Quickstart

A dartobsgen run has three pieces: a ObsGenConfig saying when and where, a DataSource saying where the observations come from, and generate_obs_sequences() to drive one window at a time.

import datetime
from dartobsgen import ObsGenConfig, CrocLakeSource, generate_obs_sequences

config = ObsGenConfig(
    start=datetime.datetime(2010, 5, 1),   # model run start; first analysis is 06Z
                                           # (or say it directly: first_analysis=...)
    end=datetime.datetime(2010, 5, 3),     # last analysis time (inclusive)
    lat_min=5,   lat_max=60,
    lon_min=-100, lon_max=-30,
    obs_types=["ARGO_TEMPERATURE", "ARGO_SALINITY"],
    assimilation_frequency=datetime.timedelta(hours=6), # assimilate every 6 hours
    output_dir="./obs_output",
)

source = CrocLakeSource(
    crocolake_path="/path/to/crocolake/",
    dart_path="/path/to/DART/",
)

# Sequential
written_files = generate_obs_sequences(config, source)

# Parallel (all CPUs)
written_files = generate_obs_sequences(config, source, max_workers=None)

# Parallel (fixed number of workers)
written_files = generate_obs_sequences(config, source, max_workers=4)

print(written_files)

generate_obs_sequences returns the paths of the files it actually wrote. Windows containing no observations are silently skipped, so the returned list can be shorter than the number of analysis times.

Where to go next

• Time windows — the start vs first_analysis distinction, and how each window is bounded. Worth reading before your first real run.

• Output file naming — the {S} seconds-of-day token and custom timestamp formats.

• Observation types — the three naming styles obs_types accepts, and the supported types.

• Parallel generation — including the if __name__ == "__main__": requirement.

• Data sources — CrocoLake, NNJA-AI, and synthetic observations.