FarmBeats: IoT for Agriculture

FarmBeats: IoT for Agriculture
Ranveer Chandra
Microsoft Research
[email protected]
ABSTRACT
Fueled by increasing population and upward social mobility, the demand for food is expected
to double by 2050 [9]. Coupled with receding water levels and shrinking arable land, this poses a major
challenge for agricultural production. A recent study by the International Food Policy Research Institute
(IFPRI) showed that data driven agricultural techniques [3] can increase global crop yields by as much
as 67%, and reduce food prices by nearly half by 2050. This can help reduce food scarcity by up to 36%
[4,8].
The most promising data-driven approach is Precision Agriculture [6]. It refers to the technique
of measuring intra- and inter-field variability, and using this information to adapt farming practices. For
example, a farmer can use a variable rate seeder to plant seeds closer in the more fertile parts of the
farm, or use a precision irrigation system depending on soil moisture in different regions. Over time,
data can indicate useful practices in the farms and make suggestions based on previous crop cycles.
Precision agriculture techniques usually require a map with information about each location in
the farm, for example, the soil temperature, soil moisture, nutrient levels, etc. We refer to this map as
the precision map of the farm [1]. State of the art solutions for precision agriculture either use in-ground
sensors or camera-equipped unmanned aerial vehicles (UAVs)1 to build this map. However, both these
approaches have limitations. In-field sensors use either cellular or satellite modems to connect to the
Internet, and hence cost several hundred dollars per sensor. Unless these sensors are deployed densely
within a farm, the estimated precision map can be very inaccurate. On the other hand, camera-equipped
UAVs can measure vegetation indices (NDVI), but cannot estimate the information that ground sensors
can measure, like soil moisture and nutrient levels.
In this paper, we present FarmBeats, a low-cost end-to-end IoT system for agriculture that uses
sparse ground sensors and camera-equipped UAVs to enable precision agriculture applications. The
FarmBeats system uses machine learning algorithms to train a model for predicting the soil features
from the UAV’s video and ground sensor readings. We then use the trained model to update the spatiotemporal precision map as subsequent sensor readings are received. Therefore, the farmer has a near
real-time view of the farm even though the last drone flight was done hours or days ago. To the best of
our knowledge, FarmBeats is the first system that can combine the temporal data from sensors, with the
spatial data from drones to construct an instantaneous spatio-temporal precision map of the farm.
1
For ease of exposition, we use drone and UAV interchangeably for the rest of the paper. We discuss quadcopter
based drones, since they are much more affordable, and offer the ability to hover over a farm. However, most of
our techniques are applicable to fixed wing UAVs as well.
However, enabling this functionality poses several systems challenges. First, to overcome poor
connectivity within the farm, we need a low-cost alternative to cellular and satellite networks to collect
sensor data from the field. Second, to maximize the area covered by UAVs in a single flight, we need
to improve their battery life. In particular, to reduce the cost of FarmBeats we use an inexpensive UAV
(an off-the-shelf quadrotor), which has a flight time of less than 20 minutes. Finally, we need to carefully
partition the computations that are performed on the IoT gateway and the cloud. The drone videos are
several gigabytes in size, and the farmer’s house usually has a slow and unreliable Internet connection
[5,7]. So, it is infeasible to transfer the entire video to the cloud. At the same time, we would like the
data to be available in the cloud for long-term, and cross-farm analytics.
The FarmBeats system addresses the above challenges, and is the first end-to-end IoT system
we are aware of that interfaces with both in-ground sensors and UAVs, and that too in weakly connected
settings.
First, to drive down the cost of the sensors, we use the unlicensed TV White Spaces (TVWS)
to collect data [2]. Our key insight is that in farms and rural areas, there is a lot of unused TV spectrum.
Since transmissions in these low frequencies propagate very far, and through crops and canopies, they
are ideal for data collection in farms. Just like Wi-Fi connects devices in our homes, our vision is that
the long range TVWS can be used to connect all sensors and farm equipment, to a gateway device or a
repeater a few miles away, through to the Internet.
Second, we present two new techniques to improve the battery life of drones. First, we use a
new path planning algorithm that minimizes the time taken to cover a region. Second, our algorithm
adapts the yaw of the drone based on wind, such that the drone uses the wind to help accelerate or
decelerate, instead of fighting the wind.
Finally, to avoid shipping gigabytes of drone video and sensor data to the cloud over a weak
Internet connection in the farm, we run video processing locally on a low-cost PC, or a similar machine,
that we call the FarmBeats gateway. The FarmBeats gateway is located at the farmer’s home/office and
collects information from the ground sensors in the farm over a local TVWS network. FarmBeats
gateway compresses the drone video and sensor data into summarized precision maps of the farm. It
first converts the drone video into a detailed orthomosaic of the farm, which is an order of magnitude
smaller than the video. It then interpolates the sensor readings to form a spatio-temporal heatmap of the
farm, and only sends these heatmaps to the cloud. This further reduces the size of data to be shipped to
the cloud by an order of magnitude. By doing this, FarmBeats gateway can enable long-term as well as
cross-farm analytics, while still continuing to provide near real-time analytics locally to the farmer
through a web-based interface.
We have built FarmBeats and deployed it on a large scale (100 acre) and a small scale farm (5 acre)
in United States. Our deployments primarily focus on two applications: precision irrigation (varying
irrigation across the farm based on current moisture levels) and precision pH (maintaining appropriate
pH values for crops). To summarize our deployment insights:

Our deployments ran over a period of 3 months in each of the farms and collected over 1 million
sensor measurements. The only periods of unavailability of our sensors were caused by multiple
days of no sunlight, causing our solar power based sensors to run out of battery.




FarmBeats gateway achieved a median compression of 1000 times from an aerial drone video
to the sensor precision maps of the field.
FarmBeats gateway produces high resolution overviews of the field for a farmer to inspect
details of different areas in the farm.
Compared to existing solutions, FarmBeats’s sensors are 5 times less expensive.
FarmBeats’s efficient flight planning algorithm increased the coverage of a drone by
approximately 30% for the same battery capacity.
Note that, in its current state, FarmBeats is an efficient, persistent and highly available data
collection system. While farmers are already finding our deployments useful, we believe that we have
only scratched the surface of what this data can deliver. With more deployments and advanced data
analytics techniques, FarmBeats can enable both small and large scale farmers to practice data driven
agriculture sustainably. Finally, we note that although most of our system description is focused on
agriculture, the design of FarmBeats can easily be extended to other weakly connected scenarios, such
as mining, forestry, and others.
REFERENCES
[1] Farms.com. Precision maps: http://www.farms.com/precisionagriculture/precision-maps/.
[2] Federal Communications Commission. https://www.fcc.gov/general/white-space-databaseadministration.
[3] H. C. J. Godfray, J. R. Beddington, I. R. Crute, L. Haddad, D. Lawrence, J. F. Muir, J. Pretty, S.
Robinson, S. M. Thomas, and C. Toulmin. Food security: The challenge of feeding 9 billion people.
Science, 327(5967):812–818, 2010.
[4] International Food Policy Research Institute. Agricultural technologies could increase global crop
yields as much as 67 percent and cut food prices nearly in half by 2050, 2014.
[5] Kansas State University. Weak internet connectivity in rural areas hindering agricultural production,
2016.
[6] N. D. Mueller, J. S. Gerber, M. Johnston, D. K. Ray, and J. A. F. Navin Ramankutty. Closing yield
gaps through nutrient and water management. Nature, 490(11420):254–257, 2012.
[7] M. Pols. To run their businesses, farmers need reliable internet connections.
[8] M. W. Rosegrant, J.Koo, N. Cenacchi, C.Ringler, R. D. Robertson, M. Fisher, C. M. Cox, K. Garrett,
N. D. Perez, and P. Sabbagh. Food security in a world of natural resource scarcity: The role of
agricultural technologies. International Food Policy Research Institute (IFPRI), 2014.
[9] United Nations General Assembly. Food production must double by 2050 to meet demand from worlds
growing population, innovative strategies needed to combat hunger, experts tell second committee,
2009.
ABOUT THE AUTHOR
Ranveer Chandra is a Principal Researcher at Microsoft Research where he is leading an
Incubation on IoT Applications. His research has shipped as part of multiple Microsoft products,
including VirtualWiFi in Windows 7 onwards, low power Wi-Fi in Windows 8, Energy Profiler in
Visual Studio, and the Wireless Controller Protocol in XBOX One. He is active in the networking and
systems research community, and has served as the Program Committee Chair of IEEE DySPAN 2012,
and ACM MobiCom 2013.
Ranveer is also leading the battery research project, and co-leading the white space networking
project at Microsoft Research. He was invited to the FCC to present his work on TV white spaces, and
spectrum regulators from India, China, Brazil, Singapore and US (including the FCC chairman) have
visited the Microsoft campus to see his deployment of the world’s first urban white space network. As
part of his doctoral dissertation, Ranveer developed VirtualWiFi. The software has been downloaded
more than 750,000 times and is among the top 5 downloaded software released by Microsoft Research.
It is shipping as a feature in Windows since 2009.
Ranveer has published more than 75 papers, and filed over 100 patents, more than 80 of which
have been granted by the USPTO. His research has been cited by the popular press, such as CNET, MIT
Technology Review, Scientific American, New York Times, WSJ, among others. He has won several
awards, including best paper awards at ACM CoNext 2008, ACM SIGCOMM 2009, IEEE RTSS 2014,
USENIX ATC 2015, and Runtime Verification 2016 (RV’16), the Microsoft Research Graduate
Fellowship, the Microsoft Gold Star Award, the MIT Technology Review’s Top Innovators
Under 35, TR35 (2010) and Fellow in Communications, World Technology Network (2012). Ranveer
has an undergraduate degree from IIT Kharagpur, India and a PhD from Cornell University.