Identifiability: 9 of 25 air-chemistry rates can be learned
A simple tool shows which parts of the chemical model can be identified before fitting data.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the paper
- Authors
- Wenqing Peng, Michael Boy, Zhi-Song Liu
- Universities and research institutions
- Not yet supplied in verified metadata; the Brief does not guess.
Institution metadata: OpenAlex record ↗
What they did and found
Researchers used a small box model of atmosphere chemistry to test identifiability. They found 9 of 25 rate coefficients are learnable with the current design; the rest cluster in fast-reaction groups where only ratios matter.
Why it matters
The results help plan data collection and show where online adjustments can improve corrections without changing the whole model.
The study used a common air-chemistry model and asked which reaction rates you can learn from real data. They checked which parameters stay close to true values when noise is added.
Online tuning of a neural correction for biogenic emissions reduced errors in several settings, especially at lower temperatures.
What remains uncertain
Findings are local to the tested design and conditions. The identifiability and drift depend on the observation setup; extrapolation to larger mechanisms may behave differently. Concentrations must stay positive for the math to work, and real-world initial states can affect results.
Original sources · 1
- KPPax: preprocess differentiable kinetic ODE for machine learning in atmospheric chemistry ↗arXiv · 2026-10-27
Check the original paper for its authors, methods, version and access terms.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.