Representing Ensembles

1
Representing Ensembles

• Set of standard name proposals by Paco
  Doblas-Reyes October 2006 to describe multi-model
  forecast ensembles
• Need to represent quantities such as
• source, e.g., HadCM3
• institution to specify where the model was
  developed
• experiment_id, e.g. say which IPCC scenario was
  used

2
Representing Ensembles

• Source and institution are already global
  attributes allow them also to be standard names
• Experiment_id is proposed as a standard name
  and as a new global attribute

3
Representing Ensembles

• There is general agreement that these types of
  metadata do need to be represented in CF
• There has been a huge amount of discussion as to
  how we go about it

4
Representing Ensembles

• Standard names vs Standard metadata
• External vocabularies
• Realization_weight
• Ensemble axis

5
Standard names or standard metadata?

• Proposed by Bryan Lawrence
• Standard names describe a physical quantity such
  as temperature, wind speed, i.e., something you
  measure or model
• The ensemble metadata are not physical variables
  they tell you how the physical data were
  obtained

6
Standard names or standard metadata?

• To distinguish between the two concepts we should
  introduce another managed vocabulary analogous
  to, but distinct from, standard names
• Call the new vocab standard metadata
• But can we always be clear about which vocab a
  new name should be placed in?
• If not, we should not introduce the new vocab

7
External vocabularies

• An extension of the standard metadata concept
• It is not sensible for CF to re-invent the wheel
  if well defined and maintained vocabularies exist
  elsewhere, e.g., IPCC emissions scenarios
• CF should point to the external vocab by giving a
  URL in the standard_metadata attribute
• The idea has generally been well received
  provided we are confident of the quality and
  continuing availability of the external vocab

8
External vocabularies

• An extension of the standard metadata concept
• It is not sensible for CF to re-invent the wheel
  if well defined and maintained vocabularies exist
  elsewhere, e.g., IPCC emissions scenarios
• CF should point to the external vocab by giving a
  URL in the standard_metadata attribute
• The idea has generally been well received
  provided we are confident of the quality and
  continuing availability of the external vocab

9
Realization_weight

• Proposed by Jamie Kettleborough, originally as a
  standard name modifier
• Realization is an ensemble member to which a
  weight is applied for a statistical process
• The weights would be contained in an ancillary
  variable so that realization and
  realization_weight can be associated
• Jonathan has proposed that this should in fact be
  a standard name

10
Ensemble axis

• The concept of an ensemble axis emerged early in
  the discussion
• This would be an axis whose dimension is equal to
  the number of ensemble members
• Ensemble metadata would be associated with the
  data using auxiliary coordinate variables with
  standard names of source, institution, etc.

11
Ensemble axis

• Steve Hankin and Balaji have both raised some
  concerns with how the ensemble axis can be
  applied to multi-model ensembles
• File size
• File aggregation
• More than one unlimited dimension?
• Loss of interoperability by moving to 5
  dimensions
• Need for regridding to a common grid

12
Ensemble axis

• However, there is now consensus that the concept
  of an ensemble axis can work for ensemble members
  on the same grid
• This allows many use cases to be represented,
  e.g., many initial condition ensembles and
  perturbed physics ensembles will be produced by
  the same model

13
Ensemble axis

• The axis will be string valued, therefore will
  not vary monotonically in the same way as a
  spatiotemporal axis
• Steve Hankin has proposed that the ensemble axis
  be clearly identified by an attribute such as
  axisensemble axis_typeenumerated
• Steve has further proposed that the discussion of
  multi-model ensemble metadata should move to
  trac/wiki