How do the background removal methods work?

Top Hat smoothing

This uses a technique that isolates the highest values of the spectra from the background (a technique called morphology). This is a quick method but does not account very well for peaks that vary in size along the spectrum.

Loess smoothing

Also known as "locally weighted scatter plot smoothing" splits the spectrum up into chunks to estimate the background and then fits polynomials together along this to create a smoothed background curve. This is computationally slower but gives a more even result for peaks that vary in size along the spectrum.