Upload
elijah-stephens
View
216
Download
2
Embed Size (px)
Citation preview
1
CloudNet data comparison
These slides demonstrate the suggested comparison strategy for CloudNet data, although they are applicable for other model - observation comparisons.
They should be viewed with the CloudNet User Requirement Document.
Damian Wilson, Met Office and Jean-Marcel Piriou, Meteo-France
2
The problem
1
2
3
4
IWC
LWC
CF
Z
A
B
C
ModelsWell defined Model-like fields
Well defined Obs-like fields
Observation sites
How do we transform between fields?
Processing to combine ice categories etc.
Processing to remove clutter etc.
3
Observations to models
1
2
3
4
IWC
LWC
CF
Z
A
B
C
ModelsWell defined Model-like fields
Well defined Obs-like fields
Observation sites
We could transform obs-like fields into model-like fields
Algorithms
Assumptions
4
Models to observations
1
2
3
4
IWC
LWC
CF
Z
A
B
C
ModelsWell defined Model-like fields
Well defined Obs-like fields
Observation sites
Or from models to observations. The algorithms might not be reversible.
A-1 A I
New algorithms, assumptions
5
Direct comparison
1
2
3
4
IWC
LWC
CF
Z
A
B
C
ModelsWell defined Model-like fields
Well defined Obs-like fields
Observation sites
It might be possible to transform directly from the model, but not for all models and obs fields.
No new assumptions
6
Unavailable information
1
2
3
4
IWC
LWC
CF
Z
A
B
C
ModelsWell defined Model-like fields
Well defined Obs-like fields
Observation sites
If data is absent then different transforms are required.
New algorithms, assumptions.
7
Sources of error
1
2
3
4
IWC
LWC
CF
Z
A
B
C
ModelsWell defined Model-like fields
Well defined Obs-like fields
Observation sites
Error in transformsInitial
conditions and forward model can produce errors
Measurement and estimation error
CloudNet wishes to assess forward model errors.
8
Recommendations Sets of well defined quantities should be identified,
which correspond closely with variables available in models and measurements available from observing sites.
Models and observations should store data in their processed state.
Algorithms should be developed to transform in either direction.
These algorithms are not necessarily reversible.
9
Recommendations
Algorithms should transform between variables which are readily available from different types of models and observation sites, so transforms are not site or model specific. This will help future comparisons with other sites and models.
If a piece information is not available from a model or site then a different algorithm must be developed.
10
Recommendations Comparisons can be carried out in both model
and observation space by using the transforms. Each would provide different sorts of information.
It may be possible in specific, limited circumstances to transform directly from a model to observations with the addition of no new assumptions. Such comparison is also of value and a model should supply information to do this if this is possible.
11
Recomendations
Errors should be assessed for each part of the comparison.