Data-driven upscaling of biogenic fluxes from eddy covariance (EC) sites to the global scale is a powerful complementary approach to process-based models for the derivation of global flux estimates. Nevertheless, significant uncertainties arise due to specific methodological choices such as data availability, data source differences, machine learning model differences, and feature selection. Accurately quantifying these uncertainties from diverse sources is essential for providing error estimates of the simulated fluxes. These uncertainties not only improve our general understanding of carbon cycle processes but also directly inform atmospheric inversions, which can use the upscaled net ecosystem exchange (NEE) as a prior. However, most existing data-driven global carbon flux products focus solely on flux estimates or provide incomplete uncertainty assessments limited to a few sources.

In this study, we introduce a comprehensive framework for quantifying the uncertainties associated with carbon flux upscaling across potential sources. The framework involves three key steps: (1) pre-ensemble generation, (2) screening, and (3) uncertainty attribution. First, we construct ensemble members by training machine learning models with varying configurations, which include climate forcing datasets, feature combinations, subsets of EC sites, machine learning algorithms, and their hyperparameters. The experiments are supported by the recently developed data-driven modeling framework FLUXCOM-X, which enables a wide range of experiments with diverse methodological choices. We crafted a feature set that includes about 300 features to capture both current and historical state information. To capture the site representativeness uncertainty, we sample subsets from global EC sites based on geolocation and feature space. Additionally, we will also investigate different machine learning models and the variation of hyperparameters to generate the ensemble. Second, ensemble members that have a low contribution to the ensemble variance will be eliminated while we retain the most representative ones. We employ a feature selection algorithm, HybridGA, to screen important subfeature sets from near-infinite combinations. Moreover, we screen other ensemble members by assessing the distribution and spread of members. Finally, we will attribute uncertainties to various categories from the perspectives of machine learning and process-based modeling, and potential strategies to reduce these uncertainties are discussed. The framework is initially used to evaluate spatiotemporal NEE uncertain patterns in Europe, and will subsequently expand globally. Additionally, the estimated biogenic carbon flux uncertainty will be assessed with independent products. This work not only advances our understanding of the sources and patterns of upscaled flux uncertainties but also enhances the robustness of posterior estimates in atmospheric inversion models.

How to cite: Yang, Q., Walther, S., Nelson, J., Duveiller, G., Hamdi, Z., and Jung, M.: An uncertainty quantification framework for data-driven carbon flux upscaling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20406, https://doi.org/10.5194/egusphere-egu25-20406, 2025.

Mapping in-situ eddy covariance measurements (EC) of terrestrial carbon and water fluxes to the globe is a key method for diagnosing terrestrial fluxes from a data-driven perspective. We describe the first global products (called X-BASE) from a newly implemented up-scaling framework, FLUXCOM-X. The X-BASE products cover the globe at 0.05° spatial resolution for every hour and include estimates of CO2 net ecosystem exchange (NEE) and gross primary productivity (GPP).

Compared to previous FLUXCOM products, the new X-BASE NEE better reconciles the bottom-up EC-based NEE and estimates from top-down atmospheric inversions (global X-BASE NEE is -5.75±0.33 PgC yr-1). The improvement of global NEE was likely only possible thanks to the international effort to improve the precision and consistency of eddy covariance collection and processing pipelines, as well as to the extension of the measurements to more site-years resulting in a wider coverage of bio-climatic conditions. However, X-BASE NEE shows low inter-annual variability, which is common to state-of-the-art data-driven flux products and remains a scientific challenge. With 124.7±2.1 PgC yr-1, X-BASE GPP is slightly higher than previous FLUXCOM estimates, mostly in temperate and boreal areas, and temporal patterns agree well with TROPOMI-based SIF.

Many further opportunities for development exist. We will outline how the new FLUXCOM-X framework provides the necessary flexibility to experiment, diagnose, and converge to more accurate global flux estimates. Pathways of exploration include methodological choices in the selection and processing of eddy-covariance and satellite observations, their ingestion into the framework, and the configuration of machine learning methods.

How to cite: Nelson, J. A., Walther, S., Kraft, B., Gans, F., Duveiller, G., Weber, U., Hamdi, Z., Zhang, W., and Jung, M. and the FLUXCOM-X Team: X-BASE: terrestrial carbon and water flux products from FLUXCOM-X, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16220, https://doi.org/10.5194/egusphere-egu25-16220, 2025.

Given the ongoing climate change and the increasing frequency of extreme weather events, accurately assessing their impacts on crop productivity is crucial for developing adaptation strategies to mitigate negative impacts and ensure sustainable food security in the future. Process-based crop models are the preferred tools to simulate and predict crop yields under climate change. However, due to the simplified representations of complex biophysical processes, these models generally introduce uncertainty when used to account for crop yield losses. Integrating process-based crop models with data-driven machine learning methods shows great promise. In our study, we are developing a hybrid crop model, particularly the carbon cycle components (photosynthesis, carbon allocation, soil carbon decomposition, etc.), based on the state-of-the-art process-based vegetation model LPJmL (Lund-Potsdam-Jena managed Land). The empirical processes and parameters in the carbon cycle of LPJmL are replaced or augmented with neural networks. The resulting hybrid crop model can leverage information from observational data to simulate previously unresolved processes while maintaining the process-based understanding. We showcase how the hybrid crop model generalizes from the LPJmL to capture the carbon cycle under unseen climate conditions.

How to cite: Lin, Y., Gelbrecht, M., Badri, M., Hess, P., Bathiany, S., and Boers, N.: Hybrid modelling for crop carbon cycle, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17604, https://doi.org/10.5194/egusphere-egu25-17604, 2025.

Using neural networks to upscale eddy covariance measurements is a common approach to obtain global estimates of net ecosystem exchange (NEE) and thereby the land carbon sink. Unfortunately, this approach suffers from a limited representativeness of eddy covariance sites of the global picture, resulting in discrepancies between such data-driven bottom-up estimates of the land-atmosphere fluxes in comparison to independent top-down products from atmospheric inversions. Here, we introduce a novel method to bridge both approaches: recalibrating the last neural network layer in a Bayesian synthesis inversion. In other words, we find the least squares estimate of the last neural network layer weights, by first transporting the deep features and then inverting the covariance matrix of transported features to obtain a least squares estimator against atmospheric observations. This approach is possible because atmospheric tracer transport of CO₂ is a linear operator with respect to the surface fluxes. It is also computationally tractable due to a small number of degrees of freedom, namely just the regression coefficients for the approximately 50 deep features. For comparison, modern CO₂ inversions typically model the land surface flux with over 1000 parameters, which requires them to leverage variational or ensemble approaches for optimization.

The NEE estimates recalibrated using atmospheric data differ significantly from those obtained through pure eddy covariance training within the FLUXCOM-X framework. Namely, the recalibrated estimates show increased agreement with observational data from atmospheric measurement stations, when transported with the atmospheric transport model TM3. Surprisingly, this agreement does not necessarily arise from a greater agreement of global flux maps with results from the Jena CarboScope inversion. Here, the approach may suffer from low robustness of deep features or from regridding fluxes to a lower resolution before transporting them. We discuss ways to alleviate these limitations and outline what our results mean for improving neural network estimates of NEE.

How to cite: Benson, V., Jung, M., Glauch, T., Jin, Y., Kraft, B., Marshall, J., Reimers, C., Winkler, A. J., and Reichstein, M.: Recalibrating neural network estimates of net ecosystem exchange in a Bayesian synthesis inversion, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17329, https://doi.org/10.5194/egusphere-egu25-17329, 2025.

Understanding the role of land surface physics and biogeochemistry is crucial for improving climate models and weather prediction, particularly in the context of long-term variability, local feedbacks, and extreme events. Accurate boundary conditions—such as land cover (LC) and land use (LU)—are key to enhancing the realism of climate simulations by better representing land-atmosphere interactions that influence surface energy balance, and ecosystem processes. Moreover, they provide the substratum for a realistic representation of the terrestrial carbon cycle components, such as vegetation and soil biogeochemistry.

The CERISE project aims to produce high-resolution (1 km) LC and Leaf Area Index (LAI) datasets covering the period 1925–2020, contributing to novel reanalysis datasets (e.g., ERA6-Land), and seasonal forecasts (e.g., SEAS6). In the first phase, we reconstructed historical LU and LAI by leveraging machine learning (ML) models to downscale coarse-resolution LU datasets (LUH2f, HILDA+). Our workflow integrates multiple ML techniques, such as Random Forest and XGBoost, to train models over high-resolution LC and LAI satellite observations, while actively exploring methods to enhance both performance and interpretability. To capture monthly LAI variations from annual LU inputs, we developed an auxiliary network to model intra-annual variability. Initial results show promising performance in reconstructing LC and LAI across various test years and regions, demonstrating the feasibility and robustness of this ML-based approach for historical reconstructions.

Future phases, including the CONCERTO and TerraDT projects, will extend this work to generate consistent high-resolution LU datasets for the historical (1850-present) and future scenarios (present–2100), supporting CMIP7 climate simulations and scenario-based studies. These efforts will incorporate additional auxiliary data (e.g., elevation, soil types, climate indices) to improve feature representation and develop autoregressive models that account for long-term temporal dependencies and dynamic changes. Ultimately, our goal is to build a robust ML-based emulator for generating scalable, high-resolution land surface boundary conditions to support digital twin applications, thereby enhancing climate simulation and prediction capabilities.

How to cite: Mozaffari, A., Materia, S., Huggannavar, V., Teckentrup, L., Ayan, I., Tourigny, E., and Donat, M.: Reconstruction and downscaling of historical land surface boundary conditions with Machine Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4375, https://doi.org/10.5194/egusphere-egu25-4375, 2025.

Over the past two decades, machine learning (ML) has become a key tool in carbon cycle research, offering new methods to quantify fluxes, map carbon stocks and turnover, and disentangle processes like photosynthesis and respiration. Early efforts with classical ML models enabled scalable integration of remote sensing and ground-based observations, significantly reducing uncertainties. More recent advancements in deep learning and hybrid modeling approaches now support multi-scale analyses, integrating diverse datasets across terrestrial, oceanic, and atmospheric domains.

However, the quest for a comprehensive ML framework faces persistent challenges. Confounding factors in observational data complicate the identification of key drivers of carbon fluxes, while causal modeling remains underexploited. Extrapolation in space and time, integrating heterogeneous data sources, ensuring robust uncertainty quantification, and balancing predictive power with interpretability are further challenges.

This talk reviews major milestones and explores whether an all-encompassing ML solution is within reach—or if tailored approaches addressing specific challenges are the more realistic path forward.

How to cite: Reichstein, M.: Machine Learning and the Carbon Cycle: Chasing the Holy Grail, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17063, https://doi.org/10.5194/egusphere-egu25-17063, 2025.

Eddy Covariance (EC) method provides a valuable opportunity to monitor greenhouse gases, enabling informed decisions on climate change mitigation. Despite the abundance of EC data, explainable machine learning (ML) methods have not been effectively utilized to study the complex nature of methane (CH₄) fluxes, especially heterogeneity of emissions within ecosystems. This study explores the application of random forest ML model to analyse CH₄ flux spatiotemporal heterogeneity using flux data from the Ess-soo restored peatland in Estonia. This site, 30 years ago abandoned peat extraction area, was restored in 2021. To study CO₂ and CH₄ fluxes, open path EC analysers (LI-7500 and LI-7700, LICOR Biosciences) were installed in 2023. Additionally, CO₂ and CH₄ fluxes were measured biweekly using chamber method with the LI-7810 trace gas analyser (LICOR Biosciences) from 12 sampling spots in the EC footprint area. Other parameters such as water pH, electrical conductivity, dissolved oxygen concentration, temperature, oxidation reduction potenital, pH, and water level were conducted.

Chamber measurements revealed significant spatial CH₄ heterogeneity within EC flux footprint. The mean CH₄ flux from chamber measurement points during the summer months was 0.052 ± 0.013 µmol m^-2 s^-1 with a range of -0.001 to 0.555 µmol m^-2 s^-1. Looking into whole year EC dataset, main driver for CH₄ flux was water temperature. Day and nighttime fluxes responded differently to environmental changes, with air temperature and wind speed being significant drivers for day and night, respectively. The random forest model predicted CH₄ heterogeneity considerably better than general linear models performed (R² = 0.31 and 0.10, respectively). Besides identifying the main drivers, ML models can also combine EC and chamber measurements to detect hotspots and moments that are overlooked by EC alone. In that case, high spatial or temporal resolution  remote sensing data (e.g. LiDAR, Sentinel-1, Sentinel-2) was used. For instance, topographic wetness index calculated from LiDAR data in all points within EC flux footprint, was combined with water level—an important driver of both EC and chamber CH₄ fluxes. This information, together with chamber data was used to train ML models to estimate CH₄ fluxes spatially and temporally.

This work brings out the advantages in using ML and high spatial and temporal resolution remote sensing data to study CH₄ flux heterogeneity in wetlands. However, more testing is needed to see if these methods give similar results in other wetland sites.

How to cite: Tamm, I., Yildiz, K., Uuemaa, E., Pindus, M., Kull, A., and Kasak, K.: Using explainable machine learning to study restored peatland CH4 flux heterogeneity, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6531, https://doi.org/10.5194/egusphere-egu25-6531, 2025.

The extensively distributed grasslands of the Qinghai-Tibet Plateau (QTP) play a vital role in the global carbon cycle and climate regulation. Gross primary productivity (GPP), a crucial indicator of ecosystem carbon sequestration capacity, remains highly uncertain partly due to neglecting the memory effects of environmental conditions (i.e., the influence of past states on current GPP). Moreover, existing models have difficulty in simultaneously handle multidimensional spatio-temporal data and dynamic climate responses, leading to simulation deviations and exacerbating uncertainties. Here, we integrated climate and vegetation data with time series characteristics and spatial characteristics to simulate the GPP of alpine grassland on the QTP, by developing a deep learning model CNN-LSTM that combined Convolutional Neural Networks (CNNs) and Long Short-Term Memory networks (LSTM). The conclusions were as follows: (a) The CNN-LSTM model effectively captured spatial patterns using CNNs and temporal dependencies with LSTMs, incorporating memory effects to consider the impact of past environmental conditions. This integration enhanced GPP simulation accuracy and improved the model's ability to capture interannual variability. (b) The training and optimization of the CNN-LSTM models revealed that the comprehensive memory effect length of GPP on historical climate and vegetation dynamics operates in a 4-month timescale, with the memory effects of GPP varied across environmental variables in both duration and intensity. (c) During 2001–2021, The mean annual GPP of the alpine grassland in QTP was 332.29 g C m^-2 a^-1, with a growth rate of 1.84 g C m^-2 a^-2. (d) Precipitation exhibited relatively longer durations and higher intensities compared to other factors, and the interannual variability of GPP was mainly influenced by water conditions. This study highlights the importance of integrating environmental memory into GPP modeling, which would enhance our comprehension of the mechanisms driving GPP and the impacts of climate change on carbon cycling in terrestrial ecosystems.

How to cite: Zhang, Q. and Zhou, T.: Deep learning-based identification of environmental memory effects on gross primary productivity of alpine grasslands in Qinghai-Tibetan Plateau, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5574, https://doi.org/10.5194/egusphere-egu25-5574, 2025.

Terrestrial ecosystems gain carbon through photosynthesis and lose it through respiration in autotrophs and heterotrophs. Continuously measured values of carbon fluxes of a forest ecosystem, particularly net ecosystem exchange (NEE) could be used as a general indicator of forest ecosystem functioning. Subsequently, quantification of the ecosystem functioning as a response to changes in the microclimate and environmental variables is necessary to frame sustainable adaptive measures and conservation policies. The Himalayan Chir- Pine (Pinus roxburghii Sarg.) is a gregarious, fire-resistant, indigenous tree species, often form pure forests and having the characteristics of high regeneration potential.  The Chir-Pine is widely distributed across the western and central part of the Indian Himalayan Region and thereby acts as a major control of land-atmosphere processes. In the recent years, studies have provided insights on sub-daily to annual scale interactions of Chir-Pine ecosystem with microclimatic and environmental variables, and it was reported that Chir-Pine ecosystem is a heat dominating ecosystem with high carbon sequestration potential. However, almost no information is available on environmental drivers resulting carbon sequestration of Himalayan Chir-Pine ecosystem. In this context, it is widely reported that the data driven models are well suited for identifying and prioritizing drivers for ecosystem carbon exchange. Therefore, this study is aimed at developing a data-driven model for predicting day-time net ecosystem exchange of a Chir-Pine forest of central Himalaya, Uttarakhand, India. And further aims to quantify driver-response relationship between net ecosystem exchange (NEE) and micro-climatic variables using machine learning classifiers. In order to address the objectives, high frequency (30-min) day-time observations of NEE and micrometeorological parameters during March, 2020 to December, 2022 are collected and compiled from a 30 m eddy covariance tower situated at Kosi-Katarmal, Almora, Uttarakhand, India (29º38'22"N, 79º37'2"E). Subsequently, four machine learning algorithms such as K-nearest neighbor, Naïve Bayes, support vector machine and decision trees are used to predict the day-time NEE using individual and combinations of predictors such as rainfall, net radiation, air temperature, soil moisture and soil temperature. To obtain a robust model, 100 times bootstrapping has been performed in each simulation where 2/3^rd of the dataset is used for training the model and rest is used for testing.  The model performance during training and testing has been assessed using receiver operator characteristics and the prioritization of the driver impacting NEE is carried out by identifying highest area under curve (AUC) value during model testing. The initial results indicate that the decision tree classifier is the best model amongst the four selected model for predicting day-time NEE of Chir-Pine ecosystem, and the best predictors having high AUCs are air-temperature, net-radiation and soil moisture. The prediction of the NEE through data-driven models of Chir-Pine ecosystem is expected to be beneficial for quantifying the regional scale extent of change in carbon fluxes under warmer scenarios.

How to cite: Lohani, P., Mukherjee, S., and Pundir, S.: Investigation of eco-hydrogical processes influencing Himalayan Chir-Pine net ecosystem exchange using machine learning classifiers, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14709, https://doi.org/10.5194/egusphere-egu25-14709, 2025.

Accurate estimation of carbon dioxide (CO₂) flux is essential for better understanding of the global carbon budget and its impact on climate changes, which would further suggest strategies for emission reduction. Bottom-up approaches, while fundamental, often face challenges in capturing the complexities of CO₂ fluxes due to uncertainties in emission inventories and limitations in representing spatio-temporal variability of CO₂ flux across diverse regions. In contrast, top-down methods, which combine simulations and observations with inverse modeling approach, offer powerful tools for dynamically constraining CO₂ flux estimates. In comaparison to limited in-situ observations, satellite provides broader spatial coverage and therefore it is expected to enhance the flux estimation. In this study, we apply an integrated flux inversion framework NISMON-CO₂ to a CO₂ inversion with column-averaged dry air mole fraction of CO₂ (XCO₂), stored in NIES Level 2 product from the Greenhouse gases Observing SATellite (GOSAT) measurements and assess general features of the inversion results by comparing with an already established surface in-situ data inversion. NISMON-CO₂ incorporates NICAM-TM (Nonhydrostatic ICosahedral Atmospheric Model-based Transport Model) for forward simulation, coupled with a 4DVar (four dimensional variational) data assimilation system for inverse computations. The 4DVar leverages the adjoint of NICAM-TM alongside the quasi-Newtonian optimization scheme. GOSAT, a Japanese satellite launched in 2009 by Japan Aerospace Exploration Agency (JAXA) in collaboration with the Ministry of the Environment (MOE) and the National Institute for Environmental Studies (NIES), provides high quality greenhouse gas mesurements from space to study their global distribution.

The prior flux data include four fluxes: fossil fuel emissions from GridFED (Gridded Fossil Emission Dataset), biomass burning emissions from Global Fire Emission Database (GFED), biospheric fluxes (gross primary production, respiration and land use change) from the Vegetation Integrative SImulator for Trace gases (VISIT) and air-sea exchange flux data from Japan Meteorological Agency (JMA). In this study, meteorological data that drive NICAM-TM is updated to the Japanese Reanalysis for Three Quarters of a Century (JRA-3Q) from JRA-55. Several numerical experiments are conducted for the period since April, 2009 till date to understand the performance of the inversion by analyzing the consistency of the resultant flux and concentration. This study illustrates the power of integrating satellite-derived products to provide comprehensive CO₂ flux estimates, significantly enhancing our understanding of CO₂ dynamics at global and regional scales.

Keywords: CO₂ flux estimation, GOSAT, XCO₂, 4DVar, NICAM, transport model, satellite data assimilation.

How to cite: Maity, S., Niwa, Y., Saeki, T., Someya, Y., and Yoshida, Y.: Global CO2 flux estimation using NISMON-CO2 and GOSAT for carbon cycle analysis improvement, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1822, https://doi.org/10.5194/egusphere-egu25-1822, 2025.

A better understanding of the country-scale terrestrial biospheric carbon dioxide (CO₂) budget is crucial for formulating national climate policies aimed at limiting carbon emissions. The scarcity of continuous and dense regional CO₂ measurements in India poses a significant challenge to accurately quantifying its carbon budget. Moreover, there are no observation-based studies investigating the regional carbon-climate interactions and carbon cycle response due to large-scale climatic events currently exist. In this study, we use the OCO-2 satellite atmospheric CO₂ column (XCO₂) retrievals, Solar Induced Fluorescence (SIF) and various observational data to study the Indian terrestrial biosphere’s response to large-scale climatic events such as El Niño-Southern Oscillation and Indian Ocean Dipole (IOD). The XCO₂ data was assimilated in an ensemble of eight global top-down CO₂ flux inverse models as part of the OCO-2 v10-Ext multi-model intercomparison project. We found a decrease in Indian terrestrial biosphere carbon uptake during El Niño and an increase during La Niña and positive IOD events. The increase in carbon uptake, driven by pIOD and La Niña events (~150 TgC) accounts for approximately one-quarter of India's annual fossil fuel carbon emissions. Studies indicate that the frequency of pIOD and La Niña events may rise under future global warming scenarios. This can potentially enhance the capacity of the Indian terrestrial biosphere to sequester more atmospheric carbon. Satellite-derived carbon-climate constraints over India as found in this study provide critical insights for developing effective strategies to achieve net-zero emissions in the future.

Acknowledgements: OCO-2 v10-Ext MIP modelers.

How to cite: Singha Roy, E., Philip, S., and S. Johnson, M.: Impact of Global Climatic Phenomena on the Carbon Exchange Dynamics of the Indian Terrestrial Biosphere, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2991, https://doi.org/10.5194/egusphere-egu25-2991, 2025.

Accurate national terrestrial net ecosystem exchange estimates are crucial for the global stocktake. Net ecosystem exchange estimates from different inversion models vary greatly at national scale, and the relative impacts of prior fluxes and observations on these inversions remain unclear. Here we estimate the net ecosystem exchange of 51 land regions for the 2017-2019 period, focusing on the 10 largest countries, using prior fluxes from 12 terrestrial biosphere models and XCO₂ retrievals from GOSAT and OCO-2 satellites as constraints. The average uncertainty reduction for the 10 countries increases from 37% with GOSAT and 45% with OCO-2 to 50% with combined observations, indicating a trend towards robust estimates. At finer spatial scales, even with combined observations, the uncertainty reduction is only 33%, i.e., the prior flux dominates the estimates. This finding underscores the critical importance of integrating multi-source observations and refining prior fluxes to improve the accuracy of carbon flux estimates.

This study provides valuable insights for improving atmospheric inversions in the future, and offers a deeper understanding of the inversion results for the carbon cycle community. Additionally, the improved estimates of carbon fluxes for the 10 largest countries presented here can inform policy makers in making more informed decisions regarding climate and carbon management strategies.

How to cite: Zhang, L., Jiang, F., He, W., Wu, M., Wang, J., Ju, W., Wang, H., Zhang, Y., Sitch, S., and Chen, J. M.: Improved estimates of net ecosystem exchanges in mega-countries using GOSAT and OCO-2 observations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3937, https://doi.org/10.5194/egusphere-egu25-3937, 2025.

In this study, a high-resolution CO₂ data assimilation (DA)-forecast system was developed to improve atmospheric CO₂ concentration simulations in East Asia. The Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) was used for simulating regional CO₂ concentrations and the ensemble adjusted Kalman filter (EAKF) in modified Data Assimilation Research Testbed (DART) was used for assimilating CO₂ concentration observations. To evaluate the performance of the developed DA-forecast system, observing system simulation experiment (OSSE) was performed in January and July 2019. Four experiments, which assimilated pseudo surface CO₂ observations from four observation site networks, were conducted to avoid the influence of observation site distributions. In January and July 2019, the ratios of the root mean square error (RMSE) to the ensemble total spread for surface CO₂ concentrations were 1.00 and 0.97, respectively. By assimilating surface CO₂ concentrations, the bias and RMSE of simulated CO₂ concentrations reduced by 1.23 ppm and 1.24 ppm in January and 1.41 ppm and 1.84 ppm in July, implying the stability of the developed DA-forecast system. Among four experiments, the experiment with an evenly distributed observation site network showed the smallest RMSE for surface CO₂ concentration. The RMSE of the experiment with the existing CO₂ observation network was greater than that with the evenly distributed observation network, but was smaller than that without DA. While the DA using the evenly distributed observation network showed the best performance for simulating CO₂ concentrations in East Asia, the DA using the existing surface CO₂ observation network also improved CO₂ simulation performance compared to the experiment without DA.

Acknowledgments

This study was supported by a National Research Foundation of Korea (NRF) grant funded by the South Korean government (Ministry of Science and ICT) (Grant 2021R1A2C1012572) and the Yonsei Signature Research Cluster Program of 2024 (2024-22-0162).

How to cite: Seo, M.-G. and Kim, H. M.: Development and evaluation of high-resolution regional CO2 data assimilation-forecast system in East Asia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5697, https://doi.org/10.5194/egusphere-egu25-5697, 2025.

Land cover change has direct implications on natural greenhouse gas emissions, as land-atmosphere interactions are function of the changing heterogeneity of the surface. Rapidly changing ecosystems, such as the Arctic, where permafrost wetting and draining is taking place in different regions in the northern latitudes, underlines the necessity of assessing patch-level emissions of greenhouse gases to better estimate net total fluxes. In this study, we combine high-resolution modelling of the atmospheric boundary layer with inverse modelling concepts to constrain land-atmosphere exchange fluxes at local to landscape scales, and explore relationships between different land cover types within heterogeneous landscapes and the net exchange processes between surface and atmosphere. We use EULAG (EUlerian LAGrangian), an established Large-Eddy Simulation model, to simulate high-resolution flow patterns induced by heterogeneous permafrost surfaces, and apply inversion techniques to infer the fluxes of the corresponding patch type forming the mixed land cover. Uncrewed Aerial Vehicles (UAV)-based grid surveys of gas concentrations are used to benchmark the spatial variability of modeled concentrations using EULAG, where we optimize for surface fluxes associated with each patch. We present a case study at Stordalen Mire in subarctic Sweden, where we use UAV measurements of methane and carbon dioxide mole fractions, and implement this inversion method to differentiate the flux rate signatures from different patch types, namely palsa, bog, and lakes. The inferred fluxes were validated with patch-level chamber measurements of methane and carbon dioxide. Our model evaluation shows a good match between modeled and observed concentrations while the resulting patch-level fluxes agree with the observed fluxes from chamber measurements. Our novel technique shows promising results in inferring patch type flux heterogeneity while facilitating the application of inversion methods to high resolution atmospheric models.

How to cite: Yazbeck, T., Schlutow, M., Bolek, A., Triches, N. Y., Wahl, E., Heimann, M., and Göckede, M.: Coupling large-eddy simulations with UAV measurement through inversion technique to estimate patch-level fluxes from heterogeneous tundra landscapes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8137, https://doi.org/10.5194/egusphere-egu25-8137, 2025.

A reliable representation of the diversity and variability of terrestrial ecosystems, both natural and managed, is crucial to the accurate simulation of their present and future roles in biogeochemical cycles and global climate. In this study we compare the treatment of vegetation distributions and of photosynthetic pathways (C₃ versus C₄) of both natural vegetation and crops across Earth System Models (ESMs) in the 6th Coupled Model Intercomparison Project (CMIP6). Of the 11 CMIP6 models reporting variables on crop and C₃ versus C₄ distribution use, 10 models use the crop distributions of the Land Use Harmonization v2 (LUH2) dataset, which has an increase of ~188 and ~254% in C₃ and C₄cropland, respectively, from 1850 to 2014. The models simulate a 10% decrease in the area coverage of natural vegetation with the C₃ photosynthetic pathway but disagree on the trend of C₄. The impact on carbon isotopic discrimination from simulated C₃ and C₄ GPP trends only, not accounting for physiological effects, is generally to drive a decreasing trend in discrimination, especially in models with increasing C₄ vegetation cover, opposite to the trend derived from atmospheric data. Our findings suggest that implementation of C₃ and C₄ vegetation area abundance and GP of C₃ and C₄ vegetation contribute to uncertainty in land carbon fluxes and need further constraints and improvement in ESMs.

How to cite: Ovwemuvwose, J., Graven, H., and Prentice, C.: Uncertainty in Land Carbon Fluxes Simulated by CMIP6 Models from Treatment of Crop Distributions and Photosynthetic Pathways, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10249, https://doi.org/10.5194/egusphere-egu25-10249, 2025.

The biosphere-atmosphere CO₂ exchange is the largest carbon flux in the global carbon cycle, yet substantial uncertainties remain in quantifying gross primary production (GPP) and ecosystem respiration (Re). Top-down atmospheric inversion modeling provides a powerful approach to reduce the uncertainties in surface fluxes through a combination of atmospheric observations and transport modeling. However, as during nighttime mixing process of the atmosphere is weakly developed and hard to simulate in atmospheric transport models, atmospheric inversions typically rely on afternoon observations when both GPP and Re occur simultaneously, making it challenging to isolate their individual contributions.  In order to disentangle the respiration signals and simultaneously utilize previously unused observational data, we established a novel algorithm for the identification of night-time mixing height, based on the temporal variation of virtual potential temperature from ICOS tower measurements. The method is validated using profile information on greenhouse gases. We then integrated CO₂ signals below the diagnosed mixing height and incorporated these partial column increment as observational operators in CarboScope-Regional (CSR), a Bayesian inverse modeling framework. This enhanced inversion scheme enables improved quantification of ecosystem respiration (and, by extension, GPP), bringing about a better understanding and constrains on the the role of biological fluxes in European carbon budgets.

How to cite: Xu, Y., Galkowski, M., Munassar, S., Ho, D. (.-H., Koch, F.-T., and Gerbig, C.: Investigating ecosystem respiration CO2 signals using night-time ICOS tower observations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10599, https://doi.org/10.5194/egusphere-egu25-10599, 2025.

The global Stocktake, a fundamental component of the Paris Agreement tracking progress on national mitigation actions, collects the Nationally Determined Contributions (NDCs) generated through the means of annual national inventories. Carbon capture through the natural ecosystem is essential to reach the Paris Agreements and thus it is crucial to understand the interaction of the atmosphere/biosphere and its changes with global warming. We present the model performances of our regional inversion system over France for the year 2022, with a special focus on an extreme drought event that impacted southern Europe during the summer. Our inversion system optimizes CO2 fluxes from fossil fuel and biogenic fluxes at higher spatiotemporal resolutions over France (3km, hourly). The Lagrangian Particle Dispersion Model (LPDM) developed running in a backward-in-time model, driven by meteorological inputs from a 3-km run of the Weather Research Forecast Model (WRF-Chem), establishes the transport of CO2 molecules. Employing a Bayesian inversion technique, we optimize prior CO2 flux estimates by integrating tower footprints and ICOS atmospheric measurements into a newly developed inversion framework, combining block matrix decomposition and adaptive mesh refinement. We infer the prior flux estimates using the TNO high-resolution fossil fuel inventory and biogenic CO2 fluxes produced by the Vegetation Photosynthesis Respiration Model (VPRM). We start by evaluating the WRF-chem model performances at high resolution compared to low resolution simulations. Then we assess the meteorology and CO2 exchanges over continental France throughout the year 2022. With the Lagrangian Model, we can explore the actual ICOS network constraints by determining the share of biogenic and fossil fuel sources at each tower of the ICOS network. We discuss here how our inversion system could help constrain the regional distribution of CO2 fluxes, including sub-annual variations at seasonal and monthly timescales to track current climate change impacts (forest fires, droughts), and the effects of emission mitigation policies. Finally, we determine potential networks of surface stations (extension of the current ICOS network) to enable the monitoring of CO2 fluxes and emissions at policy-relevant scales over continental France.

How to cite: D'angeli, C., Lauvaux, T., Matajira Rueda, D., Che, K., Abdallah, C., Bazi, H., Ciais, P., Ramonet, M., Lopez, M., and Rivier, L.: Towards a high-resolution inversion system over France using in-situ observations , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13589, https://doi.org/10.5194/egusphere-egu25-13589, 2025.

Belgium’s national greenhouse gas (GHG) inventory currently relies on a bottom-up approach, but incorporating top-down methods using atmospheric observations and inverse modeling offers significant potential to improve the understanding of CO₂ and CH₄ emissions. The VERBE project aims to develop such a system tailored for Belgium by combining satellite, ground-based remote sensing, and in situ observations from the Integrated Carbon Observation System (ICOS) network with inverse modeling techniques. As part of this effort, we start by assessing the ability of the atmospheric transport model to accurately reproduce the spatiotemporal distribution of GHGs in this region.  

We employed the Weather Research and Forecast model coupled with chemistry in its Greenhouse Gas configuration (WRF-GHG) to simulate the Western Europe region, with a focus over Belgium, from June to August 2018. Simulations were conducted at horizontal resolutions of 9 km and 3 km over two domains. In comparison with meteorological data from Automatic Weather Stations in Belgium and ICOS sites, our results indicate that the WRF-GHG simulation is capable to capture the variations of the near surface meteorological fields (temperature, wind speed and wind direction) very well, especially for temperature.

The simulated CO₂ and CH₄ are compared with near-surface concentrations at different heights from four ICOS sites around Belgium and with column-averaged dry-air concentrations from the Total Carbon Column Observing Network (TCCON) site in Orléans, France. While WRF-GHG successfully reproduces most observed variations, discrepancies were identified. These include an overestimation of the CO₂ peak values at most ICOS sites and an overall underestimation of near-surface CH₄ concentrations by 20-30 ppb at three of the four ICOS sites. Additionally, the TCCON comparison revealed a significant deviation in XCO₂ in early June, likely due to inaccuracies in biogenic fluxes which are calculated based on the Vegetation Photosynthesis and Respiration Model (VPRM). For XCH₄, we find an increasing bias towards the end of summer, possibly related to the background signal.

We will present the latest results of our analysis, including additional observational data and updates to the model configuration aimed at improving model-data agreement such as the integration of the TNO high-resolution fossil fuel inventory and refinements to the VPRM fluxes.

How to cite: Wang, J., Callewaert, S., Desmet, F., Zhou, M., and De Mazière, M.: Towards a greenhouse gas emission monitoring and Verification system for Belgium (VERBE): Evaluation of WRF-GHG simulations with observational data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18007, https://doi.org/10.5194/egusphere-egu25-18007, 2025.

Urban areas are significant contributors to global CO₂ emissions, and simulating CO₂ dispersion in these regions, especially near emission hotspots, presents considerable challenges due to the complex dynamics at small scales. High-resolution simulations are crucial for accurately capturing the dispersion of CO₂ in urban environments. As part of the Carbon Atmospheric Tracer Research to Improve Numerics and Evaluation (CATRINE) project, this study employs the Weather Research and Forecasting model with the Large-Eddy Simulation mode (WRF-LES) to simulate CO₂ concentrations over the Paris area, aiming to improve plume simulation accuracy. The study evaluates the model's performance in urban environments and investigates the added value of LES by comparing simulation results with those from mesoscale configurations. A series of simulations using five nested domains, with resolutions ranging from 8.1 km to 100 m, were performed to examine the sensitivity of plume structures to model resolution. The study also investigates the propagation of errors when running a pseudo-data CO₂ inversion using high-resolution 100-m resolution WRF outputs to generate data, but inverting using lower resolution simulations (300-m and 900-m resolutions). The focus is on understanding how resolution influences inversion model results and quantifying aggregation errors introduced when aggregating higher-resolution outputs to coarser resolutions. 

Preliminary findings emphasize the advantages of LES in capturing complex plume features, reducing numerical diffusion, and producing more concentrated, well-defined CO₂ plumes. Resolution intercomparisons highlight that higher resolutions better capture sharp concentration gradients, localized dispersion patterns, significantly outperforming traditional mesoscale models. Additionally, WRF model outputs were validated against observations from various sources, including the Paris Mid-cost CO₂ sensor network, total column of CO₂ measurements from EM27 and Total Carbon Column Observing Network (TCCON), and wind LIDAR data from six stations across Paris and Île-de-France, collected during the URBISPHERE project. Future studies will extend this research to other urban cities, utilizing different LES models such as WRF-LES, Micro-HH, and Parallelized Large-Eddy Simulation Model (PALM). Intercomparing these models will provide performance metrics on model resolution when assimilating complex urban plumes combining multiple diffuse sources and point sources, thereby further refining the accuracy of CO₂ dispersion models for urban emissions monitoring and climate mitigation strategies.

How to cite: Rafalimanana, A., Lauvaux, T., Abdallah, C., Che, K., Ramonet, M., Doc, J., Laurent, O., Lopez, M., Raznjevic, A., Krol, M., Järvi, L., David, L., Sanchez, O., Christen, A., Grimmond, S., and Morrison, W.: Investigating the Benefits of Large-Eddy Simulation for Simulating Urban CO2 Emissions Using WRF-LES Over the Paris Area, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18221, https://doi.org/10.5194/egusphere-egu25-18221, 2025.