Data Denial Experiment - Remove Observations from Macquarie Island

Background

Macquarie Island is an important observation location as demonstrated by FSOI results. As a complement to FSOI Chris Tingwell (ref?) carried out a test by removing the observation from the island. He saw very little change in the forecast. This is a single-cycle test.

Targeted observation vs. Kelly et al. (200?)

Limitation: In all the FSOI tests in which Macquarie Island was identified as having relative large impact energy norm was computed for Australian domain. Will this result carry over if the norm area is South Pacific, South America, etc.?

Experimental set-up

Control

• Suite ID: u-be286
• sonde and surface varobs files contain observations for the call sign, 94998

Experiment

• Suite ID: u-be334
• From sonde and surface stationlists rejected all observations from Macquarie Island (WMO ID 94998; BOM station number 300004; 54.6208° S, 158.8556° E)
• Macquarie Island seems to do 2 radiosonde observations a day: 11 and 23 (Check????)
• sonde varobs file contains MDI for the call sign, 94998
• surface varobs file does not contain observations for the call sign, 94998

Diary

u-be286

Cycle time	Failed task	Reason for failure	Action taken
20171215T0600Z	gl[mu]_ops_process_background_ahiclear	Not known	Reset to succeeded
20171231T1800Z	gl[mu]_ops_process_background_airs	stderr has,

Delayed replication with compression | Reset to succeeded |

u-be334

Cycle time	Failed task	Reason for failure	Action taken
20171215T0600Z	gl[mu]_ops_process_background_ahiclear	Not known	Reset to succeeded
20171231T1800Z	gl[mu]_ops_process_background_airs	stderr has,

Delayed replication with compression | Reset to succeeded |

Results

My working hypothesis is that areas of difference will propagate downstream with increasing forecast lead time,

• For a particular basetime sequence of forecast hours

  • At the start of the trial period (20171201T06 - 20180101T06) when the background was similar in both runs the forecast difference would be less. However for basetimes further away from the start basetime there's less memeory - i.e. background diverges between 2 runs and hence analysis difference would be great and hence forecast difference would be greater

• Simply looking at difference for a particular basetime may not yield anything. Better way to bring out differences is to average all the differences for each forecast range. This is based on my hunch: at 00 forecast hour absence of Macquarie Island obs should produce differences around the Island. This difference will move eastward while being amplified