EXCITED Machine Learning Workflow

An open workflow for creating machine learning models for estimating the global biospheric CO2 exchange.

Using globally available data (e.g. ERA5, MODIS) as well as site-level data (Fluxnet) a model for the gross primary production and respiration can be trained. Additionally, the workflow aims to better constrain the CO2 exchange in terrestrial ecosystems on longer timescales using estimates from inverse models (e.g., CarbonTracker) as additional input data.

The workflow is currently geared towards carbon exchange (CO2), but can also be adapted for use with other fluxes such as CH4 or N2O.

The full workflow is split over several notebooks. These guide you through all steps required, from preprocessing, to model training and finally the dataset production.