Python interface for FMI open data
pip install fmiopendataFor grid datasets install also eccodes. Both the library and
Python bindings are needed. The former is easiest to install with
conda and the latter via pip:
conda install eccodes -c conda-forge
pip install eccodesFor radar datasets rasterio is needed. It can be installed with
pip, although usage of miniconda is strongly encouraged.
This library provides two very simple scripts that list all the available data on FMI open data WMS and WFS services.
Create wms.html that lists all the available WMS layers:
wms_html.pyThe Layer ID strings are used to identify WMS (image) datasets. Examples to be added.
Create wfs.html that lists all the available WFS layers:
wfs_html.pyThe Query ID is the handle that can be used to request data from WFS stored queries. See examples below.
- Download and parse latest soundings
- Download and calibrate latest radar reflectivity (dBZ) composite
- Download and parse latest lightning data
- Download and parse grid data
- Download and parse observation data
from fmiopendata.wfs import download_stored_query
snd = download_stored_query("fmi::observations::weather::sounding::multipointcoverage")Now snd.soundings is a list of the soundings, and the data within each can be reached from the attributes:
sounding = snd.soundings[0]
sounding.name # Name of the sounding station
sounding.id # Station ID of the sounding station
sounding.nominal_time # Nominal time of the sounding
sounding.start_time # Actual start time of the sounding
sounding.end_time # Actual end time of the sounding
sounding.lats # Numpy array of the measurement location latitudes [degrees]
sounding.lons # Numpy array of the measurement location longitudes [degrees]
sounding.altitudes # Numpy array of the measurement location altitudes [m]
sounding.times # Numpy array of the measurement times [datetime]
sounding.pressures # Numpy array of measured pressures [hPa]
sounding.temperatures # Numpy array of measured temperatures [°C]
sounding.dew_points # Numpy array of measured dew points [°C]
sounding.wind_speeds # Numpy array of measured wind speeds [m/s]
sounding.wind_directions # Numpy array of measured wind directions [°]More refined search results can be retrieved by passing a list of arguments that will be added to the WFS request URL.
snd = download_stored_query("fmi::observations::weather::sounding::multipointcoverage",
["place=Jokioinen"])This parser requires the rasterio library to be installed. All radar queries will work.
from fmiopendata.wfs import download_stored_query
composites = download_stored_query("fmi::radar::composite::dbz")Now composites.data is a list of the individual reflectivity composites and composites.times is a list of the nominal measurement times.
import numpy as np
composite = composites.data[0]
# Download the image data
composite.download()
# Calibrate the data from image values to dBZ
# Calls `composite.download() if data are not already downloaded
composite.calibrate()
# Get mask for area outside the radar reach
# Calls `composite.download() if data are not already downloaded
mask1 = composite.get_area_mask()
# Get mask for invalid data
# Calls `composite.download() if data are not already downloaded
mask2 = composite.get_data_mask()
# Mask all the invalid areas using the above masks
composite.data[mask1 | mask2] = np.nan
# Plot the data for preview
import matplotlib.pyplot as plt
plt.imshow(composite.data[0, :, :])
plt.show()The following attributes are available in the underlying Radar class for all radar data:
Radar.data # The downloaded data, if .download() has been called
Radar.elevation # Measurement angle for single radar datasets
Radar.etop_threshold # Reflectivity limit for `etop` datasets
Radar.label # Clear-text name for the data
Radar.max_velocity # Maximum wind speed for `vrad` datasets
Radar.name # Name of the dataset
Radar.projection # WKT projection string for the dataset
Radar.time # Nominal measurement time of the dataset
Radar.unit # Unit of the calibrated data
Radar.url # Direct download URL for the datafrom fmiopendata.wfs import download_stored_query
# These both give identical results, the first one is much faster
lightning1 = download_stored_query("fmi::observations::lightning::multipointcoverage")
lightning2 = download_stored_query("fmi::observations::lightning::simple")The following attributes hold the lightning data.
lightning1.latitudes # Latitude of the lightning event [° North]
lightning1.longitudes # Longitude of the lightning event [° East]
lightning1.times # Time of the lightning event [datetime]
lightning1.cloud_indicator # Indicator for cloud flashes (1 == cloud lightning)
lightning1.multiplicity # Multiplicity of the lightning event
lightning1.peak_current # Maximum current of the lightning event [kA]
lightning1.ellipse_major # Location accuracy of the lightning event [km]The example below uses fmi::forecast::harmonie::surface::grid for demonstration. Also other
grid stored queries should work, as long as the data are available in GRIB format.
This parser requires eccodes library to be installed.
import datetime as dt
from fmiopendata.wfs import download_stored_query
# Limit the time
now = dt.datetime.utcnow()
# Depending on the current time and availability of the model data, adjusting
# the hours below might be necessary to get any data
start_time = now.strftime('%Y-%m-%dT00:00:00Z')
end_time = now.strftime('%Y-%m-%dT18:00:00Z')
model_data = download_stored_query("fmi::forecast::harmonie::surface::grid",
args=["starttime=" + start_time,
"endtime=" + end_time,
"bbox=18,55,35,75"])Here we limit both the time frame and area. By default the WFS interface would offer all data fields, full globe and all time steps, which would result in a huge amount of data. It's better to split the retrieval in smaller time intervals and do the downloading in parallel and start processing the each chunk of data as soon as the download completes.
The above call doesn't download any actual data, it just collects some metadata:
print(model_data.data)
# -> {datetime.datetime(2023, 1, 6, 6, 0): <fmiopendata.grid.Grid object at 0x7f8fc9aaaaa0>,
# datetime.datetime(2023, 1, 6, 9, 0): <fmiopendata.grid.Grid object at 0x7f8fc9aaaad0>}Here the dictionary keys are the model initialization times.
The structure of the underlying Grid class is:
Grid.data # Parsed data, more info below
Grid.delete_file() # Delete the downloaded GRIB2 file
Grid.download() # Download the data
Grid.end_time # Last valid time of the data [datetime]
Grid.init_time # Initialization time of the data [datetime]
Grid.latitudes # Latitudes of the parsed data [Numpy array]
Grid.longitudes # Longitudes of the parsed data [Numpy array]
Grid.parse() # Download and parse the data
Grid.start_time # First valid time of the data [datetime]
Grid.url # The URL where the data are downloaded fromNow, download and parse the data from the latest run:
latest_run = max(model_data.data.keys()) # datetime.datetime(2020, 7, 7, 12, 0)
data = model_data.data[latest_run]
# This will download the data to a temporary file, parse the data and delete the file
data.parse(delete=True)The data.data dictionary now has the following structure:
# The first level has the valid times
valid_times = data.data.keys()
print(list(valid_times))
# -> [datetime.datetime(2023, 1, 6, 9, 0),
# datetime.datetime(2023, 1, 6, 10, 0),
# datetime.datetime(2023, 1, 6, 11, 0),
# datetime.datetime(2023, 1, 6, 12, 0),
# datetime.datetime(2023, 1, 6, 13, 0),
# datetime.datetime(2023, 1, 6, 14, 0),
# datetime.datetime(2023, 1, 6, 15, 0),
# datetime.datetime(2023, 1, 6, 16, 0),
# datetime.datetime(2023, 1, 6, 17, 0),
# datetime.datetime(2023, 1, 6, 18, 0)]The second level of the dictionary has the data vertical levels, here the earliest time step is shown:
earliest_step = min(valid_times)
data_levels = data.data[earliest_step].keys()
print(list(data_levels))
# -> [0, 2, 10]Lets loop over these levels and print the dataset names and the units.
for level in data_levels:
datasets = data.data[earliest_step][level]
for dset in datasets:
unit = datasets[dset]["units"]
data_array = datasets[dset]["data"] # Numpy array of the actual data
print("Level: %d, dataset name: %s, data unit: %s" % (level, dset, unit))This will print:
Level: 0, dataset name: Mean sea level pressure, data unit: Pa
Level: 0, dataset name: Orography, data unit: m
Level: 0, dataset name: Surface net thermal radiation, data unit: J m**-2
Level: 0, dataset name: Surface net solar radiation, data unit: J m**-2
Level: 0, dataset name: Time-integrated surface direct short wave radiation flux, data unit: J m**-2
Level: 2, dataset name: 2 metre temperature, data unit: K
Level: 2, dataset name: 2 metre dewpoint temperature, data unit: K
Level: 2, dataset name: 2 metre relative humidity, data unit: %
Level: 10, dataset name: Mean wind direction, data unit: degrees
Level: 10, dataset name: 10 metre wind speed, data unit: m s**-1
Level: 10, dataset name: 10 metre U wind component, data unit: m s**-1
Level: 10, dataset name: 10 metre V wind component, data unit: m s**-1
Level: 10, dataset name: surface precipitation amount, rain, convective, data unit: kg m**-2
Level: 10, dataset name: Convective available potential energy, data unit: J kg**-1
Level: 10, dataset name: Total Cloud Cover, data unit: %
Level: 10, dataset name: Low cloud cover, data unit: (0 - 1)
Level: 10, dataset name: Medium cloud cover, data unit: (0 - 1)
Level: 10, dataset name: High cloud cover, data unit: (0 - 1)
Level: 10, dataset name: Global radiation flux, data unit: W m**-2
Level: 10, dataset name: Visibility, data unit: m
Level: 10, dataset name: 10 metre wind gust since previous post-processing, data unit: m s**-1
Level: 10, dataset name: Surface solar radiation downwards, data unit: J m**-2
The whole structure Grid.data[valid_time][level][dataset_name] with 'data' and 'units'
members is common to all grid type WFS data.
The data arrays will have invalid values replaced with np.nan.
import datetime as dt
from fmiopendata.wfs import download_stored_query
# Retrieve the latest hour of data from a bounding box
end_time = dt.datetime.utcnow()
start_time = end_time - dt.timedelta(hours=1)
# Convert times to properly formatted strings
start_time = start_time.isoformat(timespec="seconds") + "Z"
# -> 2020-07-07T12:00:00Z
end_time = end_time.isoformat(timespec="seconds") + "Z"
# -> 2020-07-07T13:00:00Z
obs = download_stored_query("fmi::observations::weather::multipointcoverage",
args=["bbox=18,55,35,75",
"starttime=" + start_time,
"endtime=" + end_time])The structure of the returned MultiPoint class is
MultiPoint.data # The observation data
MultiPoint.location_metadata # Location information for the observation locationsThe data dictionary has the following structure, which is designed to be used for
collecting data for distinct timesteps:
# The observation times are the primary key
print(sorted(obs.data.keys()))
# -> [datetime.datetime(2023, 1, 6, 14, 5),
# datetime.datetime(2023, 1, 6, 14, 6),
# datetime.datetime(2023, 1, 6, 14, 7),
# ...
# datetime.datetime(2023, 1, 6, 15, 4)]
# The next level has the names of the observation stations as keys
latest_tstep = max(obs.data.keys())
print(sorted(obs.data[latest_tstep].keys()))
# Will print a list of weather station names, e.g. "Kustavi Isokari", which we'll use
# as an example below
# On the third level we find the names of the observed parameters
print(sorted(obs.data[latest_tstep]["Kustavi Isokari"].keys()))
# -> ['Air temperature',
# 'Cloud amount',
# 'Dew-point temperature',
# 'Gust speed',
# 'Horizontal visibility',
# 'Precipitation amount',
# 'Precipitation intensity',
# 'Present weather (auto)',
# 'Pressure (msl)',
# 'Relative humidity',
# 'Snow depth',
# 'Wind direction',
# 'Wind speed']
# And on the last level we find the value and unit of the observation
print(obs.data[latest_tstep]["Kustavi Isokari"]["Air temperature"])
# -> {'value': -6.7, 'units': 'degC'}To get the location for the stations, one can use the location_metadata dictionary:
print(obs.location_metadata["Kustavi Isokari"])
# -> {'fmisid': 101059, 'latitude': 60.7222, 'longitude': 21.02681}It is also possible to collect the data to a structure more usable for timeseries
analysis by adding "timeseries=True" to the arguments:
from fmiopendata.wfs import download_stored_query
obs = download_stored_query("fmi::observations::weather::multipointcoverage",
args=["bbox=25,60,25.5,60.5",
"timeseries=True"])Now the data will be organized by location:
print(sorted(obs.data.keys()))
# -> ['Helsinki Malmi lentokenttä',
# 'Helsinki Vuosaari Käärmeniementie',
# 'Helsinki Vuosaari satama',
# 'Järvenpää Sorto',
# 'Sipoo Itätoukki']
# The times are as a list of datetime objects
times = obs.data['Helsinki Malmi lentokenttä']['times']
# Other data fields have another extra level, one for values and one for the unit
print(len(obs.data['Helsinki Malmi lentokenttä']['Air temperature']['values']))
# -> 71
print(obs.data['Helsinki Malmi lentokenttä']['Air temperature']['unit'])
# -> 'degC'This parser supports at least the following stored queries:
fmi::forecast::hirlam::surface::obsstations::multipointcoveragefmi::forecast::oaas::sealevel::point::multipointcoveragefmi::observations::airquality::hourly::multipointcoveragefmi::observations::mareograph::daily::multipointcoveragefmi::observations::mareograph::instant::multipointcoveragefmi::observations::mareograph::monthly::multipointcoveragefmi::observations::mareograph::multipointcoveragefmi::observations::radiation::multipointcoveragefmi::observations::wave::multipointcoveragefmi::observations::weather::cities::multipointcoveragefmi::observations::weather::multipointcoveragestuk::observations::air::radionuclide-activity-concentration::multipointcoveragestuk::observations::external-radiation::latest::multipointcoveragestuk::observations::external-radiation::multipointcoverageurban::observations::airquality::hourly::multipointcoverage