ElGhawi et al. (2025) Hybrid‐Modeling of Land‐Atmosphere Fluxes Using Integrated Machine Learning in the ICON‐ESM Modeling Framework

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Journal of Advances in Modeling Earth Systems
• Year: 2025
• Date: 2025-12-01
• Authors: Reda ElGhawi, Christian Reimers, Reiner Schnur, Markus Reichstein, Marco Körner, Nuno Carvalhais, Alexander J. Winkler
• DOI: 10.1029/2025ms005102

Research Groups

Information not available from the abstract.

Short Summary

This paper develops Hybrid-JSBACH4, a novel hybrid modeling approach that integrates data-driven neural network parameterizations, trained on eddy-covariance flux measurements, into the mechanistic JSBACH4 land surface model. This integration significantly improves the simulation of land-atmosphere water and carbon fluxes by reducing biases in transpiration and gross primary production compared to the original JSBACH4.

Objective

• To develop and demonstrate a hybrid modeling approach that integrates data-driven flexible parameterizations, based on eddy-covariance flux measurements, into mechanistic land surface models (specifically JSBACH4), thereby improving the simulation of land-atmosphere water and carbon fluxes.

Study Configuration

• Spatial Scale: Site-specific (forest and grassland sites).
• Temporal Scale: Hourly.

Methodology and Data

• Models used: JSBACH4 (land component of ICON-ESM), Hybrid-JSBACH4 (integrating feed-forward neural networks), feed-forward neural networks (for parameterizations of stomatal conductance, maximum carboxylation rates, and maximum electron transport rates).
• Data sources: Eddy-covariance flux measurements (FLUXNET), original JSBACH4 output (for pre-training/emulation).

Main Results

• Hybrid-JSBACH4 successfully reconstructs original JSBACH4 parameterizations for stomatal conductance, maximum carboxylation rates, and maximum electron transport rates.
• The mean hourly residuals of transpiration (T) in Hybrid-JSBACH4 with respect to FLUXNET observations vary between -0.1 and 0.15 kg m⁻² h⁻¹ for forest and grassland sites, which is an improvement over JSBACH4 residuals varying between -0.3 and 0.2 kg m⁻² h⁻¹.
• The mean hourly residuals for Gross Primary Production (GPP) of Hybrid-JSBACH4 with respect to observations vary between -0.5 and 0.5 gC m⁻² h⁻¹ for forest and grassland sites, compared to original JSBACH4 residuals ranging between -1.0 and 0.5 gC m⁻² h⁻¹.
• Hybrid-JSBACH4 improves the representation of plant physiological responses and reduces biases in transpiration and GPP simulations under varying atmospheric dryness and water availability conditions.

Contributions

• Development of a novel hybrid modeling approach (Hybrid-JSBACH4) that effectively integrates data-driven flexible parameterizations (using neural networks) into a traditional mechanistic land surface model (JSBACH4).
• Demonstration of significant improvement in simulating land-atmosphere water and carbon fluxes (transpiration and GPP) by substantially reducing biases compared to the original mechanistic model.
• Provides a pathway for constructing observation-informed modeling of land surface processes, addressing the rigidity of traditional mechanistic parameterizations and enhancing model accuracy.

Funding

Information not available from the abstract.

Citation

@article{ElGhawi2025HybridModeling,
  author = {ElGhawi, Reda and Reimers, Christian and Schnur, Reiner and Reichstein, Markus and Körner, Marco and Carvalhais, Nuno and Winkler, Alexander J.},
  title = {Hybrid‐Modeling of Land‐Atmosphere Fluxes Using Integrated Machine Learning in the ICON‐ESM Modeling Framework},
  journal = {Journal of Advances in Modeling Earth Systems},
  year = {2025},
  doi = {10.1029/2025ms005102},
  url = {https://doi.org/10.1029/2025ms005102}
}