Eumetnet postprocessing benchmark dataset demo¶

Installation with pip (if needed)

In [ ]:

#!pip install matplotlib
#!pip install climetlab
#!pip install climetlab-eumetnet-postprocessing-benchmark

Loading climetlab

In [ ]:

import climetlab as cml

and matplotlib

In [ ]:

import matplotlib.pyplot as plt

Example with a 2 metre temperature field deterministic forecast¶

Download of a deterministic high-resolution forecast (the 2 metre temperature)

In [ ]:

ds = cml.load_dataset(
    "eumetnet-postprocessing-benchmark-training-data-gridded-forecasts-surface",
    date="2017-12-02",
    parameter="2t",
    kind="highres",
)

The ds object returned is a climetlab datasource that can be converted to various format. Note that the data are cached temporarily on your disk so next time you ask for this forecast I might be faster.

For example, conversion to a xarray :

In [ ]:

fcs = ds.to_xarray()
fcs

Plotting using climetlab and magics (still in development)

In [ ]:

cml.plot_map(ds[5], # plot of the 5th time step
             foreground=dict(
                 map_grid=False,
                 map_label=False,
                 map_grid_frame=True,
                 map_grid_frame_thickness=5,
                 map_boundaries=True,
                ),
            )

Retrieving the observations corresponding to the forecasts (in xarray format)

In [ ]:

obs = ds.get_observations_as_xarray()
obs

Plotting the observations with magics

In [ ]:

cml.plot_map(obs.t2m[0,0,5,0], # plot of the 5th time step
             foreground=dict(
                 map_grid=False,
                 map_label=False,
                 map_grid_frame=True,
                 map_grid_frame_thickness=5,
                 map_boundaries=True,
                ),
            )

Compute the difference between the forecast and the observations

In [ ]:

diff = fcs - obs
diff

and plot

In [ ]:

cml.plot_map(diff.t2m[0,0,5,0], # plot of the 5th time step
             foreground=dict(
                 map_grid=False,
                 map_label=False,
                 map_grid_frame=True,
                 map_grid_frame_thickness=5,
                 map_boundaries=True,
                ),
            )

Save the forecast and observations on the hard drive as a netcdf

In [ ]:

fcs.to_netcdf('fcs_hr_2017-12-02.nc')
obs.to_netcdf('obs_2017-12-02.nc')
!ls

Plotting the observations and the forecast at 4 different points (forecasts are solid lines, observations are dashed lines)

In [ ]:

points = [(20, 20), (20, 70), (70, 20), (70, 70)]

plt.figure(figsize=(10, 8))

for point in points:
    line = fcs.t2m.isel(latitude=point[0], longitude=point[1], time=0, surface=0, number=0).plot()
    obs.t2m.isel(latitude=point[0], longitude=point[1], time=0, surface=0, number=0).plot(ls='--', color=line[0].get_color())
    

Example with a total precipitation accumulated over the last 6 hours ensemble forecast¶

Download of an ensemble forecast (the precipitation)

In [ ]:

ds = cml.load_dataset(
    "eumetnet-postprocessing-benchmark-training-data-gridded-forecasts-surface-processed",
    date="2017-12-02",
    parameter="tp",
    kind="ensemble",
)

In [ ]:

fcs = ds.to_xarray()
fcs

In [ ]:

obs = ds.get_observations_as_xarray()
obs

Plotting the observations and forecast ensemble mean at 4 different points (forecasts are solid lines, observations are dashed lines)

In [ ]:

points = [(20, 20), (20, 70), (70, 20), (70, 70)]

plt.figure(figsize=(10, 8))

for point in points:
    line = fcs.tp6.isel(latitude=point[0], longitude=point[1], time=0, surface=0).mean('number').plot()    
    obs.tp6.isel(latitude=point[0], longitude=point[1], time=0, surface=0, number=0).plot(ls='--', color=line[0].get_color())
    

Example with a 2 metre temperature field ensemble reforecast¶

Download of an ensemble reforecast (the 2 metre temperature)

In [ ]:

ds = cml.load_dataset(
    "eumetnet-postprocessing-benchmark-training-data-gridded-reforecasts-surface",
    date="2017-12-28",
    parameter="2t"
)

In [ ]:

fcs = ds.to_xarray()
fcs

In [ ]:

obs = ds.get_observations_as_xarray()
obs

Plotting the observations and forecast ensemble mean at 4 different points (forecasts are solid lines, observations are dashed lines) in 1997 with the 2017 model version !

In [ ]:

points = [(20, 20), (20, 70), (70, 20), (70, 70)]

plt.figure(figsize=(10, 8))

for point in points:
    line = fcs.t2m.isel(latitude=point[0], longitude=point[1], time=0, surface=0).mean('number').plot()
    obs.t2m.isel(latitude=point[0], longitude=point[1], time=0, surface=0, number=0).plot(ls='--', color=line[0].get_color())
    

Example with a total precipitation accumulated over the last 6 hours ensemble reforecast¶

Download of an ensemble reforecast (the precipitation)

In [ ]:

ds = cml.load_dataset(
    "eumetnet-postprocessing-benchmark-training-data-gridded-reforecasts-surface-processed",
    date="2017-12-28",
    parameter="tp"
)

In [ ]:

fcs = ds.to_xarray()
fcs

In [ ]:

obs = ds.get_observations_as_xarray()
obs

Plotting the observations and forecast ensemble mean at 4 different points (forecasts are solid lines, observations are dashed lines) in 1997 with the 2017 model version !!

In [ ]:

points = [(20, 20), (20, 70), (70, 20), (70, 70)]

plt.figure(figsize=(10, 8))

for point in points:
    line = fcs.tp6.isel(latitude=point[0], longitude=point[1], time=0, surface=0).mean('number').plot()        
    obs.tp6.isel(latitude=point[0], longitude=point[1], time=0, surface=0, number=0).plot(ls='--', color=line[0].get_color())
    