Check if variance properly computed in L1

[awhoward]

Based on an initial examination, it appears that the variance in the 2D files (which includes contributions from signal and read noise) is not being properly computed in the L1 files. For L1 files, it seems that variance is just computed as abs(signal), not the sum of variance over the extracted 2D region (possibly using weights per pixel in the extracted stripe depending on the extraction method).

[awhoward]

Here's an example comparing the variance and flux arrays for SKY in a 15-min observation of a faint target (Kepler-1654, V=13.4). It appears that variance = abs(flux). We should fix this.

[awhoward]

I think it's in the compute_variance() method referenced on line 444 of this code. I can't find where SpectralExtraction even computes the variance for each extracted L1 pixel from a region of the variance array in the corresponding 2D array.

I'm putting a note here for myself or someone else as I need to step away.