Agriculture: Fertile ground for analytics and innovation

                             By Joseph Byrum, Syngenta, Senior R&D and Strategic Marketing Global Product Development, Innovation and Delivery


December 2015


The most fertile ground for operations research today is agriculture. That may seem like a surprising claim, considering data analytics continue to help save companies billions of dollars and move billions of packages and passengers around the world—and even so, there’s no lack of opportunity in these arenas. While advanced mathematical techniques have proved invaluable across diverse industries, operations research has yet to move in and dominate the field of agriculture, where it can play a leading role in feeding billions of people who may otherwise lack food security.


That’s no exaggeration. Within the next few decades, the global population will grow by nearly two billion. Although the world produces a tremendous amount of food today, it’s nowhere near enough to feed everyone. By 2050, caloric demand will increase by 70 percent and crop demand for human consumption and animal feed will double (World Resource Institute). The problem isn’t just that we need more food. This food must be produced in the context of formidable resource constraints, providing better nutrition for more people in the face of rapid environmental change while also cutting back our overuse of natural resources, ecosystems and the climate.


Consider, for instance, that crops require irrigation, and that more crops will require even more water—water we do not have. By 2030, an estimated 40 percent of water demand is unlikely to be met. On top of that, one out of every five acres of arable land is already degraded.


Food and agribusiness comprise a $5 trillion industry that accounts for 10 percent of global consumer spending, 40 percent of employment, and 30 percent of greenhouse-gas emissions. This massive industry does not change easily, but change—transformative innovation—is precisely what’s needed.


Meeting the future population’s entire demand for food will require disruption of the current trends. Over the last 50 years, agricultural technology has evolved largely along the lines of bigger and faster. New plows, better tractors, superior combines. Even recent advances in genetics require greater inputs to take advantage of superior yields.


In the next 50 years, we must be smarter by taking advantage of the operations research and analytics revolution, and its capability for making easier and more precise decision making. Data analytics can increase the efficiency of food production by optimizing the entire agricultural ecosystem. This quantified approach would use sensing, input modulation and analytics to enhance the efficiency of producing the world’s food.


Analytics and agriculture today

To better understand where operations research and analytics can take us in the future, it’s worth exploring the current state-of-the art innovations in analytics and agriculture, where the focus is on expanding access to and making more sophisticated use of information. For example, granular data for every ten-meter-by-ten-meter square of a field combined with the analytical capability to integrate various sources of information such as weather, soil and market prices has the potential to increase crop yield and optimize resource usage, thus lowering cost.


A wealth of agricultural information is gathered and distributed by means of smartphones, portable computers, GPS devices, RFID tags and other environmental sensors. Already, automation technologies such as GPS steering are being used to operate balers, combines and harvesters. RFID technologies track livestock to enhance food safety. Since 2010, European sheep farmers are required to tag their flocks and the European Commission has suggested the extension of this practice to cattle. RFID technologies also provide new possibilities for harvest asset management. By adding RFID tags, bales can be associated with measured properties such as weight and moisture level. Mobile communication networks and technologies, which are now commonly deployed in many areas around the world, have become a backbone of pervasive computing in agriculture.


Agriculture is thus becoming a knowledge-intensive industry. As farmers need to obtain and process financial, climatic, technical and regulatory information to manage their businesses, public and private institutions cater to their needs and provide corresponding data. The U.S. Department of Agriculture supplies information as to prices, market conditions or newest production practices. Internet communities, such as e-Agriculture, allow users to exchange information, ideas or procedures related to communication technologies in sustainable agriculture and rural development. So far, however, much of the research and development in this regard has focused on sensing and networking rather than on computation, analytics and optimization. Therein lies the opportunity.


Contributions to analytics in agriculture have mainly applied off-the-shelf techniques available in software packages or libraries without developing specific frameworks and algorithms. This state of affairs has only recently begun to change. While early work in analytics in agriculture focused on the design of relational databases, more recent approaches consider semantic web technologies, for instance in pest control, farm management or the integration of molecular and phenotypic information, for breeding. Others consider recommender systems and collaborative filtering to retrieve personalized agricultural information from the web or the use of web mining in localized climate prediction.


Geo-information processing plays an important role in computational agriculture and precision farming. Research in this area considers mobile access to geographically-aggregated crop information, region-specific yield prediction or environmental impact analysis.


Applications like these require advanced remote sensing or modern sensor networks. This includes distributed networks of temperature and moisture sensors (deployed in fields), orchards and grazing land that monitor growth conditions or the state of pasture. Space or airborne solutions make use of technologies such as thermal emission and reflection radiometers or advanced synthetic aperture radar to track land degradation or to measure and predict levels of soil moisture.  Other agricultural applications include plant growth monitoring and automated map building. A particularly interesting sensing modality consists in airborne hyper-spectral imaging which records spectra of several hundred wavelengths per pixel. With respect to plant monitoring, this makes it possible to assess changes of pigment compositions due to metabolic processes. This in turn allows for remotely measuring phenotypic reactions of plants due to biotic or abiotic stress. That’s important, because [this allows a grower to both detect a stress such as disease or insect pressure and optimize the use rate of chemicals for control in a sustainable manner].


Hyper-spectral imaging is being increasingly used for near-range plant monitoring in agricultural research. It enables basic research into the molecular mechanisms of photosynthesis, but is also used in plant phenotyping, which can help as an approach toward understanding phenotypic expressions of drought stress. Classical image analysis and computer vision techniques are being used in agriculture, too. Examples include automated inspection and sorting in production facilities, the detection of the activity of pests in greenhouses or the recognition of plant diseases.


Finally, artificial intelligence techniques are increasingly applied to address questions of computational sustainability. Work in this area considers algorithmic approaches toward maximizing the utility of land, enabling sustainable water resource management and the learning of timber harvesting policies. Thanks to the increased use of modern sensors, corresponding solutions have to cope with exploding amounts data recorded in dynamic and uncertain environments where there are typically many interacting components.


We’ve only begun to scratch the surface of what can be done with artificial intelligence (AI) techniques in agriculture. Most work in this area so far has not involved specifically trained data scientists. From the point of view of analytics, more efficient and accurate methods are surely available. Yet, computer scientists entering the field must be aware that methods they bring have to benefit researchers and practitioners in agriculture.


The need for practicality

Practitioners “out in the fields” need tools that yield results they can work with, ideally on mobile devices that run in real time to assist in their daily work. From the perspective of farming professionals, purely theoretical concepts or mathematical abstractions are of little use. They face real problems that can be addressed using scientific methods and advanced computing, but the tools need to be adapted to their needs in a way that produces tangible results.


The world’s food producers are technology oriented people, but practicality for them remains a core value. They know their business, and if a new technology does not fit into their work flows, they will either ignore it or wait until it meets their needs. Thus, there is a great need for more information technology training in the industry.


Advanced technology, properly harnessed, creates the ultimate in practicality. In our own plant breeding work at Syngenta, we have seen the power of customized operations research tools for delivering concrete results. We’ve used data mining and pattern recognition to breed elite varieties of seeds that deliver higher yields, and the results speak for themselves. Before we took full advantage of the power of analytics, we realized an average annual increase in yield across our portfolio of about 0.8 bushels per acre. That average is now closer to 2.5.


We will realize more than $287 million in cost optimization for Syngenta Seeds Product Development during the period from 2012 to 2016 from our operations research tools.  What that means is, we would have had to invest an additional $287 million to achieve the same level of genetic gain that we are realizing with the tools. Our goal is to do more with less, which helps fulfill our Good Growth Plan commitments of reducing agriculture’s impact on the environment and people that produce the crops we need, while helping to ensure a growing global population will have enough food for future generations.


The scale of opportunities in agriculture

This offers just a glimpse of the future potential for analytics in our industry. We expect that, with the availability of more computational power combined with sensing and networking technologies, new forms of farming may emerge thanks to operations research and data analytics.


The opportunities brought about by data collection and analytics have touched every market, from health care to retail. While the agricultural supply chain may not at first seem to be a prime target for optimization, it should be. From early stage research to farms and end-user customers, the agricultural supply chain is heavily reliant on small improvements in operational efficiencies and processes in order to increase crop yields, manage risk and create greater profit. This is particularly true for large-scale agribusiness where commodity crops are involved and small process adjustments have large impacts in terms of production.


The key to success is figuring out a business model that captures value from data at scale. In part, that is because the data are captured by disparate players in different parts of the value chain, such as seed companies, equipment manufacturers, traders and software developers. Managing and capitalizing on the critical data points is likely to require strategic partnerships and acquisitions, and potentially a reshaping of the industry structure. Meanwhile, emerging markets still lack high-quality, reliable data on production and demand. Establishing a systematic mechanism to capture the data could offer additional value-creating opportunities. In particular, rapid expansion of mobile technologies in rural populations could allow farmers in these areas to greatly improve productivity based on access to better information.


This is where the operations research community can have a big impact. Getting involved in agriculture is much more than just a ripe business opportunity. It’s lending a hand to solve one of the toughest challenges that humanity faces. Billions of lives depend on the coming analytics revolution in agriculture, and we hope the INFORMS community will rally around the cause.