# GPROF Validation Data

The validation data consists of reference measurements from ground based radar
and co-located results from the GPROF V5, GPROF V7, GPROF-NN 1D and GPROF-NN 3D
retrievals. All resuls have been mapped to quasi equidistant 5km x 5km grids that
follow the GMI orbit.

The validation data comes from two sources: (1) Gauge-calibrated radar measurements
over CONUS and (2) the ground radar on the Kwajalein atoll, which are not gauge
corrected. For now I kept them separate but their file structure should be identical.

In addition to that, the data is split into time periods. The folder ``results_db``
contains the results from the current database period 2018-10-01T00:00:00 until
2019-10-01T00:00:00. The folder ``results_db_1`` contains the results from the
year following the database period. I working on adding one more year, but for
this we currently lack the required ancillary data available

## File format

The data is in NetCDF4 format with the results from the reference data and retrievals
in different groups. The available groups are:

1. ``reference``: This group contains the precipitation fields measured by the ground
    radar. In this group the variable 'surface_precip' corresponds to the reference
    surface precip  and 'surface_precip_avg' to the GMI footprint averaged
    precipitation. Depending on whether the source is MRMS or the Kwajalein radar,
    this group also contains additional information from the reference data
    (radar quality index, feature mask, etc.).

2. ``gprof_v5``: This group contains the GPROF retrieval results from the current
   version of GPROF.

3. ``gprof_v7``: This group contains the retrieval results from the updated version
   of GPROF.
4. ``gprof_nn_1d``: This group contains the retrieval results from the GPROF-NN 1D 
   retrieval.
5. ``gprof_nn_3d``: This group contains the retrieval results from the GPROF-NN 3d
   retrieval.
6. ``simulator``: For the results from the database year, this contains the surface
   precip from the retrieval database on which all new GPROF retrievals are
   based.