Write Dataset Documentation

[sadamov]

Compile a detailed but concise documentation about the COSMO-dataset we are using.

• Including all preprocessing steps in this repo: https://github.com/MeteoSwiss-APN/fcn-cosmo
• Information from the nvidia report:

Data
The data used in this project was produced by the data assimilation system KENDA of MeteoSwiss.
KENDA generates operational analysis files by running COSMO for 1 hour while assimilating
observations from measurement stations and weather balloons. This dataset represents the longest
homogeneous timeseries stored in the archive and runs between December 2015 and October 2020.
The basic set of equations in COSMO are prognostic Eulerian equations for momentum, heat, total
mass, mass of water substance and the equation of state.
• Spatial discretization of the data is a 2.2kmx2.2km grid of size 390x582 with 1hr temporal
resolution.
• COSMO uses a rotated pole projection with a new NorthPole defined at pollon=-170.0, pollat
= 43.0
• Vertically, the model uses the SLEVE terrain-following coordinate with 60 levels.
• Currently, we are using in total 8 free-atmosphere variables ('FI', 'T', 'U', 'V', 'RELHUM', 'TQV',
'hsurf', 'PP') on 21 model levels and 5 surface variables ('U_10M', 'V_10M', 'PMSL', 'PS',
'T_2M').
Both the variable selection and the selection of vertical levels (Section 3.2) is subject to further
research. A balance must be struck between physical accuracy and computational limitations.
Note: The dataset is currently not open-source and requires access to either the Meteoswiss or the
CSCS infrastructure.

• The cosmo-model.org docs
• all other sources deemed important