What makes a "sample" (i.e. those that should be analysed together)?

[agstephens]

• dachar version: *
• Python version: *
• Operating System: Linux

Description

We are analysing groups of Datasets together in samples of the overall population. Here are some thoughts on what makes a sample. They are expressed as the facets that are changing where all others stay the same...

CMIP5:

• product: to extend/merge between output1 and output2
• institute & model: they are essentially a single facet spread across two facets.
• experiment: compare or merge data across experiments. This would need specific subsets rather than __all__, because only some will make sense, e.g.: rcp*, or historical+rcp45.
• ensemble member: to analysis across different ensemble members.
• frequency: to look at one variable across different time frequencies.
• variable: to analyse multiple variables

CORDEX:

• domain: to compare one variable projected on to comparable domains, such as AFR-44 and AFR-44i
• institute & driving model: they are essentially a single facet spread across two facets.
• rcm model: to compare the outputs from different RCM models
• institute & driving model & rcm model: to look across the ensemble of both driving and RCM models.
• experiment: compare or merge data across experiments. This would need specific subsets rather than __all__, because only some will make sense, e.g.: rcp*, or historical+rcp45.
• ensemble member: to analysis across different ensemble members.
• frequency: to look at one variable across different time frequencies.
• variable: to analyse multiple variables

CMIP6:

• institute & source: they are essentially a single facet spread across two facets.
• experiment: compare or merge data across experiments. This would need specific subsets rather than __all__, because only some will make sense, e.g.: rcp*, or historical+rcp45.
• ensemble member: to analysis across different ensemble members.
• frequency: to look at one variable across different time frequencies.
• variable: to analyse multiple variables

NOTES about CMIP6:

• I don't think that we need to run the analyses over different activities (i.e. "MIPs"), because the experiments will cover all the variation.