Initialisation of tr_ancien

[Laurent Fairhead]

The variable tr_ancien is not presently initialised. Hence when the model is compiled with NaN initialisation (typically in debug mode), the model crashes in physiq_mod.F90 on line 2204:

d_tr_dyn(:,:,iq-nqo)=(tr_seri(:,:,iq-nqo)-tr_ancien(:,:,iq-nqo))/phys_tstep

Writing tr_ancien in the restart file (as q_ancien or ql_ancien are written) could lead to potentially oversized restart files in case the atmospheric checmistry model is run. It also needs some work as the routines involved cannot deal with a variable that is 4-dimensional (indexed with the tracer number) and writing it out one tracer at a time would need to define the tracer name so that there is no mix-up when reading them back in.

Present solution is therefore to initialise d_tr_dyn to 0 (as ii is only used for diagnostics in the physics) and discuss things with Anne and David.

[Laurent Fairhead]

For the moment, until a solution is found, we initialise tr_ancien to tr_seri at the first time step, thus the first values of d_tr_dyn will be zero but will then be correct