Research Focus LAND

AI4PEX is focused on enhancing our understanding of how terrestrial ecosystems respond to climate change and the feedback of increased atmospheric CO2 levels to the climate system. The project aims to reduce uncertainties and enhance process representation, namely:

1. Hybrid Modelling and History Matching: to better predict the instantaneous vegetation responses to water and heat stress.
2. Leverage Deep Learning: approaches, such as Long-Short Term Memory networks, to simulate phenology and enhance online deep learning frameworks to represent plant carbon dynamics and explore tree mortality drivers.
3. Temperature Sensitivity of Decomposition: Address the challenge of understanding how temperature affects soil decomposition, which is crucial for ecosystem carbon turnover and land-atmosphere carbon responses to warming.
4. Land-Atmosphere Feedbacks: Improve the representation of processes that control energy feedbacks to the atmosphere, including regional climate extremes and land carbon uptake, to reduce uncertainties in projected warming trends.

By focusing on these areas, AI4PEX will to provide a more accurate representation of ecosystem dynamics and feedbacks in climate models.

Here's a brief overview of how is planned:

Task T1.3: Observations and Data Products on Land

[lead: MPG; contributions: ETH, UVEG, ULUND]

• Objective: Improve forest response modelling in Earth System Models (ESMs) using global data on vegetation dynamics and C cycling.
• Approach: Utilize ground-based observations and latest available time series data and develop high-spatiotemporal-resolution products for land water, energy, and C fluxes.
• Focus Area:
  1. soil and vegetation water content.
  2. surface energy balance, such as albedo.
  3. land surface temperature; vegetation greenness and mortality.

Task T2.4: ML-based Parameterisations for Land Processes

[lead: MPG; contributions: CNRS-IPSL, ULUND]

• Objective: Enhance land surface models (LSMs) with machine learning to improve biosphere-atmosphere interactions.
• Approach: Develop ML-enhanced representations of three key land surface processes and implement them in LSMs.
• Focus Areas:
  1. Ecosystem response to water and heat stress.
  2. Phenological shifts due to climate changes.
  3. Vegetation mortality and soil C dynamics.

Task T3.4: Constraining and Quantifying Model Uncertainty for the Land

[lead: MPG; contributions: CNRS- IPSL, METO]

• Objective: Address uncertainties in land surface models.
• Methods: Constrain model parameters using observations, leverage modular model-data integration, combine ML with process-based models.
• Focus Area:
  1. Short, medium, and long-term processes linked to C uptake and turnover.

Task T4.4: Evaluation of Land Surface Processes Relevant for Climate Feedbacks

[lead: ULUND; contributions: MPG, UVEG]

• Objective: Evaluate model performance on key land processes.
• Tools: Use modern ML methods such as explainable AI and symbolic regression.
• Focus Area:
  1. Water use efficiency.
  2. C turnover and forest dynamics, including tree mortality.
  3. Carbon-water coupling.

Task T6.2: Projections of Impacts with ML-enhanced Terrestrial Models

[lead: MPG; contributions: CNRS- IPSL, ULUND]

• Objective: Use ML-enhanced observation-constrained models to project future ecosystem changes.
• Approach: Models will be forced by CMIP7 climate data, evaluated against historical observations, and compared to predecessor simulations.
• Focus Area:
  1. C uptake.
  2. C and water use efficiency.
  3. Ecosystem productivity and resilience.

These tasks collectively aim to reduce uncertainties in climate projections and improve the representation of land-atmosphere feedbacks.