Ranking paddocks is extremely important in New Zealand, we rely on grass to feed the...
Public Lab is an open community which collaboratively develops accessible, open source, Do-It-Yourself technologies for investigating local environmental health and justice issues.
Public Lab chatroom
Reset your password
Read more: publiclab.org/n/9971
Ranking paddocks is extremely important in New Zealand, we rely on grass to feed the cows profitably, and only use other feeds to supplement the diets. Therefore you need to know which paddock has the most grass (and how much) in order to make sure you can feed the cows enough to make milk. We rank paddocks by measuring how much grass is in each, and drawing a feed wedge.
At present, pasture covers are measured (in pastoral based farming systems like New Zealand, Australia and Ireland) by taking a device known as a rising plate meter out on farm, and walking around the farm. The meter isn't sophisticated - it's a plate attached to a stick. You place the stick on the ground and the plate will sit on top of the grass. The difference between the bottom of the stick and the bottom of the plate gives you a measure of height (and to some
extent, density). By taking many measurements (20-40) per paddock, and converting with a formula. you end up with a representative measure on how much grass is present in the paddock. In NZ, this is measured in kilograms of dry matter per hectare (kgDM/ha).
Here's a link to some more details if you are interested:
There are other methods of measuring grass too, there's an automated trailer pulled behind a quad bike known as a CDAX unit, and this uses photoeyes to determine the height of grass passing through it. It's susceptible to overestimating when there are lots of stalks (and not much green matter). There's also the 'look and guess' method where farmers will simply use their eye to measure how much they think is there. A project ~10 years ago looked at 'pastures from space' using NIR, but the cloud cover in NZ was prohibitive to getting regular measurements. The rising plate meter is pretty standard, and although it has a reasonable margin of error, if operated by the same person it will give you a good ranking of paddocks on farm.
We thought it would be interesting to investigate 'nearer space' NIR - using a near-infrared camera mounted on a quad bike or other vehicle to quickly measure pasture cover. We think if it was easier to measure covers and less time consuming, more farmers would do it and the dairy farming industry would be more profitable.
A distance shot of a paddock:
Example shot used for analysis:
We took my modified A495 out to a dairy farm and took at least 20 photos per paddock for three paddocks. The same paddocks were also measured for average cover using a plate meter. The photos were taken at about waist height, pointed right in front of my feet (so a photo of the same place as we'd put a plate meter down). As expected (and the same as a plate meter) a single photograph does not correlate with the average cover in the paddock as measured by the plate meter. However looking at the average plate meter cover vs NDVI average found:
This is only the results from 3 paddocks, but is encouraging: the mean NDVI increases with the actual cover. The standard deviation could potentially be used as a measure of uniformity across the paddock. I am now planning to do more paddocks, and in a more uniform way to test the correlations further.
A few other things I'd like to test that I'd like to hear people's thoughts on:
- modifying a gopro with the infragram filter - there is some detail on the wiki but it seems like I may have to replace the lens, rather than simply remove a filter
- using dual cameras - one with an NIR filter, the other without, mounted such that they have the same detail in frame
- using two cameras as above, one after the other on the same paddock (may have problems with natural light variation)
Please feel free to contact me at kat |at| sonic.net.nz (or post here) if you can help out.
The full set of photographs can be downloaded here (90MB). There's also a CSV file with the following NDVI information for each paddock/photo: min, max, mean, standard dev, median. The NDVI for each pixel was calculated as: ((nir + green) - (2 x blue)) / ((nir + green) + (2 x blue));
I did this Help out by offering feedback!
None yet. Be the first to post one!
Wow, fantastic. Could you share some of the source images, maybe in a zip file, or samples from your sets of the paddock photos?
Is this a question? Click here to post it to the Questions page.
Sure, have provided a zip file link with all the source, and posted some pictures into the article itself. Not sure why the focus sucked on the distance shots, I think I just had the camera on the wrong autofocus setting.
There's been some discussion of blurry images and how to fix this to some degree in the mailing lists: https://groups.google.com/forum/#!searchin/plots-infrared/blurry/plots-infrared/cqX3zqfzes8/LPGeGuCMwuQJ
Sounds like a great application for NIR imagery. I think you may even benefit by putting your camera back up in the air --> balloon mapping or a small drone plane/multirotor depending on the size of area you're surveying. the recent release of the PiNoir camera-board for the Raspberry-Pi SBC (which PublicLab was involved with) may be a really good solution as you won't have to deal with focus and it has really good resolution (5 MP) which should be plenty for what you're doing as the res should be higher than satellite images (i am not up to date with exact resolutions anymore).
You must be logged in to comment.
This is marked this as an activity for others to try.
Try it now
Click here to add some more details.
How long does this activity take?
How hard is this activity?
What kind of activity is it?
What is it's current status?