Noise estimates are estimates of the standard deviation of the noise in each channel of the integrated image, assuming a normal distribution of the noise.
Location estimates are estimates of the central value, or the most probable value, for each channel of the integrated image. Normally, since deep sky astronomical images are unimodal sets with a strong central tendency (that is, with a unique and very prominent histogram peak), you can identify these values as approximations to the location of the main histogram peak for each channel.
Scale estimates are estimates of statistical dispersion, or variability, for each channel. They tell you how much your integrated data varies around a central value, which is the location estimate described above.
SNR estimates are estimates of signal-to-noise ratio for each channel of the integrated image. These values are merely informative, non-robust, and often inaccurate. Don't use them to compare different integration results. I am considering removing these values from the final integration summary in a future version of the tool.
Reference noise reduction refers to estimates of the effective noise reduction (ENR) achieved in the final integrated image with respect to the integration reference image (the first one in the list), for each channel. The ENR function is a robust estimate of quality. It is described in the documentation.
Median noise reduction gives the median of ENR functions computed for the final integrated image with respect to all integrated images.
Your goal is to maximize ENR in the integrated result with the required rejection of outlier pixels. If you are sure that the reference image (the first one) is the best one in SNR terms, then concentrate on maximizing reference ENR values only. If you are unsure (as usually happens), maximize median ENR.
