Current state-of-the-art estimates of emissions from vegetation fires were mainly derived from burned area datasets from medium-resolution satellite sensors and those burned area datasets under-detect small fires. Hence, fire carbon emissions could be much larger than previously estimated. This is critical as medium-resolution burned area datasets are frequently used to evaluate, improve and calibrate global fire-enabled vegetation models. Calibrating global vegetation-fire models with biased estimates of burned area might hence result in an underestimation of the role of fires in global vegetation dynamics and the carbon cycle. Furthermore, knowledge about fire carbon emissions mostly relies on satellite observations of burned area that are combined with simulations from ecosystem models of fuel loads (biomass) and combustion. Alternative approaches to estimate fire emissions make use of observations of fire radiative power or fire radiative energy as a proxy for fire emissions. However, both approaches make little use of information about fire type, or aspects of fire behaviour related to smouldering and flaming combustion.

These limitations in the current data basis on fire emissions, demonstrates the need to explore the information from the Sentinels: The Sentinel-5p TROPOspheric Monitoring Instrument provides several observations related to fire emissions such as the absorbing aerosol index, aerosol layer height, nitrogen dioxide (NO2), carbon monoxide (CO) and formaldehyde. The Sentinel-3 Sea and Land Surface Temperature Radiometer allows the mapping of active fires and fire radiative power. The Sentinel-3 Ocean and Land Colour Instrument allows the mapping of fire-induced land cover changes (e.g. burned area, fire severity) at medium resolution and to retrieve pre- and post-fire vegetation properties such as leaf area index or fractional vegetation cover. The Sentinel-2 Multispectral Instrument allows the mapping of fire-induced land cover changes at a higher spatial resolution (10-20 m). The Sentinel-1 C-band Synthetic Aperture Radar allows for the estimation of surface soil moisture and can serve as a proxy for the moisture content of surface fuels. Based on the complementary information from the Sentinels, we are currently developing a series of products to better characterise fuel conditions, fire behaviour and fire emissions at a high spatial resolution and for individual fires.

Vegetation fuel loads and combustion completeness are estimated using a novel fuel data integration framework. Therein we combine surface reflectance and vegetation information from Sentinel-3, Sentinel-2 and Proba-V; land cover and above-ground biomass from the ESA Climate Change Initiative and from the Copernicus Land Service; vegetation optical depth; and soil moisture from Sentinel-1 and Metop/ASCAT. The approach uses an empirical allometry model to estimate the fuel loads of different biomass compartments of trees and herbaceous vegetation by using total above-ground biomass and leaf area index as input. Additionally, surface fuels are estimated by combining land cover, leaf area index and above-ground biomass with databases of ground observations. This information is used to provide information about fuel loads in bottom-up estimates of fire emissions.

Fire behaviour and burned area are quantified using a novel mapping of individual fires based on thermal anomalies and the diurnal fire cycle from Sentinel-3 and of burned area estimates from Sentinel-2. Additionally, the morning (10 am) and evening (10 pm) overpasses from Sentinel-3/SLSTR are combined with mid-afternoon (1:30 pm) and night-time (1:30 am) overpasses from VIIRS to track individual wildfires as they evolve. The resulting maps track fire behaviour, type and size and will enable direct estimates of fire emissions based on estimates of fuel loads and combustion completeness from a combination of modelling and top-down constraints.
Fire effects on atmospheric composition are quantified by contrasting observations of trace gases (CO, NO2, formaldehyde) and aerosols from Sentinel-5p with model results from the Copernicus Atmosphere Monitoring Service (CAMS) modelling system. A model for atmospheric composition is necessary to be able to evaluate estimated emissions against satellite observations of atmospheric composition, i.e. to provide top-down constraints on fire emission estimates. We compare aerosol plume altitude against retrievals of aerosol layer height, to constrain plume dynamics, as well as observed and modelled CO and NO2 to constrain their emissions. The analysis helps to quantify the magnitude and uncertainties from top-down fire emission estimates, and will lead to improvements in the parametrisation of the model fire plume dynamics.

The integration of these developments based on Sentinel-1, -2, -3 and -5p will enable us to estimate fire emissions at high spatial resolution and in the long-term to provide estimates of emissions from individual fires. This information will be used in the future to constrain global fire models and hence to advance the understanding of the role of fires in the global carbon cycle.
We acknowledge the European Space Agency for funding the Sense4Fire (sense4fire.eu) project.

We report on the development and application of the new DALEC & BETHY (D&B) model that is performed within ESA's Land surface Carbon Constellation (https://lcc.inversion-lab.com) study. This new community model is designed to simulate a range of satellite and in-situ observations through dedicated observation operators. A suite of observation operators allows the simulation of solar induced fluorescence, Fraction of Absorbed Photosynthetically Active Radiation, Vegetation Optical Depth from active and passive sensors, and surface layer soil moisture – and thus the assimilation of those data streams. To our knowledge, the D&B assimilation framework will be the first to combine such a large and diverse array of observational constraints moving beyond site scale to regional applications. D&B builds on the strengths of each component model in that it combines the dynamic simulation of the carbon pools and canopy phenology of DALEC with the dynamic simulation of water pools, and the canopy model of photosynthesis and energy balance of BETHY. The model uses an hourly time step, except for the water balance, which (currently) is simulated at a daily time step. We present an evaluation of the model performance against a range of in-situ observations at two well-instrumented sites at which field campaigns are carried out: (1) Sodankylä, Finland, located in a boreal evergreen needleleaved forest biome and (2) Majadas de Tietar, Spain, located in a temperate savanna biome. The model performance will also be assessed against a range of satellite observations for approximately 500 km x 500 km regions around each site. The model is embedded into a variational assimilation system that adjusts a combination of initial pool sizes and process parameters to match the observational data streams. For this purpose the D&B assimilation system is provided with efficient tangent and adjoint code. We will show initial data assimilation experiments at site scale.

At the canopy level, gross primary productivity (GPP) represents the major global carbon uptake from the atmosphere. GPP can be derived at site level by partitioning direct observations of net ecosystem exchange by eddy covariance. For a global picture these sparse measurements can be upscaled with machine learning techniques, to give globally dense estimates. However these estimates hinge on having the right predictors available. While vegetation indices such as NDVI or EVI are valuable predictors, providing an indication of the overall vegetation greenness, Solar Induced Fluorescence (SIF) has been shown to relate better to photosynthetic activity. However, time series of SIF are still very short, their spatial resolution is generally much coarser than desired, and the signal-to-noise ratio is lower than that of reflectance-based indices. With the Copernicus program and the fleet of Sentinel satellites, there is a possibility to improve the last two. Together, Sentinel-2 (MSI), Sentinel-3 (both OLCI and SLSTR), and Sentinel-5P (TROPOMI), measure a variety of indices at complementary spatial, temporal, and spectral resolutions. While TROPOMI estimates SIF, that is directly linked to photosynthesis at daily quasi-global coverage, Sentinel-2 and Sentinel-3 instruments have much higher spatial resolutions, but coarser temporal resolutions and provide information about heat stress, greenness, chlorophyll concentration and landscape heterogeneity. The objective of this work is to synergistically capitalize on these complementary characteristics to yield enhanced GPP estimates at 1 km spatial resolution. Our derived high resolution SIF is benchmarked against flux tower GPP across selected areas with diverse ecosystems that are within the Sen4GPP project. We aggregate Sentinel-2 and downscale TROPOSIF to a common grid of 1km for Sentinel 3 SLSTR. By aggregating Sentinel-2 we derive a heterogeneity index from the high resolution information. For downscaling we explore a variety of methods ranging from statistical (including machine learning) to more explainable approaches based on process understanding. Distributing the coarse resolution SIF signal depends on investigating and understanding the dependencies and sensitivities of SIF to environmental and remote sensing indices (that come generally from Sentinel-2 and Sentinel-3). This allows us to present a spatial resolved GPP estimation at 1 km scale across Europe that is based on the synergistic use of the Sentinel (2, 3, 5P) satellite measurements.

Primary Production is the main driving factor of the terrestrial carbon cycle. Knowing how much atmospheric carbon dioxide is converted into biomass is critical information for scientists and policy makers. In this context, Earth Observation (EO) is an essential technology to quantify Gross and Net Primary Production (GPP & NPP). While EO based primary production quantification has a multi-decadal history, the scientific and technical advances that can support EO based GPP and NPP have been strongly progressed in latest years. Enhanced inputs are available, pre-processing techniques are improved, new generation state-of-the-art GPP models are emerging, in-situ data is becoming operationally available and cloud processing platforms are capable to deal with large datasets.

Available operational GPP/NPP products are currently focused on medium resolution grids (e.g. Copernicus DMP at 300m and MODIS GPP/NPP at 250m). There is however a clear demand for high resolution GPP/NPP at larger scales, for example the support LULUCF (Land Use, Land-Use Change and Forestry) reporting. With the launch of the High Resolution Vegetation Phenology and Productivity (HR-VPP) service, Sentinel 2 fAPAR data (a critical input for Light Use Efficiency (LUE) GPP models) is now operationally available over the European continent. This opens up the possibility to explore and prototype high resolution GPP/NPP products at larger scales. Within this context, the operational Copernicus DMP model has been applied on HR-VPP data at site level and test products generated at tile basis. The accuracy of the products have been assessed using the in-situ data available through the ICOS network. This work highlighted a number of insights in high resolution GPP/NPP modelling, including the requirements for enhanced pre-processing techniques, ancillary data requirements, model structure and in-situ data required. The lessons learned from this work will be presented and complemented with technical and scientific conclusions from the TerrA-P project, where Sentinel 3 data was exploited for GPP/NPP modelling at European to global scale.

The integration of global land surface remote sensing and in-situ measured ecosystem carbon fluxes
through machine learning approaches offers a unique data-driven perspective to diagnose the carbon cycle. Different Earth Observation (EO) data sets contain specific information on structural and/ or physiological vegetation conditions, or on the status of land surface e.g. in terms of moisture conditions. Every single EO product alone addresses only individual aspects of the complex system, and can be confounded by other factors. The synergistic combination of complementary EO products therefore offers the greatest promise for improvements in our data-driven modelling capacities of land surface productivity. We use the new implementation of the statistical flux upscaling framework Fluxcom (Tramontana et al. 2016, Jung et al. 2020) and tailored cross-validation experiments to analyse the the individual and synergistic contributions of different EO data sets to site-level prediction accuracy for terrestrial carbon fluxes. Each of the EO predictor variables receives a dedicated and careful preprocessing in terms of quality checks and gap-filling. Meteorological observations from the sites can be included as additional predictor variables in the experiments. Next to their overall importance for prediction skill, we are interested in understanding the impacts of the EO data sets on different scales of carbon flux variability (e.g. diurnal, seasonal, seasonal anomalies, inter-annual, and between sites) and to what extent differences in acquisition properties play a role for the model estimates.
First results for the example of MODIS LST_cci indicate that it strongly contributes to prediction accuracy of gross primary productivity (GPP) for all time scales, but this is particularly the case for inter-annual variations. The contribution of MODIS LST is even slightly larger than the one of site-level air temperature, with the remarkable exception of GPP anomalies, where the prediction accuracy is low without meteorological information. In times of dry anomalies, the model strongly profits from LST as a surrogate for moisture availability. Conversely, in models without meteorological information, LST mostly acts as a proxy for light availability and improves GPP accuracy for wet anomalies as well. Regarding the impact of acquisition properties of MODIS we find that the variability in viewing geometry and overpass time does not affect predicted site-level GPP. However, thermal measurements as done by MODIS can only inform us about land surface conditions in the absence of clouds, which generates a bias towards clear-sky conditions. Failing to account for this bias in availability of MODIS LST will result in 50% higher predicted GPP values and an increase in relative prediction error to more than 100% for overcast days.
We will also present results on the individual and synergistic added values of other EO data sets, such as land surface temperature from geostationary satellites (LST_cci product from the Seviri instruments), SIF from the GOME2 and Tropomi instruments, and SMOS VOD and soil moisture for data-driven estimates of carbon fluxes – overall and under water stress - which are the focus of ongoing work in the ESA Living Planet Project ‘Vad3e mecum’ (‘Vegetation and drought: towards improved data-driven estimates of ecosystem carbon fluxes under moisture stress’).
The lessons learned from the site-level cross-validation experiments will guide the production of more accurate gridded estimates of gross and net carbon fluxes for Europe and the globe within Vad3e mecum. Those are of great relevance to increase our process understanding of terrestrial productivity and will contribute to improved characterisation of biogenic fluxes, e.g. in atmospheric inversions.

Terrestrial ecosystems have absorbed more than one-third of cumulative anthropogenic emissions during the past decades, mitigating global warming (Friedlingstein et al., 2020). Earlier studies suggest that an increase in photosynthesis in response to elevated atmospheric CO2 concentration (eCO2) has the potential to increase the strength of the terrestrial carbon sink (that is, CO2 fertilization), providing a negative feedback on the growth of atmospheric CO2 concentration (Schimel et al., 2015; Sitch et al., 2015). Such CO2 fertilization effects have been reported at forest inventory plots and in short-term CO2 spring experiments (Pretzsch et al., 2014; Hubau et al., 2020; Terrer et al., 2021). However, it remains unclear whether and to what extent rising CO2 concentration influences vegetation carbon stocks and its long-term changes at the global scale. Here, we used two sets of newly satellite-based above-ground biomass (AGB) products, i.e., BIOMASCAT (https://eo4society.esa.int/projects/biomascat/) and Xu et al. (2021), and gross primary productivity (GPP) measurements from 89 long-term FLUXNET sites, and isolated the CO2 fertilization effects on AGB and GPP from the effects of concurrent anthropogenic climate change, land-use change, and nitrogen decomposition over the period 2000-2019 using multiple linear regression. The observation-based independent AGB sensitivity and site-scale GPP sensitivity to eCO2 were then used to constrain the modelled global AGB sensitivity to eCO2 from 13 dynamical global vegetation models (DGVMs) using an emergent-constraints approach. Our constrained estimates from two satellite AGB products and FLUXNET GPP measurements show convergent results, that is, the magnitude of global CO2 effects on AGB is 5.5 – 6.6% on average (ranging from 1.7 to 9.5%). This value suggests a substantial CO2 fertilization effect on global AGB changes, but that is around 20% lower than the modelled ensemble means of 7.7% from DGVMs and 8.1% from Earth System Models (ESMs) ensembles. A direct implication is that CO2 fertilization alone could reduce atmospheric carbon by 1.9 – 2.26 PgC yr-1 per hundred ppm of eCO2 (ranging from 0.6 to 3.2 PgC yr-1 [100ppm]-1), ignoring climate change effects and vegetation adaptation. Although relying on a simple statistical approach we provide a robust estimate of the carbon dioxide removal by the terrestrial vegetation that can partially offset the increased carbon emissions from fossil fuel emissions. These results emphasize the role of legacy Earth Observations in constraining global carbon cycle diagnostics, contributing to understand and predict potential climate mitigation from vegetation biophysical feedbacks.