A vision for digital agriculture at scale: Internet of Small Things to the rescue
Digital transformation leaves no stone unturned as it remakes the world for the better. Agriculture is no exception, and is witnessing a digital revolution that aims to bring the most advanced practices and analytics to the field (pun intended). The transformation is coming just in time — the world’s expected population increase to 9 billion by 2050 will only amplify the challenge to agriculture to feed and nourish everyone in the face of dwindling arable land and water sources.
The digital reshaping of agriculture will bring heightened visibility and control. Using Internet-of-Things (IoT) technologies can control an agricultural ecosystem at the local, regional, and even global level, leading to higher crop productivity; better decision making around farming practices like planting, adding soil supplements, and harvesting; and improved water and energy utilization.
A Lattice Analogy
Our vision is what we call LATTICE — a smart, connected farm where networks of sensors continuously collect data, which gets processed and analyzed in near real time (with latency of only a few tenths of a second), and refined by machine learning to adapt practices for higher crop yields and throughput. The name LATTICE draws its inspiration from lattice arrays that can be integrated into innovative, cohesive structures.
This vision depends on integration — a seamless synchronization of IoT solutions, data processing, actionable analytics, and policymaking. IoT devices will collect such information as soil nutrient composition, water, and fertilizer needs. This data will be processed and analyzed on-site by “edge” computation — in which the data is processed by or near the devices themselves (near-data processing) — and uploaded to the cloud, where it can be combined with additional information like weather conditions and market prices. The data analytics will help farmers monitor crop growth; decide what, where, and when to grow; and devise strategies for increased yield. Machine learning and deep neural networks will continually crunch the big data flow to “learn” and fine-tune the analytics and recommendations.
Digital AI or Full-Stack AI
Our innovation is what I call “Digital AI,” in which we make AI more application-focused by compressing (“pruning”) the algorithms to make them fit into the embedded devices — think supercomputers in your palm — in order to achieve on-device computation. On-device computation results in a reduced need for data transfer, and so higher cost savings — cellular IoT plans can get quite expensive as they often are priced per device and reliable wireless connectivity still is not ubiquitous. Transferring data from its sources (e.g., sensors) to edge devices can consume more than 99 percent of a device’s energy, so decreasing data transfer by using near-data processing also produces significant energy savings.
Sensor-Edge-Cloud Continuum
Think of the LATTICE hierarchy as sensor nodes → edge devices → cloud virtual machine instances. The goals are to bring latency-sensitive analytics closer to the data source, and to move heavier analytics that don’t need near-real-time data processing to higher rungs of the data-flow spectrum, such as on beefier edge devices or in the cloud.
Even greater benefits to agricultural productivity and energy and water conservation can accrue if data processed by Digital AI is aggregated beyond a single farm to encompass a large number of farms. Network effects increase with the number of farms and the wealth of data they provide. For example, the Farmers Business Network Inc. has shown how valuable it is to share data via a crowdsourced database of agricultural input costs and performance benchmarks. Further progress in this area depends on resolving privacy concerns and the legal status of agricultural data.
Gaps in data engineering and machine learning also need to be addressed. These issues include further optimizing the processing and analytics load across sensors, edge devices, and the cloud so it can run on resource-constrained, battery-powered devices; achieving higher-performance cloud computation; optimizing compute-intensive visual workloads from drone imaging; resolving ethical questions around sharing farm data with entities like agricultural companies and insurance providers; and developing secure end-to-end agro-analytics.
Industry Partnership
Our vision for digital agriculture at scale grew out of a collaboration with Microsoft Research and Microsoft Azure. The project has received funding from the National Science Foundation (NSF) Cyber-Physical Systems (CPS) program on machine learning for secure agro-analytics (which operates under the aegis of the NSF Directorate for Computer and Information Science and Engineering, CISE). Further, our cloud and edge optimization work is funded by Adobe Research and Amazon AI, focusing on the edge-cloud continuum for graceful partitioning.
Teaching and Evaluation Testbed
Finally, to foster innovation in digital agriculture and IoT for teaching, we have a new National Institute of Food and Agriculture (NIFA) higher education challenge grant to support inculcating computational thinking in both undergraduates and graduates in our program. One innovative feature in this pursuit has been the development of podcasts to go with traditional teaching.
Our work is being experimentally validated at Purdue farms, such as the Throckmorton Purdue Agricultural Center (TPAC), which consists of more than 830 managed acres, and eight Purdue Agricultural Centers (PACs) across the state of Indiana, run by Purdue’s Agronomy Center for Research and Education (ACRE). These efforts leverage our living IoT testbed, where the fitness of sensors can be challenged by harsh conditions. To alleviate this factor, we are developing technologies to remotely analyze the health of these sensor nodes.
In these testbeds, we can create innovative data analytics solutions, some of which can hopefully be applied at scale across the globe for a sustainable and secure future. With 690 million people — almost 10 percent of the world population — who still do not have enough to eat, according to the World Food Programme, the digital transformation of agriculture cannot come soon enough.
Somali Chaterji, PhD
Assistant Professor
Department of Agricultural and Biological Engineering
Director, Innovatory for Cells and Neural Machines (ICAN)
Leadership Team member, Purdue-WHIN (Wabash Heartland Innovation Network)
College of Engineering,
Purdue University
