Land use change and land management accounts for approximately 14% of total global anthropogenic CO2 emissions, predominantly due to forest fluxes, but with an uncertainty of around ±50%, it is the most uncertain term in the carbon budget1,2. The Land Use Land Use Change (LULUCF) sector is unique because it not only includes anthropogenic Greenhouse Gas (GHG) emissions but also considers anthropogenic sinks such as removals from afforestation/reforestation and, to different extents, forest conservation on lands considered to be “managed”. The numerous approaches to estimate LULUCF each have their own scope, intentional use, input datasets and methods. Recent analysis found a difference of ~5.5GtCO2yr-1 between national GHG inventories (NGHGI) and global models3. This difference can largely be explained by the extent to which each approach considered the forest sink to be “managed” and thus anthropogenic3.
It has been hoped that Earth Observation-based estimates might reduce the uncertainty in the LULUCF flux. They can support climate policy by providing local, spatial detail to aid countries’ reporting and verification of their NGHGI as well as potentially providing a benchmark to evaluate the land sink from global models in a globally consistent manner. However, as noted in the (draft) IPCC 6th Assessment Report, there are large differences between CO2 flux estimates based on activity data of forest cover loss and gain from global Earth Observation data4 and other methods1-3. The comparison in the draft IPCC 6th Assessment Report showed that, globally, Earth Observation-based estimates4 find a considerable global forest sink of -6.7 GtCO2 yr-1 during the period 2001 – 2019 in non-intact (i.e. managed) forests. In contrast, the global LULUCF flux is estimated to be a small net sink of -1.1 ±1.0 GtCO2 yr-1 in NGHGIs3 and a considerable net source of 5.7 ±2.6 GtCO2 yr-1 in global bookkeeping models2 for the period 2010 - 2019. This Earth Observation result therefore increases uncertainty in the LULUCF net flux, as noted in the (draft) IPCC 6th Assessment Report. Understanding the reasons for the differences is crucial if Earth Observation is to be used in the context of supporting countries inventories and national/global policy.
Here we conduct a detailed comparison of the methods and datasets used to produce inventories and the Earth Observation-based estimates of forest-relaxed carbon fluxes2. We aim to (i) quantify the differences between estimates of forest-related GHG fluxes based on Earth Observation data and NGHGI; (ii) provide an understanding of the potential reasons for the differences on a (sub)country-level scale; and (iii) provide recommendations for policy makers and inventory compilers on the best approach to using global Earth Observation within reporting and verification.
We use different land cover masks to compare fluxes from the Earth Observation study4 across equivalent areas of forest that could be considered managed according to national inventories. Results show the Earth Observation based estimates of the non-intact forest flux to be a large net carbon sink, whilst all other methods suggested the LULUCF sector to be a small net source on a global scale, with similar findings regionally. We have carried out an initial detailed analysis focusing on Brazil and the Amazonian biome where there is in-country remote sensing data available at high spatial and temporal resolution, which are used during the compilation of the NGHGI. We have identified several reasons for differences and discuss implications for improving methods as well as comparability between Earth Observation approaches and inventories.

References:
1 Jia et al., Chapter 2 Land-Climate interactions. In: Climate Change and Land: an IIPCC special report on climate change, desertification, land degradation, sustainable land management, food security and greenhouse gas fluxes in terrestrial ecosystems. Editors: Shukla et al. 2019.
2 Friedlingstein et al. Global Carbon Budget 2021. Earth Syst. Data Discuss (preprint), https://doi.org/10.5194/essd-2021-386, in review, 2021.
3 Grassi, G. et al. Critical Adjustment of land mitigation pathways for assessing countries’ climate progress. Nature Climate Change. 11, 425-434 (2021). https://doi.org/10.1038/s41558-021-01033-6
4 Harris, N.L. et al. Global maps of twenty-first century forest carbon fluxes. Nat. Clim. Chang. 11, 234–240 (2021). https://doi.org/10.1038/s41558-020-00976-6

Biomass, and especially forest biomass, plays a crucial role in sequestering and storing carbon and is a key component of both national Greenhouse Gas (GHG) inventories and the global carbon budget. Accordingly, mapping aboveground biomass is a priority of space agencies such as NASA, ESA and JAXA and is highlighted by the several new and upcoming satellite missions, including GEDI, ICESat-2, BIOMASS, ALOS-2, ALOS-4 and NISAR. Using satellite biomass products seems a natural progression for country Parties when preparing their reports of GHG emissions and removals from the forest sector to the United Nations Framework Convention on Climate Change (UNFCCC). In particular considering the new 2019 Refinement to the 2006 IPCC Guidelines that include generic guidance on the use of biomass density maps for GHG inventories. However, the availability of multiple satellite products can be confusing to policy users and national technical teams. The Earth Observation (EO) global biomass monitoring community recognizes that widespread uptake of these biomass products requires their differences to be addressed, their accuracies known, and associated metadata provided to derive estimates compliant with the IPCC guidelines on bias and uncertainty. Furthermore, products need to be flexible to adhere to the land representation according to national definitions. A dedicated international effort coordinated through the Committee of Earth Observing Satellites (CEOS) and the Global Forest Observation Initiative (GFOI) R&D component is addressing some of the existing limitations that jeopardize the successful uptake of biomass products by the UNFCCC, including by national GHG inventory teams. The objectives of the CEOS Agriculture, Forestry and Other Land Use (AFOLU) group, and specifically its Biomass Harmonization Team, are to develop i) “best available” harmonized global forest biomass estimates from the next generation of biomass maps as input to the Global Stocktake (abstract by Duncanson et al.) and ii) examples of the practical implementation of the 2019 Refinement to the 2006 IPCC Guidelines to enhance the uptake of these maps by countries in their reporting to the UNFCCC.

Here we demonstrate these two objectives are only successfully achieved if the disconnect between producers and users is addressed and the interface for collaboration between EO global biomass monitoring, GHG inventory, and National Forest Inventory (NFI) experts is created. For the first objective, NFI plot data together with national expertise are required to understand and quantify the accuracy and usefulness of the maps. To achieve both objectives, we show how a collaborative interface was created between CEOS agencies and national teams from several countries (e.g., Japan, Paraguay, Peru, Solomon Islands, Wales UK) and emphasise the preparatory work required for an effective dialogue between the different groups. First, it is necessary to recognize that only individual countries, given their national circumstances, can determine if satellite-based data and derived products developed over large areas are suitable for use in national GHG inventories. To determine needs and requirements regarding the potential use of space-based biomass data we analyzed the information available in national submissions to the UNFCCC and in the reports from the technical review process, in particular in the section covering areas for future technical improvement. The second step is to understand the IPCC Guidelines, which IPCC variables we need biomass information for, what methods are used and assumptions made by national teams, and reflect on possible procedures to derive those estimates with associated bias and uncertainty according to the Guidelines. Finally, a mapping exercise between variables, areas for improvement of the estimates for those variables, and the characteristics of the available products support the discussions on how the data needs to be handled and presented following IPCC guidance and principles. We conclude with a demonstration on how this strategy and interface between CEOS agencies and national technical teams led by SilvaCarbon provided a few first examples of the successful uptake of maps in reporting to the UNFCCC and of the practical implementation of the 2019 IPCC Refinement.

Numerous global and regional biomass maps have been produced in recent years by multiple international programs using a combination of Earth observation, ground, and airborne data. Among these efforts, ESA’s Climate Change Initiative (CCI) Biomass project, has released global biomass maps for four epochs. Recent and upcoming space missions even have biomass maps as dedicated target products. Examples include the ESA Biomass mission (P-band SAR) and the NASA Global Ecosystem Dynamics Investigation (GEDI) mission (waveform lidar). The potential usefulness of such maps for greenhouse gas (GHG) reporting has been acknowledged by the Intergovernmental Panel on Climate Change (IPCC) in the form of new guidance on their use for national GHG inventories.

A common denominator of all the maps published to date, possibly with the exception of some maps produced by the GEDI mission, is that they lack the metadata required to estimate the uncertainty of map-based estimates of biomass or carbon stock for arbitrary geographical regions. As a consequence, estimates obtained directly from these maps fail to comply with the IPCC good practice guidelines, a serious shortcoming.

Biomass maps may be used in their existing forms to enhance estimates for any region of interest only if there already exists a ground-based inventory, such as a national forest inventory program, that has adopted a probabilistic field sampling design. In this case, the value of the maps as auxiliary data can be realized in the form of increased precision of the estimates, which can be substantial. However, many countries finding global biomass maps attractive in their efforts to estimate biomass, do not have extensive ground sampling programs.

In this study, being part of the ESA “Sentinel-1 for Science Amazonas” project, we demonstrate our inability to profit from an existing global biomass map – the ESA CCI 2017 biomass map – when estimating carbon stock across a 4.1 million square kilometer study region in the Brazilian Amazon Biome. To validate the map, we independently estimated mean carbon stock per unit area using an existing random sample of 501 airborne lidar transects with an approximate length of 12 km each collected by the Brazilian EBA project (“Estimativa de biomassa na Amazôni”), a coincident sample of 224 field plots, and hybrid estimators. A standard error and a confidence interval were also estimated for the mean estimate. Following the terminology of the Global Forest Observation Initiative’s “Methods and Guidance Document”, we considered this a “greater quality” estimate and, therefore, useful for validation purposes. We compared the lidar-based estimate to a mean estimate obtained from the CCI 2017 biomass map via “pixel counting”, i.e., estimating the mean biomass across all map units and applying a conversion factor to obtain a mean carbon stock estimate. Despite a substantial difference between the lidar-based and the map-based estimates, we were unable to conclude that this difference was statistically significant because the map-based estimate is just a point estimate without any accompanying information on the uncertainty.

Further, we discuss the nature of the metadata map producers must deliver with their map products to facilitate rigorous statistical inference based solely on the map products. We also illustrate the limitations of the uncertainty information that some of the current maps provide and argue that this information is insufficient to such a degree that uncertainty will be greatly underestimated.

National forest inventories (NFI) provide forest-related biomass and carbon information for country forest monitoring and greenhouse gas (GHG) accounting systems. While many tropical countries are actively working on their NFIs, many still struggle to establish a complete national inventory or to guarantee frequent updates of their NFI owing to budget constraints, inaccessibility of areas, institutional circumstances, developing capacities, internal conflicts, lack of initial information on forest resources and the like. This has limited the usefulness of their national aboveground biomass (AGB) estimates and ultimately hinders the accuracy, completeness and consistency of reporting country greenhouse gas (GHG) emissions and removals under the international climate frameworks as well as the formulation of domestic mitigation targets.

At the same time there has been significant progress in developing large-area biomass density maps to characterize the distribution of forest carbon stocks around the globe. These efforts are further developed through dedicated space-based biomass missions that increasingly provide open access and continuous coarse-scale biomass information. The extent to which global information on space-based estimates of aboveground biomass (AGB) can support national forest biomass monitoring and GHG accounting is still under investigation. In the presentation we cover the approaches and assess whether the use of a global biomass map as a source of auxiliary information can produce a gain in precision of sub-national AGB estimates. For that purpose, we made use of model-assisted estimators that best accommodated the country NFI sampling design, and we explored hybrid inferential techniques to account for additional sources of uncertainty associated with the integration of the remote sensing products (in our case the ESA CCI biomass map) and the NFI plot data. We will present results from two case studies with rather different NFI designs and forest characteristics: Peru and Tanzania.

For the case of Peruvian Amazonia, we found that the CCI biomass map tends to overestimate AGB across the Peruvian Amazonia. The most striking result was that, after calibrating the map using NFI data, it substantially increased the precision of our model-assisted stratum-wise AGB estimates by as much as 150% at the strata-level, and 90% at the Amazon level. even though the initial relationship between the plot-based biomass and space-based estimates of biomass varied for different strata and was not that strong for some strata. Our results show that more precise AGB estimates can be achieved by integrating NFI and space-based biomass data (i.e. through a locally calibrated remote sensing product) in the tropics. In the hybrid inferential analysis, we found that the very different spatial support for the NFI plots compared to the remote sensing-based units contributed 86% of the total uncertainty in the map-NFI integration. The uncertainties concerning the reference data at plot level owing to measurement error and allometric model prediction uncertainty contributed 13%. When accounting for these uncertainties, the precision of our AGB estimates was still increased by as much as 90% and 60% at the stratum and Amazon levels, respectively. Our results show that Peru could benefit from the application of global biomass maps that contribute to more precise and complete AGB estimates for GHG monitoring and reporting, particularly while the NFIs is still incomplete.

The analysis for the case of Tanzania follows a similar framework but the estimation procedures have been modified because of the different NFI plot and sampling designs. The quantitative results are not final at the time of abstract submission but will be included in the presentation to compare both country conditions.

Terrestrial CO2 fluxes from land use, land use change, and forestry (LULUCF) accounted for about 12% of total anthropogenic CO2 emissions in the last 20 years, while land simultaneously provided a natural sink for about 29% of all CO2 emissions (Friedlingstein et al., 2021). Comparisons of anthropogenic LULUCF emissions in global models and in country reports to the United Nation Framework Convention on Climate Change (UNFCCC) revealed a substantial gap between both estimates, globally amounting to about 5.5 Gt CO2/year (Grassi et al., 2021). This gap was mainly attributed to discrepancies in areas considered as managed land in models and country reports, and to the (partial) inclusion of natural CO2 fluxes (e.g. from carbon removals by CO2 fertilization, forest fires, insect outbreaks) on managed land in many of the country reports (Grassi et al., 2021).
Building on this proposed explanation, we provide a disaggregation of country-reported CO2 emissions from LULUCF into contributions from anthropogenic and natural CO2 fluxes on country level, considering eight countries with high emissions from LULUCF. We focus on natural fluxes in managed forests since the majority of natural CO2 removals occurs in forested areas. For each country we use process-based models to estimate the natural CO2 fluxes on managed forests (which we identify through a mask of non-intact forest due to lack of information on the spatial distribution of managed forests in the country reports) and add them to the model-based LULUCF emission estimates. This approach is in line with the methodologies used by almost all investigated countries to estimate CO2 fluxes from LULUCF, which imply that natural fluxes on managed land are included in their CO2 flux estimates.
In the majority of the eight countries investigated, including natural fluxes on managed forests substantially reduces the gap between model estimates and country reports of LULUCF emissions, highlighting that the methodology suggested by Grassi et al. (2021) provides a feasible approach to make estimates more compatible also at the country level. Countries include about half of the domestic natural CO2 fluxes in their LULUCF emission reports, which shifts the reported CO2 fluxes downward, i.e., towards lower emissions or, accordingly, towards larger sinks. Large gaps present in Russia and the USA can be closed almost completely by adding natural fluxes on managed forests to model-based LULUCF emissions, revealing that the CO2 sinks from LULUCF reported by these countries are predominantly due to natural fluxes on managed forests. Also in the EU, Indonesia and China, which is the country that has the largest gap and reports the largest CO2 sink of the investigated countries, the gap is considerably reduced by including natural CO2 fluxes on managed forests. These results highlight that the methodological discrepancies between country reports and model estimates of LULUCF emissions are primarily due to accounting definitions and need to be reconsidered in a proper assessment of the country contributions to the global climate mitigation targets, as planned in the Global Stocktake in 2023.
While the presented approach provides an important step forward in bringing together model estimates and country reports, it does not achieve a complete closing of the gap. For some countries, estimates from models and country reports still differ substantially, such as for China, where differences might be due to overoptimistic estimates of the actual effects of afforestation on CO2 fluxes in the country report, underestimations of the afforested area in the input datasets used by models to calculate LULULCF fluxes, and/or limitations in the capability of models to fully integrate the large-scale afforestation in China. There are several potential reasons for the remaining gaps, including incomplete reporting by countries, uncertainties in historical land use dynamics, and model limitations. Moreover, most countries report the areas considered as managed without explicit information on their location, which prevents a precise spatial identification necessary for correctly quantifying natural fluxes on managed forests in models. For some of these factors, remote-sensing products might provide independent and spatially explicit estimates through satellite-derived classifications of land use and land cover change, quantifications of changes in biomass, or identification of managed forest areas. Additionally, the near real-time availability of satellite data might be useful for providing a temporal extension of country reports, which are usually published with a lag of several years. Remote-sensing products might thus constitute an additional, strong pillar in establishing a sound and viable methodology for a translation between model estimates and country reports of anthropogenic CO2 emissions from LULUCF.


References:

Friedlingstein, P., Jones, M. W., O'Sullivan, M., Andrew, R. M., Bakker, D. C. E., Hauck, J., Le Quéré, C., Peters, G. P., Peters, W., Pongratz, J., Sitch, S., Canadell, J. G., Ciais, P., Jackson, R. B., Alin, S. R., Anthoni, P., Bates, N. R., Becker, M., Bellouin, N., . . . Zeng, J. (2021). Global Carbon Budget 2021. Earth Syst. Sci. Data Discuss., 2021, 1-191. https://doi.org/10.5194/essd-2021-386

Grassi, G., Stehfest, E., Rogelj, J., van Vuuren, D., Cescatti, A., House, J., Nabuurs, G.-J., Rossi, S., Alkama, R., Viñas, R. A., Calvin, K., Ceccherini, G., Federici, S., Fujimori, S., Gusti, M., Hasegawa, T., Havlik, P., Humpenöder, F., Korosuo, A., . . . Popp, A. (2021). Critical adjustment of land mitigation pathways for assessing countries’ climate progress. Nature Climate Change, 11(5), 425-434. https://doi.org/10.1038/s41558-021-01033-6

The need for accurate information to characterize the dynamics of forest cover at the tropical scale is widely recognized particularly to assess carbon losses from deforestation and forest degradation (Achard & House 2015). In particular, the contribution of the degradation is a key element for REDD+ activities and is still missing in most national reports due to lack of reliable information at such scale. The main scientific gaps in the estimation of carbon losses at the pantropical scale from Earth Observation data remain on (i) integrating temporal dynamics of various types of forest disturbances; (ii) assessing and combining uncertainties from ‘activity data’ (deforestation and degradation) and from emission factors (in particular in relation to biomass maps and degradation processes).

To address these gaps, we use a wall to wall tropical moist forest change product (Vancutsem et al. 2021, hereafter called TMF) at 30 m resolution, which depicts both deforestation and degradation over the past 3 decades and allow a better understanding of the interlinkage between the two processes. We combine the TMF annual changes with a pan-tropical map of aboveground biomass density (AGB) at 30 m resolution valid for year 2000 (Harris et al. 2021) to quantify the annual losses in above-ground carbon stock associated with degradation and deforestation in tropical moist forests over the period 2001-2020. The carbon loss due to direct deforestation is accounted as full carbon loss while for degradation we consider partial loss of the initial carbon stocks with a range from 20 to 75% depending on the intensity of degradation processes (Andrade et al. 2017). We also account for induced carbon losses from forest fragmentation and edge effects as an indirect consequence of deforestation (Silva Junior et al., 2020) by considering that forest areas in a 120m edge lose 50% of their biomass linearly over a 50 years period (Brinck et al. 2017). Carbon losses due to deforestation happening after a prior degradation event or edge effect are accounted as the carbon stock remaining after degradation or fragmentation effect. Finally we assess the sensitivity of our approach by replacing the AGB map of year 2000 from Harris et al (2021) with the ESA CCI Biomass map for year 2010 at 100m resolution (Santoro et al. 2021).

This approach allows to produce estimates of annual loss in above-ground carbon stock associated with deforestation and degradation in tropical moist forests. Our initial results show that deforestation and forest degradation led to losses of 580 TgC / year and 365 TgC / year respectively for the period 2011-2020 when using the Harris et al. map, or 395 TgC / year and 267 TgC / year respectively when using the ESA CCI map. These estimates show a lower contribution of degradation to the total carbon loss than recently reported from a coarser resolution study (Baccini et al 2017), i.e. 38% (Harris et al.) or 44% (ESA CCI) from our study versus 69% for period 2003-2014 from this previous study. Further analysis will be performed to better assess the sensitivity of the estimates to the AGB maps and scenarios used for the degradation and fragmentation processes.

We intend to quantify the uncertainties of area estimates from the TMF product following GFOI guidelines (GFOI, 2020). To carry out this accuracy assessment, a stratified random sampling scheme is used to create a reference dataset of 6000 sample plots (with Landsat pixel size). The sample is optimized to target omission errors of disturbances in stable forest areas (by using a buffer zone around changed strata) and commission errors in disturbed forest areas. For each sample plot, the most recent high resolution image that is available from Google Earth platform, or from Planet and Sentinel 2 databases is visually interpreted as well as the time series of available Landsat images from 1990 to 2020. The reference sample will be used to produce unbiased estimates of activity data with uncertainty range and related estimates of Carbon losses through the combination of the sample with biomass maps.



References

Achard F, House JI (2015) Reporting carbon losses from tropical deforestation with Pantropical biomass maps. Environ. Res. Lett. 10, 101002

de Andrade RB et al. (2017) Scenarios in tropical forest degradation: carbon stock trajectories for REDD+. Carbon Balance Manage 12, 6.

Baccini A et al. (2017) Tropical forests are a net carbon source based on aboveground measurements of gain and loss. Science 358, 230–234

Brinck K et al. (2017) High resolution analysis of tropical forest fragmentation and its impact on the global carbon cycle. Nat Commun 8, 14855.

Hansen MC et al (2013) High-resolution global maps of 21st-century forest cover change. Science 342, 850–853

GFOI (2020) Methods and Guidance from the Global Forest Observations Initiative Edition 3.0.

Harris NL et al (2021) Global maps of twenty-first century forest carbon fluxes. Nature Climate Change 11, 234–40

Santoro, M.; Cartus, O. (2021): ESA Biomass Climate Change Initiative (Biomass_cci): Global datasets of forest above-ground biomass for the years 2010, 2017 and 2018, v2. Centre for Environmental Data Analysis, 17 March 2021. doi:10.5285/84403d09cef3485883158f4df2989b0c.

Silva Junior CH et al (2020) Persistent collapse of biomass in Amazonian forest edges following deforestation leads to unaccounted carbon losses. Science Advances.

Vancutsem C et al (2021) Long-term (1990–2019) monitoring of forest cover changes in the humid tropics. Science Advances.