# Weather Forecast Models

## 1. METEO FRANCE's Forecast Models

Météo-France produces several models to forecast the weather. These models are based on physical equations modelling the physical behaviours influencing the weather, such as atmospheric or oceanic phenomenons... They compute a lot of weather parameters (temperature, wind speed, humidity...) at different points on the globe in 4 dimensions (latitude, longitude, height and time).

Each model runs several times a day at a fixed time. Each execution is called a 'run'. Each time-step for which we have weather forecasts is called a 'valid time'.

On the globe's surface, the points are spaced with regular latitude and longitude steps : this mesh of points is called a grid. On the vertical axis (height), the points are not necessarily regurlarly spaced; the vertical levels can be expressed in height levels in meters or in isobar levels in hPa. The isobar are levels for which the pressure is constant.

The 2 main models of Météo-France are AROME and ARPEGE. They do no have the same purposes but they are complementary.

• ARPEGE is used to forecast the weather at a high scale in space and time.
• AROME is used to forecast the weather at a finer scale in space and time.
Model Name Horizontal Spatial Resolution Number of runs per day Time scale
ARPEGE 0.1° (~10km) 4 forecasts up to 102 hours ahead
AROME 0.025° (~1km) 8 forecasts up to 48 hours ahead

All these characteristics are the result of compromises between the need for precision and for long time scale and the computing times and costs.

In pratice, ARPEGE is used to forecast the weather on Europe and AROME is used to forecast the weather on France.

In the dataset, we provided the results of 2D models (forecasts are on a grid longitude x latitude) and 3D models (forecasts are on a grid longitude x latitude x height).

Warning

Some files in the model files are incomplete/corrupted. It comes from the database of our corporation. There are archiving anomalies. We are working on a new version of MeteoNet and will take into account these anomalies. Meanwhile, you can detect these anomalies by checking the file sizes. For a given file category (zone, level, model type), the files must have exactly the same size. So if some files are smaller than the other one, then these files have anomalies. This has been reported in kaggle in this discussion https://www.kaggle.com/katerpillar/meteonet/discussion/157206 and in Github (https://github.com/meteofrance/meteonet/issues/14).

## 2. 2D Parameters

### 2.1 Description

Two 2D weather models are available in the dataset :

• the fine-mesh French weather model named 'AROME'.

• the large-mesh French weather model named 'ARPEGE'.

For each geographical zone, the two models were run each day at 00h and their forecasts range from 00h to 24h, with a time step of 1h.

For each model, the data is stored in 4 different GRIB files, depending on the vertical level of the parameters :

File Name Vertical Level Forecasted Parameters
2m 2 meters Temperature (K), Dew point (K), Relative humidity (%)
10m 10 meters Wind speed (m.s-1), Wind direction (°), U and V wind components (m.s-1)
P_sea_level sea level Mean sea level pressure (Pa)
PRECIP ground level Total precipitation (kg.m-2 which is equivalent to mm) since the beginning of the model run

Warning

The total precipitation parameter forecast begins at 1h, and not 0h as the other parameters. For now it is about total precipitation since the beginning of the model run. It is planned in a future version to modify that parameter in order to have the total precipitation between two time steps instead.

Info

When you open a GRIB file with the xarray library, a new associated .idx file is created.

### 2.2 Files organization

The files are stored with 4 compression levels :

• two per year (.tar.gz and .tar)
• two per month (.tar.gz and .tar).

Then, the data are stored, in this order, by :

1. Weather forecast model : AROME or ARPEGE
2. Vertical level : 2m, 10m, P_sea_level or PRECIP
3. One file per day

Example

Here is the path for the NW zone, year 2016, January, AROME model, 2m vertical level, day 1:
/data/NW/weather_models/2D_parameters/
NW_weather_models_2D_parameters_2016.tar.gz/
NW_weather_models_2D_parameters_2016.tar/
NW_weather_models_2D_parameters_2016/
NW_weather_models_2D_parameters_201601.tar.gz/
NW_weather_models_2D_parameters_201601.tar/
201601/AROME/2m/
arome_2m_NW_20160101000000.grib

## 3. 3D Parameters

### 3.1 Description

One 3D weather model is avaible : the large-mesh French weather model named 'ARPEGE'.

For each geographical zone (NW and SE), the model was run each day at 00h and its forecasts range from 00h to 24h. The time-step between forecasts vary depending on the time of the day : from 0h to 12h, forecasts are available each hour, and from 12h to 24h, forecasts are available every three hours. Indeed, the model is more precise for the first hours of the days, closer to the moment the model was run, and its precision degrades as time passes.

The 3D data is stored in 2 different GRIB files, depending on the vertical level of the parameters :

File Name Vertical Levels Forecasted Parameters
height 20, 100, 500, 875, 1375, 2000 and 3000 m Pressure (Pa)
isobar 1000, 950, 925, 850, 700, 600 and 500 hPa Temperature (K), Pseudo-adiabatic potential temperature of the wet bulb (K), Relative humidity (%), Wind speed (m.s-1), Wind direction (°), U and V wind components (m.s-1), vertical velocity (Pa.s-1), geopotential (m2.s-2)

### 3.2 Files organization

The files are stored with 4 compression levels :

• two per year (.tar.gz and .tar)
• two per month (.tar.gz and .tar).

Then, the data are stored, in this order, by :

1. Weather forecast model : ARPEGE only
2. Vertical level : isobar or height
3. One file per day

Example

Here is the path for the NW zone, year 2016, January, ARPEGE model, isobar level, day 1:
/data/NW/weather_models/3D_parameters/
NW_weather_models_3D_parameters_2016.tar.gz/
NW_weather_models_3D_parameters_2016.tar/
NW_weather_models_3D_parameters_2016/
NW_weather_models_3D_parameters_201601.tar.gz/
NW_weather_models_3D_parameters_201601.tar/
201601/ARPEGE/3D_isobar/
arpege_3D_isobar_NW_20160101000000.grib