Drones for better potato yields
Advancing the use of drone imagery as a more sophisticated tool for precision agriculture.
Drones, or unmanned aerial vehicles (UAVs), are becoming increasingly popular with crop growers as an easy way to take a look at their fields. Now, researchers are fine-tuning the use of imagery from drones as a more advanced tool for precision management of Canadian potato fields.
This work with drone imagery is part of a major five-year project to boost potato yields. “About three years ago, Potatoes New Brunswick and McCain Foods Canada came to me and said they were having issues in terms of potato productivity in Eastern Canada,” says Bernie Zebarth, a research scientist with Agriculture and Agri-Food Canada (AAFC) at the Fredericton Research and Development Centre.
“The data across North America show a slow but steady average increase over time in potato yields, going up about five hundredweight per acre per year. But in Eastern Canada, that is not happening – the crop insurance data for New Brunswick suggest our yields are either stagnant or perhaps even decreasing slightly over time.
“That’s a serious concern for industry. For example, a lot of New Brunswick’s potato production is for French fries for export. To export, you have to be competitive. If you start losing yield, then you become less competitive. So they asked us to work with them to see what could be done about it.”
The project aims to increase yields by addressing the variability in productivity within potato fields. “The project has three main objectives. First, can we develop ways of mapping the variability in plant growth and yield in potato fields? Second, for the areas of the fields that are not performing well, can we identify why? And third, can we overcome those yield limitations?” Zebarth says.
“In Eastern Canada, precision agriculture is an exciting new area for us, and this is part of what we’re doing in the project,” he adds. So, they are trying out tools like drones and yield monitors to map in-field variability as a first step towards managing that variability through precision agriculture.
According to Yves Leclerc, McCain’s director of agronomy for North America, managing in-field variability is key to improving yields and profitability for potato growers. “We need to understand profitability not only at the field level but also the subfield level, and manage fields at that level. It is no longer enough to manage a field based on the average conditions; we need to be more precise [to achieve a field’s full yield potential].”
The project’s initial phase, in 2013 and 2014, took place in New Brunswick only. For 2015 to 2017, the research is also taking place in Prince Edward Island and Manitoba. For the first two years, Potatoes New Brunswick, McCain Foods Canada, AAFC and New Brunswick’s Enabling Agricultural Research and Innovation program funded the project. With the expanded project, two additional agencies have come on board: the PEI Potato Board and the Manitoba Horticulture Productivity Enhancement Centre Inc. The project’s lead agency is Potatoes New Brunswick.
To try to remedy yield limitations, the project team is looking at a wide variety of practices such as compost applications, fumigation, fall cover crops, nurse crops, furrow de-compaction and, in Manitoba, variable rate irrigation.
“We’re looking at everything from drones and drone imagery, to the thousands of holes we’re digging to look at soil compaction, to compost applications, to soil salinity – everything. We want to make sure that there’s literally no stone unturned,” says Matt Hemphill, executive director of Potatoes New Brunswick. “There is no one-size-fits-all and no magic bullet in this process because we have such variability in soil and weather conditions.”
Drone imagery – advantages and hurdles
A drone with a specialized camera and advanced software can be used to map in-field variability. The drone flies over the field in parallel passes to capture the entire field in a series of overlapping images. The camera can be set up to capture particular wavelengths of light; for instance, near-infrared wavelengths may be of interest because healthy plants reflect more near-infrared light than stressed or dead plants. So a near-infrared image of a field could potentially be used to identify patches where the plants are stressed due to problems like disease or low nutrient levels.
“For instance, if we could use the imagery to identify the parts of a field that aren’t deficient in certain nutrients, then a potato producer wouldn’t need to waste time, energy and money in putting fertilizer products on those parts of the field,” Hemphill says. “Imagine, instead of broadcasting X number of tonnes per acre of lime across the whole field, you maybe only have to apply it to 25 per cent of the field. You can imagine how quickly those savings would add up. The same goes for other input costs.”
Zebarth explains that, before drones became available, the main way to map vegetation patterns was with satellite imagery, which has advantages and disadvantages compared to drone imagery. “One advantage of satellite imagery is that it’s calibrated [so the imagery data is easier to use in advanced analysis]. But there are two big problems with satellites. The first one is that you can’t control when they capture imagery of your field – they only fly over the field every so often, and they can’t see through the clouds. In New Brunswick, we have a lot of cloud cover. The other disadvantage of satellite imagery is its low resolution; a ‘pixel’, an individual point of information, represents an area of maybe five by five to 30 by 30 metres in size [on the ground]. So we’ve never used satellite imagery very much to look at crops here in the East.”
Drones avoid those disadvantages. The user can choose where and when to capture the imagery, as long the operator flies the drone safely and legally. (Anyone operating a drone in Canada must follow the rules set out in the Canadian Aviation Regulations and must respect all federal, provincial/territorial and municipal laws related to trespassing and privacy.) Cloud cover isn’t as much of an issue for drones because they fly below the clouds. And drone imagery is at a much higher resolution, about seven to 10 centimetres, depending on how high the drone is flown.
These advantages are making drones popular with crop growers. “People are already using drones, mostly for qualitative assessment. Drones are fantastic for that,” Zebarth notes. “One such application is to have a visual look at your field. For example, if you fly the drone when the crop is beginning to emerge, you can really see the field’s variability – where the crop is already emerging and doing well, where it is just emerging, and where it hasn’t emerged yet.”
Another qualitative application is to target field scouting. Zebarth explains, “In the image of the field, you might see a patch that looks different. The imagery won’t tell you what the problem is; it will just tell you something is there. So you can go out to the field and look at that patch to identify the problem.”
However, the project team wants to take drone imagery a step further. “We’d like to get to where it’s a more sophisticated tool,” Zebarth says. “We want to determine quantitative differences, like trying to develop relationships between the imagery and things like yield and leaf area index, and so on.”
McCain Foods has its own drone, camera and software for this type of geo-referenced field mapping, and it has trained two of its employees to operate the drone, one as a pilot and the other as a spotter. For the project, they are flying the drone over commercial potato fields in New Brunswick. So far, they’ve collected imagery for about 50 fields.
They fly the drone over each field several times during the growing season. The first flight is when the soil is still bare, so they can look for differences in soil moisture and drainage across the field. The subsequent flights are timed to capture the crop during early and late emergence, mid-season, and early and late senescence. Zebarth says, “So we’re looking at: do we have variation in the soil, the early canopy growth, and the canopy die-down.”
One hurdle in their quantitative use of drone imagery is to correctly stitch together all the individual images from the drone’s flight over the field. “The drone might take perhaps 50 to 100 different images of the field. Those images have to be pieced together, which is called ‘mosaicking.’ If you have a discrete object, like a house or a fence in the images, mosaicking is not too difficult because you can easily find that object in the different images and align them. But if all you have in the image are rows of potato plants, there is not much to align with,” Zebarth notes. “There is mosaicking software to do that, but it’s not perfect. So we’re working with one of the drone companies to figure out how to get the images almost perfectly lined up so you can get down to looking almost at the [individual] plants.”
A second hurdle relates to calibration. “The drone’s camera is actually measuring how bright the light is in different bands; it is taking pictures that have red, green and blue, like a regular camera, but it may also have near-infrared. So it gives you a number from one to 255 in each of those bands, but it is just a relative number. To calculate things like vegetation indices, such as the NDVI [normalized difference vegetation index], you can use a relative brightness to get a relative value of the NDVI. However, you have to convert it to a reflectance value to get a true value for the NDVI that you can compare across fields or measurement dates. That’s where it has to be calibrated,” Zebarth explains.
The researchers are making good progress with overcoming both of these hurdles. Plus, they are testing over 20 different vegetation indices to determine what each index is sensitive to and which ones work best for potato fields in New Brunswick.
“For instance, one index might be mostly sensitive to how much coverage there is of green leaves, whereas another one might be more sensitive to how much chlorophyll is in the leaves,” Zebarth says. They’ve already found that the NDVI, a common index for measuring vegetation cover, isn’t the best choice for potato crops. “For potatoes, the NDVI reading initially goes up as the canopy develops, but after the canopy reaches a certain density, the NDVI becomes insensitive.” Some of the other indices don’t have that drawback.
Once the researchers complete this work in the coming months, they’ll have a much better idea of how they can use the drone imagery. The information from this work could also help agronomists, crop advisors and growers with an interest in quantitative uses of drone imagery in potato production.
“In the long run, drone imagery is going to be a tool to add to our arsenal of tools, for sure,” Leclerc says.
Progress on agronomic findings
One of the project’s key agronomic findings so far is the degree of variability in potato fields. “We are seeing a lot more variability in our fields than we had expected. Although some fields are relatively uniform, other fields have pretty dramatic variation,” Zebarth says.
The results so far indicate that, in New Brunswick, much of the variability is due to the soil. “What we think has been going on is a gradual decline in soil health over decades,” Zebarth says.
The wet spring in 2013 emphasized some of this soil variability. He says, “When we visited the field sites, we would see places in some fields where there wasn’t a single plant. It looked like problems with poor drainage or loss of soil structure or low soil organic matter. So that is why a lot of the remedies that we’re trying are ways to improve soil health, like compost applications or changing crop rotations.”
Leclerc notes, “The biggest challenge is determining the underlying causes of the differences in productivity. In some cases it’s fairly easy to pinpoint, especially [with the wet conditions] in the project’s first year, but in other cases it is a lot more difficult. We are examining that aspect with very precise soil and subsoil analysis.”
Once a field’s variability is mapped and the causes of its differences in productivity are understood, then the field’s management zones can be defined and managed. “We can work on those management zones to improve the limitations, which are most likely soil-related. Or, if that cannot be done, the idea is to manage the different zones differently. So perhaps we might back off in terms of inputs on the lower productivity zones and reallocate those resources to the higher productivity zones, and look at changing the spacing perhaps on the higher productivity zones to take advantage of their higher yielding capabilities,” Leclerc says.
After the trials with the various management practices are completed, the project team will analyze the results to see which practices provide the most consistent benefits, how effective they are, and under what conditions they are most effective, and to determine which options make the most economic sense.
“At the end of the day, we need to look at the input costs and profitability,” Hemphill says. “The outcome needs to work for the growers and the processors in order for the industry to remain sustainable.”