I will try to clarify my problem. I have observations (NDVI, T) that I want to use to estimate sm. However, I only have a complex physical model that links sm and T like,
T = f(x1, sm)
Luckily, there is evidence that x1 ca be approximated as x1 = A_ + B*NDVI, but I need to infer values for the coefficients. To do this, I have several sites and periods in which I have also data about he variable I want to estimate (sm); let’s call these data sm_obs and these periods calibration. Once the distributions for A_ and B_ are obtained, I want to use them to estimate sm (real run time).
To run everything together seem a great idea. I just need to write a model in which sm is first data (sm_obs) and them a random variable sm.