emissions databases.

Only FAOSTAT provides both agriculture and FOLU emissions. Summing the values in Table 2 and 4 results in 6.14 and 7.00 GtCO2 eq. year−1 AFOLU emissions globally and for the tropics over the period 2011–2018. The global value is much lower than the 12 GtCO2 eq. year−1 estimated by Lamb et al. (2021) for 2018. The latest IPCC AR6 report offers net global emission values of 6.49 GtCO2 eq. year−1 for LULUCF only (11% of the global 59 GtCO2 eq. year−1 GHG emissions) (Figure TS.6 in Pathak et al. (2022)); thus, the emissions presented here are much lower than the AR6 estimate. The main reason for these lower emissions is they do not account for CH4 and N2O emissions from savannah burning and peatland and forest fires. Of the three FOLU emission datasets in our analysis, only FAOSTAT explicitly includes emissions from forest fires, but these are relatively small (0.26 GtCO2 eq. year−1 for 2011–2018) and only account for CO2 emissions only. In addition, CH4 and N2O emissions from drained peatland soils and wetlands over managed lands are still not accounted for in the databases and in the recent AR6 estimates, although these emissions can be high (Hergoualc’h et al. 2020; Swails et al. 2021). Integrating emissions from fires, peatlands, and wetlands in the country reporting and global databases would also provide a more complete assessment of the AFOLU emissions.

5 Conclusion

This study provided a detailed comparative analysis of the emissions estimates from available global datasets for the AFOLU sector. The total net FOLU emissions for the tropical regions ranged between 1.17 and 5.47 GtCO2 year−1 compared to global values of −2.03 to 5.16 GtCO2 year−1 over the period 2011–2018 in the four analysed datasets. Disaggregated analyses show that these emissions are mainly from deforestation (2.96 to 7.16 GtCO2 year−1) and partly from other land-use activities (0.45 to 1.56 GtCO2 year−1), which exceed the net forestland sink (−0.53 to −3.08 GtCO2 year−1) and make the tropical region a net source of emissions for the FOLU sector. In terms of trend, three of the five data sources (updated H&N2017, FAOSTAT, and NGHGIs) indicate a decline in the net emissions over the tropics, which is also present globally. The net agricultural pantropical emissions range between 3.08 and 3.70 GtCO2 eq. year−1, with all the data sources showing an increase. Emissions from deforestation and biomass burning are the most uncertain for the agricultural sector are highly uncertain for the tropical region.

Our analyses show that for global FOLU assessments, efforts have been made to provide more detailed datasets that allow for an assessment of not only net FOLU emissions, but also gross fluxes for the different contributing activities to this sector. Despite these advancements and the availability of more years of data, we find that there are still important gaps and that many emission sources are still not fully accounted for in the global databases. This is not new as Roman-Cuesta et al. (2016a) had already pointed out to the incomplete accounting of sources, such as from forest fires and peatland emissions, and not including non-CO2 in the FOLU estimates, particularly N2O and CH4 emissions from forest fires. Given that most of the considered databases are likely to play a major role in the global stocktake, efforts should shift towards ensuring that all emission sources are fully accounted for, and more consistency is developed. Moreover, there is an urgent need to offer geographically appropriate AFOLU emissions factors, so that land managers, policy makers, and society can efficiently concentrate efforts and resources towards the existing hotspots of emissions. Aggregated country estimates for greenhouse gas emissions remain a barrier for targeted action and improving mitigation governance. In that line, spatial data such as those published by Roman-Cuesta et al. (2016b) and Harris et al. (2021) show a way forward. Nevertheless, our analysis shows that the existing databases can be an important resource in providing a benchmark for NGHGIs in assessing the plausibility of their estimates, while investing in more detailed, complete, and accurate data is a key priority to underpin the current and future global stocktakes, considering both national to global level assessments.