What can you do with Agricultural Multispectral Imagery?

[SKR]

Good morning,

I’ve been mapping my farm using an RGB camera for a few years now, and I’ve found a few uses to justify the time/cost:
It can generate topographical maps, which can be used for drainage information or for biomass calculations,
It can be useful in checking emergence of crops,
It can be used to scout fields and identify problem areas that are visible to the eye.

I’m curious about adding a multispectral camera to my toolkit, but I’m having a hard time identifying what information it will give me that can result in action.

To those of you who work with Multispectral Imagery in agriculture, what are you using it for? Can you make topdressing maps by identifying under fertalized areas? Can you map out hard to see weed patches from the air? What spectrums are useful beyond academic curiosity?

If you’ve found a way to utilize additional sensor information I would love to hear about your workflow.

[Saijin_Naib]

Yes, to all of that, but it can be difficult.

You’ll want a platform that has a Downwelling Light Sensor (Sunshine sensor), or can be calibrated using a reflectance target. You’ll want the collect to collect variance to be controlled for if you want to compare trends over time for the fields.

For the most common algorithm, NDVI, you just need four bands (RGB+NIR). Atmospheric interference isn’t really a concern at like 400ft AGL or less, and aerosols/particulate shouldn’t be much of a bother either if you’re not collecting during rain/fog/snow/fire/dust storms
NDVI from sUAS is also really convenient since you can do ground-level spot-checks and truth assessments with a handheld sensor like a GreenSeeker or other such NDVI unit.

Those four bands also have other algorithms that are pretty robust you can use.

Moisture stress (directly) is hard unless you start really reaching into the mid/long infrared or thermal, but those bands tend to inflate the size/weight/cost of the sensors dramatically. Besides, you can get proxy metrics by looking for depressed NDVI or even just lower NIR absolute reflectance (or use another band ratio).

Inter-species detection is going to be a challenge with just four bands. You normally want a hyperspectral sensor if you’re doing that work specifically, but again, with even NDVI or NIR reflectance you can normally spot “weeds” (let’s just call them what they actually are: native vegetation) by their usually significantly-higher vigor (NIR reflectance/NDVI metrics).

[SKR]

Thanks for the information! I’d like to wrap my head around this, but don’t really have the knowledge to know what questions to ask.

Would you or someone else who does this work be willing to roughly outline a workflow for a specific use-case? It probably won’t line up with my needs, but just some steps and software would be helpful as a starting point for me.

[Saijin_Naib]

Sure, ask what you need help with and hopefully folks can help out.

[SKR]

I’m really not sure what questions to ask, because I’m not exactly sure what I’m looking for, but let’s run an example or two.

Let’s say I’m growing wheat. But I’ll take information for any crop.

How do you time your flights? Do you try to fly/map within a range of days after emergence? Based on plant stage? Based on Growing Degree Days? What information are you using the tool to gather in that first flight?

If I mapped my field and identified some weedy areas that I wanted to make into a prescription for spraying, how do you turn that map into a prescription? Is it all proprietary software based on your model of sprayer, or is there a software pipeline from WebODM through QGIS and I could learn?

If I mapped my field with the intention of identifying areas that need topdressing, how does that show up in the multispectral imagery? Would it just be “an area that looks different” and I’d have to ground-truth it to see what’s going on?

[Saijin_Naib]

How you time your flights really depends upon what you’re trying to do.

Are you scouting early-season? Where I worked, we did some limited sUAS scouting, but mostly we used my sUAS surveying for making prescriptions and checking (in a few cases) trials.

How you get that prescription onto your equipment is highly dependent upon the equipment… Not only the make, but the model (and sometimes, even the firmware version ). You’ll need to know exactly what you’re putting it into to know what formats it accepts, how to write them out to the storage, and how to sideload them.

In general, yes, it is all proprietary software. You can skirt some of it and make prescriptions, fields, boundaries, guidance lines, etc. in QGIS and then load them, but reversing how they do it is time-consuming and a bit of a game of whack-a-mole. I, unfortunately, was not able to take any of my tooling with me when I left that job.

There are a number of QGIS plugins you can try, though.

If you’re looking to apply nutrients, you’ll be looking for depressed plant vigor. How you assess that depends upon your sensor and what indices are appropriate.

You can look at just gross NIR reflectance as a metric, or you can use NDVI with an appropriate sensor and look for areas of lower NDVI.

Whether or not you ground-truth it is up to you. I developed an index for our sensor (GoPro Hero4 Black + KolariVision Infrared Full-Spectrum Conversion + KolariVision Infrared Blue/NIR NDVI Filter) that was very highly correlated to leaf/ground-level NDVI, but was not equivalent because I suck at math and couldn’t figure out how to normalize it

So, once we bore out my index, we stopped ground-truthing, but again, that depends upon you.

Do you have implement-level NDVI, like GreenSeeker? If so, you might as well collect the data as you’re running over the field to spray or do other work, right?