###################
Observation vectors
###################

In pyCIF, the observation vector :math:`\mathbf{y}` is stored as a directory,
organised by observation type and species.  Each file follows the same
format as the :doc:`input monitor files <monitor>`.

.. code-block:: text

    obsvect/
    ├── {obs type 1}/
    │   ├── {species 1}/
    │   │   └── monitor.nc
    │   └── {species 2}/
    │       └── monitor.nc
    └── {obs type 2}/
        ├── {species 1}/
        │   └── monitor.nc
        └── {species 2}/
            └── monitor.nc


NetCDF format
^^^^^^^^^^^^^

Each :bash:`monitor.nc` uses NetCDF groups to separate the two types of
information:

:maindata:
    Numerical data directly used by the inversion:

    - **obs** — observed values
    - **obserror** — observation errors (:math:`\sigma_\varepsilon`)
    - **sim** — simulated equivalents (forward run)
    - **sim_tl** — simulated increments (tangent-linear run)
    - **adj_out** — adjoint sensitivities
    - **incr** — observation-space increments

:metadata:
    All other information about each observation point: date, station,
    network, parameter, lon, lat, alt, level, i, j, tstep, tstep_glo,
    dtstep, dtstep_glo.
    See :doc:`monitor` for the full description of each field.

Each :bash:`monitor.nc` is formatted as illustrated in the following
example (real file from a CHIMERE run):

.. literalinclude:: dump_monitor.txt