Data sources

A data source answers one question: given an assimilation window, write the observations that fall inside it to an obs_seq file. Everything else — the window arithmetic, the file naming, the parallelism — is generate_obs_sequences()’ job, so swapping sources changes one object and nothing else.

Source	Observations from	Status
CrocLakeSource	CrocoLake parquet database (ARGO, GLODAP, SprayGliders)	supported
NNJASource	NNJA-AI cloud archive (NOAA/NASA Joint Archive)	under development
PerfectModelSource	synthetic, via DART’s perfect_model_obs	under development
ObsSeqSource	a bank of existing obs_seq files	stub

To write your own, see Adding a new data source.

• CrocoLake
  • Usage
  • Observation types
  • Parallelism
• NNJA-AI
  • Install the extra dependency
  • Usage
  • Supported NNJA obs types
  • Custom error variances
  • GCS mirror options
• Synthetic observations via perfect_model_obs
  • Prerequisites
  • ObsNetworkEntry fields
  • Driving it from model output: state_provider
  • Controlling observation time within a window
  • MOM6 example
  • Parallel execution
• Adding a new data source
  • Optional: check_coverage