Professional Documents
Culture Documents
Ph.D. Dissertation
Contribuciones al Modelado y
Simulación de Sistemas de Cultivo
Intensivos
Universidad de Almerı́a
Departamento de Informática
Universidad de Almerı́a
Departamento de Informática
To my parents
To my sister
“One certainly cannot make an investigation, or do it well, without zeal, a
determination to direct one’s eyes downward and a thirst for knowledge,
and without shedding the ugly mantle of pretentiousness and becoming a
willing pupil”
(Chairman Mao Zedong)
ix
Agradecimientos
xi
Resumen
xiii
xiv Resumen
xv
xvi Abstract
house crop growth taking into account economic and energetic criteria, in
such a way that the difference between the gross benefit obtained by the
sale of the production and the associate costs to the climate and irrigation
actuators are maximized.
The fulfillment of some specific objectives has been required to account
for the main objective of this thesis, such as designing, studying and mod-
elling of a CO2 storage, carbonic enrichment and heating system based in a
cheap fuel as the biomass which comes from the greenhouse residues, study-
ing and modelling those greenhouse variables that characterise crop growth,
so variables that affect to the substrate hydric balance. These models are
integrated in a two layers hierarchical control architecture that control the
system composed by crop, irrigation, and climate. The upper layer solves an
optimisation problem as a function of the awaited production and associate
costs or the desired date of harvesting. This optimisation problem maximize
an objective function that represents the benefit obtained based on the cli-
matic and irrigation variables that affect the growth of the plants, providing
the set points that must follow these climatic variables throughout the cam-
paign. The lower layer includes the controllers that try to cancel set point
tracking errors (these set points are those calculated by the upper layer).
The thesis has not been limited to the development of models, but it
has required an important work of experimentation in greenhouses during
several years, with the purpose of collecting data required for the calibration
and validation of models. Some of the developments included in the thesis
have been implemented in commercial control systems through technology
transference contracts.
Contents
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Agradecimientos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
Resumen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxvi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Intensive growing systems complexity . . . . . . . . . . . . . . . . . . . . . 2
1.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Research scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Main contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Outline of this Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
xvii
xviii Contents
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
List of Figures
xxi
xxii List of Figures
3.16 Calibration curve for the water content sensors using different
EC values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.17 Water consumption validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.18 Real and accumulated XLAI estimation by TOMGRO . . . . . . . 76
3.19 Total and root dry weight estimation by TOMGRO . . . . . . . . . . 76
3.20 Climatic and water balance data for three days chosen as an
example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.21 A three-day example for the different virtual sensors . . . . . . . . . 78
3.22 One-day example for the different virtual sensors . . . . . . . . . . . . 78
xxv
xxvi List of Tables
AC Activated Carbon
AER Automatic Control, Electronic, and Robotics Research
Group
AgEng Agricultural Engineering
AR Auto-Regressive
ARM Automatic Control, Robotics, and Mechatronic Re-
search Group
ARIMA Auto-Regressive Integrated Moving Average
ARMA Auto-Regressive Moving Average
ARX Auto-Regressive with eXogeneus inputs
ARMAX Auto-Regressive Moving Average with eXogeneus in-
puts
BJ Box-Jenkins
CEA Automatic Control Spanish Comitee - Comit Español
de Automática(in spanish)
CEA CAJAMAR Agronomic and alimentary Studies Center - Centro de
Estudios Agroalimentarios (in spanish)
CEDER Renewable Energies Development Centre - Centro de
Desarrollo de Energı́as Renovables (in Spanish)
CeiA3 The Agrifood Campus of International Excellence -
Campus de Excelencia Internacional Agroalimentario
de Andalucı́a (in Spanish)
xxvii
xxviii Acronyms
“A Chemical Engineer would not design a chemical plant with its process controls,
etc., without having a model of the chemical process in the plant to form the basis
of design. But the Agricultural Engineer whose business is to design environmental
control systems for the largest chemical factory in the world (the transformation of
light energy and chemicals into food) has no adequate model of the system he must
work with.”
(R.B. Curry [35])
1
2 1 Introduction
these such as lack of uniformity, in both the products and crop environments,
with a very mountainous terrain, and adverse weather variables [134].
Currently, Almerı́a’s agriculture faces major demands for quality, product
variety, and strong competition. Such demands force farmers to improve the
production process, which usually leads to rise in the harvest final cost [137].
This issue increases the risk assumed by farmers and lead to a reduction in
the profits. Automation, as a first step, and the improvement in the structures
and growth systems, genetic engineering, and the utilization renewable energy
systems, allowed a cost reduction and better results in total production [147].
Consequently, there is much interest at present in introducing technology,
not only in the production process, but also in the agricultural marketing
chain, from germination of seeds to consumer sales [132].
when the temperature has to decrease, and heating at night, when the tem-
perature has to be raised [13]. On the other hand, when the ventilation is
used to control the temperature, others variables as humidity are affected.
In view of the crop growth, others climate variables like relative humidity
must be also introduced in control systems to maintain the water content in
the greenhouse air within a certain range and avoid extreme situations. Very
high relative humidity increases the appearance and development of crypto-
gamic diseases [41], lowers transpiration [148], which reduces absorption of
water and nutrients, and could generate a deficit of elements such as calcium
[105]. On the other hand, with low relative humidity, the transpiration rate
rises, increasing the water demand, which could cause hydric stress, closing
the stomatas, and thereby reducing photosynthesis [167]. The best results
in relative humidity control have been obtained increasing natural or forced
ventilation (when outdoor conditions are more favorable), with heating sys-
tem (increasing the temperature reduces the risk of condensation), and using
dehumification and humidification systems (reducing and increasing the air
water content, respectively). These actions decrease the risk of condensation
presence on the roof or on the plants [41].
As it is commented above, the dynamic behaviour of the greenhouse cli-
mate is a combination of physical processes involving energy transfer and
mass balance. These processes are affected by the outlet environmental con-
ditions, structure of the greenhouse, type and state of the crop, and on the
effect of the control actuators based on predefined setpoints [15, 16]. Also,
such processes have a strong relationship among them, so daylight temper-
ature and humidity control is performed by convective air exchange between
the outside and inside [13]. This exchange rate coupled with CO2 taken by the
crop during photosynthesis determine the concentration of CO2 in the green-
house. This rate of CO2 assimilation (i.e. photosynthesis) is further modified
by the amount of CO2 presents in the surrounding air of the plant, and
hence influences on the production of agricultural crops. That assimilation
rate, linked to the low ventilation rates that presents the classical Mediter-
ranean greenhouses overall with low wind velocity, provokes a decrease of
the CO2 concentration [39, 61, 96], uptake limitation and suppression of the
photorespiration, producing a growth deficit that is increased when the crop
reaches its maximum development [142].
To avoid such situations, some authors propose different techniques of
carbon enrichment in order to overcome these problems and achieve the final
increase crop production [82, 143, 150]. An increase of CO2 concentration
can be archived relatively easily in a closed environment such as a greenhouse
[9, 30, 157]. Carbonic enrichment, i.e. supply of CO2 or rich combustion gas, is
a powerful tool to enhance the horticultural production, so the photosynthesis
rate increase translates into a significant rise in production [104, 143, 150, 180,
200]. It is also essential to optimize carbon fertilization strategy responses,
performing them in the absence of other limiting factors where the best results
6 1 Introduction
are obtained when radiation levels are high i.e. the photosynthetic rate is
higher when temperature and CO2 concentrations rise along with an increase
in available solar radiation [89, 90, 180]. Carbon enrichment not only affects
the photosynthesis rate, also it increases water consumption efficiency by
gradually reducing the stomata-opening rate, decreasing water loss caused
by transpiration (water vapor released by plants), which has influence the
humidity content [180] and the water dynamics [142].
In this line, the optimization based on the knowledge of the water dynamics
is basic for plants development; water is a constituent of vegetable tissues, a
solvent, a reagent, keeps cellular turgor [85, 187, 107], and is an excellent me-
dium for temperature regulation [187]. Between 80 and 90% of fresh weight
of plants is water. Furthermore, water is a limited resource in many agricul-
tural areas. This resource limitation is made worse due to the recent rapid
expansion of the surface area occupied by greenhouses in the Mediterranean
Basin. It is becoming a more important consideration in the sustainability
of the greenhouse-based system in Almerı́a [152]. So, the water needs to be
supplied precisely due to an excess in water supplied might result in radi-
cular asphyxiation, substrate flooding or subterranean water contamination
[48, 49]. Conversely, a hydric deficit can lead to a decrease in production and
might even be dangerous to crop growth. Hence, automatic irrigation control
systems are fundamental tools in supplying water to the crop at the required
amount and frequency [107, 171]. An inadequate water supply results in an
immediate reduction of vegetative growth. Therefore, it is important to ana-
lyze the hydric balance in a crop because this analysis helps to understand
the substrate-plant-atmosphere continuum dynamics and makes it possible
to efficiently manage water [85, 107, 171]. This analysis can be done conside-
ring: water-content in substrate, water uptake, transport from root to leaves,
and movement from leaves to the atmosphere (transpiration) [152].
Finally, it is important to notice that radiation, by extension Photosyn-
thetic Active Radiation (PAR), is not a limiting variable in warm climates.
It is usually controlled with shade screens and artificial lights, and its use
is not much extended in warm climates. In this Thesis and in most of the
works include in it, this variable is only consider as a system disturbance
[59, 111, 134].
the experience of the people who work in the specific field [71, 32, 197]. The
availability of this information would help to optimize crop growth (produc-
tion), lower costs and assist in decision-making at the right time. Hence, the
grower has to choose what the optimal conditions for plant growth are, and
to convert them in control specifications. In general, to control the green-
house environment, a grower has to deal with temperature, air humidity,
CO2 concentration, and radiation, in a system with strong physical relation-
ships (e.g. temperature and air humidity) which influence on the greenhouse
environment control [133].
Furthermore, that control decisions are modified by the market price fluc-
tuations and the environmental rules to improve water-use efficiency, to re-
duce fertilizer residues in the soil (such as the nitrate contents) [129] or in-
crease the use of renewable energies [21]. Therefore, the optimal production
process in a greenhouse agrosystem may be summarized as the problem of
reaching the following objectives: an optimal crop growth (bigger production
with better quality), reduction of the associated costs (mainly fuel, electricity,
and fertilizers), reduction of residues (mainly pesticides and ions in soil), and
the improvement of water use efficiency [189]. From an objective function,
with the use of climate, fertirrigation and growing models, along with optim-
1.2 Problem statement 9
are the two models that help understand the water flow from substrate (ir-
rigation) to the water vapour releases to the atmosphere (transpiration). On
the other hand, the biomass-based system hybrid-model was established due
to the designed and installation of a patented biomass heating system with
flue gases recovery for carbonic enrichment. This model allows the incorpor-
ation of this system to the hierarchical scheme, allowing the introduction
of the continuous and discrete dynamics that such facility has. This system
was added to the greenhouse with the aim to reduce external fossil fuel de-
pendence, increasing the added horticultural production value and improving
sustainability, reducing the energy consumption.
energy sources because of the nature of their energy demand adapts perfectly
to the guidelines of, e.g., radiation solar or wind resource availability [151].
In this sense, there are many experiences in agricultural holdings of a viable
and effective use of autonomous systems of electrification and pump powered
by renewable energy [21, 122, 123]. In the case of greenhouses, themselves
can be considered as a Sun energy transformation devices, because of the
presence of crops in its interior makes the treatment of flows of mass and
energy is to be much more sophisticated than in a thermal solar collector,
[1, 17]. This analogy of departure, along with driving generic in this type of
holdings of optimization of consumption - water, energy, materials,...- acts
on behalf of the implantation of renewable sources in greater measure than
in other production environments.
In this line, the use of biomass as an energy source has undergone sig-
nificant developments over recent years due to the various initiatives imple-
mented by organisations and institutions operating renewable energy policy
[23]. The 2011-2020 Spanish Renewable Energy Plan (PER)[68] highlighted
biomass energy use as one of its priorities and performed an initial assess-
ment in which the incorporation of biomass linked to intensive agriculture
[146, 147, 195].
Biomass use would alleviate an important environmental problem and
could help to control nocturnal greenhouse temperatures during cold periods.
In the case of greenhouses, biomass consumption can achieve negative CO2
balances by reusing flue gases from combustion emitted from crop-enrichment
processes. The installation of a biomass heating system following elements:
biomass, combustion chamber storage tank transport, boiler (steam, hot wa-
ter, thermal oil), biomass transportation equipment from the tank to the com-
bustion chamber, auxiliary heat recovery, gas purification, ashes extraction
(in the biomass boiler ashes are removed periodically) and, additionally for
this thesis, an equipment recovery of gases for carbon enrichment [146, 147].
1.3 Objectives
input (ARMAX and NARMAX), have been also considered. In this case, an
effort in the design of appropriate input signals to identify and estimate both
the parameter and its structure was carried out.
Following these lines, the main objectives of this thesis focus on modelling
of intensive growing system. As a particular case, the resulting models will be
included in a general hierarchial control strategy designed for greenhouses.
For this purpose, it has been necessary to cover the following specific tasks:
This Ph.D. dissertation has been developed within the framework of six re-
search projects:
1.5 Main contributions 13
Among all the works performed during the development of this PhD thesis,
the following ones have been included in the current dissertation:
[152] J.A. Sánchez-Molina, F. Rodrı́guez, J.L. Guzmán, and J.A. Ramı́rez-
Arias. Water content virtual sensor for tomatoes in coconut coir substrate
for irrigation control design. Agr Water Manage. 151: 114-125. 2015c. In
this work, three different virtual sensors were developed to estimate the hy-
dric balance in crop growth from climatic data and the water supplied: (1)
based on first principles, and (2) based on linear black-box and (3) non-linear
black-box. Thus, the aim of this paper is to provide several alternatives of
virtual sensors that help in the design of irrigation controllers based on the
14 1 Introduction
vironmental point of view and, (ii) as much as the crop produces in this
conditions and the quality of that production.
The data used in this research was acquired from the Cajamar Foundation
Experimental Station greenhouses in El Ejido, Almerı́a Province, Spain (2o
43W, 36o 48N, and 151 m a.s.l.). The crops grew in a multispan Parral-type
greenhouse (Fig. 2.1a). The greenhouse is 877 m2 (37.8 x 23.2 m) with a
variable height (between 2.8 and 4.4m), having a polyethylene cover. The
structure of the greenhouse is symmetric in area and the roof runs from East
to West, and crop rows are aligned north–south.
Respect to the actuators, the greenhouse counts with automated ventila-
tion with windows in the north and south walls (Fig. 2.1b), and flap roof win-
dow in each span (Fig. 2.1c), 20 x 10 threads x cm−1 mesh bionet anti-insect
screen, an heating system with an aerothermal generator of 95 kW (Ernaf
19
20 2 Material and Methods
Fig. 2.1 Greenhouse used in this work: a. Greenhouse, b. side windows with anti-
insect screen and the engine, and c. roof windows with anti-insect screen
For the development of this Thesis, different cycles of tomato and pepper
crops with different varieties were used from 2005 to 2014 (Fig. 2.2a). Fur-
thermore, these cycles had different lengths: (i) short cycles, a spring cycle
and autumn-winter separated in the same year (around 120 days each), and
(ii) long crop cycle, one per year (approx. 220 days). The crops were grown
in two different soilless cultures in 15 l rockwool slabs (100 cm x 20 cm x
7.5 cm, Fig. 2.2d), and in 40 l coconut coir slabs (130 cm x 35 cm x 9 cm,
Fig. 2.2b), both enclosed tightly in plastic, placed in gutters with a 1% lon-
gitudinal slope. Crop density was 2.01 plants m−2 (6 plants per slab, 1.9 m
between rows of slab and 0.5 m between slabs in the same row).
2.1 Greenhouse environment 21
Fig. 2.2 Crop growing system: a. tomato crops in production, b. coconut coir slabs
and black soil net , c. emitters, and d. seedlings in rockwool cubes on rockwool slabs
Seedlings, always grown in rockwool cubes, were transplanted into the slabs
before the inflorescence of the first truss was visible. Plants were vertically
supported by nylon cord guides, and were regularly pruned to remove all
auxiliary shoots leaving only the main stem. The irrigation is automated by a
demand tray, which is formed by two crop bags. Standard complete nutrient
solutions, the recommended one by Cajamar Foundation researchers, were
applied in all irrigations through a drip irrigation system (3 emitters per slab,
discharge rate of 3 l h−1 , Fig. 2.2c). The nutrient solutions were prepared in
accordance with established local practices. The nutrient solution was applied
to maintain a target drainage fraction of approximately 25 %, in accordance
with local practice. Drainage water was not re-circulated.
Climate sensors The diurnal and nocturnal temperature, humidity, and CO2
of this greenhouse were the controlled variables. During the crop season, sev-
eral internal and external measurements were monitored continuously. Out-
22 2 Material and Methods
side the greenhouse, a weather station measured air temperature and relative
humidity with a ventilated sensor (Vaisala HMP45P), solar radiation (Delta-
Ohm LP PYRA 03), photosynthetic active radiation with a silicon sensor
(PAR, Kipp&Zonnen PAR Lite), rain detector, CO2 concentration (Vaisala
GMP222), wind direction (Met One 020C-L), and wind speed (DeltaT AN3).
The cover temperature (Thermopars T type) sensors were located on the east
(two sensors), and west (two sensors) sides. During the experiments, the in-
door climate variables were also taken, especially air temperature, and relat-
ive humidity (Vaisala HMP45P), solar radiation with a pyranometer (Delta-
Ohm LP PYRA 03), photosynthetic active radiation (PAR, Kipp&Zonnen
PAR Lite), soil temperature (Decagon Devices RT-01), and CO2 concentra-
tion (Vaisala GMP222).
Fig. 2.3 Sensor network: a. CO2 sensor, b. wind speed and wind direction, c. micro-
lysimeter, d. rain detector, e. EC-5 sensor, and f. external PAR and global radiation
router that sends all the information to the data acquisition computer where
a Supervisory Control And Data Acquisition (SCADA) was designed and
installed, section 2.1.2.1.
Irrigation was periodically applied throughout each day of each crop. Ir-
rigation frequency was controlled with a demand tray system; fixed volumes
were applied. The demand tray system, using water level sensors, is the most
commonly used system for automatically activating irrigation of soilless crops
in SE Spain. A water level sensor is installed in a small water reservoir in
which the volume of water (and therefore the surface level) is in equilibrium
with the water content of the substrate. When the water level in the reservoir
decreases, through crop uptake, to the physical level of the sensor, irrigation
is activated. The physical height of the sensor is adjusted by the grower on
the basis of measured drainage volumes and experience. This method can be
used once the crop root system is established.
The main objectives of the SCADA system are the supervisory analysis of the
system and the validation of control algorithms, and the deviation analysis
between the real and the simulated behaviour of the greenhouse, derived from
the assumed hypothesis in its disturbances and another uncertainties. Other
objectives of the SCADA system are:
Fig. 2.4 Design of the SCADA system: a. Main screen, b. modules and sensors
configuration, c. graph for differents variables , d. start up variables, e. Matlab script
for control, and f. assays configuration screenshot
For the growth model, it was necessary to know the evolution of leaf area in-
dex and the dry weight (DW). It was determined through the leaf area meas-
urements of each plant removed for biomass task, the pruning, and deleafing
were also take into account. The samplings were made up of a destructive
sampling of five randomly selected plants every 21 days, duration accorded
in the research protocol. The choice of 21 days is twofold: first, it was the
sufficient time to find growth differences; and second, to avoid the elimination
of too much vegetal stuff in the greenhouse which could end in a modification
in the climate or transpiration measurements.
In biomass the process the following measurements were taken again: num-
ber of nodes, leaf area, number of fruits per bunch, fresh and dry weight of
leaves, stem, and fruits (Fig. 2.6e). The plant material and fruits were intro-
duced into a drying oven where they remained for 24-48 h (depending on the
phenological state, Fig. 2.6f) at a temperature of 65o C. Based on this, the
dry matter of leaves, stems, and fruits were determined by analytical balance.
The matter of leaves and secondary stems pruning came from the 12 selected
plants for samplings while kept in production. Once removed from the plant,
the fresh and dry weight was taken, such as biomass. In the case of pruning,
stems, and leaves are measured separately. Both, the bare and the pruning
are carried out for the leaf area index measures, executed, as in the biomass,
through electronic planimeter (Delta-T Devices Ltd, Fig. 2.6d). At the end of
the crop, the twelve plants from each of the four drainage trays were removed
and separated into leaves, stem and fruit which were weighed, and the dry
matter (DM) content determined. For each drainage tray, total shoot DM
was determined from the sum of total DM at final sampling, the combined
26 2 Material and Methods
DM mass of all pruned material, and the combined DM mass of total fruit
production.
Fig. 2.6 Crop and irrigation measurements facilities: a. Drainage tray and bucket,
b. Dripper bucket, c. laboratory , d. planimeter, e. weight scale, and f. drying oven
Drainage was collected in five metallic drainage trays (Fig. 2.6a), each tray
contained two slabs of plants (total of 12 plants and six drippers). Each drain-
age tray was positioned mid-way along the length of a crop row; the crop rows
were selected to ensure representative sampling of the greenhouse. Drainage
was collected separately and measured. All the containers and buckets for
collecting drainage were covered with white polyethylene sheeting to reduce
incoming radiation thereby minimising evaporation. The total volume of nu-
trient solution applied to the cropped area was measured with a flow meter
that was read manually each day before irrigation beginning. Three drippers
were placed in individual buckets (Fig. 2.6b). Representative samples of ap-
plied nutrient solution, for the previous day, were taken from each bucket
daily.
For this work, a biomass-Based system has been designed and tested, which
consists of a greenhouse residue heating system to control the nocturnal tem-
perature at low cost, and a CO2 storage system, where CO2 generated from
the heater is stored to be use during the diurnal photosynthesis process. The
energetic and economic efficiency is improved by reusing greenhouse organic
material (biomass, Fig. 2.7b) and CO2 generated from the heating system.
The main design methodology, system description and test carried on for
heating, CO2 storage and CO2 enrichment system are shown in section 5.1.
2.2 Biomass-Based system 27
Fig. 2.7 Heating biomass-based system: a. Biomass boiler, b. tomato pellets used,
c. Heating distribution system, and d. aerial pipes
The heating circuit consists of a main supply pipe, a main return pipe and
distribution pipes along the crop lines formed by two polyethylene parallel
pipes. A calefaction pump and a continuous mixing valve is used to control
the pipe temperature. A Pt100 temperature sensor was installed in each pipe.
The distribution loop runs around the canopy lines placed 10 cm above the
ground.
The boiler (Fig. 2.7a) operates in accordance with the greenhouse heating
demand, normally at night. The flue gases generated are driven to storage
28 2 Material and Methods
by the action of a compressor (Fig. 2.8a). The system is responsible for CO2
retention from the flue gas, releasing the rest into the atmosphere. CO2 sub-
traction allows a reduction in the volume of gas stored. The capture is per-
formed using an activated carbon (AC) adsorption bed (Fig. 2.8c). This AC
bed is employed because it is an inexpensive adsorbent, simple to operate,
with easy adsorbent regeneration for cycles of adsorption / desorption con-
trolled by pressure changes (compressor action). During the day, the CO2 is
released into the greenhouse with the aim to increase production, increasing
photosynthesis by carbonic enrichment. The full process has two different
control-loops: (1) the CO2 capture from the flue gases and (2) the release of
the CO2 stored. Both processes are carried out at two distinct times of the
day.
So, the system counts with: a compressor (Fig. 2.8a), heat exchangers, an
AC storage tank (Fig. 2.8c), a proportional valve (Fig. 2.8b), sensors (pres-
sure, air pipe temperature, CO2 and flow), distribution pipes (Fig. 2.8c),
and a three-way valve. The CO2 system works with two different processes:
capture and enrichment. Both procedures use the same actuator, the propor-
tional valve after the tank. During capture, the AC tank pressure is controlled
by this valve and pressure sensors located on the outlet and inlet tank pipes;
flow and CO2 sensors were also added to improve process information. The
last step in the CO2 storage system is a three-way valve, which releases the
remaining gases into the atmosphere. The same sensors are used to control
2.3 Modelling methodology 29
the CO2 mass flux released during enrichment; in this case, the flow is intro-
duced inside the greenhouse by altering the same three-way valve. The full
biomass-based system for heating and CO2 enrichment is shown in Fig. 2.9.
In addition to the energy balances that take place into the greenhouse,
other chemical phenomena such as mass transfers or the biological photo-
synthetic rate of the cell that is produced in presence of solar radiation and
CO2 , must be considered [123, 21]. Being solar radiation the most relevant
meteorological factor due to its influence on the main variables of the sys-
tem, being necessary to develop models to forecast climatic evolutions. On
the other hand, there are other variables that describe elements which can be
measured. They can be used as boundary conditions, thus reducing the com-
plexity of the modeling problem as the number of ODEs is reduced. However,
due to technical or economic reasons, sometimes several of these variables are
not measured, being necessary to estimate them. It makes the calibration a
complex task for the technicians that spend many hours in it. Due to the
large set of unknown parameters that this kind of models normally have, it
is difficult to obtain their values using a unique search technique with the
compound model.
The main interest of this Thesis is in the modeling of inside air CO2 con-
centration, crop growth and water-balance variables, based on fundamental
principles. The meteorological and irrigation inputs/outputs and the actu-
ator status data are used to calibrate such models without losing the physical
meaning of the processes involved in the balance equations. Once finished the
model election, the parameters of the equations are estimated in two phases:
(i) The search space of the physical sensor parameters was determined using
an iterative sequential algorithm to minimize a least-square-error criterion
between the real and the estimated water content (sequential iterative al-
gorithms), and (ii) a calibration process based on genetic algorithms.
Sensitibity analysis As modern computing capabilities increase, consider-
able effort is devoted to the development of computational tools for the ana-
lysis, design, control, and optimization of complex physical systems [125]. A
critical, and sometimes expensive, first step in the process of computer-aided
control is the formulation of a detailed and accurate model of the system. The
physical performance of the design is modeled by a system of mathematical
equations, usually ordinary or partial differential equations. For greenhouses
applications, analysis of the climate or irrigation requires efficient construc-
tion and manipulation of complex relationships along with fast and accurate
numerical methods for solving differential equations. Generally, there are a
huge number of physical parameters that engineer have to adjust (either
manually or within a computer simulation) to fix the model. In applications
such as greenhouse design, growth and control, some design variables may
have more influence than others in the final result. Consequently, research-
ers then become interested in how sensitive the state variables are to small
changes in the system variables [22].
Sensitivity analysis consists of a set of tools that can be utilized in the
context of optimization, optimal design, or simply system analysis to assess
the influence of parameters on the state of the system. Continuous sensit-
2.3 Modelling methodology 31
Unlike the models based on first principles, which need a prior knowledge of
the phenomena that take place in the system, the models based on black-grey
box are relatively easy to identify by an appropriate design of experimental
tests to be performed in the real system, although their validity and per-
formance strongly depend on the quality of data used for their structure and
parameter selection. This kind of models allow to capture linear dynamics by
32 2 Material and Methods
1. Choice of the model inputs that are suitable to consider for a particular
dynamic of the system. As was discussed above, the pH of culture is mainly
related to the CO2 injections and the available solar radiation.
2. Design of the excitation signals that allow to observe the non-linearities of
the process. For this purpose, signals with different frequencies and amp-
litudes were used to cover the dynamics around all operation conditions of
the system, obtaining the non-linear features of the process while avoiding
dangerous situations for the culture.
2.3 Modelling methodology 33
3. Choice of the model architecture and model order. In this case, a NARMAX
model was chosen due to its ability to represent non-linear dynamics by
a simple structure that can be used in an one-step prediction. A second-
order was selected, since a higher order increases the dimensionality of the
problem and hence its identification difficulty.
4. Identification of the model structure where several terms are selected from
a set of possible candidate terms. An algorithm composed of an Akaike
information criterion and a test of statistic significance was performed to
choose the model structure of the system reducing the computational time
dedicated to these procedures.
5. Estimation of the model parameters. The terms chosen in the previous step
are multiplied by some coefficients related to the parameters that must be
properly estimated. An optimization technique, such as orthogonal least
squares, was used to estimate the parameters of the non-linear structure.
6. Validation of the model with a different data set from the training data
set, in order to evaluate the bias error and the variance error correctly.
Chapter 3
Publications on Water Balance
35
36 3 Water Balance
imizing the dispensed water while fulfilling crop needs. This paper shows
how system identification techniques can be applied to obtain nonlinear vir-
tual sensors for estimating transpiration. The greenhouse used for this study
is equipped with a micro-lysimeter, which allows to continuously sample the
transpiration values. The micro-lysimeter is a good piece of equipment for re-
search, but it is expensive and requires maintenance. The paper presents the
design and development of a virtual sensor to model the crop transpiration,
hence avoiding the use of this kind of expensive sensors. The resulting vir-
tual sensor is obtained by dynamical system identification techniques based
on regressors taken from variables typically found in a greenhouse such as
global radiation and vapor pressure deficit. The virtual sensor is thus based
on empirical data. In the paper, some effort has been made to eliminate some
problems associated with grey-box models: advance phenomenon and over-
estimation. The results are tested with real data and compared with other
approaches. Better results are obtained with the use of non-linear Black-box
virtual sensors. This sensor is based on global radiation and vapor pressure
deficit (VPD) measurements. Predictive results for the three models are de-
veloped for comparative purposes.
3.1.1 Introduction
dangerous for the crop growth. Hence, automatic irrigation control systems
are fundamental tools to supply water to the culture in the required amount
and frequency. Moreover, as water is a limited resource in many agricultural
areas, optimizing productivity through efficient and adequate irrigation is
a basic objective. In order to design a good automatic irrigation system,
the following questions must be answered: what should the frequency of the
irrigations be, and how much water should be applied in the irrigation? To
answer these questions, it is necessary to know how much water should be
applied to replenish the losses due to the transpiration during the plant’s
respiration.
Measuring the water lost by transpiration is a way of obtaining the plant’s
water demand. This estimation of transpiration in different species grown
in greenhouses has been developed by [10] on ornamentals, [168], [73], [19],
and in tomato [10], [109] with geranium crops, [105] in cucumber, [170] for
rose cultivation, [194] for chrysanthemum, and [182] in Gerbera pots, among
others.
In most of these works, the micro-lysimeter became the basic measurement
device to record the water losses in crops, subtracting the water content in
an instant (t) by the water content in another instant (t − 1). However, in the
same way, in many occasions the measurements are not continuous due to
the irrigation process or during the water drainage. Furthermore, it is seldom
used by farmers since this device is expensive to acquire and to maintain.
From an operational point of view, it is important to find alternatives to this
irrigation system gadget. Thus, virtual sensors based on transpiration become
a good option to reduce total system cost, especially in the agriculture sector
where profit margins are so narrow. Such sensors must be based on sensors
that are typically installed in greenhouses for climate control (temperature,
humidity, and solar radiation), and thereby reducing the installations costs.
Virtual sensors become a very efficient and powerful tool that has been
successfully used in other fields [87, 126, 175]. These sensors utilize models in
order to estimate features from low-cost measurements. Ideally, the virtual
sensors should be simple, and obtainable from data collected. It should not
require extensive training. Virtual sensors are useful in replacing physical
sensors, thus reducing hardware redundancy and acquisition cost, or as part
of the fault detection methodologies by having their output compared with
that of a corresponding actual sensor. Virtual sensors may be developed based
on mathematical models obtained directly from the Physics of the system and
first principles. In many cases, such mathematical models are unavailable, or
their exact parameter values are unknown, or they are too complicated to
be used. For this reason, the development of virtual sensors often has to be
based on system identification [63].
The purpose of the current study is to develop a virtual sensor to in-
fer transpiration from other easily measured variables. For this purpose, the
38 3 Water Balance
The irrigation control systems are essential tools to provide water to the crop
in the required amount and frequency. Moreover, water is a limiting resource
in many agricultural areas. In such places, it should be a basic objective to
optimize their management and productivity through adequate and efficient
irrigation. The proposed control algorithm design (Fig. 3.1) is a hierarchical
control system, consisting of two levels:
amount that activates irrigation (setpoint). The irrigation starts when this
fixed value is exceeded, and finishes as soon as the amount of water lost has
been replenished.
A virtual sensor requires an accurate calibration and validation of the in-
stantaneous transpiration measurements at each sampling instant. The meas-
urement of transpiration includes direct (observation, porometer, lysimeter,
...), and indirect methods (water budget, energy balance, ...). The indirect
methods are more complicated for prediction and verification of the transpir-
ation values because of the meteorological factors involved. The transpiration
data can be collected through direct methods such as:
covering it with waterproof material. This method can be used with small
plants and crops in soilless culture. The results are expressed in grams or
milliliters of water transpired per leaf area per unit time.
This section is devoted to describe the main features of three different types
of virtual sensors for transpiration designed to replace the micro-lysimeter
in the automatic irrigation system. The aim of such virtual sensors is to
substitute the expensive micro-lysimeter in measuring the transpiration and
controlling the irrigation.
Fig. 3.2 shows the input variables of a virtual sensor for crop transpira-
tion: (1) global (Solar) radiation, (2) Vapor Pressure Deficit (VPD), which
is a function of the temperature (T) and the relative humidity (RH), and
(3) Leaf Area Index (LAI). These variables are measured by two different
sensors typically installed in commercial greenhouses: a psychrometer for the
temperature and the relative humidity (VPD), and a pyranometer for the
solar radiation.
The transpiration model needs to calculate the LAI. The LAI measure-
ments are taken in a non-continuous way. In this paper, a simplified TOM-
GRO model [128] adapted to the Mediterranean conditions is utilized to
estimate the LAI. The simplified TOMGRO model needs the temperature as
the input.
Based on the architecture of Fig. 3.2, three different virtual sensors are
considered in this paper. The first one is based on a pseudo-physical struc-
ture. The others have a more empirical structure with linear and non-linear
behaviors respectively. All three kinds are described in the following.
where the coefficients CA and CB are the parameters dependent on the crop.
In [10], it was observed that the coefficient CB increases with the XLAI ,
and furthermore, the coefficient adopted different values during the day due
to oscillations in stomatal resistance (eq. (3.2)). This daily oscillations was
corrected with two different parameters for diurnal (CBD ) and nocturnal
(CBN ).
The parametric virtual sensors (or black box) are diagrams capable of
representing any system without having any knowledge about the physical
process dynamics. Parametric virtual sensors are not obtained (at least not
completely) from the application of physical laws. These virtual sensors are
3.1 Virtual sensors for designing irrigation 43
Fig. 3.3 Graphical user interface of “ident”, a part of the System identification
Matlab-Toolboxr [92].
B(z) C(z)
A(z)y(t) = F (z) u(t − nk) + D(z) e(t) (3.3)
where, y(t) is the transpiration, u(t) is the different input variables, and e(t)
is the estimation error. A(z), B(z), C(z), D(z), E(z), and F (z) represent the
polynomials that define the output (transpiration), inputs, and the estima-
tion error. The order of the polynomial equations is defined by regressors,
where na outlines the outputs, nb is the order of the input, and nk is the
delay of the radiation solar (nkVGR ) and the vapor pressure deficit (nkVV P D ).
System Identification Matlab’s Toolbox [92] was used for the identification
process. The ARX, ARMAX, OUTPUT ERROR, BOX JENKINS, and FIR
formulations were tried out. The differences among these virtual sensors are
the way in which the inputs, outputs and disturbances are defined with para-
metric equations. The System Identification Toolbox of Matlab(R) software
was used to obtain the virtual sensor. Fig. 3.3 shows the main interface of
the Toolbox and the whole process to obtain a virtual sensor. The top of the
figure displays the calibration and validation process interface, and the bot-
tom shows the output analysis process. The toolbox allows to process data,
estimate the parameters of different types of structures, and validate virtual
sensors using different strategies.
For the identification process is only required to have data from experi-
ments. The data can be handled in the time domain or frequency domain,
and the experiments can have one or multiple inputs and/or outputs [60].
y(t) = f (y(t − 1), y(t − 2), y(t − 3), ..., u(t), u(t − 1), u(t − 2), ..) (3.4)
where y(t) is the output variable in t time; u and y are the different input,
and output variables (regressors); and f is the nonlinear function. Computer
regressors from the current, past input values, and past output data. The
current transpiration value is predicted through as a weighted sum of past
values, and current and past inputs values. With such information the equa-
tion becomes as:
y(t) = [a1 , a2 , ..., ana , b1 , b2 , ..., bnb ][y(t − 1), y(t − 2), ...,
(3.5)
, y(t − na), u(t), u(t − 1), ..., u(t − nb − 1)]T
where y(t − 1), y(t − 2), ..., y(t − na), u(t), u(t − 1), ..., u(t − nb − 1 are the re-
gressors, so called the delayed inputs and outputs. Nonlinear ARX regressors
can be both delayed input-output variables and more complex nonlinear ex-
pressions of delayed variables. The nonlinearity estimator block maps the
regressors to the virtual sensor output using a combination of nonlinear and
linear functions (Fig. 3.4).
Fig. 3.4 The nonlinear ARX computes the output y in two stages [92]. u(t) are the
model inputs, and y(t) are the model outputs.
The research data used in this work have been obtained from greenhouses
located in the Experimental Station of Cajamar Foundation, in El Ejido, in
the province of Almeria, Spain (2◦43’ W, 36◦48’ N, and 151 m elevation).
The crops grows in a multi-span ”Parral-type” greenhouse (Fig. 3.5); [128].
The greenhouse has a surface of 877 m2 (37.8 x 23.2 m), polyethylene cover,
automated ventilation [13] with lateral windows in the northern and southern
walls, flap roof window in each span, mesh-protected anti-trips ”bionet” of
20 x 10 thickness, and night heating applied with a 95 kW hot air heater
that is programmed to maintain the minimum temperature above 14 o C.
The greenhouse orientation is E-W; with crop rows aligns N-S. Cropping
conditions and crop management are very similar to those in commercial
greenhouses.
Climatic parameters are continuously monitored within the greenhouse.
Outside the greenhouse a meteorological station was installed, in which air
temperature, and relative humidity, solar and Photosynthetic active radiation
(PAR), rain detector, wind direction, and velocity measurements were taken.
The cover temperature sensors were located on the faces oriented to the east
(two sensors), and west (two sensors).
During the experiments, the inside climate variables were also taken,
among which stand out: air temperature, and relative humidity with a vent-
ilated psychrometer (model MTH-A1, ITC, Almeria, Spain), solar radiation
with a pyranometer (model MRG-1P, ITC, Almeria, Spain), and Photosyn-
thetic active radiation (PAR) with a silicon sensor (PAR Lite, Kipp–Zonnen,
Delft, The Netherlands). Among all the climate sensors installed in the green-
house, only solar radiation and psychrometer was used for the transpiration
virtual sensors.
Regarding, the daylight air temperature and humidity are controlled the
top and side windows controlled through PI controller [136]. Potentiometers
allow for knowing the window’s position in each control instant. The night
air temperature and humidity is controlled by the windows and the heating
system [136]. Setpoints of both systems are established at 24 o C [13], and 14
o
C for the ventilation and heating respectively. All the actuators are driven
by relays designed for this task.
All climatic data was recorded every minute with a personal computer. The
acquisition system is formed by two different National Instrument Compact-
Fieldpointsrconnected through Ethernet protocol.
For the growth model, it was necessary to know the evolution of leaf area
index. It was determined through the leaf area measurements of each plant
removed for biomass task, the pruning, and deleafing were also take into ac-
count. The biomass were made up of a destructive sampling of five randomly
3.1 Virtual sensors for designing irrigation 47
Fig. 3.5 Greenhouse facilities used for the experiences performed in this work. From
left to right and from top to bottom on the figure: Greenhouse, CO2 sensor, Solar
and PAR radiation, Heating system, Solar and PAR radiation inside the greenhouse,
and the tomato crop lines.
selected plants every 21 days, duration accorded in the research protocol. The
choice of 21 days is twofold: first, it was the sufficient time to find growth dif-
ferences; and second, to avoid the elimination of too much vegetal stuff in the
greenhouse which could end in a modification in the climate or transpiration
measurements.
In biomass process were measured against: number of nodes, leaf area,
number of fruits per bunch, fresh and dry weight of leaves, stem, and fruits.
The plant material and fruits were introduced into a drying oven where they
remained for 24-48 h (depending on the phenological state) at a temperat-
ure of 65o C. Based on this, the dry matter of leaves, stems, and fruits was
determined by analytical balance. The matter of leaves and secondary stems
pruning came from the selected plants for biomass while kept in production,
once removed from the plant, the fresh and dry weight was taken, such as bio-
mass. In the case of pruning, stems, and leaves are measured separately. Both
the bare and the pruning are carried out for the leaf area index measures,
executed, as in the biomass, through electronic planimeter (Delta-T Devices
Ltd).
Micro-lysimeter was the system chosen to take the transpiration measure-
ment in the present paper [128]; (Fig. 3.6). The device consists in two elec-
48 3 Water Balance
Fig. 3.6 Lysimeter installed in the greenhouse for the transpiration calculation.
Fig. 3.7 Testing the values of transpiration measured with daily consumption data
in a autumn-winter cycle. It represents the daily accumulated transpiration and the
daily average of the five trays with twelve plants consumption measurements
50 3 Water Balance
Fig. 3.8 Testing the values of transpiration measured with daily consumption data
in a spring-summer cycle. It represents the daily accumulated transpiration and the
daily average of the five trays with twelve plants consumption measurements
Once obtained the transpiration values, the next step was searching a substi-
tution of this system with transpiration virtual sensors based in the Penman–
Monteith (P-M). The equation combines the energy balance with the vapor
convective transport. In the last years, different physical and pseudophys-
ical virtual sensors (grey box) based in the P-M equation have been de-
veloped and tested by different authors. For this paper, the pseudophysical
P-M simplification proposed by [10] was chosen, because of the good res-
ults obtained by [105] for pepper crops in the same conditions. In this paper,
these good results were got fitting the parameters differentiating summer and
winter seasons and with the different crop development stages. Furthermore,
3.1 Virtual sensors for designing irrigation 51
a delay is reported between measured and predicted values for some particu-
lar conditions. The causes of the observed delay were explored and a climate
variables dependency was found as other authors assert ([50, 105]). On the
other hand, the proposed black-box dynamic virtual sensors would be used
to design events based fertirrigation controller. In order to use modern con-
trol algorithm, the use of dynamic virtual sensors joined system identification
technics are presented as a alternative to physical or pseudophysical virtual
sensors. The proposed virtual sensors incorporate the dynamics of transpira-
tion and will be of varying complexity, beginning with linear black-box virtual
sensors fitted to data. Non-linear virtual sensors based in system identifica-
tion was tried out obtaining good results. Nonlinearities will be introduced
in an ad-hoc form, relying on a priori knowledge about the system.
exists, it does not reach a significant quota, with which in the first analysis it
could be concluded that correlation would not be necessary, however, it is re-
commended. Also was demonstrated the presence of a phenomenon known as
delay, which is translated to an advancement of the dynamics of the estimated
values for real ones, as other authors asserted ([105, 50]).
To calibrate grey-box virtual sensor described in section 3.1.3.1, a growth
model is required to try out the LAI estimations. The simplified model TOM-
GRO [128] rises as an option to remove the complexity of the full virtual
sensor proposed by [75] and make them available online control systems while
retaining their physiological characteristics [76]. The parameter that influence
the dynamics of the XLAI was calibrated and validated, first by [128] and
later by [48] in the same greenhouse for tomato crops.
The first processing step was to eliminate the middle and both trends
of inputs and outputs. It is worth noting, XLAI was not introduced into
the system as an entry. The reason for this is that the rate of leaf area
index remains constant on the same day, lacking the dynamic of remaining
input variables. To reach this conclusion many tests with the Matlab toolbox
was realized. First, XLAI was introduced in the virtual sensor as a regressor
(XLAI (t − i)) obtaining a bad fix in the output. A regressor based in the
equation of solar radiation reaching a given depth in canopy was tried out,
eq. (3.6), but with the same results, such results was the expected [10]:
where VGR is the global radiation reaching the crop (W m−2 ); −Ck is
the extinction coefficient of radiation (unitless); and XLAI leaf area index
(m2crop m−2
soil ). As noted above, LAI remains constant during a chosen day
as the parameter (Ck ), that means the regressor is constant during a day,
depending exclusively of the radiation.
This virtual sensor was also evaluated by using an estimated VDP as a
function of the inside temperature and the relative humidity. Despite this
estimation, the virtual sensors obtained with these two variables do not im-
prove those calibrated only using VPD. In the end, the radiation and vapor
pressure deficit remains as the unique inputs.
In order to obtain the number of terms in the regressors, it is necessary
to obtain a model trying out different combinations. Besides the number
of terms obtained, the non-linearities was estimated through the different
possibilities: wavelet network, tree partitions, and sigmoid network. Of all this
ways to obtain the non-linearities were tested, the wavelet network remained
with the best fit joining a non-linear block and a linear block. The Table 3.3
shows the parameters of the resulting virtual sensor.
54 3 Water Balance
All the available data has been used for the validation, more than 1 million
for each variable. A total of 9 different Spring-summer and Autumn-winter
seasons was used. Fig. 3.9, 3.10, and 3.11 show an example of the results
obtained in the validation process of grey-box, and linear and non-linear
dynamic virtual sensors in the different cycles. The virtual sensors validation
can be seen in the Fig. 3.9. The presence of a phenomenon known as delay
is demonstrated in the Fig. 3.10. In the end, the Fig. 3.11 samples a day
detailed with the three virtual sensors. In some instances, system dynamics
is not well captured by the virtual sensors, as happens in the Fig. 3.10. This is
caused by the difficulty in transpiration calculation using the micro-lysimeter.
Furthermore, as in the Fig. 3.9, the transpiration has a similar behaviour to
the sunlight; rising during the morning, falling down in the afternoon, and
remaining almost constant at night.
The grey-box virtual sensor has a good behavior in the dynamic, even
though there exists a small overestimation of the estimated values in the
start and the end of the crop seasons. Even though this overestimation really
exists, it does not reach a significant quota, with which in the first analysis
it could be concluded that correlation would not be necessary, although it
is recommended. Furthermore, it was demonstrated the presence of a phe-
nomenon known as delay (Fig. 3.10), which translates to an advancement of
the dynamics of the estimated values for real ones, as other authors asserted
[50, 105].
3.1 Virtual sensors for designing irrigation 55
Fig. 3.10 One full day with delay in 2007-2008’s autumn-winter. The figure starts
at midnight.
light ones. The transpiration is so low that it is difficult to take it with the
micro-lysimeter due to the sensor precision, 1 g.
Moreover, this dynamic virtual sensor does not show the anticipation phe-
nomenon showed in the static virtual sensor. It is the resistance of the plant
transpiration resulting in a delay of about the same processes that produce
it, graphically shown as a delayed action on transpiration estimated.
For all the seasons included in this work (from 2004’s Autumn to 2008-
2009’ Autumn), the virtual sensor’s goodness was obtained. This goodness for
a data series is calculated through the minimum mean square error (MMSE).
In all cases the dynamic virtual sensor obtained good results (MMSE<6%),
as can be seen in the three validation errors Table 3.4 and Fig. 3.12.
In the tables (Table 3.5, 3.6, and 3.7) show a full review of errors of each
model.
3.1 Virtual sensors for designing irrigation 57
Fig. 3.12 Virtual sensor error comparison, minimum mean square error in g
m−2 min−1 .
3.1.6 Conclusion
Significantly, the research was conducted over three years, for a total of eight
cycles of cultivation in those which took the different climatic and physiolo-
58 3 Water Balance
gical variables of a tomato crop. All this data was taken at intervals of one
minute, making a total of approximately over one million data for each vari-
able. The main objective, as noted at the beginning of this paper, is the
transpiration virtual sensor of tomato crop for the design of irrigation control-
lers. For this occasion we had a measurement system based on transpiration
weighing the difference that occurs on a scale from one moment to another,
and to measure water loss with a good approximation.
The next task was to seek a virtual sensor that would replace the system
based on the weight difference caused by the loss of water by the crop. One
based on sensors that are typically installed in greenhouses; temperature,
humidity, and solar radiation. The aim is to reduce total installation costs
and to avoid the constant maintenance the scales require.
After preliminary assessment of some of these virtual sensors. Non-linear
ARX had a better fit and, in the end, was the best election for the irrigation
virtual sensor proposed. This virtual sensor has good results in the calibration
and validation of the virtual sensor. An average error of 5%, for all cycles
taken, shows how the choice of this virtual sensor was successful. But it also
presents some problems, such as the overestimation at night occurs. On the
other hand, linear and non-linear black-box virtual sensors have demonstrated
the absence of a advance phenomenon. It was translated to an advancement
of the dynamics of the estimated values for real ones.
System identification techniques was chosen to obtain a dynamic virtual
sensor. M atlab rsoftware package reached as a good option to work with the
identification technics. A large number of different non-linear dynamic virtual
sensors structures was tested. The proposed dynamic virtual sensors require
only two inputs; global radiation and vapor pressure deficit, thus eliminating
the inclusion of the XLAI . This virtual sensor presents the possibility of using
modern control algorithms that cannot be used with the grey-box sensor.
As summary:
3.1 Virtual sensors for designing irrigation 59
• Grey-box virtual sensor has a good fixing as advantage, but some disad-
vantages overestimation at different moment of the year, worse final error
result, and the advance phenomenon.
• The black-box virtual sensors obtain better results. Also allow to eliminate
the grey-box problems: advance phenomenon and overestimation. Only
appears an overestimation at nocturnal periods.
• Non-linear black-box had the best results.
This paper has dealt with the transpiration from an industrial point of
view, as a process in which there are entries and exits. The crop itself, and
some aspects of the climate inside the greenhouse, are considered disturbances
affecting the dynamics. This shifts the focus away from the strictly agronomic,
agronomy classic, which studies the exchanges that occur in the greenhouse
by static virtual sensors based on fundamental principles, without including
the system dynamic effect.
Acknowledgments
Abstract The main objective of this work has been the development of a
virtual sensor (VS) based on the water balance dynamics of tomato crops in
greenhouses, using coconut coir substrate as an example in this case. Such
sensors are used to provide a viable and economical alternative to expensive
or impractical sensors. The final virtual sensor is the result of combining:
crop growth, where the water is distributed in structural and non-structural
plant biomass (storage); the substrate, which is considered to be composed
of a single layer; and the water loss caused by transpiration and drainage. In
this work, two ECH2O-EC-5 sensors (Decagon Devices) were calibrated and
then used to determine the substrate water content; and a microlysimeter
was installed to continuously sample the water supplied, the drainage, and
the crop water loss values. The VS took into account the water supplied, the
amount of water in three stores (the substrate, the root and the aerial part of
the crop), the climate and the water loss. The water dynamic was determined
by system identification techniques and by physical virtual sensors, which
considered the water balance from a holistic point of view - as a sub-model for
a customizable interface between crop growth and the plant ecosystem. The
influence of both the crop and the climate conditions on the water balance
was analysed and the virtual sensors were evaluated giving good final results.
3.2.1 Introduction
possibility of another virtual sensor that estimates crop water loss through
transpiration - as developed by the research group members [148].
Water balance has been studied and described in terms of the energy state of
water, which allows one to describe the flow rate through the soil [20] or sub-
strate [85]. In this kind of flow, water velocity is proportional to the potential
gradient [20], where the proportionality constant is a substrate characteristic;
thus, the same potential gradient will produce different flows in different sub-
strates [20]. Hence, the water potential in the substrate (ψh,s ) is the sum of
the potential:
64 3 Water Balance
where ψm,s is the matric potential, ψg,s is the gravitational potential and
ψos,s is the osmotic potential. In this work, the gravitational potential (ψg,s )
can be ignored given that the substrate height is less than 7 cm [20]. The
water potential remains as a function of the matric potential and osmotic
potential:
ψh,s = ψm,s + ψos,s (3.8)
In addition, for substrate characterization, one of the important factors is
the water retention capacity (Kirda et al., 2004), which depends on physical
characteristics: porosity, structure, particle size and distribution [4, 20]. Sub-
strates with a particle size of between 1 and 10 mm have little variation in the
amount of water stored but this capacity increases when the particle size is
less than 1 mm [4]. The water retained in the substrate has a nonlinear rela-
tionship to the matric potential and also presents the hysteresis phenomenon
[20]. Knowledge of this relationship has been the subject of many studies
[36, 20, 38, 181] and is useful in formulating water balance virtual sensors.
The proposed model is a generic one, which considers the water balance
from a holistic point of view. It is dynamic, explanatory and simple. It takes
into account the amount of water in three stores: the substrate, the root and
the aerial part of the crop (including leaves, stems and fruit). The state vari-
ables are the substrate (XAG,s ), root (XAG,r ), and the water content of the
aerial part XAG,d . This virtual sensor (VS) can be changed to accommodate
different conditions [177]:
where FAG,r−d is the root to crop water flow, FAG,d−atm the water flow
from crop to atmosphere, and FAG,s−r from substrate to root:
where, ψh,s , ψh,r and ψh,d are the substrate, root and aerial part of the
plant potentials, respectively. rAG,s−r is the flow resistance of the substrate
to the root, and gAG,r−d is the root to crop conductivity. rAG,s−r is defined
by eq. 3.25, in which Csors and Crsr are parameters that define the resistance
between the substrate and the root, Ckwrsr is a root resistance parameter,
Kse is the substrate hydraulic conductivity, and r is the root density.
where, Cagm is the form parameter in the water retention curve, and KrSe
is the hydraulic conductivity relative to the substrate:
where Cag , and Cagn are the water retention curve form parameters. The
different water potentials are defined as follows:
where Cbs is the calibration parameter, (θs ) the relative water content
of the substrate and (θas,max ) the substrate capacity. The potential at the
root and crop are defined by the osmotic (ψos,r , ψos,d ) and pressure potential
(ψpr,r , ψpr,d ) through the following expressions:
where CR is the universal gas constant, Xpr the root dry matter, Ca is
the water density, and Xta,k the air temperature in Kelvin. fneoa,r is the
osmotically-active fraction in the roots, S the stored total mass, Cpr,r a para-
meter that affects the pressure component of water potential in the root and
Cε the cell wall rigidity parameter. Likewise, the crop potential is defined as:
where, Xps is the crop dry matter and fneoa ,d is the fraction of osmotically-
active storage in the canopy described by [177]:
where Cf a,r is the osmotic fraction ratio and the root molar mass, and
Cf a,d the osmotic fraction ratio and the crop molar mass. To estimate the
soil water flux, it is necessary to calculate which part of this water is lost to
drainage, thus the following equation is used:
where, fdr is the percentage of irrigation water lost to drainage. Water flux
to the atmosphere, FAG,atm , was described by [148] using a virtual sensor
based on the Jolliet and Bailey equation [74]. They observed that tomato
3.2 Water content virtual sensor 67
crop transpiration depends on the radiation (VSR ), the vapour pressure deficit
(VV P D ) and the crop growth state defined by XLAI - thereby simplifying the
PenmanMonteith equation [198]. This virtual sensor accurately predicts the
water flux produced between the plant and the atmosphere in tomatoes, and
can be estimated using the following equation:
where C is the latent heat of vaporization (kg C−1 ), Ck is the light ex-
tinction coefficient for crops (this is related to the leaf inclination angle and
the leaf arrangement with regard to the XLAI , providing an indication of the
plants efficiency in intercepting solar radiation), XLAI is the leaf area index
(m2 [leaves] m2 [soil]), VV P D is the vapour pressure deficit (kPa), and VSR is
the global radiation reaching the crop (Wm2 ). The coefficients CA (unitless)
and CB (kg m2 h1 kPa1 ) are constants that are dependent on the crop. To
obtain more reliable results from the virtual sensor, the parameter CB is ob-
tained for diurnal (CBD ) and nocturnal (CBN ) periods through calibration.
where y(t) is the water content, u(t) is the input variable (this variable will
be represented by a vector in case of multiple inputs) and e(t) is the estim-
ation error. A(z), B(z), C(z), D(z), E(z), and F (z) represent the matrices
determined by means of off-line parameter identification methods and defines
the output (water content), inputs, and the estimation error:
68 3 Water Balance
y(t) = f (y(t1), y(t2), y(t3), ..., u(t), u(t1), u(t2), ..) (3.38)
where y(t) is the output variable in t time; u(t−j) and y(t−j) are the past
values of the input and output variables (regressors); and f is the nonlinear
function. The current water balance value is predicted as a weighted sum of
past values, along with current and past inputs values. With this information
the equation becomes [91]:
y(t) = [a1 , ..., ana , b1 , , ..., bnb ][y(t1), ..., y(tna ), u(t), ..., u(tnb 1)]T (3.39)
where y(t1), ..., y(tna), u(t), ..., u(tnb 1) are the regressors, the so-called
delayed inputs and outputs. Nonlinear ARX regressors can be both delayed
inputoutput variables and more complex nonlinear expressions of delayed
variables. The nonlinearity estimator block maps the regressors to the vir-
3.2 Water content virtual sensor 69
tual sensor output using a combination of nonlinear and linear functions. One
of the characteristics of the calibration of the linear and nonlinear black-box
virtual sensor is that the first validation is done during the identification,
since the trial and error processes must be carried out to choose the virtual
sensor.
XLAI /dt = ρδe(β(Xn −Nb )) /(1 + e(β(Xn −Nb )) ), if XLAI < CLAI,max (3.40)
where p1 is a parameter that weighs the loss of dry matter per node after
it reaches the maximum leaf area (CLAI,max ) and CRn is a function that
defines the crop growth. Hence, the estimation of the total dry weight must
be modulated by the effects of the electrical conductivity (fEC , [97]):
XP Se = f ECXP S (3.44)
where XP R is the root dry weight, and fR is the root growth fraction.
The research data used in this work have been obtained from greenhouses
located in the Experimental Station of the Cajamar Foundation, El Ejido, in
the province of Almerı́a, Spain (2o 43’ W, 36o 48’ N, at a 151 m elevation). The
tomato crops are grown in a multi-span ”Parral-type” greenhouse (Fig.3.14).
The greenhouse has a surface area of 877 m2 (37.8 x 23.2 m) with a polyethyl-
ene cover. The daytime air temperature and humidity are managed using the
top and side windows, which are controlled via a PI controller [136]. In addi-
tion, the biomass-based heating system allows one to control the night-time
air temperature [146]. Setpoints for both systems are established at 24 o C,
and 10o C for the ventilation and heating, respectively.
Fig. 3.14 Greenhouse facilities used for the experiments performed in this work.
From left to right and from top to bottom: the greenhouse, the droppers and EC-
5 sensor, Solar and PAR radiation sensors, the irrigation system, Solar and PAR
radiation sensors inside the greenhouse, and the microlysimeter.
3.2 Water content virtual sensor 71
different cycle to that used for validation), the XLAI and XP S real data were
taken every 21 days. The obtained daily rate value remained constant for
every minute of that day, thus allowing one to estimate the hydric balance
using irrigation and climatic data. The minimum value of the root mean
square error was used for calibration, trying different values sequentially in
each iteration. The parameter estimates are shown in Table 3.12 in Section
3.2.4. For the leaf area index calculation, cultural activities were taken into
account, included in the model estimation as a simple subtraction of the
eliminated XLAI value.
For the virtual sensors calibration, all the data gathered minute-by-minute
during the months of October 2011 and May 2012 were used and separated
into groups for validation and calibration. In the first phase, the search space
of the physical sensor parameters was determined using an iterative sequen-
tial algorithm to minimize a least-square-error criterion between the real and
the estimated water content (sequential iterative algorithms). In the second
phase, a calibration process based on genetic algorithms, the parameters were
obtained (Tables 3.10 and 3.11). Conversely, the proposed black-box dynamic
virtual sensors would be used to design advanced controllers. In order to
employ a modern control algorithm, joined dynamic virtual sensors were
presented as an alternative to physical or pseudo-physical virtual sensors.
The proposed virtual sensors incorporated transpiration dynamics and were
of varying complexity, beginning with linear black-box virtual sensors fitted
to data using system identification techniques (see Table 3.8).
Different substrate samples were taken and placed in an oven at 105 C, for
24 h. Then, these samples were used to obtain the EC-5 calibration curve. For
this, known water amounts were added to the dried substrate. Furthermore,
the sensor readings were recorded continuously, allowing one to relate the real
water supplied with the sensor output. This test was repeated five times to
obtain a wider data range. Additionally, in order to test the sensor response to
electrical conductivity, the same assays, known water amounts, were carried
out but with different electrical conductivities (0, 1, 3, 5, 8, 12 and 15 dS
m−1 ). Five different assays were conducted for each EC value (35 in total).
Fig. 3.15 shows the calibration of the two ECH2O-EC-5 sensors used in this
work. The calibration curve shows how the sensors exhibited linear behaviour
in measuring the water content. It is important to notice that this linear
regression was carried out with an EC below 8 dS m−1 .
As shown in Fig. 3.15, and 3.16, the EC-5 sensors showed good perform-
ance in measuring the substrate water content, but had the drawback of
being sensitive to CE from 8 dS m−1 (the same value as indicated by the
manufacturer), which is not a problem for greenhouse crops given their salt
limits are below these levels. Once calibrated, the sensors were installed in
the greenhouse.
The next step was the validation of the estimated water consumption (Fig.
3.17) obtained by the microlysimeter data (irrigation supplied and drainage
water), and the real water consumption measured, which is explained above:
74 3 Water Balance
Fig. 3.15 Calibration curve for the water content sensors Moreover, calibration
curves were plotted using water with different electrical conductivities (0, 3, 5, 8,
12, 15 dS m-1, Fig. 4).
Fig. 3.16 Calibration curve for the water content sensors using different EC values
The lineal regression results show a high R2 value and a slope with a value
close to one, which would mean that for an instant t, both values are very
3.2 Water content virtual sensor 75
TOMGRO validation was carried out with the data from the same cycle
as that obtained for the climatic and water content. In view of the results
shown in Fig. 3.18a and 3.18b, it can be concluded that the TOMGRO model
quite closely approximates what happens to the XLAI dynamics throughout
the crop cycle, as can be observed on the days when the XLAI real data were
taken (marked with a square). In addition to this, Fig. 3.18a shows the steps
produced by different prunings being carried out; these are expressed as a
significant step due to the XLAI being removed in such a process. Fig. 3.18b
displays the accumulated XLAI , in which the pruning are not considered, but
are added to the final result. In conclusion, the leaf area index was estimated
acceptably. Both figures clearly show that the simulation behaviour is closer
to the measured data. It is important to note that the crop was transplanted
late, in October, when the temperature is low for crop adaptation and growth
is slower that in a normal cycle starting at the end of August. This trans-
planting delay could have provoked the overestimation in the estimated data.
Fig. 3.19 shows the results of the total dry weight (g m−2 ) obtained for
the validation season. Real measurements were taken from four groups of
five plants in the greenhouse (+), and are shown along with the average
76 3 Water Balance
value of these measurements (o). The results properly describe the plant
growth, although an overestimation is presented, especially in the central-end
period of the growing season. This was expected, as were the XLAI dynamics
observed due to the transplanting date. The root dry weight estimated by
simulation is shown in Fig. 3.19b. The absence of root dry weight real data
meant that it was estimated as a function of the total dry weight (fR ).
The models were validated on different data sets from those used for calib-
ration and identification. The proposed virtual sensors incorporate the water
content dynamics. Nonlinear virtual sensors, based on system identification,
were tried out obtaining good results. Nonlinearities were introduced in an
3.2 Water content virtual sensor 77
Fig. 3.20 Climatic and water balance data for three days chosen as an example
Fig. 3.21 relates to the estimated virtual sensor and measured water con-
tent under different conditions. The results obtained show how all the VS
exhibit good behaviour when the crops are photosynthesising. This is a crit-
ical period in which the substrate water content is influenced by the irrigation,
drainage and the crop assimilation. All of this leads to continuous changes in
the hydric balance. Nonetheless, the dynamics were simulated satisfactorily.
Fig. 3.22 shows the dynamics on a selected day; it can be seen how different
VS capture all the changes produced in the substrate water content. Finally,
78 3 Water Balance
the validation carried out presents satisfactory results in general terms, since
the different VS demonstrate acceptable behaviour in periods when the water
content varies little each day or night; and even present correspondingly good
dynamics when changes are greater. Moreover, when the model behaviour
does not approach the measured data, this is primarily due to a significant
change occurring at the first and second irrigations of the day, when the
irrigation frequency increases.
Table 3.9 contains the errors obtained between the simulations of the three
VS and the measured data. The goodness-of-fit for the three data series is
3.2 Water content virtual sensor 79
calculated using the minimum mean square error (MMSE), the maximum
error, and the mean error. In all cases, the virtual sensor obtained good
results (MMSE<8%), as can be seen in the three validation errors (Table
3.9). Although the average error was low, around 8 %, in 84 % of the data,
the absolute error was less than 3 % of water content expressed in volume.
Virtual XLAI Time Virtual sensor Max Mean σ MMSE Water content
sensor intervals series error error interval
0 - 0.7 48.254 Physical VS 1.96 0.844 0.732 0.736 [66.1-72.3]
Physical 0.7 – 1.5 51,323 Physical VS 1.52 0.778 0.711 0.699 [66-72.6]
> 1.5 72,554 Physical VS 1.67 0.761 0.673 0.691 [64.9-75.2]
0 - 0.7 48.254 ARX441 2.01 0.731 0.758 0.654 [66.1-71.8]
Linear 0.7 – 1.5 51,323 ARMAX33211 0.74 0.702 0.698 0.621 [66.3-72.5]
black-box > 1.5 72,554 ARMAX34211 1.18 0.687 0.757 0.598 [65.3-75.7]
0 - 0.7 48.254 NARX331 1.11 0.779 0.621 0.681 [66.4-71]
Non-linear 0.7 – 1.5 51,323 NARX231 0.98 0.644 0.789 0.563 [66.2-72.3]
black-box > 1.5 72,554 NARX441 1.06 0.578 0.614 0.477 [65.3-75.8]
Table 3.9 Virtual sensor error calculation (% volume, water content)
Finally, the NARX obtained a good fit due to the nonlinear component,
which adds complexity to the virtual sensor. In this work, this nonlinear
component was added to the water content virtual sensor as a result of the
strong nonlinear behaviour in the system inputs, and got the best fit (see
Table 3.9). The linear black-box VS also obtained good results, even though
it was a less sophisticated VS. The first-principle-based VS uncertainties and
the parameter calibration problems could be the cause, whereby the VS res-
ults obtained were substantially improved by the empirical VS. These results
can be checked looking at Table 3.10.
3.2.4 Conclusions
Three virtual sensor validations have been presented with satisfactory res-
ults in general terms, since the different VS exhibited acceptable behaviour
during these periods. The water content showed little variation day-to-day,
and the simulated water content obtained good results; the average error
(MMSE) was low, at around 8 %, also presenting good dynamics. Thus, the
resulting virtual sensors are considered as powerful tools to be used for control
design purposes in irrigation systems.
This paper has dealt with the hydric balance from an industrial point of
view, as a process in which there are entries and outputs. The crop itself,
and some aspects of the climate inside the greenhouse, are considered as
disturbances which affect the dynamics.
Acknowledgements
Appendices
3.2 Water content virtual sensor 81
83
84 4 Crop growth modelling and simulation
4.1.1 Introduction
The greenhouse is ideal for farming because these variables can be manip-
ulated to achieve optimal growth and plants development [129]. Crops use
solar radiation, CO2 the concentration in the surrounding air, water and
nutrients to produce biomass (roots, stems, leaves, and fruits) through the
photosynthesis process. Moreover, the greenhouse crop growth can be simpli-
fied, assuming that plants are healthy, and that receive adequate water and
fertilizer at all times [75]. Thus, crop development is primarily determined
by climatic variables of the environment. So, the growth may be managed,
controlling the main climate variables: temperature, humidity and CO2 [148].
Thus, the climate control inside the greenhouse is important for veget-
ative growth and for fruit production. In the mild climate countries, night
temperature is controlled by heating system [136] and daylight temperature
control is performed by convective air exchange between the outside and in-
side [13]. This exchange rate coupled with CO2 taken by the crop during
photosynthesis determine the concentration of CO2 in the greenhouse. When
photosynthetic rates are higher, the concentration of CO2 falls below the at-
mospheric producing a growth deficit that is increased when the crop reaches
its maximum development [142]. Some authors propose different techniques
4.1 Growth model for sweet pepper 85
are drawn. Finally, an annex section have been added with information and
values of the parameter and variables.
The model calibrated and validated in this paper is based on the works of
[100], who presents a mechanistic dynamic model validated in six experiments
in greenhouses in France and The Netherland. It is built on the generalist
model of [54], called INTKAM and is derived from the cucumber growth
model develop by [98]. Leaf gross photosynthesis is estimated with model
of [44] and stomatal conductance according to [118]. Crop photosynthesis is
4.1 Growth model for sweet pepper 87
In the following sections are shown all the model equations. Crop photosyn-
thesis (section 4.1.2.1) is estimated from leaf photosynthesis (section 4.1.2.1)
and leaf area index (section 4.1.2.1). Maintenance respiration (section 4.1.2.2)
uses the different organs dry wight to calculate the carbohydrate losses. Crop
dry weight (section 4.1.2.3) is calculated from photosynthesis and respiration,
and different organs dry weight estimation (section 4.1.2.4) is estimated from
fractions of the total daily dry weight.
For this work, the time step of calculation of crop photosynthesis and
maintenance respiration is one minute, while the step time of calculation of
dry matter production is one day. The plant dry matter is divided between
different plant organs according to growth factors.
Leaf Photosynthesis
The leaf photosynthesis (mol m−2 s−1 ) is calculated with the LAI and
climatic variables. The equation of [100], based in the works of [44], is used
to determine the assimilation of CO2 by the enzyme Rubisco (ribulose-1,5-
bisphosphate carboxylase oxygenase):
Therefore, three equations are shown, one for each variable (Vc , Vq , and
Vs ,). The leaf photosynthesis will take the minimum value of them which will
be determined by the most limiting factor in each moment (4.2).
This enzyme Rubisco has a dual behavior, catalyzing two opposing pro-
cesses. The first process, called carboxylation (Vc ), is the CO2 fixation to
organic form. In contrast, the plant O2 fixation process (photorespiration), is
that through which a part of the assimilated carbon is lost. In the equ. (4.3),
the assimilation of CO2 by the enzyme Rubisco is determined as:
∗
Vc = Vcmax C+KC−Γ
(1+ O
) (4.3)
c K0
∗
C−Γ
Vq = eq αIf C+2Γ ∗ (4.4)
where eq is the light quantum efficiency (µmol mol−1 ); α is the leaf absorption
for PAR (-); If is PAR flux incident on a unit leaf area (µmol m−2 s−1 ); and
the intercellular concentration of CO2 , C, is calculated from the concentration
of CO2 in the greenhouse.
Besides the limitations of light and the ability of the enzyme Rubisco,
[33] includes another limiting factor; the accumulation of photosynthetic
products. [33] asserts the rate of production of a product decreases when
the concentration product increases. This limitation was assumed simply re-
ducing the capacity of the enzyme Rubisco (Vs ) in a half [174]:
90 4 Crop growth modelling and simulation
Vc
Vs = max
2
(4.5)
X
LAIini
XLAI = XLAImax ( XLAI )e−CpLAI ∗XT S (4.6)
max
where X LAImax is the maximum value that LAI can take; X LAIini is the initial
value of LAI; C pLAI is a tuning parameter of the Gompertz function; and
finally, X T S , is the thermal time taken by the plant (0 C day−1 ). This thermal
time is calculated like the accumulative sum of diary average temperature
minus a base temperature of ten degrees.
dWT
dt = fc (Pcrop − Rm ) (4.9)
Once the total dry weight is estimated, the weight of each different plant
organs can be obtained. The diary total dry weight will be multiplied by the
diary fraction of assimilates directed to each organ. Therefore, the sum of the
four fractions always has to take the value of 1.
Daily fruits dry weight (Wf ; gDW m−2 day−1 ) and diary roots dry weight
(Wr ; gDW m−2 day−1 ) are estimated multiplying the diary total dry weight
(dWT /dt) by the diary fraction of dry weight for fruits, and roots. This is
represented by equ. (4.10) for fruits dry weight and equ. (4.11) for roots dry
weight [56]:
dWf
dt = ff dW
dt
T (4.10)
dWr
dt = fr dW
dt
T
(4.11)
where WT is the total dry weight of the plant, and fr and ff are the fractions
of the total daily dry weight to be led to the fruits and roots, respectively.
92 4 Crop growth modelling and simulation
The leaves (Wl ) and stems (Ws )dry weight are calculated in the same
way that fruits and roots. The fractions of total dry weight for leaves and
stems (fl and fs ) have been calculated as linear correlations on function of
the fraction of the fruits, omitting the initial phase of the vegetative growth
of the plant, as follows:
dWl
dt = fl dW
dt
T
(4.12)
dWs
dt = fs dW
dt
T
(4.13)
[56, 57] observed that the sweet pepper crop is characterized by periodic
fluctuations in fruit production, even under constant environmental condi-
tions. These temporal variations in production can be approximated by a
sinusoidal function, f , to determine the daily fraction of the total dry weight
that is directed to the fruits and also to the roots. This function takes diffe-
rent values for fruits and roots obtaining two functions which fit satisfactory
and follow an opposite trend but a similar amplitude:
f = a + bsin( 2π(days+c)
d ) (4.14)
where the a parameter represents the average fraction of dry weight in fruit
on the entire production cycle. The b parameter is the amplitude of variation
of the values of the fraction devoted to fruits and roots. The parameter c
represents the phase of the sine (in days). Parameter d can be approximated
as the lifespan of the fruit. Finally, the variable days corresponds to the
number of days after planting.
The assumption of sinusoidal functions identical but opposite in phase to
the fruits and roots implies that the sum of the fractions involved in stems
and leaves is constant, and therefore, the sum of the four fractions (fruits,
leaves, stems and roots) must be equal to 1, to fully distribute the weight
between plant organs.
Once the total dry weight of the plants is obtained, the organs dry weights
are obtained multiplying by the fraction. Fig. 4.3 shows the dry weight frac-
tions of the different organs involved. It is noticed that the fractions of fruits
and roots are opposite and equal in amplitude, being regulated by the sinus-
oidal function f . In the initial period of vegetative growth of the plant dry
matter is distributed only among the roots, stems, and leaves, due to fruits
growth is delayed.
Fig. 4.4 shows the fractions of leaves and stems (fl and fs ). These fractions
have been estimated as linear correlations on function of the fraction of the
fruits, omitting the initial phase of the vegetative growth of the plant.
4.1 Growth model for sweet pepper 93
The research data used in this work was located in the Experimental Sta-
tion of Cajamar Foundation (El Ejido) at the province of Almeria in SE
Spain (2o 43’ W, 36o 48’ N, and 151 m elevation). The crops were grown in
94 4 Crop growth modelling and simulation
The validation process includes data taken in a third greenhouse used in the
same period of time, also located in the Experimental Station of Cajamar
Foundation, and with the same characteristics.
Fig. 4.5(a) represents the LAI obtained by the model and the Leaf Area Index
data sampled from measurements taken in the greenhouse for six days during
the season.
The results describe correctly the LAI increasing, but with a little over-
estimation. The use of more data for the calibration could produce better
adjustment of parameters of Gompertz function and better results.
The model estimates the gross photosynthesis with the climate data and LAI,
and the maintenance respiration in order to obtain total dry weight of the
day. Fig. 4.5(b) shows the results of total dry weight (g m−2 ) obtained for the
validation season. Actual measurements are taken at four groups of plants in
the greenhouse, and are shown along with the average value of these meas-
urements. The results describe properly the plant growth, although presents
an underestimation, especially in the central-end period of the season. This
fact was expected, as was also observed in the two calibration seasons.
96 4 Crop growth modelling and simulation
(b) Total Dry Weight, (g m−2 ). (c) Fruits Dry Weight, (g m−2 )
(d) Total Fresh Weight, (g m−2 ) (e) Fruits Fresh Weight, (g m−2 )
The fruit dry weight estimated by simulation is shown in Fig. 4.5(c). The
figure represents the real measurements (four samples and the average value)
and the estimated dry weight. Fig. 4.5(c) shows that an underestimation
is obtained in the middle-end of the season, although the final simulated
results fit to the real data. The fruit dry weight makes it possible to obtain
an approximate value of greenhouse performance which is a valuable data for
growers.
The results of leaves and stems dry weight have not been validated, since
no real measurements were available for these weights. Only the final value
was sampled for leaves dry weight, 350 g m−2 . The estimated value for leaves
dry weight obtained for the season was around 330 g m−2 , so this value was
slightly underestimated. The final stems dry weight was situated in 348 g
m−2 , and it is well-estimated by the model (350 g m−2 ).
4.1 Growth model for sweet pepper 97
The fresh weight is the actual weight of crop or fruits, being the sum of the
dry weight plus the water content of element in question. Therefore, it is the
main result for determining the crop profitability. The content of water in
the plants and fruits is not constant.It will vary depending on the maturity
or stage of development of the crop growth. The content of total fresh weight
and fruit fresh weight based on dry weight content along the season has been
obtained empirically.
Fig. 4.5(d) shows the results obtained by the simulation for the total fresh
weight during the validation season. An underestimation of the values for
much of the season is observed. Correct fresh weight values are obtained only
at the beginning and at the last stage of the season. The same behaviour is
obtained for the calculation of fruit fresh weight (Fig. 4.5(e)).
The Table 4.1 shows a full review of errors of each model. the goodness
of fit for the Leaf Area Index, Total and Fruits Dry Weight, and Total and
Fruits Fresh Weight was obtained. This goodness for a data series is calculated
through the relative mean error (RME). An average error between 4 to 9 %
for fruit dry and fresh weight, LAI and total dry weight, shows how the model
calibration was successful. However, it also presents some problems, such as
a underestimation in the middle-end of the season. These errors could be
produced by the lack of a richer and bigger group of calibration data.
The table 4.1 shows promising results for the fruits dry weight, which is
the most important one from an economic point of view. The values of dry
weight of leaves and stems are not very meaningful for the grower and their
decisions over the greenhouse climate conditions.
The results obtained by the crop growth model can be useful for support the
growers in the decision-making in the management of the greenhouse climate
in order to keep an optimal crop growth.
98 4 Crop growth modelling and simulation
The software has been developed using C++ programming language under
the Integrated Development Environment C++ Builder [67]. The resulting
software shows four different parts (plus data and results) can be distin-
guished in the development framework:
(b) Climate data introduction in- (c) Growth data introduction inter-
terface. face.
Moreover, for climate data insertion have been considered diverse options
according to the availability of climate data or the users purpose (Fig. 4.6(b)):
100 4 Crop growth modelling and simulation
• Use of the climate data pattern stored in the system. In this case, there is
no need of data introduction since the pre-established data in the software
will be used.
• Climate data introduction through a plain text file (with extension .txt).
This file has to contain the values of temperature, radiation and CO2
concentration in rows. The values must be separated by blank spaces or
tabs and each row represents the samples collected in a specific time. Before
uploading the file, the time step between samples has to be marked on the
application.
• Manual introduction of climate data. The last option is to type the climate
data values in tables. The time step between samples can be a day, a week
or the whole season. Moreover, two values have to be introduced for each
sample, one of them for daytime average value and other for nighttime
average value. Another interesting option is that the radiation can be
generated automatically using the solar radiation model.
In addition, the user can introduce values of solar radiation outside the
greenhouse or indoors. In the first case, the use of shading meshes and the
greenhouse transmissivity coefficient have to be indicated for the estimation
of the radiation inside the greenhouse.
As commented above, the real growth data available can be inserted into
the system in order to be compared graphically with the results calculated
by the model. The system accepts the growth data of total dry weight, fruits
dry weight, total fresh weight and fruits dry weight. The growth data input is
carried out by means of a plain text file or writing the values in tables where
the time step is a day. Some daily values can be ignored leaving blank the
corresponding cell in the table or writing the value -1 in the case of using a
plain text file.
Furthermore, the growth model application permits the modification of
the values of model parameters (Fig. 4.6(d)). This is very useful for analyzing
each one of the parameters and readjusting its values. The software shows
the parameters together with their units, value range and a short description.
Finally, the simulation execution presents several options. The simulation
can run in a single step for the whole season or in daily steps. It also allows to
define the time step as a range of dates or a number of days. After finishing
the simulation, results are shown graphically. Fig. 4.6(e) shows results of a
simulation as example.
4.1 Growth model for sweet pepper 101
Growers are starting to see the advantages to install climate control systems.
These tools are tedious and complex and growers are complained about the
difficulty and amount of time it took to solve the tasks. The development of
a decision support tool was aimed for in order to choose the right applicable
climate control setpoints. The DSS is provided to help growers better under-
stand the effect of the different variables on growth. The different options
provide to introduce climate data make this DSS an interesting and valid
tool to be used by growers and researchers. Theses options allows users to
try out different combinations of data and to compare results. The tool dis-
plays valuable information for growers through graphics and results reports.
The first step in the DSS is to define the simulation. In this stage, the main
information about the greenhouse localization and size, and crop season dur-
ation and density are introduced. For this example, four different strategies
were tested (Table 4.2). The climate data was introduced from greenhouse
data (GD), constant values for day and night temperature or CO2 concen-
tration were modified in the greenhouse data file (strategies 2, 3, and 4):
The simulation is the last process to obtain the growth data. Fig. (4.7(a);
4.7(b); 4.7(d); 4.7(c)) show the fruits fresh weight in each strategy tried out.
With those results can be drawn that the dynamics of the strategies change
in slope, but have a similar behaviour. In the end, the strategy 3 had better
results in production due to the better climate conditions for the crop growth.
The DSS was done for a number of greenhouse decision problems as crop
growth management or harvest scheduling could be developed on the burden
of the environmental impact of the various climate regimes.
Table (4.3) shows as the modification of the climate conditions (temper-
ature or CO2 ) provoke changes in the final production.
Moreover, greenhouses with CO2 enrichment, natural ventilation and heat-
ing get a better performance to produce fruits, reducing the water consump-
tion due to a lower leaf transpiration and leaching fraction [142]. This is
translate in a higher relationship between fruits weight and the total plant
weight, even the leaf growth (LAI) is lesser comparatively. The effect of tem-
perature or CO2 on fruits weight, LAI and on other crop development issues
can be studied by growers, so the need for balanced growth in the long run
102 4 Crop growth modelling and simulation
suggests that the desired integral value should be coupled to the available
light.
In addition, it is possible to generate a summary inform, in pdf format,
which contains the numeric values abstract at the end of the season to pro-
gress with the decision analysis and highlight the most significant factors
in the choice among the different alternatives. For the future remains the
possibility to export simulation data to a file.
4.1 Growth model for sweet pepper 103
4.1.6 Conclusion
The research and studies on the development of growth models for the sweet
pepper crop is still scarce, if compared with other greenhouse crops such
as tomatoes. In addition, the problem of fluctuation in the production was
found.
The results obtained by the model present underestimation found in the
central periods of simulated seasons. Even considering this, the data pro-
duced by the model compared to available data are satisfactory. The model
can correctly simulate dry matter production based on photosynthesis and
partitioning between different organs of the plant.
In the same way, the calibration and validation seasons belong to the same
time period. It means that climate data are very similar to each other, and
therefore the response of the model under different climatic conditions is not
known.
The tool provides production results depending on the variations in the
climatic conditions of the greenhouse environment which allow an optimum
pepper growth control. The DSS has been developed using object-oriented
methodology, which facilitates the reuse, modification or extension of each of
its components.
This paper presents a growth model of sweet pepper cultivation based
on different studies by other authors. This model provides a good basis for
further research and development in the future. In addition, the software
implementation of the model, can be a useful tool for growers. This software
can be used to understand the pepper crop growth process for teaching or
researching and to get an optimal production in greenhouses.
Note that the incorporation of a greenhouse climate model based on ex-
ternal conditions to evaluate the effect of different greenhouse structures on
the pepper growth. Furthermore, it is exploring the possibility of extending
the tool with other horticultural varieties as cucumber or tomato.
Future works
A future line of the present work is the models evaluation in controlled exper-
imental conditions including salinity management. Furthermore, the models
will be tried out in crop with similar agronomical characteristics like cucum-
ber. The decision support system will be implemented as a web-based tool
to be used with teaching purposed.
104 4 Crop growth modelling and simulation
Acknowledgements
[!htbp]
Symbol Units Description Value Reported Variable
aF [–] Fruits DW fraction 0.48 0.5 Wf ,l ,s
aR [–] Roots DW fraction 0.17 0.17 Wf ,l ,s
alf a [–] PAR Leaf absorption 0.76 0.8 Pleaf
b [–] Fractions amplitude 0.17 0.17 Wf ,l ,s
c [days] Sin phase [60, 25] [60, 22] Wf ,l ,s
d [days] Fruits life cycle [79] [76] Wf ,l ,s
eq [µmolmol−1 ] Quantum light efficient [0.063] [0.06] Pleaf
fc [g PS/ g CH2 O] Conversion factor [0.29] [0.28] Wf ,l ,s
Kc [µmolmol−1 ] CO2 Michaelis constant [300] [300] Pleaf
Ko [µmolmol−1 ] O2 Michaelis constant [300] [300] Pleaf
O [µmolmol−1 ] O2 partial pressure [21000] [21000] Pleaf
pLAI [–] Gompertz parameter [0.0018] [0.00161] LAI
τ(25) µmolmol−1 [2600] [2600] Pleaf
Vcmax [µmolmol−1 s−1 ] Max carboxylation [200] [200] Pleaf
Table 4.4 Model parameter outline.
[!htbp]
Symbol Units Description Variable
C [µmol mol−1 ] Intercellular [CO2 ] Pleaf
ff [–] Dry weight fruits Wf
fl [–] Dry weight leaves Wl
fr [–] Dry weight roots Wr
fs [–] Dry weight stems Ws
Γ∗ [µmol mol−1 ] CO2 compensation point Pleaf
If [µmol mol−1 s−1 ] PAR flux incident Pleaf
XLAI [m2 m−2 ] LAI LAI
XLAImax [m2 m−2 ] LAI initial LAI
XLAIini [m2 m−2 ] LAI max LAI
Pcrop [g CH2 O m2 h−1 ] Crop Photosynthesis [–]
Pleaf [µmol mol−1 ] Pleaf Pleaf
Rm,r [g CH2 O m2 h−1 ] Rm , T=25o C [–]
Rm [g CH2 O m2 h−1 ] Maintenance respiration [–]
TS [o C day] Thermal time Rm,r
Vc [µmol mol−1 s−1 ] CO2 assimilation Pleaf
Vq [µmol mol−1 s−1 ] Low light levels Pleaf Pleaf
Wf ,l ,r ,s ,T [g DW] Fruits, Leaves, Roots, Stems DW [–]
WT [g DW] Total DW [–]
Table 4.5 Model variables resume.
107
108 5 Renewable energy applications in greenhouses
5.1.1 Introduction
Agriculture is one of the sectors with the greatest potential for utilising en-
ergy transformation from renewable sources such as biomass, although it
has to be said that the existence of crops makes treatment of mass and
energy flows more sophisticated than in other applications [1, 17, 68]. Des-
pite this, the presence of other factors such as water consumption, energy
requirements, crops, etc... favour the implementation of renewable systems
more than in other production environments. However, when any project is
considered which integrates renewable sources in greenhouses, the following
considerations need to be determined from the outset: (1) The minimum in-
terference conditions on greenhouse production processes by the renewable
systems installed and (2) the choice of a self-consumption operation mode,
setting the renewable energy production to the specific needs of each farm.
In greenhouses, internal temperature control is of significant economic im-
portance. In the biomass system approach, one of the goals it to reduce the
farmer’s costs - a good system design can translate into a significant pro-
duction cost reduction, as heating represents about 30% of the total [116].
It is also, therefore, of great interest to increase the efficiency of commercial
systems in order to increase production [94] and reduce costs associated with
heating but with no structural modifications. Furthermore, in countries with
mild climates, daylight temperature control is performed by convective air
5.1 Development of a biomass-based system 109
exchange between the outside and inside gases [82], which occurs through
the windows. In greenhouses, when photosynthetic rates are higher, the CO2
concentration falls below the atmospheric concentration producing a defi-
cit. It increases even more when the crop reaches maximum development
[143]. Assays performed in Mediterranean greenhouses indicate that the CO2
concentration decreases by up to 20% compared to atmospheric CO2 concen-
tration [96], even when the windows are opened and it is enhanced by lower
wind speed. As a means of addressing this deficit, some authors propose va-
rious carbon enrichment techniques in order to overcome these problems and
achieve increased crop production [39, 61]. Conversely, many authors assert
that the photosynthesis rate enhances, thus increasing CO2 concentration
above the atmospheric value (0.035-0.038% volume fraction) to 0.1% volume
fraction. In this range, the photosynthesis rate becomes optimal and cannot
be surpassed by increasing the CO2 concentration [39, 61, 117]. This pho-
tosynthesis rate increase translates into a significant rise in production in
accordance to that shown in [9, 51, 82, 89, 117, 143, 142, 157], even can be
found examples in other ecological systems like microalgae bioreactors [7].
It is also essential to optimize carbon fertilization strategy responses, per-
forming them in the absence of other limiting factors - the best results are
obtained when radiation levels are high i.e. the photosynthetic rate is higher
when temperature and CO2 concentrations rise along with an increase in
available solar radiation [3, 7, 143, 157]. Carbon enrichment not only affects
the photosynthesis rate, it also increases water consumption efficiency by
gradually reducing the stomata-opening rate, decreasing water loss caused
by transpiration [142].
This work proposes a combined system, which consists of a greenhouse
residue heating system to control the nocturnal temperature at low cost,
and a CO2 storage system, where CO2 generated from the heater is stored
for use during the diurnal photosynthesis process. Notice that in both con-
trol processes, nocturnal temperature and diurnal CO2 supply, the energetic
and economic efficiency is improved by reusing greenhouse organic material
(biomass) and CO2 generated from the heating system.
The research data used in this work have been obtained from a greenhouse in-
stalled in the Experimental Station of Cajamar Foundation, in El Ejido, in the
province of Almeria, Spain (2o 43’ W, 36o 48 N, with a 151 m elevation). This
greenhouse is a multispan ”Parral type” greenhouse with a surface area of
110 5 Renewable energy applications in greenhouses
To achieve the desired goals, the system can be operates burning at night,
because of the design requirements and implementation of a CO2 capture and
storage system, which took into account the final goal of applying this on an
industrial scale. The energy produced in the biomass combustion provides
heat (about 17 MJ kg−1 ), which is transferred to the water and CO2 content
from the combustion flue gases. The system is responsible for gas purification
5.1 Development of a biomass-based system 111
Fig. 5.1 Examples of biomass used in this paper. From left to right and from top to
bottom on the figure: almond shells, olive-pit, and tomato residues and pine pellets.
through CO2 capture, releasing the rest into the atmosphere. The use of the
capture system allows to reduce the volume of gas stored under pressure,
obtaining more concentrated CO2 gas. After looking at the different systems
used in industry to capture CO2 , activated carbon was chosen. The choice
was based on several aspects such as it being an inexpensive adsorbent, the
simplicity of the operation, and the adsorbent regeneration in adsorption
or desorption cycles by pressure and temperature changes with low energy
consumption.
Residue Boiler
The boiler power was determined based on the energy requirements, 1.9
MJ m−2 , determined with the aim of getting a thermal increase in the green-
house of 4 to 5 o C. This rise was calculated from the minimum nocturnal
temperature referred to for tomato plants (12 ◦ C) and Almeria’s average
nocturnal minimum temperature in winter of 8.3 o C (Table 5.1). Tm is the
average monthly temperature (o C), Tm,max the average monthly daylight
maximum temperature (o C), Tm,min the average monthly night minimum
temperature (o C), and RH the average monthly relative humidity (%). With
these data, the daily heat consumption was determined using the following
empirical correlation:
112 5 Renewable energy applications in greenhouses
2
Qd = 0.0497∆Tmin − 0.001∆Tmin + 1.107 (5.1)
where Qd is the daily heat consumption, (MJ/m2 day ), and ∆Tmin is the
minimum increase in greenhouse temperature (o C). The heating consumption
is 0.131 kW m−2 with an estimated heating time of 4 h day−1 . Given the
energy available, and assuming that the performance of commercial biomass
boilers is around 90% (thermal efficiency), the minimum power for the blower
must be 145.55 kW. Based on these results, a Missouri 150000 boiler with a
power of 150 kW was installed. This model has a water accumulation capacity
of 300 L with automatic water temperature control, an adjustable pressure
air supply for combustion of 200 - 800 hPa, and an automatic endless screw,
which allows flow rates from 15 to 80 kg h−1 .
The main heating system has the following elements (Fig. 5.2): a biomass
boiler, a mixing valve, a pump, and aerial pipes.
The circulation pump maintains a constant water flow through the heating
circuit. The water speed in the distribution loops is about 0.08 m s−1 , ensur-
ing a flow of 25 m3 s−1 . The heating circuit consists of a main supply pipe, a
main return pipe and distribution pipes along the crop lines formed by two
polyethylene parallel pipes. The main steel pipe is 10 cm in diameter and
the secondary plastic one is 5 cm. A specific 1.86 kW calefaction pump and
a continuous mixing valve is used to control the pipe temperature. A Pt100
5.1 Development of a biomass-based system 113
temperature sensor was installed in each pipe, main and return. The distri-
bution loop runs around the canopy lines placed 10 cm above the ground.
A proportional three-way valve was installed which allows more control over
the mixing ratio between hot water from the boiler and cooled water return-
ing from the distribution loops. The greenhouse temperature setpoint varied
between 9 and 14 o C, concurrent with requirements and enough to avoid
plant growth from stopping.
CO2 system
Fig. 5.3 Outline of the CO2 storage and enrichment system with control loop
of the day. The CO2 system works with two different processes: capture and
enrichment. Both procedures use the same actuator, the proportional valve
after the tank. During capture, the AC tank pressure is controlled by this
valve and pressure sensors located on the outlet and inlet tank pipes; flow
and CO2 sensors were also added to improve process information. The last
step in the CO2 storage system is a three-way valve, which releases the re-
maining gases into the atmosphere. The same sensors are used to control
the CO2 mass flux released during enrichment; in this case, the flow is in-
troduced inside the greenhouse by altering the same three-way valve. Based
on these sensor measurements, we can calculate: (1) the adsorption capacity
of the AC storage, and (2) the amount of CO2 injected into the greenhouse
(desorption).
The physical properties of these combustion processes for the various kinds
of biomass are given in Table 5.2, different assays were carried out in order
to evaluate them.
As the results of this test show, tomato pellets have an excessive ash con-
tent, more so than wood [119] meaning some additional problems might occur
during combustion in the boiler. Furthermore, their moisture content is higher
than for wood pellets. Despite these facts, the gross calorific power is similar
and it was possible to start combustion. The residues used have very similar
GCV compared to the dry mass of other biomasses more commonly used in
combustion applications. The main problem is their high moisture content
after cutting which leads to very poor calorific value. However, this moisture
5.1 Development of a biomass-based system 115
Heating system
The heater (Fig. 5.2) was set using a combustion gas analyzer (IMR
2000/2800P). In the assays, different combinations of primary and secondary
air flow, as well as tomato pellet feed speed were tried and their resulting
combustion efficiencies are shown in Table 5.3:
Despite the fact that a combination existed where it was possible to obtain
fewer losses, the final combination chosen was: 16.0 kg h−1 intake of tomato
biomass, 93.8 primary air m3 h−1 , and 93.6 secondary air m3 /h. This boiler
116 5 Renewable energy applications in greenhouses
Fuel mass Primary air Secondary air Heat losses CO2 Flue gases
velocity (kg h−1 ) flow (m3 h−1 ) flow (m3 h−1 ) (%) (% Volume) T (o C)
16.01 93.81 93.61 38.81 3.91 198.21
16.0 187.6 93.6 45.3 2.7 213.6
16.0 93.8 322.4 40.1 3.1 212.1
16.0 187.6 322.4 58.9 1.9 196.7
33.4 93.8 93.6 26.0 7.0 289.9
33.4 187.6 93.6 42.9 5.2 361.2
33.4 93.8 322.4 29.1 6.3 291.0
33.4 187.6 322.4 51.8 4.3 355.5
33.4 93.8 561.6 23.9 7.0 266.2
33.4 187.6 561.6 43.6 5.7 391.3
56.3 93.8 561.6 31.4 7.0 341.1
56.3 187.6 561.6 21.9 14.0 432.0
1
Working conditions chosen for this paper
Table 5.3 Different combinations tried for boiler optimization conditions
setting was used because it was the optimum for the lowest feed speed res-
ulted in less consumption during this period. The setting chosen gave the
following results: 3.9% volume fraction of CO2 , 38.8% heat loss, a combus-
tion temperature of 198.2 o C. The level of noxious gases as NOx and SO
are low, below the maximum values allowed by Andalusian legislation. The
mass airflow stream produced in the boiler was 246 kg h−1 . The silo used for
storing available biomass would have to have sufficient storage capacity for
4-5 days.
CO2 system
Fig. 5.4 Capture capacity (g kg−1 ) as a function of pressure (kPa) at 10, 20, and
30o C
storage operation conditions were set at 190 kPa and 20 ◦ C using absorption
and costs analysis, at which levels absorption was 21 g kg−1 (Fig. 5.4). The
gas flowing through the particle tank causes a 16.2 Pa of frictional pressure
drop. This pressure drop experienced by the gas passing through the particle
bed is calculated using the following equation [86]:
P12 −P22 = (2RT1 )/M [150 µG(1−ϵ)2 /(ϵ3 pd2 )+1.75G2 (1−ϵ)/(pdϵ3 )]Lb (5.2)
where P1 is the inlet pressure in Pa; P2 is the outlet pressure (Pa); R is the
gas constant (8.314 m3 Pa K−1 mol−1 ); T1 , the inlet temperature (K); M ,
the molecular weight (kg mol−1 ); ,ϵ the gas viscosity, in kg•m−1 •s−1 ; G, the
mass speed (ρvo ); iϵs the porosity; Lb , the bed length (m) and pd, the particle
diameter (m). In this way, the isothermal compression 5.3 of combustion gas
requires an input of 89.45 kJ kg−1 , and for the compression of 50.97 kg h−1
is required about 1.27 kW [86].
kJ kg−1 . Assuming the same 60% in the blower, the power must be at least
2.48 kW [86].
Heating supply
The control system algorithms required a prior study of the system prop-
erties and dynamics. Many assays on the real system were carried out to test
its performance and to acquire enough data dynamic richness to obtain mo-
dels by system identification, Fig. 5.6 shows the evolution of tests covering
one night in winter. During these tests, the heated water in the boiler had a
set-point fixed at 65 ◦ C, step changes were then performed in the valve aper-
ture until the steady state were reached, and the responses observed. Varying
operating conditions and varying the proportional valve opening (control sig-
nal) led to important changes in the process response (water temperature,
Fig. 5.6a).
Fig. 5.6 Step response and response to an output disturbance. From left to right
and from top to bottom on the figure: Step signal assay and the main and return
temperatures, inside and outside humidity, inside and outside temperature, heating
system and wind speed.
Fig. 5.6a shows several steps introduced, in which it is evident how the
inside greenhouse temperature varies when the changes introduced in the
control signal are performed. The disturbances action, overall the return
pipe temperature, make difficult to differentiate the steady state. Disturbance
variables have a dominant role and coherent effect on forming the greenhouse
environment (Fig. 5.6b,Fig. 5.6c, and Fig. 5.6d). These figures also show dis-
turbances acting on greenhouseconditions during the assays even though the
120 5 Renewable energy applications in greenhouses
CO2 supplied
The progress of CO2 adsorption is shown in Fig. 5.7 and Fig. 5.8. Comple-
mentary to this, Fig. 5.7a and Fig. 5.8a show the progress of CO2 adsorption
in the tank. The total amount of CO2 captured in this process was 16.3 kg and
23.05 kg, respectively. The average molar CO2 concentrations observed once
the active carbon was considered saturated, were 6.9% and 9.3% (Fig. 5.7b
and Fig.5.8b), respectively. Therefore, this was the average concentration of
gas coming out of the boiler. The elapsed time to fill up the AC completely
was approximately 250 min in both cases.
Several stages are observed in the capture process by following CO2 con-
centration over time. Initially, the AC bed is completely empty; therefore
any CO2 coming into the bed is adsorbed and there is no observed CO2 in
the gas coming out of the tank, the pressure is increased (Fig. 5.7d and Fig.
5.8d) controlled by the proportional valve (Fig. 5.7e and Fig. 5.8e). Once
the active carbon bed starts to become saturated, adsorption performance
begins to fall until zero (Fig. 5.7c and Fig. 5.8c). By this time, the CO2
concentration is the same as the CO2 concentration of the flue gases. It is
important to remark that Fig. 5.7c shows negative values because of the CO2
released from the tank is higher that the CO2 concentration measured in
the boiler smokes. Such situation can be due to, the active carbon was full
of CO2 and lost its capture capacity, and a part of that CO2 was released
joined to the CO2 contained in the smokes. This situation helped to establish
the CO2 released setpoint and thus to set the end of the capture process.
The total CO2 captured was in accordance with the amount recommended
in the bibliography, but in the first assay (Fig. 5.7) was produced a decrease
in the system performance due to the air came in to the tank warmer than
expected.
The CO2 concentration dynamic in the greenhouse is revealed in Fig. 5.9a
and Fig. 5.10a. The flow introduced in the greenhouse is given in Fig. 5.9c
and Fig. 5.10c, while the CO2 captured in a ‘perfect’ greenhouse (without
external exchange) is given in Fig. 5.9e and Fig. 5.10e. In addition, a comple-
5.1 Development of a biomass-based system 121
mentary mass balance was performed to study the CO2 captured in the AC
tank (Fig. 5.9g and Fig. 5.10g). The released amount in these assays was 31.4
and 20.5 kg, respectively. The CO2 concentration increased over time inside
the greenhouse. The maximum achieved CO2 concentration was three times
122 5 Renewable energy applications in greenhouses
released CO2 decreased, the CO2 concentration inside the greenhouse began
to fall because of photosynthetic consumption and losses through the win-
dows and holes in the plastic (a common problem in parral-type greenhouses).
Table 5.4 shows the flows produced in the system during CO2 capture, on
which the use of exchangers had a great influence.
124 5 Renewable energy applications in greenhouses
During CO2 capture, the flue gases coming from the combustion gases are
driven into the AC storage by a blower. Pressure is the critical parameter at
this stage, since the CO2 adsorption yield in activated carbon increases with
pressure. For this reason, the control objective in this step is to regulate the
pressure inside the AC-bed, at or about a preset value, while the flue gases
pass through. The control device is a proportional valve located at the AC
storage outlet. For the control algorithm, a PI controller with restrictions
would be implemented. These restrictions consist of several variables limit-
ing the capture process: (1) the flue gas temperature, which is used to show
when the burner is working so the capture process can begin; (2) the flue gas
temperature, which is basic because of blower limitations, it has a maximum
working temperature of 65 o C according to the manufacturer’s specifications;
and (3) the maximum CO2 in the outlet pipe; when this value is exceeded, the
AC-storage is full and the capture process stops. For a while after the burner
has started, bad combustion is produced and the flue gases contain a small
percentage of CO2 , and high concentrations of undesirable gases like carbon
monoxide (CO) and unsaturated hydrocarbons. To avoid this situation the
capture starts when the flue gas temperature in the chimney becomes higher
than 190o C and is stable. The flue gas flow is blown through pipe with an
exchanger to reduce temperature. Moreover, the flue gases cooled by the
exchanger are warmed by the action of the blower; another exchanger was
therefore installed between the tank and the blower. The blower forces the
gas flow into the bulk filled with active carbon while the proportional valve
placed after the tank maintains the pressure (190 kPa) inside the tank. Un-
der these pressure conditions, the activated carbon captures CO2 while other
gases are released out of the tank when the proportional valve is opened
to regulate pressure. A three-way valve was installed between the greenhouse
and the system allowing gases to be driven either to the greenhouse or outside
depending on the situation. The three-way valve is directed outside during
capture and towards the greenhouse during enrichment. The CO2 molar frac-
tion is measured at the tank output. The AC-bed is saturated when the CO2
concentration in the outlet pipe is close to the maximum CO2 concentration
in the tank.
126 5 Renewable energy applications in greenhouses
During the injection stage, the main aim is to maintain the CO2 concen-
tration in the greenhouse around the setpoint. When the enrichment process
starts (with the daylight), and the CO2 inside the greenhouse is below the
setpoint. Then, the CO2 retained in the adsorption bed is released opening
the proportional valve. This action provokes a pressure drop inside the AC
storage tank. When the tank pressure is lower than a minimum, which means
that all the CO2 retained has been released. Then, the blower is activated to
release the CO2 retained in the storage free volume. The blower stays on until
that the CO2 measured in the outlet tube is lower than an established value.
Besides the CO2 setpoint in the greenhouse, two certain conditions determine
the control action: (1) A minimum pressure at the outlet of the tank from
which the blower is activated, and (2) A minimum CO2 concentration in the
pipe, when the system reached this minimum therefore implies that the tank
has released all the carbon dioxide captured.
5.1.4 Conclusions
This work shows a design methodology for a heating, CO2 storage and
CO2 enrichment system. The main objective of this paper was to try out a way
of storing the CO2 from the flue gases. To achieve this, various sub-objectives
were met - the design of the heating system, the boiler fixing, and the testing
of different kinds of biomass fuels. The dual-mode biomass-based system has
been developed as an ecological system in order to reduce petrol-based fuel
demand, achieving cost savings from using chipper fuel as pellets from green-
house plants residues (tomato and pepper). The utilization of greenhouses
residues was evaluated and compared with other biomass fuels as pine pel-
lets or olive-pits obtaining good results. Once achieved the biomass residues
to use, the next step was to create a protocol for the design of a biomass
system to produce heat and CO2 . To achieve the design protocol was taken
into account the necessity to reduce the final installation cost. In this way,
the number of devices to use and their characteristics was deeply studied to
get a good results. For this design was assessed the optimal requirements of
CO2 for the crop and the minimal heat necessities for tomato crops on the
southeast coast of Spain. The setpoints for the different system was thought
to obtain an efficient use of the energy, trying to moderate the electrical en-
ergy consumption. Active carbon demonstrated being a cheap and efficient
tool to CO2 storage and later release to the greenhouse. In the end, simple
5.1 Development of a biomass-based system 127
control systems was tried out for both systems, CO2 and heating. The results
of these test show a good system behaviour. The hybrid system is available
to generate heat and CO2 as inputs in a cheap operational way.
Acknowledgements
Appendices
128 5 Renewable energy applications in greenhouses
5.2.1 Introduction
Agriculture is one of the sectors with the greatest potential for energy used
from renewable sources such as biomass, although the existence of crops
makes treatment of mass and energy flows more sophisticated than in other
applications (Abdel-Ghany & Al-Helal, 2011; Boulard & Baille, 1987). Des-
pite this, the presence of other factors such as irrigation, CO2 enrichment,
heating, refrigeration, etc. . . favour the implementation of renewable systems
more than in other production environments. However, when any project is
considered which integrates renewable sources in greenhouses, the following
considerations need to be determined from the outset: (1) The minimum in-
terference conditions on greenhouse production processes by the renewable
systems installed and (2) a self-consumption operation mode design, setting
the renewable energy production to the needs of each farm. This work pro-
poses hybrid model of a biomass-based system (Fig.5.11, [146]) that consist
in: (i) a CO2 storage system, where (ii) CO2 generated from the heater is
stored for use during the diurnal photosynthesis process (enrichment), (iii)
the CO2 is taken by the crop, and (iv) the CO2 lost by natural ventilation.
130 5 Renewable energy applications in greenhouses
The boiler (Fig.5.11) has two different inlet air streams which supply O2 for
combustion. One is placed towards the basement of flame, where combus-
tion is developed, and it is driven to it. The other one is injected to a higher
place. Furthermore, it is possible to adjust the biomass intake though a screw
conveyor. The combustion efficiency, the flue gases temperature and its con-
sequent CO2 production depend on the combination of these three streams.
The setting chosen gave the following results: 3.9% volume fraction of CO2 ,
38.8% heat loss, a combustion temperature of 198.2 [146].
5.2 Hybrid modelling for a biomass-based system 131
The produced amount of CO2 was estimated taking into consideration meas-
ured flow of gasses coming out heater, their concentration and flow of burnt
biomass:
ṁair
Fair = (5.5)
M M air
where Fair is the molar air flow; ṁair the mass air flow; M M air the molecular
mass of air, FCO2 is the Molar CO2 flow; andx CO2 e : CO2 concentration.
where mCO2 is the mass CO2 flow; M M CO2 the CO2 molecular mass, Y CO2 :
CO2 production yield; air : mass air flow. CO2 production yield of considered
biomass is directly estimated from CO2 concentration in gasses stream com-
ing out boiler (xCO2e ), CO2 (MMCO2 ) and air (MMair ) molecular masses.
Several assays were performed in order to check the required time to com-
plete a capture in the AC tank and its CO2 adsorption capacity. To get this
goal, first, CO2 was completely removed from tank and then, capture process
was performed until concentration in the stream coming from boiler became
constant. Pressure operation was 1.9 bar [146]. Tomato pellets was used in
this assays. Total CO2 captured is estimated considering a matter balance in
which both of gasses flow and its CO2 concentration were taken into account
for streams coming from boiler and the tank:
∑
M CO2,ads = t
t=0 [Mw,CO2 (PP,i FCO2 )/(RT )(XCO2 out − XCO2 in )] (5.7)
where MCO2,ads is the total adsorbed mass in active carbon bed; Mw,CO2 the
CO2 molecular mass; PP,i the pressure inside tank; FCO2 the gas flow com-
ing out the tank; R the universal gas constant; T the gas temperature;
XCO2out the molar CO2 concentration in the stream coming out from tank;
XCO2in the molar CO2 concentration in the stream coming into tank.
132 5 Renewable energy applications in greenhouses
FCO2,gr
M CO2,en = (5.8)
cs,gr
[( )2 ( )
X −P
ϕvent = cven,d √Vven,area−lat
2
Vven,area−roof
2
2cg cven,h t,aX̄i t,e +
Vven,area−lat +Vven,area−roof
( )2 ]
Vven,area−lat Vven,area−roof 2
+ 2 cven,w Pws,e + ϕlosses
(5.10)
A well-known model is the TOMGRO [75, 128]. In such model, the photo-
synthesis is given by the next equation:
( )
Ccnvf ot Fmax ftf ot (XT a )
M CO2,f ot = CK
[ ] (5.12)
(1−Cm )Fmax +αCK V −RF A
ln (1−Cm )Fmax +αCK V −RF A(−CK XLAI ) − Rm
.
where CK is the extinction coefficient of light through the canopy, Cm is the
coefficient of transmission of light by leaves, VRF A is photosynthetic active
radiation, α is the efficiency of utilization of leaf photosynthetic radiation,
XLAI is the leaf area index, and Ccnvf ot is a coefficient for converting units.
On the other hand, the respiration equation is given by:
∫ tf
(X −20)/10
Rm = Q10 ta rm (XP S − XM )dt (5.13)
ti
.
where Xta is the air temperature, Q10 is the respiration sensibility to the
temperature y rm is the maintenance respiration coefficient, ti initial time y
tf final time.
Operation modes BL PV 3W VT BH
1. Out of service 0 0 0 0 0
2. Cooling 0 0 0 1 0
3. Capture 1 0 0 0 1
4. Enrichment 0 1 1 0 0
5. Capture & Proportional valve 1 1 0 0 1
6. Enrichment & Blower 1 1 1 0 0
7. Enrichment & Ventilation 0 1 1 1 0
8. Enrichment, Ventilation & Blower 1 1 1 1 0
Table 5.6 System operation mode actuator configuration.
Under these pressure conditions, the activated carbon captures CO2 while
other gases are released out of the tank when the proportional valve is opened
to regulate pressure. If the AC tank pressure is higher than 1.9 bar, the
system switches to mode 5, controlling the pressure with the proportional
valve installed in the outlet pipe.
The boiler can be paused changing the system to mode 1. A three-way
valve was installed between the greenhouse and the system allowing gases to
be driven either to the outside during capture and towards the greenhouse
during enrichment. On the other hand, in the enrichment subsystem, the
main aim is to maintain the CO2 concentration in the greenhouse around
the setpoint. If the greenhouse demands CO2 to enrich, it means the crop
requires CO2 to carry out the photosynthesis. Then happens, the system
5.2 Hybrid modelling for a biomass-based system 135
Several assays were performed in order to evaluate the greenhouse CO2 con-
centration dynamic. For the capture process, the assays consist in introducing
the flue gases into the activated carbon bed until saturation, repeating the
operation for different pressure valve openings and different external climate
conditions. On the other hand, in order to calibrate and validate the enrich-
ment and ventilation model, different CO2 injections were performed during
night (with plants) and with the greenhouse without crops, aiming that there
were not any CO2 consumption by plants for their photosynthesis. These as-
says were hold until CO2 concentration of stream coming out AC tank got
zero. Finally, the TOMGRO model was calibrated by [152] for the same con-
ditions.
136 5 Renewable energy applications in greenhouses
Once finished the different submodels validation, the hybrid model was
validated for a normal working day (Fig. 5.13). For it, the subsystems were
included or excluded from the plant during its operation by means of manip-
ulating valves, defining their current operation mode (Table 5.6, Fig. 5.12):
cooling (mode 2, 7 and 8), enrichment (mode 4, 6, 7, and 8), and capture
(mode 3 and 5). The plant operation modes, together with the actuators
signal value related to each of them, are shown in Table 5.6, where are rep-
resented with a value equal to 1 when its mode is the current one, and zero
in any other case. The transitions among the operation modes are defined in
Fig. 5.12. The Fig. 5.13 shows how the three subsystems are interconnected
during the real plant operation and, furthermore, show the model validation
with a good fixed of the real system dynamic.
For the future, it would be necessary the incorporation of the full system,
including the greenhouse heating process with warm water which come from
the biomass boiler.
Chapter 6
Other publications
The author of this PhD thesis has collaborated actively in some publications
with other researchers:
137
138 6 Other publications
• Examples in robótics:
7.1 Conclusions
• The seek of transpiration models to estimate how much water the crops
release to the atmosphere which can be used with two different object-
ives: (i) to design irrigation controllers and (ii) to estimate air humidity
content. The research was conducted over eight crop cycles in those which
took the different climatic and physiological variables of a tomato crop.
The transpiration measurements were obtained by weighing the difference
that occurs on a system formed by two scales, microlysimeter, which is an
expensive and performance device that should be placed by the models
calibrated. The resulting models, or virtual sensors, are based on sensors
that are typically installed in greenhouses: temperature, humidity, CO2 ,
and solar radiation. The aim is to reduce the total installation costs and
to avoid the constant maintenance that the scales require. Thus, two di-
fferent virtual sensors were tested: (i) Grey-box virtual sensor has a good
fixing as advantage, but some disadvantages such as the overestimation
at different moments of the year, the necessity to use a growth model to
estimate the leaf are index, and the advance phenomenon by which the
estimated value has a delay in relation with the real ones. (ii) The black-
box virtual sensors obtain better results and also allows the elimination
of the grey-box problems: advance phenomenon and overestimation. An
overestimation only appears during nocturnal periods. In this case, two
139
140 7 Conclusions and future works
• The water content and the transpiration virtual sensors obtained will allow
the design of irrigation controllers with different control strategies. The
objective is not only to reduce the water supplied as much as possible, but
also to determine the right moment to be supplied.
• Respect to the growth models, a future line of the present work is the
models evaluation in controlled experimental conditions including irriga-
tion and salinity management. Furthermore, the models will be evaluated
with others greenhouse crops like cucumber or aubergine. Decision support
systems will be implemented to be used with training purposes.
• The biomass-based system designed during the realization of this disserta-
tion has been tested in many assays, even a first approach as hybrid model
have been presented. For the future remains the inclusion of the heating
system in the hybrid model to complete the full system modelling. Once
finished, the development of advance control algorithms will be performed
with the aim to reduce the system energy consumption.
• In order to introduce more complexity to the hierarchical scheme, the hu-
midity modelling will be addressed with the inclusion of a dehumification
system based on water condensation. This facility will provide a degree of
freedom to the system keeping the relative humidity within certain range,
avoiding extreme situations that provokes the reaching of cryptogamic dis-
eases or stress problems. In the first assays, the device has demonstrated
a high power to reduce the humidity in winter greenhouse conditions with
good dynamics. The next step will be to model the humidity introducing
the effect of the dehumification.
• The joint control of temperature and relative humidity with two different
inputs, the greenhouse ventilation and the dehumification system. Further
to the problems that arise to control two variables at the same time, there
is also a highly negative correlation, so that changing the value of one of
them involves the variation of the other. Using both actuators, the control
of one of them may cause an undesirable behavior, being off the limits.
All these facts lead to evaluate multivariate control techniques, based on
systems with multiple inputs and multiple outputs (MIMO).
• The introduction of new facilities to the hierarchical algorithm will be
completed with the incorporation of energy demand prediction and dis-
tribution models. Thus, different wattmeters have been installed in the
7.2 Future Works 143
References
[58] J. Goudriaan and HH. Van Laar. Modelling potential crop growth
processes. Kluwer Academic Publishers, Dordrecht, The Netherlands,
1994.
[59] J.K. Gruber, J.L. Guzmán, F. Rodrı́guez, C. Bordons, M. Berenguel,
and J.A. Sánchez. Nonlinear mpc based on a volterra series model for
greenhouse temperature control using natural ventilation. Control Eng
Pract, 19(4):354–366, 2011.
[60] J.L. Guzmán, D.E. Rivera, S. Dormido, and M. Berenguel. An interact-
ive software tool for system identification. Advan Eng Soft, 45(1):115
– 123, 2012.
[61] X. Hao, Q. Wang, and S. Khosla. Responses of greenhouse tomatoes
to summer co2 enrichment. Acta Hortic, 797:241–246, 2008.
[62] Y. Hashimoto. Computer integrated system for the cultivation process
in agriculture and horticulture. Approach to Intelligent Plant Factory,
pages 175–196. The computerized greenhouse. Eds. Hashimoto Y., Bot
G., Tantau H.J., Day W., Nonami H. Academic Press, 1993.
[63] G. Heredia and A. Ollero. Virtual sensor for failure detection, iden-
tification and recovery in the transition phase of a morphing aircraft.
Sensors, 10(3):2188–2201, 2010.
[64] E. Heuvelink. Evaluation of a dynamic simulation model for tomato
crop growth and development. Ann Bot-London, 83(4):413–422, 1999.
[65] T. Hoch and J. Agabriel. A mechanistic dynamic model to estimate
beef cattle growth and body composition: 1. model description. Agr
Syst, 81(1):1 – 15, 2004.
[66] M.P. Hoffmann, A. Castaneda-Vera, M.T. van Wijk, K.E. Giller,
T. Oberthur, C. Donough, and A.M. Whitbread. Simulating potential
growth and yield of oil palm (elaeis guineensis) with palmsim: Model
description, evaluation and application. Agr Syst, 131(0):1 – 10, 2014.
[67] J. Hollingworth, B. Swart, P. Gustavson, and M. Cashman. Borland C
Builder 6 developer’s guide. Sams, Indianapolis, Ind., 2003.
[68] IDAE. Plan for renewable energies in Spain 2011-2020. Spanish Min-
istry of Industry, Energy and Turism - IDAE, 2011.
[69] I. Ioslovich and P.O. Gutman. Evaluation of experiments for estimation
of dynamical crop model parameters. Bull Math Biol, 69(5):1603–1614,
2007.
[70] I. Ioslovich, M.I.R.S. Moran, and P.O. Gutman. Identification of a non-
linear dynamic biological model using the dominant parameter selection
method. J Franklin Inst, 347(6):1001–1014, 2010.
[71] B.K. Jacobson, Pierce H. Jones, J.W. Jones, and J.A. Paramore. Real-
time greenhouse monitoring and control with an expert system. Comput
Electron Agr, 3(4):273 – 285, 1989.
[72] Y.W. Jame and H.W. Cutforth. Crop growth models for decision sup-
port systems. Can J Plant Sci, 76(1):9–19, 1996.
150 7 Conclusions and future works
[73] R. Jemaa, T. Boulard, and A. Baille. Some results on water and nutrient
consumption of a greenhouse tomato crop grown in rockwool. Acta
Hortic, 408:137–146, 1995.
[74] O. Jolliet and B. Bailey. The effect of climate on tomato transpira-
tion in greenhouses: measurements and models comparison. Agr Forest
Meteorol, 58:4362, 1992.
[75] J. Jones, E. Dayan, L.V. Allen, and H. Challa. A dynamic tomato
growth and yield model (tomgro). Transactions of ASAE, 34:663–672,
1991.
[76] J. Jones, Y. Hwang, and A. Kenig. Developing a simplified crop model,
chapter in: Optimal Environmental Control for Indeterminate Green-
house Crops. Eds. Seginer I., Jones J.W., Gutman P., Vallejos C.E.
BARD Research Report, 1995.
[77] J.W. Jones, A. Kenig, and C.E. Vallejos. Reduced state-variable tomato
growth model. Transactions of ASAE, 42:255–265, 1999.
[78] T. De Jong. Natural Ventilation of Large Multi-span Greenhouses, 1990.
[79] P.G.H. Kamp and G.J. Timmerman. Computerized environmental con-
trol in greenhouses. A step by step approach. IPC Plant, The Nether-
lands, 1996.
[80] D.J. Kilpatrick and R.W.J. Steen. A predictive model for beef cattle
growth and carcass composition. Agr Syst, 61(2):95 – 107, 1999.
[81] C. Kittas, T. Boulard, and G. Papadakis. Natural ventilation of a
greenhouse with ridge and side openings: Sensitivity to temperature
and wind effects. Transactions of ASAE, 40(2):415–425, 1997.
[82] H.P. Kl˙Model-based control of co2 concetration in greenhouses at am-
bient levels increases cucumber yield. Agrl Forest Meteorol, 143:208 –
216, 2007.
[83] H.P. Kläring and D. Schwarz. Model based control of concentration
of nutrient solution in glasshouse tomato cultivation. Acta Hortic,
507:127–132, 1999b.
[84] H.P. Kläring, D. Schwarz, and A. Heibner. Control of nutrient solu-
tion concentration in tomato crop using models of photosynthesis and
transpiration- a simulation study. Acta Hortic, 450:329–334, 1997.
[85] P. Kramer and J. Boyer. Water Relations of Plants and Soils. Academic
Press, San Diego,, 1995.
[86] O. Levenspiel. Engineering flow and heat exchange. Springer, New
York, 2nd ed. edition, 1997.
[87] Haorong Li and James E. Braun. Decoupling features and virtual
sensors for diagnosis of faults in vapor compression air conditioners.
Int J Refrig, 30(3):546 – 564, 2007.
[88] T. Li, E. Heuvelink, T.A. Dueck, J. Janse, G. Gort, and L.F.M. Mar-
celis. Enhancement of crop photosynthesis by diffuse light: Quantifying
the contributing factors. Ann Bot-London, 114(1):145–156, 2014.
References 151
[89] R. Linker, P.O. Gutman, and I. Seginer. Robust controllers for simul-
taneous control of temperature and co2 concentration in greenhouses.
Control Eng Pract, 7(7):851–862, 1999.
[90] R. Linker, I. Seginer, and P.O. Gutman. Optimal co2 control in a green-
house modeled with neural networks. Comput Electron Agr, 19(3):289–
310, 1998.
[91] L. Ljung. System Identification: Theory for the user. Second edition.
Prentice Hall, 1999.
[92] L. Ljung. System Indentification Toolbox TM 7 Getting Started Guide,.
MathWorks, 2010.
[93] L. Ljung and T. Glad. Modeling of Dynamic Systems. Prentice-Hall,
Englewood Cliffs, NJ, 1994.
[94] J.C. López, A. Baille, S. Bonachela, and J. Pérez-Parra. Analysis and
prediction of greenhouse green bean (phaseolus vulgaris l.) production
in a mediterranean climate. Biosyst Eng, 100(1):86 – 95, 2008.
[95] I.L. López-Cruz, J. A. Sánchez-Molina, A. Ramı́rez-Arias,
F. Rodrı́guez-Dı́az, and J.L. Guzmán-Sánchez. Local and global
sensitivity analysis of a greenhouse crop transpiration model. In Book
of abstracts of Greensys 2011. ISHS, 2011.
[96] P. Lorenzo, C. Maroto, and N. Castilla. Co2 in plastic greenhouse in
almeria. Acta Hortic, 262:165–170, 1990.
[97] J.J. Magán, Gallardo M., Thompson R.B., and Lorenzo P. Effects of
salinity on fruit yield and quality of tomato grown in soil-less culture in
greenhouses in mediterranean climatic conditions. Agric Water Man-
age, 95(9):1041–1055, 2008.
[98] L.F.M. Marcelis. A simulation model for dry matter partitioning in
cucumber. Ann Bot-London, 74(1):43–52, 1994.
[99] L.F.M. Marcelis, F. Buwalda, J.A. Dieleman, T.A. Dueck, A. Elings,
A. De Gelder, S. Hemming, F.L.K. Kempkes, T. Li, F. Van Noort,
and P.H.B. De Visser. Innovations in crop production: A matter of
physiology and technology. Acta Hortic, 1037:39–46, 2014.
[100] L.F.M. Marcelis, A. Elings, M.J. Bakker, E. Brajeul, J.A. Dieleman,
P.H.B. de Visser, and E. Heuvelink. Modelling dry matter production
and partitioning in sweet pepper. Acta Hortic, 718:121–128, 2006.
[101] L.F.M Marcelis, E Heuvelink, and J Goudriaan. Modelling biomass pro-
duction and yield of horticultural crops: a review. Sci Hortic, 74(12):83
– 111, 1998.
[102] R. Martin-Clouaire, P.J. Schotman, and M. Tchamitchian. A survey of
computer-based approaches for greenhouse climate management. Acta
Hortic, 406:409–423, 1996.
[103] L. Maruster, N.R. Faber, R.J. Jorna, and R. van Haren. Analysing
agricultural users patterns of behaviour: The case of optiras, a decision
support system for starch crop selection. Agr Syst, 98(3):159 – 166,
2008.
152 7 Conclusions and future works
[104] B. Matysiak and J. Nowak. Carbon dioxide and light effects on pho-
tosynthesis, transpiration and ex vitro growth of homalomena emerald
gem plantlets. Sci Hortic-Amsterdam, 57(4):353 – 358, 1994.
[105] E. Medrano, P. Lorenzo, M. C. Sánchez-Guerrero, and J. I. Montero.
Evaluation and modelling of greenhouse cucumber-crop transpiration
under high and low radiation conditions. Sci Hortic, 105:163175, 2005.
[106] J.A. Sánchez-Molina Mena, R., F. Rodrı́guez, J.L. Guzmán, F. G.
Acién, J. V. Reinoso, and J.C. López. Designing a modulable and
scalable scada system for climate control and irrigation in greenhouses.
In Agroingenierı́a 2015, 2015. (Accepted).
[107] H. Mohr and P. Schopfer. Plant Physiol. Springer, Berlin, 1995.
[108] J. Monteith. Evaporation and environment. the state and movement
of water in living organism. In 19 th Symposium of the Society of
Experimental Biology, pages 205–234, 1965.
[109] J. Montero, A. Antón, P. Munoz, and P. Lorenzo. Transpiration from
geranium grown under high temperatures and low humidities in green-
houses. Agr Forest Meteorol, page 323332, 2001.
[110] J.I. Montero, J.C. López, and M. Teitel. Developments in covering
materials for intensive horticulture: Technical properties and recycling
opportunities. Acta Hortic, 1015:269–280, 2014.
[111] A.P. Montoya, J.L. Guzmán, F. Rodrı́guez, and J.A. Sánchez-Molina.
Hybrid modeling and control of greenhouse temperature through aerial-
pipes and air-fan heater systems. In Agricultural Engineering - CIGR-
AgEng 2012: Agriculture and Engineering for a Healthier Life, page
C1392.
[112] A.P. Montoya, J.L. Guzmán, F. Rodrı́guez, and J.A. Sánchez-Molina.
Temperature control in greenhouses using natural ventilation: Dis-
trubances compensation by feedfoward control. In Actas de Horticul-
tura, volume 64. Orihuela (Spain), February 20-22, 2014.
[113] A.D. Moore, J.R. Donnelly, and M. Freer. Grazplan: Decision support
systems for australian grazing enterprises. iii. pasture growth and soil
moisture submodels, and the grassgro {DSS}. Agr Syst, 55(4):535–582,
1997.
[114] J.C. Moreno, M. Berenguel, F. Rodrı́guez, and A. Baos. Robust control
of greenhouse climate exploiting measurable disturbances. Proc. 15th
IFAC World Congress, page 365, 2002.
[115] Y. Mualem. A new model for predicting the hydraulic conductivity of
unsaturated porous media. Water Resour Re, 12(3):513–522, 1976.
[116] M. Nachidi, A. Benzaouia, and F. Tadeo. Temperature and humid-
ity control in greenhouses using the Takagi-Sugeno fuzzy model. In
IEEE International Conference on Control Applications, pages 2150–
2154. Munich, Germany, 2006.
[117] E. M. Nederhoff. Technical aspects, management and control of co2
enrichment in greenhouses. Acta Hortic, 268:127–138, 1990.
References 153
[118] E.M. Nederhoff, A.A. Rijsdijk, and R. de Graaf. Leaf conductance and
rate of crop transpiration of greenhouse grown sweet pepper (capsicum
annuum l.) as affected by carbon dioxide. Sci Hortic, 52(4):283 – 301,
1992.
[119] Ingwald Obernberger and Gerold Thek. Physical characterisation and
chemical composition of densified biomass fuels with regard to their
combustion behaviour. Biomass Bioenerg, 27(6):653 – 669, 2004.
[120] J.P. Parra, E. Baeza, J.I. Montero, and B.J. Bailey. Natural ventilation
of parral greenhouses. Biosyst Eng, 87(3):355–366, 2004.
[121] G.D. Pasgianos, K.G. Arvanitis, P. Polycarpou, and N. Sigrimis. A non-
linear feedback technique for greenhouse environmental control. Com-
put Electron Agr, 40(1-3):153–177, 2003.
[122] M. Pérez and J. A. Sánchez-Molina. Energı́as renovables en los inver-
naderos. Cuadernos de estudios agroalimentarios (CEA CAJAMAR),
03:181–210, 2012a.
[123] M. Pérez, J. A. Sánchez-Molina, J. Pérez, and A. Callejón. Experiences
with the exploitation of photovoltaics on greenhouse covers. Plasticul-
ture, 9:24–38, 2012b.
[124] J. Pérez-Parra, A. Ramı́rez-Arias, M. Berenguel, and F. Rodrı́guez.
Ventilation rate models of mediterranean greenhouses for control pur-
poses. Acta Hortic, 719:197–204, 2006.
[125] F. Pianosi and T. Wagener. A simple and efficient method for global
sensitivity analysis based oncumulative distribution functions. Environ
Model Soft, 67:1–11, 2015.
[126] J.C. Ponsart, D. Theilliol, and C. Aubrun. Virtual sensors design for
active fault tolerant control system applied to a winding machine. Con-
trol Eng Pract, 18(9):1037 – 1044, 2010.
[127] R. Rabbinge, S.A. Ward, and H.H. Van Laar. Simulation and systems
management in crop protection. Plant Growth Regul, 10:169–170, 1991.
[128] A. Ramı́rez, Rodrı́guez, M. Berenguel, and E. Heuvelink. Calibration
and validation of complex and simplified tomato growth models for
control purposes in the southeast of spain. Acta Hortic, 654:147154,
2004.
[129] A. Ramı́rez-Arias, F. Rodrı́guez, J.L. Guzmán, and M. Berenguel. Mul-
tiobjective hierarchical control architecture for greenhouse crop growth.
Automatica, 48(3):490 – 498, 2012.
[130] J.A. Ramı́rez-Arias. Hierarchical multiobjective control of greenhouse
crop production. PhD thesis, University of Almerı́a, 2005. (in spanish).
[131] C. Rodrı́guez, J.L. Guzmán, F. Rodrı́guez, M. Berenguel, and M.R.
Arahal. Diurnal greenhouse temperature control with predictive con-
trol and online constraints mapping. In IFAC Conference on Control
Methodologies and Technology for Energy Efficiency. Vilamoura, Por-
tugal, 2010.
[132] F. Rodrı́guez. Modelling and Hierarchical Control of Greenhouse Crop
Production. PhD thesis, University of Almerı́a, 2002. In Spanish.
154 7 Conclusions and future works
ing high pressure fog systems and phytocontrol strategy. Acta Hortic,
797:279–284, 2008.
[158] I. Seginer. Some artificial neural network applications to greenhouse
environmental control. Comput Electron Agr, 18:167–186, 1997.
[159] I. Seginer. The penmanmonteith evapotranspiration equation as an
element in greenhouse ventilation design. Biosyst Eng, 82 (4):423439,
2002.
[160] I. Seginer, T. Boulard, and B.J. Bailey. Neural network models of the
greenhouse climate. J Agr Eng Res, 59:203–216, 1994.
[161] I. Seginer and A. Sher. Optimal greenhouse temperature trajectories for
a multi-state variable tomato model., pages 153–172. The computerized
greenhouse. Eds. Hashimoto Y., Bot G., Tantau H.J., Day W., Nonami
H. Academic Press, 1993.
[162] J.H. Shin, J.S. Park, and J.E. Son. Estimating the actual transpiration
rate with compensated levels of accumulated radiation for the efficient
irrigation of soilless cultures of paprika plants. Agric Water Manage,
135:9–18, 2014.
[163] T.H. Short and G. Van Duyne. Naturally ventilated greenhouses for
mediterranean climates. Acta Hortic, 434:229–235, 1996.
[164] N. Sigrimis, P. Antsaklis, and P. Groumpos. Control advances in agri-
culture and the environment. IEEE Contr Syst Mag, 21(5):8–12, 2001.
[165] N. Sigrimis, P.N. Paraskevopoulos, K.G. Arvanitis, and N. Rerras. Ad-
aptive temperature control in greenhouses based on multirate-output
controllers. In 14th IFAC World Congress, pages 571–576. Beijing,
China, 1999.
[166] C.J.T. Spitters, H. Van Keulen, and D.W.G. Van Kraalingen. A simple
and universal crop growth model simulator: Sucros87. Simulation
Monographs, 32:147–181, 1989.
[167] C. Stanghellini. Evapotranspiration in greenhouses with special refer-
ence to mediterranean conditions. 335.
[168] C. Stanghellini. Transpiration of Greenhouse Crops. IMAG Publica-
tion, Wageningen, Holland, 1987.
[169] C. Stanghellini and F. Kempkes. Steering of fogging: Control of humid-
ity, temperature or transpiration? Acta Hortic, 797:61–68, 2008.
[170] R. Suay, P. Martı́nez, D. Roca, M.Martı́nez, J. Herrero, and C. Ramos.
Measurement and estimation of transpiration of a soilless rose crop and
application to irrigation management. Acta Hortic, 614:625–630, 2003.
[171] L. Taiz and E. Zeiger. Plant Physiology. Sinauer Associates Publishers,
USA, 1998.
[172] H.J. Tantau and D. Lange. Greenhouse climate control: an approach
for integrated pest management. Comput Electron Agr, 40(13):141–152,
2003.
[173] F. Tap. Economics-based optimal control of greenhouse tomato crop
production. PhD thesis, Wageningen University. The Netherlands, 2000.
References 157
[190] G. Van Straten and E.J. Van Henten. Optimal greenhouse cultivation
control: Survey and perspectives. In IFAC Proceedings Volumes (IFAC-
PapersOnline), volume 3, 2010.
[191] A. Van Vosselen, H. Verplancke, and E. Van Ranst. Assessing water
consumption of banana: Traditional versus modelling approach. Agr
Water Manage, 74(3):201–218, 2005.
[192] B.H.E. Vanthoor, C. Stanghellini, E.J. van Henten, and P.H.B. de Vis-
ser. A methodology for model-based greenhouse design: Part 1, a green-
house climate model for a broad range of designs and climates. Biosyst
Eng, 110(4):363 – 377, 2011.
[193] B. Von Elsner, D. Briassoulis, D. Waaijenberg, A. Mistriotis, Chr.
Von Zabeltitz, J. Gratraud, G. Russo, and R. Suay-Cortes. Review
of structural and functional characteristics of greenhouses in european
union countries: Part i, design requirements. J Agr Eng Res, 75(1):1–16,
2000a.
[194] W. Voogt, J. Kipp, R. de Graaf, and L. Spaans. A fertigation model
for glasshouse crops grown in soil. Acta Hortic, 537:495–502, 2000.
[195] J. Werther, M. Saenger, E.-U. Hartge, T. Ogada, and Z. Siagi. Com-
bustion of agricultural residues. Prog Energ Combust, 26(1):1–27, 2000.
[196] D Whitley, T Starkweather, and C Bogart. Genetic algorithms and
neural networks: optimizing connections and connectivity. Parallel
Comput, 14(3):347 – 361, 1990.
[197] C.P. Yialouris, H.C. Passam, and A.B. Sideridis nad C. Métin. A multi-
lingual expert system for the diagnosis of pests, diseases and nutritional
disorders of six greenhouse vegetables. Comput Electron Agr, 19(1):55–
67, 1997.
[198] Seung-Hwan Yoo, Jin-Yong Choi, and Min-Won Jang. Estimation
of design water requirement using fao penman-monteith and optimal
probability distribution function in south korea. Agr Water Manage,
95(7):845 – 853, 2008.
[199] Q. Zhang and C.H Wu. Application of fuzzy logic in an irrigation con-
trol system. In Proc. of IEEE International Conference on Industrial
Technology, pages 593–597, 1996.
[200] Zhiming Zhang, Lihong Liu, Min Zhang, Yongsong Zhang, and Qiaomei
Wang. Effect of carbon dioxide enrichment on health-promoting com-
pounds and organoleptic properties of tomato fruits grown in green-
house. Food Chem, 153(0):157 – 163, 2014.