Diva QC

From GHER

Jump to: navigation, search

We propose a four-step approach for the quality control of EMODNET and SeaDataNet products obtained with DIVA.

The first checks can be routinely performed and do not require a specific knowledge of the studied region.

Contents

Internal consistency check

These tests aims to detect incorrect analysis parameters.

  • Check the vertical and temporal coherence of:
    • the correlation length L
    • the signal-to-noise ratio λ
    • the analyzed and background fields, through simple statistics (mean value, mean gradient, etc)

Large variations may be the manifestation of a wrong evaluation of the analysis parameters during the application of DIVA.

  • Compare the mean distance between observation and compare it with L
  • Compute RMS between two successive layers (depth and time)

Simple visual checks (by non-experts)

These tests also allows one to detect incorrect analysis parameters, but also isolated outliers. (A mechanism to report the outliers to the data provider might be an idea to develop)

  • Visualize analyzed and error field
  • Check the misfit between data and analysis

Consistency with a priori knowledge

These tests require a knowledge concerning the studied variables.

  • Check the range of the field value and gradient compared to the range of data (if enough), or with values by experts. The range can be a function of:
    • depth
    • time of the year (month, season)
    • location (basin, ocean, etc)
  • Compare the regional product with global climatology World Ocean Atlas, if the studied variables are present in the WOA.
  • Check the correlation between WOA and Emodnet parameters

Validation by regional experts

  • Visual validation of various horizontal and vertical sections
  • Comparison of data products with information from scientific literature
  • Comparison with regional climatologies (if existing)


Validation work-flow (proposed)

  • The originator uploads the product to a test site.
  • He checks the product and possibly update the products.
  • Two regional experts (project members or not) are invited to comment on the products.
  • The data product maybe need to adapted to address the comments.
  • If two regional experts are satisfied, the data products will be transfered on the official OceanBrowser web-site.

Documents

  • Troupin, C. (2011). Detection of wrong analysis with Diva.

http://hdl.handle.net/2268/114186

Personal tools