just starting to vary parameters seems to be a poor choice for calibrating models. This rail and error approach has some serious drawbacks:
-it will much time
-no guarantee on succes
-you will achieve a non-reproducible result
-when using the metric 'acceptable match' your result will be defined by subjectives choices
I suggest you apply a sound metric (simplest example leasts squares)) using some optimisation algorithm (e.g. genetic algorithm, or Levenberg-Marquart), there is software freely available to do this (e.g. PEST). This will result, next to a vector of parameters, also provide information on bias in our model (due to either a wrong proces model, wrong input or systematic errors in the measurements you are using for calibration).
Thank you so much for your answer and your recommendations, i will try the software you told me, and if i have any other questions i will come back to you.