Which filter do I need to perform NDVI measurements?

Question

I have Nikon D70s DSLR camera and i would like to modify it for NDVI (Normalized Difference Vegetation Index) measurment. I am looking for the filter that can pass effectively RED and NIR band but block others. From the modified D70s (after installing new filter), I should be able to get RED and NIR as a separate bands of a image. I would like to get suggestions for the filter for my purpose. Thanks in advance.

Answer 1

NASA tells me that to capture NIR, you'll need a filter cutting at about 700nm, such as Hoya R70, Marumi 700HB or B+W 092 (that last one is actually 695nm). The diameter should be chosen the same as thread diameter of lens you're going to use (if there are several lenses, either get all different sizes, or the one with biggest diameter and step-up rings for smaller diameter lenses).

I should be able to get RED and NIR as a separate bands of a image

I'm not quite sure what was meant by this, but if you meant you'd like to capture NIR in one color channel and visible light in another, then unfortunately there's no filter that could do exactly that. It would require you to replace the Bayer filter on sensor, i.e. the microfilters on each sensel.

However, my experiences with a Olympus C-2020Z (known for its hot mirror doing a lousy job when blocking IR) with a Hoya R72 and tells me that IR affects all color channels. So when letting in red+IR, you'd have IR in blue/green channels and red visible light+IR in red channel. By subtracting average of other channels from red (perhaps multiplied by some factor found experimentally through testing with IR only), you'd get approximate value for visible light only.

Answer 2

I was in Valencia for the AgEng/CIGR 2012 conference where I saw a presentation about adapting a DSLR to do exactly this. The paper contains the details how to get the best possible Red/NIR colour space for NDVI and shows application of the results. You need to remove the ir cut filter and replace with a ~600nm long pass filter.

http://cigr.ageng2012.org/images/fotosg/tabla_137_C1122.pdf

Then you do linear combinations of the the 3 colour bands. Their work base the calculations on actual measured spectral response curves, which you might not have the gear for. The result depends on the white balance but I have tried this technique without actually doing all this on a Basler ACA640-90GC camera. I got good result incorporating it into the bayer demosaiqueing, separating the NIR channel into the blue colour and red into the red colour and nullifying the green. The hard part is balancing the intensity balance of the two bands, since the response to NIR in the cmos is about 25%. I also found that the result is very sensitive to the bayer conversion algorithm. I found that Basler's SDK actually use nearest neighbor, which makes very bad NDVI images.

Example image - before and after converting to NDVI index:

And an example when using the Basler bayer conversion:

Answer 3

Five years on this topic remains of research interest, particularly regarding drone carried cameras, so another answer is probably of some value. The AgEng2012 site linked above has been removed but this paper is the best I've come across:

Rabatel, G.; Gorretta, N.; and Sylvain, L. Getting simultaneous red and near-infrared band data from a single digital camera for plant monitoring applications: Theoretical and practical study. Biosystems Engineering 117(1):2–14, January 2014. DOI 10.1016/j.biosystemseng.2013.06.008.

Methodology and findings are similar to what Michael mentions: converted camera, ~600nm red filter, sensitivity to demosaicing, linear band calculations, and improved spatial resolution compared to NDVI specific cameras at the expense of some SnR limitations.

Answer 4

Newport.com, thorlabs.com, or edmundsoptics all sell longpass filters that can be used for you app.

It's been my experience that the NIR bandpass of the green and blue Bayer filters are almost the same ( less than 10 digital numbers difference at up to 90% of the sensors dynamic range). The red channel NIR bandpass is slightly smaller than the blue and green. But it's going to depend on the camera manufacturer.

You might also want to work with raw data images as the built in demosaicing process can create artifacts. NN demoaicing might not look the prettiest but it retains the trues spectral signal of the scene.