Computers and Electronics in Agriculture 162 (2019) 134–142

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

journal homepage: www.elsevier.com/locate/compag

Original papers

Greenhouse environment modeling and simulation for microclimate control T

a b a,c a b
Dongdong Ma , Neal Carpenter , Hideki Maki , Tanzeel U. Rehman , Mitchell R. Tuinstra ,
⁎
Jian Jina,
a
Department of Agricultural and Biological Engineering, Purdue University, West Lafayette, IN 47907, USA
b
Department of Agronomy, Purdue University, West Lafayette, IN 47907, USA
c
Health and Crop Sciences Research Laboratory, Sumitomo Chemical Co. Ltd., Hyogo 665-8555, Japan

A R T I C LE I N FO A B S T R A C T

Keywords: Greenhouse plant science assays have been impacted by microclimates which causes significant level of noise to
Greenhouse microclimate control plant growth measurement data. Researchers and scientist have been randomizing pots locations, which helps to
Radiation temperature re-distribute the noise, but does not remove the noise. The impacts from microclimates can be eliminated by
Simulation shuffling plants, but there has been no study on the optimization of shuffling pattern, such as the frequency and
Pot position
moving distance for the pots. It is important to quantitatively study the microclimates in the greenhouse, so we
Plant phenotyping
can optimize the shuffling pattern accordingly. The aim of this study was to propose a computer modeling
approach for simulating microclimate in the greenhouse, and then use the simulation result to optimize pot
movement distance and frequency. A computational greenhouse model was developed using inputs from real
design, materials and location of a Purdue Lily greenhouse in West Lafayette, Indiana. Microclimate variables,
including ambient temperature and lighting radiation over 24 h and 7 days were predicted with the simulation
model. Thermometers and lighting sensors were also distributed in the greenhouse for the ground-truth mea-
surements over a seven-day period. Comparison of measured microclimate variables with predicted variables
obtained from the computational model demonstrated that the simulation model could precisely predict tem-
peratures and light radiation at any time, and at different positions in the greenhouse. Optimized pot movement
frequency and distance were then determined with the simulation result. The new shuffling pattern can remove
over 90% of the microclimate variance but could save more than 95% shuffling efforts compared with non-stop
movement.

1. Introduction et al., 1998).

Researchers have kept exploring methods to eliminate microclimate
At present greenhouse facilities are prevalent crop research tools impact in the greenhouse. The whitening of greenhouse method was
with the advantages of easily achieving optimal plant growth and re- introduced by Baille et al. (2001). Influence of whitening a greenhouse
latively more uniform environment compared to the field tests (Vásquez roof on microclimate and canopy behavior was studied during summer
et al., 2015; Zhang et al., 2019). However, due to the intrinsic green- in a greenhouse located in the coastal area of eastern Greece. Another
house features, the climate within the same greenhouse is still hetero- innovative work was conducted by Brien et al. (2013) by designing
geneous and can be considered as a microclimate (Alain, 1988). The greenhouse containing plants on conveyor systems to account for var-
non-uniform distributions of temperature, relative humidity, carbon iation of microclimate. For the relocation scheme of plants to be suc-
dioxide concentration and irradiation caused by microclimate have cessful, it is required that plants spent an equal amount of time in each
recently received attention by several researchers (Ahemd et al., 2016; microclimate, preferably during comparable growth stages. In addition
Alain, 1988; Baille et al., 2001; Castro et al., 1991; Körner et al., 2007) to redesigning greenhouses, the technique of optimal control of
and have been proven to introduce the nuisance impacts on the growth, greenhouse environment had also been studied.
production and quality of the crops. Especially, temperature and irra- To control greenhouse environment, typical modern greenhouses
diation can be considered as two major factors relating to microclimatic were equipped with ambient control to accomplish optimal growing
effects (Gonzalez-Real and Baille, 2000; Lindquist et al., 2005; Stone environment, greenhouse environment management was dependent on

⁎
Corresponding author at: Agricultural and Biological Engineering Department, Purdue University, 225 South University Street, West Lafayette, IN 47907, United
States.
E-mail address: [email protected] (J. Jin).

https://doi.org/10.1016/j.compag.2019.04.013
Received 5 November 2018; Received in revised form 3 April 2019; Accepted 9 April 2019
Available online 13 April 2019
0168-1699/ © 2019 Elsevier B.V. All rights reserved.
D. Ma, et al. Computers and Electronics in Agriculture 162 (2019) 134–142

the overall greenhouse temperature manipulation by measuring the air

temperature at a singular point using commercially available climate
monitoring devices. This solution was working well in some way until Imaging Tower
some researchers realized local climate should be measured and con-
trolled to ensure uniformity of plants (Ahonen et al., 2008; Körner
et al., 2007; Radojević et al., 2014). A single measurement could not be
a representative of the entire greenhouse and could not provide the
detailed information about the temperature gradient and distribution
prevailing in the greenhouse (Castro et al., 1991; Körner et al., 2007).
Moreover, monitored air temperature could not represent the crop
temperature as it is comparatively different than the greenhouse air
temperature due to the inherent nature of heat transfer (Körner et al.,
2007). Heat transfer within the environment dispersed slowly at vari-
able rates. This phenomenon effects crop temperature which was highly
variant because of different microclimatic components, such as varia-
tion in speed of heating or cooling and difference in intercepted ra-
diation by different locations within the same greenhouse (Taki et al.,
2016).
Fig. 1. Overall greenhouse geometry and plants layout. 3D coordinates re-
To further invest and control the local crop temperature, a simula-
present x as north, y as west, and z as zenith.
tion model developed by Körner et al. (2007) characterized the time-
depending crop microclimate from a greenhouse macroclimate. Based
on the temperature close to the leaf surfaces collected periodically by a greenhouse facility was covered with transparently clear glass along
PT100 thermometer that were located above the crop, the model pre- with an automated high throughput imaging and a belt conveyor-based
dicted greenhouse microclimate focusing on crop temperatures in ver- plant transferring system (Fig. 2). Despite the existing microclimatic
tical transect of the crop canopy. Another similar study on the tem- effects, installation of the large lighting box right across the North wall
perature distributions in greenhouse was performed by Teitel et al. further escalated these effects. The temperature of the greenhouse was
(2008). The air temperature distribution inside the greenhouse regulated using evaporative cooling system consisting of two ventila-
equipped with vertical roof openings was investigated in this study. The tion fans (installed on the South wall of greenhouse) and rectangular
numerical computational methods were useful to predict local tem- evaporative cooling pads (installed on the North wall) (Fig. 2). The
perature distribution for greenhouse climate control purpose. However, greenhouse was equipped with the 17 sodium lights, each having a
these mentioned models mainly focused upon temperature and does not power rating of 1000 Watts. As the girders underneath the glass roof of
include radiation data, of which was not also equally distributed within the greenhouse caused the shades on the plant canopies, therefore they
a greenhouse environment. At the same time, the application of simu- were also introduced in the simulation model.
lation model on solving real greenhouse microclimate problem was
rarely discussed. In this study, a greenhouse microclimate simulation 2.2. Model development
model was developed to quantitatively analyze the temperature and
light radiation change over 24/7 at any location of the greenhouse. The The explicit finite element method is particularly suitable for nu-
simulated greenhouse was an innovative greenhouse which was merical simulation of heat transfer problems (Korioth and Versluis,
equipped with an automated, high-throughput imaging system and a 1997). The radiative heat exchange between greenhouse interior sur-
conveyor belt system (Fig. 2). In this greenhouse, plants were placed on faces such as the convective heat exchange within indoor air, the heat
the conveyor belt. By running the conveyor continuously, all the plants exhaust by wet wall cooling system, and sun light radiation effect can
would be shuffled and receiving uniform environment treatment during be readily associated with corresponding elements. The change in the
growth, but with a high cost. With the microclimate simulation result, greenhouse temperature distribution and radiation absorption on the
we were able to achieve the same uniformity but with a much lower surface of the plant can be modeled by using heat transfer and radiation
cost by adjusting conveyor “run” and “stop” time intervals. energy.
In this paper, a simulation model was presented which precisely The heat transfer between different objects can be expressed by
calculates the microclimate effect in greenhouse. The model was vali- conduction, convection and radiation phenomena (Kimball, 1973).
dated with experimental data from regular radiation and temperature Assuming the simulation model has constant thermal properties, the
sensors. Furthermore, the model showed its potential to optimize con- heat conducted by the solid medium can be described by the classic
veyor system settings in the greenhouse with lowest cost while keep the Fourier’s heat conduction (Eq. (1)) equation (van Eck et al., 2016).
same efficiency in eliminating microclimate impacts. During the last
∂T
decades, microclimate study for greenhouse climate control is always ρCP + ∇·q = Q
∂t (1)
one of the hottest topics in agronomy field. Automated greenhouses
−3 −1 −1
with conveyor systems have been developed and applied in many plant where ρ is the density of the solid material (kg m ), Cp (W m K ) is
breeding researches (Ge et al., 2016; Gehan et al., 2017; Golzarian the specific heat capacity, ∂T is time-based temperature derivative (K/
∂t
et al., 2011). This model developed provides the basis for further in- s), Q (W m−3) is the amount of heat generated and q is the conduction
−2
vestigation of designing next generation of greenhouse applying uni- heat flux (W m ) which can be represented by Eq. (2).
form macroclimate and microclimate environment on the plants.
q = − k ∇T (2)
−1 −1 −1
2. Materials and methods where k (W m ·K ) is the thermal conductivity, ∇T (K m ) is three-
dimensional temperature distribution.
2.1. Greenhouse geometry and plants layout In addition to heat conduction in the solid, the heat transfer on the
surface is also important which is affected by heat convection and ra-
The simulation study was designed and carried out in Lily 13-4 diation (Baxevanou et al., 2010). As the site under study was equipped
greenhouse (latitude 40.4259°N, longitude 86.9081°W) at Purdue with evaporative cooling system, therefore the heat dissipated by all
University, West Lafayette, Indiana (Fig. 1). The N-S orientated air-exposed surfaces through the forced heat convection can be

135
D. Ma, et al. Computers and Electronics in Agriculture 162 (2019) 134–142

Fig. 2. Purdue Lily 13-4 Greenhouse with automated, high-throughput imaging system and a belt conveyor-based plant transportation system.

modeled using heat transfer coefficient (Eq. (3)). Table 1

Properties of materials used in the greenhouse microclimate simulation model.
Q = A ∗ h (T − Tambient) (3)
Model inputs Heat Capacity Thermal Surface
2 −2 −1
where A is surface area (m ), h is heat transfer coefficient (W m K ), J kg−1 K−1 conductivity emissivity
T is the object temperature (K), and Tambient is ambient air temperature W m−1 K−1
(K).
Glass of greenhouse 703 1.38 0.93
In the greenhouse system, radiative exchange between two objects Imager tower 1470 0.18 0.6
can be a major factor leading towards the non-uniform environment Plant 3000 5.48 0.85
regarding both temperature and radiation absorption of plants (van Eck Roof girders 900 201 0.3
et al., 2016). Temperature under direct sunlight is significantly higher (aluminum)
Ground (concrete) 880 1.8 0.94
than that of other places. On the other hand, because solar radiation is
the main resource of energy for the plants to carry out the photo-
synthesis, the nonuniform spread of solar radiation has always been
Table 2
main part of microclimate impact. The overall radiosity exchange (Eq. Overall of running and break time for conveyor movement optimization.
(4)) between two objects (van Eck et al., 2016) can be defined as
Item 1 2 3 4 5 6 7 8
Rij = Ai Fij (Ji − Jj ) (4)
Running Phase (s) 10 20 40 80 160 320 640 1280
where Rij is the power transmitted (W) from body i to body j, Ai is Pause Phase (s) 0 60 120 240 480 960 1920 3840
the surface area (m2) of body i, Ji is total radiative flux (W m−2) leaving
surface i, Jj is total radiative flux (W m−2) leaving surface j and Fij is the
view factor from body i to body j and can be expressed as (Eq. (5)) 2.3. Computation

radiation leaving Ai and hitting Aj The simulation model was programmed in the COMSOL 5.2b
Fij =
total radiation leacing Ai (5) (COMSOL Inc., Burlington, MA, USA), which is a cross-platform finite
element analysis, solver and Multiphysics simulation software. In
For a certain temperature difference between two surfaces the
COMSOL 5.2b, four radiation modules are included, and this study used
transmitted radiation from one surface to another can be defined as (Eq.
the “Heat Transfer with Surface-to-Surface Radiation” module. In the
(6))
application area, three equations (Eqs. (1), (2) and (7)) were applied to
Ai εi (σTi4 − Ji ) account for the heat transfer in greenhouse including heat transfer in
Rij =
1 − εi (6) solids, heat transfer in fluid, and surface to face radiation (van Eck
et al., 2016). Referring to Fig. 3, initial boundary for the greenhouse
where Ai is the thermal energy (W) leaving surface i, εi is the emissivity was set to measured ambient air temperature (Frei, 2016). Eq. (3) was
(m2) of surface i, and σ is the Stefan Boltzmann constant with a value of used to set the flux boundary condition for both inside and outside of
5.67 × 10−8 (W m−2 K−4). the greenhouse.
This means radiation flux between two objects can be defined by the The domain was meshed with Lagrange triangular quadratic ele-
Stefan-Boltzmann Law (Eq. (7)). ments. Triangular elements were selected to allow for local mesh re-
Rij = Ai εi σ (Ti4 − T j4 ) finement (Fig. 4). With integration of elements, COMSOL computed the
(7)
view factor with finite element method as backend. All the simulations
where Ti is the temperature (K) of surface i, and Tj is the temperature were carried out on a ThinkPad workstation P300 (Lenovo PC inter-
(K) of surface j. national, Morrisville, Morrisville, NC, USA) equipped with 16-gigabytes
As the problem is governed by the heat transfer equation (Eq. (1)), (GB) of random-access memory (RAM) and a 3.70 GHz Intel® Xeon™
therefore, the difference between the energy moving in and out of the E1270 processor.
system govern the final temperature change. By combining all equa-
tions and boundary conditions together, Eq. (1) in terms of the overall 2.4. Optimization for greenhouse conveyor movement based on simulation
energy balance can be written as (Eq. (8)) result
− k∇T = h (T − Tambient) + εi σ (Ti4 − T j4 ) (8)
Once the simulation results were obtained, the model data was used
The physical properties of the materials used in this study are to optimize the movement of the conveyor. Initially, the simulation
summarized in Table 1, which referred from COMSOL material library model calculated the temperature and radiation by assuming that the
(see Table 2). position of all targets was fixed. However, to incorporate and optimize

136
D. Ma, et al. Computers and Electronics in Agriculture 162 (2019) 134–142

at 1.5 m, and so on. Values of all coordinates were determined by their

real positions from model shown in Fig. 5. The conveyor running speed
was set to be 12 m/min and the time required to start and stop the
conveyor was neglected for this study as they were relatively small
compared to the total movement time. Since the initial position (X0) of
each plant, speed of conveyor (v), and the direction of the conveyor
were known, the time-based position (Xt) of the plant can be calculated
by linear position function (Eq. (9)).

Xt = vt + X0 (9)

The microclimate information regarding both temperature and ra-

diation varied in spatio-temporal domain. The expressions of tem-
perature and radiation could be written as temperature (Xt, t) and ra-
diation (Xt, t), respectively. Once the target plant and running time
were determined, the real-time position (Xt) of the plant was calculated
from Eq. (9). Based on Xt, two closest plants from 38 fixed position
Fig. 3. The air flow inside and around house related with temperature and could be selected which were marked as X1 and X2. At the same time,
radiation. simulated temperature and radiation results were also obtained as
T1(X1, t), T2(X2, t), R1(X1, t), and R2(X2, t). Then the real time tem-
perature T (Xt, t) and radiation R (Xt, t) of plant could be calculated by
linear interpolation method (Eqs. (10) and (11)).

R2 (X2 , t ) − R1 (X1 , t )
R (Xt , t ) = R1 (X1 , t ) + (Xt − X1 ) ∗
X2 − X1 (10)

T2 (X2 , t ) − T1 (X1 , t )
T (Xt , t ) = T1 (X1 , t ) + (Xt − X 1) ∗
X2 − X1 (11)

For either running and pause phase, there are 8 step levels for dif-
ferent movement frequencies and distances. For example, if a combi-
nation of 40 s running and 240 s pause phase is selected, then the
conveyor can move for 40 s followed by a 240 s rest period and then
process can be repeated. Finally, there were 64 combinations from
8 × 8 step levels. For each set of combination, the accumulated radia-
tion energy received by 38 plants was calculated by summing up R(Xt,t)
Fig. 4. Greenhouse is divided into sub faces by meshing, of which facets are
from Eq. (10). Different possible combinations were compared to op-
triangles. timize the conveyor movement in terms of the highest energy efficiency
as suggested by lowest coefficient of variation (Eq. (12)) and smallest
running time ratio.
the effect of the conveyor movement, the changing temperature and
radiation over the movement should be calculated based on each as- Standard Deviation
Coefficient of variation (CV) = × 100
sumed settings of conveyor running time, break time, speed, and the Average (12)
initial position.
There were 38 plant samples distributed in the greenhouse (Fig. 5).
Each plant was labeled and its position along the conveyor was marked. 2.5. Measurements and validation
For instance, plant 1 was set at 0 m, plant 2 was at 0.75 m, plant 3 was
To validate the simulated results, the temperature and the radiation
data were collected from each pot with the help of Xiaomi flower care
sensor (Xiaomi Inc., Beijing, China). With embedded temperature and
radiation sensors (Fig. 6a), Xiaomi flower care sensor (Model:
HHCCJCY01HHCC) provided real-time temperature and light intensity
recordings. As shown in Fig. 6b, the sensor was inserted into the soil of
the pot, with actual plants when collecting temperature and radiation.
The data was continuously collected for seven days from Aug 09, 2018
to Sep 04, 2018 and logged at 1 Hz with the help of single board
Raspberry pi 3 computer (Raspberry Pi foundation, Cambridge, United
Kingdom).

3. Results and discussion

By solving the combined heat transfer equation based on energy

balance, the simulation model developed in Section 2 enabled a de-
tailed analysis of the temperature and/or radiation change, and dis-
tribution caused by microclimates in the greenhouse. The analysis re-
veals significant environmental condition differences existing at
Fig. 5. Layout of simulated plants in the greenhouse. different locations within greenhouse.

137
D. Ma, et al. Computers and Electronics in Agriculture 162 (2019) 134–142

Temperature sensor

Radiation sensor

Fig. 6. (a). The layout of Xiaomi flower care sensor including temperature and radiation sensors; (b). The sensor was inserted into the soil of the pot during data
collection.

Fig. 7. Simulated radiation (W/m2) at four different time points during a day, (a) 8 am, (b) 12 pm, (c) 18 pm and (d) 20 pm. The radiation value is shown in grey-
scale color legend.

3.1. Simulated temperature and radiation received relatively less radiation energy. The radiation prediction per-
fectly met the expectation. For example, at 8am Plant1 was at the west-
The 3-D simulated radiation profiles for the greenhouse illustrated north corner in this greenhouse. Because of the imaging tower, it did
the spatio-temporal variation of the radiation (Fig. 7). As the sun rises not receive enough radiation energy (Fig. 8b). On the other hand, at
from the east, Fig. 7a demonstrated that the east facing parts of 6 pm Plant 4 was under the shadow of imaging tower. The complex
greenhouse were much brighter indicating they received more radiation structure of this greenhouse led to nonuniform radiations received by
energy in the morning. Also, the west oriented shadow was clearly plants.
shown in Fig. 7a. At noon, due to the high quantity of solar radiation, The 3-D simulated temperature profiles for the greenhouse were
overall color of Fig. 7b was brighter. Fig. 7c showed the radiation shown in Fig. 9. Like Fig. 7, temperature changes were clearly illu-
distribution at 6 pm (just about the sunset) with the shadow being or- strated over both time and space. In the morning, the overall tem-
iented to east due to the west oriented sky. At 8 pm, when it was ap- perature was relatively low (Fig. 9a). The east face of the greenhouse
proaching night, the radiation became very weak in the greenhouse. showed higher temperature because of the direct incident solar radia-
For more specific radiation distribution information, we selected tion in the morning. At noon (Fig. 9b) and in the afternoon (Fig. 9c), the
five simulated plants (Fig. 8a). These five sample were from different temperature of the greenhouse was high. Particularly for a specific part
locations of the greenhouse covering solar radiations from low to high. like cement floor within greenhouse, the temperature was up to 45 °C
The temporal distribution of the radiation for the selected five plants due to the continuous exposure to the incident sunlight and its intrinsic
was plotted (Fig. 8b) and indicated that the plants under shadow properties of small heat capacity.

138
D. Ma, et al. Computers and Electronics in Agriculture 162 (2019) 134–142

Fig. 8. (a) Positions of 5 samples labeled from top view of greenhouse, (b) simulated radiation (W) of plant as function of time (radiation energy is calculated for the
total integrated plant body surface).

Then six simulated plants were chosen for example from different led to nonuniform temperature distribution. In this simulation, we did
locations covering temperatures from low to high. As shown in Fig. 10a, not consider plant evaporation. Meanwhile, we used thermal properties
six plants were chosen, of which temperature change based on timeline of “wood” as parameter inputs for plant in the simulation model (From
were extracted and plotted in Fig. 10b. In this simulation, plant’s COMSOL 5.2b material library). That was the reason the simulated
temperature was largely affected by position even in the same green- temperature reached as high as 45 °C. The main idea of simulation was
house. Plants displayed lower temperature when they were exposed to to quantity the variation from heterogenous environment in the
the cooling wall from north side of greenhouse (Plant 3, 5 and 6). The greenhouse. We assumed the simulated results was still able to be re-
Plant 2 and 4, which were blocked by imaging tower from cool wall, presentative for showing the environmental variances.
obviously had higher temperature. For all the plants in Fig. 10b, they
also had a temperature change profile along the timeline. They
achieved highest temperature point around 3–4 pm in the afternoon. As 3.2. Model validation
September is still in the summer, the hottest part of the day during the
summer is typically between 3 p.m. and 4:30 p.m. ET depending on The greenhouse model was simulated based on assumption of sunny
cloud cover and wind speed. The complex structure of this greenhouse day and constant wind speed. For validation purposes, we chose seven
days from Sep 03, 2018 for the ground truth data collection, as the

Fig. 9. Simulated temperature (°C) at four different time points during the day, (a) 8 am, (b) 12 pm, (c) 18 pm and (d) 20 pm. The temperature value is shown in grey-
scale color legend.

139
D. Ma, et al. Computers and Electronics in Agriculture 162 (2019) 134–142

Fig. 10. (a) 6 sample positions labeled from top view of greenhouse. (b) Simulated temperature of plant as function of temperature (°C).

weather matched environmental assumptions. As introduced in the treatment following the same idea as the “accumulate radiation”. The
Methods, the data was logged at 1 Hz by Xiaomi flower care sensor. key idea of this simulation model was to predict the heterogenous en-
Therefore, it was 1 measurement per minute. In total, there were 1440 vironment (temperature and radiation) in the greenhouse. Therefore,
instantaneous data points for one day, accumulated from 12:00 am to we referred “accumulated temperature” as the accumulated impacts
the 12:00 am of the next day. from temperature on the plants by summing up the measured tem-
For radiation, based on the collected data and simulated result, we perature data points. Accumulated heat based on the temperature
calculated the diurnal accumulated radiation value and made com- prediction and ground truth measurement for 20 samples was de-
parison between simulated and measured radiance. In the study, we termined from the matching time period. The R2 between the mea-
were focusing on quantifying how variant of microclimate (radiation surement and prediction was 0.8819 (Fig. 12). This result showed that
and temperature) is. From this point of view, the “accumulated radia- the accumulated heat based on temperature could be quantified using
tion” and “accumulated temperature” were the suitable choices, as well simulation model, and the variation from different spots could be
as easy to be compared. There were 20 samples in Fig. 11 covering low clearly presented.
to high radiation regions. x-axis was the simulated radiation from
software COMSOL 5.2b, and the original unit was W. Therefore, the 3.3. Use model to drive movement of the conveyor
unit of simulated accumulated radiation was J/day. On the other hand,
the y-axis was the measured radiation by Xiaomi flower care sensor, The simulation result data was used to compare 64 different com-
and the original unit is Lux. Therefore, the unit of measured accumu- binations of running and break time settings. The goal is to find the
lated radiation was Lux/day. The results of the linear regression in- optimized conveyor “run” and “stop” time intervals, so that the mi-
dicated that measured and simulated radiation were significantly cor- croclimate effect is still mostly eliminated while the conveyer running
related with R2 of 0.9053. cost is significantly reduced. Fig. 13 showed the microclimate variance
The “accumulated temperature” meant the “integration of tem- in each case as coefficient of variation (CV) of accumulated radiation
perature over time”. It was used in this study as an evaluation of heat energy. The table was colored as a gray-scale heatmap to clearly show

Fig. 11. Accumulate radiation results: real versus simulated measurements.

140
D. Ma, et al. Computers and Electronics in Agriculture 162 (2019) 134–142

Fig. 12. Accumulate temperature results: simulated versus real measurements.

the different CVs for the 64 combinations of settings. Generally, plants conveyor system.
in the greenhouse would receive more uniform radiation energy with
longer running time and shorter break time: for same break time, CV
became smaller when the running time increased, since longer moving 4. Conclusions
distance should help eliminate microclimate effect; Oppositely, under
the condition of same running time, CV became larger with increasing A numerical greenhouse temperature and radiation simulation
of break time, since lower moving frequency will make the plants suffer model was developed for the purpose of greenhouse microclimate
more from microclimates. To indicate uniformity of radiations, we as- control. The model successfully simulated temperature and radiation
sumed CV larger than 1 to be high-variance whereas those with a CV profiles over time and space in Purdue University’s Lily 13-4 green-
lower than 1 are considered low-variance. After filtering all low-var- house. By applying the simulation results in optimization of conveyor
iance cases, the optimal choice could be made by selecting the best system movement in Lily 13-4, this study showed new insights into how
energy efficiency combination. For example, conveyer setting with to quantitatively use simulation data to minimize greenhouse micro-
“10 s running time + 8 min break time” would have fairly low energy climate. The proposed model was validated by comparing simulated
cost among all 64 combinations while still keeping low CV as 0.91. For radiation and heat result with ground truth measurements from tem-
that, unnecessary conveyor movement was avoided. Shifting from ori- perature and light sensors distributed in the greenhouse. The R square
ginal setting with continuous running to new setting “10 s running between prediction and measurement is 0.91 for accumulated light
time + 8 min break time” saved more than 95% energy from the radiation and 0.88 for accumulated heat.
As an application, results from the model showed its potential to

Fig. 13. Coefficient of variation from different running and break time combinations for conveyor movement optimization.

141
D. Ma, et al. Computers and Electronics in Agriculture 162 (2019) 134–142

optimize pots movement in the greenhouse with lower cost while greenhouse experiments with special reference to greenhouses containing plants on
keeping same efficiency in eliminating microclimate problem. It will conveyor systems. Plant Methods 9 (1), 5.
Castro, V., Isard, S.A., Irwin, M.E., 1991. The microclimate of maize and bean crops in
greatly benefit the current automated greenhouses with conveyor sys- tropical America: a comparison between monocultures and polycultures planted at
tems, which have been developed and applied in many plant breeding high and low density. Agric. For. Meteorol. 57 (1–3), 49–67.
researches (Ge et al., 2016; Gehan et al., 2017; Golzarian et al., 2011). Frei, W., 2016. Thermal modeling of the air flow inside and around your house. Retrieved
from https://www.comsol.com/blogs/thermal-modeling-of-the-air-flow-inside-and-
In Purdue’s Lilly 13-4 greenhouse case, shifting from original setting around-your-house (date accessed: Oct 31, 2018).
with continuous running to optimal setting “10 s running time + 8 min Ge, Y., Bai, G., Stoerger, V., Schnable, J.C., 2016. Temporal dynamics of maize plant
break time” suggested by the model could save more than 95% energy growth, water use, and leaf water content using automated high throughput RGB and
hyperspectral imaging. Comput. Electron. Agric. 127, 625–632.
from the conveyor system. In situations when pots movement is im- Gehan, M.A., Fahlgren, N., Abbasi, A., Berry, J.C., Callen, S.T., Chavez, L., Sax, T., 2017.
practical, this modeling method can still help the researchers to quan- PlantCV v2: Image analysis software for high-throughput plant phenotyping. PeerJ 5,
titatively estimate the severity of microclimate impact, so to set proper e4088.
Golzarian, M.R., Frick, R.A., Rajendran, K., Berger, B., Roy, S., Tester, M., Lun, D.S., 2011.
expectations on the signal-over-noise level when analyzing the plant
Accurate inference of shoot biomass from high-throughput images of cereal plants.
growth measurement data. Plant Methods 1–11.
Gonzalez-Real, M., Baille, A., 2000. Changes in leaf photosynthetic parameters with leaf
Acknowledgements position and nitrogen content within a rose plant canopy (rosa hybrida). Plant, Cell
Environ. 23 (4), 351–363.
Kimball, B., 1973. Simulation of the energy balance of a greenhouse. Agric. Meteorol. 11,
This project was supported by Department of Agricultural and 243–260.
Biological Engineering, Purdue University. The authors would like to Korioth, T., Versluis, A., 1997. Modeling the mechanical behavior of the jaws and their
related structures by finite element (FE) analysis. Crit. Rev. Oral Biol. Med. 8 (1),
thank Libo Zhang (PhD. student, Agricultural and biological 90–104.
Engineering Department, Purdue University) and Liangju Wang Körner, O., Aaslyng, J.M., Andreassen, A.U., Holst, N., 2007. Microclimate prediction for
(Postdoc, Agricultural and biological Engineering Department, Purdue dynamic greenhouse climate control. HortScience 42 (2), 272–279.
Lindquist, J.L., Arkebauer, T.J., Walters, D.T., Cassman, K.G., Dobermann, A., 2005.
University for their assistance during the system design and sensor data Maize radiation use efficiency under optimal growth conditions. Agron. J. 97 (1),
collection. 72–78.
Radojević, N., Kostadinović, D., Vlajković, H., Veg, E., 2014. Microclimate control in
greenhouses. FME Trans. 42 (2), 167–171.
References Stone, P., Sorensen, I., Wilson, D., 1998. Radiation interception accounts for the effects of
plant population on maize yield. In: Paper Presented at the Proceedings Annual
Ahemd, H.A., Al-Faraj, A.A., Abdel-Ghany, A.M., 2016. Shading greenhouses to improve Conference of Agronomy Society of New Zealand 28. pp. 9–10.
the microclimate, energy and water saving in hot regions: a review. Sci. Horticult. Taki, M., Ajabshirchi, Y., Ranjbar, S.F., Rohani, A., Matloobi, M., 2016. Modeling and
201, 36–45. experimental validation of heat transfer and energy consumption in an innovative
Ahonen, T., Virrankoski, R., Elmusrati, M., 2008. Greenhouse monitoring with wireless greenhouse structure. Inf. Process. Agric. 3 (3), 157–174.
sensor network. In: Paper Presented at the IEEE/ASME International Conference on Teitel, M., Ziskind, G., Liran, O., Dubovsky, V., Letan, R., 2008. Effect of wind direction
Mechtronic and Embedded Systems and Applications, 2008. MESA 2008, pp. on greenhouse ventilation rate, airflow patterns and temperature distributions.
403–408. Biosyst. Eng. 101 (3), 351–369.
Alain, B., 1988. Greenhouse microclimate and its management in mild winter climates. van Eck, R., Klep, M., van Schijndel, J., 2016. Surface to surface radiation benchmarks. In:
In: Paper Presented at the International Symposium on Protected Cultivation of Paper Presented at the COMSOL Conference, 12–14 October 2016, Munich, Germany,
Ornamentals in Mild Winter Climates 246. pp. 23–36. pp. 1–6.
Baille, A., Kittas, C., Katsoulas, N., 2001. Influence of whitening on greenhouse micro- Vásquez, L., Iriarte, A., Almeida, M., Villalobos, P., 2015. Evaluation of greenhouse gas
climate and crop energy partitioning. Agric. For. Meteorol. 107 (4), 293–306. emissions and proposals for their reduction at a university campus in Chile. J. Cleaner
Baxevanou, C., Fidaros, D., Bartzanas, T., Kittas, C., 2010. Numerical simulation of solar Prod. 108, 924–930.
radiation, air flow and temperature distribution in a naturally ventilated tunnel Zhang, L., Maki, H., Ma, D., Sánchez-Gallego, J.A., Mickelbart, M.V., Wang, L., Jin, J.,
greenhouse. Agric. Eng. Int.: CIGR J. 12 (3–4), 48–67. 2019. Optimized angles of the swing hyperspectral imaging system for single corn
Brien, C.J., Berger, B., Rabie, H., Tester, M., 2013. Accounting for variation in designing plant. Comput. Electron. Agric. 156 (September 2018), 349–359.

142