A bird’s-eye view of India’s farmland
Amid the glitz of satellite and drone-powered remote sensing in agriculture, India’s farmers may be better served by a lower-tech option.
Mark Jeunnette’s engineering career has taken him to some far-flung places. He’s spent time in Nepal working on energy access and in Ethiopia designing products for small scale farmers. Now, he’s building cameras at MIT.
“It’s more accurate to say I build camera arrays,” he clarifies. “I think of building a camera as taking it down to the electronics board and sensor array, and then building the optics. I’m taking products you can buy off the shelf for normal imaging applications and combining them into one box.”
That mish-mashed box of off-the-shelf imaging hardware could have a big impact on a lot of lives. Jeunnette’s cameras are being used as part of a project (and his PhD) at MIT’s Tata Center for Technology and Design to improve field and crop information for the 85 percent of farmers in India tilling small plots of land. When affixed to the wheel of a two-seater Cessna airplane, the cameras take a series of black and white images of India’s farmland from a height of about 1,200 meters. Shading variations between the patchwork of fields in each image can indicate issues like flooding or crop health.
Jeunnette’s approach to aerial imaging may sound less glamourous than other modern technologies, like satellite imaging or drone surveillance. But while drones and satellite images both have practical applications in highly sophisticated farm operations, neither are well suited to India’s agriculture sector, Jeunnette says.
“A lot of farmers practice agriculture the way their fathers and grandfathers did,” he explains.
India’s agriculture sector accounts for 50 percent of livelihoods in the country, according to the 2015–2016 national agriculture census. Most of these livelihoods are made on farms less two hectares in size, with little to no high-tech machinery and few informational resources for support.
“[Farmers’ issue is] the lack of a reliable source of information for what to do when a problem crops up during the season, such as a pest or a drought,” says Jeunnette.
The government provides extension services that aim to give small farmers personalized recommendations from extension centers. Less that 10 percent of farmers report using those services, however, according to a 2016 report from India’s Ministry of Statistics and Programme Implementation. This is in large part due to a lack of available agronomists and the high cost of reaching farmers in such remote areas.
Other organizations are working on solutions to fill the gap by providing services via text message, smart phone or call-in centers, but for the most part, farmers resort to contacting neighboring farmers and local input suppliers when they need help.
“The general consensus is that input suppliers have a conflict of interest and will end up selling the farmer whatever will make the most money, not what’s necessarily best for the farmer,” Jeunnette says.
Planes, drones and satellites
The key ingredient missing from government and private e-extension services is field-level information. Jeunnette says he has heard of projects trying to collect data from farmers directly by mobile phone. “But the cost of making and distributing phones with special cameras for measuring crop health is prohibitive when you’re talking about over 100 million farmers,” he adds. “Without special equipment, you’re limited to the problems that the farmer can see with his or her own eyes.”
Instead, remote sensing struck Jeunnette as a more efficient and promising way to collect this type of data.
There are already plenty of use cases for remote sensing in agriculture, construction and other sectors, which are experimenting with high resolution imaging technologies as a cost-effective and scalable way to monitor and analyze business operations. Drones, or unmanned aerial vehicles, have attracted a lot of buzz and funding in recent years in particular.
Indeed, venture capital for agriculture drones hit $389 million in 2015 — a 237 percent increase from the previous year — in response to the technology’s perceived potential in helping farmers gather granular field data and operate more efficiently. Drones have demonstrated the capability of simple, visual discoveries, such as a water pipe leak in one corner of an irrigated field, and of more complex tasks, like measuring the reflectance of light off leaves — an indicator of plant health.
The drawback to drones, however, is that the relatively small models available on the market are not well suited to large farming operations. Deploying and charging the drones and then downloading the acquired images is both time and manpower-intensive. What’s more, the information they provide often comes too late to make a significant difference to farmers or to warrant the cost of the technology.
In theory, agriculture drones could be well suited for India’s farms, where the average plot size is small. But Jeunnette says most farmers don’t need highly detailed data about their own fields.
“The assumption is that he is already intimately aware of the spatial differences in his field, such as which parts drain better or are prone to flooding,” he explains. “Instead he wants a heads up about plant stress before he can see it with his own eyes.”
For that, farmers need access to comparative data on what is happening on nearby farms or regionally. Visually, that information could be captured by drones’ flying capabilities. “You can stand at one point on a farm and turnaround and count 20 to 30 fields in your immediate field of view, all easily in walking distance,” Jeunnette explains.
In practice, he found that collecting a broader dataset from drones is inefficient.
“I took five drones and a single operator in a van, drove to a location, launched all five simultaneously, let them fly their patterns and return autonomously, downloaded the data, updated new flight plans, drove to the next location, and sent them out again,” he recounts. “At that point, in terms of area coverage, drones couldn’t compete with manned aircraft that fly 4,000 feet [1,200 meters] above ground level.”
He continues: “I think drones will eventually get there, but with the operational complexities and regulations in India that do not allow pilots to fly them out of line of sight, or a single pilot to fly multiple drones, it will still be a while until they’re useful for this purpose.”
On the other end of the imaging spectrum, Jeunnette considered satellite imagery, which is also gathering pace as a tool for the agriculture industry. A number of startups are using satellite images to provide farmers with field-level insights and produce higher level analytics, such as regional crop yield predictions, which are of interest to crop insurers, commodities traders, hedge funds, and governmental agencies.
The benefits of satellite imagery are that it is relatively inexpensive to acquire, and it can cover a large amount of ground in a short period of time. The images tend to be low resolution and are in limited supply, however, because they can only be collected for a target area when the satellite passes over. Cloud cover blocking the ground view is also a problem. As a result, there has been some disappointment in advanced agriculture markets over satellites’ utility for farming intelligence. Jeunnette says he felt similarly about using satellite imagery for India’s smallholder farmers.
“It would be really challenging to get enough information from one image a week at that scale, especially as during the rainy season, cloud cover would be a big issue,” he says. “In the Indian context, I felt that the long return time combined with the low resolution meant satellites [weren’t] suited to the task.”
The latest wave of agriculture tech innovators is trying to mitigate some of the drawbacks of satellite imagery by applying heavy layers of image processing and using machine learning and novel sensing techniques to get farmers actionable information. For Jeunnette, using off-the-shelf cameras and a small airplane seemed like an all-around more reliable way to get the information he sought.
“My goal is to be able to distinguish between the different [small farming] plots, and for that I don’t need more than four- to five-meter resolution, which is still pretty high compared to satellite imagery,” he explains.
Black and white
To gather this kind of data, Jeunnette’s approach requires mounting several cameras on the side of a chartered airplane and flying up a couple of times per week to snap images of the farmland below.
For the cameras, Jeunnette opted to build them himself, to ensure the image resolution would be adequate. There were options Jeunnette could purchase on the market, but they were expensive and didn’t offer the flexibility to change filters to capture different wavebands needed to distinguish between different type of crops the planes might fly over, the time of year, or the different indices the data would be used in.
“Most of the cameras available on the market come with a predefined set of wavebands and cost at least $3,500, though they are getting cheaper,” he explains.
Even with the cameras, Jeunnette has chosen a low-tech option. He uses black and white cameras, because he has found that the red-green-blue (RGB) filter built into color cameras interferes with the narrow-wavebands of light he is trying to capture.
A typical RGB camera will capture three different wavebands between 400 and 700 nanometers, with each waveband about 100 nanometers wide. A black and white camera with a narrow-band filter can capture bands of light that are just 10 nanometers wide.
“Without a filter in front of it, a black and white camera can be sensitive from 400 nanometers, up to 900 or 1,000 nanometers, which gets into the near infrared spectrum,” he explains. “Some of the narrow-band filters I use are actually passing light outside of the visible spectrum, allowing me to see things the farmer can’t see on the ground.” This, in turn, allows for more specificity in image analysis.
Flying at 1,200 meters, however, yields lower resolution and larger pixel images than drone imaging would. Jeunnette explains that he has had to figure out how to expediently process the flight images he collects, compile the pixels together, and then “still do something with it.”
The solution he is experimenting with is converting the images into a collection of crop indices for each farm.
Among the most common indices in use by agronomists and plant breeders is the Normalized Difference Vegetation Index (NDVI), which offers a general measure of crop vigor or crop stress by looking at light reflectance off of crop leaves. (Light reflectance gives an indication of a plant’s chlorophyll activity, a key indicator of plant health.) Jeunnette’s cameras can capture the two wavebands — one in the visible spectrum and the other in the near-infrared spectrum — needed to measure a field against this metric.
For the data to be useful, it has to be collected and analyzed regularly. “The flight surveys are meant to be flown regularly and often during the [growing] season, up to two times per week,” Jeunnette explains. The images collected from each survey allow him to build charts that reveal each farm’s crop index as it changes over a growing season. “It’s that data that would be interpreted by extension services and agronomists to provide advice to farmers.”
The widest reach
For Tata Trusts, one of India’s oldest philanthropic organizations and founder of MIT’s Tata Center, Jeunnette’s idea for low-tech aerial imaging stands out among its hundreds of small farm interventions.
“The Tata Trusts employ a wide range of strategies to improve farmer livelihoods, from providing irrigation equipment to agronomist expertise, but they are always seeking to make their efforts more scalable,” explains Jason Prapas, the Tata Center’s Translational Research Director. Remote imaging can potentially gather information on hundreds if not thousands of small farms at a time. It can also reduce the cost of providing information that farmers need by “digitizing the crucial but costly step of observing the state of farms over time,” Prapas adds.
Jeunnette is quick to note that his work is still in the research phase. So far, he has built two cameras and tested them on several chartered flights. He plans to conduct a full season of flights and image collection this year.
Flight testing has come with its own set of challenges. Jeunnette can collect the type of images he wants on two- or four-seater, propeller-powered Cessna 172s — “the most manufactured planes in the world” — but finding chartered flights in the regions of India he needs isn’t easy.
“Small scale aircraft are generally quite few and far between,” he says. “Part of the challenge in implementing my strategy is finding that confluence of an area of land that has the right aircraft and pilots around, and where extension services are connected to the farmers.”
The reason proximity to extension services is so critical is because those providers, whether government or private, are likely to be Jeunnette’s main data customers. Data aggregators serving farmers in the U.S. and Europe are able to sell their services directly to farmers, mostly on a subscription basis. A direct model is unlikely to be successful in India, however, because few farmers would be able to afford it.
The current spectrum of government extension services in India varies from state to state. A number of private organizations including Microsoft Research, Tata Consultancy Services, Mahindra Group, and a number of non-profit organizations, are also developing e-extension services in parallel to or in partnership with government efforts. “They’re trying to do a lot with SMS and IVR (interactive voice response), and some have created really powerful channels,” says Jeunnette.
But while they’re delivering weather and crop-pricing information effectively, they fall short on tailored information for farmers’ crop health. “That’s where I saw a niche — to support these extension services in personalizing individual information,” Jeunnette adds.
The crop indices that Jeunnette compiles from the images he aggregates can be translated into crop health metrics that extension service providers can analyze and use to make recommendations. The maps help identify where farmers are experiencing crop stress, for example. Then the crop indices can be analyzed alongside other data the extension services have on hand to identify the source of the stress, whether that’s from water, nitrogen deficiency, pest infestation, or disease attack. From that, they can provide recommendations directly to farmers.
“My research purposefully ends here because the extension services have the agronomic knowledge to translate the indices and wavelengths into recommendations for farmers,” explains Jeunnette. He adds that leaving recommendations to other experts allows his data to be as objective and independent as possible.
Weighing the impact
Of course, whether or not Jeunnette’s research ultimately helps farmers depends on who uses it. Government extension services are unlikely to pay enough to cover the costs of collecting and compiling the data. Jeunnette and the Tata Center therefore have to consider what private actors could. Large NGOs, like the World Bank and the World Food Programme, and philanthropic organizations with programs committed to smallholder farmer livelihoods are an obvious choice.
“Many groups expend a significant amount of resources to collect farmer data manually,” Prapas says. “These groups could divert some of these resources to paying for imagery.”
On the private sector side, Jeunnette adds that the same players that are interested in crop yield predictions in advanced agricultural markets could be interested in data for India: crop insurers or commodities traders, for example.
Prapas cautions that “there are ways this data could be used to impact farmers negatively” — for instance, if agrochemical companies bought and used it to target farmers with specific products that they may not need. This is something that Jeunnette wants to hedge, since farmers in some parts of the country are already overly dependent on input suppliers.
“While I think the risk of data acquisition [itself] being biased is relatively low, the goal of supporting farmers needs to be the driving factor in the use of the data, which may mean implementing protections for the most vulnerable stakeholders — the farmers,” he says.
Both he and Prapas point out that staying ahead of negative impacts are something all social enterprise initiatives have to do. Jeunnette adds that he feels personally motivated to do that as well.
“Watching out for those conflicts of interest is definitely a big part of what I’d want to be aware of,” he says. “My personal interest and motivation in agriculture stems from its value as something so essential to humans and their survival.”
This case study is featured in Demand’s Winter 2018 issue. It was written by Jessica Pothering, Demand’s managing editor, with reporting from Louisa Burwood-Taylor, editor of AgFunderNews.
