How to implement a new model

In the CIF, a given numerical model has two main components:

1. a black-box:

   it includes physical, chemical, transport processes, etc.; in most cases it is an executable compiled from Fortran or C sources; it computes outputs (concentration fields, footprints, etc.) from prescribed inputs (fluxes, meteorological fields, boundary conditions, etc.)

2. a Python interface to the model:

   it produces the model inputs, reads the outputs and runs the executable properly.

The Python interface to a model is implemented as a pyCIF model class.

Prior to starting to implement your model, please make sure you have the following elements:

1. a working version of your model (compilable sources), in forward mode, and if applicable in adjoint and tangent-linear modes

2. a test directory with compatible inputs and outputs for your model

3. (optional) an observation file you use to compare with your model; it is also possible to generate random observations with pyCIF

In the following, you will learn how to implement a model to pyCIF step-by-step.

• Include the model sources to the CIF
• Create and register the plugin module
• Create a domain for your model
• Define the “mapper” of your model
  • 1. Setting an empty mapper
  • 2. Setting the mapper with output concentrations
  • 3. Running the model executable
    • 3.1. compile
    • 3.2. make_auxiliary
    • 3.3. run
  • 4. Setting the mapper with input fluxes
• Generate a monitor file compatible with the CIF
• Pre-process observations to fit in model realm
  • Temporal fit to the model time steps
  • Horizontal fit to model domain
• Preparing inputs
  • native2inputs
• Comparing outputs to observations
  • native2inputs
  • run
  • outputs2native