Problem using previous samples to new inference in new model context

Questions v5
3

Thank you for the answer. The general scheme I was trying to implement was:

  1. to estimate the posterior of some ancillary variables (A_, B_) given T and NDVI (both input data) in the case I know sm_toEst (calibration).
  2. then to use these posteriors of A_ and B_ to better estimate sm_toEst as a function of T and NDVI.

I understand now that if I define a new prior that does not depend on anything else, the model is sampling from this new prior and all calibration is lost. I just don’t understand how to use the posteriors of A_ B_ in a new inference problem.