Biegel et al. (2025) Unrecognised water limitation is a main source of uncertainty for models of terrestrial photosynthesis

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

Identification

• Journal: Repository for Publications and Research Data (ETH Zurich)
• Year: 2025
• Date: 2025-12-01
• Authors: Samantha Biegel, Konrad Schindler, Benjamin Stocker
• DOI: 10.3929/ethz-c-000789603

Research Groups

Not explicitly stated in the provided text, but the study involved a network of globally distributed sites (N = 104).

Short Summary

This study investigates how models accounting for temporal structure impact the prediction of ecosystem photosynthesis (GPP), comparing mechanistic, memoryless deep learning (MLP), and recurrent neural network (LSTM) models. It finds that while both deep learning models outperform the mechanistic one, the LSTM leverages learned temporal dependencies to achieve lower error during periods of drought and frost, and performs better than MLP in dry environments.

Objective

• To investigate how models that account for temporal structure impact the prediction of ecosystem gross primary production (GPP) across diverse ecosystems and climates, with a focus on effects by low rooting zone moisture and freezing air temperatures.

Study Configuration

• Spatial Scale: Globally distributed network of 104 sites.
• Temporal Scale: Time-series measurements, covering periods affected by temporal patterns such as drought and frost.

Methodology and Data

• Models used:
  • Mechanistic, theory-based photosynthesis model
  • Memoryless Multilayer Perceptron (MLP)
  • Recurrent Neural Network (Long Short-Term Memory, LSTM)
• Data sources:
  • Time-series measurements of ecosystem fluxes (Gross Primary Production, GPP)
  • Meteorological variables
  • Remotely sensed vegetation indices

Main Results

• Both deep learning models (MLP and LSTM) outperform the mechanistic photosynthesis model.
• The overall performance of MLP and LSTM is similar, with an R² of 0.79 for spatial out-of-sample predictions.
• During periods affected by temporal patterns such as drought and frost, the LSTM shows lower model error than the MLP and an LSTM with shuffled input, demonstrating an advantage from learned temporal dependencies.
• Generalisation patterns reveal that the LSTM tends to be more successful than the (time-agnostic) MLP in simulating GPP in dry environments.
• A large variability in model skill across relatively dry sites remained, which was not resolved by the inclusion of additional Earth observation data, though overall performance improved.
• Insufficient information on the exposure and response to water stress and related effects on GPP appear to be dominant sources of error for modelling ecosystem fluxes globally.

Contributions

• Quantifies the benefit of incorporating temporal dependencies into machine learning models for predicting ecosystem photosynthesis, particularly under environmental stress conditions like drought and frost.
• Provides a comparative analysis of mechanistic, memoryless deep learning, and recurrent deep learning models for GPP prediction across a wide range of global ecosystems.
• Highlights critical data and model limitations concerning the representation of water stress impacts on ecosystem function, emphasizing the need for improved data and models in this area given increasing hydroclimatic extreme events.

Funding

Not explicitly stated in the provided text.

Citation

@article{Biegel2025Unrecognised,
  author = {Biegel, Samantha and Schindler, Konrad and Stocker, Benjamin},
  title = {Unrecognised water limitation is a main source of uncertainty for models of terrestrial photosynthesis},
  journal = {Repository for Publications and Research Data (ETH Zurich)},
  year = {2025},
  doi = {10.3929/ethz-c-000789603},
  url = {https://doi.org/10.3929/ethz-c-000789603}
}