Hi Pauline, > 1) Silly question but just to be sure: can you confirm that the > retrieval unit is defined in: Q.ABS_SPECIES(2).UNIT ? (for example > for abs_species(2)). and if nothing is mentioned, VMR is the default > unit for retrieval ?
UNIT is not used for species not retrieved. Is it possible to leave UNIT undefined for species to be retrieved? If yes, then this is a bug. > 2) When you say qpack automatically converts the VMR to the selected > unit when creating x_a, it means that if I select logrel for > retrieval unit, my x_a vector in oem.m is expressed in log rel and > not VMR anymore. Then, the direct radiative transfer of ARTS is > called in : > > calc_y_and_j( comfun, Q, R, O, xnorm, x, iter ) feval( comfun, Q, R, > x, iter ); > > when we call this function x is in log rel in my example. Can you > confirm that then, in another routine x is again converted to VMR to > call the direct code of ARTS to simulate the TB ? Yes, this conversion is handled by arts_x2QR. > 3) My retrieved state and Jacobians are expressed according to the > UNIT of retrieval defined in Q.ABS_SPECIES(2).UNIT In my example, can > you confirm this will be log rel ? Yes. > It is quite complicated to convert B matrices computed from ensemble > assimilation, we need to compute them directly at the good unit and > it costs a lot to run ensemble assimilation. They are normally > provided in the UNITS used in the NWP models (kg/kg or log(kg/kg)) Some conversions of B should be quite straightforward, or? The conversion from kg/kg to VMR (for a given pressure) depends somewhat on temperature, but I think that can be ignored. If you select a reference state in kg/kg, you can also map the B matrix to rel or logrel quite accurately. If your a priori changes, and you want to use rel or logrel, you need convert B for each case. Or do I miss something? Bye, Patrick _______________________________________________ qpack mailing list [email protected] https://www.sat.ltu.se/mailman/listinfo/qpack
