In theory, running future simulations offline is no different than running current simulations, assuming you have the driver files. Getting the driver files can be a bit more tricky.

It seems that Patricia Cadule, Fabienne Maignan (working from files pre-treated by Patricia), and Nicolas Viovy are the three people who know something about future driver data (that is, driver data taking from the results of CMIP5 simulations for the years 2006-2100). Working from the files that Nicolas has used is also a bit tricky depending on which version of ORCHIDEE you are using, since they are currently formatted for his input driver and so require a little bit of change to work with the trunk. We did this first by running the following bash script on Curie.

!/bin/sh

export inputfilename="ipsl_rcp4p5_2006-2010.nc"
years=(2006 2007 2008 2009 2010)
sindex=(1 366 731 1096 1461) 
eindex=(365 730 1095 1460 1825) 
varnames=(LWdown Psurf Qair Rainf SWdown Snowf Tmax Tmin Wind)

for ifile in 0 1 2 3 4 
do 
    echo "On year ${years[$ifile]}"
    export outputfilename=ipsl_rcp4p5_${years[$ifile]}.nc
    rm -f $outputfilename 
    for ivar in {0..8}
    do
	rm -f temp.nc
	echo "On ${varnames[$ivar]},${sindex[$ifile]},${eindex[$ifile]} "
	ncea -F -d time,${sindex[$ifile]},${eindex[$ifile]} /ccc/scratch/cont003/dsm/p529viov/isimip/ipsl/i${varnames[$ivar]}_$inputfilename temp.nc
	ncks -A -o $outputfilename temp.nc
	rm -f temp.nc
    done
done

The above script combines the forcing from 2006-2010 that Nicolas uses into annual files. From there, additional work still needs to be done (and the script needs to be modified to cover the years 2011-2100). In particular,

1) The order of the variables is not necessarily the same as the CRU-NCEP files. ORCHIDEE is very picky about the input files, so you'll have to make sure that the order of the variables and all meta data/axis are exactly the same as found in the existing CRU-NCEP files to make it work.

2) All years are daily, not 6-hour timesteps. In theory, ORCHIDEE can use a weather generator to account for this, but don't be surprised that the files are smaller than CRU-NCEP (all years have 365 days).

3) There are no Wind_N and Wind_E variable, only WIND. It seems like a good idea is to assign the value of WIND to Wind_E, and then set Wind_N equal to zero, since apparently in the offline mode the wind is always converted to the magnitude and therefore the direction does not matter.

4) There is no tair, just TMIN and TMAX. A simple average of TMIN and TMAX for each point/timestep and setting that to be tair in the output file probably suffices, but should be checked with some test simulations afterwards.