What I want to do
Due to the issues with blue channel infrared leakage, I was interested to see if it was possible to use the green channel data in place of the blue channel data to create useful NDVI alternatives. (I suggested this approach in this comment.)
My attempt and results
As you can see in the lead image, the green channel does seem to be detecting reflected infrared (as this graph suggests it will), but much less than the neighboring R and B channels. Unfortunately it also seems to detect less reflected light on non-vegetation, so maybe the approach is doomed from the start...
Anyhow, here is the image with the B and G channels swapped (which I did in order to use the Infrapix prototype web utility), plus the "NDVI" alternative and the infrared channel:
(see the full data here: http://infrapix.pvos.org/show/blue-green-swap.png)
And here is a larger version of the full alt-NDVI:
Questions and next steps
Basically it seems not to work, from what I can tell -- vegetation is not highlighted vs. the non-vegetation background.
Perhaps by looking at these histograms of the raw image (working from Chris Fastie's post, we can see if there really is enough differentiation to understand differences in health:
What do you think?
Update: I did one more where I ran auto-levels on the green channel to boost the range, which may not be "scientifically justified" but it did produce better contrast in the resulting alt-NDVI. It still did not, however, seem to differentiate plants: