Hi Hugh,
Just a side note.
Carefully looking to the blue part of your spectra, the orange line is on top, while it is in the middle in the red part, for example. So I suggest there is tilt of the continuum slope.
My first impulse was thinking it should be a problem of color calibration due to different height above horizon and causing atmospheric extinction. This effect is known to tilt the slope of the continuum.
In astronomical photometry, the blue and red part of light, esp. the ratio of the different parts in color spectrum, will vary with height of the target above horizon. Atmospheric (color) extinction is amplified, if the calibration star and the target have been observed through different air mass. The effect is also observed to have seasonal influence (Gutiérrez-Moreno et al., 1982). The effect should also have a significant influence on observation of the eclipse. Therefore, a careful selection of the reference star and time of observation to calibrate the instrumental response is important.
Conclusion
Instead of using the H-alpha intensity for comparing spectra, try to normalize spectra based on continuum intensity to study possible reasons of the effect. If you want to study H-alpha intensity only, normalize spectra based on the base level of the continuum around H-alpha.
Literature
Gutiérrez-Moreno, A., G. Cortes, and H. Moreno.
A study of atmospheric extinction at Cerro Tololo Inter-American observatory. Publications of the Astronomical Society of the Pacific 94.560 (1982): 722.
http://adsabs.harvard.edu/abs/1982PASP...94..722G
Good luck,
Thilo