dtcenter / MET

Model Evaluation Tools
https://dtcenter.org/community-code/model-evaluation-tools-met
Apache License 2.0
74 stars 22 forks source link

Enhance MET library code to enable Point-Stat and Ensemble-Stat to apply filtering logic when encountering multiple observations in the time window at the same location. #104

Closed dwfncar closed 12 years ago

dwfncar commented 12 years ago

See MET-Help ticket: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=52626

When processing GDAS prepbufr observations and setting the quality marker to a value of 9, it appears that "duplicate" observations are coming out of PB2NC. This task is to investigate this issue more fully. It may be the case that when an observation occurs with multiple quality marks, all versions of the observation are being written to the output file, rather than the one at the top of the stack... but that's just a theory.

[MET-104] created by johnhg

dwfncar commented 12 years ago

Write a patch. Discussed with Jonathan Case.
Expect this may need another issue for follow-on work to put option in config file. by bonny

dwfncar commented 12 years ago

Tressa and Tara gave guidance on 4/16/2012 to add a new option to the point_stat and ensemble_stat config files. The new option, called duplicate_handling, would assume one of three values:
  'none' for the legacy behavior which did not try to detect duplicates
 
'unique' to detect and remove all but a single observation that are identical in every way except station id
 * 'single' to detect observations that are identical in every way but station id, valid time and observation value and use only the observation with the valid time closest to that of the model data
In the second two cases, using a verbosity level of 4 or above, a report of all duplicate observations will be printed, showing the one(s) that were used for verification. by pgoldenb

dwfncar commented 12 years ago

Most of the new code to detect and handle duplicates is in place in the PairBase class, which is being tested using hard-coded flag values for the unique and single features pending completion of the point_stat and ensemble_stat vx_config upgrade. The design for the reporting feature of the duplicate detector is still being worked out, with little or no code written yet for that feature. by pgoldenb

dwfncar commented 12 years ago

Added functionality to vx_statistics PairData to detect and handle duplicate point obs in two different ways and wired functionality into point_stat and ensemble_stat. The duplicate functionality is controlled using the duplicate_flag in the config file for each tool, with three options supported: none, unique and single. by pgoldenb