Understanding variant label in CMIP5 and CMIP6 

Ensembles (Variant label) in CMIP5 (CMIP6) have been assigned to help identify the different simulations within a set of experiments. To locate specific experiments out of the various CMIP5/CMIP6 simulations global attributes have been defined. These are:

• Experiment id: Experiment id is to identify Root experiment for the simulation. For example; historical, ssp scenarios (ssp126, ssp245, ssp370 etc.), dcppA-hindcast, dcppB-forecast etc.

• Sub-experiment id: Sub-experiment for defined ensemble, otherwise ‘None’. Sub-experiments is not defined in CMIP5. In CMIP6 it is defined mainly for DCPP experiments. For example; in dcppA-hindcast experiments s1961 defines the year in which experiment is initialized. 

• Variant label: For an experiment, Realization Index, Initialization Index, Physics Index and Forcing index are used to uniquely identify simulations of an ensemble of runs contributed to a single model.

1. Realization Index (k): (>=1) Index distinguishing among members of an ensemble of simulations that differ only in their initialization point (temporally), i.e. initialized from different points in a control run.

Example: Simulations with ‘Natural forcing alone’ and ‘Anthropogenic + Natural forcing’ starting at the same point in time will have the same realization number. This may not be true always for all the experiments even within same model, for example, DCPP experiments (initialized every year, so at different point of time can have same realization number), piControl experiments etc.

However, across different models same realization number doesn't ensure the experiments in all the models have been initialized at the same point.

2. Initialization Index (l): (>=1) This index is meant to distinguish between various algorithms used to impose initial conditions on a forecast or to distinguish between observational datasets that are used for the initialization of a forecast experiment, otherwise it is assigned ‘1’.

3. Physics Index (m): (>= 1) For a single physics version of the model, this index is assigned the value ‘1’. It is essential that there should be consistent assignment of Physics Index across all simulations of a particular model.

Note: 

• Models with substantial differences in physics from others should also be assigned a different source id, not just Physics Index.

• In CMIP5 above three numbers were used in variant label to differentiate between set of experiments in the given model. However, in   CMIP6, one more number is introduced, listed below.

4. Forcing Index (n): (>=) This index distinguishes between runs following standard protocols of a single CMIP6 experiment, but with different forcing.

Example: Two historical simulations, one forced with CMIP6 recommended forcing datasets and another forced by different dataset should be distinguished by this very index.

Variant Label is represented as r <k> i <l> p <m> f <n>. For example r2i1p3f233,     where k = 2, l = 1, m = 3, n = 233.