First forward simulation with the toy Gaussian Model

Configuration and set-up

Here we generate pyCIF outputs for a forward run using the Toy Gaussian Model. Follow the steps below. Please refer to this page for instructions on how pyCIF handles paths.

1. Set up a YAML configuration:

   • copy the reference YAML file available here to a directory of your choice

   • the YAML is ready to use; you only need to modify the following variables:

     • rootdir: path to the CIF sources

     • outdir: path where the outputs will be generated

     pyCIF automatically substitutes these variables wherever they appear in the YAML; see here for details

2. Run pyCIF based on your Yaml file:

   python -m pycif path-to-your-yaml
   

Outputs

Explore the results in workdir; if you used the default content of the yaml, you should find files in ${outdir}/fwd_long_noMCF/ that is structured as follows:

${outdir}/fwd_long_noMCF/
├── config_fwd_long_noMCF.yml: a copy of the yml used for the run
├── pycif.logtest: the log file
├── VERSION: a log file giving the version (git commit ID) used for the run
├── controlvect/: directory storing data useful for the control vector
    └── correlations/
        └── horcor_hpixels_12x18_cs-999_cl500.bin:
                binary file storing the horizontal correlation matrix used in this case
├── datavect/: all input data
    ├── concs/: directory with concentrations
                (empty as observation file is randomly generated here)
        └── CH4/
    ├── fluxes/
        └── CH4/
            └── dummy_flx.txt: fluxes as generated from the formula in the yml
    └── meteo/
        └── dummy_meteo_%Y%m%d.txt: meteo files randomly generated
├── model/
    ├── H_matrix.pickle: saved H matrix after computation
    └── Pasquill-Gifford.txt:
            reference correspondance file between
            Pasquill-Gifford classes and Gaussian parameters
├── obsoperator/
    └── fwd_0000/: runs are identified by the mode (fwd, tl or adj) and an ID number
        ├── controlvect.pickle: a binary file with the control vector values used for this run
        ├── controlvect/: the same with netCDF user-friendly format
        ├── %Y-%m-%d_%H:%S: sub-directories for every individual model runs (one per day)
            └── Pasquill-Gifford.txt:
                    link to the original Pasquill-Gifford.txt file
                    in the model directory
        ├── obsvect/:
            directory with a structure following available observation
            components and species; Therein, for each component/species
            a monitor.nc with simulated concentrations
        └── chain/: empty in that case; can be used by models that need to chain sub-simulations

└── obsvect/: the final observation vector with perturbed simulations

The example produces five days of simulations using a Gaussian model with arbitrary meteorological conditions, fluxes distributed according to the formula \((cos(lat / 500) + sin(lon / 1000)) \cdot ((lat / 1000)^2 + (lon / 1000)^2)\), and 5 stations randomly distributed over the domain.

Going further

You can play with these parameters to see the different possible outputs:

• nstations in the paragraph datavect/components/concs/parameters/CH4: you can reduce or increase the number of virtual sites

• flx_formula in the paragraph datavect/components/fluxes/parameters/CH4: you can change the formula used to generate fluxes; accepted arguments to be changed are:

  • variables: zlon or zlat

  • period: a typical horizontal length scale

  • operations: product or sum or square

Many other examples using the Toy Gaussian model with various configurations (inversions, forward runs, adjoint tests, etc.) are available here.