MESSy Symposium 2026: Advancing Aviation Climate Metrics with Machine Learning

Last week in Karlsruhe (Germany), the Modular Earth Submodel System (MESSy) Symposium 2026 brought together atmospheric scientists and Earth system modelers to discuss the latest advancements in climate simulation. Representing the F4EClim project, Katarina Grubbe Hildebrandt from TU Delft presented cutting-edge research aimed at minimizing the climate impact of aviation through smarter, data-driven flight routing.

The Challenge: Beyond CO2​

While carbon dioxide emissions from aircraft are a well-known driver of global warming, non-CO2​ emissions—such as nitrogen oxides (NOx​) and contrail cirrus—actually contribute a massive share of aviation's total climate impact. However, the climate response to NOx​ is highly variable depending on exactly where and when it is released.

To optimize flight paths for minimal climate impact, scientists use Algorithmic Climate Change Functions (aCCFs). While version 1.0 of these functions was a massive breakthrough, it was geographically restricted to the North Atlantic, limited to summer and winter seasons, and lacked uncertainty estimates.

ACCFs V2.0: Expanding geographical-seasonal scope and quantifying uncertainty

Katarina’s poster, titled "Towards NOx​-O3​ aCCFs V2.0: A Machine Learning Approach Using Lagrangian Simulations in EMAC," detailed how the F4EClim team is breaking past these limitations.

By deploying a Gaussian Process probabilistic surrogate model, the new framework scales up aviation climate optimization across several fronts:

• Broader Scope: Moving beyond the North Atlantic to include Europe, the Middle East, and North America, spanning all four seasons (winter, spring, summer, and fall).

• Predictive Accuracy: The model is trained on extensive climate-chemistry simulations using the EMAC atmospheric model to predict instantaneous radiative forcing (iRF) from NOx​ based on local weather conditions.

• Feature Engineering: The research identifies key meteorological indicators like geopotential height (which reflects transport pathways) and the solar zenith angle (which influences photochemistry), to accurately gauge climate impact.

• Quantifying Uncertainty: Crucially, version 2.0 introduces probabilistic uncertainty estimates, giving flight planners a much clearer understanding of predictability and risk.

  
  

Next Steps at F4EClim

The F4EClim team is currently validating the surrogate model’s performance across consecutive days, different weather patterns, and varying altitudes. Ultimately, these refined aCCFs will be integrated into flight planning software to help the aviation industry chart eco-efficient, climate-optimized trajectories in real time.