the old drift scan programs from the HP1000 are in

/net/bootes/usr/HartRAO/src/nccs/doc/continuum/hpcontinuum/scan

the programs dscnX.ftn were my versions, and included a simple method of
estimating the peak height in a drift scan by fitting a gaussian to the
top half of the peak:

the observing program averaged the incoming data so that there were always
12 samples per half power beamwidth.  This level of smoothing produces
less than 1% decrease in the peak value, as I recall.

After the baseline was flattened, the routine GFLUX (extracted to file
gaus_peak_fit.f) would then compute the peak height of the gaussian,
centred on a specified point.

the routine DOGFIT in peak_fit.f called gaus_peak_fit repeatedly for a
range of centre positions on either side of where the peak should be, and
then used the biggest value as being the correct fitted gaussian.

crude and simple, but no convergence problems with iterative fitting.