Using the same input Collection 2, Level-2 Landsat data set (spatially and spectrally subset, values below 0 and above 1 masked, both WGS 1984 UTM Zone 11N, pixels reportedly 30 m) to perform Linear Spectral Unmixing, I observe different ROI pixel coverage and reported areas depending on whether I use ENVI Classic + IDL (64-bit) or "New" (ENVI 5.6.3 + IDL 8.8.3, 64-bit).

When I use the ROI Tool with the three ROIs that have 6, 6, and 3 pixels (which I created in ENVI "New"): When I apply those ROIs to the aforementioned Landsat file in ENVI Classic, I see that those 6, 6, and 3 pixel grouping line up with the Landsat pixels and the ROI Tool reports the expected areas of: 5400, 5400, and 2700 sq m, respectively. When I apply those same ROIs to the same Landsat file in ENVI "New," the ROI Tool reports ROI coverage of 6, 6, and 3 pixels but also "pixel counts" of 21, 20, and 9 respectively, and the ROIs have different shapes in the image (e.g., polygons of three pixels instead of a single pixel). Additionally, the imagery does not have defined pixel edges in ENVI "New" (the way it does in Classic), so I don't see that ROI pixels are lining up with image pixel edges.

Unsurprisingly, when I run Linear Spectral Unmixing and output the values from my generated reference sampling grids, I receive slightly different results from ENVI Classic than "New."

I understand that ENVI Classic is being phased out, so I hesitate to use it and that is why I have been doing my work in the "New" ENVI. However, I see the visuals that I expect in Classic, and now I am worried about the data outputs from ENVI "New."

Have you come across these disparities between the two? What is your recommendation regarding which to trust moving forward?