I see callbacks as being useful to define standardized entry and exit routine for a model. These callbacks could be used to perform user-specific data validation, and on the back-end, clean-up. The timing of these calls would be left up to the growth model (hence the need for a callback). If we found that callbacks could not be ported, we could still spec out a calling sequence that the client could use.
The other use for callbacks that I can see, is to perform custom processing within a growth cycle (wood/product quality and mortality models come to mind). The growth model(s) would call these user-defined functions at predetermined points/times.