Develop an AI/ML-based tool to assist in setting station-specific QA/QC thresholds for sensors (including water level, shore stations, etc.). The tool would establish thresholds and possibly quantify the confidence of those values. The tool could utilize historical sensor timeseries data, sensor characteristics, sensor metadata, nearby stations, and site-specific context, if available. 

Expected Outcomes

  • Prototype AI/ML tool generating station-specific QARTOD thresholds
  • Documented workflow that can be used with the prototype tool to test it on multiple stations

Skills required

  • Python
  • Machine learning / AI
  • Familiarity with QARTOD tests

Expertise

Intermediate

Topic Lead(s)

Aaron Mau, Marta Wolfshorndl

Relevant links

Functioning Prototype

TBD

Workflow

Tasks:

  1. Find example training datasets
  2. Choose potential models
  3. Create training workflow and reproduce results
  4. Scale up to more datasets

Deliverables:

  1. auto_thresh module in ioos_qc
    • This includes example model weights and coeffs, training code to export coeffs
  2. Walkthrough Jupyter notebook for building your own model
    • In docs/examples

Updates