that is one way to do it and DBE can certainly handle complex gradients. the only "problem" is that when you evaluate a background model from a dark sky image or a single sub, it's fairly obvious when you have it right - the background mainly looks smooth and linear.
with an integrated image from a light polluted site, the gradient can become complex. for instance if you shoot all night into the light dome of a city, the integrated LP gradient will probably be curved at the bottom. again, DBE can handle it but in order to make sure that you are not subtracting real signal you have to have some idea of what the background should look like. if you have a multi-night project with LP and moon, well, the gradient gets even more complex.
you can do DBE on your individual calibrated subs before integration, but because of the nature of DBE each one has to be done by hand. the ABE tool is automatable with ImageContainer but then that's another background extraction tool to learn. and then the parameters that you calculate for one sub might not work well with another sub, so there's still some hand work to do, evaluating the output of ABE and perhaps partitioning your subs into groups for which a single ABE configuration makes sense.
given the complexity of all that, it's probably better to start with the integrated image with no pre-background subtraction and see if you can get a result that you like...
rob
