Carbon Emissions and Removals from Forests: New Estimates, 1990-2020

. NTrends in ational, regional and global , regional, and national CO 2 emissions and removals from forests were estimated , for the period 1990–2020,, using as input the are estimated by FAO and disseminated in FAOSTAT. using We document a major product update, based on the new country reports data from tpublished byof the Global Forest Resources Assessment (FRA) 2020, which replaces and updates previous information published under the FRA 2015, based on country reports, providing new information with respect to the previous FRA published in 2015. The FAOSTAT new FAO estimates, based on a simple carbon stock change approach, provideupdate published information on net emissions and removals from forests in relation to, in total as well as by component processesestimates were derived separately for two components: a) emissions from deforestationnet forest conversion; and b) emission and removals from remaining forest lands, including new forests resulting from natural expansion or afforestation. The estimatesResults indicate a significantshowconfirmed a significant reduction of the emissions fromin global net deforestation emissions from net forest conversion, a proxy for deforestation, over the study period, , though at . However, the emission reduction is slower rates than previously assessed, i.e., from a mean n average of 4.3 in the 1991–2000 to 2.9 Gt CO 2 yr -1 during 1991–2000, to an average of 2.9 Gt CO 2 yr -1 during 2016–2020 in 2016–2020. . At the same time, Excluding

drainage of peatlands, generating roughly 4-5 Gt CO2 yr -1 in recent decades (e.g., Tubiello, 2019). Additional important 1 anthropogenic emissions and removals of CO2 are located directly on forest land, in relation to processes linked to forest 2 management or degradation. 3 There is nonetheless significant disagreement between carbon cycle models on the one side, and national greenhouse gas 4 inventories (NGHGI) on the other, on the quantification of the combined emissions and removals of CO2 from all these land 5 processes, though it is being increasingly shown that most differences are due to boundaries and definitional issues (e.g., Grassi 6 et al., 2018;2021). Greatly simplifying and limiting our scope to forests, terrestrial carbon cycle models have tended to focus 7 on the CO2 emissions from deforestation and forestry activities (land use change processes defined under the term ELUC), while 8 NGHGI have typically added removals on forest land beyond those linked to forestry practices, which the models tend not to 9 consider anthropogenic. These forest removals in NGHGI counterbalance the positive emissions, resulting in near-zero 10 estimated total net contributions of forests to the atmosphere (Grassi et al., 2018). Beyond the critical issues of the differences 11 in boundaries and definitions between the two approaches, which are addressed elsewhere (e.g., Grassi et al., 2021), there is a 12 significant need to improve the underlying activity input data used by both approaches. To this end, the Food and Agriculture 13 Organization of the United Nations (FAO) collects, analyses and disseminates at regular intervals a wealth of country-based 14 forest statistics through its Global Forest Resources Assessment (FRA), describing the status of forests with data at country, 15 regional and global level (FAO, 2020a). FRA activity data of forest land area and carbon stock serve as critical inputs for 16 estimates of forest carbon fluxes by FAO (Federici et al., 2015;FAO, 2020b) and other major international efforts (e.g., 17 Friedlingstein et al., 2019;IPCC, 2019;Houghton and Nassikas, 2017). This paper describes the forest statistics available at 18 FAO to estimate emissions and removals of CO2 from forests that, being based on a simple though powerful (and replicable) 19 carbon stock change method, generate data that can serve as boundary conditions to help evaluate more complex terrestrial 20 carbon model results and NGHGI data. Our analysis highlights new trends based on the use of FRA 2020 input data, 21 documenting the differences with respect to the previous use of FRA 2015. Finally, it compares results to national data 22 independently reported by countries to the United Nations Framework Convention on Climate Change (UNFCCC). 23

Material and Methods 24
The estimates of CO2 emissions and removals from forests made by FAO and published in FAOSTAT (FAO 2020b) are 25 computed by applying a simplified carbon stock change method based on the 2006 IPCC Guidelines for National Greenhouse 26 Gas Inventories (IPCC, 2006). Previous estimates covered the period 1990-2015, using as inputs activity data from the FRA 27 2015(Federici et al., 2015. This work extends the FAO estimates of emissions and removals to 2020, by adding new input 28 data for the period 2015-2020, while incorporating any revision in time series that may have occurred in the FRA 2020 with 29 respect to FRA 2015. In describing the methods used in this work, we also discuss their limitations and uncertainties and the 30 scope for comparing FAO estimates to UNFCCC country data. 31 cover information derived from remote sensing can results in differences of up to 20% at regional level, largely due to the 23 difficulty of mapping land cover characteristics to land use status (FAO, 2020c). For well-defined forest land areas, typical 24 uncertainties in national forest inventories may be nonetheless an order of magnitude smaller. For lack of additional knowledge 25 of how uncertainty in local measurements carried out at national to regional levels, we applied the generic uncertainty suggested 26 by IPCC for FAO activity data (20%) to the forest land area and biomass stock data used in this work. 27

28
In terms of comparison with UNFCCC data, we note that the FAO forest land use definitions used herein may differ from those 29 used by countries for reporting their national GHG inventories (NGHGI), for instance in relation to minimum forest area 30 thresholds or in criteria to assign land use status. Furthermore, country reporting to UNFCCC of emissions and removals data 31 is limited to areas of managed forest, as per IPCC guidelines, while the FAO land use definitions comprise both managed and 32 unmanaged forests, as discussed above. In practice, such differences may often be small, considering that a large portion of 33 the world's forest land area in many countries is administratively regulated. Finally, we note that the FAO forest land area 1 considered herein does not track separately, as done instead in UNFCCC reporting, the two-sub-components forest land 2 remaining forest land (FL-FL) and newly converted forest land. This is often overlooked in the literature, where FAO estimates 3 of forest land emissions and removals may be incorrectly compared to UNFCCC data for FL-FL (e.g., Petrescu et al., 2020). 4 5

Emissions and Removals 6
The estimates presented herein provide information on total net emissions and removals from forests, in total as well as by 7 component processes. Specifically, for each country a and total carbon stock Ba, the total forest emissions/removals, ERa, were 8 computed as a simple carbon stock change, as follows: 9 10 ERa(ti) = -ΔCa(ti) = -[ Ba(ti) -Ba(ti-1)] = NFCa(ti) + FLa(ti) (1) 11 12 Where biomass stock information was derived from the FRA 2020 as indicated in the previous section, and ti = 1990, 2000, 13 2010, 2015, 2020 represent FRA years. The minus sign was used to adhere to the convention of considering emissions as 14 positive fluxes to the atmosphere, corresponding to decreases in forest carbon stock-and vice-versa to consider removals as 15 negative fluxes, i.e., from the atmosphere into forest land, corresponding to increases in forest carbon stock. We note that the 16 estimates in equation (1) are robust as well as easily replicable by anyone having access to FRA data. At the same time, it is 17 noted that the FAO carbon stock change estimates include only two of the five carbon pools typically reported by countries 18 according to IPCC. This difference may affect the magnitude of the estimated C stock changes, although likely not the sign, 19 because of biophysical linkages across carbon pools. The net forest signal to the atmosphere, ER, was split into two mutually 20 exclusive components, specifically emissions from net forest conversion, NFC, and emissions/removals from forest land, FL 21 ( Fig. 1). 22 23

Emissions from Net Forest Conversion 24
For each country a, total carbon stock Ba, and time period ti, the emissions from net forest conversion, NFCa(ti) in equation (1)  25 were computed as the positive carbon flux to the atmosphere associated with net forest land area loss, tracked separately for 26 sub-categories naturally regenerating forest, NRa, and planted forest, PLa as follows: Thus net forest conversion tracks losses of both naturally regenerating (including primary and secondary forests) and planted 31 forest areas. It should be noted that in cases when net forest land area change is positive, indicating net area gains, NFC is zero 32 by definition and the relevant emissions/removals are instead accounted for on forest land (see next section). A number of 33 limitations apply to the computation of emissions in (2), First, results are limited by the lack of carbon stock data by forest 1 sub-component, resulting in the need to apply a single value for both naturally regenerating forest and planted forest. 2 Considering that the majority of forest area losses in the FRA 2020 pertain to the natural forest component, however, the use 3 of a single carbon density value in (2) is not a significant issue to this end. At the same time, carbon stock density can be 4 expected to be higher in natural forests than the average biomass stock (which also includes carbon stock in plantations), 5 implying that the NFC emissions computed in (2) are likely underestimates. Furthermore, we note that equation (1) above does 6 not depend on the availability of carbon stock values by forest sub-component. A second important limitation to equation (2)  7 is that forest losses are computed net of forest area gains taking place over the same period. The underlying activity data used 8 as input do not in fact allow separate tracking of gross gains and losses. Thus in terms of comparison to UNFCCC, FAO net 9 forest conversion data would roughly correspond to the sum of UNFCCC-reported land use changes from forest land to non-10 forest land, for those countries using the so-called 'IPCC approach 1' to land use representation, which like our estimates relies 11 on net area changes. By contrast, use of more accurate national forest inventories, with more detailed identification of gross 12 area fluxes, would generate larger differences between FAO estimates and the corresponding UNFCCC country data for this 13 category. Finally and importantly, estimates in equation (2)

Emissions and Removals on Forest Land 20
For any country a, total carbon stock Ba, and time period ti in equations (1) and (2) above, the emissions/removals on forest 21 land, FLa(ti), were computed as the residual between total forest carbon stock change and net forest conversion, as follows: 22 23 FLa(ti) = ERa(ti) -NFCa(ti) ( 3)  24   25 The emissions/removals computed in (3) represent the net carbon flux to or from the atmosphere located within the boundaries 26 of forest land area, arising from a combination of carbon stock and forest area changes between successive FRA periods. These 27 changes in principle may arise from both anthropogenic and natural causes, including legacy effects of deforestation prior to 28 the study period, afforestation, forest management, climate signals, as well as the impacts on plant growth of nitrogen 29 deposition and increased atmospheric CO2 concentrations. As discussed above, we associated an uncertainty level of 50% to 30 estimates in equation (3), consistently with those computed for the emissions from net forest conversion and in line with the 31 uncertainty used in the literature. 32 Within the differences highlighted above, with regards to land accounting approaches and differences in national forest 1 definitions, the FAO emissions/removals on forest land largely correspond to those used by countries in their reporting to 2 UNFCCC with respect to forest land. 3 4

Comparisons to UNFCCC data 5
A final consideration on the limitations of the approach presented herein concerns the underlying drivers of the 6 emissions/removals estimates, i.e., whether they could be labelled as anthropogenic or natural fluxes. On the one hand, the 7 definitions underlying equation (1)-(3) make the association impossible within our approach. On the other, a bit more can be 8 said in practice. This is because human intervention is typically required to determine land use changes-for instance the 9 establishment of specific activities, for instance agriculture, preventing natural forest regrowth and recovery following forest 10 loss. To this end, and within the limitations discussed above, net forest conversion, representing permanent forest loss in the 11 FAO statistics, can be considered virtually all anthropogenic in nature, hence a good proxy for human-driven deforestation. 12 Conversely, only a portion of the emissions/removals estimated on forest land can be considered anthropogenic. At the same 13 time, recent work shows that the anthropogenic portion of this component can be substantial, once the concept of 'managed 14 land' is expanded beyond forestry practices to include all forest areas except in very remote places (Grassi et al., 2021). 15 Nonetheless, because of the above complexities, we chose not to determine a priori the anthropogenic portion of our 16 emissions/removals estimates. Instead, we complemented our analysis of results with a comparison between our estimates of 17 emissions and removals and the anthropogenic fluxes submitted by countries to UNFCCC. In this context, although it is 18 recognized that countries report data to both FAO and UNFCCC, we reserve herein the term 'country data' to the 19 emissions/removals reported by countries to the UNFCCC. 20 To this end, we used country data accessed at the UNFCCC data portal (UNFCCC, 2020) and complemented with information 21 from national Biennial Update Reports (BURs). While data from Annex I countries (AI) are fairly complete over the period 22 1990-2018, data from non-Annex I (NAI) countries are sparse, although becoming increasingly available through BURs. 23 Given these data limitations, a full comparison was possible only for Annex I countries for the FRA periods 1990FRA periods -20002001-24 2010and2011-2015. First, we compared results of equation (3) with aggregate Annex I reporting of emission/removal for 25 the category '4.A Forest land' (UNFCCC, 2020). To gain further insights, we also separately analyzed emissions/removals on 26 forest land reported by individual countries in their national GHG inventories (NGHGIs), focusing on those reporting large 27 sinks, i.e., Canada, Russian Federation and the United States of America among Annex I parties, and China among non-Annex 28 I parties. We also compared our results for net forest conversion to available non-Annex I country data from Brazil and 29 Indonesia, representing large emission sources, according to FAOSTAT estimates respectively the first and third emitters in 30 this category (FAO, 2020b). Unfortunately, no BUR submissions have been made so far by the Democratic Republic of 31 Congo-the second largest emitter from deforestation according to FAOSTAT data-which therefore could not be included 32 7 in this comparison exercise. Data for NAI countries were sourced from China's second Biennial Update Report (2018), Brazil's 1 third Biennial Update Report (2019) and from Indonesia's second Biennial Update Report (2018). 2 3

Structure of the datasets on emissions-forest land and online access 4
The FAO emissions and removals estimates and associated area information statistics are disseminated in the FAOSTAT 5 Emissions Land Use/ Forest Land domain as yearly statistics, over the period 1990-2020 (FAO, 2021), for 220 countries and 6 territories. Annual mean fluxes are obtained by dividing the outcomes of (1)-(3) by the relevant time-period underlying FRA 7 intervals, i.e., by 5 or 10 years. They therefore refer to the following periods : 1991-2000; 2001-2010; 2011-2015; and 2016-8 2020. For completeness, values for the year 1990 were set equal to the averages computed for 1991-2000, and the full period 9 of analysis was referred to as 1990-2020. Data include, by country and year, forest land area and area of net forest conversion 10 (in 1000 ha), emissions from net forest conversion; emissions/removals on forest land; and total emissions/removals from 11 forests (in kt CO2). The carbon stock in living biomass (in Mt C) is available under the FAOSTAT database, Inputs/Land Use 12 (FAO, 2020c). Data are disseminated by country, by standard FAO regional aggregations and special groups, including the 13 Annex I and non-Annex I country grouping relevant to UNFCCC reporting. 14

Results 15
We present below the main findings of annual CO2 emissions/removals estimates from net forest conversion, forest land, and 16 their aggregate, total emissions and removals from forest, for the period 1990-2020, computed for more than 200 countries 17 and territories, based on equations (1)

Emissions from Net Forest Conversion 22
Results show that carbon fluxes to the atmosphere from net forest conversion were significant, with world-total means of 3.7 23 Gt CO2 yr -1 for the period 1990--2020, and almost entirely located in non-Annex I countries, which contributed more than 90 24 % of the world total (Table 1). In terms of temporal trends, the global mean decreased by 20% from 1990 to 2015, from 4.3 to 25 3.3 Gt CO2 yr -1 , less than previously estimated over the same period using the FRA 2015 (-40 %). It decreased by another 10% 26 to 2.9 Gt CO2 yr -1 during 2016-2020. For the period 2016-2020, the Americas and Africa were nearly equal contributors, but 27 with markedly opposite trends compared to the period 1991-2000. Specifically with respect to the two time periods, emission 28 in the Americas nearly halved, from 2.2 to 1.3 Gt CO2 yr -1 , while they increased in Africa, from 0.9 Gt to 1.1 CO2 yr -1 . Asia 29 8 was the third region in terms of emissions from net forest conversion, showing a slight decrease, from 0.6 Gt to 0.4 CO2 yr -1 1 over the same time periods (Fig. 2). 2 3

Emissions and removals on forest land 4
Emissions/removals on forest land showed a net sink over the entire period 1990-2020, with a mean removal of -3.3 Gt CO2 5 yr -1 globally. This forest carbon flux was nearly equally divided between Annex I (-1.8 Gt CO2 yr -1 ) and non-Annex I countries 6 (-1.5 Gt CO2 yr -1 ) (Table 1). Additionally, we computed that the new FAO estimates indicated a stronger forest sink than 7 previously estimated using FRA 2015 data, i.e., on average 1.0 Gt CO2 yr -1 (35 %) stronger, due to larger estimated sinks in 8 Europe (dominated by trends in Russian Federation) and Asia (China). 9 At the same time, the estimated global forest land sink weakened in strength over the study period, with the world total mean 10 decreasing from -3.5to -2.6 Gt CO2 yr -1 , i.e., about 20 % decrease from 1990 to 2020. The period 2011-2015 represented an 11 exception to this decreasing trend, showing the strongest forest sink over the study period, with mean world total rates of -4.0 12 Gt CO2 yr -1 . In terms of regional distribution and averaged over the period 1990-2020, Europe, the Americas and Asia nearly 13 equally contributed to the estimated forest land removals, within a narrow range of -1.0 to -1.2 Gt CO2 yr -1 , with Europe 14 (including the Russian Federation) being the largest contributor. Conversely, forest land in Africa was a source to the 15 atmosphere, with mean emissions increasing significantly from 2000 to 2015, i.e., from 1.4 to 43 Mt CO2 yr -1 (Fig. 3). By 16 associating net forest land emission to forest degradation, as done in Federici et al. (2015), our results suggest a significant 17 relative increase in forest degradation (defined as carbon stock reduction in forest land) in Africa over the last twenty years. 18

Total emissions and removals from forests 19
Our estimates show that the net effects of emissions from net forest conversion and removals on forest land were a small net 20 source of CO2 emissions to the atmosphere, with a world total mean of 0.4 Gt CO2 yr -1 over the 1990-2020 period. This new 21 estimated value was significantly less than reported earlier based on FRA 2015 data (Table 1). It is further of interest to note 22 that the estimated small global source was the result of a balance of larger fluxes: a net sink on forest land, largely located in 23 in UNFCCC Annex I countries (-1.5 Gt CO2 yr -1 ), counterbalanced by a net emission source from net forest conversion, mainly 24 in non-Annex I countries (1.9 Gt CO2 yr -1 ). 25 A more detailed analysis focusing on trends over time (Fig. 4) revealed two notable new findings of our analysis with respect 26 to previous results. First, the period 2015-2020 saw a reversal of the decreasing trend in non-Annex I sources and the increasing 27 trend in Annex I sinks seen for the period 1990 to 2015. Specifically, non-Annex I sources from net forest conversion began 28 to increase again in 2016-2020, from 1.3 to 1.6 Gt CO2 yr -1 , while Annex I sinks on forest land began decreasing in strength, 29 from -2.0 to -1.3 Gt CO2 yr -1 . Second, and remarkably, forests acted as a net overall sink of atmospheric CO2 during 2011-30 2015, averaging nearly -0.7 Gt CO2 yr -1 , largely a result of decreased emissions from net forest conversion in this period, 31 counterbalanced by a strong sink on forest land. Conversely, FAO had previously estimated for the same period, based on FRA 1 2015 input data, a net emission source of 1.1 Gt CO2 yr -1 (Table 1). 2 3

Comparisons with UNFCCC 4
Forest Land 5 As discussed in the methodology section, we first compared our estimates of emissions/removals on forest land to data reported 6 by Annex I countries, i.e., for category "4.A Forest land" in their national GHG inventory (UNFCCC, 2020). In the aggregate, 7 e.g., summing up all country data and averaging over the period 1990-2015, our estimates agreed in both sign and magnitude 8 with the UNFCCC country data (14 % relative absolute error). Specifically, our estimates indicated a mean sink of -1.8Gt CO2 9 yr -1 vs -2.2 Gt CO2 yr -1 reported. Using the FRA 2015 in earlier work (Federici et al., 2015) had given a 33 % smaller sink 10 (Table 2). Our estimates were particularly well aligned with country reporting for the period 2010-2015, i.e., within 5 %, 11 predicting a sink on forest land of -2.1 Gt CO2 yr -1 vs -2.2 Gt CO2 yr -1 reported. As in the previous case, earlier sink estimates 12 based on the FRA 2015 were 40 % smaller (Fig. 5). 13 Comparisons of estimated emissions/removals on forest land for specific countries with large reported sinks confirmed the 14 good overall agreement found for Annex I parties in aggregate. For instance, on average over the period 1990-2015, our 15 estimates of forest land sinks for the Russian Federation were within 5 % of those reported by the country NGHGI. Agreement 16 with NGHGI data was even closer after the year 2000, i.e., for the period 2001-2010 our estimates indicated a mean sink on 17 forest land of -800 Mt CO2 yr -1 versus country data of -750 Mt CO2 yr -1 , and a mean sink of -730 Mt CO2 yr -1 versus -680 Mt 18 CO2 yr -1 for the period 2011-2015 (Fig. 6). Comparisons for the USA were also encouraging, albeit with larger differences 19 than found for the Russian Federation. On average over the period 1991-2010, the FAO estimates were of a 25 % smaller sink 20 on forest land compared to the NGHGI country data. Averaged over the period 2011-2015 our estimates were 29 % smaller 21 than the country data, or -460 Mt CO2 yr -1 and -650 Mt CO2 yr -1 , respectively. 22 We performed comparisons for China, using data from the country's Second Biennial Update Report (2018), to extend our 23 analysis to non-Annex I countries reporting large sinks on forest land. Specifically, we used national data on total removals 24 from LULUCF for the period 2011-2015. We concluded that China LULUCF data were a good proxy for forest land data, 25 considering that: 1) zero emissions from net forest conversion were indicated in the same BUR; and 2) emissions from cropland 26 and grassland-the other main component of LULUCF within a national inventory-were likely small, as indicated by 27 independent emissions estimates published in FAOSTAT (FAO, 2020b). Within these assumptions, our estimates of a sink on 28 forest land in China for the period 2011-2015 agreed well with country data (within 20 % of country data), i.e., -710 Mt CO2 29 yr -1 compared to -840 Mt CO2 yr -1 reported to UNFCCC (Fig. 6). 30 Conversely, our estimates of emissions/removals on forest land did not agree well to those reported by Canada. Our results 31 indicated a net source on forest land, declining from 2000 to 2015, whereas the NGHGI country data reported a progressively 32 smaller sink over the same period (Fig. 6). Specifically for the period 2011-2015, our estimates indicated a weak net source, 33 about 23 Mt CO2 yr -1 , compared to a net sink of -150 Mt CO2 yr -1 in the country data. Finally, our estimates for the most recent 1 period, 2016-2020, for which however there is no available NGHGHI data yet from the country, began to show a sink on 2 forest land, of -80 Mt CO2 yr -1 , thus indicating a possible alignment with NGHGI data in recent years. A possible reason for 3 the discrepancies found in this case may relate to differences in land use definitions, particularly those related to managed 4 forest land. For the purpose of the NGHGI, in fact, the area of managed forests defined by Canada is 65 % of the total forest 5 land area reported to FAO (Canada's 7th National Communication and 3rd Biennial Report, 2017;Ogle et al., 2018). 6 7 Net forest conversion 8 We also compared estimates of emissions from net forest conversion with data reported to UNFCCC. As discussed in the 9 methodology section, FAO estimates of emissions from net forest conversion are proxies for deforestation emissions data. The 10 two countries for which relevant data were available were Brazil and Indonesia. For Brazil, we compared our estimates of net 11 forest conversion directly to deforestation emissions from the country's BUR. For Indonesia, we compared our estimates to 12 sum of LULUCF emissions arising from land use change to cropland and grassland-assuming, in line with current 13 understanding of deforestation trends in this country, that land converted to cropland and grassland in Indonesia originated 14 largely from loss of forest land area. For Indonesia, for the period 1991-2000, our estimates of emissions from net forest 15 conversion greatly overestimated country data for deforestation, by over a factor of 10 (Fig. 7). Conversely, for the more recent 16 period 2011-2015, they were on average within 25 % of country data, specifically 180 Mt CO2 yr -1 vs country data of 165 Mt 17 CO2 yr -1 . Our estimates further suggested a 50% decrease in emissions from net forest conversion in the period 2016-2020, for 18 which however BUR data are not yet available (Fig. 7). 19 For Brazil, our estimates were in good agreement (within 10 %) of country data over the period 1990 to 2015, i.e., on average 20 1.4 vs. 1.5 Mt CO2 yr -1 reported data (Fig. 7). More in detail by decade, our estimates were 1.4 vs 1.9 Gt CO2 yr -1 during 1991-21 2000 and 1.6 vs 1.6 Gt CO2 yr -1 over 2001-2010. Conversely, for the period 2010-2015, our estimates of emissions from net 22 forest conversion were significantly higher than reported in the BUR. 23

Discussion 24
The availability of new forest area and carbon stock data from the FRA 2020 enabled a new analysis of the role of forests in 25 generating CO2 emissions and removals at country, regional and global level, during the period 1990-2020. In particular, the 26 new information allowed us, for the first time in the literature, to estimate emissions and removals relative to the most recent 27 decade, covering the period 2011-2020. Our findings indicate that in the decade just concluded the net contribution of forests 28 to the atmosphere, representing the combination of emissions from net forest conversion and removals on forest land, was very 29 small, i.e., an overall emission sink of less than -0.2 Gt CO2 yr -1 , estimated for the first time in the literature for this period. It 30 nonetheless resulted from the balance of large global fluxes of opposite sign, namely mean net forest conversion emissions of 31 3.1 Gt CO2 yr -1 , counterbalanced by mean net removals on forest land of -3.3 Gt CO2 yr -1 (Table 1). Both fluxes, and hence 32 the overall net near zero balance for forests, were shown to be in very good agreement with the data reported by countries in 1 national GHG inventories, and in line with independent findings by Grassi et al. (2021). At the same time, the consistency of 2 our estimates with those of terrestrial carbon cycle models were limited to the anthropogenic carbon flux from forests to the 3 atmosphere (i.e., IPCC, 2019). Results further showed that, with respect to the previous decade 2001-2010, emissions from 4 net forest conversions had decreased by 15 %, while removals on forest land had decreased by 5 %. Further analysis of the 5 underlying FRA 2020 data (not shown) indicated that such decreases were due to a reduced pace of natural expansion and 6 afforestation in Annex I countries, which have functioned historically (1990-2020) as forest sinks, as well as a decrease in 7 forest loss in non-Annex I countries, which have represented the bulk of deforestation. The new estimates also show that over 8 the earlier period 1991-2010 forests were a smaller net source of emissions than previously calculated (Federici et al. 2015). 9 largely due to much stronger sinks on forest land estimated using the new FRA 2020 as opposed to FRA 2015 data, respectively 10 for Europe (+ 0.7 Gt CO2 yr -1 ) and Asia (+ 0.6 Gt CO2 yr -1 ). 11

12
The main new finding of this work is the large estimated sink on forest land over the period 2011-2015, averaging -4.0 Gt CO2 13 yr -1 , causing the overall net negative carbon flux from forests highlighted in the results section. Notable contributors to this 14 forest land sink were the Russian Federation, USA, China, Indonesia and India, which all had stronger carbon uptake compared 15 to the previous 2001-2010 period. Comparisons with country data reported to the UNFCCC support our estimates, indicating 16 that they represent an improvement compared to previous results. In particular, the good agreement between our new estimates 17 and country NGHGI data on emissions/removals on forest land and emissions from net forest conversion suggests that the 18 definition of forest land area underlying both FAO and UNFCCC reporting was consistent across the countries considered, 19 i.e., they considered most of the forest land area reported to FAO as managed for UNFCCC purposes-confirming the analysis 20 provided in the methodological section of this paper. This implies that, limited to the countries tested and within the range of 21 limitations discussed earlier in this paper, the estimates of emissions and removals from forests provided in this paper can be 22 considered largely anthropogenic. Finally, the good agreements found between our estimates and country reports support the 23 finding of a large anthropogenic sink on forest land for the period 2011-2015, leaving open the possibility, in need of 24 verification in coming years, that even when considering deforestation, the world forests were a small sink, rather than a source 25 of atmospheric carbon during this period. In fact, the discussed progressive reduction of the overall forest source observed 26 across the two most recent decades is consistent with these findings. 27

Data availability 28
The emissions and removals data, alongside with input activity data of forest land area and carbon stock, are disseminated in 29 FAOSTAT (FAO, 2021). An exact replica of the data used for this paper is available as open access at 30 http://doi.org/10.5281/zenodo.3941973 (Tubiello, 2020). 31

Conclusions 1
Estimates of CO2 emissions and removals from forests were updated based on the most recent FRA 2020 data and by applying 2 a simple yet robust, transparent and easily replicable carbon stock change approach. Over the period 1990-2020, result 3 confirmed known country, regional and global trends, providing additional detail to specific dynamics while extending 4 available information to the period 2016-2020. Importantly, the new estimates allowed to characterize for the first time forest 5 emissions and removals for the decade just concluded, 2011-2020, showing that in this period the net contribution of forests 6 to the atmosphere was very small, sink i.e., less than -0.2 Gt CO2 yr -1 . This near-zero balance was nonetheless the result of 7 large global fluxes of opposite sign, namely net forest conversion emissions of 3.1 Gt CO2 yr -1 counterbalanced by net removals 8 on forest land of -3.3 Gt CO2 yr -1 . 9 Tables  1   2   Table 1. Estimates of total emissions and removals from forests (ER), net forest conversion (NFC) and emissions/removals on forest 3 land (FL) for World, Annex I and non-Annex I totals, based on FRA 2020 and FRA 2015 (Gt CO2 yr -1