PeakWeather Demo

import matplotlib.dates as mdates
import matplotlib.pyplot as plt
import matplotlib

import peakweather

print(peakweather.__version__)

from peakweather import PeakWeatherDataset

0.2.2

Loading the data

To get the dataset, simply initialize a PeakWeatherDataset object. This will, by default, download the data in the current working directory, unless it was previously downloaded.

dataset = PeakWeatherDataset()

stations.parquet: 57.3kB [00:00, 194kB/s]
installation.parquet: 90.1kB [00:00, 348kB/s]
parameters.parquet: 8.19kB [00:00, 34.5kB/s]
disclaimer.txt: 8.19kB [00:00, 36.4kB/s]
2017.parquet: 56.9MB [00:01, 41.3MB/s]
2018.parquet: 57.3MB [00:01, 39.9MB/s]
2019.parquet: 57.5MB [00:01, 40.5MB/s]
2020.parquet: 57.1MB [00:01, 41.2MB/s]
2021.parquet: 56.5MB [00:01, 39.7MB/s]
2022.parquet: 56.8MB [00:01, 40.7MB/s]
2023.parquet: 57.5MB [00:01, 38.6MB/s]
2024.parquet: 57.5MB [00:01, 41.0MB/s]
2025.parquet: 45.1MB [00:01, 38.6MB/s]

With get_observations you can load the timeseries data as a dataframe or array. It is possible to load specific spatial and temporal subsets and to obtain a binary mask that specifies the availability of each measurement.

df, mask = dataset.get_observations(
    stations=["ABO", "GRO", "KLO"],  # list of stations
    parameters=["temperature", "wind_speed", "humidity"],  # list of weather parameters
    first_date="2024-08-02 16:32",
    last_date="2024-08-06 23:26",
    return_mask=True,
)

df.head()

nat_abbr	ABO			GRO			KLO
name	temperature	wind_speed	humidity	temperature	wind_speed	humidity	temperature	wind_speed	humidity
datetime
2024-08-02 16:40:00+00:00	19.400000	2.0	67.000000	24.100000	0.8	72.500000	26.299999	2.8	50.000000
2024-08-02 16:50:00+00:00	19.600000	2.7	67.300003	23.100000	0.9	79.099998	25.500000	3.2	56.200001
2024-08-02 17:00:00+00:00	19.200001	2.0	71.199997	22.900000	1.4	80.800003	25.400000	2.7	55.799999
2024-08-02 17:10:00+00:00	19.000000	0.9	73.900002	23.000000	2.3	78.300003	25.200001	2.7	57.000000
2024-08-02 17:20:00+00:00	19.100000	1.5	71.099998	22.799999	2.3	78.199997	25.000000	2.3	58.000000

mask.head()

nat_abbr	ABO			GRO			KLO
name	temperature	wind_speed	humidity	temperature	wind_speed	humidity	temperature	wind_speed	humidity
datetime
2024-08-02 16:40:00+00:00	True	True	True	True	True	True	True	True	True
2024-08-02 16:50:00+00:00	True	True	True	True	True	True	True	True	True
2024-08-02 17:00:00+00:00	True	True	True	True	True	True	True	True	True
2024-08-02 17:10:00+00:00	True	True	True	True	True	True	True	True	True
2024-08-02 17:20:00+00:00	True	True	True	True	True	True	True	True	True

Visualization of the data

def plot_weather_parameters(df, stations, parameters):
    fig, ax = plt.subplots(len(parameters), figsize=(12, 7))

    time_freq = mdates.date2num(df.index[1]) - mdates.date2num(df.index[0])
    time_freq_minutes = int(time_freq * 24 * 60)

    colors = matplotlib.colormaps["tab10"]

    for i, param in enumerate(parameters):
        for station in stations:
            ax[i].plot(
                df.index,
                df[station, param],
                label=f"{station} {param}",
                color=colors(stations.index(station)),
            )
        ax[i].set_title(f"{param.capitalize()} ({time_freq_minutes} min)")
        ax[i].set_ylabel(f"{param.capitalize()} ({dataset.parameters_table['unit'][param]})")
        ax[i].legend()
        ax[i].grid(True)

    plt.tight_layout()
    plt.show()

plot_weather_parameters(df, stations=["ABO", "GRO", "KLO"], parameters=["temperature", "wind_speed", "humidity"])

Time series windowing

The get_windows function extracts sliding windows from the time series using a look-back window w and forecast horizon h. This prepares the data in a format that’s easy to use with framework-specific datasets and data loaders (e.g., PyTorch, TensorFlow, JAX).

This returns arrays of shape [n_w, w, n_s, n_f] for the inputs x and [n_w, h, n_s, n_f] for the outputs y, where n_w is the number of windows (or examples/samples), n_s is the number of stations and n_f is the number of features.

windows = dataset.get_windows(
    window_size=24,  # number of lookback time steps
    horizon_size=6,  # number of lead times to be predicted
    parameters=["temperature", "wind_speed", "humidity"],
    stations=["ABO", "KLO", "GRO", "LUG"],
)

# [num_windows, num_time_steps, num_stations, num_parameters]
print(f"Windows x shape: \t{windows.x.shape}")
print(f"Windows mask_x shape: \t{windows.mask_x.shape}")
print(f"Windows y shape: \t{windows.y.shape}")
print(f"Windows mask_y shape: \t{windows.mask_y.shape}")

Windows x shape:        (461923, 24, 4, 3)
Windows mask_x shape:   (461923, 24, 4, 3)
Windows y shape:        (461923, 6, 4, 3)
Windows mask_y shape:   (461923, 6, 4, 3)

Temporal re-sampling and NWP forecasts

We can initialize the dataset to resample the time series to a coarser frequency, such as from the default 10-minute interval to hourly. This is useful for reducing the variability of highly dynamic parameters like wind speed, aligning with lower temporal resolutions, or shortening sequence length for training.

Hourly data also allows us comparing with the forecasts from the numerical weather prediction (NWP) model, which are provided as hourly forecasts within PeakWeather.

Each parameter has a default aggregation method used during temporal resampling, which can be overridden using the aggregation_method parameter.

dataset = PeakWeatherDataset(
    freq="h",
    aggregation_methods={"temperature": "mean", "wind_gust": "max"},
    extended_nwp_pars=["temperature", "wind_u", "wind_v"],
    compute_uv=True,
)

temperature.zarr.zip: 1.28GB [00:26, 48.7MB/s]
wind_u.zarr.zip: 1.69GB [00:35, 47.5MB/s]
wind_v.zarr.zip: 1.70GB [00:34, 49.1MB/s]

Just like before, let’s look that temperature and wind speed for the stations of Adelboden, Grono and Kloten, this time with an hourly temporal resolution.

df, mask = dataset.get_observations(
    stations=["ABO", "GRO", "KLO"],
    parameters=["temperature", "wind_speed", "humidity"],
    first_date="2024-08-02 16:32",
    last_date="2024-08-06 23:26",
    return_mask=True,
)

plot_weather_parameters(df, stations=["ABO", "GRO", "KLO"], parameters=["temperature", "wind_speed", "humidity"])

Let’s now compare data from the stations with forecasts from the NWP model. The ICON-CH1-EPS NWP model produces 11 ensemble members for an horizon of up to 33 hours ahead. The NWP model is initialized every 3 hours and by setting split="nwp_test", we can generate a subset of the test set with the windows aligned to the NWP model.

Other options are split="train" or split="test". In those cases, only observations windows will be produced, with a time delta between them that depends on the selected dataset frequency.

windows = dataset.get_windows(
    window_size=24,  # number of lookback time steps
    horizon_size=24,  # number of lead times to be predicted
    stations=["KLO", "ABO"],
    split="nwp_test",
    parameters=["temperature", "wind_u", "wind_v"],
    nwp_parameters=["temperature", "wind_u", "wind_v"],
    drop_extra_y_pars=False,
)

# [num_windows, num_time_steps, num_stations, num_parameters]
print(f"""
Observations x shape:       \t{windows.x.shape}
Observations mask_x shape:  \t{windows.mask_x.shape}
Observations index_x shape: \t{windows.index_x.shape}
Observations y shape:       \t{windows.y.shape}
Observations mask_y shape:  \t{windows.mask_y.shape}
Observations index_y shape: \t{windows.index_y.shape}\n
""")

# [num_windows, num_time_steps, num_stations, num_parameters, num_ensemble_members]
print(f"NWP model fcasts shape:    \t{windows.nwp.shape}\n")

print(f"""
{windows.x.shape[0]}\tNumber of windows
{windows.x.shape[1]}\tLookback window size
{windows.y.shape[1]}\tLead times to predict
{windows.x.shape[2]}\tNumber of considered stations
{windows.x.shape[3]}\tNumber of observed parameters
{windows.nwp.shape[3]}\tNumber of NWP forecasted parameters
{windows.nwp.shape[-1]}\tNumber of NWP ensemble members\n
""")

print(
    f"{(windows.index_x[1][0] - windows.index_x[0][0]).total_seconds() / 3600}h"
    "\t Time delta between windows"
)

Observations x shape:           (3003, 24, 2, 3)
Observations mask_x shape:      (3003, 24, 2, 3)
Observations index_x shape:     (3003, 24)
Observations y shape:           (3003, 24, 2, 3)
Observations mask_y shape:      (3003, 24, 2, 3)
Observations index_y shape:     (3003, 24)


NWP model fcasts shape:         (11, 3003, 24, 2, 3)


3003    Number of windows
24      Lookback window size
24      Lead times to predict
2       Number of considered stations
3       Number of observed parameters
2       Number of NWP forecasted parameters
3       Number of NWP ensemble members


3.0h     Time delta between windows

fig, ax = plt.subplots(2, figsize=(10, 7))
colors = matplotlib.colormaps["tab10"]
w = 18
station = 1
index_to_station = {0: "KLO", 1: "ABO"}
# Plot temperature
ax[0].plot(
    windows.index_x[w],
    windows.x[w, :, station, 0],
    label=f"Input {index_to_station[station]} observations",
    color="black",
    linestyle="dashed",
)

for ens_member in range(windows.nwp.shape[0]):
    if ens_member == 0:
        ax[0].plot(
            windows.index_y[w],
            windows.nwp[ens_member, w, :, station, 0],
            label=f"NWP forecasts at {index_to_station[station]}",
            color=colors(0)
        )
    else:
        ax[0].plot(
            windows.index_y[w],
            windows.nwp[ens_member, w, :, station, 0],
            color=colors(0)
        )

ax[0].plot(
    windows.index_y[w],
    windows.y[w, :, station, 0],
    label=f"Target {index_to_station[station]} observations",
    color="black",
)
ax[0].set_title("Temperature")
ax[0].set_ylabel("Temperature (°C)")
ax[0].legend()
ax[0].xaxis.set_major_formatter(mdates.DateFormatter("%Y-%m-%d %H:%M"))
ax[0].xaxis.set_major_locator(mdates.AutoDateLocator())
ax[0].tick_params(axis="x", rotation=30)
ax[0].grid(True)

# Plot wind speed from u and v componented (eastward and northward)
ax[1].plot(
    windows.index_x[w],
    dataset.get_wind_speed(
        u=windows.x[w, :, station, 1], v=windows.x[w, :, station, 2]
    ),
    label=f"Input {index_to_station[station]} observations",
    color="black",
    linestyle="dashed",
)

for ens_member in range(windows.nwp.shape[0]):
    nwp_wind_speed = dataset.get_wind_speed(
        u=windows.nwp[..., 1], v=windows.nwp[..., 2]
    )
    if ens_member == 0:
        ax[1].plot(
            windows.index_y[w],
            nwp_wind_speed[ens_member, w, :, station],
            label=f"NWP forecasts at {index_to_station[station]}",
            color=colors(1)
        )
    else:
        ax[1].plot(
            windows.index_y[w],
            nwp_wind_speed[ens_member, w, :, station],
            color=colors(1)
        )

ax[1].plot(
    windows.index_y[w],
    dataset.get_wind_speed(
        u=windows.y[w, :, station, 1], v=windows.y[w, :, station, 2]
    ),
    label=f"Target {index_to_station[station]} observations",
    color="black",
)
ax[1].set_title("Wind speed (from u and v)")
ax[1].set_ylabel("Wind speed (m/s)")
ax[1].legend()
ax[1].xaxis.set_major_formatter(mdates.DateFormatter("%Y-%m-%d %H:%M"))
ax[1].xaxis.set_major_locator(mdates.AutoDateLocator())
ax[1].tick_params(axis="x", rotation=30)
ax[1].grid(True)

plt.tight_layout()
plt.show()

PeakWeather can also output sliding windows as Xarray datasets, allowing the data to remain tightly coupled with its coordinates for easier indexing and analysis.

icon_windows = dataset.get_windows(
    window_size=24,
    horizon_size=24,
    stations=["KLO", "ABO"],
    split="nwp_test",
    parameters=["temperature", "wind_u", "wind_v"],
    nwp_parameters=["temperature", "wind_u", "wind_v"],
    as_xarray=True,
).nwp

icon_windows

<xarray.Dataset> Size: 19MB
Dimensions:      (reftime: 3003, lead: 24, nat_abbr: 2, realization: 11)
Coordinates:
  * reftime      (reftime) datetime64[ns] 24kB 2024-10-02 ... 2025-10-13
  * lead         (lead) timedelta64[ns] 192B 01:00:00 ... 1 days 00:00:00
  * nat_abbr     (nat_abbr) <U5 40B 'KLO' 'ABO'
  * realization  (realization) int64 88B 0 1 2 3 4 5 6 7 8 9 10
Data variables:
    temperature  (reftime, lead, nat_abbr, realization) float32 6MB 12.39 ......
    wind_u       (reftime, lead, nat_abbr, realization) float32 6MB 0.1056 .....
    wind_v       (reftime, lead, nat_abbr, realization) float32 6MB 0.9805 .....
Attributes:
    LICENSE:    CC-BY-4.0
    parameter:  Air temperature, Eastward wind, Northward wind.
    period:     2017-01-01:2025-10-13
    source:     MeteoSwiss (ICON-CH1-EPS)
    time_zone:  UTC

xarray.Dataset

• Dimensions:
  • reftime: 3003
  • lead: 24
  • nat_abbr: 2
  • realization: 11
• Coordinates: (4)
  • reftime
    (reftime)
    datetime64[ns]
    2024-10-02 ... 2025-10-13

    array(['2024-10-02T00:00:00.000000000', '2024-10-02T03:00:00.000000000',
           '2024-10-02T06:00:00.000000000', ..., '2025-10-12T18:00:00.000000000',
           '2025-10-12T21:00:00.000000000', '2025-10-13T00:00:00.000000000'],
          shape=(3003,), dtype='datetime64[ns]')

  • lead
    (lead)
    timedelta64[ns]
    01:00:00 ... 1 days 00:00:00

    array([ 3600000000000,  7200000000000, 10800000000000, 14400000000000,
           18000000000000, 21600000000000, 25200000000000, 28800000000000,
           32400000000000, 36000000000000, 39600000000000, 43200000000000,
           46800000000000, 50400000000000, 54000000000000, 57600000000000,
           61200000000000, 64800000000000, 68400000000000, 72000000000000,
           75600000000000, 79200000000000, 82800000000000, 86400000000000],
          dtype='timedelta64[ns]')

  • nat_abbr
    (nat_abbr)
    <U5
    'KLO' 'ABO'

    array(['KLO', 'ABO'], dtype='<U5')

  • realization
    (realization)
    int64
    0 1 2 3 4 5 6 7 8 9 10

    array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10])

• Data variables: (3)
  • temperature
    (reftime, lead, nat_abbr, realization)
    float32
    12.39 12.33 12.49 ... 6.494 5.208

    unit :

    Celsius

    array([[[[12.39447  , 12.3315735, 12.486664 , ..., 12.478241 ,
              11.595825 , 13.185699 ],
             [11.352417 , 11.2778015, 11.448517 , ..., 11.343903 ,
              10.832092 , 11.161224 ]],
    
            [[12.417175 , 11.90033  , 12.466217 , ..., 12.34259  ,
              11.270874 , 12.828918 ],
             [10.033325 , 10.3573   ,  9.980804 , ...,  9.958252 ,
               9.6536255,  9.916748 ]],
    
            [[11.997314 , 10.96402  , 12.21109  , ..., 11.69812  ,
              11.425446 , 12.549652 ],
             [ 9.436707 , 10.159271 ,  9.841888 , ...,  9.149719 ,
               9.318451 ,  9.72879  ]],
    
            ...,
    
            [[ 9.719971 , 10.075012 ,  9.332947 , ...,  9.63739  ,
               7.9854126,  9.055969 ],
             [ 5.9054565,  5.941162 ,  5.57605  , ...,  5.8287354,
    ...
             [ 1.9482727,  1.8123169,  3.2970276, ...,  4.2848816,
               4.210663 ,  3.437378 ]],
    
            ...,
    
            [[ 9.119446 , 10.139954 ,  9.935913 , ...,  9.059937 ,
               9.272614 ,  9.16568  ],
             [ 6.994385 ,  6.324463 ,  5.927063 , ...,  7.0769653,
               6.7413025,  6.3589783]],
    
            [[ 8.497803 ,  9.439728 ,  9.233215 , ...,  8.311584 ,
               8.995056 ,  8.668579 ],
             [ 6.4318237,  6.557831 ,  5.255615 , ...,  6.3203125,
               6.642456 ,  6.2197876]],
    
            [[ 8.140106 ,  8.903351 ,  8.675171 , ...,  7.8661804,
               8.94812  ,  8.435211 ],
             [ 5.277252 ,  7.2450867,  5.026184 , ...,  6.5594788,
               6.4939575,  5.207611 ]]]],
          shape=(3003, 24, 2, 11), dtype=float32)

  • wind_u
    (reftime, lead, nat_abbr, realization)
    float32
    0.1056 1.32 ... -0.004663 -0.1612

    unit :

    m/s

    array([[[[ 1.05593748e-01,  1.31973469e+00,  1.92507543e-02, ...,
               1.18351185e+00,  1.73285615e+00,  1.16441429e+00],
             [ 8.83410633e-01,  2.08902311e+00,  1.27283442e+00, ...,
               7.48575211e-01,  3.14314067e-01,  3.04182386e+00]],
    
            [[ 2.24287295e+00,  2.44967270e+00,  8.95613492e-01, ...,
               3.05198717e+00,  2.53871632e+00,  2.37895536e+00],
             [ 1.89828396e-01, -7.85575509e-01, -8.75264168e-01, ...,
              -6.84412181e-01, -1.07066381e+00, -3.82492065e-01]],
    
            [[ 2.57831526e+00,  2.79495430e+00,  1.87868655e+00, ...,
               4.91730738e+00,  2.65069485e+00,  2.23019028e+00],
             [ 1.63980454e-01, -2.64358789e-01, -1.60281420e-01, ...,
               4.84399125e-02, -9.65699017e-01,  3.57032597e-01]],
    
            ...,
    
            [[-2.53415585e-01,  1.31669074e-01, -3.45357269e-01, ...,
              -1.74989402e-01,  3.94368470e-01, -4.34733838e-01],
             [-5.03578842e-01, -3.68093848e-01, -5.46747208e-01, ...,
    ...
             [ 1.56278104e-01, -6.50587827e-02,  2.57337153e-01, ...,
               1.37796089e-01, -1.20016493e-01,  5.61462753e-02]],
    
            ...,
    
            [[ 6.95043385e-01,  1.63107908e+00,  8.15451920e-01, ...,
               4.89257276e-01,  1.06548667e+00,  9.59462941e-01],
             [ 2.09125459e-01, -1.52243182e-01,  2.99225062e-01, ...,
               4.94545639e-01,  4.11954433e-01, -2.52443906e-02]],
    
            [[ 1.16918802e+00,  1.04401934e+00,  8.38310182e-01, ...,
               9.44927931e-01,  8.91971886e-01,  9.42664802e-01],
             [-6.63717464e-02,  1.89769566e-02, -6.80902302e-02, ...,
               3.75799328e-01, -1.36862606e-01,  1.51633397e-02]],
    
            [[ 1.01285696e+00,  6.53321564e-01,  1.29664326e+00, ...,
               1.05189562e+00,  8.51567984e-01,  1.26235414e+00],
             [-4.36559245e-02,  3.21070403e-01, -1.91405311e-01, ...,
               3.21752459e-01, -4.66327742e-03, -1.61183521e-01]]]],
          shape=(3003, 24, 2, 11), dtype=float32)

  • wind_v
    (reftime, lead, nat_abbr, realization)
    float32
    0.9805 0.05836 ... -0.12 0.07474

    unit :

    m/s

    array([[[[ 9.80516911e-01,  5.83623685e-02,  1.65535355e+00, ...,
               6.67569876e-01, -5.05961001e-01,  1.45959795e+00],
             [ 2.34656954e+00,  2.29104948e+00,  1.71622372e+00, ...,
               2.67293549e+00,  2.88092899e+00,  1.22632742e+00]],
    
            [[ 5.94174206e-01, -1.41338766e-01,  4.69693154e-01, ...,
               5.45814276e-01,  1.19907808e+00,  1.00468075e+00],
             [ 1.53220475e+00,  3.03296566e+00,  1.95059836e+00, ...,
               2.42263794e+00,  2.83574486e+00,  1.43073285e+00]],
    
            [[-1.43525195e+00, -1.63397551e+00, -8.24894071e-01, ...,
              -2.95301765e-01, -1.24631321e+00, -1.00014758e+00],
             [-5.76261878e-01,  1.86911500e+00,  1.79322883e-01, ...,
              -2.20061198e-01,  1.21987653e+00, -6.28389895e-01]],
    
            ...,
    
            [[-5.07508814e-01, -9.11781847e-01, -1.74083650e-01, ...,
              -9.54754055e-01, -8.52862477e-01, -4.81635094e-01],
             [-5.26523665e-02, -6.74390733e-01, -2.77386248e-01, ...,
    ...
             [-8.81975703e-03,  1.06652193e-01, -1.05682209e-01, ...,
               8.83272365e-02,  6.01558539e-04,  1.15962394e-01]],
    
            ...,
    
            [[-1.13143647e+00, -6.06379092e-01, -5.04909575e-01, ...,
              -9.85798717e-01, -7.39710033e-01, -1.54179811e+00],
             [ 2.90912092e-01, -2.71023363e-02, -7.16824532e-02, ...,
               5.29371738e-01,  3.48246247e-01,  2.61015147e-01]],
    
            [[-1.06447589e+00, -9.77661014e-01, -6.19859040e-01, ...,
              -4.89928961e-01, -1.14697075e+00, -2.13117409e+00],
             [-2.67029088e-02, -2.25271568e-01, -1.98731825e-01, ...,
               1.50959045e-01,  1.48656130e-01,  1.86846759e-02]],
    
            [[-1.25979209e+00, -1.10429311e+00, -6.04349494e-01, ...,
              -8.41145813e-01, -8.98327053e-01, -1.89175034e+00],
             [-1.18726432e-01, -2.32368156e-01, -4.72060554e-02, ...,
              -1.80072144e-01, -1.19953848e-01,  7.47393817e-02]]]],
          shape=(3003, 24, 2, 11), dtype=float32)

• Indexes: (4)
  • reftime
    PandasIndex

    PandasIndex(DatetimeIndex(['2024-10-02 00:00:00', '2024-10-02 03:00:00',
                   '2024-10-02 06:00:00', '2024-10-02 09:00:00',
                   '2024-10-02 12:00:00', '2024-10-02 15:00:00',
                   '2024-10-02 18:00:00', '2024-10-02 21:00:00',
                   '2024-10-03 00:00:00', '2024-10-03 03:00:00',
                   ...
                   '2025-10-11 21:00:00', '2025-10-12 00:00:00',
                   '2025-10-12 03:00:00', '2025-10-12 06:00:00',
                   '2025-10-12 09:00:00', '2025-10-12 12:00:00',
                   '2025-10-12 15:00:00', '2025-10-12 18:00:00',
                   '2025-10-12 21:00:00', '2025-10-13 00:00:00'],
                  dtype='datetime64[ns]', name='reftime', length=3003, freq=None))

  • lead
    PandasIndex

    PandasIndex(TimedeltaIndex(['0 days 01:00:00', '0 days 02:00:00', '0 days 03:00:00',
                    '0 days 04:00:00', '0 days 05:00:00', '0 days 06:00:00',
                    '0 days 07:00:00', '0 days 08:00:00', '0 days 09:00:00',
                    '0 days 10:00:00', '0 days 11:00:00', '0 days 12:00:00',
                    '0 days 13:00:00', '0 days 14:00:00', '0 days 15:00:00',
                    '0 days 16:00:00', '0 days 17:00:00', '0 days 18:00:00',
                    '0 days 19:00:00', '0 days 20:00:00', '0 days 21:00:00',
                    '0 days 22:00:00', '0 days 23:00:00', '1 days 00:00:00'],
                   dtype='timedelta64[ns]', name='lead', freq='h'))

  • nat_abbr
    PandasIndex

    PandasIndex(Index(['KLO', 'ABO'], dtype='object', name='nat_abbr'))

  • realization
    PandasIndex

    PandasIndex(Index([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10], dtype='int64', name='realization'))

• Attributes: (5)

  LICENSE :

  CC-BY-4.0

  parameter :

  Air temperature, Eastward wind, Northward wind.

  period :

  2017-01-01:2025-10-13

  source :

  MeteoSwiss (ICON-CH1-EPS)

  time_zone :

  UTC

Static attributes

Each station is either a rain_gauge, which measures precipitation (and sometimes temperature), or a meteo_station station, which records multiple weather parameters. Every station has a known geographic location, an elevation above sea level, and additional static attributes derived from topographic data interpolated to its position.

The full topographic features at a \(50m\) resolution can be loaded by initializing the PeakWeatherDataset dataset with the extended_topo_vars parameter.

dataset.stations_table

	station_name	latitude	longitude	station_height	swiss_easting	swiss_northing	ASPECT_2000M_SIGRATIO1	WE_DERIVATIVE_2000M_SIGRATIO1	TPI_2000M	SN_DERIVATIVE_10000M_SIGRATIO1	dem	SN_DERIVATIVE_2000M_SIGRATIO1	SLOPE_10000M_SIGRATIO1	ASPECT_10000M_SIGRATIO1	SLOPE_2000M_SIGRATIO1	STD_2000M	STD_10000M	TPI_10000M	WE_DERIVATIVE_10000M_SIGRATIO1	station_type
nat_abbr
ABE	Aarberg	47.057969	7.285350	444.00	2.588355e+06	1.211894e+06	310.688416	0.010961	-6.219757	-0.011349	442.315613	-0.009424	0.872107	318.209137	0.828149	0.000000	44.313626	-38.296173	0.010144	rain_gauge
ABO	Adelboden	46.491703	7.560703	1321.38	2.609372e+06	1.148939e+06	124.863129	-0.167379	-53.303345	-0.026807	1317.771851	0.116605	1.702339	25.582031	11.529667	120.940262	374.830623	-443.723877	-0.012833	meteo_station
AEG	Oberägeri	47.133636	8.608206	724.43	2.688729e+06	1.220956e+06	190.899567	0.010865	-20.030273	-0.013760	724.173462	0.056423	1.099379	315.811829	3.288559	27.269887	141.396220	-165.792114	0.013376	meteo_station
AFI	Andelfingen	47.604669	8.670289	360.00	2.692617e+06	1.273392e+06	193.235062	0.004745	-12.622437	-0.001833	357.506256	0.020173	0.299592	290.519653	1.187198	0.000000	31.905980	-55.071655	0.004897	rain_gauge
AGATT	Attelwil	47.265233	8.050519	475.00	2.646308e+06	1.235106e+06	96.557785	-0.020930	-14.010315	-0.009250	474.737488	0.002406	0.593415	333.268860	1.206895	0.000000	75.500339	-105.056763	0.004659	rain_gauge
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
WYN	Wynau	47.255025	7.787475	421.99	2.626406e+06	1.233849e+06	319.106384	0.021612	-16.944305	-0.002912	421.595306	-0.024955	0.167600	5.381668	1.890799	19.067440	16.002325	-33.209930	-0.000274	meteo_station
ZER	Zermatt	46.029272	7.752433	1638.35	2.624297e+06	1.097574e+06	126.072998	-0.222615	-173.187622	0.023503	1645.671875	0.162173	2.437473	123.514168	15.398737	213.252085	469.596215	-791.969727	-0.035491	meteo_station
ZEV	Zervreila	46.578797	9.118797	1738.00	2.728781e+06	1.159992e+06	109.527702	-0.111410	-240.550537	-0.028125	1738.876587	0.039513	2.217149	43.410706	6.741610	146.893540	324.236704	-571.910034	-0.026606	rain_gauge
ZWE	Zweisimmen	46.550511	7.384917	936.00	2.595880e+06	1.155471e+06	273.443573	0.167828	-93.677673	-0.005795	939.551575	-0.010099	2.334606	278.171326	9.543953	118.914438	316.598639	-477.413879	0.040355	rain_gauge
ZWK	Zwillikon	47.290092	8.431958	461.00	2.675139e+06	1.238164e+06	166.458527	-0.001038	-20.879974	0.011795	461.351562	0.004308	1.648120	245.798660	0.253896	0.000000	93.148896	-49.712555	0.026244	rain_gauge

302 rows × 20 columns

Detailed Dataset Initialisation

In the previous section, the dataset was initialised by relying on certain default parameters. Below you can find an example of more detailed initialisation. Refer to the official documentation for additional details.

ds = PeakWeatherDataset(
        root=None,  # Path to the dataset
        pad_missing_values=True,  # Pad missing values with NaN
        years=None,  # Years to include in the dataset (None for all)
        parameters=None,  # Parameters to include in the dataset (None for all)
        extended_topo_vars="none",  # Optional extended topographic variables
        extended_nwp_pars="none",  # Optional extended NWP model (ICON) variables
        imputation_method="zero",  # Method for imputing missing values
        freq="h",  # Frequency of the data (e.g., "h" for hourly)
        compute_uv=True,  # Compute u and v (zonal and meridional) components of wind
        station_type="meteo_station",  # Which station type to load (None for all)
        aggregation_methods={'temperature': 'mean'} # Use specific temporal aggregation
    )

Topographic features

The dataset comes with a digital elevation model (DEM) and topographic features at a \(50m\) resolution. They can be downloaded with the extended_topo_vars parameter. The values are provided with the swiss EPSG:2056 coordinate system. As an example, let’s look at the DEM and aspect of the terrain over the swiss radar domain.

ds = PeakWeatherDataset(
        extended_topo_vars=["DEM", "ASPECT_10000M_SIGRATIO1" ]
)

ds.load_topography()['topo_DEM']

ASPECT_10000M_SIGRATIO1.zarr.zip: 417MB [00:08, 48.2MB/s]
DEM.zarr.zip: 557MB [00:10, 51.2MB/s]

<xarray.Dataset> Size: 863MB
Dimensions:  (y: 14002, x: 15402)
Coordinates:
  * x        (x) float64 123kB 2.225e+06 2.225e+06 ... 2.995e+06 2.995e+06
  * y        (y) float64 112kB 8.1e+05 8.1e+05 8.101e+05 ... 1.51e+06 1.51e+06
Data variables:
    dem      (y, x) float32 863MB ...
Attributes:
    crs:                 EPSG:2056
    description:         Topographic descriptors computed from a blend of Dig...
    domain:              Swiss radar domain
    history:             Produced by topo-workflow
    source:              DHM25, DTM, NASADEM
    spatial_resolution:  50 m

xarray.Dataset

• Dimensions:
  • y: 14002
  • x: 15402
• Coordinates: (2)
  • x
    (x)
    float64
    2.225e+06 2.225e+06 ... 2.995e+06

    array([2224975., 2225025., 2225075., ..., 2994925., 2994975., 2995025.],
          shape=(15402,))

  • y
    (y)
    float64
    8.1e+05 8.1e+05 ... 1.51e+06

    array([ 809975.,  810025.,  810075., ..., 1509925., 1509975., 1510025.],
          shape=(14002,))

• Data variables: (1)
  • dem
    (y, x)
    float32
    ...

    crs :

    EPSG:2056

    units :

    m

    [215658804 values with dtype=float32]

• Indexes: (2)
  • x
    PandasIndex

    PandasIndex(Index([2224975.0, 2225025.0, 2225075.0, 2225125.0, 2225175.0, 2225225.0,
           2225275.0, 2225325.0, 2225375.0, 2225425.0,
           ...
           2994575.0, 2994625.0, 2994675.0, 2994725.0, 2994775.0, 2994825.0,
           2994875.0, 2994925.0, 2994975.0, 2995025.0],
          dtype='float64', name='x', length=15402))

  • y
    PandasIndex

    PandasIndex(Index([ 809975.0,  810025.0,  810075.0,  810125.0,  810175.0,  810225.0,
            810275.0,  810325.0,  810375.0,  810425.0,
           ...
           1509575.0, 1509625.0, 1509675.0, 1509725.0, 1509775.0, 1509825.0,
           1509875.0, 1509925.0, 1509975.0, 1510025.0],
          dtype='float64', name='y', length=14002))

• Attributes: (6)

  crs :

  EPSG:2056

  description :

  Topographic descriptors computed from a blend of Digital Elevation Models

  domain :

  Swiss radar domain

  history :

  Produced by topo-workflow

  source :

  DHM25, DTM, NASADEM

  spatial_resolution :

  50 m

import matplotlib.pyplot as plt

da = ds.load_topography()['topo_DEM']["dem"][::5, ::5]  # skip every 5 pixels

fig, ax = plt.subplots(1,2, figsize=(12, 5))
ax[0].imshow(
    da.values,
    origin="lower",
    extent=[
        da.x.min().item(),
        da.x.max().item(),
        da.y.min().item(),
        da.y.max().item(),
    ],
    aspect="auto",
)

ax[0].set_xlabel("x")
ax[0].set_ylabel("y")
ax[0].set_title("DEM")

da = ds.load_topography()['topo_ASPECT_10000M_SIGRATIO1']['ASPECT_10000M_SIGRATIO1'][::5, ::5]  # skip every 5 pixels
ax[1].imshow(
    da.values,
    origin="lower",
    extent=[
        da.x.min().item(),
        da.x.max().item(),
        da.y.min().item(),
        da.y.max().item(),
    ],
    aspect="auto",
)


ax[1].set_xlabel("x")
ax[1].set_ylabel("y")
ax[1].set_title("ASPECT")

plt.show()