Internally, the radiative transfer integration is always performed with
Double Precision although the underlying line parameters on which the
calculations are based (with the exception of line position) are
rarely known to better than 1%, i.e. single precision.
Double Precision output is only really
necessary when considering transmittances close to 1.0 (in which case
writing absorption (= 1-transmittance) spectra
would be better),
or when comparing
different RFM runs with minor perturbations.