SIAM News Blog
SIAM News
Print

FarmBeats: Improving Farm Productivity Using Data-Driven Agriculture

By Zerina Kapetanovic, Ranveer Chandra, Tusher Chakraborty, and Andrew Nelson

The global demand for food is expected to increase by 70% by the year 2050 compared to 2010 levels. Achieving this increase in food production has become even more challenging as the resources we rely on are starting to diminish. For instance, water levels are receding, the amount of arable land is decreasing, and climate change has become more imminent.

Data driven agriculture techniques can help alleviate the world’s food problem by reducing waste in resources, increasing yield, and ensuring sustainable farming practices. In particular, studies have shown that precision irrigation techniques can increase yield by 45% all the while reducing water intake by 37% [1]. Such results extend to other precision agriculture techniques as well. While the efficacy of data driven agriculture has been demonstrated, these techniques are sparsely adopted in today’s farming practices. This is primarily due to the expensive cost of data collection and the challenging environment of typical farming locations.

Figure 1. FarmBeats System. The FarmBeats system using sensor, drones, TV Whites Spaces, Edge, and cloud solutions to provide insights to farmers and enable data-driven agricultural techniques.

To enable data driven agriculture, a seamless data collection system is needed. In other words, this would be an end-to-end IoT system where sensors collect data, such as soil moisture or temperature, and stream to the cloud to perform data analytics. In turn, providing insights for farmers to enable precision agriculture techniques. For example, soil moisture data can be used to determine where water should be applied and where it is not needed. However, enabling an IoT system for agriculture faces several significant challenges, those being power, connectivity, and overall system cost.

Power: Since most farms are located in rural areas, having access to a power source on the farm can be challenging. Relying on renewable energy, such as solar is a possible solution, but can be unreliable depending on weather conditions. For instance, a strong trend of cloudy or rainy weather could cause solar energy to be depleted and in turn disable the IoT system.

Connectivity: Another location specific challenge is Internet connectivity. On average, there is little to no Internet connectivity in farming locations. Perhaps the farmer has Internet access at their home or office, but it cannot reach the farm that is several miles away. Aside, from simply having Internet connectivity, the communication between sensor nodes and the IoT base station must be long range in order to have a practical implementation for large-scale deployments. Lastly, the available Internet connection is often prone to weather related outages and can render the system offline from days to weeks.

Cost: Most data driven agriculture solutions that are available today are too expensive for the average farmer. A single sensor can cost 350 USD with a recurring fee. Considering that there are over 500 million small-holder farmers across the world, this is more often than not unattainable. Thus, it is crucial that the overall cost of the IoT deployment is affordable for all farmers without compromising quality of data.

FarmBeats is an end-to-end AI and IoT system for agriculture that helps address these challenges and enable precision agriculture techniques. In particular, FarmBeats gathers data from sensors, cameras, and drones, to produce actionable insights for farmers. Moreover, it can extend Internet coverage to the farm and is resilient towards weather related outages or power variability. Thus far the FarmBeats system has been deployed worldwide, including locations in the United States, India, Africa, and China to name a few. FarmBeats is being used for a variety of agriculture applications such as storage monitoring, generating precision sensor maps of crop fields, animal monitoring, and much more. To get a better understanding of what all FarmBeats encompasses we explain the main components of the system and provide a case study of a FarmBeats deployment on Nelson farm in Spokane, WA.

Overview

The FarmBeats system consists of sensors to gather soil, plant, and weather data, aerial imagery using drones, TV White Spaces for connectivity, and IoT Edge Device, and Cloud components.

Sensors: Soil, plant, and weather sensors are connected to data acquisition hardware which is called sensor node. Sensor node supports both industrial- grade analog and digital sensors. These sensors are robust against weather. Besides, the sensor node supports sensors having cable length up to 500m which facilitates longer coverage by a single sensor node. Sensors nodes are solar powered, and communicate either via LoRa or a TVWS Radio to an IoT Gateway, which then transmits the data to the cloud.

Drones: The FarmBeats system supports drones to collect aerial imagery. This imagery is used for observation, and also for combining ground sensor data with aerial imagery to create continuous heatmaps [3].

TV White Spaces: FarmBeats leverages a technology called TV White Spaces (TVWS) [2] to extend Internet coverage across several miles of farmland. TVWS is unused TV spectrum which can be leveraged to enable Wi-Fi like connectivity over long distances. Using TVWS is particularly compelling for agriculture scenarios because there are many unused TV channels in rural areas, in comparison to metropolitan areas where TV towers are typically located. Moreover, because TV signals operate in the lower frequency bands (UHF and VHF) they can travel through dense crop canopy at long distances.

Edge: Instead of transmitting all the data to the cloud, which could be quite prohibitive, we send it to an Azure IoT Edge device. This is a PC form factor device that does processing of the data, such as stitching an orthomosaic using drone imagery, on the Edge, instead of transmitting gigabytes of data to the cloud.

Cloud: The FarmBeats cloud components aggregate data from sensors, drones, and other cloud sources, such as weather stations or satellites. This data is displayed to the farmer, and further insights are generated using AI and machine learning techniques.

Figure 2. FarmBeats Deployment. A TV whitespace base station and FarmBeats sensor box deployed on Nelson Farm.

Deployment Site

In Spokane, WA sits Nelson farm which spans approximately 7900 acres across 45 miles. This is one of the many farming locations using the FarmBeats system. At Nelson farm the growing season runs between the months of March and July, where production is focused around dry-land wheat, lentil, peas, and garbanzo beans. The day-to-day tasks the farmer performs ranges from spraying, fertilizing, seeding to marketing crops and researching crop varieties. Given this, the farmer is particularly interested in using the FarmBeats system to enable precision spraying, fertilizing, and seeding as well as informing his marketing decisions and researching crop varieties that would perform best in his microclimate.

In this deployment, the farmer has Internet connectivity at his home but it does not come close to providing coverage for the vast acreage of his farm. While connectivity is an issue, there is power available at a few location across the farmland. We utilize this and address the connectivity constraints by deploying a TVWS network. A TVWS base station connects to the Internet at the farmers’ home and extends the coverage to TVWS clients that are located in areas with power available. These TVWS links between the base station and client can range 10s of miles.

With the TVWS network enabled, FarmBeats sensor boxes are deployed across Nelson farm to collect data using sensors such as ambient temperature, atmospheric pressure, soil moisture and temperature, wind speed and direction to name a few. Every sensor box is powered by a single battery and backed by solar power, which allows battery maintenance to be quite minimal. Moreover, the FarmBeats sensor boxes are designed to be modular in the sense that sensor types of various kinds (analog or digital) can be used and a variety of communication schemes can be leveraged to transmit data. At Nelson farm the sensor boxes communicate using LoRa, a standard IoT protocol, and also TVWS depending on distance between the sensor box and the nearest TVWS client. For example, sensor boxes that are several miles away would rely on TVWS since LoRa has a communication range of up to 2.5 miles. Each sensor box, whether using TVWS or LoRa, transmits data that reaches the Edge device sitting in the farmers’ home.

Now we have enabled a seamless data collection system. However, in order to have an accurate representation of soil moisture, for example, across an entire field we would need to deploy a vast amount of sensors. This becomes costly and such a large deployment of sensors can get in the way of the farmer’s daily tasks. Instead, FarmBeats uses a sparse sensor deployment coupled with drone imagery to create precision heat maps of the farm.

At Nelson farm, the farmer performs weekly drone flights that capture images, which are then used to stitch panoramic overviews of the entire farm. Machine learning and vision algorithms are used to combine the orthomosaic with the sensor data and in turn, create precision heatmaps of soil moisture, temperature, and other data types. All of this processing is performed locally at the Edge. The reasoning behind this is twofold. First, if all of the data analytics are performed in the cloud and the system suffers from a power or connectivity fault, then no insights can be provided to the farmer to aid in enabling precision agriculture techniques. Second, shipping sensor data and images to the cloud requires a readily available high bandwidth connection which can be difficult to provide on most farms, even with decent bandwidth the size of the files can still take too long to upload to help the farmer make decisions in time. The FarmBeats system creates summaries of the data that get shipped to the cloud and in turn, decreases the uplink data from several gigabytes to kilobytes and is capable of providing these insights in offline scenarios.

Farmer Experience

Thus far the FarmBeats has been operating on Nelson farm for 11 months and has provided numerous insights to improve productivity. For instance, the panorama overviews of the farm and precision heatmaps help the farmer make intelligent decision when it comes to day-to-day tasks. For instance, they enable precision pesticide or herbicide application, where chemicals are only applied where they are needed, which saves cost and improves overall plant health. Specifically, the application of the most expensive chemicals was reduced by 90% and cost savings increased by 15%. FarmBeats also provides micro-climate predictions of temperature and wind speed and direction, which has immediate impact on the farm. The micro-climate across Nelson farm can have temperature differences of over 10 degrees. When the temperature drops below freezing within 24 hours of some field operations, it can cause a decrease in crop yield by up to 50% and significant revenue loss. Being able to predict such freezing temperatures accurately was crucial in May 2019. The farmer was able to change his plan of action in time when the FarmBeats system predicted below freezing temperatures and his weather apps did not. Moreover, wind microclimate predictions have proven beneficial in running operations, where equipment can be sent to locations that have the least amount of wind and keep conducting field operations rather than completely halting all tasks.

Conclusion

FarmBeats uses innovative techniques that overcome the many challenges of resource constrained environments such as lack of power and reliable Internet connectivity to enable data-driven agriculture techniques. We provided a case study of a FarmBeats deployment on Nelson farm and demonstrated how FarmBeats can improve productivity and help farmers develop more sustainable farming practices.


Zerina Kapetanovic presented this research as part of a minisymposium at the 2019 SIAM Conference on Computational Science and Engineering, which took place earlier this year in Spokane, Wash.


References

[1] Almarshadi, M. H., & Ismail, S. M. (2011). Effects of precision irrigation on productivity and water use efficiency of alfalfa under different irrigation methods in arid climates. J. Appl. Sci. Res., 7(3).

[2] Bahl, P., Chandra, R., Moscibroda, T., Murty, R., & Welsh, M. (2009). White space networking with wi-fi like connectivity. SIGCOMM Comput. Commun. Rev., 39(4), 27-38.

[3] Vasisht, D., Kapetanovic, Z., Won, J., Jin, X., Chandra, R.,  Sinha, S.,  Kapoor, A., Sudarshan, M., & Stratman, S. (2017). Farmbeats: An iot platform for data-driven agriculture. In 14th USENIX Symposium on Networked Systems Design and Implementation (NSDI 17), 515–529, Boston, MA. USENIX Association.

Zerina Kapetanovic is a Ph.D. student in the Electrical and Computer Engineering Department at the University of Washington, focusing on low-power wireless communication and sensing. In 2015, she joined the Microsoft’s FarmBeats team where she works on hardware research and IoT system deployments. Ranveer Chandra is a chief scientist at Microsoft Azure Global where he leads the FarmBeats program. His prior research as principal researcher at Microsoft Research has been shipped as a part of multiple Microsoft products. Tusher Chakraborty is a research fellow at Microsoft Research. His research interests include sensor-enabled embedded systems, IoT, and WSNs. He is part of the Microsoft FarmBeats team where he focuses on TVWS for IoT systems. Andrew Nelson is a software engineer and 5th-generation farmer in Eastern Washington.  He manages his farm and runs a private software consulting company.

blog comments powered by Disqus