As far as I can tell, the only data they used were the data fed into the computational model as inputs. To be clear this technique can be really useful though only if units are good -and- conclusions need to be severely restricted which I don't really see well expressed