Welcome, Guest
Username: Password: Remember me
Forum on operational GLAMEPS

TOPIC: possible data flow problem in HIRLAM sub-ensemble

possible data flow problem in HIRLAM sub-ensemble 7 years 1 month ago #947

  • Xiaohua Yang
  • Xiaohua Yang's Avatar
  • OFFLINE
  • Administrator
  • Posts: 195
  • Thank you received: 4
Last week when analysing behaviour of the recent GLAMEPS ensembles it came to our attention that, the HIRLAM sub-ensembles with both physical options (KF and STRACO) seem to suffer an obvious problem which seem to be associated with possible flaw in cycling data flow. The manifestation of the problem can be seen in the attached observation verification time series of control member of HIRLAM-KF and part of the KF-sub-ensemble members for T2m during October month, in which the T2m bias for all perturbed KF members are degrees colder than that of control member. (Only 7 perturbed ensemble members out of 12 are selected here for plot due to technical limitation. A separate test did show that other perturbed members show consistent tendencies as shown here).

This is clearly against expectation, and also in contrast to a more 'normal-looking' curves for ECEPS-ensemble members and for ALARO-subensemble members, all of these can be seen in the interface

hirlam.org/portal/glameps/WebgraF/ObsVer...l?choice_ind=Surface
(select T2m and 'daily error evolution, Oct, ECeps- alaro- kf-, st- subensembles)

It looks to me that there is something to correct with the current GLAMEPS suite. More work and discussion shall hopefully follow to address the issue, which shall lead to an improvement of the GLAMEPS ensemble forecast for T2m.
PS_00201210_00000000_ALL_TT_0-08da4aa4442d6c9e90b4a6deccb287b0.png

Re: possible data flow problem in HIRLAM sub-ensemble 7 years 4 weeks ago #964

  • Xiaohua Yang
  • Xiaohua Yang's Avatar
  • OFFLINE
  • Administrator
  • Posts: 195
  • Thank you received: 4
Two separate corrections, both of them presumably associated with the performance issue of T2m forecast in GLAMEPS, have now been detected and corrected yesterday for the daily GLAMEPS suite,
1. a script error in GLAMEPS HIRLAM-Fg that disabled surface assimination for perturbed members;
2. a source code change to enable results of surface assimilation for HIRLAM sub-ensemble members to have impact. Prior to this correction, the surface assimilation cycling with HIRLAM sub-ensembles has had no effect.

We'll continue to monitor and update the trend in T2m from GLAMEPS runs.

Re: possible data flow problem in HIRLAM sub-ensemble 7 years 3 weeks ago #966

  • Xiaohua Yang
  • Xiaohua Yang's Avatar
  • OFFLINE
  • Administrator
  • Posts: 195
  • Thank you received: 4
The following figour appear to confirm that the latest update/bug correction on surface assimilation cycling for HIRLAM subensembles in GLAMEPS has resulted an improved behaviour for T2m bias error. The following example plot show time series of the T2m forecast error for 7 perturbed STRACO ensemble members in comparison to control member (in red), with the correction took place on Nov 9. The gap between perturbed members and control for bias in T2m are now closing.

More of such examples are shown in the verification monitoring interface for GLAMEPS control members on
hirlam.org/portal/glameps/WebgraF/ObsVer...l?choice_ind=Surface
PS_00201211_00000000_ALL_TT_0.png
Time to create page: 0.071 seconds