LMDZ work on the Cordex South America domain:
From G. Krinner:
My intention is to use observed SST and sea-ice for the present runs and observed SST and sea ice plus anomalies from coupled runs for the projections. Is that also what has been done / is planned for the other applications? This approach is actually a bit tricky over sea-ice, but probably those of you who run the model for Southeast Asia, South America and Europe do not really care for that aspect.
My most relevant papers in this respect are both in J. Climate (Krinner et al. 2008 and 2014) : http://journals.ametsoc.org/doi/abs/10.1175/JCLI-D-13-00367.1 and http://journals.ametsoc.org/doi/abs/10.1175/2007JCLI1690.1 To those who really would like to have a look at the papers and cannot download them for some reason, I can send these papers of course, but I don't want to overload everybody's mailboxes with too much stuff.
In the 2008 paper, I essentially show that for present-day simulations, it is better to use observed sea surface conditions than those directly taken from a coupled model. Surprise, surprise. And I describe how I calculate the sea-ice concentration anomalies (which are a bit trickier to calculate than SST anomalies) for "downscaling" the climate projections.
I think that an interesting aspect of my 2014 paper here is that I carried out Antarctic climate projections with a number of different prescriped SST and sea-ice data sets, coming from different coupled models (all the same scenario, SRES-A1B). Because there is only one present-day reference run (with observed SSTs), it is relatively cheap to carry out a number of future runs - there is no need to redo a present-day run with boundary conditions for each driving coupled model.
PhD subject on the topic