That's a tough one. With the IR blocking filter removed, all three color channels (all pixels, regardless of which color array filter they are under) will capture some near infrared light. So the camera records three channels, each with mostly information about red, green, or blue light, but also with some unknown amount of NIR (and some UV as well). Trying to figure out how much of that brightness in each channel is from NIR light might be futile. Here are a couple examples of what the three color channels see when the IR block filter is removed (but each camera is different):
.

.
That's why single camera NDVI systems use a filter. A red filter that lets no blue light through allows only NIR light (and some UV) to reach the blue channel (the pixels under blue CFA filters). The red and green channels are a mix of visible and NIR light, but it still sort of works. To get a mostly pure NIR channel and a mostly pure red channel, two cameras are needed.

Chris

Thanks for your comment, this is what I was afraid of; there is no relatively simple way to isolate specific wavelengths and furthermore if this was done those wavelengths that were to be used for red and green would contain infrared information thus providing false (or at least less accurate) NDVI readings. Hmmm, looks like it's the two camera system which would produce more true NDVI readings than one camera with a blue filter.

Thanks again!

What about if you added a Schott BG3 filter behind or in front of the lens, then you would have an infrablue response curve (mostly IR in the red channel, mostly visible in green and visible+IR in blue channel).