Overview

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.

As explained on the main page, we train models on two separate datasets. Fluxnet site-based data for hourly fluxes to resolve smaller spatial and temporal scale fluxes, and CarbonTracker inverse modelling data to achieve higher accuracy on the longer-term.

As the Fluxnet data is (half-)hourly, some preprocessing is required. This is performed in the first two notebooks. The CarbonTracker data is at a lower frequency, and more ready to immediately use for analysis.

Fluxnet-based hourly model:

1. Preprocessing Ameriflux data
2. Preprocessing ERA5
3. Fluxnet model training
4. Hourly dataset production

CarbonTracker-based monthly model:

1. CarbonTracker model training
2. Monthly dataset production