Applications of Precision Agriculture and Digital Technology in Agroecology and Small Farms

Key Definitions

Agroecology — “The integrative study of the ecology of the entire food systems, encompassing ecological, economic and social dimensions” (Francis et al., 2003).

Precision Agriculture — Using technology to do “the right management practice at the right place and the right time” (Mulla, 2016).

Technology Lock-In — Inefficiencies in the market that result from a past choice that lock that decision into place, preventing competition from potentially more advantageous innovation (Arthur, 1989).

Market Failure — The failure of an economic system to result in desirable outcomes (Bator, 1958).

Artificial Intelligence (AI) — The automation of “activities that we associate with human thinking, activities such as decision-making, problem solving, learning…” (Bellman, 1978)

Computer Vision — “The automatic extraction, analysis and understanding of useful information from a single image or a sequence of images” (BMVA, 2017).

Remote Sensing — Scanning by UAVs, satellites, or other aircrafts to obtain information about an object or system

Rapid Phenotyping — The use of remote sensing to characterize plant qualities quickly for breeding.

Hyperspectral Imagery — Pictures that typically use hundreds to thousands of wavebands, instead of the typical red, green, and blue.

Multispectral Imagery — Pictures typically using 3 to 10 wavebands

Algorithmic Accountability — The concept and debate of whether programmers, companies, and institutions are responsible for the impact of their code

The future of farming is highly debated. At stake is the health of our planet and the lives of billions around the globe. We’ll need to use all of the resources at our disposal to effectively feed the world while protecting the environment. Although digital technology cannot solve all of agriculture’s challenges, it has a critical role to play. Unfortunately, in our current landscape, many of the digital tools are only accessible and applicable to larger farms, but it doesn’t have to be this way. With the right investments, precision agriculture and digital technology can assist the diverse cropping systems found on many smallholder farms around the globe.

Historically, agroecology has emerged as a counter-argument to intensive, high-input agricultural systems with short-term goals (Wezel et al., 2009). When referring to technology, previous authors have focused more on genetic engineering than digital technology (Vanloqueren, 2009). In contrast, this essay will focus on changing digital technological innovations that were built for industrial monocultures, and adapting them to more sustainable methods of farming.

Agroecology and precision agriculture have often been placed as philosophical rivals. Some have even proposed that due to scalability and diversity, many aspects of precision agriculture are not applicable to small-scale, diverse farms (Mulla, 2016). While that may be true today, the circumstances of our present do not dictate the opportunities of the future. Some researchers have made significant efforts to adapt technology such as neural networks (Daniel et al., 2008), harvesting equipment, and other digital technology to agroecology (Marel & Huyge, 2017). So, while technology has mostly been made for monoculture systems in the past, early dives into high technology to agroecology are small but promising. I will present arguments for how precision agriculture can assist agroecological systems by exploring precision agriculture techniques that have been used in diverse cropping systems and discuss additional areas where precision agriculture could improve outcomes for agroecological farming techniques.

In some areas, smallholder farmers will be able to take significant gains from large-scale precision agriculture, but in others, the fundamental technology needs to be shifted. Technological lock-in has meant that it is hard to use current technology on alternative farming methods, but that is not to suggest that it would be impossible to implement technology in these systems. Instead, we should view this situation as a market failure, and provide investment in research to correct it. Although technology can refer to many aspects of innovation we are going to focus on digital and mechanical technology, such as algorithms, software, sensors, robotics, and artificial intelligence. Other areas of innovation include soil conservation practices, other farming practices, genetics, and business models where research and investment could greatly benefit small-scale farmers around the world. I will discuss the opportunities for future investment and address the barriers to achieving this vision.

Technological lock-in, means that it is difficult to introduce new practices to revolutionize the industry.

User Experience and Software Design

When envisioning the food system of the future, science fiction has implanted a vision of factory-grown, highly processed food for space travel. While that is one possible future, we must evaluate all aspects of that and come to our future that is divorced from what we imagine it to be and instead find the solutions that best meet the challenges on planet Earth. The food system of the future must conserve resources, protect the environment, meet the nutritional needs of the earth, be affordable, pay workers well, and be overall enjoyable.

Likewise, while developing all this software, we must center the farmer, the consumer, and their needs. If we make technology without the input of those who use it, the projects are doomed to fail. I was once talking to a farmer, and he said he could “wallpaper his barn with how much data he had, but there was nothing to do with it.” Most farmers don’t go into farming to become data scientists, and many farmers have strong resistance to technology’s influence in their profession but can see some uses (Odom, 2010).

Additionally, we must resist the temptation to simply use technology because we are searching for practical applications for theoretical techniques that have been developed. In particular, remote sensing, computer vision, and machine learning have repeatedly used agriculture as a potential application of their work. While these techniques certainly are useful in many cases, we must resist temptation as computer scientists to only apply techniques because we can and use agriculture as our experimental playground without regard to the real needs of farmers. The application of digital technology in agriculture must take a needs-first approach. We must take the time to listen to the pains of farmers and researchers with an open mind and provide suggestions only when we think our skill sets provide the best solutions. A common pitfall for those studying and working in technology is to metaphorically treat every problem as a nail for the technology hammer.

To make precision agriculture work for small scale farms, it also needs to be affordable. The economics also have to work out that it saves the farmer time, money, or resources and is not just another thing to manage. Luckily, many aspects of digital technology are highly scalable. A great application could be in record keeping or disease identification. Additionally, by making many of these tools accessible, we can empower suburban and urban homeowners to grow some of their own food and take advantage of the highly productive soils they live on. This will all make it easier to farm on small plots of land in urban and suburban areas that are now allocated to grass.

We must also consider that small-scale farming varies greatly across the globe and thus precision agricultural adoption has also varied (Zhang, 2002). Some technology, like variable-rate fertilizers depending on soil need, might make sense on most farms globally. Others, like large-scale machinery, might not be accessible to parts of the developing world. We must also invest more work into farmer decision-support systems that integrate sensing systems and help overall farm operation (Mcbratney, 2005).

Sensors and Remote Sensing

Sensors and Remote Sensing are so far only available to large, monoculture farms, adjusting this is important for small farms.

A substantial barrier to effective management of soils, livestock, and crops is getting real-time data on the conditions on the farm. Multispectral remote sensing from drones can scan fields on a daily basis to search for irregularities and diseases. This is evidenced by studies by in which they x, y and z. By looking at wavelengths beyond the human eye, we can see heat and hints at the chemical structure reflected, which can help in the identification of ripeness and diseases. Hyperspectral imagery can assist in the detection of some diseases and damage (Zhao, 2016; Yang, 2016; Reynolds, 2016). This can allow a more reactionary approach which eliminates diseased plants before the disease becomes widespread, and eventually eliminate the need to do preventive pesticide sprays that can harm both farm workers and beneficial insects.

A key component to making these advancements impactful is long-term, thorough, and diverse datasets that illustrate the growing conditions of not just a few commercially dominating crops, but information on a wide, genetically-diverse set of crops that these systems can call upon to infer decisions about new varieties that come their way. Unless information is available on more than just a few key currently economical crops, precision agriculture will just further enforce the domination of large industrial monocultures instead of enabling a sustainable, diverse future.

Right now satellites do not operate at a resolution that is suitable to small-scale farming. It will take much longer for satellites to be applicable to small-scale agriculture but it will eventually become so. The size and cost of satellites are shrinking while the resolution is increasing (Denis, 2017). The cost of expanding this technology to small-scale farms will be minimal if the data and algorithms are already available. However, it is important to note that many algorithms such as row detection that work well in monocultures will likely fail in diverse, small farms. Additional investment will be needed to adjust row detection to polyculture and intercropping systems.

In addition, remote sensing techniques can be helpful in developing breeding programs that focus on aspects such as nutrition content and other crop quality aspects (Mcbratney, 2005). Through a technique called rapid phenotyping, which uses remote sensing to rapidly gather data on plant traits for breeding, new crops can be developed with diverse traits that are helpful to small farmers.

Finally, one of the most obvious and well-used applications of precision agriculture is to more accurately apply nutrients and water to the soil. With better developed sensors, farmers will easily be able to test the soil, often using small robotics, and the organic nutrients they adding to the soil to assure they are building the soil up and not contributing to excess nutrient run-off. In addition, they’ll be better able to sense water content and incorporate weather forecasts to ensure better water distribution and application.

Data, Open Source Software and “Right to Repair”

The diversity of agroecological systems theoretically is much larger than for conventional farming. Thus, one could argue it is much harder to develop machinery that is applicable to every application. In order to break this paradigm, we must open up both the software and hardware of these systems for adjustment, so that farmers and technicians can customize the applications they use for their specific farms. Some farmers have been fighting for a “Right to Repair”, citing that manufacturers have purposely restricted their ability to fix and modify the machinery they buy (Fitzpatrick, 2017).

Farmers must also be able to access the data they collect from their farm to customize the information they receive back. If data is just collected and maintained by a few companies with proprietary software, it will never be available to small farmers who make up too segmented of a market-share for large corporations to focus on. As more of the world rapidly gains access to smartphones, data on one’s own farm in cooperation with open data can lead to management insights such as recognizing weeds, diseases, measurements and access reference databases (Maurel and Huyghe, 2017). Although, data needs to become more available and standardized for research and development in agroecology to occur (Shekhar, 2017) (Maurel and Huyghe, 2017).

Food Supply Chains, Storage, and Waste Reduction

Data-driven monitoring and function of food supply chains could reduce waste, increase food safety (Golan, 2004), and reduce food costs. Currently, if a foodborne illness is detected, it can take months to trace back to a farm, if it is even done, but digital technology like CDC’s FoodNet can help (King, 2006). Technology can also provide connections to reduce food waste like MIT’s FoodCam and programs like Copia. Most of the software developed today for supply chains is made for large aggregators and inaccessible to local farmers. Software, such as Restaurants Eat Direct (RED) Food, could help by directly connecting farmers with restaurants or consumers while cutting out much of the innovation cost. Another area of investment is making checklists that assist organic farmers in record keeping for certification, lowering the labor put into record keeping.

Artificial Intelligence

Artificial intelligence (AI), especially machine learning, is extraordinary at learning systems as they are, but are rarely used in agriculture for developing new models. We can call these two types of goals reductionist and expansive learning (also called computational creativity). In reductionist learning, machine learning sets out to optimize a set of solutions within the limited constraints given but does not create any new content beyond what is given. In any of these systems, machine learning will still be very poor at accounting for externalities without considerable effort. In short, when we put AI to the task of optimizing a failed agricultural model, it will do exactly what we just told it to. They are are models of artificial intelligence that avoid these basic pitfalls — that can behave creatively if we allow them to, but we must do so with purposeful intention.

If we want machine learning to play a purposeful role in the development and design of new agricultural models, we must fundamentally shift how we use the technology. Although it is still useful to use a reductionist machine learning model at times, like to classify diseases or deficiencies, this does not fully unlock the capabilities of the technology. If over time, we can better model plants and farms as ecological systems, we could allow machine learning to assist in the design process as has been done with the design of bridges and quadcopters (Walsh, 2016). Skilled simulations would be able to model years of agricultural practices in a fraction of the time, developing new seed mixtures and plant arrangements that are non-linear and would be unintuitive to human farmers and researchers. Increased data collection with sensor integration will assist in this process and the customization of farms.

In addition, we must better develop and utilize the methods of computer vision that we use. Most techniques today are limited to a single crop and a few symptoms, and highly dependent on ideal conditions (Barbedo, 2016). The diverse ecosystems that comprise many diverse, small-scale farms are hardly ideal for these touchy algorithms and techniques. In the real world, diseases, pests, and nutritional deficiencies can result in very similar symptoms. Better techniques are being developed, but they need more investment, data, and utilization to be effective.

In particular, better techniques are needed for accounting for varied light conditions, such as shadows and glare, for in-field conditions. Methods should also be able to account for varied camera angles and plant positions.

Pest Management

Hyperspectral imagery is challenging the tendency in conventional agriculture to use preventative pesticides to protect against future pest infestations. New technology is enabling quicker detection that will allow for a faster reaction to pest infestations (Mulla, 2016). Adopting these techniques along with a reactionary instead of a preventive model can lower pesticide use with the effect of reducing costs, protecting the environment and result in faster plant growth (Bongiovanni, 2004). Using near-infrared cameras, we will be able to detect the first stages of infestation, rather than waiting for visual cues (Alves et al, 2015). Image processing also shows promising signs of being able to identify insects and report pests without human labor (Korinšek et al, 2016; Ding & Gram, 2016). Recommender systems will allow farmers to better process the data they have and make sustainable decisions (Mahaman, 2002).

This is one area where small farms can take direct gains from large-scale precision agriculture. However, right now most of the data available is on commodity crops (Venugoban, 2014) (Qing, 2012) (Karimi, 2006) like corn and soybeans and not on the fruits and vegetables that small-scale farmers rely on. Particularly this is true in developing countries where small research budgets have prevented precision agriculture research, where many of the smallholder farmers of the world live. Besides a few key cases (Boniecki, 2016), machine learning has focused on Bayesian classifiers which only separate targets into two categories. Instead, we should focus on training machine learning systems on a variety of pests and develop the wider knowledge needed for diverse system. Overall, we must enhance robotics and our remote sensing efforts to include many more varieties of crops, including specialized and heirloom varieties.

Labor, Automation, and Robotics

In order to bring down the price of food to make it accessible to the poor, the operations must be cost-effective. Since there are often labor shortages and it contributes an incredible amount of cost to production, robotics has a place to step in. However, as stated before, it must be low-cost, customizable, and applicable. Small weeding, harvesting, and planting robots have been made, but need a lot more investment to thrive. These small robots will also be adjusted to different methods such as no-till farming, and to handle diverse ecosystems.

Material science and the miniaturization of batteries will also play a large role. It will allow for more renewable resources, such as sunlight, to be used to fuel the farming robots of the future. These will include both ground vehicles and Unmanned Aerial Vehicles (UAVs). Ground vehicles will assist in manual labor and also be able to sense additional aspects of the soil.

UAVs present incredible opportunities for remote sensing, but regulations prevent genuinely autonomous operation, and more work needs be done on processing the data effectively for useful insights. Sensors need to become cheaper and more accurate to allow for real-time data collection in ways that accessible to farmers. Advancements in artificial intelligence and computer vision will enable agricultural robots to navigate their environment better and make intelligent conclusions.

Conclusion

Digital technology alone cannot solve the challenges faced by small-scale farmers. However, through the use of algorithms, software, sensors, robotics, and artificial intelligence, digital technology can help assist in conserving resources and making sustainable, affordable food to feed the world. It presents the opportunity to monitor complex ecosystems and use AI to come up with unique solutions never before imagined.

Digital technology also presents the opportunity to take many of the practices that were previously only practical to small farms and scale them up in ways that were previously too labor intensive. It also has the potential to create incredibly inequality where just a few corporations own the code that feeds our entire planet. Additionally, if these developments only help a handful of large-scale farms, it introduces barriers that could lead to further consolidation. Although many advancements in precision agriculture could benefit all farmers, there are some key areas where the market is likely to reach technological lock-in and investment in research and development could create a more equitable, sustainable food system.

References

Arthur, W. Brian. “Competing technologies, increasing returns, and lock-in by historical events.” The economic journal 99.394 (1989): 116–131.

Bator, Francis M. “The anatomy of market failure.” The quarterly journal of economics 72.3 (1958): 351–379.

Barbedo, Jayme Garcia Arnal. “A review on the main challenges in automatic plant disease identification based on visible range images.” Biosystems Engineering 144 (2016): 52–60.

Barbedo, Jayme Garcia Arnal. “Digital image processing techniques for detecting, quantifying and classifying plant diseases.” SpringerPlus 2.1 (2013): 660.

Bellman, Richard. An introduction to artificial intelligence: Can computers think?. Thomson Course Technology, 1978.

BMVA. “What is computer vision?” www.bmva.org/visionoverview. Accessed December 23, 2017.

Bongiovanni, R., and J. Lowenberg-Deboer. “Precision Agriculture and Sustainability.” Springer. Precision Agriculture, Aug. 2004. Web. 28 Mar. 2016.

Boniecki, P., K. Koszela, H. Piekarska-Boniecka, J. Weres, M. Zaborowicz, S. Kujawa, A. Majewski, and B. Raba. “Neural Identification of Selected Apple Pests.” Computers and Electronics in Agriculture, Jan. 2015. Web. 28 Mar. 2016.

Cook, S. E., et al. “Is precision agriculture irrelevant to developing countries.” Precision Agriculture (2003): 115–120.

Daniel, J., Andrés, P. U., Héctor, S., Miguel, B., Patrick, V., & Marco, T. (2008). A survey of artificial neural network-based modeling in agroecology. Soft Computing applications in industry, 247–269.

Denis, Gil, et al. “Towards disruptions in Earth observation? New Earth Observation systems and markets evolution: Possible scenarios and impacts.” Acta Astronautica 137 (2017): 415–433.

Ding, Weiguang, and Graham Taylor. “Automatic Moth Detection from Trap Images for Pest Management.” Automatic Moth Detection from Trap Images for Pest Management. Computers and Electronics in Agriculture, n.d. Web. 28 Mar. 2016.

Fitzpatrick, Alex. “Apple and Farmers Fight in Right to Repair Movement.” Time, 22 June 2017, time.com/4828099/farmers-and-apple-fight-over-the-toolbox/.

Francis, Charles, et al. “Agroecology: the ecology of food systems.” Journal of sustainable agriculture 22.3 (2003): 99–118.

Galindo, Pau Aragó, et al. “Participative site-specific agriculture analysis for smallholders.” Precision agriculture. 13.5 (2012): 594–610.

Golan, Elise, Barry Krissoff, and Fred Kuchler. “Food traceability.” Amber Waves 2.2 (2004): 14.

Karimi, Y., et al. “Application of support vector machine technology for weed and nitrogen stress detection in corn.” Computers and electronics in agriculture 51.1 (2006): 99–109.

King, Lonnie J. “CDC food safety activities and the recent E. coli spinach outbreak.” Testimony before the Committee on Health, Education, Labor and Pensions, United States Senate (2006).

Korinšek, G., M. Derlink, M. Virant-Doberlet, and T. Tuma. “An Autonomous System of Detecting and Attracting Leafhopper Males Using Species- and Sex-specific Substrate Borne Vibrational Signals.” Computers and Electronics in Agriculture, n.d. Web. 28 Mar. 2016.

Maurel, V. B., & Huyghe, C. (2017). Putting agricultural equipment and digital technologies at the cutting edge of agroecology. OCL, 24(3), D307.

McBratney, Alex, et al. “Future directions of precision agriculture.” Precision agriculture 6.1 (2005): 7–23.

Mulla, David. “Twenty Five Years of Remote Sensing in Precision Agriculture: Key Advances and Remaining Knowledge Gaps.” Biosystems Engineering, n.d. Web. 28 Mar. 2016.

Odom, William. “Mate, we don’t need a chip to tell us the soil’s dry: opportunities for designing interactive systems to support urban food production.” Proceedings of the 8th ACM Conference on Designing Interactive Systems. ACM, 2010.

Qing, Yao, et al. “An insect imaging system to automate rice light-trap pest identification.” Journal of Integrative Agriculture. 11.6 (2012): 978–985.

Raghavan, Barath, et al. “Computational Agroecology: Sustainable Food Ecosystem Design.” Proceedings of the 2016 CHI Conference Extended Abstracts on Human Factors in Computing Systems. ACM, 2016.

Reynolds, Gregory, Carol Windels, Ian MacRae, and Soizik Laguette. “Remote Sensing for Assessing Rhizoctonia Crown and Root Rot Severity in Sugar Beet — Plant Disease.” Plant Disease, n.d. Web. 28 Mar. 2016.

Schultz, A., R. Wieland, and G. Lutze. “Neural networks in agroecological modelling — stylish application or helpful tool?.” Computers and Electronics in Agriculture 29.1 (2000): 73–97.

Shekhar, Shashi, et al. “Intelligent Infrastructure for Smart Agriculture: An Integrated Food, Energy and Water System.” arXiv preprint arXiv:1705.01993 (2017).

Shekhar, Shashi, et al. Agriculture Big Data (AgBD) Challenges and Opportunities From Farm To Table: A Midwest Big Data Hub Community† Whitepaper. 14 Dec. 2017, www-users.cs.umn.edu/~shekhar/talk/2017/NIFA_whitepaper_AgBD_final_12_14_2017.pdf.

Tokekar, Pratap, et al. “Sensor planning for a symbiotic UAV and UGV system for precision agriculture.” IEEE Transactions on Robotics 32.6 (2016): 1498–1511.

Toyoshima, Morio. “Trends in satellite communications and the role of optical free-space communications.” Journal of Optical Networking 4.6 (2005): 300–311.

Vanloqueren, Gaëtan, and Philippe V. Baret. “How agricultural research systems shape a technological regime that develops genetic engineering but locks out agroecological innovations.” Research policy 38.6 (2009): 971–983.

Venugoban, Kanesh, and Amirthalingam Ramanan. “Image classification of paddy field insect pests using gradient-based features.” International Journal of Machine Learning and Computing 4.1 (2014): 1.

Walsh, Jeff. “Machine Learning in Design: The Evolution of AI.” Redshift, 8 Dec. 2017, www.autodesk.com/redshift/machine-learning/.

Xie, Yiqun, et al. “Spatially Constrained Geodesign Optimization (GOP) for Improving Agricultural Watershed Sustainability.” Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17), Workshop on AI and OR for Social Good, San Francisco, CA, USA. 2017.

Yang, Chenghai, Gary Odvody, Alex Thomasson, Thomas Isakeit, and Robert Nichols. “Change Detection of Cotton Root Rot Infection over 10-year Intervals Using Airborne Multispectral Imagery.” Computers and Electronics in Agriculture, n.d. Web. 28 Mar. 2016.

Zhang, Chunhua, and John M. Kovacs. “The application of small unmanned aerial systems for precision agriculture: a review.” Precision agriculture 13.6 (2012): 693–712.

Zhang, Naiqian, Maohua Wang, and Ning Wang. “Precision agriculture — a worldwide overview.” Computers and electronics in agriculture 36.2 (2002): 113–132.

Zhao, Yun, Yong He, and Xing Xu. “A Novel Algorithm for Damage Recognition on Pest-infested Oilseed Rape Leaves.” Computers and Electronics in Agriculture, Nov. 2012. Web. 28 Mar. 2016.