Grayson,
The NDVI images look excellent. Just to clarify a few things, I think the following are true:

1. The NIR camera has had its IR block filter removed.
2. The 850 nm and 720 nm filters are glass filters in front of the lens of the NIR camera.
3. These are good optical filters which pass no wavelengths shorter than 720 or 850 nm, so pass absolutely no visible light.
4. The histograms are of the entire NIR photos.
5. You used CHDK to capture RAW image data.
6. The color table for the NDVI images is NDVIBlu2Red.lut:

Some questions for you:

1. Which channel did you use for visible light to compute NDVI?
2. Which channel did you use for NIR light to compute NDVI?
3. Did you stretch either or both histograms? How much?
4. Did you use the RAW files? How did you convert them to jpg or tiff?

This is a really revealing test. I conclude that:

1. The green channel in the A2500 is not very sensitive to NIR light > 720 nm, but does record some. In other words, the Bayer filter over all the green pixels does not pass as much NIR light as the filter over the blue and red pixels.
2. The red Bayer filter passes more NIR light between 720 and 850 nm than does the blue Bayer filter.
3. The red and blue Bayer filters pass the same amount of NIR light longer than 850 nm.
4. The processing of RAW files to tiff requires assumptions about white balance and other things. Your histograms could look quite different with assumptions other than the ones made. So statements 1 to 3 above might not really be true.
5. Your photos include lots of bright sky. If sky was not included, this would increase the dynamic range of the part of the scene with vegetation. Then there might be more information about NDVI differences within the vegetated part of the scene.
6. Considering that you used only part of the camera's dynamic range for the vegetated part of the photos, there is a lot of good plant heath information in the NDVI images.
7. I agree that the 850 nm filter NDVI images has slightly increased NDVI values (shifted to the right on the scale). That could be an artifact of several processing steps.
8. It would be good to see histograms of just a homogeneous area of vegetation. That might be easier to interpret.
9. This is very cool.

Chris

I have pictures of three different kinds: true color, 560nm IR filter, and 720nm IR filter, and I've been having the difficulty of choosing which channel from IR photo to choose. How did you choose? Which channel did you use with the 720nm IR filter? Did you or didn't you modify your IR camera, to take off the IR block filter?

And another question for all you guys, I have to exactly equal cameras, and was considering modifying one. But I am not sure whether to choose the blue filter, or the wratten 25A filter to insert inside the camera once I remove de IR block filter.

Can I close the camera with NO filter, and try them on the outside, like a normal filter?

Thanks Rita

Rita,
If you are using two cameras, one unmodified and one to capture an NIR image, the best filter to use in the NIR camera is one that blocks all visible light, like the Wratten 87 or your 720 nm IR filter. With the factory IR block filter replaced with a Wratten 87, all three channels in that camera will record only pure NIR light. Any of the three channels will make good pure NIR images. One of the channels might be brighter than the others because that color Bayer filter is passing more NIR. That is probably the one to use (unless it is overexposed).

The Wratten 25A and the blue filters (Rosco 2007, 2008, 74, BG3, Wratten 47B) should replace the IR block filter only if your goal is to compute NDVI from a single camera. However, the Wratten 25A will produce a pure NIR channel (blue), so taking pairs of photos with an unmodified camera and a Wratten 25A NIR camera will allow pretty much the same NDVI result as using an unmodified camera and a pure NIR camera like with Wratten 87. You will have to use the blue channel from the Wratten 25A camera for NIR, and you will have a second channel with red visible light (red channel), although the one from the unmodified camera will not be contaminated with NIR. If the unmodified camera malfunctions, the photos from the Wratten 25A camera, by themselves, can be used to compute a good facsimile of NDVI. This is not true with a pure NIR camera.

The Wratten 25A filter will give you a purer NIR channel (blue) than a blue filter (with which the red channel will be mostly NIR but with some red). NDVI is traditionally computed using a red band for visible light, and that's what the Wratten 25A allows. Using the red channel for visible light avoids the greater scattering of blue light, so photos of very distant subjects can be less hazy with the Wratten 25A. Both filter choices will result is a visible light channel that has some NIR in it. The only way to avoid that is to use two cameras.

It doesn't matter whether the filter is placed inside the camera or in front of the lens. Only a small piece of filter is required inside the lens, and it is protected. Placing the filter in front of the lens requires a larger piece, it can get scratched, and it is not protected from direct rays of sunlight (it might need a lens hood to prevent glare, although this is not relevant if all photos are vertical mapping photos). But if the filter is on the outside, you can swap filters or use the camera without a filter (e.g., for IR spectroscopy). Placing any gel, film, or plastic filter inside the camera will risk degrading the image quality by altering the optical path. But removing the IR block filter without replacing it with something can have similar effects. Some Powershots have adapters that allow using standard, round, screw-on, glass filters. These will allow good image sharpness, and some common filters are not expensive. Most small Powershots do not allow this. Chris

Hi

The first pictures I have are with only one camera and an unmodified one.

Now I bought the second camera, same model as the first and was considering modifying it. So my question remains. On this NDVI did you modify one camera? Did you use the filters inside the camera or in front of the lens. And how did you choose the channel to compute NDVI...

Thanks Rita

Hallo Rita. In my setup One camera is not modified. The other camera I removed the IR block filter inside. I use the 720nm or the 850nm filter on the outside. My cameras will be in a drone so I should not have problem with any glare. Maybe if I go over water but not so much NDVI to do there. (Wait a moment maybe I can see some alge growth, that will be another project). For protection: before my drone (plane) will land the lens will be retracted in to the fuselage. In the visible picture I use the red channel. In the filtered picture I use the red channel. For my camera you can even use the blue channel in the filtered picture. If you look a my histogram thy overlay almost 100%. Regards Grayson.

Thanks that is exactly what I wanted to know.

I am going to modify one camera, and wanted to test different filters, but was unsure if I could do it in front of the lens or only on the inside.

You are using them outside which is great (although I just bought a wratten 25A today) and the 720nm I already have...

Anyway thanks, that was just what I needed to know.

Best regards, Rita

Hello Chris. Seems I need to be more precise on my notes, answer to your question.

NIR camera has the block filter removed. 720 and 850 nm filter are glass filter placed on the outside.

The quality of the filters, I trusted what is written on the box, Filters where bought at a proffesional camera shop. cost arround 25 euro 38 us dollar.

The histograms are of the NIR pictures. Made in gimp with no alteration.

I use chdk to get raw images. format DNG. I use ufraw to convert them to tiff. ufraw --out-type=tiff. the result on this page are in jpeg else the picutres will be huge. For future test I will host all my pictures online, if any one else want to play with them. NDVI are created with tiff as input.

I did not know that withe balance will have an effect. Need to look in to that.

Color table is NDVIBlu2Red.lut

I used red channel IR visible. I used blue channel for IR, I made a mistake and forgot to set to red channnel. I made a test with red channel and there is no visual diffrences in the NDVI image. If you look at the histogram you can see they have the same profile. But for future test I always will use red channel.

I do not know what you mean with: Did you stretch either or both histograms? How much? I let gimp create the histogram. Only streching I did is make the window a bit larger.

You request: It would be good to see histograms of just a homogeneous area of vegetation. Will come, just waiting for the same wheater conditions as last time.

Chris any idea if NDVI will work over water. Can I see alge growth. If yes i want to try to see how farm land runoff will affect surrounding water alge content. In holland we always have alge problem in the summer. Even toxic level where people are not allowed to swim in lakes.

Regards Grayson.

Thanks Grayson,

I was referring to stretching the histograms in the Fiji plugin. I forget what the default is, but it can make a big difference.

GIMP might allow you to marquee an area of grass in your current photos and see the histogram for just that area.

If the aquatic algae is floating it should be revealed by NDVI. If it is near the surface, and there is a lot of it, it might show up. I'm not sure how deep you can see into water and get similar results in both an NIR and visible image so that you can compute NDVI. If the algae is blue-green algae you will need to out-think it. Some common blue-green algae absorb light only at the red end of the spectrum. So there should be a difference in NDVI depending on whether you use the red or blue channel for visible light. Demonstrating that difference could confirm that the signal is from blue-green algae.

Just so you know, 720nm is in the "red edge" region so you're not computing NDVI. You are computing the Normalized Difference Red Edge Index (NDRE).

It's good to know about the NDRE. Instead of computing the difference between reflected NIR and reflected red light, I think NDRE replaces reflected red (600 to 700 nm) with a band at the edge of red and NIR, somewhere between 690 and 730 nm. This has been shown to be more sensitive to early plant stress than NDVI.

When we use a "720 nm" filter, it is a filter that passes all wavelengths longer than 720 nm (all NIR) and blocks all shorter wavelengths (all visible light). So that filter is used to produce a pure NIR band between 720 and about 900 nm (upper limit depends on the camera). This band is used as the NIR information in NDVI.

NDVI = (NIR – Red)/( NIR + Red)

NDRE = (NIR – Red_edge)/( NIR + Red_edge)

It is easy to capture a red edge band with an unmodified camera and a Wratten 89B filter. That filter passes only wavelengths longer than 690 nm, and the IR block filter in the camera passes only wavelengths shorter than about 730 nm (depends on cmaera). Not much light would get through both filters (and the Bayer filters) so exposures might have to be long. The red channel could record a good red edge band.

In my case I also want to capture visible images, will help the farmer to identify the problem. Can use my drone for mapping and 3d vegation maps with http://ecosynth.org/. The pic look so nice. If this red edge is a big problem for my cam I still can use 850nm filter. Regards Grayson.

The alignment of the blue and red spectral response below 850nm that you are seeing is consistent with the spectral response curves that have been documented for other cameras, for instance at http://www.maxmax.com/camera_technical.htm. The low spectral sensitivity of the the green band is also consistent; however, note the comment on the maxmax page that the spectral response is different that the published spectral responses of the RGB-filtered channels at the sensor; that "may be due to other processing that occurs in the camera after you take a picture." For a stab in the dark, and to the best of my shaky understanding, the Bayer array has two green sensors in every 2x2 block to take advantage of the solar spectral maximum in the green for luminance (https://en.wikipedia.org/wiki/Bayer_filter). I guess that in the visible that using green as a component of chromaticity might require scaling down the relative contribution of the green pixels, e.g. by approximately 1/2. In the infrared there is no reason for doing this, and it seems to throw away useful spectral information. If, as I understand the situation, the raw image literally contains the raw sensor readouts, it should be possible to recover more spectral sensitivity from the green band. However, the de-mosaicing software for "developing" raw imagery was developed for visible images, so getting a better green channel requires more software modification than I know how to do. (I've spent some trying to figure out how to fool Raw Therapee into giving up a better IR green band, without success)

I just used a StellarNet spectrometer to find the transmittance of hot filters from the Canon A2200 & A2500. Both hot filters rapidly fall off before 700nm. The Wratten 89B only seemed to transmit above 700nm so the two together don't transmit any light. The Rosco 2007 with the hot filter only transmits blue light. Any ideas on a camera model with a hot filter that transmits up to 730nm, as Chris suggested? Has anyone had success creating a red-edge bandpass?