092

Internet of Things Based Weather-Soil Sensor

Station for Precision Agriculture
Ahmed A. El-magrous1, Jason D. Sternhagen1, Gary Hatfield2, Qiquan Qiao1,*
Ahmed.elmagrous@sdstate.edu, Jason.sternhagen@sdstate.edu, Gary.hatfield@sdstate.edu, Quiquan.qiao@sdstate.edu
1
Department of Electrical Engineering and Computer Science
2
Department of Mathematics & Statistics
Jerome J. Lohr College of Engineering,
South Dakota State University, Brookings, SD 57007

Abstract— This paper presents the design, development, and system is effective, practical, and would enable farmers to
testing of a customizable and cost-effective Weather-Soil Sensor make better informed decisions in their operations.
Station (W-SSS) for use in Precision Agriculture based on high
accuracy sensors, wireless communication, cloud data storage, and II. W-SSS HIGH LEVEL OVERVIEW
computation technology. The W-SSSs operated from July 25, 2018,
to September 15, 2018, using an off-grid power system, Arduino The W-SSS system consists of the hardware needed to
microcontroller, and Wi-Fi connection to the cloud. Sensor data
collect, store, and transmit the data as well as the data
quality was evaluated using several statistical techniques. The data
management to validate, format, and present to the user. A
obtained from the weather-soil stations illustrate the differences in
unique feature of the W-SSS is that it can be configured as a
weather and soil conditions both relative to the local weather station
singular, Super-Node system or as a group of Nodes
as well as those within a field. Knowledge of these differences would
allow a farmer to vary planting densities relative to soil conditions,
communicating with the Super-Node. A block diagram of a
irrigation control, as well as pest/herbicide management.
W-SSS Super-Node and Sensor-Node system is shown in Figs.
1 and 2, respectively.
Keywords --- Super-Node, Sensor-Node, Internet of Things

I. INTRODUCTION

Precision Agriculture (PA) is a farm management concept

that utilizes data and data analysis to support the farmer’s
decision-making process [1-3]. Monitoring crops for
environmental conditions and disease detection typically
requires an on-site visit by experts who go to the field, measure Fig. 1. W-SSS Super-Node block diagram.
soil temperature and moisture, and take soil and crop samples
for laboratory analysis [4]. However, manual collection of the The sensor complement of each W-SSS can be individually
samples results in data variations due to incorrect and tailored to meet the specific PA requirements. For example,
imprecise measurements [5-7]. Farmers in the U.S.A. and rainfall, wind speed, and solar insolation may be consistent
around the world are already using smart technologies to vary across a field necessitating only one sensor for each parameter
seed density and measure yields across a field but lack an located on one Super-Node. However, soil conditions and
topography can vary dramatically thus requiring one Super-
inexpensive real-time monitoring tool for weather conditions,
Node and several Nodes with additional soil moisture, soil pH,
soil properties, and crop health [8, 9].
and soil temperature sensors.
Internet of Things (IoT) is a network of smart internet
connected objects and devices which offers a simple and secure
method to remotely access sensor data and wirelessly control
the physical world [10]. IoT-based technology is an enabling
technology for precision agriculture which can connect fields,
growers, and experts in real time [11]. Real time field
monitoring can help farmers make the right decisions based on
site-specific information collection.
This paper presents the design, development, and testing of
a customizable and cost-effective Weather-Soil Sensor Station
(W-SSS) to collect real time site-specific maps of soil
properties, climate conditions, and crop information for the Fig. 2. W-SSS Sensor-Node block diagram.
2018 soybean growing season. Preliminary tests show this

978-1-7281-0927-5/19/$31.00 ©2019 IEEE

 093
III. HARDWARE DEVELOPMENT AND DESIGN
The hardware component of the W-SSS consists of three
parts: station sensors, microcontroller, and power
management.
A. Station Sensors
In PA, there are many promising applications based on
weather conditions, soil properties, and IoT technology that
farmers can utilize to enhance crop yield. For instance, the
optimum time to irrigate, fertilize, control weeds, and detect
diseases. Each application requires the monitoring of unique
parameters; therefore, the W-SSS system is customizable to
meet various PA applications.
Nine site-specific weather-soil parameters were selected for
the prototype W-SSS Super-Node. These parameters,
commonly used in PA applications, included air temperature,
humidity, precipitation, wind speed and direction, solar
radiation, soil temperature and moisture at different depths (4”
and 11”), and leaf wetness. The sensors and their parameters
are shown in Table 1. A sensor for each of these parameters
was identified with preference given to high relative accuracy,
common availability, and low cost.
TABLE I
SENSOR PARAMETERS AND ACCURACY
B. Microcontroller
The Arduino Mega 2560, was chosen to interface with the
sensors, Wi-Fi, real time clock, and SD card as it is widely
available and the software easy to use. It has 54 digital I/O, 16
analog inputs, SPI and I2C communication, and regulated 5V
and 3.3V supplies [12]. The individual weather and soil
sensors were interfaced to the Arduino using a Rugged-
Circuits Aussie Shield breakout board with push connectors
[13].
C. Power Management
The W-SSS Node and Super-Node systems need to operate
independently and unmanned in remote areas for the entire
growing season. This necessitated an off-grid power system
comprised of a 10W photovoltaic panel, charge controller, and
a 55 Ah lead acid battery (Fig. 3).
The design of the power system began by determining the
total daily load of the W-SSS. The power consumption of the
sensors and Arduino Mega were based on a 100% duty cycle
and the duty cycle of the Wi-Fi module was assumed to be
50%. The total estimated power consumption (1.4W) was
multiplied by 24 hours to obtain the daily energy requirement
of 33.6 Wh.
The energy input of a photovoltaic system is the solar
insolation for the operating area. PV-Watts is a tool developed
by NREL to estimate the energy production of a grid-
connected photovoltaic system [14]. The average daily solar
insolation estimated by PV-Watts for the location of the
weather stations was 5.5 kW/m2/day. Since a sunny day is
likely but not guaranteed during the summer and the actual
output of the photovoltaic system is reduced by losses in the
charge controller, a recharge factor of 1.3 and a charge
controller loss factor of 1.1 was used [15]. The resulting
photovoltaic module size was calculated as 8.5W. Given that
8.5W is not a common size, and, since an increase in
photovoltaic module output increases reliability, a 10W panel
was purchased.
A 12 V sealed lead acid battery was chosen due to cost and
local availability. The battery was sized for 10 days of
autonomy; days it can operate the W-SSS system without the
sun to recharge the battery. For a 12V battery constrained to a
maximum of 50% depth of discharge, a capacity of 55 Ah was
required.
Fig. 3. Power system components
IV. SYSTEM SOFTWARE AND DATA
MANAGEMENT
The Arduino Integrated Development Environment (IDE)
was used to write the W-SSS software due to ease of access
and availability of open-source libraries [12]. Each sensor
required the development of a “setup” and a “loop” portion of
code. Fig. 4 shows a generic flow chart for the operation of
each sensor.
A. Communication and Data Transmission
The W-SSS Sensor Nodes transmit their data to the Super-
Node where the data packets are combined. The aggregated
 094
data is communicated to the cloud server using an Adafruit
model TWNIC1500 Wi-Fi module. The TWNIC1500 is an
IEEE® 802.11 b/g/n network controller and considered ideal
for IoT applications [16]. If local Wi-Fi is available, this
communication is direct, otherwise, a Wi-Fi hot spot is
required (Fig. 5).
Fig. 4. Software flowchart.
Fig. 5. W-SSS Sensor-Node and Super-Node data transmission path.
B. Data Security and Privacy Management
In a wireless sensor network, data packets are travelling
among sensor nodes to reach the gateway or the Super-Node;
then to the server database via the internet. During the packet’s
journey from the sensor, to the SD card, and to the cloud, it is
susceptible to theft or distortion from illegal access. Since this
information is valuable and growers are concerned about
privacy, robust and durable protection procedures were utilized
to cover the network entities communication to provide data
privacy and data integrity [17-19].
1) Data Stored Locally in SD Card.
Due to concerns about data loss during the transmission
process, an SD card was used in each W-SSS to store the
aggregated data. An RSA (Rivest, Shamir, and Aleman)
cryptography algorithm was used to encrypt the data saved into
the SD card to prevent unauthorized access. The advantages
of using an RSA cryptography algorithm include 1) RSA is an
asymmetric key with a simple algorithm; 2) low power
consumption; 3) RSA is suitable for any network type [20, 21].
Fig. 6. Symmetric encryption conceptual diagram.
2) Data Transmission from Super-Node to the Cloud.
The TWNIC1500 Wi-Fi module includes enterprise
security protocols WPA/WPA2 which ensure that data packets
are safely transferred from the W-SSS to the cloud without
corruption [16].
C. Mobile User Interface for Data Access
An application enabling the grower to use their mobile
phone to access field data was developed. This app offers a
variety of options to view current weather-soil status, hourly
data averaging, and historical data.
Fig. 7. Mobile application for accessing real-time field information.
V. W-SSS SUPER-NODE DEPLOYMENT
A. Selection of W-SSSs Deployment Locations
Two W-SSSs Super-Node prototypes were developed and
deployed at a soybean field in Red Rock, SD. W-SSS1 and W-
SSS2 were located at 43° 38.747 N 96° 29.478 W at an
elevation of 1555.1 feet and 43° 38.747 N 96° 29.478 W at an
elevation of 1555.1 feet, respectively. A soybean field was
 095
chosen because it is a common crop in the area and prone to
weather related diseases such as white mold [22]. During an
interview conducted on July 11, 2018, Assistant Professor
Emmanuel Byamukama, extension plant pathologist, South
Dakota state University, stated that a diverse weather
condition, soil properties, and land topographies are a real test
W-SSSs performance. Therefore, the specific sites within this
field were chosen to assess potential weather variations due to
minor differences in elevation (Fig. 8).
Fig. 8. Low elevation deployment station (a) and high elevation deployment
station (b).
B. System Operation and Maintenance
The W-SSSs operated from July 25, 2018, to September
15, 2018. During this time, each W-SSS recorded ten weather-
soil parameters every 30 seconds. Data was concurrently
stored internally in an SD card and sent to a server database in
the cloud. The W-SSSs were visited weekly to assess the status
of the battery, power management system, the physical
condition of the system, and to adjust the height and position
of the leaf wetness sensor to match the current crop canopy
(Fig. 9).
Fig. 9. Leaf wetness sensor mounted on a movable arm to enable close
monitoring of crop canopy.
VI. RESULTS AND ANALYSIS
A. Data Collection and Transmission
Two large data sets, each with more than 60,000 data
points, were collected from W-SSS1 and W-SSS2. The data
saved internally on the SD card was compared to that stored in
the cloud to determine if all data stored was also transmitted
and received. Fig. 10 shows the data loss rate of 0.013% and
0.009% for the W-SSS1 and WSSS2, respectively. It was
determined that the source of error was due to connectivity
between the Wi-Fi shield and the internet hotspot. If no
internet connection was available after data collection, the data
was saved to the SD card but not transmitted to the cloud.
A future improvement to the data transmission process is
the addition of a binary Transmission Confirmation Tag (TCT)
to the end of each data packet. If an internet connection is
available at the time of transmission, the TCT value is a 1.
However, if a connection is unavailable during the intended
transmission, the value is a 0. If at a later point in time, a
connection would become available, those data packets which
were not transmitted would be re-transmitted.
Fig. 10. Data transmission rate from W-SSS to cloud.
B. Data Validation
Since precision agriculture relies on quality information
collected from a field to make informed decisions, several
statistical techniques were used to assess the validity of the
data provided by the W-SSSs.
Data collected from a field may include errors such as
noise, drift, and outliers. The typical reasons for these errors
include environment effects, power problems, hardware
damage, software problems, network problems, and security
attacks [23]. R language is a statistical programming tool that
was used to search for four common types of potential errors:
out of range, stuck-at, outliers, and missing values [23, 24].
1) Out-of-Range Fault
Every sensor has a maximum and minimum possible value
that can be measured [23]. For example, the humidity sensor
cannot have a valid output above 100% RH. Table 2 shows
minimum and maximum sensed value from each applied
parameter in the W-SSS. Data values out of the sensing range
were considered an out-of-range error. The data was analyzed
and verified that no sensor recordings for any parameters were
outside of the manufacturer’s specifications.
TABLE 2
MINIMUM AND MAXIMUM VALUES FOR THE APPLIED SENSORS’ PARAMETER
 096
2) Stuck-at Fault
Data that becomes “stuck” at a constant value either within
or outside the expected range is generally considered a Stuck-
at fault or constant fault [23]. For this work, a series of data
that exhibits zero or nearly zero variation for a period of time
will be considered as a stuck-at fault [25]. The calculated
variance for temperature in randomly selected periods of an
hour, day, and week shows values of 0.79, 192.24, and 206.91,
respectively. Since the variances were greater than zero, this
is strong evidence that the sensed data had no stuck-at fault
errors. However, some parameters such as relative humidity,
wind speed, sun radiation, soil moisture, and leaf wetness have
some constant values by nature at a specific period or at
specific weather and soil conditions. For example, the solar
radiation at night will be zero; yet this would not be considered
a fault. Accordingly, this assessment technique was not
appropriate for these parameters.
3) Outliers
Data which are within the sensor range but lying an
abnormal distance from adjacent data points are considered
outliers [26]. The types of outliers investigated include
Additive, Innovation, Level Shift, and Temporary Change [27,
28]. The statistical functions in R language were used to search
the data set for TC, AO, LS, and IO outliers.
Fig. 10 shows the three TC potential outliers observed in
the temperature data. These potential outliers were a
significant increase or decrease in a temperature value in a
series compared to the previous value. For example, on August
22, 2018, at 11:00 PM, there was a decrease in temperature
from 73.14 °F at index 524 to 56.16 °F at data index 525. Two
potential outliers with a dramatic increase in temperature were
observed at exactly 9:00 AM on August 25 and 29. No other
types of outliers were observed.
Fig. 10. Three potential outliers observed in the W-SSS2 temperature data.
4) Missing Values Fault
Missing data values can occur in sensor networks due to
problems such as sensor quality, power instability, sensor
failure, and sensor interference [29]. No missing values were
observed.
C. Site-Specific vs. Local data
General weather information is available to growers from
several sources such as government stations, local TV and
radio. However, the specific weather and soil characteristics
vary from region to region. This lack of field-specific
information makes it difficult to make precise decisions related
to crop management. The nearest local weather station, KFSD,
is located in Sioux Falls, SD and approximately 10 miles from
the field where W-SSS super nodes were deployed. Fig. 11
compares temperature, relative humidity, and wind speed
information recorded from W-SSS1, W-SSS2, and KFSD.
While the data trends are very similar, there are clear
differences in weather conditions. For example, the daily
temperature excursions are wider in the soybean field than in
the city. This is likely due to the thermal mass of the concrete
and asphalt in the city of Sioux Falls moderating the
temperature (heat island effect).
Fig. 11. Comparison of W-SSS1, W-SSS2, and KFSD temperature, relative
humidity, and wind speed.
D. Site-Specific Data Variations
Site specific information especially important for PA is
lacking from the local weather station. This includes field
parameters such as soil moisture, soil temperature, and leaf
wetness. While some stations do provide some of this
information, due to varying soil conditions, this information is
not precise. Furthermore, the conditions within a specific field
can also vary tremendously [30]. While the geographic
distance separating W-SSS1 and W-SSS2 is only about 0.32
miles, their data represent very different field conditions (Fig.
12). The W-SSS1 is located at a higher elevation in the field
which results in lower leaf wetness, lower soil moisture, and
higher soil temperature. Accordingly, the W-SSS2 location
likely receives more rainfall runoff and lower evaporative
losses which results in higher leaf wetness and higher soil
moisture. Knowledge of these differences would allow a
farmer to vary planting densities relative to soil conditions,
irrigation control, as well as pest/ herbicide management.
Fig. 12. Comparison of W-SSS1 W-SSS2 soil moisture, soil temperature, and
leaf wetness.
 097
VII. CONCLUSIONS
A weather-soil station was developed and successfully
deployed for the 2018 growing season. The data obtained from
the weather-soil stations illustrates the differences in weather
and soil conditions both relative to the local weather station as
well as those within a particular field. Knowledge of these
differences, readily available on a cell phone application, are
expected to facilitate and improve PA. Statistical and machine
learning techniques could be integrated into the app for various
PA applications that can make significant improvements in
decision making strategies.
ACKNOWLEDGEMENTS
Thanks go to the following for their support: Dr.
Emmanuel Byamukama, Assistant Professor & SDSU
Extension Plant Pathologist, Mr. Philip Schumacher, Science
and Operations Officer at the National Weather Service, Sioux
Falls, SD, and Mr. Anthony Bly for allowing the W-SSS to be
operated in his field.
REFERENCES
[1] B. Kay, "Rates of change of soil structure under different cropping
systems," in Advances in Soil Science 12: Springer, 1990, pp. 1-52.
[2] D. B. G. Williams, M. J. George, and L. Marjanovic, "Rapid detection
of atrazine and metolachlor in farm soils: gas chromatography–mass
spectrometry-based analysis using the bubble-in-drop single drop
microextraction enrichment method," Journal of agricultural and food
chemistry, vol. 62, no. 31, pp. 7676-7681, 2014.
[3] P. Schröder et al., "Intensify production, transform biomass to energy
and novel goods and protect soils in Europe—a vision how to mobilize
marginal lands," Science of The Total Environment, 2017.
[4] M. Glantz, "Guidelines for establishing audits of agricultural-
environmental hotspots," Food and Agricultural Organization of the
United Nations (FAO)2003.
[5] T. J. Lybbert and D. A. Sumner, "Agricultural technologies for climate
change in developing countries: Policy options for innovation and
technology diffusion," Food Policy, vol. 37, no. 1, pp. 114-123, 2012.
[6] Y. Kebede, K. Gunjal, and G. Coffin, "Adoption of new technologies in
Ethiopian agriculture: The case of Tegulet-Bulga district Shoa
province," Agricultural economics, vol. 4, no. 1, pp. 27-43, 1990.
[7] J. R. Okalebo, K. W. Gathua, and P. L. Woomer, "Laboratory methods
of soil and plant analysis: A working manual second edition," Sacred
Africa, Nairobi, p. 21, 2002.
[8] S. Swinton and J. Lowenberg-DeBoer, "Evaluating the profitability of
site-specific farming," Journal of production agriculture, vol. 11, no. 4,
pp. 439-446, 1998.
[9] J. M. Barcelo-Ordinas, J.-P. Chanet, K.-M. Hou, and J. García-Vidal,
"A survey of wireless sensor technologies applied to precision
agriculture," in Precision agriculture’13: Springer, 2013, pp. 801-808.
[10] H. Kopetz, "Internet of things," in Real-time systems: Springer, 2011,
pp. 307-323.
[11] D. Puri. (2016). IoT-based precision agriculture reduces the
guesswork in farming. Available:
https://www.networkworld.com/article/3151154/internet-of-things/iot-
based-precision-agriculture-reduces-the-guesswork-in-farming.html
[12] Arduino. Arduino MEGA 2560 & Genuino MEGA 2560. Available:
https://www.arduino.cc/en/Main/ArduinoBoardMega2560?setlang=en
[13] R. R. CIRCUITS. (2016). MEGA Aussie Shield Pro. Available:
https://www.rugged-circuits.com/news/
[14] N. R. E. Laboratory, "PVWatts Calculator," 2018.
[15] P. E. Network. Operation of Lead Acid Batteries. Available:
https://www.pveducation.org/pvcdrom/lead-acid-batteries/operation-
of-lead-acid-batteries
[16] Atmel, "ATWINC1500 Wi-Fi Network Controller - Software Design
Guide," PDF 2015, vol. 240.
[17] A. F. B. Federation. (2016). Privacy and Security Principles for Farm
Data. Available: https://www.fb.org/issues/technology/data-privacy/
[18] X. Du and H.-H. Chen, "Security in wireless sensor networks," IEEE
Wireless Communications, vol. 15, no. 4, 2008.
[19] B. Singh, P. Singh, and V. Dhaka, "Sensor data encryption protocol for
wireless network security," Global journal of computer science and
technology, 2012.
[20] R. Kayalvizhi, M. Vijayalakshmi, and V. Vaidehi, "Energy analysis of
RSA and ELGAMAL algorithms for wireless sensor networks," in
International Conference on Network Security and Applications, 2010,
pp. 172-180: Springer.
[21] R. L. Rivest, A. Shamir, and L. Adleman, "A method for obtaining
digital signatures and public-key cryptosystems," Communications of
the ACM, vol. 21, no. 2, pp. 120-126, 1978.
[22] R. Harikrishnan and L. d. J. P. D. Rio, "A logistic regression model for
predicting risk of white mold incidence on dry bean in North Dakota,"
vol. 92, no. 1, pp. 42-46, 2008.
[23] J. Ravichandran and A. I. Arulappan, "Data validation algorithm for
wireless sensor networks," International Journal of Distributed Sensor
Networks, vol. 9, no. 12, p. 634278, 2013.
[24] R. C. Team, "R: A Language and Environment for Statistical
Computing," ed: R Foundation for Statistical Computing, Vienna,
Austria, 2017.
[25] K. Ni et al., "Sensor network data fault types," vol. 5, no. 3, p. 25, 2009.
[26] K. Ni et al., "Sensor network data fault types," ACM Transactions on
Sensor Networks (TOSN), vol. 5, no. 3, p. 25, 2009.
[27] IBM®. Outlier types, IBM Knowledge Center. Available:
https://www.ibm.com/support/knowledgecenter/fr/SS3RA7_15.0.0/co
m.ibm.spss.modeler.help/ts_outliers_overview.htm
[28] M. Shcherbakov, A. Brebels, N. Shcherbakova, V. Kamaev, O. M.
Gerget, and D. Devyatykh, "Outlier detection and classification in
sensor data streams for proactive decision support systems," in Journal
of Physics: Conference Series, 2017, vol. 803, no. 1, p. 012143: IOP
Publishing.
[29] L. Gruenwald, H. Chok, and M. Aboukhamis, "Using data mining to
estimate missing sensor data," in Seventh IEEE International
Conference on Data Mining Workshops (ICDMW 2007), 2007, pp. 207-
212: IEEE.
[30] S. Malcolm, E. Marshall, M. Aillery, P. Heisey, M. Livingston, and K.
Day-Rubenstein, "Agricultural adaptation to a changing climate:
economic and environmental implications vary by US region," 2012.