Use the full-sampling SpeX PRZ spectrum, rather than the Line et al. down sampled

[gully]

The full-sampling ones are available in the SpeX Prism Library

[gully]

I did this, it looks good!

[gully]

Green is the Line et al. spectrum, Blue is what's in the SpeX spectral archive.

By the way, here is why we are using the native sampling spectrum:

The spectra within the SpeX SXD library are oversampled relative to a spectral resolution element, and therefore each pixel is not an independent sample. Using the full oversampled data would result in overconstrained results. We therefore sample every few pixels. We choose the sampling length based on the autocorrelation length scale of the residuals of a typical model fit to the data. This is 2.7 pixels. We thus take every third pixel to be statistically independent. An alternative approach would be to model the covariant error structure of the oversampled data through a Guassian process (e.g., Czekala et al. 2014).

• Line et al. 2015

[gully]

That is, our system can tolerate covariant error structure in the oversampled data.