African rangelands contribute to the livelihoods of hundreds of millions of pastoralists over large parts of Africa, providing livestock-based food, essential proteins (meat, milk) and income from livestock sales. The ESA Rangeland Monitoring for Africa Using Earth Observation – Continental Demonstrator (RAMONA) project aims to map status and dynamics of herbaceous vegetation across African rangelands at high spatial resolution (10) m. It will serve the purpose of generating user-ready products supporting management and planning of grazing resources across the continent. It is based on integrated analysis of high temporal and spatial resolution data from the Sentinel satellites 1-3. Net primary productivity is estimated using light-use efficiency models driven by satellite observations and climate data, parameterized using eddy covariance data from flux towers, and parameters on respiration from state of the art dynamic vegetation models. Phenology of vegetation will be estimated using TIMESAT, and novel near-real time methodology for estimating start-of-season will be developed to enable rapid information to livestock herders. Product validation is based on biomass sampling in several test areas. The data will contribute to an improved understanding of carbon dynamics and resource availability in Africa, where both these task include challenges. Challenges of proper calibration of methods estimating biomass productivity and phenology due to the extensive and diverse rangeland ecosystems of Africa, and challenges due to the inherent lack of systematic and standardized collection of in-situ data for calibration and validation. RAMONA is funded by ESA and carried out in collaboration between the universities at Aarhus (DK) and in Lund (SE) together with DHI (DK) and GeoVille (AT).

Terrestrial landscapes are increasingly heterogeneous due to land use and land use change, often with a mosaic of ecosystems with varied carbon stocks, plant traits, dominant processes, and management. Incorporating this complex spatial variation is a major challenge for diagnosing carbon dynamics and understanding landscape resilience to climate change. Earth observation can provide vital insights into the state and dynamics of the land surface but does not directly provide C cycle diagnostics. Process models can represent the full C cycle and allow for exploration of counter-factuals and decision support towards sustainable landscape management. Combining the two sources of information – EO and process models - provides a route forward. This approach is termed Model Data Fusion (MDF). Model data fusion (MDF) approaches have demonstrated their capacity to address C diagnostic challenges by combining the strengths of earth observation (EO) and process modelling. There are in-depth tests of MDF at individual sites with detailed in situ data that provide proof of concept. There are other tests across global domains generating aggregated analyses at coarser grid resolution, demonstrating capacity to operate at scale. In MDF, diagnostic rigor depends on using data at appropriate scales fused with process models to ensure a valid calibration and initialisation of the model, and on providing MDF outputs with clear characterisation of error. The challenge this talk addresses is in applying model-data fusion approaches at fine resolutions (e.g ha to km2) across regional domains (e.g. nation). Here in situ data are sparse and so data constraint is reliant on EO. We identify two challenges for regional analyses: (i) can EO data can provide robust information on management factors?; and (ii) what are appropriate scales for analysis and which are the scale-variant processes that must be managed? A key uncertainty in complex landscapes is diagnosing the role of management factors, including decision making and activity of farmers. For instance, is it possible to use MDF to infer the effect of variations in livestock numbers, and grass harvesting, on C budgets across pastoral landscapes in northern Europe? Another challenge area is how to manage the trade-offs in computation and resolution. High resolution analyses can be limited by computational capacity, so what is the appropriate aggregation approach to simplify landscapes yet maintain their key variability for diagnostics?  A key science question here is to determine the scale variant and invariant processes in the C cycle and manage the variant processes appropriately. We address the first challenge by assessing the potential of MDF to provide robust analyses of C dynamics in managed grasslands across Great Britain (GB). We combine EO data and biogeochemical modelling by implementing a probabilistic MDF algorithm to (1) assimilate leaf area index (LAI) times series (Sentinel-2), (2) infer defoliation instances (grazing, cutting) and (3) simulate livestock grazing, grass cutting, and C allocation and C exchanges with the atmosphere. The algorithm uses the inferred information on grazing and cutting to drive the model's C removals-and-returns module, which partitions grazed biomass into returns to the soil as manure, and losses through livestock respiration and growth.  The MDF algorithm was applied for 2017-2018 to generate probabilistic estimates of C pools and fluxes at 1855 fields sampled from across GB. The algorithm was able to effectively assimilate the Sentinel-2 based LAI time-series (overlap=80%, RMSE=1.1, bias=0.35) and predict livestock unit (LU) densities per area that correspond with independent agricultural census-based data (r=0.68, RMSE=0.45 LU/ha, bias=-0.06 LU/ha). The mean total removed biomass across all simulated fields was 6 (+/- 1.8) t/ha/yr. The simulated grassland ecosystems were mostly C sinks in 2017 and 2018; the net biome exchange (NBE) was -191 (+/-81) (2017) and -49 (+/-69) gC m-2 y-1 (2018). Our results show that the 2018 European summer drought reduced the strength of C sinks in GB grasslands and led to a 9-fold increase in the number fields that were annual C sources (NBE>0) in 2018 (18% of fields) compared to 2017 (2% of fields). The field-scale analysis showed that management in the form of timing, intensity and type of defoliation were key determinants of the C balance of managed grasslands, with cut fields acting as weaker C sinks compared to grazed fields. Nevertheless, extreme weather, such as prolonged droughts, can convert grassland C sinks to sources. We conclude that there is significant potential to use high resolution EO data within MDF to identify management activities and their impacts on field C balance. We directly address the second challenge by quantifying the sensitivity of simulated carbon fluxes in a mixed-use landscape in the UK to the spatial resolution of the model analysis. We test two different approaches for combining EO data into the CARDAMOM Model-Data Fusion (MDF) framework, assimilating time series of satellite-based Earth Observation (EO) derived estimates of ecosystem leaf area and biomass stocks to constrain estimates of model parameters and their uncertainty for an intermediate complexity model of the terrestrial C cycle. In the first approach, ecosystems are calibrated and simulated at pixel-level, representing a "community average" of the encompassed land cover and management. This represents our baseline approach. In the second, we stratify each pixel based on land-cover (e.g. coniferous forest, arable/pasture etc.), and calibrate the model independently using EO data specific to each stratum. We test the scale-dependence of these approaches for grid resolutions spanning 1o to 0.05 o over a mixed land-use region of the UK. Our analyses indicate that spatial resolution matters for MDF. Under the "community-average" baseline approach biological C fluxes (primary production, respiration) diagnosed by CARDAMOM are insensitive to resolution. However, disturbance fluxes exhibit scale-variance that increases with greater landscape fragmentation, and for coarser model domains. In contrast, stratification of assimilated data based on fine-resolution land-use distributions resolved the resolution dependence, leading to disturbance fluxes that were approximately double the baseline experiments. The differences in simulated disturbance fluxes were sufficient to drive alternative interpretations of the terrestrial C balance: in the baseline experiment the live C pools suggest a strong C sink, whereas in the stratified experiment, the live C pools were approximately in steady-state as the C gains from NPP were balanced by losses due to the higher simulated harvest fluxes focused in conifer woodlands. We also find that stratifying the model domain based on land-use leads to differences in the retrieved parameters that reflect variations in ecosystem function between neighbouring areas of contrasting land-use. The emergent differences in model parameters between land-use strata give rise to divergent responses to future climate change. Accounting for fine-scale structure in heterogeneous landscapes (e.g. stratification) is therefore vital for ensuring the ecological fidelity of large-scale MDF frameworks. The need for stratification arises because land-use places strong controls on the spatial distribution of carbon stocks and plant functional traits, and on the ecological processes controlling the fluxes of C through landscapes, particularly those related to management and disturbance. Given the importance of disturbance to global terrestrial C fluxes, together with the widespread increase in fragmentation of forest landscapes, these results carry broader significance for the application of MDF frameworks to constrain the terrestrial C-balance at regional and national scales.

talk

talk

talk

talk

talk

talk

The high latitude regions are expected to be largely impacted by climate change, as northern ecosystems are sensitive, and the changes in climate will be pronounced. The resulting impacts on the carbon cycle need to be studied throughout the Arctic region to predict changes in carbon sequestration and storage pools. The measurement network in the high latitudes is sparse; therefore, the data from remote sensing observations can be of great help in the carbon cycle studies. We have investigated the carbon cycle at Sodankylä Scots pine forest (67.4 °N, 26.6 °E), located 100 km north from the Arctic circle, with the help of a terrestrial biosphere model and in-situ and remote sensing observations. The site has a 20-years time series of micrometeorological CO2 and energy flux data. We collected in-situ observations including snow depth and tower-based sun-induced chlorophyll fluorescence (SIF), and remote sensing estimates of foliar chlorophyll content per unit leaf area (Cab), TROPOSIF and cryosphere-related variables, such as soil thaw and snow melt. Carbon, nitrogen, energy, and water cycles were simulated with a novel terrestrial biosphere model, QUantifying Interactions between terrestrial Nutrient CYcles and the climate system (QUINCY). We aim at 1) understanding how the cryospheric dynamics influence the carbon and nitrogen cycling in spring and summer, 2) evaluating the SIF implementation in the QUINCY model and 3) evaluating the use of Cab to constraint the nitrogen cycle modelling at site level. Preliminary results show that QUINCY is able to capture the seasonality of SIF. The comparison between remotely-sensed and simulated Cab (based also on a larger dataset) raises the need for modifying the model representation of Cab, that is at the moment not showing enough variability between different sites. The different data streams help to improve the model and our future predictions for boreal forests at high latitudes.

talk

Observations of atmospheric column average dry air mole fractions of CO2 (XCO2) depend on accurate knowledge of the number density of dry air in the atmospheric column, which requires accurate knowledge of surface pressure. A robust understanding of surface elevations allows us to assess and ensure the quality of XCO2 measurements from space. For XCO2 retrievals from the NASA OCO-2 satellite, prior estimates of surface pressure are used that are adjusted to the average elevation in the satellite footprint based on a Digital Elevation Model (DEM). A global empirical bias correction is applied to the retrieved XCO2, after the fact, that has the effect of moving the density of dry air in the column closer to that implied by the prior surface pressure and away from that implied by the retrieved surface pressure. As a result, an accurate DEM is essential for defining an appropriate bias correction for OCO-2 retrievals. In this analysis, we specifically focus on the northern high latitude regions because, until recently, DEMs for these regions have been based on extremely limited data and little was known about their accuracy. Polar orbiting satellites such as OCO-2 have great potential for providing an abundance of observations over northern high latitude regions if issues of retrieval quality can be addressed. Different DEMs for the northern high latitude regions (north of ~60°N) can differ by as much as 50 m, corresponding to differences in surface pressure of ~5-6 hPa and differences in retrieved, bias-corrected XCO2 up to 4 ppm. We will present an evaluation of the relative merits of DEMs used in OCO-2 ACOS B10 and B11 retrievals, as well as results from the Arctic DEM at 32 m resolution and the Copernicus 30 m DEM. We will then explore the impact of using different DEMs on OCO-2 retrieved XCO2 in northern high latitude regions.

The Arctic Observing Mission (AOM) is a satellite mission concept that would use a highly elliptical orbit (HEO) to enable dense and frequent observations of greenhouse gases (GHGs), air quality, meteorological variables and space weather north of the usable viewing range of geostationary (GEO) satellites. AOM is envisioned as a Canadian-led mission to be implemented with international partners. AOM is currently undergoing a pre-formulation study to refine options for the mission architecture and advance other technical and design aspects, investigate socio-economic benefits of the mission and better establish the roles and contributions of prospective partners. AOM would use an Imaging Fourier Transform Spectrometer (IFTS) with 4 near-infrared/shortwave infrared (NIR/SWIR) bands to observe hourly CO2, CH4, CO and Solar Induced Fluorescence (SIF) spanning cloud-free land from ~40-80°N during daylight. The rapid revisit is only possible due to cloud avoidance using ‘intelligent pointing’, which is facilitated by the availability of real-time cloud data from the meteorological imager and the IFTS scanning approach. Simulations suggest that these observations would improve our ability to detect and monitor changes in the Arctic and boreal carbon cycle, including CO2 and CH4 emissions from permafrost thaw, or changes to northern vegetation carbon fluxes under a changing climate. This presentation will give an overview of AOM with a focus on the GHG instrument, its expected capabilities and its potential for carbon cycle science and monitoring.

talk

talk

talk

talk

talk

talk

talk

talk

talk

talk

talk

Introduction Forests have increasingly multifunctional roles. The increased prominence of forests in climate change mitigation and bioeconomy strategies, and the rising importance of sustainability in forest policies brings new challenges for forest management. At the same time, forests are facing environmental challenges due to changing climatic conditions (e.g. storms, drought, pest attacks) and there is a growing demand for both wood and non-wood forest products. Furthermore, forestry stakeholders have numerous legal and voluntary obligations (e.g. certification schemes) to monitor and report the usage and status of forests. In this situation, acquiring up-to-date spatially explicit information on forest characteristics is more crucial than ever. Up-to-date information on the current status of forests is not only needed for regulative monitoring and operational planning purposes, but it forms a basis for future management planning and policy formulation. Modelling of future forest development in different management and climate scenarios support decision making from small scale forest management level to large scale forest policies. Most countries have National Forest Inventory (NFI), and other wide scale field measurement campaigns that provide information on the status of the forests to support management planning and policy decisions. However, field plot measurements alone do not provide sufficient wall-to-wall estimates of forest characteristics in fine spatial detail. For this reason, remotely sensed datasets are increasingly used in large forest inventories (Kangas et al. 2018). Earth Observation based approaches combining field data with satellite data can be used to estimate forest characteristics in a spatially explicit manner for large areas. While < 5 m spatial resolution Earth Observation data is generally not free of charge, there is an increasing variety of freely available 10-30 m spatial resolution datasets suitable for forest monitoring purposes. These include optical datasets (such as Sentinel 2 and Landsat 8) and radar datasets (such as the C- band Sentinel 1). These data allow estimation of forest structural variables (like tree species, height, stem volume) and site fertility that can be used to monitor the state of the forests for large areas. While both optical and radar data have been found suitable for structural forest variable estimation, optical data has been more used for the purpose. Particularly the European Space Agency Sentinel 2 satellite, with its 10 wavelength bands in 10-20 m spatial resolution and high temporal frequency (5-days repeat time at the equator, 2-3 days in mid latitudes) provides optimal data for large scale forest monitoring. One of the advantages of using EO is that data acquisition is repeated over time and repeated satellite-based estimates could be integrated together to increase the accuracy of the estimates. However, forests are dynamically changing over time and forest models need to be used to integrate repeated measurements of forest status. Process based forest ecosystem models (Fontes et al. 2011; Minunno et al. 2019), provide optimal means for forecasting forest development under different management options as well as changing climatic conditions. Modern computational techniques such as data assimilation (DA) allow to combine model predictions and data from multiple sources, considering also the associated error. In the case there are repeated remotely sensed measurements available for a forest area, it is possible to initialize the model with the first data acquisition and forecast the state of the forest for the date of the second data acquisition. The two sources of information are combined considering their relative uncertainty. The new estimate will constitute the starting point for a new data assimilation and the process can be repeated every time new data become available. The main objective of this study is to develop two frameworks that allow us to assimilate repeated measurements of high resolution (10-30 m) remotely sensed data into a simple process-based forest model. A framework will be used to estimate site fertility that is site specific and remains relatively stable over time; another framework will be used to update forest structural variables that dynamically change over time. Materials and methods Satellite data Sentinel-2 Level 2A (L2A) satellite data (together with field measurements) was used to estimate forest variables that form a key input into a process based ecosystem model. Three Sentinel-2 tiling grid tiles ‘Vaasa’, ‘Tampere’ and ‘Oulu’ were used in the study, providing a north-south spread through the boreal zone. The sites were selected taking into account 1) the availability of field data and 2) the availability of early (2016) Sentinel-2 data. For each of the three tiles, Sentinel-2 L2A composite images were created for 2016 and 2019 using imagery acquired between June and September. Each observation (a pixel in a given image) was evaluated according to four criteria: cloudiness, haze, shadows, and resemblance to usual pixels observed. Based on the evaluation, a weight was given for each observation and these weights were used to average the observations to produce the final image. Field data Field plot measurements were used as 1) training data for the forest variable model creation and 2) reference data for accuracy assessment. Altogether, 5054 field plots were used, 2508 for the creation and evaluation of the 2016 model, and 2546 for the 2019 model. The field data plots had been measured by the Finnish Forest Centre (https://www.metsakeskus.fi/node/321). Eight variables were extracted from the field plot database and used in the forest structural variable model creation: Basal area (B), Diameter at breast height (D), Height (H), Volume (V), Pine proportion (P%), Spruce proportion (Sp%), Broadleaf proportion (Bl%) and fertility class (Site). PREBAS and model emulator. The forest model used in this study is PREBAS (Minunno et al., 2019), a combination of a light use efficiency model, PRELES (Mäkelä et al., 2008; Peltoniemi et al., 2016) and a forest growth model (CROBAS, Mäkelä, 1997; Valentine & Mäkelä, 2005) based on the pipe model theory. PREBAS predicts carbon and water balances and the growth of forests and it has been calibrated for the pan-boreal region (Minunno et al., 2016; Minunno et al., 2019). In this study, PREBAS was initialized using the structural forest variables estimated with the Probability method using Sentinel 2 and field data. The forest structure inputs included basal area (B), diameter at breast height (D), height (H), pine proportion (%P), spruce proportion (%P), deciduous species proportion (%D) and fertility class. In addition to PREBAS, we used model emulators (modEm) that mimicked PREBAS outputs reducing the computational load. The emulators were regression models fitted to simulate PREBAS outputs (B, D, H, %P, %S,%D) at the time of the second satellite measurement (2019, t2) using as independent variables the structural forest variables and fertility class at the first satellite measurement (2016, t1). Data assimilation frameworks We developed two data assimilation frameworks that, through the implementation of a few steps, allow to combine repeated forest structure estimates from Sentinel 2 data and forest model predictions. One framework was used to update the forest structural variables (state variables: B, D, H, %P, %S and %D) (ForVarDA); while the other framework was used to update the estimates of site fertility class (STDA). The Bayesian approach was used for data assimilation. Bayesian method relies on probability theory and allows to account for the uncertainties in measurements and model structure. The Bayesian theorem allows updating the conditional probability of an event (posterior distribution) combining prior information with new data encoded in the likelihood. Since the state of forests evolves over time, we used a sequential Bayesian filtering, i.e. unknown distributions (state of the forests) are recursively estimated overtime using incoming measurements and models. We used the forest model PREBAS and its emulators to project the estimates from t1 (2016) to t2 (2019). The model prediction of the forest state at t2 was the prior, while the s2 based estimates for 2019 represented the new data (likelihood). The prior and the likelihood were approximated by multivariate normal distributions and combined using the Kalman filter, i.e. the posterior estimates for 2019 were the weighted averages of the prior and the likelihood and their uncertainties represented the weights.    Results & Discussion The data assimilation frameworks, developed in this work, allowed obtaining updated spatially explicit estimates of forest structural variables, site class, carbon balance and forest growth with their associated uncertainty. The new maps enclosed all the available information acquired by repeated earth observations (2016 and 2019) and model simulations. Forest variable estimates achieved through Sentinel 2 data, model forecasts and data assimilation showed consistent patterns across the tiles. Furthermore data assimilation significantly reduced the uncertainty of all forest structural variables and species coverage. The use of process-based model in DA is somehow hampered by the load of the calculations. In this work, by means of PREBAS emulators, we were able to overcome the computational challenges (Dietze et al., 2013; Fer et al., 2018) of DA and we assimilated data over large areas at high resolution. The reliability of emulators in mimicking PREBAS outputs is conditional to the time step of DA and the ecosystem considered. For boreal forests, a DA time step of 3-5 years is suitable; in more productive and fast growing ecosystems, such as tropical forests, where complex processes occur at higher temporal resolution, the emulator performances should be tested and, eventually, a shorter DA timestep should be considered. With the increasing availability of free of charge high (10-20 m) resolution satellite datasets, annual or biannual monitoring frequency can be considered feasible even in areas with high cloud coverage. Combining data assimilation with field measurements would be always desirable because it would allow a continuous validation of the data-model integrated system and a better quantification of uncertainties. In addition, combined use of field measurements and remotely sensed datasets allow wall-to-wall analysis of the distribution of results. Forest state is dynamically changing over time therefore model predictions need to be used to integrate repeated measurements of forest status. Fusing information from both 2016 and 2019 under a Bayesian framework makes our estimates more consistent and dependable. Satellite measurements and model predictions proved to be consistent. Furthermore, DA has the great advantage of helping to reduce and identify bias errors that could characterize both the initialization data and model predictions. Potential biases in the initialization propagate through model forecasts, however, by means of data assimilation, the biases in the data can be easily identified and their impact becomes marginal once the DA process covers multiple data acquisitions. Our results show how data assimilation can increase the accuracy of forest monitoring. In fact, the impact of biased estimates of the last data acquisition (s2019) for D and H of the Oulu tile was limited in the data assimilation framework because the weight of more accurate data prevailed. In comparison with satellite-based estimates in 2019, data assimilation results showed obvious lower predictive uncertainty, as the probability density distribution was shifting to a high kurtosis. The uncertainty was reduced because the product of two multivariate normal PDFs is proportional to the PDF of another multivariate distribution of smaller variance. Data assimilation gives the opportunity to make use of all the available information to update the state of forests. Data integration can be repeated systematically over time, providing a useful tool for forest monitoring and management planning (Saad et al. 2017). In this study we used satellite-based estimates at high resolution; in the framework developed here, multiple data sources can be integrated at different spatial (from landscape to country scale) and temporal resolutions. For instance, national forest inventories, eddy-covariance fluxes, lidar and UAV data are some of the sources of information that could also be included in the data assimilation process. The use of process-based models gives the possibility to integrate information about the physiological status of the forests integrating data on drought stress, pest disturbances (Bastos 2020). The frameworks presented here can be applied to any kind of model and data and have a great potential for future applications.

talk

talk

talk

talk

Understanding carbon sources and sinks across the Earth’s surface is fundamental in climate science and policy; thus, these topics have been extensively studied but have yet to be fully resolved and are associated with massive debate regarding the sign and magnitude of the carbon budget from global to regional scales. Developing new models and estimates based on state-of-the-art algorithms and data constraints can provide valuable knowledge and contribute to a final ensemble model in which various optimal carbon budget estimates are integrated, such as the annual Global Carbon Budget paper. Here, we develop a new atmospheric inversion system based on the four-dimensional local ensemble transform Kalman filter (4D-LETKF) coupled with the GEOS-Chem global transport model to infer surface-to-atmosphere net carbon fluxes from Orbiting Carbon Observatory-2 (OCO-2) V10r XCO2 retrievals. The 4D-LETKF algorithm is adapted to an OCO-2-based global carbon inversion system for the first time in this work. On average, the mean annual terrestrial and oceanic fluxes between 2015 and 2020 are estimated as −2.02 GtC yr−1 and −2.34 GtC yr−1, respectively, compensating for 21% and 24%, respectively, of global fossil CO2 emissions (9.80 GtC yr−1). Our inversion results agree with the CO2 atmospheric growth rates reported by the National Oceanic and Atmospheric Administration (NOAA) and reduce the modelled CO2 concentration biases relative to the prior fluxes against surface and aircraft measurements. Our inversion-based carbon fluxes are broadly consistent with those provided by other global atmospheric inversion models, although discrepancies still occur in the land-ocean flux partitioning schemes and seasonal flux amplitudes over boreal and tropical regions, possibly due to the sparse observational constraints of the OCO-2 satellite and the divergent prior fluxes used in different inversion models. Four sensitivity experiments are performed herein to vary the prior fluxes and uncertainties in our inversion system, suggesting that regions that lack OCO-2 coverage are sensitive to the priors, especially over the tropics and high latitudes. In the further development of our inversion system, we will optimize the data-assimilation configuration to fully utilize current observations and increase the spatial and seasonal representativeness of the prior fluxes over regions that lack observations.

This study presents an assessment of the terrestrial carbon budget for Australia continent based on a regional inversion approach using the Orbiting Carbon Observatory-2 (OCO-2) satellite data for 2010-2015 as a contribution to the REgional Carbon Cycle Assessment and Processes (RECCAP2) project. Our inversion approach uses the best available bottom-up flux estimates for Australia, which biosphere fluxes are derived from the Community Land Surface Model (CABLE) in BIOS3 set-up. CABLE BIOS3 relies on regional drivers and observations and offers better and more realistic quantification of the land fluxes than global simulations. OCO-2 assimilated fluxes suggest that Australia's semi-arid ecosystems such as savanna and regions with sparse vegetation can act as stronger carbon sinks (compared to the bottom-up estimates), likely influenced by climate drivers such as rainfall and temperature.

We provide a roadmap to deliver frequently updated estimates of greenhouse gases emissions and removals for each country and key sectors, constrained by Earth Observations, including quantification of uncertainties related to each country's characteristics (e.g. forest cover, importance of land use change, role of methane emissions). By combining observable components of individual CO2, CH4 and N2O fluxes, we show that it is possible to quantify the budget of these three greenhouse gases with a low latency at national and even sub-national scales and from the ocean, including differentiation of natural and anthropogenic fluxes. Developing such near real-time estimates would transform our ability to track changes of emissions separately over unmanaged and unmanaged lands, to assess trends of deforestation emissions and carbon losses from ecosystem disturbances, and will make it possible to identify emissions hotspots for prioritising mitigation actions. Concrete applications for carbon anomalies in 2021 and 2022 in response to emissions variations and natural GHG fluxes will be provided

The Second Regional Carbon Cycle and Processes study (RECCAP2) will inventory greenhouse gases for 10 land and 5 ocean regions, in addition to several additional focal areas. RECCAP2 is a continuation of RECCAP1, extending the time period to cover 2010-2019 and also the greenhouse gases to include carbon dioxide as well as methane and nitrous oxide. Involving over 200 scientists from around the globe, the greenhouse gas budgets are developed using top-down atmospheric inversions and bottom-up process models, inventories or data-driven approaches. The regional budgets aim to provide both scientific insights into processes that regulate greenhouse gas sources and sinks, and also to inform policy by integrating inventory information and being used to develop global syntheses that can be used to interpret trends and dynamics of atmospheric concentrations. A key component of RECCAP2 is the role of lateral transport of carbon dioxide via riverine processes and aquatic burial as well as from the trade of goods. Considering these lateral fluxes is a key component of understanding the differences between top-down and bottom-up methods as well as closing the global budgets. This presentation will cover the motivation for RECCAP2 and main results so far.

Changes in terrestrial ecosystem carbon storage can immediately affect global atmospheric CO2 concentrations by both releasing and removing carbon, with both direction of fluxes having large uncertainties in magnitude and spatial and temporal variability. These uncertainties are mostly due to the complexity of estimating two main additive fluxes: first, the net flux of carbon from land use, land use change and forestry (LULUCF) () that is estimated to be a net source of carbon to the atmosphere, and second, the net flux of carbon driven by environmental changes (e.g. climate) () that is estimated to be a sink. For , emissions are mostly from deforestation and degradation and carbon removals from secondary forest regeneration or afforestation. For , emissions are from natural or indirect human disturbances (e.g. droughts, storms, climate change), and removals from vegetation and soils in primary forests and nonforest ecosystems assumed to be free of human influence. In reality, all terrestrial ecosystems are impacted by CO2 fertilization, N deposition and increasing temperatures.   Distinguishing between land use and environmental effects is important for UNFCCC recommended carbon accounting and mitigation policies, yet difficult in practice.   Using remote sensing observations, advances in AI, and data assimilation models, separating carbon fluxes from land use activities and environmental effects have become a reality. Here, we present a data-model framework based on forest inventory data, satellite observations of forest cover change and carbon storage, and data assimilation models to quantify emissions and removals from deforestation, degradation, and regeneration. We highlight that the complexity of forest carbon stock changes, lagged emissions, attributions, and uncertainty dealt with in most scientific studies, may have confounding and adverse effects on policy applications and investments for carbon credits.   However, focusing only on changes of aboveground live biomass from land use and environmental effects and committed versus lagged emissions will increase the confidence in the MRV system for policy and the carbon market. If we are to expect forests to play a significant role in mitigating climate change, we should develop the path from science to policy to market to be the least uncertain and complex.

We use Optimal Estimation (OE) to quantify methane fluxes based on total column CH4 data from the Greenhouse Gases Observing Satellite (GOSAT) and the GEOS-Chem global chemistry transport model. We then project these fluxes to emissions by sector at 1 degree resolution and then to each country using a new Bayesian algorithm that accounts for prior and posterior uncertainties in the methane emissions. These estimates are intended as a pilot dataset for the Global Stock Take in support of the Paris Agreement. However, differences between the emissions reported here and widely-used bottom-up inventories should be used as a starting point for further research because of potential systematic errors of these satellite based emissions estimates. We find that agricultural and waste emissions are ~263 +/- 24 Tg CH4/yr, anthropogenic fossil emissions are 82 +/- 12 Tg CH4/yr, and natural wetland/aquatic emissions are 180 +/- 10 Tg CH4/yr. These estimates are consistent with previous inversions based on GOSAT data and the GEOS-Chem model. In addition, anthropogenic fossil estimates are consistent with those reported to the United Nations Framework Convention on Climate Change [80.4 Tg CH4/yr for 2019]. Alternative priors can be easily tested with our new Bayesian approach (also known as prior swapping) to determine their impact on posterior emissions estimates. We use this aproach by swapping to priors that include much larger aquatic emissions and fossil emissions (based on isotopic evidence) and find little impact on our posterior fluxes. This indicates that these alternative inventories are inconsistent with our remote-sensing estimates and also that the posteriors reported here are due to the observing and flux inversion system and not uncertainties in the prior inventories. We find that total emissions for approximately 57 countries can be resolved with this observing system based on the degrees-of-freedom for signal metric (DOFS > 1.0) that can be calculated with our Bayesian flux estimation approach. Below DOFS of 0.5, estimates for a countries total emissions are more weighted to our choice of prior inventories. The top five emitting countries (Brazil, China, India, Russia, USA) emit about half of the global anthropogenic budget, similar to our choice of prior emissions but with the posterior emissions shifted towards the agricultural sector and less towards fossil emissions, consistent with our global posterior results. Our results suggest remote sensing based estimates of methane emissions can be substantially different (although within uncertainty) than bottom-up inventories, isotopic evidence, or estimates based on sparse in situ data, indicating a need for further studies reconciling these different approaches for quantifying the methane budget. Higher resolution fluxes calculated from upcoming satellite or aircraft data such as the Tropospheric Monitoring Instrument (TROPOMI) and those in formulation such as the Copernicus CO2M, MethaneSat, or Carbon Mapper can be incorporated in our Bayesian estimation framework for the purpose of reducing uncertainty and improving the spatial resolution and sectoral attribution of subsequent methane emissions estimates.

Extreme fires affect the Earth’s climate through their extensive carbon emissions over a short period and the slow recovery of ecosystems. The tendency for increased extreme fires highlights the challenges we face to understand fire carbon budgets and their interaction with climate due to substantial uncertainties in our current emission monitoring capability. Here, we improved a satellite-based global atmospheric inversion system, based on the latest MOPITT (Measurements of Pollution in the Troposphere) version 9 carbon monoxide retrievals, to reconstruct global fire CO2 release from 2000 to 2021. The inversion results show that CO2 release from boreal fires, which typically account for 10% of global fire emissions, has been increasing by 6.6 ± 4.5 Mt C year−1 since 2000. In 2021, boreal fires contributed 0.48 Gt C (equivalent to 23%) to global fire carbon emissions, by far the highest over the MOPITT satellite record. The increasing fire emission trends and record anomalies are mainly due to the increasing encroachment of fires over boreal forests at northern high latitudes, an upward trend that began a decade ago and occurred concurrently with the emergence of warmer and drier fire seasons. A new indicator of tree cover fraction-weighted climatic water deficit suggests that boreal fire emissions rose sharply when this variable reached critical low negative values. 2021 was an abnormal year because boreal North America and Eurasia experienced synchronous lowest climatic water deficit, which never happened before in the studied period. An increase in climate-fire feedback challenges meeting the internationally agreed climate mitigation targets. Our study calls for monitoring capacity to evaluate boreal fire carbon budgets to deepen our understanding and projection of the impact of fires on the build-up of atmospheric CO2.

Estimates of fire carbon emissions mostly rely on satellite observations of burned area (BA) that are combined with model simulations of fuel loads (biomass) and fuel consumption. Alternative approaches make use of fire radiative energy (FRE) or of observations of atmospheric trace gases to estimate fire emissions. However, current state-of-the-art datasets on fire emissions show large differences in emission estimates, which originate from the used approach, resolution of burned area data, or from the used estimates of fuel consumption, emission factors, or FRE-conversion factors. However, those approaches make little use of information about fire type, or aspects of fire behaviour related to smouldering and flaming combustion. Here, we present an alternative approach to estimate fire emissions that aims to integrate a wealth of satellite observations and derived products by reconciling bottom-up approaches based on BA and FRE and top-down approaches based on observations of atmospheric trace gases. We use observations from Sentinel-1, Sentinel-2, Sentinel-3 OLCI and ESA CCI land cover and biomass products and various in situ databases to quantify vegetation and surface fuel loads and fuel moisture. Observations from Sentinel-2, Sentinel-3 SLSTR, and VIIRS are used to quantify the temporal development of individual fires and to derive BA and FRE. We then use this information to classify different fire types and combine the information with the estimated fuel loads to quantify fuel consumption and fire emissions. The estimated fire emissions of each fire type are then integrated with the Copernicus Atmospheric Monitoring Service (CAMS) forecast model to allow an evaluation against observations of CO, NOx and aerosols from Sentinel-5p TROPOMI. We test and apply our approach for various study regions in the Amazon, southern Africa, central Asia, and Siberia. Our results demonstrate that the retrieved bottom-up emissions are in good agreement with CO concentrations as observed by Sentinel-5p. Those novel estimates of fuel loads and moisture, fire behaviour and fire emissions will further constrain the role of fire dynamics and emissions in the global carbon cycle.

Data from the Advanced Scatterometer (ASCAT) on the Metop series of satellites can provide a new perspective on global vegetation dynamics, and specifically the role of vegetation in the carbon and water cycles. Here we will illustrate the benefits of ASCAT data for monitoring vegetation dynamics and constraining parameters in the terrestrial carbon balance. The theory behind the ASCAT slope will be introduced to demonstrate its direct value as a microwave observable. Results will be presented to illustrate how ASCAT slope dynamics vary over a range of land cover types, revealing vegetation dynamics across temporal scales. Radar backscatter data are sensitive to the dielectric properties of vegetation, as well as the number, distribution and geometry of scatterers in the vegetation layer. The relationship between backscatter and incidence angle is determined by the degree to which total backscatter is due to surface scattering, volume scattering and multiple scattering. Temporal variations in the incidence angle dependence of backscatter therefore provide information on vegetation dynamics, specifically changes in above ground biomass, structure and moisture content. Over deciduous evergreen forests in Amazonia, slope dynamics follow changes in radiation, corresponding to the limiting factor of radiation on vegetation activity for this region. Over the more heterogeneous vegetation of the Cerrado, slope dynamics vary per land cover type and can be explained by the variability in limiting factors to vegetation activity. Over croplands, slope follows seasonality in precipitation and Equivalent Water Thickness from GRACE. For shrubs and forests, slope has its peak during the dry season simultaneously with a peak in radiation. The deeper rooting depth enables them to increase photosynthesis and leaf development slightly before or at the onset of increasing radiation. For the 2010 and 2015 Amazonia droughts, no clear impact of droughts were found over deciduous evergreen forests. In the Cerrado, the response in drought varied between 2010 and 2015 due to the timing of the drought. The 2015 drought occurred during the wet season, and positive anomalies in slope were observed over natural vegetation, as increased radiation allowed for increased photosynthesis. In 2010 the drought occurred during the dry season. Similar to what was found by Liu et al. (2018) in passive microwave VOD, positive anomalies were observed before and during the beginning of the 2010 drought, whereas negative anomalies occurred at the end and after the drought.   An analysis of the ASCAT slope, compared to model simulations and ground data from the Sodankyla ICOS site, has been conducted within the scope of the ESA Land Carbon Constellation project. Results from this boreal forest region in Northern Finland show that slope dynamics are influenced by freezing temperatures and snow, revealing the transition in seasonality as well as phenological changes during the summer months. During the 2018 drought, positive anomalies in slope were found. This is consistent with results found by Bastos et al., (2020), who demonstrated that increased temperature, drier than average conditions and increased radiation led to increased vegetation growth as modelled with several vegetation models and observed with SMOS L-VOD. To benefit terrestrial carbon cycle modelling and science, ASCAT slope can be assimilated directly into land surface models to constrain states and parameters related to the fast and slow water and carbon fluxes. Results from the ESA Land Carbon Constellation project will be presented to demonstrate that the measurement operator required for assimilation can be determined using several approaches, and is analogous to that which would be used for VOD. In addition, it will be shown that a machine learning approach can be highly effective in capturing indirect relationships between model states and the ASCAT slope, e.g. when vegetation water content is not directly modelled. The availability of ASCAT data since 2007, and the guaranteed availability of data from the SCA instrument on Metop-SG means that the ASCAT slope series will provide a uniquely continuous, consistent dataset spanning several decades. This is essential to study terrestrial carbon processes at scales relevant for climate science. Bastos, A., Ciais, P., Friedlingstein, P., Sitch, S., Pongratz, J., Fan, L., Wigneron, J.P., Weber, U., Reichstein, M., Fu, Z., Anthoni, P., Arneth, A., Haverd, V., Jain, A.K., Joetzjer, E., Knauer, J., Lienert, S., Loughran, T., McGuire, P.C., Tian, H., Viovy, N., Zaehle, S., n.d. Direct and seasonal legacy effects of the 2018 heat wave and drought on European ecosystem productivity. Science Advances 6, eaba2724. https://doi.org/10.1126/sciadv.aba2724 Liu, Y.Y., van Dijk, A.I.J.M., Miralles, D.G., McCabe, M.F., Evans, J.P., de Jeu, R.A.M., Gentine, P., Huete, A., Parinussa, R.M., Wang, L., Guan, K., Berry, J., Restrepo-Coupe, N., 2018. Enhanced canopy growth precedes senescence in 2005 and 2010 Amazonian droughts. Remote Sensing of Environment 211, 26–37. https://doi.org/10.1016/j.rse.2018.03.035

talk

talk

talk

Extreme events can result in sudden carbon losses equal to many years of carbon sequestration. It has been suggested that changes in the frequency and intensity of extreme events due to climate change could reduce the terrestrial carbon sink, but such consequences are highly uncertain. A definitive answer requires better quantification of the impacts of extreme events on the carbon cycle. Here, I summarize how recent advances in space-based carbon cycle observations have made it possible to monitor carbon cycle perturbations on a finer scale, leading to a more detailed understanding of the response of ecosystems to extreme events. And discuss how future space-based observations will allow for robust monitoring of carbon cycle perturbations from extreme events.

As Earth system models (ESMs) become increasingly complex, there is a growing need for comprehensive and multi-faceted evaluation, analysis, and diagnosis of model results. Better representation of biogeochemistry–climate feedbacks and ecosystem processes in ESMs is essential for reducing those uncertainties during the remainder of the 21st century and beyond. Model–data comparison and integration activities are required to inform improvement of land carbon cycle models and the design of new measurement campaigns aimed at reducing uncertainties associated with key land surface processes. Space-based observation play a key and growing role in constraining model processes and in providing data for benchmarking model performance. The International Land Model Benchmarking (ILAMB) Package was designed to facilitate systematic and comprehensive model–data comparison and improve understanding of factors influencing model fidelity. We used ILAMB to benchmark and intercompare terrestrial carbon cycle models coupled within ESMs used to conduct historical simulations for the Fifth and Sixth Phases of the Coupled Model Intercomparison Project (CMIP5 and CMIP6). Results indicated that the suite of CMIP6 land models exhibits better performance than the suite of CMIP5 land models in comparison with observations for a variety of biogeochemical, hydrological, and energy-related variables. These improvements are partially attributed to reductions of biases in temperature, precipitation, and incoming radiation, suggesting that free-running atmosphere models in these ESMs also improved; however, biases in some regions increased. An analysis of forcing variables, prognostic land variables, and relationships from variable-to-variable comparisons indicated an overall improvement in most CMIP6 models, with relationships for some models exhibiting the greatest improvement in ILAMB scores, suggesting that improved model process representation in some models, and likely increased model complexity, contributed to improved model performance. We further analyzed the degree to which the range of model uncertainties may have been reduced for CMIP6 land models as compared with CMIP5 land models. These results offer a baseline from which improved benchmarks, informed by space-based observations, may help further constrain land and ocean models in future CMIP activities.

Droughts in a warming climate have become more frequent and extreme thereby moving the study of forest responses to droughts from the margins of eco-hydrology to a pressing societal issue. Changes in vegetation water content (VWC) have been linked to a range of tree responses to drought stress, including fluxes of water and carbon, mortality, flammability, among other examples. VWC can be retrieved from remote sensing measurements, particularly at microwave frequencies using radar and radiometry. We demonstrate that such microwave observations are not just sensitive to biomass changes (as commonly assumed) but also vary with water stress through its influence on relative water content. Key frontiers through which microwave vegetation observations have the potential to enable quantification of forest responses to water stress are highlighted. To validate remote sensing observations of VWC at landscape scale and to relate them to data assimilation model parameters, an ecosystem-scale analog of the pressure–volume curve (PVC) is introduced. In this framework, the PVC is encoded in a non-linear relation between average leaf or branch water potential and water content. The sources of variability in these ecosystem-scale PVC and their relation to forest response to water stress are discussed. To what extent diel, seasonal, and decadal dynamics of VWC reflect variations in different processes relating the tree response to water stress are discussed. VWC can also be used to infer belowground conditions—which are notoriously difficult to observe directly through remote sensing. Opportunities for geostationary or other multi temporal spaceborne observational systems for VWC, which, when combined with existing datasets, can capture diel and seasonal water dynamics are presented in the context of advancing understanding of global forest vulnerability to future droughts.

Carbon cycle feedbacks represent large uncertainties in climate change projections, and the response of soil carbon to climate change contributes the greatest uncertainty to this. Future changes in soil carbon depend on changes in litter and root inputs from plants and especially on reductions in the turnover time of soil carbon (τs) with warming. An approximation to the latter term for the top one metre of soil (ΔCs,τ) can be diagnosed from projections made with the CMIP6 and CMIP5 Earth System Models (ESMs), and is found to span a large range even at 2°C of global warming (−196 ± 117 PgC). Here, we present a constraint on ΔCs,τ, which makes use of current heterotrophic respiration and the spatial variability of τs inferred from observations. This spatial emergent constraint allows us to halve the uncertainty in ΔCs,τ at 2 °C to −232 ± 52 PgC.

Interactions between the climate and carbon-cycle remain one of the largest uncertainties in climate prediction. The most uncertain aspect of the interaction is the modulation of biosphere-atmosphere exchanges by physical drivers. These modulations are likely heterogeneous in space and time and often act in concert. Satellites can measure many proxies of these interactions such as variability in concentration, vegetation indices, soil moisture, biomass etc. The question arises what combination of remote sensing capabilities is needed to constrain these feedbacks. This talk sketches a framework for assessing an observing system. We will show that modulations of production are amenable to remote sensing while responses of respiration are more difficult. I will make some suggestions on the path to a combined approach.

We assess the detectability of COVID-like emissions reductions in global atmospheric CO2 concentrations using a suite of large ensembles conducted with an Earth system model. We find a unique fingerprint of COVID in the simulated growth rate of CO2 sampled at the locations of surface measurement sites. Negative anomalies in growth rates persist from January 2020 through December 2021, reaching a maximum in February 2021. However, this fingerprint is not formally detectable unless we force the model with unrealistically large emissions reductions (2 or 4 times the observed reductions). Internal variability and carbon-concentration feedbacks obscure the detectability of short-term emission reductions in atmospheric CO2. COVID-driven changes in the simulated, column-averaged dry air mole fractions of CO2 are eclipsed by large internal variability. Carbon-concentration feedbacks begin to operate almost immediately after the emissions reduction; these feedbacks reduce the emissions-driven signal in the atmosphere carbon reservoir and further confound signal detection.

The assimilation of terrestrial carbon cycle remains difficult because of a lack of flexibility and modularity of the underlying terrestrial ecosystem models and also because of computational challenges when going to high frequency (daily) resolution. I will present a community-driven framework that should permit assimilation of data and uncertainty quantification at scale, while allowing learning new physical relationships such the dependence of photosynthesis on water stress.

Deforestation in the Brazilian Amazon has cleared over 20% of its original extent, but it is not the only land use change impact. Forest degradation — through selective logging, fires and fragmentation — likely affects a similar area, however its impacts on the carbon cycle and ecosystem responses to droughts remain uncertain. Here we investigate the potential carbon uptake of degraded forests and study how the heterogeneity of forest structures affect the Amazon forest response to droughts. We conducted a series of 1°×1° simulations for the Brazilian Amazon using the Ecosystem Demography Model (ED2), a dynamic vegetation model that represents the horizontal heterogeneity of canopy height, vertical foliage profile, and functional diversity of forest canopies. The model was initialized with observed forest structure from 541 airborne lidar transects (12.5 km × 300 m) that randomly sampled the Brazilian Amazon. We explored three scenarios: (1) retaining current forest structure without further changes; (2) natural recovery of forests; (3) expansion of degradation and deforestation. We found that degraded forests have high potential of carbon accumulation, but low-biomass forests show stronger negative responses to drought on both gross primary productivity and evapotranspiration. Expansion of forest degradation could cause more areas in the Amazon to become susceptible to drought-related stress. This work shows the substantial role of variation in forest structure, caused by natural and anthropogenic drivers, in mediating carbon-climate feedbacks. We will also discuss opportunities and challenges to leverage recent and upcoming space missions and advances in vegetation models to better understand how structural and functional diversity of forests modulate the land carbon and water cycles.

Most earth system models from the Coupled Model Intercomparison Project show relatively high levels of carbon accumulation in terrestrial ecosystems during the satellite era, with past work attributing much of this change to the influence of rising levels of carbon dioxide on photosynthesis as well as biosphere responses to changing nutrient availability and land use. New time series of aboveground biomass and forest cover change from Landsat and other sensors provide important new constraints on this set of model processes. Here we describe some recent work from boreal, temperate, and tropical forests describing long-term trends in aboveground live biomass and forest cover. In boreal forests of western North America, we show that rates of forest carbon accumulation are considerably lower than state-of-the-art land surface model estimates, highlighting the importance of fire in modulating interannual and decadal trends. Using these observations as an “emergent constraint” provides evidence that model estimates of the high northern latitude terrestrial carbon sink may be overestimated. In temperate forest ecosystems across California, we document declining levels of conifer forest cover that are closely tied to climate-driven increases in drought stress and burned area, with models often predicting forest trends with the opposite sign. For tropical ecosystems, we combine information from new satellite-derived maps of aboveground biomass, reanalysis, and CMIP6 estimates of the regional climate impacts of forest cover change to estimate the carbon-climate feedback associated with deforestation. Together, these studies suggest that climate change impacts on disturbance regimes are important and that a mechanistic representation of these processes in earth system models is key to reducing uncertainties associated with projections of future change.