Thanks again! That reply from @aseyboldt is very complete (I am saving it). But I made a mistake, I did not mean a function but a theano vector (continuum_true), very sorry for that…
It should be then something like a Simpson’s rule:
To calculate the green area in the image. Then, I add it to the model computation of the red area. This sum I can compare with the observational total line area without breaking the Bayesian statistics 
Do I need to declare a theano ops to do that?