BG3.13 | Novel methods for bridging process-based and data-driven modelling of carbon, energy, and water fluxes from leaf to continental scales
Novel methods for bridging process-based and data-driven modelling of carbon, energy, and water fluxes from leaf to continental scales
Convener: Mana Gharun | Co-conveners: Alexander J. WinklerECSECS, Christina Bogner, Holger Lange, Gregory Duveiller, M. Piles, Rossella Guerrieri
| Thu, 27 Apr, 14:00–15:45 (CEST)
Room 2.17
Posters on site
| Attendance Fri, 28 Apr, 14:00–15:45 (CEST)
Hall A
Orals |
Thu, 14:00
Fri, 14:00
A fundamental understanding of biosphere-atmosphere interactions is an invaluable asset for accurately representing the terrestrial carbon, water, and energy cycles. Multiple processes determine how the exchange of mass, energy, and momentum scale from leaf to plant, to ecosystem, and eventually to the entire globe. Challenges remain in robustly formulating the mechanistic underpinnings of these biogeochemical processes across all these scales and improving process-based modelling efforts without falling into a complexity trap. At the same time, we are facing increasing availability of data at multiple scales, ranging from leaf-level measurements (e.g., gas exchange), tree-level measurements (e.g., sap flow and tree growth, dendroecology), ecosystem-level measurements (eddy covariance towers, UAVs, aircrafts) to synoptic Earth observations from space. These are opening to utilize machine learning algorithms and data-driven modelling as an alternative to process-based approaches. Though sometimes very successful in fitting observations, they are often plagued by lack of interpretability and physical consistency. However, recent developments like interpretable machine learning, physics-aware regression or causal inference models are attempts to mitigate these issues.
This session invites studies that improve our overall understanding of biosphere-atmosphere interactions by combining observations at different temporal and spatial scales as well as their seamless integration into modelling strategies. In addition to empirical multi-scale observations of carbon, energy and water fluxes, we invite research that explores data-driven diagnostics and constraints for model evaluation (e.g., Emergent Constraints), data-driven parameterizations in mechanistic models (e.g., Earth system models) and other developments of machine-learning / hybrid modelling strategies (e.g., fusion of data-driven approaches and mechanistic models, interpretable machine learning, causal inference) for an integrated understanding of carbon, energy and water fluxes across scales.

Orals: Thu, 27 Apr | Room 2.17

On-site presentation
Christiane Werner, Laura Meredith, and S. Nemiah Ladd and the B2WALD Team

Ecosystem response to drought present a complex interplay between regulation at the leaf, plant, and ecosystem scale as well as soil-plant-atmosphere interactions and feedbacks. While single leaf or tree fluxes can be continuously measured, other processes such as changes in below ground carbon allocation or shifts in root-water uptake depth under drought are difficult to assess.

To trace ecosystem scale interactions, we implemented a whole-ecosystem labelling approach in the Biosphere 2 Tropical Rainforest. In the Biosphere 2 Water, Atmosphere, and Life Dynamics (B2-WALD) experiment, we applied an ecosystem scale drought and tracing carbon allocation and dynamics of volatile organic compounds (BVOC), CO2 and H2O fluxes and their isotopes from leaf, root, trunks, soil and atmospheric scales.

Drought sequentially propagated through the vertical forest strata, with a rapid increase in vapor pressure deficit in the top canopy layer and early dry-down of the upper soil layer but delayed depletion of deep soil moisture. Gross primary production (GPP), ecosystem respiration (Reco), and evapotranspiration (ET) declined rapidly during early drought and severe drought. Interactions between plants and soil led to distinct patterns in the relative abundance of atmospheric BVOC concentrations as the drought progressed, serving as a diagnostic indicator of ecosystem drought stress.

Ecosystem 13CO2-pulse-labeling showed that drought enhanced the mean residence times of freshly assimilated carbon, indicating down-regulation of carbon cycling velocity and delayed transport form leaves to trunk and roots. Despite reduced ecosystem carbon uptake and total VOC emissions, plants continued to allocate a similar proportion of fresh carbon to de novo VOC synthesis, as incorporation of 13C into both isoprene and monoterpenes remained high.

A 2H-labelled deep-water label during severe drought provided a unique opportunity to evaluate transit times and legacy effects during the recovery phase. Combined with an in situ approaches allowed to monitor the isotopic composition in soils, tree xylem and transpiration at high temporal resolution. Drought-sensitive canopy trees strongly reduced water fluxes during early drought, while drought-tolerant trees increased their relative contribution to total water flux. Interestingly, all deep-rooted canopy trees taped into deep-water reserves, but spared deep water reserves until severe drought and exhibited long transit times of 2-6 weeks until d2H-labelled water was transpired. This was partially due to stem water refill exceeding the onset of transpiration after drought release.

These data highlight the importance of quantifying drought impacts on forest functioning beyond the intensity of (meteorological) drought, but also taking dynamics response of hydraulic regulation of different vegetation and soil compounds into account. Such data set can be used for carbon and water partitioning from the metabolic to ecosystem scale and help disentangling belowground processes to better parameterize models.

Werner et al. 2021, Science 374, 1514 (2021), DOI: 10.1126/science.abj6789

How to cite: Werner, C., Meredith, L., and Ladd, S. N. and the B2WALD Team: Understanding drought dynamics of carbon and water fluxes from leaf to ecosystem scales in an experimental tropical forest, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9115,, 2023.

On-site presentation
Quentin Beauclaire, Simon De Cannière, François Jonard, and Bernard Longdoz

One of the most efficient ways to estimate carbon assimilation at the ecosystem scale is based on the observations of sun-induced chlorophyll fluorescence (SIF) from satellites coupled with empirical relationships between SIF and gross primary production (GPP). However, there is still a lack of knowledge about the influence of physiological and environmental factors on these relationships. Recently, a process-based light response (MLR) model was developed from the perspective of the light reactions of photosynthesis to mechanistically determine the ecosystem photosynthetic activity from SIF measurements above the canopy. In addition to GPP, the MLR model can also be used to predict the latent heat (LE) flux when coupled with a stomatal conductance model like the unified stomatal optimality (USO) model. The goal of this research is to calibrate and test the MLR-USO model, as well as evaluate its performance at the field scale.

The MLR-USO model was used to estimate GPP and LE at the ICOS station in Lonzée, Belgium (BE-Lon), which was equipped with a field spectrometer (Fluorescence box –FLOX) installed above a winter wheat crop to measure SIF between February and July 2022. The application and evaluation of the coupled MLR-USO model at the canopy scale requires (i) to determine the value of the MLR model parameters, (ii) to calibrate the USO model on previous cropping seasons of winter wheat at BE-Lon, (iii) to calculate the broadband (640-850 nm) SIF emitted by all photosystems II (PSII) from in-situ measurements of directional observed SIF at 760 nm, and (iv) to compare GPP and LE estimates from the MLR-USO model with eddy covariance (EC) flux tower GPP and LE.

The results of this study demonstrate that SIF at the canopy scale captured the dynamics of both water and carbon exchanges over a wide range of environmental conditions, including light and water limiting conditions. Consequently, the MLR-USO model performed well when compared to EC data (RMSEGPP=6.56 μmolm-2s-1 / RMSELE=22.79 Wm-1 at half hourly timescales and RMSEGPP=4.61 μmolm-2s-1 / RMSELE=26.75 Wm-1 at daily timescales). These results support the use of this integrated model containing both a stomatal conductance and a photosynthesis module as an important step towards a better understanding and quantification of carbon and water fluxes at the ecosystem scale, providing also key information for the interpretation of satellite-based SIF.

How to cite: Beauclaire, Q., De Cannière, S., Jonard, F., and Longdoz, B.: Mechanistic modelling of gross primary production and latent heat flux using SIF observations in different water and light limitation conditions, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5508,, 2023.

On-site presentation
Mariette Vreugdenhil, Susan Steele-Dunne, Xu Shan, Thomas Kaminski, Mika Aurela, Emanuel Bueechi, Wouter Dorigo, Wolfgang Knorr, Juha Lemmetyinen, Nemesio Rodriguez-Fernandez, Marko Scholze, Tea Thum, and Mathew Williams

Combining data from in situ measurements, remote sensing and models can provide new insights on global vegetation dynamics, specifically on the role of vegetation in the carbon and water cycles. Here we will demonstrate the benefits of combining Metop Advanced SCATterometer (ASCAT) C-band radar backscatter observations with in-situ and model data for monitoring vegetation dynamics and constraining parameters in terrestrial carbon stock and flux simulations. 

The slope of the relation between backscatter and incidence angle of Metop ASCAT data is sensitive to vegetation dynamics over the Amazon region and North-American grasslands, as demonstrated in previous studies by Petchiappan et al. (2022) and Steele-Dunne et al. (2018).  Here we use the slope in combination with in-situ observations to analyze vegetation dynamics over the ICOS site in Sodankyla. Results from this boreal forest region in Northern Finland show that slope dynamics are influenced by freezing temperatures and snow, hindering monitoring of vegetation dynamics during these times. During periods without freezing temperatures and snow, the slope reveals phenological changes both in terms of seasonal changes and anomalies. During the 2018 drought, positive anomalies in slope were found, 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 Vegetation Optical Depth.

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. 

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., 2020. Direct and seasonal legacy effects of the 2018 heat wave and drought on European ecosystem productivity. Science Advances 6, eaba2724.

Petchiappan, A., Steele-Dunne, S.C., Vreugdenhil, M., Hahn, S., Wagner, W., Oliveira, R., 2022. The influence of vegetation water dynamics on the ASCAT backscatter-incidence angle relationship in the Amazon. Hydrology and Earth System Sciences 26, 2997–3019.

Steele-Dunne, S.C., Hahn, S., Wagner, W., Vreugdenhil, M., 2019. Investigating vegetation water dynamics and drought using Metop ASCAT over the North American Grasslands. Remote Sensing of Environment 224, 219–235.

How to cite: Vreugdenhil, M., Steele-Dunne, S., Shan, X., Kaminski, T., Aurela, M., Bueechi, E., Dorigo, W., Knorr, W., Lemmetyinen, J., Rodriguez-Fernandez, N., Scholze, M., Thum, T., and Williams, M.: Seven Frozen Trees in Sodankyla: Relating ASCAT slope to water and carbon processes over a Boreal forest using in-situ, model and reanalysis data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12005,, 2023.

On-site presentation
Mallory Barnes, Matthew Dannenberg, Steven Kannenberg, Rubaya Pervin, and Natasha MacBean

Drylands have tightly coupled water and carbon cycles due to persistent water scarcity, making them valuable systems for understanding coupled ecohydrological and biogeochemical processes. In addition, dryland ecosystems contribute significantly to the interannual variability of the terrestrial carbon sink. To better characterize dryland carbon dynamics, we present DryFlux, a machine learning upscaled product based on a dense network of eddy covariance sites in the North American Southwest. This product combines in-situ fluxes with remote sensing and meteorological data to estimate gross primary productivity in drylands while explicitly accounting for water limitation during the model development process. DryFlux outperforms existing products in capturing interannual and seasonal variation in carbon uptake when used globally. We specifically explore how machine learning techniques can accurately upscale fluxes at multiple spatial (1 km and 9 km) and temporal (daily, weekly, monthly) scales to find the best resolution for capturing spatial and temporal heterogeneity in carbon and water fluxes. In addition, we discuss how remotely sensed soil moisture from satellites can help capture biogeochemical 'hot spots' and 'hot moments' in drylands. Our findings can help us better understand dynamic carbon fluxes in drylands, as well as the spatiotemporal resolution needed to resolve water-carbon dynamics in these and other systems. Machine learning methods that explicitly incorporate water limitation in model development can contribute to a more comprehensive understanding of carbon, energy, and water fluxes at multiple scales.

How to cite: Barnes, M., Dannenberg, M., Kannenberg, S., Pervin, R., and MacBean, N.: An Ecohydrologically-informed Machine Learning Approach for Understanding Dryland Carbon Dynamics, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16789,, 2023.

On-site presentation
Victor Boussange, Pau Vilimelis Aceituno, and Loïc Pellissier

In contrast to purely data-driven statistical models, process-based ecosystem models have the potential to extrapolate beyond observed dynamics and predict their response to global change. Yet, the predictive power of process-based ecosystem models has been limited in practice because of issues with the estimation of the model parameter values and because of model inaccuracies. While inverse modelling techniques can make use of observation data to improve the estimation of parameters and, combined with model selection techniques, improve model inaccuracies, process-based ecosystem models are dependent on numerous parameters, are strongly nonlinear, and their numerical integration is computationally expensive. These characteristics, together with the nature of available observation data that may consist of shallow, incomplete and noisy time series, as well as the difficulty to obtain the model sensitivity to the parameters, have challenged the use of inverse modelling and model selection techniques in ecosystem modelling. Here, we present a machine learning (ML) framework relying on a segmentation method combined with state-of-the-art optimizers and automatic differentiation to perform inverse ecosystem modelling. The segmentation method regularizes the likelihood landscape, while the latter techniques, traditionally used in the field of artificial intelligence, greatly improve the efficiency of the inference process. We introduce PiecewiseInference.jl, a software package written in the Julia programming language that implements the ML framework, and evaluate its performance in recovering the dynamics of simulated chaotic food-webs. We show that it can efficiently estimate parameters and subsequently provide reliable forecasts based on noisy, incomplete and independent time series. Using model selection techniques, we further show that the ML framework can provide accurate statistical support for the true generating model among several candidates. We plan on utilizing PiecewiseInference.jl with long-term fish and invertebrate abundance time series to better understand the dynamical processes regulating marine communities in the Northeast Atlantic and Mediterranean Sea.

How to cite: Boussange, V., Vilimelis Aceituno, P., and Pellissier, L.: PiecewiseInference.jl: a machine learning framework for inverse ecosystem modelling, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5195,, 2023.

Virtual presentation
Doaa Aboelyazeed, Chonggang Xu, Forrest M. Hoffman, Alex W. Jones, Chris Rackauckas, Kathryn Lawson, and Chaopeng Shen

Photosynthesis plays an important role in carbon, nitrogen, and water cycles. Ecosystem models for photosynthesis are characterized by many parameters that are obtained from limited in-situ measurements and applied to the same plant types. Previous site-by-site calibration approaches could not leverage big data and faced issues like overfitting or parameter non-uniqueness. Here we developed a programmatically differentiable (meaning gradients of outputs to variables used in the model can be obtained efficiently and accurately) version of the photosynthesis process representation within the Functionally Assembled Terrestrial Ecosystem Simulator (FATES) model. This model is coupled to neural networks that learn parameterization from observations of photosynthesis rates. We first demonstrated that the framework was able to recover multiple assumed parameter values concurrently using synthetic training data. Then, using a real-world dataset consisting of many different plant functional types, we learned parameters that performed substantially better and dramatically reduced biases compared to literature values. Further, the framework allowed us to gain insights at a large scale. Our results showed that the carboxylation rate at 25°C (Vc,max25), was more impactful than a factor representing water limitation, although tuning both was helpful in addressing biases with the default values. This framework could potentially enable a substantial improvement in our capability to learn parameters and reduce biases for ecosystem modeling at large scales.

How to cite: Aboelyazeed, D., Xu, C., Hoffman, F. M., Jones, A. W., Rackauckas, C., Lawson, K., and Shen, C.: A differentiable ecosystem modeling framework for large-scale inverse problems: demonstration with photosynthesis simulations, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16693,, 2023.

On-site presentation
Álvaro Moreno-Martínez, Laura Martínez-Ferrer, Jordi Muñoz-Marí, Emma Izquierdo-Verdiguier, John S. Kimball, Steven W. Running, Nicholas Clinton, and Gustau Camps-Valls

Carbon captured via photosynthesis by vegetation is known as gross primary production (GPP). It is an important variable related to climate regulation and determines ecosystem carbon sources and sinks. GPP is routinely estimated globally by operational algorithms that combine remote sensing data at coarse spatial scales (e.g., MODIS, 500 m) and meteorological information. The need for global high-resolution operational products arises from the requirement of capturing GPP variability, which co-occurs at finer resolutions and over large areas. These specifications demand gap-free remote sensing data to obtain continuous maps, high spatial and temporal resolution, and realistic uncertainty quantification. Machine learning (ML) methods are widely used but sometimes do not fit real-world physics restrictions. Therefore, we propose a physics-aware machine learning methodology that combines 1) high spatial resolution spectra at 30m and gap-free observations derived from blending Landsat and MODIS with the HISTARFM algorithm, 2) meteorological information, and 3) in situ eddy covariance GPP estimates as reference data. The ML model further incorporates an extra regularizer that constrains the GPP estimates for improved consistency with ancillary data and covariates closely related to photosynthesis (e.g., SIF). Moreover, we rely on the HISTARFM methodology to provide well-calibrated data uncertainty estimates, which allows us to yield both epistemic and aleatoric uncertainty for the GPP estimates. The processing pipeline is fully implemented in Google Earth Engine (GEE), allowing us to estimate carbon fluxes over Europe at 30m. The methodology enables more precise and real-world carbon studies and opens the door to deriving other key fluxes at an unprecedented spatiotemporal resolution. 



How to cite: Moreno-Martínez, Á., Martínez-Ferrer, L., Muñoz-Marí, J., Izquierdo-Verdiguier, E., Kimball, J. S., Running, S. W., Clinton, N., and Camps-Valls, G.: Physics-Aware Machine Learning for Carbon Fluxes at High Spatio-Temporal Resolution and Scales, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-7439,, 2023.

On-site presentation
Paul Levine, Eren Bilir, Anthony Bloom, Renato Braghiere, Caroline Famiglietti, Alexandra Konings, Marcos Longo, Shuang Ma, Elias Massoud, Victoria Meyer, Alexander Norton, NIcholas Parazoo, Gregory Quetin, Luke Smallman, Mathew Williams, John Worden, Matthew Worden, Sarah Worden, and Yan Yang

Predicting the fate of the terrestrial ecosystems and their role in the Earth system requires a quantitative and mechanistic understanding of carbon, water, and energy exchanges between the land surface and the atmosphere. While the current generation of land surface models show skill in representing many ecosystem processes, they largely disagree in the integrated response of the terrestrial biosphere to climatic change. These disagreements may be reconciled by confronting models with the diverse and expanding suite of Earth system observations in order to better constrain the underlying processes. In light of this goal, we have implemented substantial developments to the CARbon DAta-MOdel FraMework (CARDAMOM)—a data assimilation system that optimally estimates parameters of a parsimonious ecosystem model—which expand its original scope as a diagnostic tool for estimating carbon states and fluxes into a system that can infer and predict the response of carbon, water and energy cycles to climate and CO2 concentrations at seasonal-to-decadal timescales. CARDAMOM 3.0 retains all functionality and model structures of previous versions, but now features a flagship model which includes coupled carbon, water, and energy cycles, along with semi-mechanistic representations of photosynthetic assimilation, allocation, phenology, autotrophic and heterotrophic respiration, snow and cold-weather processes, and soil hydrology. Additionally, the underlying framework was substantially updated in order to facilitate community use of CARDAMOM by simplifying the interface and increasing the ease with which users can integrate new observations and develop new model structures. With these new developments, CARDAMOM 3.0 provides a versatile tool for applying information from a broad array of Earth observation data to investigate carbon, water, and energy cycles and their responses to climate and atmospheric CO2 across the full range of terrestrial ecosystems, from leaf level to continental scales.

How to cite: Levine, P., Bilir, E., Bloom, A., Braghiere, R., Famiglietti, C., Konings, A., Longo, M., Ma, S., Massoud, E., Meyer, V., Norton, A., Parazoo, N., Quetin, G., Smallman, L., Williams, M., Worden, J., Worden, M., Worden, S., and Yang, Y.: Constraining carbon, water, and energy cycling using diverse Earth observations across scales: the CARDAMOM 3.0 approach, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10918,, 2023.

On-site presentation
Emanuel Dutra, Francisco Lopes, Jean-Christophe Calvet, Bertrand Bonan, Anna Agusti-Panareda, Souhail Boussetta, and Martin Jung

Biogenic fluxes play a fundamental role in the carbon cycle and are crucial for the land-surface water and energy cycles. These three cycles, water/energy/carbon, are coupled and interact on time-scales ranging from minutes to centuries. Among different aspects of the processes involved, Land Use and Land Cover (LULC) are extremely relevant in the estimation of biogenic carbon. Moreover, the errors found in the model’s representation of LULC effects on the lower troposphere have also been shown to limit the progress in weather and climate predictability. In this work we evaluate different configurations of two land surface models: the ECMWF ECLand and Meteo-France ISBA within the SURFEX modelling platform. The evaluation is focused on the surface energy, water and carbon fluxes using FLUXCOM as reference, as well as land surface temperature using LSA SAF satellite product. The surface offline simulations evaluation identified the added value of a revised land cover and Leaf Area Index (LAI) in ECLand in terms of Gross Primary Production (GPP) when combined with a model configuration using the Farquhar photosynthesis model. The results also suggest that time-varying LAI, prescribed in ECLand and via data assimilation in SURFEX are relevant to GPP estimates during large-scale extreme events. Limitations in the evaluation of Net Ecosystem Exchanges and terrestrial respiration arising from model uncertainties, as well as in the reference data used, suggests that flux adjustments are paramount to mitigate biases in global CO2 analysis. Finally, coupled atmosphere weather forecasts with the ECMWF model show a clear improvement of the 2-meter temperature in Eurasia during spring following the revised land cover and LAI with a negative impact during summer in the tropics, which requires further developments.

This work was developed in the framework of the CoCO2 project that has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 958927.

How to cite: Dutra, E., Lopes, F., Calvet, J.-C., Bonan, B., Agusti-Panareda, A., Boussetta, S., and Jung, M.: Evaluation of global water, energy, and carbon fluxes in ECLand and ISBA models, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9819,, 2023.

Posters on site: Fri, 28 Apr, 14:00–15:45 | Hall A

Stephen Ingram, Yann Salmon, Anna Lintunen, Teemu Hölttä, Timo Vesala, and Hanna Vehkamaki

Xylem sap exists in a state some have described as “doubly” metastable[1]: liquid water is transported from root to leaf under negative pressure, and, in some climates, below its freezing point. Sub-100 nm nanobubbles may be injected into the xylem liquid through pit membranes[2], becoming coated with Phospho- and Glycolipids in the process. Their surface properties, and therefore fate within the tree hydraulic system, remain largely unexplored.

In this work, we have used molecular dynamics simulations to produce surface tension – area isotherms of biologically relevant lipid monolayers, as a function of both temperature and negative pressure (i.e. dynamic surface tensions).

We find that glycolipid monolayers resist expansion proportionally to the rate of expansion[3]. Their surface tension increases with the tension applied, stabilising the bubble with respect to embolism. In contrast, a typical phospholipid rapidly condenses into more dense lamellar-like phase, rendering it highly resistant to tensions as high as -3.5 MPa. Mixed monolayers of the two exhibit hybrid behavior, as the glycolipids' larger head group disrupts the more ordered phase of the phospholipid. Finally, it is observed that increasing temperature also increases surface tension, at a given surface area.

[1] Lintunen, A., Hölttä, T., & Kulmala, M. (2013). Anatomical regulation of ice nucleation and cavitation helps trees to survive freezing and drought stress. Scientific Reports, 3, 1–7.

[2] Schenk, H. J., Steppe, K., & Jansen, S. (2015). Nanobubbles: a new paradigm for air-seeding in xylem. Trends in Plant Science, 20(4), 199–205.

[3] Ingram, S., Salmon, Y., Lintunen, A., Hölttä, T., Vesala, T., & Vehkamäki, H. (2021). Dynamic Surface Tension Enhances the Stability of Nanobubbles in Xylem Sap. Frontiers in Plant Science, 12.

How to cite: Ingram, S., Salmon, Y., Lintunen, A., Hölttä, T., Vesala, T., and Vehkamaki, H.: Dynamic Surface Tensions of Nanobubbles in Plant Xylem: When are they Stable?, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2012,, 2023.

Teresa E. Gimeno, Ane Umerez, Usue Pérez-López, Jon Miranda-Apodaca, Javier Porras, and Guillermo López-Castro

Plant water-use efficiency (WUE) describes the intimate link between the carbon and water cycles. WUE can be estimated using multiple methodologies concerning different spatial and temporal scales, but empirical evidence has shown that these estimates do not always agree. Two of the most widely used methodologies to estimate WUE are measurements of the ratio of photosynthesis to stomatal conductance to water, using gas-exchange, and analyses of the carbon isotopic composition (δ13C) of plant material, most often measured on plant tissues (leaves and woody stems mainly), reflecting the signal of the plant physiological status all along the organ ontogeny. In addition, in tall trees, δ13C varies greatly among leaves and thus individual measurements cannot capture whole-tree physiological status. In contrast, analyses of the phloem δ13C collected at the base of the trunk should reflect the whole-tree physiological performance. To test this novel approach under contrasting climate change scenarios, we estimated WUE: from measurements of gas-exchange, from δ13C of leaf and woody tissues, as well as from δ13C of phloem samples collected along the whole plant pathway. We measured gas-exchange and collected samples for analyses of δ13C from European beech (Fagus sylvatica) saplings grown under controlled conditions and from adult trees in the field. Saplings were subjected to four climate change scenarios, resulting from a combination of two atmospheric CO2 levels and two watering regimes. In the field, we measured WUE on adult beech trees during the unusually hot and dry growing season of 2022. Preliminary results show that phloem δ13C could serve as a good proxy of whole-plant WUE, provided that the internal leaf conductance is incorporated into the calculations.

How to cite: Gimeno, T. E., Umerez, A., Pérez-López, U., Miranda-Apodaca, J., Porras, J., and López-Castro, G.: Tracking phloem carbon isotopic composition to reconcile water-use efficiency estimates across scales in beech trees, under future climate change scenarios, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-13394,, 2023.

Luana Krebs, Susanne Burri, Iris Feigenwinter, Philip Meier, Mana Gharun, and Nina Buchmann

Forest ecosystems play an important role in the global carbon (C) cycle by sequestering a large fraction of anthropogenic carbon dioxide (CO2) emissions and by acting as important methane (CH4) sinks. Nevertheless, how the forest C sink will respond to climate change is still largely unknown. The forest-floor GHG flux is one of the major processes to consider when determining the C balance of forests. Although winter fluxes are essential to determine the annual C budget of forests, there have been very few studies that have examined long-term, year-round forest-floor GHG fluxes in high elevation forests. Especially during snowy periods, forest floor GHG fluxes are difficult to measure and are therefore often missing from studies. In this study, we used four years of forest-floor CO2, CH4 and nitrous oxide (N2O) fluxes (2017, 2020, 2021 and 2022; N2O not available for years 2021, 2022). Fluxes were measured year-round with four automatic chambers at the ICOS Class 1 station Davos, located in a subalpine coniferous forest in Switzerland. We applied random forest models to investigate the environmental drivers and to gap-fill the flux time series for calculating annual sums of CO2 and CH4 fluxes. More specifically, the aims of this study were to i) investigate the seasonal and annual variations in climate variables and forest-floor CO2, CH4 and N2O fluxes; ii) evaluate the environmental drivers of forest-floor GHG fluxes including the effect of snow cover and snow melt, and iii) calculate annual budgets of the forest-floor GHG fluxes. We hypothesized that the main drivers of soil CH4, CO2 and N2O fluxes are soil temperature and soil moisture (e.g., higher CH4 uptake in warmer and drier soils). Additionally, we hypothesized that winters with little snow and early melting can lead to reduced soil moisture later in the year, which could lead to higher CH4 uptake. First results show that the forest-floor CO2 efflux generally follows soil temperature. However, the dynamics in the CO2 efflux cannot be entirely explained by soil temperature, e.g., a large increase in CO2 efflux in 2022 compared to other years. Furthermore, we found that the forest-floor is a consistent sink for CH4, however with large short-term dynamics, and that the magnitude of the sink is mainly driven by air temperature and snow cover. N2O fluxes are very low, i.e., probably below the detection limit of our method, which is why we consider them negligible for the overall forest-floor GHG budget at our site.

How to cite: Krebs, L., Burri, S., Feigenwinter, I., Meier, P., Gharun, M., and Buchmann, N.: Year-round forest-floor greenhouse gas fluxes in a subalpine coniferous forest: drivers and budgets, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16830,, 2023.

Maren Dubbert, Shrijana Vaidya, Adrian Dahlmann, Marten Schmidt, Gernot Verch, Michael Sommer, Jürgen Augustin, and Mathias Hoffmann

Improved agricultural practices increasing the water use efficiency (WUE), reducing greenhouse gas emissions (GHG) and/or improving atmospheric C sequestration rates within the soil are crucial for an adaptation and/or mitigation to the global climate crisis. However, processes driving water (H2O), carbon (C) and GHG fluxes within the soil-plant-atmosphere continuum of agricultural used landscapes are complex and flux dynamics differ substantially in time and space. Hence, to upscale and evaluate the effects/benefits of any new agricultural practice aiming towards improving WUE, soil C sequestration and/or GHG emissions, accurate and precise information on the complex spatio-temporal H2O, C and GHG flux pattern, their drivers and underlying processes are required.

Current approaches to investigate H2O, C and GHG (CH4, N2O and CO2) dynamics as well as underlying processes driving them, are usually laborious and have to choose between high spatial or temporal resolution due to methodological constraints. On the one hand, often used eddy covariance systems are not suitable to account for small scale spatial heterogeneities, despite growing evidence of their importance. On the other hand, chamber systems either lack temporal resolution (manual chambers) or strongly interfere with the measured system (static automatic chambers). Hence, none of these systems enable a proper upscaling and evaluation of effects/benefits of new farming practices for WUE, C sequestration and GHG emissions at especially heterogeneous agricultural landscapes, such as present within inter-alia the also globally widespread hummocky ground moraine landscape of NE-Germany.

In an effort to overcome this, a novel, fully automated robotic field sensor platform was established and combined with an IoT network and remote sensing approaches. Here, an innovative, continuously operating automated robotic field sensor platform is presented. The platform was mounted on fixed tracks, stretching over an experimental field (size) which covers three different, distinct soil types. It carries multiple sensors to measure GHG and water vapor concentrations and isotope signatures of d18O and dD of water vapor. Combined with two chambers which can be accurately positioned in three dimensions at the experimental field below, this system facilitates to detect small-scale spatial heterogeneity and short-term temporal variability of H2O, C and GHG flux dynamics as well as crop and soil conditions over a range of possible experimental setups. The automated, continuous estimation of d18O and dD of evapotranspiration further provides the basis to partition water fluxes alongside the flux based partitioning of C and GHG fluxes. This particularly promotes to assess not only ecosystem but component specific water use efficiencies. Hence, this platform produces a detailed picture of H2O, C and GHG dynamics across different treatments and crop cycles, with a high-degree of accuracy and reproducibility.

How to cite: Dubbert, M., Vaidya, S., Dahlmann, A., Schmidt, M., Verch, G., Sommer, M., Augustin, J., and Hoffmann, M.: AGROFLUX: An innovative sensor platform to study high-frequency responses in water, carbon and greenhouse gas fluxes in a complex arable landscape, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2321,, 2023.

Pia Gottschalk, Aram Kalhori, Zhan Li, Christian Wille, and Torsten Sachs

The cropland carbon (C) balance at regional scale still contains high uncertainties not the least due to the problem of up-scaling C fluxes of temporarily and spatially highly divers ecosystems. The C-exchange between the terrestrial ecosystem and the atmosphere constitute the largest and most uncertain flux of the cropland C balance, as opposed to C import from organic (manure) and C export through harvest which are lower and less uncertain.

Combining satellite data with local eddy covariance CO2-flux data is commonly used to up-scale the C-exchange signal from point to regional scale across global ecosystems. Low spatial resolution products like MODIS limit their applicability and accuracy to larger homogeneous areas involving a high degree of uncertainty rather than detecting and tracing highly dynamic (farm-)field scale CO2-fluxes from space. We are using eddy-covariance CO2-flux data of an arable field in conjunction with Sentinel-2 derived vegetation indices (VI) to assess the ability of the satellite data to monitor daily net-ecosystem exchange (NEE), gross-primary productivity (GPP) and ecosystem respiration (Reco) based on a matched footprint. Simple linear regression models are built to test the ability of a range of VIs (NDVI, GNDVI, EVI, EVI2, SAVI, MNDWI, NDWI, SR, S2REP) to monitor and predict CO2-exchange for croplands. We analyze the correlation between measured CO2-fluxes and VIs over the course of the growing seasons to assess the suitability and accuracy of the VIs along the phenological year. We present a single site analysis to zoom into short-comings of this approach and how the satellite signal relates to vegetation CO2-exchange. VIs generally show a high variability in their predictive power. Still, results suggest a similarly high accuracy as mechanistic modelling approaches for suitable VIs, e.g. the cumulative C-exchange (NEE) of winter wheat of one growing season based on NDVI and GNDVI is over- and under-estimated by only 33 (15%) and 41 (18%) g C m-2 respectively.

How to cite: Gottschalk, P., Kalhori, A., Li, Z., Wille, C., and Sachs, T.: Monitoring daily cropland CO2-exchange at field scale with Sentinel-2 satellite imagery, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11596,, 2023.

Milagros Rodriguez-Caton, Ulrike Seibt, Johen Stutz, Nicholas Parazoo, Sol Cooperdock, Julia Bigwood, Mukund Palat Rao, Zoe Pierrat, Christopher YS Wong, Diego Dierick, Orlando Vargas, and Troy Magney

Tropical forests are responsible for approximately one third of the global terrestrial carbon dioxide uptake. However, the ability of tropical forests to continue to sequester carbon is threatened by climate change. Some species may adapt and become more dominant while less resilient species may not be able to adapt. These changes may ultimately impact overall ecosystem carbon gain. A better understanding of species specific traits related to leaf physiology can potentially help us predict how forest carbon uptake might change in the future. Additionally, to gain a wholistic understanding of forests carbon assimilation and tree stress it is important that we make simultaneous and complementary measurements at different temporal and spatial scales under different climatic conditions. Here we combine leaf-, canopy- and ecosystem- scale measurements to assess plant-environment interactions in the tropical forest of Costa Rica (Tower 2 of La Selva Biological Station). At the leaf scale, we evaluated seasonal and species-specific changes in chlorophyll fluorescence and reflectance-based vegetation indices [the Photochemical Reflectance Index (PRI), the Chlorophyll-Carotenoid Index (CCI) and Normalized Difference Vegetation Index (NDVI)]. We collected sun-exposed leaves from six tree species within the flux tower footprint: Pentaclethra macroloba, Virola koschnyi, Virola sebifera, Goethalsia meiantha, Sacoglottis trichogyna, and Warszewiczia coccinea. Preliminary results for the first sampled years (2022-2023) show that P. macroloba, the dominant species in this forest, has likely the highest photosynthetic capacity due to its higher electron transport rates (ETR), followed by V. koschnyi, the second most dominant. Using different metrics of photosynthetic activity, we found that most species do not show photosynthetic seasonality. However, one species, V. koschnyi, showed decreased reflectance in the visible part of the spectrum during the wetter season (35-45%), indicating increased pigment concentration and, likely, increased photosynthetic activity. G. meiantha had the lowest ETR of all species, as well as the lowest PRI, CCI and NDVI, especially during the drier season, which is coincident with a visually unhealthy colouration during this season. Our results will help improve our understanding of how different species are responding to environmental stress, in particular to increased evaporative demand, ultimately advancing our knowledge of the tropical forest carbon cycle.

How to cite: Rodriguez-Caton, M., Seibt, U., Stutz, J., Parazoo, N., Cooperdock, S., Bigwood, J., Rao, M. P., Pierrat, Z., Wong, C. Y., Dierick, D., Vargas, O., and Magney, T.: Species-specific ecophysiology within a flux tower footprint in an evergreen wet tropical forest in Costa Rica, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-13425,, 2023.

Joachim Ingwersen, Arne Poyda, Pascal Kremer, and Thilo Streck

In Poyda et al. (2019), we determined the carbon balance of six cropland sites in Southwest Germany on the basis of half-hourly net ecosystem exchange (NEE) fluxes measured with the eddy covariance (EC) method over a period of eight years from 2010 to 2017. We came up with the finding, that the sites lost on average about one ton of carbon per hectare and year. This huge loss was surprising, because the sites have been used as croplands already over several decades, and one would expect that the sites should be close to steady-state. In April 2022, we performed a soil organic carbon inventory at one of the six sites, and compared it with carbon data collected in April 2009. The soil inventory data do not give any evidence for a carbon loss in the order of one ton of carbon per hectare and year. Based on the data collected in April 2009, the carbon stock size of the plowing horizon at that time was 39.1 t C ha-1 yr-1. The carbon stock size determined in April 2022 was in the same range and amounted 39.9 t C ha-1 yr-1 with a standard error of 1.9 t C ha-1 yr-1, what means that these data indicate that the carbon budget of the cropland is indeed in or close to steady-state. In Poyda et al. (2019), the NEE flux data were gap-filled with the widely used software tool REddyProc. In REddyProc, the user has the option to apply or not to apply the friction velocity  (u*) filter before gap-filling. In Poyda et al. (2019), the u* filter was applied before gap-filling. We reprocessed the data for the year 2016, the only year with no winter time gaps, because a methanol fuel cell supplied the station with additional power, without applying the u* filter. Without applying the u* filter the cumulated annual NEE was ‑2080 kg C ha-1. Applying the u* filter increased the NEE by 836 kg C ha-1 to -1244 kg C ha-1. We did the same comparison for NEE fluxes measured at the same site over the years 2019, 2020 and 2021, and we got a similar result. The application of the u* filter increased the mean NEE by +730 kg C ha-1. The cumulated NEE was -2832 kg C ha-1 without u* filter and -2102 kg C ha-1 with u* filter. This positive bias is in the range of the EC based derived carbon loss and is able to explain a major part of the suspected positive bias. We recommend all REddyProc users to check their NEE data for this phenomena simply by processing the data once with and once without enabling the u* filter and comparing both results.

How to cite: Ingwersen, J., Poyda, A., Kremer, P., and Streck, T.: The role of REddyProc's friction velocity filter in determining the carbon budget of croplands, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-17094,, 2023.


Eddy covariance techniques are widely used to measure the net exchange of greenhouse between the surface and the atmosphere, providing high resolution, instantaneous flux measures and long-term observations, which in turn allows more accurate assessments of the ecosystem’s state. However, gaps in eddy covariance time series reduce the statistical efficiency and increase bias estimates, hampering predictions of ecosystem function. Although, several imputation techniques have been proposed to overcome these difficulties, including Marginal Distribution Sampling (MDS), the standard method of FLUXNET, MDS has limitations for filling long gaps (weeks to months). In this study, we combine MDS and machine learning imputation techniques to fill an 18-year time series of carbon fluxes. Our objective was to evaluate whether Random Forest algorithms are able to fill long-gaps and detect seasonality, as well as to identify the best predictors of ecosystem exchange, gross primary productivity, and ecosystem respiration. The eddy covariance raw-data were obtained from an experiment in an upland semi-natural grassland in the Auvergne region of France that has been managed by continuous cattle grazing under low animal stocking rate.  After raw-data processing using EddyPro software, we applied the MDS technique to half-hour data to fill the short-gaps, and then used a Random Forest (RF) algorithm to daily data to fill longer gaps. The time series was split into a training and testing dataset, and all variables describing atmospheric conditions, solar radiation, and energy fluxes were used to predict C fluxes. Random Forest models with high R2 and low prediction error increases were used to impute the long-gaps.  The cross-validation between observed and predicted values in the test dataset obtained R2 of greater than 0.85 for all carbon flux variables. Our analysis also revealed that the daily carbon flux values could be estimated using the basic meteorological variables, i.e., air temperature, precipitation, atmospheric pression, friction velocity, and wind speed, but also by energy fluxes. Finally, the imputed dataset presented similar seasonality along the years, with the highest C sequestration and respiration in the summer and spring. These results highlight the value of machine learning techniques for producing robust, long-term eddy flux data time series.

How to cite: WINCK, B., BLOOR, J., and KLUMPP, K.: Random forest algorithm for long-gap imputation in Eddy Covariance data: a case study in an upland semi-natural grassland in the Auvergne region, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8031,, 2023.

Nicolas Behrens and Mana Gharun

Natural undisturbed peat lands act as a net carbon sink while drainage and subsequent aeration of the peat layers leads to oxidation of the organic material and the release of greenhouse gases. In Germany emissions from peatlands make up over 7 % of the country’s total annual carbon emissions. However, continuous observations with a high temporal resolution in German peatlands are still rather sparse. Due to the heterogeneity of peatland ecosystem characteristics and their relations to GHG fluxes, it is a major challenge to understand and model emissions across the wider category of peatlands. Some of the frequently used models (for example in interpolation of chamber-based measurements) are straightforward and easy to implement but leave potentially valuable information aside. In this study we use high-frequency (10Hz) CO2 and H2O exchange measured with eddy covariance at a drained bog, along with a suite of meteorological, hydrological and phenological measurements to disentangle the roles of biotic and abiotic variables in peatland CO2 emissions. The site is a highly drained but intact bog with a peat body of roughly 3 m and a groundwater depth of around 60 cm, located in northwestern Germany. Preliminary results collected with manual chambers show that methane fluxes are negligible. First, we test the model performance of the commonly used rectangular hyperbolic light response curve for GPP. We then extend this model by including a greenness index derived from a time series of daily Phenocam images, allowing us to evaluate the impact of biotic drivers and their seasonality. Similarly, for Reco we test the performance of the classical exponential Lloyd & Taylor model and modify it by accounting for the underlying hysteresis observed in the response of respiration to soil temperature changes, by including hydrological drivers such as soil moisture, precipitation and water table depth. Our results advance our capacity for understanding and predicting how peatland ecosystems respond and contribute to changes in the Earth´s future climate. 

How to cite: Behrens, N. and Gharun, M.: Disentangling biotic and abiotic drivers of CO2 flux in a drained German peatland using eddy covariance flux measurements and modelling techniques, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8466,, 2023.

Yan Sun, Daniel S. Goll, Yuanyuan Huang, Philippe Ciais, Ying-Ping Wang, Vladislav Bastrikov, and Yilong Wang

Global change ecology nowadays embraces ever-growing large observational datasets (big-data) and complex mathematical models that track hundreds of ecological processes (big-model). The rapid advancement of the big-data-big-model has reached its bottleneck: high computational requirements prevent further development of models that need to be integrated over long time scales to simulate the distribution of ecosystems carbon and nutrient pools and fluxes. Here we introduce a machine-learning acceleration (MLA) tool to tackle this grand challenge. We focus on the most resource-consuming step in terrestrial biosphere models (TBMs): the equilibration of biogeochemical cycles (spin-up), a prerequisite that can take up to 98% of the computational time. Through three members of the ORCHIDEE TBM family part of the IPSL Earth System Model, including versions that describe the complex interactions between nitrogen, phosphorus and carbon that do not have any analytical solution for the spin-up, we show that MLA reduced the computation demand by 77-80%  for global studies via interpolating the equilibrated state of biogeochemical variables for a subset of model pixels. Despite small biases in the MLA-derived equilibrium, the resulting impact on the predicted regional carbon balance over recent decades is minor. Our tool is agnostic to gridded models (beyond TBMs), compatible with existing spin-up acceleration procedures, and opens the door to a wide variety of future applications, with complex non-linear models benefit most from the computational efficiency.

How to cite: Sun, Y., Goll, D. S., Huang, Y., Ciais, P., Wang, Y.-P., Bastrikov, V., and Wang, Y.: Machine learning for accelerating process-based computation of land biogeochemical cycles, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10692,, 2023.

Jongho Kim and Sujong Jeong

Quantitative assessment of the carbon cycle for terrestrial ecosystem is significant to improve our understanding on climate change, as it absorbs about 30% of annual global anthropogenic CO2 emission. To refine the carbon flux estimation, we construct a data-based model named CArbon Simulator from Space (CASS), motivated by the representative Light Use Efficiency (LUE) model VPRM. The model simply estimates carbon flux with the information of air temperature, relative humidity, photosynthetically active radiation (PAR), Enhanced Vegetation Index (EVI), and Land Surface Water Index (LSWI). CASS construct the hourly NEP dataset in 250m resolution for the Seoul Metropolitan Area using refined datasets, including PAR from the HIMAWARI8, geostationary satellite. Notably, CASS does not have Plant function type (PFT) dependency by replacing empirical coefficients with the machine learning regressor. The result confirms the increased ability to capture the spatiotemporal variation at local scale for NEP especially in the urban area. Our refined estimation of carbon flux is expected to help understanding the role of terrestrial ecosystem in climate crsis.

How to cite: Kim, J. and Jeong, S.: Estimating high spatiotemporal terrestrial carbon flux using geostationary and polar-orbiting satellites: CArbon Simulator from Space (CASS), EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10842,, 2023.

Wenli Zhao, Alexander J. Winkler, Markus Reichstein, Rene Orth, and Pierre Gentine

Drought detection is crucial for water resources management and global food security. Drought is typically detected using empirical indices, such as the Palmer Drought Severity Index (PDSI). However, those indices lack objectivity and are therefore suboptimal. Machine learning has been proven to be a powerful tool to define objective and optimal regressions, especially in hydrology. Here, we developed a machine learning (ML) model using Long Short-Term Memory (LSTM), which include memory effects, to predict the Evaporative Fraction (EF), an indicator of water stress at the surface, based on FLUXNET2015 Tier 1 eddy-covariance dataset. Compared to the widely used PDSI, EF is a more direct drought index to indicate water stress conditions. The results show that, firstly, with some routinely available variables, e.g., precipitation, net radiation, air temperature, relative humidity and other static variables like, Plant Functional Type (PFT) and soil property, the model can capture the EF dynamics, especially during the dry season. Secondly, we found there were different LSTM memory lengths across different Plant Functional Types. This indicates different rooting depth and different plant water use strategies that regulate the time scales of droughts. Our results have important implications for future water stress estimation, e.g., drought detection, in order to obtain a more direct and more accurate estimate of water stress.

How to cite: Zhao, W., Winkler, A. J., Reichstein, M., Orth, R., and Gentine, P.: An objective estimate of water stress - going beyond PDSI, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9278,, 2023.

Phillip Papastefanou, Adriane Esquivel-Muelbert, Susanne Suvanto, Stefan Olin, Thomas Crowther, Mart-Jan Schelhaas, and Thomas Pugh

Making projections of ecological systems under environmental change is central to many disciplines. Process-based models aim to represent core ecological mechanisms governing ecosystem dynamics, which can then be valuable for projecting change under novel environmental conditions. Yet, as our ecological understanding evolves, updating parameter information can be challenging. In addition, classical statistical approaches to fitting functional relationships often miss the complexity of interacting, non-linear dynamics, which can limit the predictive capacity of models. As such, a growing body of work suggests that the integration of modern machine learning might help to improve the representation of key ecological dynamics within such process-based models.  Here, we present a case study, using machine learning to identify key relationships between relative growth, biomass and mortality compared to classical regression methods. Our results suggest that the inclusion of a deep neural network (DNN) into a “theory-driven” process based global vegetation model can greatly improve model predictions of patch level forest structure and vegetation dynamics. This hybrid approach offers both the benefits of interpretability and physically-realistic structure, combined with the depth of information contained in big datasets and the flexibility of model machine learning.

How to cite: Papastefanou, P., Esquivel-Muelbert, A., Suvanto, S., Olin, S., Crowther, T., Schelhaas, M.-J., and Pugh, T.: Improving ecosystem model development with machine learning: a full hybrid approach, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9948,, 2023.

David Hafezi Rachti, Christian Reimers, Guohua Liu, and Alexander J. Winkler

Climate change and extreme weather events have far reaching consequences for terrestrial ecosystems, particularly for vegetation phenology. However, the effects of meteorological variations on phenology are still not well understood, rendering phenology modeling a major challenge. Here we adapt explainable machine learning (ML) techniques from computer vision to investigate the role of meteorological variability and its multi-scale memory on phenology.
Specifically, we develop a modelling framework using convolutional neural networks trained on wavelet transformed key meteorological variables to predict vegetation greenness. The wavelet transformation of the meteorological time series (temperature, soil moisture, and shortwave radiation) yields two-dimensional images that reflect their different frequencies across a broad spectrum from multi-year variability to synoptic time scales. We use the green and red chromatic coordinate (GCC and RCC) from the ground-based PhenoCam network as proxies for the daily state of vegetation phenology. Additionally, to compensate for calibration artifacts across the sites, we use the satellite-based normalized difference vegetation index (NDVI) for normalisation.
Explainable ML techniques, such as Integrated Gradients, in combination with the wavelet images give us insight into the importance of the various meteorological factors as well as the length and timing of the weather events for the prediction of phenology. We present first results of our modelling framework and illustrate the effects of meteorological variability, with an emphasis on spring phenology, at different time scales. In particular, we use the interpretability of our model architecture to develop hypotheses and test them with manipulation experiments. In addition, we explore the model's ability to spatially extrapolate to unseen locations during training.
Such studies are important to understand the impact of climate change on the seasonal cycle of terrestrial ecosystems and to find out whether ML with explainable techniques can lead to a better understanding and thus improvements in modelling phenology.

How to cite: Hafezi Rachti, D., Reimers, C., Liu, G., and Winkler, A. J.: Predicting Vegetation Phenology using Machine Learning based on Wavelet Transform of Meteorological Drivers, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10141,, 2023.

Johanna Kranz, Matthias Forkel, Christian Bernhofer, Matthias Mauder, and Ronald Queck

Changes in plant phenology as for example earlier leaf unfolding and delayed autumn senescence can result in variations in the carbon and water cycle. Studies investigating the impact of phenological shifts on biophysical processes such as water availability are still limited. Due to the sensitivity of radar satellite observations to both, structural and dielectric properties of the scattering materials, microwave remote sensing offers the potential to analyse structural (i.e. canopy biomass) and physiological (i.e. water status) dynamics in vegetation.

Here, we aim to derive annual water dynamics of vegetation canopies from the Sentinel-1 C-band radar backscatter signal by removing the influence of vegetation structure on the backscatter seasonality. To decouple the phenology of vegetation structure from the moisture content dynamics, a semi-empirical backscattering model (Water Cloud Model, WCM) is combined with a canopy water balance model. The WCM aims to separate contributions of soil and vegetation to the total backscatter. When introducing physical parameters for vegetation structure like leaf area index (LAI) and and moisture like leaf fresh moisture content (LFMC) to describe the vegetation backscatter, the effect of the seasonal variability of both variables on the radar signal can be assessed. The canopy water balance model estimates interception and changes in the canopy saturation and storage capacity of the vegetation using precipitation and throughfall measurements. Both models are combined to iteratively estimate measures of vegetation moisture. To calibrate the two models, we use measurements of LFMC and of canopy interception for the Tharandt ecosystem site in Germany in 2022, which is part of the ICOS and FLUXNET network. The calibrated model is then used to analyse the individual effects of both vegetation descriptors, LAI and LFMC, by fixing either one and looking at the changes in the seasonality of the S1 signal. The combined use of both models will allow to remove the structural-related changes in the Sentinel-1 radar backscatter to finally retrieve vegetation water dynamics over larger areas.

How to cite: Kranz, J., Forkel, M., Bernhofer, C., Mauder, M., and Queck, R.: Analysing the sensitivity of Sentinel-1 SAR to vegetation water dynamics using a combined model approach, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12640,, 2023.

Maryna Lukach

The benefits of the application of weather radar observations for aeroecological research are already well known to the scientific community. The advantages of long-term polarimetric weather radar observations for the detection of bird and insect migration or estimation of their abundances are used by different teams all over the world. In this context, a correct, timely, and meaningful interpretation of polarimetric weather radar observations is an important part of these studies. This interpretation requires a well-developed technique that automates the recognition of separate classes in both spatial and temporal dimensions of the data.

The study presents a novel data-driven technique for identifying different classes in Quasi-Vertical Profiles (QVPs) and in Columnar Vertical Products (CVP) based on observations made by a dual-polarization Doppler weather radar. The top-down optimal clustering is applied to the detection and identification of aeroecological classes in the QVPs and CVPs. We demonstrate its application to the NCAS X-band dual-polarization Doppler weather radar (NXPol) data and the potential of its application to the C-band data of the Met Office radar network. This technique is generally applicable to similar multivariate data from other observational instruments and will improve quantitative observation and monitoring of biodiversity in the UK.

How to cite: Lukach, M.: Potential of AI classification of the weather radar observations for aeroecological research, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16397,, 2023.