This repository collects machine learning–based parameterizations developed in the AI4PEX project, providing reusable modules for Earth system modeling.
Each parameterization is maintained as a submodule with its own documentation and license.
- Browse the available parameterizations by domain (Land, Atmosphere, Ocean).
- Follow the links to each submodule’s README for setup and usage.
- Quickstart guides (coming soon) will provide examples for applying each method to your own data and challenges.
-
NN-based Respiration and GPP from Flux Partitioning
- Quickstart on learning fluxes with NNs
-
Hybrid Variational Inference for Soil Organic Matter dynamics
-
MortNetSE parameterization pipeline
- Quickstart notebook here
Coming soon:
- Semi-parametric Hybrid Modeling (including Q10 model)
- Parameterizations of tree mortality directly from satellite and climate data
-
- Quickstarts on a data-driven cloud cover equation discovery:
-
Stochastic Recurrent Neural Network for modeling Atmospheric Regimes
- Quickstart using synthetic data
Coming soon:
- Convection parameterization trained on ClimSim data for the ICON model
- Improving vertical detail in simulated temperature and humidity
Coming soon:
- Emulation of PISCES biogeochemical model (for details, contact Edward Thornton, edward.gow-smith@meteo.fr)
- Predicting eddy energy using a CNN for use in the scale-aware GEOMETRIC eddy parameterization
- Add quickstart guides for each method
- Expand documentation and folder organization
- Review contribution guidelines
Please check out License for each submodule linked in this repository.
For questions and contributions, please reach out to the ISP at UVEG:
- Gherardo Varando gherardo.varando@uv.es
- Andrei Gavrilov andrei.gavrilov@uv.es
- Kai-Hendrik Cohrs kai.cohrs@uv.es