What are the main characteristics of CMIP5 Historical experiments?
e.g. Initial conditions? Greenhouse gases forcing?
How they differ from RCPs? And how they are similar to Observed reanalysis data?
The CMIP5 experimental protocol is documented at https://cmip.llnl.gov/cmip5/docs/Taylor_CMIP5_design.pdf
The historical experiment for a model spans model years 1850-2005 and is initialized from a control experiment for that model (forcings from 1850 are used repeatedly).
For models that are not coupled to a modeled carbon cycle, greenhouse gas forcing are based on the observed record (including ice cores) of atmospheric concentrations. For models that are coupled to a modeled carbon cycle, atmospheric CO2 concentration is predicted, and the model is forced with anthropogenic CO2 emissions (and land use change).
RCPs are extensions into the 21st Century. There are different RCPs, based on different assumptions about future greenhouse gas concentrations and/or emissions, land use, population, etc.
Reanalysis data is from a model this is constrained to be close to observations of the atmospheric state (e.g., air temp, winds, pressure, etc.) This differs from CMIP models that run unconstrained.