Tesis Rodrigo Gil

Rodrigo
Rodrigo Gil
Rodrigo Gil
Rodrigo Gil
Rodrigo
Rodrigo Gil
Rodrigo Gil
Rodrigo
GilGil GilRodrigo Gil A MODEL-BASED
FACULTY OF BIOSCIENCE ENGINEERING
FACULTY
FACULTY
FACULTY
FACULTY
FACULTY DEPARTMENT
OF BIOSCIENCE
OF
FACULTY
OF
BIOSCIENCE
FACULTY
OF
OF
BIOSCIENCE
BIOSCIENCE
BIOSCIENCE
OFENGINEERING
OF OF
BIOSCIENCE
ENGINEERING
BIOSCIENCEBIOSYSTEMS
ENGINEERING
ENGINEERING
ENGINEERING
ENGINEERING
ENGINEERING ARENBERG DOCTORAL SCHOOL
DEPARTMENT
DEPARTMENT
DEPARTMENT
DEPARTMENT
DEPARTMENT
OF
DEPARTMENT
DEPARTMENT
BIOSYSTEMS
OFOF BIOSYSTEMS
OF
OF BIOSYSTEMS
BIOSYSTEMSMEBIOS
BIOSYSTEMS
OFOF BIOSYSTEMS
BIOSYSTEMS ARENBERG
ARENBERG
ARENBERG
ARENBERG
ARENBERG
ARENBERG
ARENBERG
DOCTORAL
DOCTORAL
DOCTORAL
DOCTORAL
DOCTORAL
DOCTORAL
DOCTORAL
SCHOOL
SCHOOL
SCHOOL
SCHOOL
SCHOOL
SCHOOL
SCHOO
Celestijnenlaan
FACULTY OF BIOSCIENCE
MEBIOS200e
MEBIOS box
MEBIOS
MEBIOS2411
MEBIOS
ENGINEERING
MEBIOS
MEBIOS FACULTY OF BIOSCIENCE ENGINEERING
Celestijnenlaan
DEPARTMENT
B-3001
B-3001
B-3001
Celestijnenlaan
Celestijnenlaan
Celestijnenlaan
Celestijnenlaan
200e
B-3001
Leuven,
B-3001
B-3001 tel.
Leuven,
OF+
Leuven,
Celestijnenlaan
Celestijnenlaan
200e
box200e
200e
Leuven,
200e
2411
boxbox
32
B-3001
Belgium
Leuven,
Leuven,
B-3001
Belgium
2411
box
box
32
Belgium
200e
2411
BIOSYSTEMS
16
2411
200e
2411
boxbox
00800
Belgium
Leuven,
Belgium
Belgium
Leuven,
2411
2411
Belgium
Belgium
FACULTY
FACULTY
FACULTY
FACULTY
FACULTY
FACULTY
OF
FACULTY
OF
BIOSCIENCE
OF
OF
OF
BIOSCIENCE
BIOSCIENCE
BIOSCIENCE
BIOSCIENCE
OF
ARENBERGOFBIOSCIENCE
BIOSCIENCE
ENGINEERING
ENGINEERING
ENGINEERING
ENGINEERING
ENGINEERING
DOCTORALENGINEERING
ENGINEERIN
SCHOOL
rodrigo.gil@kuleuven.be MEBIOS
tel. +tel.
32tel.
+16
tel.
tel.
32
+3232
+16
+ 32
00800
32
16
32
tel.
16
16
32
tel.
00800
+32
32
32
00800
+00800
00800
32
16 16
32 32
00800
00800
Celestijnenlaan 200e box 2411
www.biw.kuleuven.be/biosyst/mebios
rodrigo.gil@kuleuven.be
B-3001 Leuven, Belgium
FACULTY OF BIOSCIENCE ENGINEERING
FACULTY OF tel.BIOSCIENCE
+ 32 16 32ENGINEERING
00800
AA
DEPARTMENT OF BIOSYSTEMS
rodrigo.gil@kuleuven.be ARENBERG DOCTORAL SCHOOL
A MODEL-BASED
A NITROGEN
A MODEL-BASED
A MODEL-BASED APPROACH FOR A LIFE CYCLE ASSESSMENT OF THE
MEBIOS
MODEL-BASED
Celestijnenlaan 200e box 2411 FACULTY OF BIOSCIENCE ENGINEERING
MODEL-BASED
MODEL-BASED
B-3001 Leuven, Belgium
NITROGEN EMISSIONS OF TOMATO PRODUCTION IN COLOMBIA

NITROGEN EMISSIONS
NITROGEN
NITROGEN
NITROGEN EMISSIONS OF TOMATO PRODUCTION IN COLOMBIA

NITROGEN EMISSIONS
tel. + 32 16 32 00800
A MODEL-BASED
NITROGEN
A MODEL-BASED
NITROGEN
EMISSIONS
EMISSIONS
EMISSIONS
APPROACH
A model-based approach for a life cycle
APPROACH
APPROACH
APPROACH
APPROACH
APPROACH
EMISSIONS
A model-based
AAmodel-based
Amodel-based
model-based
AAmodel-based
model-based
assessment approach
approach
of approach
approach
the approach
approach
forfor
nitrogenafor
for
life
aa life
afor
cycle
life
for
life
cycle
emissionsacycle
acycle
life
lifecycle
ofcycle
tomato
EMISSIONS
APPROACH
assessment
assessment
assessment
assessment
A assessment
assessment
of
model-based of
the
ofof
the
the
nitrogen
the
of
nitrogen
of
nitrogen
the
nitrogen
the
approach nitrogen
emissions
nitrogen
emissions
emissions
foremissions
a emissions
emissions
lifeof of
tomato
cycleof
of
tomato
tomato
tomato
ofoftomato
tomato
APPROACH
productionapproach
A model-based in Colombiafor a life cycle
production
production
productioninofin
production
productioninColombia
production
Colombia
inColombia
Colombia
inthe
inColombia
Colombia
OF
assessment
assessment the of
nitrogen nitrogen
emissions emissions
of tomato of tomato
OFFOR
OF
OF
OFFOR
FOR
TOMATO
production in Colombia
production in Colombia
FOR
FOR
FOR
OF
TOMATO
TOMATO
TOMATO
TOMATO
OFTOMATO
FOR
AFOR
TOMATO
A LIFE
A LIFE
A LIFE
A
A LIFE
LIFE
A
LIFE
LIFE
PRODUCTION
ACYCLE
PRODUCTION
PRODUCTION
PRODUCTION
PRODUCTION
LIFE
PRODUCTION
CYCLE ASSESSMENT
CYCLE
CYCLE
CYCLE
CYCLE ASSESSMENT
CYCLE
PRODUCTION
CYCLE
ASSESSMENT
ASSESSMENT
ASSESSMENT
ASSESSMENT
ASSESSMENT
IN
INCOLOMBIA
ASSESSMENT
IN COLOMBIA
IN
IN
IN COLOMBIA
COLOMBIA
COLOMBIA
COLOMBIA
IN COLOMBIA
OFOF
THE
OF THE
OF THE
OF THE
OF
OF THE
THE
THE
OF THE
Rodrigo Gil
Dissertation presented in partial
Supervisors:
fulfilment of the requirements for the Rodrigo Gil
Prof. Dr. E. Schrevens Rodrigo
Rodrigo
Rodrigo
Rodrigo
Rodrigo
Gil
Rodrigo
Gil
Gil
Gil Gil
Gil
DecemberDecember
Prof. Dr. C.R. Bojacá degree of Doctor of Bioscience

Supervisors: Dissertation Rodrigo
presented
Engineering Gil
in partial
Supervisors:
Supervisors:
Supervisors:
Supervisors:
Supervisors: Dissertation
Dissertation
Dissertation
Dissertation
Dissertation
fulfilment Dissertation
Dissertation
ofpresented
presented
the presented
presented
presented
presented
inpresented
requirementspartial
ininpartial
in
inpartial
partial
partial
inthe
for inpartial
partia
Prof.Supervisors:
Supervisors:
Dr. E. Schrevens December 2019
fulfilment
fulfilment
fulfilment
fulfilment
fulfilment
fulfilment
ofdegree
fulfilment
the
ofofthe
of
of
requirements
the
the
the
requirements
ofrequirements
of
requirements
requirements
thetherequirements
requirements
offor for
the
for
for
for
the
the
the
the
forforthe
the
2019
Prof.
Prof.
Prof.
Dr.
Prof.
Prof.
Dr.
E.Dr.
Prof.Prof.
Schrevens
Dr.
E.
Dr.
Prof.
E.
Dr.Schrevens
E.
E.
Schrevens
Dr.
C.R.Schrevens
Schrevens
Dr.
E.E.
Schrevens
Schrevens
Bojacá Dissertation of Doctor
presented Bioscience
in partial
December 2019
December 2019
December 2019
December
December 2019
December 2019
Prof. Supervisors:
Prof.
Dr.
Prof.
Prof.
Prof. C.R.
Dr.
Dr.Prof.
Dr.
Dr.
C.R.
Bojacá
C.R.C.R.
C.R.
Dr.
Prof. Bojacá
C.R.
Bojacá
Bojacá
Bojacá
Dr. Bojacá
C.R. Bojacá degree degree
degree
degree
degree of degree
Doctor
ofdegree
ofDoctor
of
ofDoctor
Doctor
Doctor
ofofDoctor
Bioscience
ofDoctor
ofBioscience
of
ofBioscience
Bioscience
Bioscience
of of
Bioscience
Bioscience
Prof. Dr. E. Schrevens fulfilment of the requirements Engineering
for the
Engineering
Engineering
Engineering
Engineering
Engineering Engineering
Engineering
December
Prof. Dr. C.R. Bojacá degree of Doctor of Bioscience

December
Engineering 2019
2019
December
December
December
December
December
December
2019
December
2019
2019
2019
20192019 2019
20192019
December 2019
A MODEL-BASED APPROACH FOR A LIFE CYCLE
ASSESSMENT OF THE NITROGEN EMISSIONS OF
TOMATO PRODUCTION IN COLOMBIA
Rodrigo GIL
Supervisor(s):
Prof. Dr. E. Schrevens
Prof. Dr. C.R. Bojacá
Members of the Examination

Committee:
Prof Dr ir J. Delcour Dissertation presented in
Prof Dr ir L. Duchateau partial fulfilment of the
Dr ir R. De Vis requirements for the degree
Prof Dr ir H. Vandendriessche of Doctor in Bioscience
Prof Dr ir R. Merckx Engineering
December 2019
Doctoraatsproefschrift nr. 1618 aan de faculteit Bio-ingenieurswetenschappen van
de KU Leuven
© 2019 Rodrigo GIL

Uitgegeven in eigen beheer, Rodrigo Gil, Bogotá
Alle rechten voorbehouden. Niets uit deze uitgave mag worden vermenigvuldigd
en/of openbaar gemaakt worden door middel van druk, fotokopie, microfilm,
elektronisch of op welke andere wijze ook zonder voorafgaandelijke schriftelijke
toestemming van de uitgever.
All rights reserved. No part of the publication may be reproduced in any form by print,
photoprint, microfilm, electronic or any other means without written permission from
the publisher.
Acknowledgements
Education has been my choice to live a worthy life. Obtaining this Ph.D. degree, I
hope to be able to provide a deserving life to my family as well. However, an
achievement of this magnitude is the result of a sum of efforts. Throughout these
years, I received support from my promotors, colleagues and family, and also, I had
important institutional support. I want to express my gratitude to each of them.
First, I want to acknowledge my promotor Eddie Schrevens for opening the doors of
this wonderful University and for his support throughout these years. To my friend,
colleague and mentor Carlos Bojacá who has unconditionally supported this process
and from whom I learned most of my professional skills. I would have never achieved
it without his determined support and guidance. Thanks Carlos. To the members of
the examination committee for their constructive feedback that improved
considerably the quality of the work. To my colleagues and friends, Luz Stella, Luis
Alejandro, Hugo who with words of encouragement and advice have contributed to
the fulfillment of this attainment. My colleagues from the Department of Basic
Sciences and Modeling, especially Miguel Ángel, Gladys and Adelina, with whom I
shared fears and frustrations during the doctoral training process. My Belgian and
Peruvian colleagues, Reindert, Dries, Diego, Ximena, Cindybell, Diana, Marcos,
Eduardo and Karin with whom I shared special moments in these distant lands.
La familia que ha sido pilar fundamental de este proceso. Quiero reconocer el apoyo
de mi esposa Adriana, su amor ha sido fundamental para cumplir esta meta. A María
Paula quien pese a su corta edad ha comprendido lo que estoy haciendo, y me ha
apoyado con sus pequeños pero significativos gestos. A Juan y Nico por ser una
inspiración más. A mi mamá por su entrañable amor y devoción. A mis padrinos
Guillermo y Ayda, a Johana, Fabián, David, la tía Leonilde, a Katty, a Jorge, gracias
por ese cariño desinteresado. A mis suegros Beyer y Susana, a Marcela, Liceth,
Nelson, quienes me han acogido en el seno de la familia. A mi tía Esther, Paola y Juan
David quienes desde Santander siempre han estado al tanto de mis avances y
retrocesos. A mi abuelita Matilde. A quienes me apoyaron en el pasado pero partieron,
especialmente a mi abuelito Adán y mi tío Roberto. A mi papá, a Luis Eduardo,
Gloria, Diego, Oscar y Catherine, gracias, especialmente por el apoyo en los últimos
meses. Nuevamente a Carlos Bojacá a quien considero mi hermano y por lo tanto
parte de mi familia, espero que sigamos recorriendo juntos los caminos de la vida.
Finally, I want to acknowledge to the Flemish Interuniversity Council, Colciencias,
KU Leuven and the University of Bogotá Jorge Tadeo Lozano for its financial support
that allowed the successful completion of the doctorate.
i
Summary
This thesis develops methodologies for a technical sustainability assessment of
agricultural systems through the integration of soil-crop modeling and standard life
cycle assessment (LCA). In addition, we propose a strategy to facilitate
communication of the results of the comparative LCAs to stakeholders. As case
studies, the proposed methods are applied to open field and greenhouse tomato
production systems in the Colombian Andes, focusing on nitrogen fertilization
practices.
The thesis begins with a detailed characterization of the selected tomato production
systems under both open field and greenhouse conditions. This characterization was
necessary since the information available for these smallholder-based cropping
systems is very limited in the Colombian context. This characterization included
biophysical factors related to climatic conditions and soil fertility level, and an
inventory of inputs, management practices, and productivity factors. Two data
collection tools, namely, surveys and a procedure based on direct observation of the
production cycles referred to as detailed follow-ups or input-output accounting, were
used to gather the information from two major production regions. Since local
growers usually do not keep records of their activities, detailed follow-ups are
recommended for obtaining reliable data under these conditions, despite their cost and
time-consuming nature. Nevertheless, as a first screening of the production systems,
surveys are very useful. As a result of this first phase, some features of the growers
were highlighted, such as a high dose of fertilizers without a proportional increase in
yield. Heterogeneity with respect to management practices and varying levels of yield
gaps, especially in the open field system, were also evident.
A comparative LCA between the two systems was carried out and included the
infrastructure, machinery, pest management, and fertilization subsystems. In this case
study, the field emissions derived from the nitrogen fertilizers were obtained through
standard methods based on empirical regression meta-analysis allowing the
modification of cropping parameters as a function of the production system. The
outcome of the LCA was not clear enough to differentiate the environmental
performance among the two production systems. For instance, the greenhouse system
showed better environmental performance according to indicators such as
acidification and eutrophication, while the opposite was observed with indicators such
as global warming potential and human toxicity potential. In this section, the
stochastic multi-attribute analysis (SMAA) was introduced as a way to build a single
indicator that would objectively rank the environmental performance of the two
systems. However, an important constraint not considered by the SMAA method had
to be overcome first, namely, the high degree of correlation between environmental
indicators. Thus, the copula method was introduced to fit the required joint-statistical
distributions for all environmental indicator based on the estimated marginal
distributions of the individual impact categories. The successful implementation of
the SMAA method, supported by the copula method, resulted in the better overall
environmental performance of the open field than the greenhouse system. In
conclusion, intensification of tomato production in Colombia by shifting to protected
ii
conditions has led to a relatively high environmental impact because of inadequate
and inefficient implementation of greenhouse technology. As a consequence, it is
necessary that greenhouse farmers optimize resource use and greenhouse
methodologies to achieve higher production and lower environmental impact.
To evaluate strategies that improve the environmental performance of tomato
production systems in terms of nitrogen fertilization, a tomato plant growth and
development model was calibrated and validated for open field and greenhouse
conditions. In the first stage, the growth and development of tomato plants under
Colombian conditions was effectively simulated by the calibrated model. This plant
model was adapted by incorporating modules to represent root growth and nitrogen
demand to couple it to a soil model. The soil model was previously calibrated and
validated under Belgian conditions and can represent water and nitrogen dynamics as
well as all N and C processes. The soil-plant model was validated for Colombian
conditions based on trials carried out on commercial fields. Although the validation
results presented varying degrees of fit, overall, the model showed a workable level
of accuracy. The resulting soil-plant model was driven by meteorological data, soil
and plant properties and management practices such as nitrogen fertilization and
irrigation. The model simulated, on a daily basis, crop growth and development, heat,
water and nitrogen flows through the soil profile, and soil nitrogen and carbon
dynamics, including emissions of environmental pollutants to the atmosphere and
groundwater.
Finally, the soil-plant model was used to evaluate a set of nitrogen fertilization
scenarios. The scenarios included currently applied strategies by commercial growers
and two alternative prospective scenarios. In the grower scenario, nitrogen
fertilization was applied based on a fixed dose, while the alternative scenarios applied
nitrogen as a function of plant demand over time. Each of the scenarios was evaluated
over 52 weather periods, corresponding to weekly plantings over a period of one year.
From each simulation, the predicted yield and the emissions of ammonia, nitrous
oxide, and nitrates were extracted and used as inputs for a posterior comparative LCA.
The results revealed that some scenarios showed better environmental performance
for some impact categories, while others performed better in a different set of
categories. Therefore, the SMAA method supplemented by the copula method was
applied again to rank the environmental performance of the scenarios. The results
showed that greenhouse production systems in which the fertilization strategy is based
on plant demands have the best environmental performance. The method proposed in
this thesis adds dynamic properties to the classic static LCA in terms of predicting
nitrogen emissions through integration with soil-crop modeling. Additionally, the
proposed framework incorporates an optimized single performance indicator method
to enhance communication of the LCA results to stakeholders and decision-makers.
iii
Samenvatting
Deze thesis ontwikkelde innovatieve methodologie voor de evaluatie van de
technische duurzaamheid van tuinbouwsystemen door integratie van bodem-gewas
interactiemodellen met standaard levenscyclus analyse (LCA). De methodologie
wordt toegepast op serre- en vollegronds tomatenteelt in twee concentratiezones in de
Colombiaanse Andes, waarbij de focus vooral ligt op de verschillende strategieën
voor stikstofbemesting.
De thesis gaat van start met een gedetailleerde beschrijving van beide systemen,
gezien er zo goed als geen data beschikbaar zijn met betrekking tot de kleinschalige
tuinbouw in de Colombiaanse Andes. Beide systemen werden biofysisch
gekarakteriseerd door een gedetailleerde beschrijving van het klimaat, de
bodemcondities en de cultuurtechnische beheersstrategie. Dit resulteerde in een
volledige boekhouding van alle biofysische in- en outputs van de systemen. Voor de
datacollectie werden enerzijds gestructureerde enquêtes en anderzijds
tweewekelijkse, directe metingen op bedrijfs- en veldniveau uitgevoerd. Als
belangrijkste resultaat werd gevonden dat in beide systemen de Colombiaanse
tomatenteler extreem hoge bemestingsdosissen gebruikt zonder dat dit resulteert in
hogere producties. Tevens werd een bijzonder grote heterogeniteit vastgested voor
managementpraktijken, vooral in de vollegrondsteelt.
In het tweede deel werd een vergelijkende LCA tussen beide systemen uitgevoerd
voor de volgende subsystemen: infrastructuur, mechanisatie, ziekte- en
plaagbeheersing en bemesting. De veldemissies veroorzaakt door de
stikstofbemesting werden met de standaard, empirische modellen, gebaseerd op meta-
data regressie-analyse, berekend. De serresystemen vertoonden een lager
eutrofierings- en verzuringspotentieel, terwijl ze duidelijk hoger scoorden dan de
vollegrondsteelt voor klimaatsopwarming en menselijke toxiciteit. In dit deel werd
‘Stochastic Multi-attribute Analysis’ (SMAA) geïntroduceerd als method om een
geïntegreerde milieu-impactindex te berekenen, waarmee de serre- en
vollegrondsteelt kunnen vergeleken worden, rekening houdend met de variantie- en
correlatiestructuur tussen de verschillende LCA impactcategorieën. Deze structuur
werd berekend met de copula methode, die de gemeenschappelijke kansdistributie
van alle milieu-indicatoren schat op basis van de geschatte marginale distributies van
de individuele impactcategorieën. De SMAA methode toonde aan dat de
vollegrondsteelt aanzienlijk minder milieu-impact genereerde dan de serreteelt. Er
kan geconcludeerd worden dat intensifiëren van de tomatenteelt, door gebruik te
maken van serres, niet de verhoopte productieverhoging, nodig om de milieu-impact
te reduceren per ton tomaten, heeft veroorzaakt. Dus het blijft hoogstnoodzakelijk om
de serretelers de nodige kennis en expertise bij te brengen zodat hun productie beter
aansluit bij de potentieel mogelijke productie in hun regio.
Voor de procesgebaseerde evaluatie van strategieën die de milieu-impact van de
stikstofbemesting in de tomatenteelt kunnen reduceren werd een gewasgroei- en
ontwikkelingsmodel voor tomaat gecalibreerd en gevalideerd voor serre- en
vollegrondsteelt. Dit gewasmodel werd uitgebreid met modules die de wortelgroei en
stikstofvraag van het gewas kunnen simuleren. Tevens werd dit gewasmodel
iv
gekoppeld aan een bodemmodel dat transport van stikstof, water en warmte samen
met de stikstof-koolstof processen kan simuleren. Dit bodemmodel werd voordien
gecalibreerd onder Belgische condities. Het bodem-gewasmodel werd gevalideerd
onder Colombiaanse condities op basis van potentiële productie-experimenten,
uitgevoerd bij verschillende telers zowel in serre als in volle grond. Dit model vertrekt
van metereologische-, bodem- en gewas-data en kan het effect van verschillende
bemestingsscenario’s voor stikstof simuleren op groei en ontwikkleing van de plant,
op verschillende toestandsvariabelen van de bodem, zoals water, stikstof en koolstof
concentraties in de verschillende bodemlagen en op de verschillende stikstofemissies
naar water en lucht.
Tenslotte werden met dit bodem-gewas interactiemodel de volgende
bemestingscenario’s voor stikstof gesimuleerd: de standaard bemesting van een
commerciële teler, twee scenario’s gebaseerd op de stikstofvraag van het gewas.
Ieder bemestingsscenario werd geëvalueerd bij 52 verschillende planttijdstippen ,
namelijk iedere week van het jaar, met als doel de klimaatseffecten op de teelt te
onderzoeken. Voor iedere simulatie werd de totale biomassa, de productie en de
fenologie gesimuleerd, alsook de bodemvariabelen en de emissies van ammoniak,
lachgas en nitraat. Deze emissies werden ingevoerd in een vergelijkende LCA
zodanig dat de empirische functies van deel twee nu vervangen kunnen worden door
dynamische, procesgebaseerde modelresultaten. De SMAA methode, gebaseerd op
een copula schatting van de gemeenschappelijke kansdistributie van all
impactcategorieën, berekende voor alle scenario’s één geïntegreerde milieu-indicator
om beide productiesytemen op een objectieve wijze te vergelijken. Uit de resultaten
blijkt dat een serreteelt onder optimale bemesting, gestuurd door de vraag van het
gewas, een aanzienlijk kleinere milieubelasting veroorzaakt dan de vollegrondsteelt.
De voorgestelde methode breidt de klassieke statische LCA uit met dynamische,
procesgebaseerde kennis op vlak van de predictie van stikstofemissies door de
koppeling van LCA met een bodem-gewas interactiemodel. Additioneel ontwikkelde
deze thesis een methode om de verschillende impactcategorieën te integreren naar één
globale milieu-indicator op basis van de copula- en SMAA methode.
v
List of abbreviations and symbols
!" Scale parameter for fruit dry matter function
!# Scale parameter for leaf dry matter function
!$" Scale parameter for dry matter function of ripe fruits
%& Empirical intercept for tropical climates (0.25)
%' Empirical slope for tropical climates (0.45)
%" Midpoint parameter for fruit dry matter function
%# Midpoint parameter for leaf dry matter function
%$" Midpoint to estimate the dry matter function of ripe fruits
(" Asymptote for the fruit dry matter function
(# Asymptote for the dry matter function
)* Covariance matrix
($" Asymptote for the dry matter function of ripe fruits
)+, Carbon dioxide concentration in air (ppmv)
-# Displacements parameter to leaf dry matter function
./$ Daily dry matter allocated to roots (g DM d-1)
.01 Deep percolation (cm) on the ith day
2345 Dry matter of ripe fruits (g DM)
607 Daily gross photosynthesis (g CH2O d-1)
603 Hourly gross photosynthesis (g CH2O h-1)
895 Extraterrestrial solar radiation (J m-2 day-1)
81 Water intercepted by the foliage (cm) on the ith day
85 Daily total solar radiation above plant canopies (J m-2 day-1)
:;<,1 Nitrogen inputs for inorganic fertilizers on the ith day
8>1 Irrigation (cm)
M@ Median from a sample
/ABC Maintenance respiration per day (g CH2O d-1)
/!DAEE5F9G53 Maximum rooting depth (cm)
/81 Nitrogen released by the mineralization process on the ith day
H;& Mineral N-soil content at beginning of the cycle
H;I Mineral N-soil content at the end of the cycle
;,,1 Dinitrogen volatilization (kg N ha-1) on the ith day
;, +1 Nitrous oxide (kg N ha-1) generated during the nitrification and
denitrification processes on the ith day
;J9KL3,1 Nitrogen losses due to leaching (kg N ha-1) on the ith day
;+M,1 Nitrogen oxides emitted (kg N ha-1) on the ith day
+1 Experimental observed value
vi
+N Experimental observations average
O;<,1 Nitrogen inputs for organic fertilizers on the ith day
01 Precipitation (cm) on the ith day
0*PQ Maximum leaf photosynthetic rate (μmol CO2 m-2 s-1)
0RK,1 Actual plant transpiration (cm) on the ith day
S9 Leaf quantum efficiency (μmol CO2 μmol-1 photon)
>;T1 Plant root N-uptake (kg N ha-1) on the ith day
>+1 Surface runoff (cm) on the ith day
U1 Model simulated value on the ith day
UVK,1 Actual soil evaporation (cm) on the ith day
UV;# Rate of senescence for leaves
UW)& Soil water content at the beginning of the cycle
UW)I Soil water content at the end of the cycle
R9" Daily effective temperature (°C)
R./" Total dry matter in fruits
R./X# Total dry matter of photosynthetically active leaves (g DM)
R./# Total dry matter in leaves
R./$ Total dry matter in roots
R./Y Total dry matter in stems
R/0K4X Daily average temperature (°C)
T21 Amount of fertilizer used by each grower (kg ha-1)
Z;[\,1 Ammonium volatilization on the ith day
]^ Light extinction coefficient
]_ Leaf light transmission coefficient
xN Arithmetic mean from a sample
a1 Yield obtained by the ith grower (kg ha-1)
µ Mean parameter
s Standard deviation parameter
a Shape parameter for gamma distribution
b Scale parameter for gamma distribution
l Scale parameter for Weibull distribution
AD Abiotic depletion (kg Sb eq)
AEF Agronomic efficiency of fertilizer use
AIC Akaike Information Criterion
AP Acidification potential (kg SO2 eq)
efgH Asymptotic depth (cm)
AT Air temperature (°C)
ATT Accumulated thermal time (°Cd)
BD Bulk density (g cm-3)
vii
Bias Model bias
BIC Bayesian Information Criterion
C d-dimensional copula function
c Soil clay content (%)
C Carbon
C:N Carbon nitrogen ratio
Capp Correction factor for the degree of dilution (dimensionless)
CEC Cation exchange capacity (cmolc kg-1)
)h Effect of vapor pressure deficit on photosynthesis (kPa-1)
Cr Current N-fertilization practices carried out by local growers
)i Coefficient of variation
Cx Correction factor indicating the crop parameters (dimensionless)
DAT Days after transplanting
DEM Digital elevation model
DL Day length (h)
./j Dry matter in fruits (g DM)
./k Dry matter in leaves (g DM)
./l Total dry matter produced that day (g DM d-1)
./f Dry matter in stems (g DM)
DSSAT Decision Support System for Agrotechnology Transfer
ECss Electrical conductivity of saturated soil-pastes (dS m-1)
EF Model efficiency
EFAGRAMMON Ammonium emissions factor for the type of synthetic fertilizer
eN2O Potential emissions of nitrous oxide (kg N ha-1)
eN2ODEN nitrous oxide generated by denitrification (kg N ha-1)
eN2OFER Emissions due to organic and synthetic sources (kg N ha-1)
eN2ONIT Nitrous oxide generated by nitrification (kg N ha-1)
m;+3 Nitrate losses to water bodies
EP Eutrophication potential (kg PO4-3 eq)
er Ammonium emissions rate depending on the type of manure
2 d-dimensional cumulative density function
FAETP Fresh-water aquatic ecotoxicity potential (kg 1,4-DB eq)
FC Field capacity
Fr Application of a fixed dose of nitrogen
FWU Follow-up data collection strategy
GH Greenhouse production system
6>V2 Growth efficiency coefficient (g DM g-1 CH2O)
GWP Global warming potential (kg CO2 eq)
HTTP Human toxicity potential (kg 1,4-DB eq)
viii
IDEAM Colombian meteorological service agency
ieNH3 Emissions due to of inorganic fertilizers (kg NH3 ha-1)
iNF Nitrogen applied through inorganic fertilizers (kg total N ha-1)
K Potassium
k Shape parameter for Weibull distribution
KCl Potassium chloride
L Rooting depth (m)
LA Leaf area per plant
oe8 Leaf area index
LCA Life cycle assessment
LCI Life cycle inventory
LCIA Life cycle impact assessment
LL Lower limit
MAETP Marine aquatic ecotoxicity potential (kg 1,4-DB eq)
masl Meters above sea level
MCDA Multicriteria decision-support analysis
/. Mahalanobis distance
H:- Inflection point at ATT
N Nitrogen
Ncr Nitrous oxide emissions from the crop residues (kg N ha-1)
NH+4 Ammonium
NO-3 Nitrate
Norg Nitrogen applied through organic sources (kg N ha-1)
NSE Nash-Sutcliffe modelling efficiency
oeNH3 Ammonia emissions due to organic fertilizers (kg NH3 ha-1)
OF Open field production system
OLDP Ozone layer depletion potential (kg CFC-11 eq)
oNF Nitrous oxide emissions due to organic sources (kg N ha-1)
ONI Oceanic Niño Index
Op Fertilization based on the plant nitrogen demand
P Phosphorus
P Water inflows (mm)
PAR Photosynthetic active radiation
PBias Percent bias
PCA Principal component analysis
pdf Probability density function
06>V. A function that corrects the maximum leaf photosynthetic rate for
suboptimal temperatures
Pi Uncertainty-based preference
ix
0o/2 Plant density (plants m-2)
POP Photochemical oxidation potential (kg C2H4 eq)
002. Photosynthetic photon flux density (μmol photons m-2 s-1)
0i0. A function that corrects the maximum leaf photosynthetic rate for
the air vapor pressure deficit
Qi Uncertainty-based indifference
RAI Rank acceptability index
RH Relative humidity (%)
RLD Root-length density (cm cm-3)
>/>2 Respiration coefficient for growing fruits (g CH2O g-1 DM d-1)
>/>o Respiration coefficient for stem and leaves (g CH2O g-1 DM d-1)
RMSE Root mean square error
RSR Root mean square error – observed standard
f Duration of the sunshine (h day-1)
SAT Soil water saturation
f(!k Scale parameter for maximum root depth function
SLA Specific leaf area
SMAA Stochastic multiattribute analysis
SOC Soil organic carbon
UOk>!- Hourly solar radiation (J m-2 s-1)
SR Solar radiation (W m-2)
SRL Specific root length (cm/g)
SVY Survey data collection strategy
SWC Soil water content
t Ton
TAN Total ammonium content in the organic fertilizer (kg of NH+4 t-1)
TEP Terrestrial ecotoxicity potential (kg 1,4-DB eq)
R/0 Hourly mean temperature (°C)
U Nitrogen uptake by the crop (kg N ha-1)
i0. Air vapor pressure deficit (kPa)
i0.o Effect of vapor pressure deficit on photosynthesis (kPa)
WS Wind speed (m s-1)
Μ Multivariate arithmetic mean
r Carbon dioxide use efficiency (μmol CO2 m-2 s-1 ppmv-1)
x
Table of contents
Acknowledgements ............................................................................................... i
Summary .............................................................................................................. ii
Samenvatting ....................................................................................................... iv
List of abbreviations and symbols ...................................................................... vi
Table of contents ................................................................................................. xi
List of Figures ................................................................................................... xiii
List of Tables ..................................................................................................... xv
Chapter 1. General introduction and objectives ................................................... 1
1.1 Introduction ......................................................................................... 1
1.2 Objectives ............................................................................................ 4
Chapter 2. Study areas and data acquisition methods .......................................... 5
2.1. Introduction ......................................................................................... 5
2.2. Data ..................................................................................................... 7
2.3. Experimental Design, Materials, and Methods ................................. 11
Chapter 3. Tomato smallholder systems in the Colombian Andes .................... 15
Abstract .............................................................................................................. 15
3.1. Introduction ....................................................................................... 16
3.2. Materials and methods ...................................................................... 17
3.3. Results ............................................................................................... 22
3.4. Discussion ......................................................................................... 32
3.5. Conclusions ....................................................................................... 35
Chapter 4. Comparative LCA between open field and greenhouse systems ..... 36
4.1. Introduction ....................................................................................... 37
4.2. Methods ............................................................................................. 39
4.3. Results and discussion ....................................................................... 44
4.4. Conclusions ....................................................................................... 54
Chapter 5. A tailor-made crop growth model for the tomato in Colombia ........ 56
Abstract .............................................................................................................. 56
5.1. Introduction ....................................................................................... 57
5.3. Results and Discussion ...................................................................... 66
5.4. Conclusions ....................................................................................... 72
Chapter 6. Soil-plant model validation under Colombian conditions ................ 73
6.1. Introduction ....................................................................................... 74
6.3. Results and discussion ....................................................................... 80
xi
6.4. Conclusions ....................................................................................... 90
Chapter 7. Evaluation of nitrogen fertilization scenarios .................................. 91
7.1. Introduction ....................................................................................... 92
7.3. Results and discussion..................................................................... 101
7.4. Conclusions ..................................................................................... 115
Chapter 8. General conclusions ....................................................................... 116
Chapter 9. Future works ................................................................................... 120
References ........................................................................................................ 122
xii
List of Figures
Figure 1. Overview of the greenhouse tomato production system in the province of
Alto Ricaurte (Boyacá, Colombia). ............................................................................. 8
Figure 2. Overview of the open field tomato production system in the province of
Guanentá (Santander, Colombia). ............................................................................... 9
Figure 3. Elevation profile of a cross-section (southwest-northeast direction) of the
Guanenta province showing the main types of parent materials. .............................. 10
Figure 4. Elevation profile between the massif of Iguaque and the Merchán
Mountains, including the main types of parent materials. ........................................ 10
Figure 5. Tomato crop cycles observed during the follow-up data collection period in
the (A) OF and (B) GH systems. Each segment represents the cycle length from
transplanting until the end of harvest. ....................................................................... 21
Figure 6. Average monthly precipitation (A), daily temperature (B), relative humidity
(C) and solar radiation (D) of the production regions of Boyacá and Santander. For
precipitation, error bars represent ± the standard deviation, while for the other
variables are the shaded area is that indicates ± the standard deviation. .................. 24
Figure 7. Scaled PCA-biplot showing the variation and correlational structure of
physical and chemical soil properties........................................................................ 27
Figure 8. Graphical representation of correlations between environmental indicators
for tomato production under greenhouse (a) and open field (b) conditions. ............. 50
Figure 9. Cumulative distribution functions of the rankings for each tomato
production system. .................................................................................................... 53
Figure 10. Rank acceptability index for the tomato production systems under
greenhouse and open field conditions. ...................................................................... 54
Figure 11. Schematics of the proposed tomato crop growth model for the open field
and greenhouse production systems, which is a simplification of the TOMGRO model
(Jones et al., 1991)..................................................................................................... 59
Figure 12. Daily average air temperature (A), relative humidity (B) and solar radiation
(C) during the calibration experiments carried out under greenhouse and open field
conditions. ................................................................................................................. 67
Figure 13. (A) Greenhouse and (B) open field tomato dry matter partitioning fractions
as a function of thermal time for each of the plant organs. ....................................... 69
Figure 14. Observed and simulated dry matter accumulation and distribution
throughout the plant organs for the calibration experiments carried out under
greenhouse and open field conditions. Vertical bars represent the estimated standard
deviations. ................................................................................................................. 70
Figure 15. Schematic representation of the nitrogen processes considered in the soil
model. Adapted from Heuts (2018)........................................................................... 74
Figure 16. Nitrogen content in the tomato plant organs as a function of the
accumulated thermal time. ........................................................................................ 81
xiii
Figure 17. Maximum rooting depth of tomato plants as a function of accumulated
thermal time determined for greenhouse and open field conditions. The points
represent the observations. The line corresponds to the fitted logistic function. ...... 82
Figure 18. Probability distribution to control the root dry matter allocated to each layer
with root presence as a function of the accumulated thermal time. .......................... 82
Figure 19. Observed and simulated dry matter accumulation and partitioning and
nitrogen content of the plant organs for the validation experiments carried out under
greenhouse (A, C) and open field (B, D) conditions. ................................................ 84
Figure 20. Observed and simulated water content of the four soil layers for the open
field and greenhouse trials......................................................................................... 87
Figure 21. Simulated (lines) and measured (± standard deviations) nitrate and
ammonium contents at a 20 cm depth throughout the experimental cycle. The modeled
nitrogen contents were obtained by adding the predicted values for the first two layers
(0-10 and 10-20). ....................................................................................................... 88
Figure 22. Observed and simulated soil temperature at different depths and the
goodness-of-fit statistics used to evaluate the model performance. .......................... 89
Figure 23. A schematic overview of the coupled soil-crop model and LCA to assess
the environmental performance of N-fertilization scenarios................................... 100
Figure 24. Kernel probability density functions describing the tomato yield of the
scenarios evaluated in the greenhouse (A) and open field (B) systems. ................. 101
Figure 25. Emissions of ammonia (NH3), nitrous oxide (N2O) and nitrate (NO3-)
during the crop cycles simulated for each scenario in the greenhouse and open field
systems. ................................................................................................................... 103
Figure 26. Effect of the transplanting date on emissions of nitrate in the greenhouse
and open field systems. ........................................................................................... 105
Figure 27. Kernel probability density functions for each impact category in the
scenarios evaluated for the greenhouse system. ...................................................... 107
Figure 28. Probability density functions for each impact category in the scenarios
evaluated for the open field system. ........................................................................ 108
Figure 29. Cumulative distribution functions of the rankings for each nitrogen
fertilization scenario. ............................................................................................... 112
Figure 30. Rank acceptability index for each of the scenarios evaluated regarding
nitrogen fertilization. ............................................................................................... 113
xiv
List of Tables
Table 1. Metadata of collected information in open field and greenhouse production
systems. ....................................................................................................................... 6
Table 2. Monthly averages and annual values (± standard deviations) for the climate
variables used to describe the weather in the provinces of Guanentá (Santander) and
Alto Ricaurte (Boyacá).............................................................................................. 11
Table 3. Fertility characterization in long-term fallow soils used for tomato production
in open field (Santander) and under greenhouse (Boyacá) production systems. The
coefficient of variation (CV) is expressed as a percentage. ...................................... 25
Table 4. Soil properties comparison between uncultivated soils and those used for
tomato production under the GH system. The coefficient of variation (CV) is
expressed as a percentage. ......................................................................................... 28
Table 5. Main features of the open field and greenhouse tomato systems determined
through surveys and detailed follow-ups. ................................................................. 31
Table 6. Inventory data (mean and standard deviation) for the subsystems considered
for the LCA, and all are allocated to one ton of tomatoes. ....................................... 46
Table 7. Emissions (kg t-1 functional unit) due to the field application of inorganic
products applied in both tomato production systems. ............................................... 47
Table 8. Average of potential environmental impacts per ton of tomato produced on
open field and greenhouse conditions in Colombian Andes. .................................... 49
Table 9. Properties of the fitted statistical distributions to the environmental indicators
distributions estimated for the greenhouse and open field tomato production systems.
................................................................................................................................... 51
Table 10. Log-likelihood, Akaike information criterion (AIC) and Bayesian
information criterion (BIC) of the estimated D-vine models for the environmental
indicators of the greenhouse and open field production systems. ............................. 52
Table 11. Parameters included in the tomato growth model ..................................... 62
Table 12. General characteristics of the on-farm experiments used to calibrate the
tomato crop growth model. ....................................................................................... 64
Table 13. Fitted parameters for dry matter allocation in leaves (DMl) and fruits (DMf),
leaf senescence rate (SENl) and ripening fruit rate (Fhvt). ......................................... 68
Table 14. Goodness-of-fit measures of the simulated dry matter per plant and per
organ by the calibrated tomato crop growth model................................................... 71
Table 15. Summary of the main crop management parameters used as inputs for the
model. ........................................................................................................................ 77
Table 16. Properties of the two soil profiles used in the validation process. ............ 78
Table 17. Goodness-of-fit statistics used to validate the model predictions. ............ 80
Table 18. Final water and mass balance predicted terms to confirm the appropriate
implementation of the water and nitrogen soil modules. .......................................... 83
Table 19. Model performance scores for dry matter accumulation and nitrogen content
for the whole plant and disaggregated per organ. ..................................................... 85
xv
nitrogen fertilizers in the scenarios evaluated for both open field and greenhouse
systems. ................................................................................................................... 104
Table 21. Median values for potential environmental impacts per ton of tomato
produced in the scenarios evaluated for open field and greenhouse conditions in the
Colombian Andes. ................................................................................................... 110
Table 22. Number of parameters, log-likelihood, Akaike information criterion (AIC)
and Bayesian information criterion (BIC) of the 36 bivariate copula estimated D-vine
models for the environmental indicators of the scenarios evaluated in greenhouse and
open field systems. .................................................................................................. 111
xvi
Chapter 1. General introduction and objectives
1.1 Introduction
Tomato production in Colombia is mainly carried out by smallholders and is mostly
marketed fresh for local customers, with an annual per capita consumption of 9.4 kg
(Bojacá et al., 2014). In Colombia, the varying agroclimatic conditions allow tomato
cultivation in a large part of the territory, as well as in open fields (OF) and
greenhouses (GH). Under OF conditions tomato production is carried out in lowlands
(warm tropics) where environmental conditions, especially in temperature terms,
allow cultivation throughout the year. In OF system small processing-type tomatoes,
locally known as chonto, are cultivated, while in GH system both, chonto and
beefsteak-type are planted. During the past few decades, tomato production has
changed from OF to GH systems, and this has been perceived as a technological
advance. Greenhouses have allowed to establish tomato crops in areas whose climatic
conditions limit open field production (cold tropics at high altitudes), and which
therefore have lower pressures of tomato pests and diseases. Greenhouses are
perceived by producers as an investment that allows them to avoid costs in pest
management issues and at the same time allows them to increase productivity by
employing higher plant densities per unit area. Currently, there is a perception among
OF tomato growers that the greenhouse production system is replacing them in the
markets.
In Colombia, the OF tomato system is rainfed and totally exposed to cold tropical
climate conditions. As a consequence, growers are constantly addressing pest and
disease issues, in addition to low soil fertility levels in some regions. Based on their
limited knowledge and resources, tomato growers are able to produce with intensive
overuse of agrochemicals. According to official statistics between 2007 and 2017, the
average area planted with OF tomatoes was 12670 ha y-1, with an average yield of
22.4 t ha-1 (Agronet, 2018).
On the other hand, the use of greenhouses has allowed the establishment of tomato
crops in low-temperature areas, regarded in the past as inadequate for tomato
production. These greenhouses are wooden structures without climate control systems
that protect the plants from adverse weather conditions such as rain or high wind
speeds, but they minimally improve the internal temperature and humidity conditions
to achieve optimum tomato yield. Despite the above, the use of low-tech greenhouses
for tomato production has boosted productivity compared to that in the open field
system. Between 2007 and 2017, the average greenhouse area with tomato production
was 1190 ha with an average yield of 115.6 t ha-1 (Agronet, 2018). However, despite
achieving a major increase in production capacity with respect to the OF system,
management practices are still considered to be suboptimal, resulting in yields far
from the potential yields and with a considerable environmental impact. As a
consequence, opportunities to optimize these systems exist from technical and
environmental points of view.
Concerns have arisen in the past regarding the environmental cost associated with the
tomato production, Bojacá et al. (2014) through a life cycle assessment (LCA)
General introduction and objectives
estimated the environmental burdens of the GH tomato production system based on

primary data collected through surveys. Their results show that fertilization is one of
the subsystems with the highest environmental impact share. Nitrogen (N)
fertilization, in particular, is strongly connected with the negative effects on the
environment, both due to the environmental burdens associated with its manufacture
(Hasler et al., 2015) and the emissions resulting from its use in the field (Brentrup et
al., 2001).
A LCA, despite being a standard methodology widely used to evaluate the
environmental impact of goods, products, and services, has limitations when applied
in the agricultural context due to its static character (Dufossé et al., 2013; Goglio et
al., 2018; Liao et al., 2014; Perrin et al., 2017). Agricultural production is
characterized by large variations mainly due to climate, spatial soil heterogeneity and
the wide array of management practices applied by growers. Moreover, in an
agricultural production system, many processes interact considerably. The inability
of the method to consider these variations and interactions is one of the greatest
challenges for the reliability of the LCA method in the agricultural sector (Liao et al.,
2014; Perrin et al., 2017).
Considering a fertilization subsystem, the current LCA method is able to consider the
impacts associated with fertilizer production, transport, and emissions caused by
fertilizer field use. The impacts derived from the manufacturing stage and background
processes are derived from secondary information stored in specialized databases
such as LCA Foods (Nielsen et al., 2003) and Ecoinvent (Ecoinvent, 2007). The
impact of transport depends on the distance between the manufacturing facilities and
the production field as well as the transport means used; however, the impact of
transport can be considered negligible compared to the impacts related to
manufacturing and field emissions (Bojacá et al., 2014; Xia and Yan, 2011). The
cornerstone is the on-field losses, which are considered to be nonpoint source
emissions due to multiple sources, such as fertilization and organic matter
mineralization, and also due to the pathways to exit the system, e.g., volatilization,
runoff, and leaching.
Therefore, as well as representing a major concern from the environmental point of
view (Bojacá et al., 2014; Brentrup et al., 2000; Nemecek et al., 2016; Perrin et al.,
2014), the prediction of N-field emissions represents a challenge for LCAs in terms
of their estimation methodology. According to Havlikova and Kroeze (2006), the
approaches that predict emissions from agriculture can be organized into three
categories: emissions factors, regression analyses, and process-based models. The
emissions factors are calibrated proportional numbers that relate the amount of a
released pollutant to an associated human activity. On the other hand, regression
analyses, based on metadata, estimate pollutant emissions involving a set of variables
that determine the potential losses (Havlikova and Kroeze, 2006). Finally, process-
based models, such as DNDC (Giltrap et al., 2010), DAYCENT (Del Grosso et al.,
2005), CERES-EGC (Drouet et al., 2011), and the work developed by Heuts (2018)
to simulate N-emissions into the environment in addition to the agricultural
production process, have been developed in recent years. However, this last approach
has been questioned due to its complexity related to the high number of required
parameters to be estimated for adaptation to local conditions, as well as by the input
datasets required to use the model, which are often unavailable. Based on these
2
considerations, it has been suggested that uncertainty in the N-emissions estimates

tends to increase when a model-based approach is used due to its propagation from
the parameter estimation phase (Nemecek et al., 2016).
Currently, emissions factor and regression analysis methods are the most frequent
approaches to estimate N-emissions from agricultural fields (Brentrup et al., 2000;
Nemecek et al., 2016), especially for LCA studies (Bojacá et al., 2014; Wang et al.,
2018). However, these approaches assume that input-output relations of agricultural
production systems and their associated impacts are relatively steady over the long
term. This scenario means that the requirements of N for a single crop or crop rotation
system is a known factor that is applied more or less constantly cycle after cycle. This
scenario assumes that mineralization and immobilization rates are relatively constant
(Brentrup et al., 2000). In developed countries, where agriculture is characterized by
a high level of technology with well-defined climatic seasons, the latter statement may
be closer to reality, but this is not the case in tropical smallholder systems such as
tomato production in the Colombian Andes. Therefore, this thesis explores the
dynamic prediction of N-emissions through a soil-crop modeling approach, which
was previously validated for local conditions. More specifically, this work aims to
evaluate the environmental impact caused by the N-fertilization subsystem in tomato
production systems. In contrast with a static LCA, in this approach, N-emissions vary
dynamically depending on the type of fertilizer, soil characteristics, management
practices and climatic conditions along the production cycle.
A second constraint related to LCAs is its use as a decision-making tool in
comparative scenarios. In a standard LCA, the results for the selected group of
indicators are presented and analyzed, but the challenge of integrating this
information is left to the decision makers (Tsang et al., 2014; Zanghelini et al., 2018).
This scenario is a challenge since each impact category is an indicator of a particular
dimension of the environmental performance of the system or product under analysis,
but the conclusions and subsequent decisions depend entirely on the correct overall
interpretation of these indicators. This task is not always straightforward, especially
in the case of comparing alternative scenarios with outstanding environmental
performances on different environmental indicators (Zanghelini et al., 2018). As a
result, an important challenge to improving the LCA methodology is the development
and operationalization of a method to generate an overall indicator that, in an integral
and objective manner, allows the performance evaluation of a system or product for
all indicators together. However, several aspects must be considered before
attempting to propose a tool that would allow the construction of such an indicator.
First, these indicators are estimated based on a group of variables, some of which are
correlated and used to compute several impact categories, generating varying degrees
of correlation among the indicators (Bojacá and Schrevens, 2010). Second, LCA
results are characterized by high variability that is reflected in asymmetric
distributions of the impact indicators when factors such as growers, climate or spatial
locations are considered, even for the same crop in a particular region (Bojacá et al.,
2014). In the present thesis, a stochastic-based method (Prado-Lopez et al., 2014;
Prado-Lopez and Heijungs, 2018) is adapted and applied to tackle the aforementioned
considerations. The resulting overall unbiased indicator is used within this thesis
framework to assess the environmental performance of current and prospective
fertilization management scenarios applied to two Colombian tomato production
systems.
3
In accordance with the above, the present thesis has the following objectives.
1.2 Objectives
The general objective is to improve the LCA method for the agricultural sector by
means of two modifications. The first modification consists of coupling soil-crop
modeling to a LCA as an alternative to dynamically predict the emissions derived
from the use of nitrogen fertilizers. The second modification considers the integration
of a method to derive multivariate joint distributions for the impact categories and a
multicriteria decision-analysis method known as stochastic multiattribute analysis
(SMAA) to build a single performance indicator in the optional normalization and
weighting LCA steps. An overall indicator should improve the comparison and
decision processes related the environmental performance of production alternatives
or management practices.
Specific objectives
• To characterize the open field and greenhouse tomato production systems in
terms of biophysical conditions and management practices carried out by
Colombian growers with special emphasis on nitrogen fertilization.
• To compare the environmental performance of the open field versus the
greenhouse production system by applying a standard LCA based on a single
indicator that incorporates the intrinsic variation in each system and the
correlation structure between the impact categories.
• To develop a climate-based tomato model with the ability to simulate tomato
growth and development based on the growing patterns and management
practices applied by local farmers.
• To validate a soil-plant model for both tomato production systems in
Colombia, resulting from the integration of an adapted version of the tomato
model and a previously calibrated soil model.
• To objectively evaluate a set of nitrogen fertilization scenarios as alternatives
to improve the current environmental performance of the tomato production
systems in Colombia.
4
Chapter 2. Study areas and data acquisition methods
Published as: Gil, R., Bojacá, C. R., & Schrevens, E. (2019). Datasets of the
environmental factors and management practices of the smallholder tomato
production systems in the Colombian Andes. Data in brief, 24, 103844.
2.1. Introduction
Datasets presented here were employed in the main work “Understanding the
heterogeneity of smallholder production systems in the Andean tropics – The case of
Colombian tomato growers” (Gil et al., 2019a). In the Colombian Andes two tomato
production systems are prevailing: in the open field (OF) and using plastic covered
greenhouses (GH) both perceived to be of low technological input level. For OF, data
were acquired in five municipalities of the Guanentá province (Santander
department), while for GH, data belong to five municipalities of the Alto Ricaurte
province (Boyacá department). The data presented here includes information on soil
parental materials and climate variables (averages ± standard deviations) relevant
from the agricultural point of view, which were calculated from historical climate
series. Fertility data on long-term fallow soils, sampled in the production areas, are
also presented. After filtering the data, 67 soil samples were obtained for OF and 70
for the GH. For GH, a dataset with the results of 38 soil samples taken inside
greenhouses were paired with the results of samples taken outside these greenhouses
in uncropped fallow areas. In the case of these soil analyses, the data correspond to
tables with the results reported by the laboratory for both, chemical and physical
variables, for each location in which soil samples were taken. In this work, the main
dataset is one that contains the inputs of fertilizers and water, and the corresponding
yields of tomato production cycles managed by local growers. This information was
collected through two data collection tools: surveys (SVY) to growers about these
aspects in their last production cycle, and through detailed follow-ups of selected
production cycles (FWU). For the OF, we collected data from 71 cycles through the
surveys and 22 through the follow-ups, while for the GH, information from 138 and
38 tomato cycles was collected through surveys and follow-ups, respectively. The
Table 1 with the results aggregated by tomato cycle is attached.
Study areas and data acquisition methods
Table 1. Metadata of collected information in open field and greenhouse production

systems.
Subject area Tropical agriculture, smallholder production
More specific subject Soil fertility, crop management, fertilization strategies

area
Type of data Tables, figures and text files.
How data were acquired Government agencies, farmer’s interviews and on-field measurements
Data format Raw, filtered and analysed
Experimental factors We present a table and datasets that describes the environmental
heterogeneity in terms of the climate and the fertility in long-term fallow
soils for two representative tomato production areas in Colombia. In terms
of the soil fertility, the soil samples corresponded to uncultivated soils,
seeking to avoid that the results were influenced by the addition of
nutrients through fertilization or amendments. We also included data to
show the effect of continuous greenhouse tomato production on the soil
chemical and physical variables; in this case, the factor corresponds to the
place where the samples were taken, i.e. inside or outside the greenhouse.
Finally, a tomato production cycles dataset with aggregated inputs and
outputs, as well as some cultivation practices is included. This dataset has
two factors associated: the production zones and the tool used for data
collection, which correspond to surveys and a direct observation procedure
called follow-ups.
Experimental features Soil fertility dataset came from a systematic sampling of the soils in each
production region. Inside and outside, uncultivated fallow soil samples
were collected for selected greenhouse tomato production units. Finally,
an extensive data collection work was carried out using two data collection
tools to learn about the main management practices used by tomato
growers, as well as a quantification of the amount of fertilizers used and
the yield, they are able to obtain.
Data source location Data from open field tomato were collected in the Guanentá province
(latitude: 6º25’18.83’’ N, longitude: 73º10’03.18’’ W; average altitude:
1370 masl) located in the Santander department, 300 km northwest of
Bogotá, Colombia’s capital.
The main area dedicated to greenhouse tomato production is located in the
department of Boyacá, specifically in the Alto Ricaurte province (latitude:
5º39’37.30’’N, longitude: 73º34’57.46’’W; average altitude: 2070 masl),
located 165 km northwest of Bogotá. There data were collected.
Data accessibility Data are attached to the article

Related research article Gil, R., Bojacá, C.R. and Schrevens, E. 2019. Understanding the
heterogeneity of smallholder production systems in the Andean tropics –
The case of Colombian tomato growers. NJAS-Wageningen Journal of
Life Sciences, https://doi.org/10.1016/j.njas.2019.02.002
Value of the Data

• The data serves to quantify the uncertainty derived from the fertility on long-
term fallow soils along with the diversity in fertilization practices applied by
smallholders.
• The data provides the possibility to compare the usefulness of standard data
collection tools such as surveys and detailed follow-ups for inventoring the
6
inputs and outputs of production systems lacking a proper record keeping

system.
• The data allows to compare the efficiency of smallholder production
systems, with different technological intensification level, and at the same
time draw a baseline for future investigations of fertilizer use efficiency.
2.2. Data
The data presented in this article serve to describe the heterogeneity associated with
tomato production systems managed by small producers in the Colombian Andes.
Data used in this work came from two of the main tomato production areas in
Colombia, located in the provinces of Guanentá (Santander department) and Alto
Ricaurte (Boyacá department), where tomatoes are cropped under open field (OF) and
greenhouse (GH) conditions, respectively. The distance between the two production
zones is 115 km. An overview of both production systems is shown in Figure 1 for
GH and in Figure 2 for OF.
7
Figure 1. Overview of the greenhouse tomato production system in the province of

Alto Ricaurte (Boyacá, Colombia).
8
Figure 2. Overview of the open field tomato production system in the province of
Guanentá (Santander, Colombia).
9
The landscape of the Guanentá province is formed by mountains and hills crossed by
rivers, which in some sectors form small riverbanks (Figure 3). For the GH zone, the
elevation profile of this study area, including the main soil parent materials, is
presented in Figure 4. In Table 2 the monthly variations of the main agroclimatic
variables are presented, from each production area. The datasets attached as
supplementary material in Gil et al. (2019a) corresponds to the fertility on long term
fallow soils defined based on a set of physico-chemical variables, see Appendix A in
Gil et al. (2019a), in both production areas (at 30 cm of depth). Also, a table with soil
analysis results of tomato production under greenhouse versus nearby uncultivated
lots is attached, see Appendix B in Gil et al. (2019a). Finally, we share data related to
crop management, fertilization and yield obtained from surveys and detail follow-ups,
collected on commercial tomato plots, see Appendix C in Gil et al. (2019a).
Figure 3. Elevation profile of a cross-section (southwest-northeast direction) of the

Guanenta province showing the main types of parent materials.
Figure 4. Elevation profile between the massif of Iguaque and the Merchán
Mountains, including the main types of parent materials.
10
Table 2. Monthly averages and annual values (± standard deviations) for the climate
variables used to describe the weather in the provinces of Guanentá (Santander) and
Alto Ricaurte (Boyacá).
Province Month Temperature Relative Precipitation Solar radiation
(ºC) humidity (%) (mm) (MJ m-2 d-1)
Guanentá Jan 18.7 ± 0.58 78.5 ± 2.59 56.7 ± 5.54 17.9 ± 0.98
Feb 19.1 ± 0.72 77.5 ± 2.71 92.6 ± 7.82 18.0 ± 1.46
Mar 19.3 ± 0.66 78.5 ± 2.26 154.8 ± 9.89 17.1 ± 1.19
Apr 19.2 ± 0.52 80.9 ± 2.10 282.2 ± 13.28 15.8 ± 0.97
May 19.2 ± 0.50 82.5 ± 2.20 295.1 ± 12.37 15.2 ± 1.15
Jun 19.0 ± 0.56 82.2 ± 2.15 232.0 ± 10.95 15.2 ± 0.82
Jul 18.7 ± 0.53 81.7 ± 1.80 223.5 ± 11.08 16.2 ± 0.76
Aug 18.7 ± 0.56 81.7 ± 2.05 260.3 ± 11.63 16.7 ± 1.07
Sep 18.7 ± 0.59 81.9 ± 1.93 259.0 ± 11.50 16.6 ± 1.10
Oct 18.7 ± 0.51 82.6 ± 1.93 304.9 ± 12.52 15.9 ± 0.94
Nov 18.9 ± 0.53 82.6 ± 1.81 211.8 ± 10.77 15.8 ± 0.86
Dec 18.7 ± 0.51 81.2 ± 1.67 93.3 ± 7.86 16.8 ± 0.86
Annual 18.9 ± 1.0a 81.0 ± 2.10a 2466.3 ± 125.2b 197.3 ± 12.16b
values
Alto Jan 16.3 ± 0.50 74.0 ± 3.80 46.3 ± 4.72 17.0 ± 1.33
Ricaurte Feb 16.6 ± 0.61 73.5 ± 4.05 76.9 ± 6.04 17.1 ± 1.45
Mar 16.7 ± 0.55 75.5 ± 3.36 120.3 ± 7.91 15.9 ± 1.65
Apr 16.8 ± 0.43 76.8 ± 3.53 150.8 ± 8.71 14.9 ± 1.05
May 16.8 ± 0.33 77.1 ± 2.94 110.5 ± 7.35 14.6 ± 0.87
Jun 16.7 ± 0.42 74.0 ± 5.27 53.6 ± 4.89 14.6 ± 0.93
Jul 16.4 ± 0.29 72.6 ± 5.71 52.0 ± 4.67 15.3 ± 0.96
Aug 16.5 ± 0.34 71.8 ± 4.41 48.2 ± 4.71 15.8 ± 1.02
Sep 16.5 ± 0.50 71.9 ± 4.49 78.7 ± 6.32 15.7 ± 1.25
Oct 16.4 ± 0.38 77.0 ± 4.60 159.2 ± 8.55 14.8 ± 0.97
Nov 16.3 ± 0.33 78.8 ± 3.23 132.3 ± 7.59 14.4 ± 1.11
Dec 16.3 ± 0.44 76.7 ± 3.60 15.3 ± 1.58 15.3 ± 1.58
Annual 16.5 ± 1.1a 75.0 ± 4.00a 1115.1 ± 78.6b 185.4 ± 14.17b
values
a
Annual average
b
Annual sum
2.3. Experimental Design, Materials, and Methods

Soil and climate data
For each production area, the elevation profiles were constructed with data extracted
from Google Earth and supplemented with information about the soil parental
materials obtained from previous studies (Ardila-Villamizar et al., 2011; Carrera et
al., 1975; Jiménez, 2003). The climate was described based on historical time series
obtained from the Colombian meteorological service agency (IDEAM). For the
Guanentá province, climate time series were obtained from a weather station placed
in Mogotes municipality (6°29'4.79"N, 72°58'28.59"W), while in Alto Ricaurte the
station was placed in Villa de Leyva municipality (5°38'37.18"N, 73°34'17.85"W).
Time series corresponded to daily records of 50 years for Boyacá (1964-2014) and of
11
57 years for Santander (1958-2015), but both presented a high proportion of missing
data. From the dataset, we excluded years with strong or very strong affection of the
El Niño (1957, 1958, 1965, 1966, 1972, 1973, 1982, 1983, 1997, 1998, 2015, 2016)
and La Niña (1973, 1974, 1975, 1976, 1988, 1989), based on the Oceanic Niño Index
(ONI) calculated and published by the National Oceanic Atmospheric Administration
(NOAA, 2016). With the filtered dataset, monthly averages (± standard deviation)
were calculated for four variables: precipitation (mm), temperature (ºC), relative
humidity (%) and solar radiation (MJ m-2 d-1). The solar radiation was estimated from
the hours of bright sunshine (hours day-1) since this variable was not included in the
original data. The incoming solar radiation was estimated from extra-terrestrial solar
radiation and relative sunshine hours, following the equation proposed by Angstrom
(1924)
f
85 = 895 t%& + %' v (1)
.o
where 85 is the daily total solar radiation above plant canopies (J m-2 day-1); 895 is the
extraterrestrial solar radiation (J m-2 day-1); %& and %' are empirical coefficients, f is
the duration of the sunshine (h day-1); DL is the day length (hours). The coefficients
%& and %' were determined according to dry tropical climate zones based on the values
proposed by Frere and Popov (1979) as 0.25 and 0.45, respectively. For extra-
terrestrial solar radiation and day length estimation, we followed the equations
proposed by Teh (2006), which take into account an eccentricity correction factor and
the local solar times.
Fertility in long-term fallow soils
Initially, in each production area, 75 soil samples were collected at 30 cm depth on
fallow plots between May and July 2015. Sampling spots were determined by a non-
aligned random sampling procedure and adjusted once on the field to sample only
uncropped soils. Based on the geographic coordinates of the sampling points, we
determined the altitude and slope using a 30 m digital elevation model (DEM). Soil
samples were processed at an accredited soil laboratory (ISO/IEC 17025:2005). The
analysis included chemical properties such as nitrate (NO-3 ), exchangeable ammonium
(NH+4 ), extractable phosphorus (P) and available potassium (K) contents, pH,
electrical conductivity of saturated soil-pastes (ECss), soil organic carbon (SOC); and
physical properties such as clay, silt and sand contents (%). Exchangeable NH+4 and
NO-3 were determined by extraction with KCl, and the solution was analyzed as
described by Bremner and Keeney (1966). Extractable P was determined by the Bray
II method, K was extracted with ammonium acetate (1:20) and determined by flame
photometry, pH in 1:1 soil:water suspension and ECss in soil saturated extract, SOC
by the wet oxidation method and texture was measured through a hydrometer. P and
K contents were transformed to P2O5 and K2O by multiplying them by 2.292 and
1.205, respectively.
Because some samples could correspond to recently cultivated soils and this would
cause over-estimates of the fertility level, a pre-treatment of the data were carried out
to detect and remove outlier records. For outlier detection the Mahalanobis distance
of each observation to the overall mean was calculated and thresholded. The
Mahalanobis distance was calculated using the following formula:
12
/.1 = w(D1 − Μ){ )* |' (D1 − Μ) for : = 1, … , (2)
where Μ is the multivariate arithmetic mean (centroid) and )* is the covariance

matrix. This distance is used to determine which observations can be considered
atypical, as shown by Filzmoser et al. (2005) for geochemical multivariate datasets.
To define the threshold for atypical observations, we took into account that /.1,
variability can be approximated with a chi-square distribution with l degrees of
freedom (Gnanadesikan, 2011). Soil samples having a /.1, greater than ÄG;&.ÉÑ
,
were
classified as outliers and removed from the dataset. Final datasets comprised 67 and
70 records for Guanentá and Alto Ricaurte provinces, respectively.
Effect of the greenhouse tomato production on soil properties
As part of the characterization, we determined the effect of GH fertilization
management on soil properties. Because tomato is the only crop grown in these
greenhouses and as such no rotation is applied, planting can be done at any time
throughout the year. Based on the above, changes in soil properties depend
exclusively on the fertilization management conducted by the growers. To analyse the
effect of GH fertilization, we took 30 cm deep soil samples inside greenhouses during
the fallow period and sampled only greenhouse soils with more than two years
dedicated to tomato production; also, on adjacent non-cultivated areas (100 to 500 m
away from the greenhouse edges) soils were sampled. The sampling sites were
randomly selected based on satellite images on which greenhouses locations were
clearly identified. Samples were analysed including the same variables used to
describe for the fertility in long-term fallow soils in a certified soil laboratory, and
also following the aforementioned methods. We took 38 pairs of soil samples in June
2013.
Fertility management practices
In the present work, two data collection tools were employed: surveys (SVY) and a
direct follow-up observation procedure (FWU). SVY consisted of a structured
questionnaire of closed-ended questions about technical aspects related to the last
tomato growing cycle such as: cropped area, plant density, cycle length, type and
amount of fertilizers applied, crop management practices, irrigation and yield.
Questions were developed by the research team and subsequently and adapted after a
test with local growers. Previously trained undergraduate students conducted the
interviews on the farm. Between 2009 and 2010, a total of 80 and 174 surveys were
carried out to randomly selected smallholder of the OF and GH technological levels,
respectively.
Among surveyed growers, a random subsample was selected for a detailed FWU
during complete production cycles. The FWU method corresponds to a detailed
observational procedure on tomato production cycles carried out by members of the
research team. With this data collection tool, detailed data were acquired on use of
fertilizers, irrigation, management practices and yield. At the beginning of each
production cycle, we measured characteristics such as cropping area, plant density,
irrigation water flow rate, and features associated to available infrastructure.
Remaining factors were recorded through weekly interviews with growers, focusing
on crop management practices including the allocation of time and resources,
13
irrigation schedule, dosing and timing of inputs used in fertilization, as well as fruit
production. This procedure was conducted from soil preparation to the end of harvest.
In OF, we recorded data for 22 tomato cycles belonging to 10 farms, with four of them
being planted consecutively on the same plots. Under GH conditions, we recorded
data for 39 cycles established in nine farms, with 10 of them being consecutives. The
smaller number of consecutive cycles in OF is due to the crop rotations scheme
applied by each grower. Obviously, the FWU method is more time consuming than
the SVY method. FWU data collection took 16 months (from October 2011 to February
2013) for the OF and 30 months (from September 2010 to March 2013) for the GH.
From both data collection tools, the data recorded were aggregated in order to obtain
the total inputs (e.g. fertilizers) employed for the tomato production along with the
total yield achieved. For commercial fertilizers, the nutrient content was obtained
from the official information showed in the label. In the case of organic fertilizers,
samples were taken and analyzed to determine the concentration of nitrogen,
phosphorus and potassium.
All variables are georeferenced with coordinates in decimal degrees taken along to
the official geodesic datum for Colombia (MAGNA-SIRGAS).
14
Chapter 3. Tomato smallholder systems in the Colombian Andes
Published as: Gil, R., Bojacá, C. R., & Schrevens, E. (2019). Understanding the
heterogeneity of smallholder production systems in the Andean tropics–The case of
Colombian tomato growers. NJAS-Wageningen Journal of Life Sciences, 88, 1-9.
Abstract
In developing countries, a common goal is to improve horticultural production
systems as a strategy to increase food security and to improve the living conditions of
these rural communities. However, smallholder-based agricultural systems are highly
heterogeneous due to the wide range of biophysical conditions to which the crops are
exposed, and the diversity in the management practices. In order to implement
programs aimed at improving the productivity of these systems it is necessary to
recognize its variability in quantitative terms. The main objective of this work was to
describe the heterogeneity associated to smallholder production systems, using as a
case study the Colombian tomatoes growers. Data were collected from two tomato
production zones located in the Colombian Andes and under two cropping systems
being the open field (OF) and the greenhouse (GH) production models. In both zones,
the climate was described based on historical records, soil samples were taken to
determine the long-term fallow soils fertility and the growers’ management practices
were inquired. We also compared two instruments for data collection, surveys and
detailed follow-ups. A higher heterogeneity in environmental conditions and
management practices was evidenced for the OF system compared to the GH system.
The fertilization strategies used by GH growers caused a significant increase in soil
nutrient content, electrical conductivity and acidity. We found a higher productivity
per square meter in the GH system, however the yield per plant was higher for the OF
system (47.5 t ha-1 cycle-1) in comparison with the GH system (87.5 t ha-1 cycle-1).
Results also indicated that follow-ups are an appropriate instrument to obtain accurate
inventories. Knowledge empowerment arises as the key point to improve the
smallholder’s productivities; in opposition to results elsewhere, where economic
constraints are highlighted as the important sources of variability and low yield.
Tomato smallholder systems in the Colombian Andes
3.1. Introduction
In developing countries, smallholder-based agricultural systems have great
importance from a socio-economic point of view. These low-tech systems are focused
on providing commodities demanded by domestic markets, at the minimum cost
without a proper business-oriented model in place. Moreover, on the technical side,
these production systems are characterized by a huge gap between actual and potential
yields (Ruben and Pender, 2004; Tittonell et al., 2008). This gap is due to multiple
factors, some of which are difficult to control, such as climate, while others rely on
decisions made by the growers related to crop management, especially factors such
as nutrient supply (Tittonell et al., 2008) and pest management. In addition, many
production areas have biophysical constraints such as low fertility in long-term fallow
soils, steep slopes, and strong climatic uncertainty, especially with regard to water
availability (Ruben and Pender, 2004). As a result, these production systems are
distinguished by a huge heterogeneity in relation to crop yields (Ruben and Pender,
2004), inputs employed (Bojacá et al., 2012) and the subsequent environmental
impact (Bojacá et al., 2014; Gil et al., 2017a). In this work, the term heterogeneity is
used to refer to a high-variation in quantitative variables, while the term diversity is
used to refer to the variability observed in categorical variables. Moreover, this
heterogeneity increases over time due to cycle-to-cycle variability in fertilization,
irrigation, pest and disease management and mechanization practices. This situation
threatens the existence of these production systems, especially when there is land use
competition for crops that offer higher profit margins and less susceptibility to
environmental factors, such as oil palm (Chenoune et al., 2016).
Currently, the need to increase food production to meet the demands of a growing
population (Licker et al., 2010), raises the interest to improve the productivity of
smallholder farming systems, as an alternative to reduce the expansion of the amount
of land under agriculture and inducing negative effects like deforestation and loss of
biodiversity (Vanlauwe et al., 2010). Smallholder systems have a greater potential to
increase productivity as compared to high-tech systems. In addition, improving the
productivity of those systems would also generate socio-economic benefits, such as
an increase in growers income, improvement of the rural infrastructure, and ultimately
raising the quality of smallholders’ lives (Ruben and Pender, 2004). However,
improving the productivity requires quantifying and understanding the current status
of the system, including its variability and uncertainty. This heterogeneity may
influence the variable performance of the new tools and technologies introduced to
improve productivity (Tittonell et al., 2010). On this basis, it has been proposed that
understanding the inherent heterogeneity of smallholder systems should be the first
task to properly orient agricultural innovations (Chenoune et al., 2016).
Acknowledging this heterogeneity will contribute to analyse other aspects such as the
environmental impact of these systems.
Nowadays, Colombia is recognized as an emerging economy with investment-grade
status; however, sectors such as agriculture still running on low technology
techniques, especially subsectors such as horticulture. The country seems to have
followed the standard policies of developing countries; which according to Ruben and
Pender (2004), has left aside investment in smallholders to focus entirely on high
potential agricultural production systems, seeking to promote rapid economic growth.
At present, with the recently signed peace agreement, a great deal of attention is
16
focused on the development of rural areas, not only those directly affected by the
conflict but the rural sector in general with special emphasis on smallholders
(Presidency of the Republic of Colombia, 2016). This will require investment in
research and development to improve these production systems. However, to allocate
resources in an efficient way, one must start with an up to date characterization
portraying how heterogeneous these systems are, while also explaining the causes of
this high-variation (Chenoune et al., 2016; Ruben and Pender, 2004).
The present work is a response to the limited body of information regarding the
diversity of smallholders’ agricultural systems in Latin America and with the certainty
that this type of knowledge helps to improve the overall performance of these systems.
We analysed the case of the tomato (Solanum lycopersicum L.) production system in
Colombia, which, in addition to being the main horticultural crop in the country, is
mostly developed by smallholders. Moreover, tomato is one of the main cropping
alternatives to promote the development of rural areas, since the country has extensive
agroclimatic zones suitable for this crop. Nowadays, tomato is mainly cultivated
under two production systems: open field (OF) and greenhouse (GH), whose planted
areas in 2014 were 14768 and 2305 ha, respectively (Agronet, 2018).
The main purpose of this work was to study the heterogeneity associated to
smallholder agricultural systems, using as a case study detailed data on Colombian
OF and GH tomato systems. The following three specific objectives are proposed: (I)
to describe the climate of the production zones under study as well as the spatial
variation of the soil properties driving its fertility; (II) to expose the variability in the
management practices under which tomato production is carried out by comparing the
information obtained from two sources of data: surveys and detailed follow-ups; and
(III) to analyse the agronomic efficiency of fertilizer use (nitrogen, phosphorus and
potassium) for each production system.
3.2. Materials and methods

The data used in this work came from two of the main tomato production areas in
Colombia located in the provinces of Guanentá in the departments of Santander (OF
system) and Alto Ricaurte (GH system) in the Boyacá department. The distance
between the two production zones is 115 km. OF tomato production is carried out
countrywide; however, five departments concentrate almost 60% of the national
production. Santander was the second largest producer of tomatoes with a cultivated
area of 1958 ha during 2014. As opposed to the OF production, GH tomato production
is concentrated in 31% of the country departments, and the most important five hold
87% of the national production. Boyacá is the main GH tomato producer in Colombia
with an area of 990 ha reported in 2014 (Agronet, 2018).
For the OF system, data was collected in seven municipalities (802 km2), namely
Curití, San Gil, Mogotes, Pinchote, Valle de San José, Páramo, and Confines.
Guerrero and Torres (2016) estimated a tomato planted area of about 320 ha in these
municipalities, based on secondary data about annual seedlings sold by local
nurseries. GH data were acquired in the municipalities of Villa de Leiva,
Sutamarchán, Sáchica, Tinjacá and Santa Sofia (441 km2); for this region, the area
under GH systems is approximately 330 ha, estimated through visual interpretation of
satellite images (Guerrero and Torres, 2016). In Alto Ricaurte, GH tomato is the main
crop, while in Guanentá the main crops are coffee (Coffea arabica L.) and sugarcane
17
(Saccharum officinarum L.) while horticultural commodities, such as tomato, take the
second place. These two systems represent the extremes of smallholder tomato
production in Colombia, from the low input and low-tech production (OF) to a system
with higher inputs and higher technology level (GH) compared to OF, but low
compared to greenhouses in other latitudes.
Data related to climate, soil fertility and main characteristics of both productions
systems with emphasis on fertilization efficiency were employed for systems
description. Data analyses quantified the heterogeneity and compared its magnitude
between and within production systems depending on the data source. Special
attention was paid to the effect of the GH tomato production on soil fertility.
Climate
For each production zone, climate data were obtained from the Colombian
meteorological service agency (IDEAM). In Alto Ricaurte, the weather station was
located in the municipality of Villa de Leiva (5°39'12" N – 73°34'39" W; 2060 masl),
while in Guanentá the station was placed in the municipality of Mogotes (6°28'34" N
– 72°58'13" W; 1680 masl). For both regions, the climate records corresponded to
daily series spanning a 50-year period for Villa de Leiva station (1964-2014) and a
57-year period for Mogotes station (1958-2015); however, it is important to highlight
that these time series contain a high proportion of missing data. For climate
description purposes, we excluded the information of years with strong or very strong
El Niño or La Niña effects, based on the Oceanic Niño Index (ONI). The climate
variables included in the analysis were the monthly average precipitation (mm), daily
mean temperature (ºC), relative humidity (%) and solar radiation (MJ m-2 d-1). In
addition to the daily or monthly average, the standard deviation was calculated as a
measure of the intra-annual variation.
Fertility in long-term fallow soils
The fertility in long-term fallow soils was described based on 75 soils samples taken
within each production area between May and July 2015. The samples were collected
up to a 30 cm depth on fallow plots. Sampling spots were determined by a non-aligned
random sampling procedure and adjusted on the field to sample only uncropped soils.
Based on the geographic coordinates of the sampling points, we determined the
altitude and slope using a digital elevation model (DEM). Soil samples were
processed in the soil laboratory of the ‘Centro de Bio-Sistemas’, belonging to the
Universidad Jorge Tadeo Lozano. Soil samples were packed in plastic bags and
transported to the soil laboratory in around two days without refrigeration. The
analysis included chemical properties such as nitrate (NO-3 ), exchangeable ammonium
(NH+4 ) and extractable phosphorus (P), available potassium (K) contents, pH,
electrical conductivity of saturated soil-pastes (ECss), soil organic carbon (SOC); and
physical properties such as clay, silt and sand contents (%). Exchangeable NH+4 and
NO-3 were determined by extraction with KCl, and the solution was analysed as
described by Bremner and Keeney (1966). Available P was determined by the Bray
II method, K was extracted with ammonium acetate (1:20) and determined by flame
photometry, pH in soil:water suspension (1:1) and ECss in soil saturated extract, SOC
by the wet oxidation method and texture was measured through a hydrometer.
18
A preliminary data inspection evidenced unusually high values in some samples,

which were detected based on the squared Mahalanobis distance and the 0.98
percentile of a chi-square distribution, with the number of variables as degrees of
freedom, to set the threshold from which a sample is considered an outlier (Filzmoser
et al., 2005; Gnanadesikan, 2011).Those samples probably corresponded to recently
cropped areas; therefore, we excluded them from the analysis. Final datasets
comprised 67 and 70 records for Guanentá and Alto Ricaurte, respectively. We used
the mean (xN) and median (M@ ) to describe the centrality, while the coefficient of
variation ()i) was used to report the variability within and between zones. To
compare the soil properties between the two zones, we first verified the data fit to a
normal distribution (Kolmogorov–Smirnov test) and the homogeneity of variances
(Levene test). For normally distributed and homoscedastic variables, the t-student test
(at 95% confidence level) was used to compare the variables between production
zones. For non-normal distributions and heteroscedastic variables, the Mann–
Whitney non-parametric U-test (at 95% confidence level) was applied.
Afterwards, a principal component analysis (PCA) was performed with the dataset of
fertility in long-term fallow soils, composed of 10 variables and 137 soil samples. We
did the PCA to arrange the soil samples along an underlying, latent soil fertility
gradient, and also to study the correlational structure between variables by means of
the cosine angle between the vectorized variables. PCA results were scaled and were
represented as biplots, in which only the first two principal components were plotted.
Effect of the GH system on soil properties
As part of the characterization, we determined the effect of GH production on soil
properties. We only considered the GH system since no rotations are involved, and
planting can be done at any time throughout the year. Based on the above, changes in
soil properties depend exclusively on the fertilization management conducted by the
growers. To analyse the effect of GH production, we took 30 cm deep soil samples
inside greenhouses with more than two years dedicated to tomato production and on
adjacent non-cultivated areas (100 to 500 m away from the greenhouse edges).
Samples were analysed at the soil laboratory of the ‘Centro de Bio-Sistemas’
following the methods described above for the soil fertility dataset. We took 38 pairs
of soil samples in June 2013. We used the U-test or t-test to compare the soil
properties inside and outside the greenhouse depending on the outcome of the
normality and homoscedasticity tests, as it was described in for fertility in long-term
fallow soils.
Production systems characterization
Two data collection tools were used to characterize the production systems: surveys
and follow-ups. The surveys consisted of face-to-face structured interviews of closed-
ended questions applied between 2009 and 2010 to 80 and 174 randomly selected
growers from the OF and GH systems, respectively. The survey instrument was
designed to collect information at both farm and plot level. At farm level, data
corresponded to available infrastructure (e.g. greenhouse features, water sources,
infrastructure), household characteristics and marketing strategies. At plot level, we
asked about technical aspects related to the last growing cycle such as cropped area,
plant density, cycle length, type and amount of fertilizers, crop management practices,
irrigation rate and yield.
19
For the follow-ups, we selected among the surveyed growers a group of farmers to
carry out detailed inquiries on consecutive tomato production cycles. At the beginning
of each production cycle, we measured some crop features such as planted area, plant
density, and features associated to the available infrastructure. Remaining features
were recorded through weekly interviews with the growers, focused on crop
management practices including time and resources allocation, irrigation schedule,
dosing and timing for inputs used in fertilization, as well as fruit production. The
follow-ups were conducted from soil preparation to the end of harvest. In Guanentá,
we collected data on 32 production cycles, from October 2011 to February 2013.
Since OF tomato is rotated with other horticultural crops, just 22 of them
corresponded to tomato cycles. For the GH system we recorded activities on 39
tomato cycles located in nine farms from September 2010 up to March 2013. As no
crop rotation is involved under GH system, all information collected was related to
tomatoes. A schematic sequence of the production cycles of the follow-up dataset can
be seen in Figure 5.
20
Figure 5. Tomato crop cycles observed during the follow-up data collection period in the (A) OF and (B) GH systems. Each segment represents
the cycle length from transplanting until the end of harvest.
21
Finally, an agronomic efficiency of fertilizer use (AEF), in which, tomato yield was
divided by the amount of nitrogen, phosphorus and potassium applied by each grower,
was calculated. For this purpose the approach proposed by Aujla et al. (2007) for
nitrogen, extended for phosphorus and potassium, was used, as shown below:
'$
!"#$% = (3)
(#$%
where, !"#$,% is the agronomic efficiency of fertilizer use for the jth fertilizer (j =
nitrogen, phosphorus or potassium) and for the ith grower, '$ is the yield obtained by
the ith grower (kg ha-1) and (#$% is the amount of fertilizer used by each grower (kg
ha-1). This !"# is reported as kg tomato yield per kg of fertilizer applied. The results
were used to compare the Colombian production systems to others elsewhere.
Data analysis was carried out in the same way as for the soil fertility dataset. After
outlier detection and exclusion, descriptive statistics such as the mean and median
were used to describe the centrality of the considered variables, while the *+ was
included to quantify the variability. As the is invariant to measurement units, it
allowed comparing the variation degree between data sources and production systems.
The final dataset for crop management contained 71 and 138 tomato cycles
characterized by surveys for OF and GH systems, respectively; while 22 (OF) and 38
(GH) cycles were characterized by detailed follow-ups. All statistical procedures
described above were implemented through the R statistical software (Team R Core,
2017).
3.3. Results
Climate
Figure 6 shows the average monthly precipitation and daily temperature throughout
an average year in both production zones. A bimodal rainfall pattern was observed for
monthly precipitation in both zones (Figure 6A). However, in Guanentá, average
(±standard deviation) annual rainfall is higher (2466.3 ± 125.2 mm) compared to Alto
Ricaurte (1115.1 ± 78.6 mm). In addition, rainfall seasons are more intense in
Guanentá province occurring between the strong dry season of December and
February and the mild dry season of June and August. This rainfall distribution
ensures a good water supply during most of the year; although topography in this zone
restricts the access to water near the mountain tops. Oppositely, the water supply is
drastically reduced in Alto Ricaurte during its two dry seasons, which occur between
December-February and June-August (Guerrero and Torres, 2016). Regarding
temperature, Guanentá has higher temperatures throughout the year as compared with
Alto Ricaurte (Figure 6B). In Guanentá, the average (±standard deviation) annual
temperature is 18.9±1.0ºC while in Alto Ricaurte is 16.5±1.1ºC. However, the
temperature exhibits a greater spatial variation, especially in Guanentá due to the
mountainous landscape. Simultaneous climatic records on two tomato crops, one
located at 1700 m (6°25'15" N, 73°11'56" W) and the other at 1150 m (6°28'55 "N,
73°06'54" O) and distanced by 11.5 km, presented average daily temperatures of 19.5
and 22.8ºC, respectively. The low average temperature in the Alto Ricaurte
production zone is the main reason why farmers use greenhouses structures for tomato
production. In contrast to precipitation, the temperature does not show a strong
seasonal trend, although the maximum temperatures are usually recorded between
22
February and March (26.9ºC in Guanentá and 23.6ºC in Alto Ricaurte). Regarding the
relative humidity (RH), the annual average (±standard deviation) is lower in Alto
Ricaurte province (75±4.1%) compared to Guanentá (81±2.1%) and variations
throughout the year follow the same trend as the rainfall (Figure 6C). Finally, solar
radiation exhibited a similar trend as the temperature (Figure 6D), without showing a
marked seasonality; however, the annual average (±standard deviation) is slightly
higher in Guanentá (16.44±1.01 MJ m-2 d-1) than in Alto Ricaurte (14.45±1.18 MJ m-
2 -1
d ).
23
Figure 6. Average monthly precipitation (A), daily temperature (B), relative humidity (C) and solar radiation (D) of the production regions of
Boyacá and Santander. For precipitation, error bars represent ± the standard deviation, while for the other variables are the shaded area is that
indicates ± the standard deviation.
24
Soils
The results of the mean differences test between both provinces indicated that all soil
properties showed significant differences with the exception of NO-3 content and SOC.
According to the results, the tomato production area in Guanentá province has sandier
textures, steeper slopes, higher content of NH+4 and higher soil acidity; but lower P
and K contents, as well as a lower ECss, as compared with Alto Ricaurte (Table 3).
Concerning data distribution, in both production areas, most of the variables had
positive skewness; that is, a higher data proportion with values below the mean were
observed.
Table 3. Fertility characterization in long-term fallow soils used for tomato production
in open field (Santander) and under greenhouse (Boyacá) production systems. The
coefficient of variation (CV) is expressed as a percentage.
Variable Unit Open field Greenhouse p-value
Mean Median CV Mean Median CV
-1
Ammonium mg kg 10.59 9.00 55 8.28 6.25 73 <0.05
(NH+4 )
Nitrate (NO-3 ) mg kg-1 5.13 3.90 94 4.44 2.95 108 0.40

-1
Bray-II mg kg 15.22 6.65 144 71.27 37.81 110 <0.05
phosphorus
(P2O5)
Exchangeable mg kg-1 158.05 127.70 72 219.10 199.98 63 <0.05

potassium
(K2O)
pH in 5.09 4.80 15 5.93 5.65 17 <0.05

soil:water
(1:1)
ECssa dS m-1 0.15 0.11 79 0.65 0.34 115 <0.05

Soil organic % 2.10 1.98 43 1.90 1.70 48 0.19
carbon
Slope % 11.01 11.01 51 5.87 5.17 57 <0.05
Sand % 44.86 43.50 29 27.53 27.70 42 <0.05
Clay % 17.66 16.70 49 29.56 28.75 37 <0.05
a
Electrical conductivity of saturated soil-pastes
Table 3 shows that NH+4 concentrations were higher than NO-3 in both Alto Ricaurte
and Guanentá provinces, which is an atypical result in agricultural soil analysis. The
larger concentration of ammonium is probably due to the uncooled transport of soil
samples (low oxygen, high humidity), which promotes the biological conversion to
NH+4 and accumulation in the sample before its soil lab processing. Although this does
not affect the concentration of total nitrogen in the soil, it restricts the interpretations
of soil fertility status based on the concentrations of NO-3 and NH+4 in this chapter.
In Alto Ricaurte skewness were more pronounced in variables such as NO-3 , P and
ECss, and to a lesser extent in NH+4 content. In Guanentá, although the skewness was
less pronounced in comparison with Alto Ricaurte, the variables with the greatest
25
asymmetry were NO-3 , NH+4 , P and K contents. The main difference between both
production zones, with respect to data distribution, was observed for terrain slope.
While in Guanentá the slope is distributed symmetrically, in Alto Ricaurte it has a
slight positive skewness. The only variables that consistently showed a symmetric
distribution were the content of clay and sand (soil texture).
Regarding the variation, the !" revealed that soil properties in both production zones
exhibited a high heterogeneity. However, with exception of the P, K and clay content,
other variables showed a greater variation degree in Alto Ricaurte. Although
variations are function of the production area, largest ones (in order of importance)
were detected for NO-3 , P, NH+4 and ECss, in both zones, but the latter with a
particularly high !" in Alto Ricaurte province (115 %). On the other hand, pH was
the variable that showed the lowest variation in both production zones; however, it
should be taken into account that pH is expressed on a logarithmic scale. The
skewness shown for most of the variables in both production zones, together with the
high !" indicate a huge heterogeneity in soil fertility, which is particularly low in
Guanentá province.
Regarding the PCA results, the total variance explained for the first two components
was 51.52% with 34.45% explained by the first component and 17.07% by the second
component. Figure 7 shows the biplot for the first two components describing the
correlational structure between soil properties. The first factorial axis discriminates
the soils by location and associates the characteristic properties to each zone.
However, data distribution on the factorial plane, moving away from the biplot center
in different directions, shows the soil heterogeneity, even within the same production
area. Different degrees of correlation were observed based on the cosine angle formed
between each pair of variables. A high positive correlation was observed between pH
and P availability; since in both zones, soils are acid (Table 3). The correlation
indicates that a lower soil acidity generates a greater availability of P. Also, a high
positive correlation was observed between the sand content and NH+4 ; this relationship
is observed because a higher proportion of sand limits the NH+4 nitrogen adsorption
by the clay fraction.
26
Figure 7. Scaled PCA-biplot showing the variation and correlational structure of

physical and chemical soil properties.
The comparison between soils located inside and outside greenhouses revealed the
impact of tomato production on soil properties in Alto Ricaurte. This comparison
allowed understanding the effect of continuous mono-cropping of tomato under GH
conditions on the soil fertility. Table 4 highlights the differences between both soil
types. The GH tomato production significantly modifies some soil properties,
increasing the nutrients content (NPK), and consequently the ECss. Beyond the
statistical significance, it is important to highlight the huge increase in some nutrient
contents; on average, NO-3 and P2O5 were 6.7 and 5.2 times higher inside GH
compared with the uncultivated soils around them. The accumulation of P in the soil
is a well-known consequence of the use of soils under greenhouses for vegetable
cultivation, as extensively analysed by Sonneveld and Voogt (2009). In these cases,
the use of simple salts sources of fertilizers instead of compound fertilizers is
recommended, in order to have a better control over the additions of this element. In
cases of high accumulations of P in the soil and a pHrange between 6 and 7, as was
determined here, it has been recommended that the application of this element be
suspended or applied at rates lower than 140 kg ha-1 yr-1. Currently, the imbalance in
nutrient content in soil, especially expressed in an increase in salinity, together with
soil-borne disease problems (Fusarium and nematodes) are driving the shift towards
systems such as raised-beds filled with new soil and hydroponics using substrates with
mixtures of rice-husk and coal. For pH, SOC, sand and clay, no significant changes
were detected as a consequence of GH tomato growing. The results showed how the
27
chemical properties on surface soil layers (0-30 cm) are more susceptible to being
modified by agricultural practices management; which agrees with what was reported
by Yemefack et al. (2005).
Table 4. Soil properties comparison between uncultivated soils and those used for
tomato production under the GH system. The coefficient of variation (CV) is
expressed as a percentage.
Uncultivated Soils under
Variable Unit soils greenhouses p-value
Mean CV Mean CV
Ammonium
mg kg-1 13.80 88 34.50 55 <0.05
(NH+4 )
Nitrate (NO-3 ) mg kg-1 10.66 133 71.82 82 <0.05
Bray-II
phosphorus mg kg-1 98.74 126 510.77 132 <0.05
(P2O5)
Exchangeable
mg kg-1 280.75 132 535.85 93 <0.05
potassium (K2O)
pH in soil:water
6.13 18 6.57 16 0.08
(1:1)
ECssa dS m-1 0.58 101 2.57 52 <0.05
Soil organic
% 1.73 67 1.95 64 0.42
carbon
Sand % 30.25 39 35.24 34 0.07
Clay % 41.76 23 41.83 22 0.97
a
Electrical conductivity of saturated soil-pastes
General characteristics of tomato production systems
The main characteristics of tomato production systems are described below; firstly,
as a general overview and afterwards the emphasis is on management practices,
particularly those related to NPK fertilization.
In the OF system, the small processing-type chonto tomato is cultivated while under
the GH system both, chonto and beefsteak-type tomatoes are cropped. In both
systems, production is carried out almost exclusively on soil (OF = 100%, GH =
99.28%), starting with four weeks old seedlings. In both zones, hybrid cultivars with
a high genetic potential are planted; under GH the two main cultivars are planted by
80.1% of the growers; while in OF there is a greater diversity of chonto cultivars, and
the chonto percentage drops to 38.1%. All tomato production is destined to fresh
consumption in the Colombian local markets. Under both systems, plants are guided
by a training system consisting of a wooden structure between which galvanized wires
are tightened. Because of crop rotation in the OF system, the training system is
installed and removed for each cropping cycle, while under the GH system a
permanent training structure is set-up. Greenhouses are naturally ventilated,
consisting of a wooden structure covered with a single sheet of polyethylene, without
28
active climate control systems. Most of the GH farmers prune stems (98.6%), to have
vigorous monopodial plants, but also senescent leaves (99.3%) and even fruits
(24.5%) are pruned; while the OF farmers omit all these practices. Unlike the GH
system, which is dedicated exclusively to tomato production, many OF smallholders
rotate tomato with other horticultural species (58.8%) such as corn (Zea mays), green
beans (Phaseolus vulgaris), cucumber (Cucumis sativus) and squash (Cucurbita spp.).
Tomato growers are also characterized by a low associativity level (no growers
associations or similar were found) and consequently, their trading power with
suppliers and customers is also low. The farmer’s educational level ranges from basic
to secondary levels (84.2%), but this does not imply that they have received
agricultural-oriented formation. Growers infrequently share information with each
other, but access to extension services is relatively high (OF = 55.3%; GH = 78.0%);
however, these services are mostly provided by agrochemical dealers (OF = 48.7%;
GH = 69.3%) whose impartiality is questionable due to their commercial interests. As
a consequence, each producer is basing his decision-making mainly on his own
experience. The growers have a relatively low experience in tomato cultivation, in the
OF system 90% of the producers have an experience less than 6 years, while in GH
this experience time reaches 10 years.
Tomato fertilization is carried out using both organic and inorganic sources. As
indicated by the detailed follow-ups, nitrogen (N) input comes mostly from inorganic
sources (OF = 83.1%, GH = 68.6%), while the remaining is applied in the form of
organic sources. In the OF system, the most used inorganic fertilizers are NPK multi-
nutrient solid fertilizers and the most frequent formula are 15-4-23 (22.8%), 10-30-
10 (20.6%) and 15-15-15 (11.0%). Under the GH system, the most used fertilizers are
soluble multi-nutrient represented by a great diversity of brands, along with nitrate
salts of calcium, magnesium, and potassium. In both systems, chicken manure (OF =
53.7%, GH = 45.7%) is the preferred organic fertilizer source. Growers commonly
use chicken manure because Guanentá province is an important area of poultry
production (Guerrero and Torres, 2016); consequently, manure is an abundant and
cheap organic source, which is used by both, OF and GH growers. The fertilization
method is different for each production area. In Guanentá, solid fertilizers are diluted
in water, sometimes mixed with organic sources or amendments such as lime, which
are manually applied per plant; this procedure is repeated several times along the
production cycle. In contrast, 100% of the GH growers apply highly soluble fertilizers
through basic fertigation systems.
Tomato production systems description
Table 5 presents the main characteristics of the two tomato production systems under
study. We emphasise on the #$% for NPK fertilizers. In the OF system, a lower
planting density is used, the production cycles are shorter, larger areas are planted and
the yield per hectare is lower as compared to the GH system. Regarding productivity,
although it is higher per square meter in the GH system, when standardized by the
planting density, we observed that the plant average yields were higher for the OF
system (4.88 kg plant-1) in comparison with GH system (2.84 kg plant-1). In the OF
system, the surveyed growers reported an average yield 18.8% lower than the one
calculated through the follow-ups; while for the GH system it was the opposite,
surveyed growers reported an average yield 38.22% higher than the one determined
with the follow-ups. Yield !" calculated from the follow-ups showed a lower
29
variability (OF = 66%, GH = 54%) than the one obtained by the surveys (OF = 120%,
GH = 75%). In the OF system the water supply by irrigation presented the lowest
median (Table 5) and simultaneously the highest !"; this is due to the dependence of
the crop on rainfall, which is highly variable. Under the GH system, the water input
comes entirely from the irrigation, and contrary to OF, it is a variable with relatively
low degree of variation. Based on these results, it was established that OF yields have
a greater degree of uncertainty in comparison to those of the GH system.
Table 5 demonstrates the unproportional amounts of fertilizers used by growers in the
production of tomatoes under greenhouses, up to 437, 410 and 665 kg ha-1 cycle-1 of
N, P2O5 and K2O respectively. The above values are given in relation to the median
of the AEF calculated on the basis of the data obtained from the detailed follow-ups.
These results justify the necessity to evaluate strategies that will lead to a reduction
in the amount of fertilizers applied (e.g. nitrogen) in order to improve the tomato
production profitability along with its environmental performance.
30
Table 5. Main features of the open field and greenhouse tomato systems determined through surveys and detailed follow-ups.
Open field Greenhouse
Production factor Unit Survey Follow-up Survey Follow-up
Mean Median CV Mean Median CV Mean Median CV Mean Median CV
Plant density plants m-2 1.3 1.2 27 1.1 1.1 28 3.0 3.1 17 3.0 3.0 15
Cycle duration days 116.0 112.0 9 114.0 109.0 15 186.0 183.0 16 166.0 166.0 12
Cultivated area m2 4986.4 4000.0 78 6004 3750 103 2522.7 2425 43 3000.6 2835 46
Yield t ha-1 43.6 35.9 53 53.7 47.5 50 117.9 111.1 45 85.3 87.5 37
-2
Irrigation lm 387.1 230.4 129 120.3 46.5 152 603.2 495.2 65 521.7 467.5 65
Precipitation mm 916.4 912.0 10 220.5 231.0 36 - - - - - -
-1
AEF, total kg kg N 645.6 254.8 215 370.1 264.5 77 241.7 183.3 82 272.6 200.2 66
nitrogen
AEF, phosphorus kg kg-1 701.6 207.5 258 278.7 183.2 79 249.8 168.9 211 435.0 213.6 123
P2O5
AEF, potassium kg kg-1 803.5 233.2 305 221.8 157.5 59 190.2 136.2 121 162.9 131.5 67
K2O
31
Agronomic efficiency of fertilizer (AEF) use

In both systems, the yield obtained per kg of fertilizer applied, changed depending of
the data collection tool (survey or follow-ups). In the OF system, a poor !"# (i.e. a
lower yield per kg of fertilizer applied) was determined through the follow-ups as
compared with results obtained for the surveys. The opposite trend is observed for
nitrogen and phosphorus in the GH system, where the survey´s showed a poor AEF
in comparison with follow-up data.
The CV calculated for both data sources also showed differences, decreasing when
the follow-up data were used. The CV´s estimated for the AEF based on the surveys
averaged 198%, whereas the one based on the follow-ups reduced to 78%. This
confirms the reduction in the uncertainty caused by the follow-up data collection tool.
However, we have to consider that the comparison between surveys and follow-ups
is limited by the difference in the degrees of freedom, which are much larger for
surveys.
In both production systems, the distribution of the !"# showed a positive asymmetry.
This skewness is induced by the components of the AEF calculation. In the OF
production system, the follow-up dataset reduced this asymmetry, which was
evidenced by a smaller difference between the median and mean. In the GH system,
although the trend was similar and reduction in asymmetry was observed, the !"#
showed a slight increase in its skewness when estimation was derived from the follow-
up dataset.
Detailed follow-ups provided more accurate information to describe not only the AEF,
but also other features related to tomato production. The weak point of follow-ups is
that the sample size is lower compared to the surveys, affecting the representativeness
of the results. However, the follow-ups allowed capturing the cropping season
variations by recording consecutive tomato production cycles in the same plots.
Finally, it should be remarked that minimum $% for AEF was 59%, even for the
follow-up dataset; this shows the high diversity in fertilization management practices.
3.4. Discussion
In the study zones, the contrast in climatic conditions shows the heterogeneous
conditions in which horticultural production takes place in Colombia. While in
Guanentá province, the climate is warm and humid, in Alto Ricaurte it is cold and
dry; both zones have bimodal rainfall distributions and strong spatial variations in
temperature caused by the mountainous landscape, especially in Guanentá (Figure 4).
Although smallholders seem to be aware of constraints imposed by the climate, they
adapt trying to find genotypes able to tolerate rough weather conditions; in OF
systems, this is especially evident by the greater diversity of cultivars planted. They
also appeal to planting during favorable periods or building low-tech structures to
protect plants from extreme conditions. However, the use of climate control
technologies to create a comfortable environment for plants is not a priority, even for
farmers that use greenhouses. As a consequence of the low-tech greenhouses used,
the internal microclimate strongly depends on external conditions (Bojacá et al.,
2009). This means that Colombian smallholder farmers, as well as in other parts of
the world (e.g. Africa), are highly susceptible to the constraints imposed by the
climate (Tittonell et al., 2007).
32
The results of the present work showed low fertility soils, especially in the OF
production zone; associated to a high heterogeneity, mainly observed in Alto Ricaurte
where tomato is cropped under GH (see Table 3). In the OF production zone, the
tropical soils are characterized by high acidity levels and low concentrations of total
N and P, mainly due to the high precipitation regime during a large part of the year
(Haileslassie et al., 2005). This general description for tomato production systems
agrees with literature reports that associate smallholder agricultural production with
low fertility and heterogeneous soils (Haileslassie et al., 2005; Tittonell et al., 2007;
Yemefack et al., 2005). Given the scant information about the effect of low-tech
greenhouse production systems handled by smallholders on soil fertility, our results
are novel by showing an unbalanced nutrient enrichment of the soil top layer. This is
the opposite of the results reported for agricultural soils managed by smallholders in
other developing countries (e.g. in Africa), in which depletion of soil fertility has been
documented (Haileslassie et al., 2005). But this increase in nutrient soil stock does not
necessarily mean an increase of the tomato yield because other factors such as specific
nutrient deficiencies or too low pH’s values constrain the effects of fertilizer
application.
The high concentration of some nutrients such as total N and P on the topsoil layer
poses a risk to the environment (Kwong et al., 2002). While soils remain covered by
greenhouses, the mobility of these nutrients will be constrained by reduced water
inputs, however, emissions to the air by volatilization and denitrification can be
expected. However, if these soils return to OF production systems a flow of the
accumulated nutrients toward water sources during the rainy seasons by runoff and
leaching will certainly occur (Kwong et al., 2002).
Regarding the data sources employed, the results showed that the survey has a greater
uncertainty ($%>120% in OF and $%>75% in GH) compared with the follow-ups. It
is important to highlight that smallholders rely on their memory to keep record of the
management practices such as fertilization. Additionally, smallholders vary their
strategies from cycle to cycle making it difficult to characterize the system with a
snapshot instrument such as the survey. On the other hand, follow-ups are more time-
consuming, expensive and the numbers of observed cycles generally are lower.
Despite the above, the follow-ups offer a higher resolution, especially concerning
input and output inventories. We suggest the follow-ups as the collection instrument
in cases where, in addition to the socio-economic characterization, data will be used
to estimate system efficiency and environmental impacts where high accuracy is
required.
The results of the present study also expose the diversity of management practices
employed by growers, leading in interaction with climate, environment and genotype
to very different final yields (Tittonell and Gilller, 2013). The results show important
variations for both AEF (CV>59%) and application methods, ranging from the
dilution of solid fertilizers mixed with organic fertilizers and amendments in OF, to
the use of fertigation under GH. This diversity in fertilization practices, along with
pest and disease management, not included in this work, caused the observed
heterogeneity in yields; similar cases have been documented in other production
systems managed by smallholders (Bojacá et al., 2012; Tittonell et al., 2007, 2005).
In this regard, it should be noted that the lower variation found in the GH production
system has been associated with less dependency on environmental conditions,
33
achieved by the plastic isolation (Gruda, 2005). Regarding to the AEF, in absence of
a local reference that relates the potential tomato yield to the amounts of NPK-
fertilizers necessary to obtain it; we use as a reference the averages calculated from
Hatirli et al. (2006), Boulard et al. (2011), Torrellas et al. (2012), and Payen et al.
(2015) for greenhouse tomato crops grown in soil; which were 359.5 (kg kg-1 N),
433.5 (kg kg-1 P2O5) and 222.7 (kg kg-1 K2O), respectively. Comparing these values
with our results (medians) it is clear that a wide margin to improve the AEF in local
systems is still present. It is known that small farmers’ knowledge is a determining
factor to decide the amount of fertilizers used, and it seems that they have the pre-
established idea that a larger fertilizers application always leads to a yields increase
(Pishgar-Komleh et al., 2017; Tittonell et al., 2008, 2007). Therefore, these results
generate concerns from both an economic and environmental point of view. Tomato
growers seem to have a high-risk aversion and decide to invest in excessive
fertilization doses to be confident that no yield reduction will occur due to this
production factor. In Colombia, the inadequate fertilization strategy by tomato
growers not only decreases the efficiency of the system but also increase soil nutrient
stocks and therefore, increase the risk of environmental pollution, as was shown for
the GH system. On the other hand, in OF systems higher fertilization rates together
with a high precipitation regime create the ideal conditions for nutrient (N and P)
losses through leaching and runoff.
Contrary to the African scenario, where low efficiencies and heterogeneity observed
in smallholder systems are associated, additionally to climate and soil natural
variations, to low fertilization rates due to economic constraints (Ruben and Pender,
2004; Tittonell et al., 2005); high productive potentials seeds and high fertilization
rates show that farmer’s endowments are not the main cause of current low yields. In
this sense, our results show a huge potential to increase productivity in tomato
production in systems managed by smallholders in the Colombian Andes. However,
achieving this purpose is conditioned by understanding the biophysical constraints in
the production zones and diversity of management practices employed by growers.
This is a well-documented prerequisite to be able to define locally adapted alternatives
aimed at increasing farmer’s yields (Tittonell et al., 2010, 2007; Vanlauwe et al.,
2010). According to Vanlauwe et al. (2010), a realistic approach should be based on
agroclimatic zones delimitation and characterization to develop standardized
practices addressed to increase productivity and this aligned with farmer’s
endowments; and at the same time chasing an increase in the farmer’s knowledge
level.
The low knowledge level of smallholders is not only related to their education level
but is also associated with other shortcomings such as restricted access to truthful and
unbiased information sources, and the lack of cooperation amongst growers.
Regarding the sources of information, our results showed that small producers access
extension services mainly from agricultural input suppliers and to a lesser extend from
universities and research institutes; while extension services from governmental
agencies are almost no existing. Grower’s exposure to low-quality information
sources creates skepticism regarding the adoption of new technologies and/or results
of research processes. Recently, Weber and McCann (2015) found that knowledge
transfer is less likely to be adopted by growers when coming from agrochemical
dealers. On the other hand, the low organization level limits the flow of information
among producers, which has been recognized as an important factor to increase the
34
level of knowledge of grower communities (Wall, 2007). To improve the farmer’s

technical skills is a challenge that should be focused on developing and implementing
strategies that lead to a knowledge empowerment situation. Smallholder growers,
beyond being able to implement a particular technology, should be formed to evaluate
the recommendations from different sources and decide rightly the most convenient
one with full awareness of the complexity of their production system. This kind of
approaches has emerged as alternatives to improve the productivity of smallholder-
base production systems in developing countries (Friis-Hansen, 2004). For that
reason, a formal training program directed to producers has been proposed as a
strategy to increase the AEF (Luo et al., 2016; Pishgar-Komleh et al., 2017). The aim
is to abandon the excessive application of fertilizers as a strategy to increase yields;
and on the contrary, to focus on improving management practices that increase AEF.
Based on our results, we propose three management practices that have already
proven to be effective to increase AEF. Firstly, to carry out a soil analysis before each
planting and combine the results with average plant demands to define the fertilization
strategy, with respect to fertilizers types, quantities, and timing (Pishgar-Komleh et
al., 2017). This is especially important in the OF system, in which annual rains modify
the soil nutrient contents. Secondly, to train the farmers to use the different types of
fertilizers correctly (Wang et al., 2018), and avoiding incorrect practices such as those
described for the OF system. Thirdly, face the soil constraints (e.g. pH) with
technically-sound practices.
3.5. Conclusions
In Colombia, smallholder tomato growers face the commercial-oriented production
by planting high production potential genotypes, which they grow in heterogeneous
climatic conditions and on soils with low fertility levels, in some cases characterized
by properties (e.g. pH) that could limit the crop response to the fertilization. Despite
this, growers resort to excessive fertilization as the main strategy to increase
production, which results in highly variable yields.
With respect to data sources, surveys consistently showed greater variation in the
amount of inputs employed in the production cycle compared to follow-ups. This is
due to the fact that smallholders do not keep formal records about input-output
relations in their production cycles, as a consequence surveys are an ineffective
instrument to capture data concerning input-output inventories. As an alternative, we
evaluated and proposed to carry out detailed follow-ups on production cycles;
although, the sample size is reduced, and it is a time-consuming method compared to
surveys.
Despite the current low fertilizer use efficiencies, there is an enormous potential to
increase the yields of smallholders based on relatively small changes in the
fertilization practices. However, this increase seems to be strongly conditioned by an
increase in farmer’s knowledge. Yield improvements demand changes in the
fertilization strategy; leaving aside over-fertilization practices and instead, adopt a
soil management strategy based on soil analysis, plant requirements and amend the
soil factors that limit the plant nutrient uptake. Until this happens, mismanagement
practices carried out by growers could be increasing the risk by creating favorable
conditions for nutrients losses to the environment.
35
Chapter 4. Comparative LCA between open field and greenhouse
systems
Gil, R., Bojacá, C. R., & Schrevens, E. (2019). Accounting for correlational structures
in stochastic comparative life cycle assessments through copula modelling. Submitted
to The International Journal of Life Cycle Assessment
Abstract
In many cases, the univariate distribution of environmental indicators does not follow
the Gaussian distribution. Moreover, these environmental indicators are correlated,
therefore, it is necessary to build joint multivariate sampling spaces for life cycle
assessment (LCA) uncertainty analyses. The copula method allows to model any type
of multivariate joint distributions. This article integrates the copula and the stochastic
multi-attribute analysis (SMAA) methods to perform the normalization and weighting
steps in a comparative agricultural LCA. An attributional LCA was performed to
compare the environmental impact of two tomato production systems (GH:
greenhouse; OF: open field) with different intensification levels. The primary data
used for the inventory came from direct data collection carried out on tomato
production cycles from two of the main production regions in Colombia. To choose
the best environmental performance system, we implemented the outranking
procedure of the SMAA method. As required by the SMAA method, initially, we
fitted skewed multivariate distributions among the environmental indicators but
accounting for their correlation structure through the copula method. Afterwards, the
standard SMAA procedure was followed, leading to the calculation of overall scores
indicating the environmental performance of the systems under comparison. After
individual LCAs were performed for each grower, the variability observed in the
primary data was propagated to the environmental indicators. The marginal
distributions of the environmental indicators showed a right skewed trend which were
fitted to gamma, log-normal or Weibull distributions as applicable. The application
of the copula method for the environmental indicators of the GH and OF systems
resulted in D-vine models consisting of 46 and 45 bivariate copulae requiring 47
parameters each, respectively. Sampling the multivariate space configured by the D-
vine models and integrating it with the SMAA method, indicated that the OF system
is more likely to have a better environmental performance with a rank acceptability
index (RAI) of 57.6% than the GH system which resulted in a 42.4%. We applied a
stochastic unbiased approach to compare the environmental performance of
agricultural systems but recognizing the correlation structure of the indicators. The
copula method introduced here can be applied to uncertainty or multi-criteria decision
analysis where correlation needs to be accounted for. Joining the copula and the
SMAA methods to produce an unbiased preference indicator allows to evaluate
scenarios in a realistic way producing results that can easily be communicated.
Comparative LCA between open field and greenhouse systems
4.1. Introduction
To date, life cycle assessments (LCA) have been established as the topmost method
to assess the environmental impact of a product, service or system. This method is
applied to all types of industries, technologies, locations, and scales of analysis. As
LCAs is governed by a series of ISO standards (ISO, 2006a, 2006b), the main stages
and components to carry out an LCA are well defined. However, the LCA method is
under constant scrutiny by a scientific community aiming to improve its reliability
and applicability, as observed by its dauntingly growing body of literature.
The optional normalization and weighting steps in the life cycle impact assessment
(LCIA) stage are among those issues under permanent debate (Kägi et al., 2016).
Normalization involves dividing the potential impact per unit of emissions by the
impact score of a reference situation per person per year (Huijbregts et al., 2003).
Weighting is carried out to obtain one single dimensionless indicator that represents
the overall performance of the indicators included in a LCIA (Kalbar et al., 2017).
Several approaches for performing the normalization and weighting steps are
available; however, the robustness and uncertainty of these approaches remain a
matter of concern by the LCA community (Pizzol et al., 2017).
Prado-Lopez et al. (2014) proposed an alternative approach to normalization and
weighting based on stochastic multi-attribute analysis (SMAA) using internal
normalization by outranking and stochastically exploring weight sets treating all
impact categories as equal. This proposal is intended to overcome the limitations,
biases and distortions produced by the current approaches. By performing a
comparative LCA of laundry detergents, the authors concluded that their approach
avoids masking criteria and includes multiple perspectives while generating better
information for decision makers. SMAA is a multicriteria decision-support analysis
(MCDA) based on the weight space, approximated by a very large sample of
randomly extracted vector of weights (Greco et al., 2017). The outcome of the method
is represented by an acceptability index, that is, the probability for a certain alternative
to become best, based on the weight distributions used by the method (Lahdelma et
al., 1998). As a stochastic approach, SMAA eliminates the subjectivity and value
judgment factors commonly present in many weighting tools, such as the standard
MCDA (Myllyviita et al., 2014).
Recently, SMAA has been considered, among the various MCDA methods available,
to support environmental decisions (e.g., Domingues et al., 2015; Islam et al., 2015;
Reza et al., 2014). However, to our knowledge, application of the SMAA method for
comparative LCAs has been minimal, at least based on the published scientific
literature (Dias et al., 2016; Reeb et al., 2016). This article builds upon the proposal
of Prado-Lopez et al. (2014) considering the implementation procedure described by
Prado and Heijungs (2018) and applied to the agricultural sector more specifically for
comparing the environmental performance of two tomato production systems,
namely, greenhouse (GH) and open field (OF) systems.
Agriculture is under permanent pressure; on the one hand, there is an increasing need
to feed the growing population worldwide; on the other hand, the current management
practices for most production systems lead to poor environmental performance (Van
Kernebeek et al., 2016). Therefore, globally, the agricultural community is working
to develop new means of production and technologies and to develop alternative
37
production systems that meet the demands of the population at all levels (Ponisio et
al., 2016). From an environmental point of view, a comparison of production systems
is a challenging task since agricultural production systems are subjected to natural
biophysical variability in time and space.
In their paper, Prado-Lopez et al. (2014) assumed lognormal distributions for each
inventory input, while in their article describing the implementation of the SMAA
method, Prado-Lopez and Heijungs (2018) assumed the distributions of the
environmental indicators were normal. The environmental performance of an
agricultural production system at a regional level entails inventory data collection for
a set of representative production units. The biophysical variability in these
production units is the one that defines the uncertainty in the inventory data and allows
the construction of actual probability density functions required by the outranking
procedure of the SMAA method instead of assuming theoretical probability
distributions.
On the other hand, the distributions of environmental indicators cannot be treated as
independent since they represent linear transformations of inventory data; therefore,
univariate sampling of these distributions does not consider the highly correlated
structure of the environmental indicators (Bojacá and Schrevens, 2010). For
comparative LCAs, it is important to address uncertainty properly because ignoring
the correlation between input parameters can lead to an incorrect estimation of the
output variance, which can result in an incorrect decision-making process (Groen and
Heijungs, 2017). Recognizing the dependence structure among environmental
indicators requires incorporating multivariate random sampling procedures for
methods such as SMAA, while recognizing that in many cases, the univariate
distribution of environmental indicators does not follow the well-known Gaussian
distribution (Bojacá et al., 2014; Heuts et al., 2012). Therefore, building multivariate
sampling spaces that consider the marginal probability distributions of the
environmental indicators are required.
To overcome the aforementioned issues related to the probability density functions of
environmental indicators, we introduce the copula method to generate multivariate
distributions that can be used in the outranking step of the SMAA method. The copula
method is a solution for the demand of non-Gaussian models in sectors such as
finance, climate research, environmental sciences, medicine and physics (Dißmann et
al., 2013). Modeling dependent observations through copulae allows combining
potentially different marginal distributions into a multivariate distribution, in which
the starting point is the marginal distributions rather than a derivation by integration
based on the parent multivariate distribution (Danaher and Smith, 2011).
The concept, and the name, of copulae was introduced in Sklar’s theorem (Sklar,
1959) as shown below. Consider a d-dimensional cumulative density function, F, with
marginal F1, …, Fd. There exists a copula C, such that:
#('1, … , '+) = .(#1('1), … , #+('+); q) (4)
where xi Î R; q Î R; i=1,…,d; with q being the copula parameters measuring the
dependence among marginal distributions. q parameters depend on the specific
copula shape selected. Therefore, the copula function acts as a binder between the
joint distribution and the marginal ones (Kayalar et al., 2017).
38
The C: 12 ⟶ 1 function represents a d-dimensional copula only if the following

properties hold (Durante and Sempi, 2010):
(i) for every j Î {1,2,…,d}, C(u) = uj when all the components of u are equal to 1,
with an exception made for the j-th one that is equal to uj Î 1
(ii) C is isotonic, i.e., C(u) £ C(v) for all u, v Î 12 , u £ v
(iii) C is d-increasing
Since the theorem provides the probability density function (pdf) of the copula from
the joint and marginal pdf’s, the copulae can be fitted to the empirical value through
maximum likelihood (Meucci, 2011). For a full description of the foundation of
copula statistics, the reader can reference the works of Joe (1997) and Nelson (1990).
Although standard Archimedean multivariable copulae are generally used for
asymmetric distributions, those are limited by being controlled by a single parameter.
In response Aas et al. (2009) introduced the use of bivariate copula as dependency
models for the distribution of a definite number of pairs of variables conditional on a
set of variables. This new approach has been called pair-copulae and was used to
construct multivariate distributions. From the methodological point of view, it was
determined that the identification of the needed pairs of variables and their
corresponding set of conditioning variables is facilitated by a sequence of trees
(Kurowicka and Cooke, 2006.). This is commonly known as trees regular vine (R-
vines) and the resulting multivariate distribution corresponding to a R-vine
distribution.
The aim of the present work is twofold: (i) to demonstrate the feasibility of performing
the normalization and weighting steps in agricultural LCAs for comparative purposes
through the SMAA method and (ii) to improve this method by using the actual
probability distributions of environmental factors, while recognizing their
dependence structure through the copula method. The proposed analytical framework
is illustrated with a case study comparing the environmental performance of tomatoes
grown in open fields and under protected conditions in Colombia.
4.2. Methods
To fulfill the objectives of this work, we first present the procedure used to perform a
standard LCA for this case study, and then, when describing the normalization and
weighting steps, we highlight the procedure used to incorporate the SMAA method
and improve the method. The goal of the present attributional LCA was to compare
the environmental impact of two tomato production systems with different
intensification levels. The LCA was performed following the ISO 14040 and 14044
(ISO, 2006a, 2006b) standards that are described in the previous sections.
The two tomato production systems compared are GH and OF systems, which can be
established across several regions in Colombia year-round. Due to the wide range of
microclimates present in the intra- and inter-mountainous regions of Colombia, open
field and greenhouse tomato crops are cultivated without distinction. From a market
point of view, there is no product differentiation related to whether the tomato comes
from an open field or from a greenhouse for the consumer. In addition, consumers
prefer the processing type cultivars, locally known as “chonto” tomatoes, which
growers plant in both production systems.
39
Goal and scope definition

The present work focuses on determining the impact associated with the tomato
production system; for this reason, only the cultivation phase was considered.
Consequently, the boundaries of the system extend from the extraction of raw
materials to the farm gate (cradle-to-gate LCA). The most relevant processes
considered were the production of fertilizers, pesticides, plastics, steel and fossil fuels.
Additionally, the impacts associated with emissions from the use of pesticides and
nitrogen fertilizers were considered. We did not include the transportation of the
materials, the elimination of waste or the impacts generated in the nursery stage. The
functional unit for the analysis was one metric ton of fresh tomatoes.
Life cycle inventory analysis
The primary data used for the inventory stage were obtained through a direct
observational procedure carried out for several tomato production cycles. For both
systems, the production cycle begins with soil tillage and ends at the last harvest.
Between September 2010 and March 2013, data were collected from 22 open field
production cycles and from 39 greenhouse production cycles. These follow-up
production plots were selected from a group of farmers who had previously been
surveyed to characterize both production systems. Growers who agreed to collaborate
with this research did not receive any technical assistance to avoid any bias that could
affect the outcomes of this study. The follow-up open field growers were in Guanentá
Province (6°25’ N – 73°10’ W, 1370 masl), while the greenhouse growers were
located in Alto Ricaurte Province (5°39’ N – 73°34’ W, 2070 masl). Both zones are
among the major tomato production regions in Colombia.
Throughout the follow-up observations, information was collected directly in the field
and through weekly interviews with the growers. The field measurements provided a
detailed inventory of the infrastructure and machinery used in the production cycle,
as well as the characteristics of the crops such as the cropped area and planting density
and of the irrigation system features. During the weekly interviews, the growers were
questioned about the activities performed in their plots during the previous week. The
most common practices included tilling, transplanting, irrigating, managing crops
(pruning and twining), fertilizing, managing pests and diseases, weeding, and
harvesting fruit.
Growers were asked about the time and resource allocation for each practice,
including the amount of inputs used and the time devoted to each activity. As
fertilization and pest and disease management practices mostly use external inputs,
the interview focused on registering the commercial name of the product, dosage and
application method. Then, the corresponding amounts of nutrients or active
ingredients were calculated. For each production cycle, the registered data were
aggregated per management practice (i.e., irrigation, fertilization, and pest
management) to construct the LCA primary data inventory. Subsequently, the
resources and emissions allocated to the functional unit were derived from the yield
of each production cycle. Secondary data, related to the background processes of
external inputs such as fertilizers and pesticides, were extracted from the Ecoinvent
v.2.0 database (Frischknecht et al., 2007).
40
Nitrogen fertilization emissions

Nitrogen impacts derived from on-field use of nitrogen fertilizer were considered in
the estimation of the potential emissions to the air (NH3 and N2O) and water (NO-3 )
and their subsequent conversion to the corresponding impact categories. The
estimation of N-emissions was based on the work of Nemecek et al. (2016), which we
considered as an updated guide with the potential to be applied under tropical
conditions because some parameters can be set locally. Next, the equations used to
estimate N-emissions together with the parameters employed in each of them were
determined.
According to Nemecek et al. (2016), NH3 emissions can be calculated using the
Agrammon model (Kupper et al., 2010). This model has two modules to estimate the
emissions derived from fertilizer use, one for organic fertilizers and one for inorganic
fertilizers. In the case of organic fertilizers, the emissions are estimated as follows:
45673 = :!6 ´ (5; + $=>>)´ $' ´ 46# (5)
where oeNH3 (kg NH3 ha-1) is the ammonia emissions caused by the use of organic
fertilizers and TAN (kg of NH+4 t-1) is the total ammonium content in the organic
fertilizer. Based on our own measurements, we set this value at 1.7 kg of NH+4 t-1 for
poultry manure, which is the most frequently applied organic source in both systems.
46# (t ha-1) is the amount of organic fertilizer applied. The other terms are correction
factors that represent the fertilizer characteristics and the application method; er is the
NH+4 emissions rate depending on the type of manure, which we set at 0.82, following
the suggestion of Nemecek et al. (2016) for poultry manure. Capp (dimensionless) is a
correction factor for the degree of dilution, which was set as zero since in Colombia,
liquid manure is not commonly applied in tomato production. Cx (dimensionless) is
the correction factor indicating the crop production parameters, which was set to 0.3
in this study to represent poultry manure applied in solid form (Nemecek et al., 2016).
On the other hand, the NH3 emissions derived from the use of inorganic fertilizers
(ieNH3, kg NH3 ha-1) were calculated based on the amount of N multiplied by an
emissions factor that depended on the type of fertilizer.
?5673 = ?6# ´ "#@AB@CCDE (6)
where iNF (kg total N ha-1) is the nitrogen applied through inorganic fertilizers and
EFAGRAMMON is an NH3 emissions factor for the type of inorganic fertilizer.
The sum of ieNH3 plus oeNH3 indicates the potential NH3 losses to the atmosphere.
In the next step, the NO-3 losses (eNO3, kg NO-3 ha-1) were estimated with the SQCB-
NO3 model proposed by Emmenegger et al. (2009), which is based on an adaptation
of the model proposed by Willigen (2000) and is recommended to be applied under
non-European conditions. The model is represented by the following equation:
L
56FG = 21.37 + K O ´(0.0037´?6#
MŃ (7)
+ 0.0000601´64;R– 0.00362´T)
The model depends on local factors such as precipitation or irrigation (P, mm) and
the ability of the soil to retain moisture, indirectly represented by its clay content (c,
%). Water inflows were determined for each production cycle based on the follow-up
41
observations in addition to rainfall measurements from a weather station that was

installed in the open field production area. Soil clay contents at 30 cm depth were
obtained from a soil analysis carried out at the beginning of each production cycle,
resulting in averages of 16.3% for the OF and 29.5% for the GH systems. The rooting
depth (L, m) was set at 0.53 m for the open field and 0.47 for the greenhouses based
on previously performed experimental measurements. iNF (kg N ha-1) represents the
amount of nitrogen applied through inorganic fertilizers, while Norg (kg N ha-1) is the
nitrogen applied through organic sources. The term U (kg N ha-1) represents the N-
uptake by the crop, which was established as 1.46 and 1.08 g N kg-1 of fresh weight
of fruits for OF and GH systems, respectively.
Finally, the N2O emissions derived from the application of fertilizers and from
processes such as nitrification and denitrification were calculated as follows:
44
56U F = K O ´ (56U FXYB ) + (56U FEZ[ + 56U F\YE ) (8)
28
56U FXYB = 0.01 ´ (46# + ?6# + 6M;) (9)
14
56U FEZ[ = 0.01 ´ K O ´ 567G (10)
17
14
56U F\YE = 0.0075 ´ K O ´ 56FG (11)
62
where eN2O (kg N ha-1) is the potential emissions of nitrous oxide, eN2OFER (kg N ha-
1
) is the emissions due to fertilization with organic (oNF, kg N ha-1) and inorganic
(iNF, kg N ha-1) sources and emissions from the crop residues (Ncr, kg N ha-1). The N
in crop residues was estimated based on the N-uptake by the crops and experimental
measurements to determine the proportion of N remaining in the crop residues, which
resulted in 41.7 and 31.3% for OF and GH, respectively. eN2ONIT (kg N ha-1) is the
nitrous oxide generated by nitrification, and eN2ODEN (kg N ha-1) is the nitrous oxide
generated by denitrification; these were calculated as fractions of the estimated
emissions for ammonia (eNH3) and nitrate (eNO3), respectively.
Impact assessment method and interpretation
As an impact assessment method, we applied the CML2001 v.2.05 method (Guinée,
2002), considering the following midpoint impact categories: abiotic depletion (AD,
kg Sb eq), acidification potential (AP, kg SO2 eq), eutrophication potential (EP, kg
LF^G_ eq), fresh-water aquatic ecotoxicity potential (FAETP, kg 1,4-DB eq), global
warming potential (GWP, kg CO2 eq), human toxicity potential (HTTP, kg 1,4-DB
eq), marine aquatic ecotoxicity potential (MAETP, kg 1,4-DB eq), ozone layer
depletion potential (OLDP, kg CFC-11 eq), photochemical oxidation potential (POP,
kg C2H4 eq), and terrestrial ecotoxicity potential (TEP, kg 1,4-DB eq).
Aggregated values of N-emissions per cycle were allocated to the categories GWP
(N2O), HTTP (NH3), AP (NH3) and EP (NO-3 , NH3 and N2O), using equivalent factors
similar to those used by Brentrup et al. (2001). On the other hand, we allocated on-
field pesticide emissions to the FAETP, MAETP, HTTP and TEP categories. On-field
emissions of pesticides to soil were converted to FAETP, HTTP and TEP using the
characterization factors suggested by Hayer and Gaillard (2010). For pesticide
emissions into water bodies, the characterization factors for the same categories plus
42
MAETP were estimated using the uniform system for the evaluation of substances
adapted for LCA purposes from the Van Zelm et al. (2009) model, in the same way
as reported by Bojacá et al. (2014).
Normalization and weighting
To compare the environmental performance between OF and GH tomato production
systems considering the aforementioned impact categories, we implemented the
SMAA method as described by Prado-Lopez and Heijungs (2018). The outranking
procedure of the SMAA method generated an overall ranking of alternatives while
considering the uncertainty in the performances. In summary, the outranking
procedure steps were (1) defining a preference function, (2) calculating the outranking
scores, (3) generating stochastic weights, (4) calculating overall scores and (5)
defining a rank acceptability index. For the detailed description of these steps, the
reader can reference Prado-Lopez and Heijungs (2018).
The inputs for the outranking procedure were the distributions of the environmental
indicators of the alternatives under consideration. Since the primary data for the OF
and GH systems were collected for several production cycles, we had the variability
required to fit statistical distributions to the ten environmental indicators. However,
as indicated at the beginning of this article, the correlation structure among the
environmental indicators should be considered if a random sampling Monte Carlo
procedure is applied, as occurs with the SMAA method.
The first step was to fit theoretical statistical distributions to each environmental
indicator. Due to the skewed nature of the data, we selected the lognormal, gamma
and Weibull probability distributions as initial options to fit the observed data, using
the package fitdistrplus (Delignette-Muller and Dutang, 2015) from the R statistical
software package (version 3.5.1., R Core Team, 2018). The final selection of the
statistical distribution for each environmental indicator was based on the Akaike and
Bayesian information criteria.
To build multivariate distributions, the starting point is the recursive decomposition
of a multivariate density into products of conditional densities (Czado, 2010). The
pair copula decompositions are graphically represented through a sequence of nested
trees, or vine trees, with undirected edges. The tree edges indicate the indices
representing the conditional copula densities. Starting with the transformation of the
original data to uniform marginal distributions, we fitted a regular vine (R-vine)
copula model to the 10-dimensional data set. The method used a joint search of an
appropriate R-vine structure and its pair copula families and estimated their
parameters as a top-down strategy in an automatic way as developed by (Dißmann et
al., 2013). Starting with the first tree, all pairwise Kendall’s t values were calculated,
and their absolute values were used as edge weight to find a tree, maximizing the edge
weight sum among all possible trees. In the next step, the families and parameters for
all pair of copulae in the top tree were selected using the smallest AIC. These choices
were then used to estimate all pairwise Kendall’s t values for edges that were eligible
to be an edge in tree 2, maintaining the proximity condition necessary for the R-vine
tree structure. The corresponding copula families and their parameters were chosen
again by AIC. The complete R-vine specification was constructed by following this
procedure and by selecting the strongest pairwise conditional dependencies first
43
(Czado et al., 2013). This method was applied using the package VineCopula
(Schepsmeier et al., 2018) included in the R statistical software package.
Once the R-vine copula model was fitted, we performed the Monte Carlo procedure
for each environmental indicator considering a sample size of 10000. Since the
marginal distributions of the R-vine copula model were uniform, we transformed back
the sampled data to the aforementioned fitted distributions.
Next, we calculated uncertainty-based preference (Pi) and indifference (Qi) thresholds
based on the propagated uncertainty of the data. An outranking score was calculated
per pairwise comparison and per environmental indicator, and the net flow was
established as the sum of these scores. Then, we defined the weights for each
environmental indicator based on a stochastic procedure by assigning a distribution
to weight the factor values, resulting in an aggregated overall score that accounted for
all possible value systems (Prado-Lopez and Heijungs, 2018).
The final overall scores were calculated as weighted sums of the net flows with the
weight factors. These overall scores indicated the relative performance to other
alternatives in the comparison, with a higher overall score indicating a preferred
performance. Finally, a probability value per production system per rank was
configured based on ranking the production system per run.
4.3. Results and discussion

Life cycle inventory
Despite the low technological level and relatively low demand of the inputs to both
systems, the infrastructure component of the greenhouse system necessarily increased
the input demand per functional unit. However, the life span of most of the materials
included in the infrastructure are allocated to several crop cycles, decreasing their
potential impact. Comparing the structures used in Spain for tomato production
(Torrellas et al., 2012) with those employed by Colombian growers, the wood GH
frame reduces its lifespan (Table 6).
The greater use of machinery in the GH system is also a distinguishing factor with
respect to the OF system. According to the growers, 64% and 72% applied
mechanized soil tillage in a GH and OF, respectively. All OF growers used gravity-
fed irrigation systems, while 61.5% of the GH growers used a fertigation system
powered by a diesel pump.
The results of the inventory showed the high application rate of fertilizers of the
tomato growers in both production systems (Table 6). These results are in line with
the country trend, which places Colombia as one of the top fertilizer consumers in
Latin American (Sánchez-Navarro et al., 2013). The main fertilizers reported for the
GH system were ammonium nitrate (0.6%), manure (25.2%) and NPK fertilizers
(74.2%). Calcium nitrate is a commonly used N-source in a GH system, but because
the emissions estimation does not include it, we decided to categorize it as an NPK
fertilizer. Manure is broadcast on the field before transplanting without incorporation
into the soil. The other fertilizers are applied through basic fertigation systems with
the nutrient solution being prepared at each fertilization time. For the OF system, the
reported fertilizers were urea (4%), manure (5%), ammonium nitrate (9.3%) and NPK
44
fertilizers (81.7%). In this production system, these solid fertilizers were diluted in
water and manually applied to each plant.
The reduced pest incidence of fungal diseases, due to the protection given by the
greenhouse, also decreased the need to spray pesticides in terms of timing and dosing.
Since pest problems are common for both systems, growers share a wide range of
active ingredients. For fungicides, the most reported active ingredients by growers
were dimethomorph, chlorothalonil and mancozeb, while for insecticides, the most
reported were imidacloprid, thiocyclam and methomyl. Despite pesticide type, higher
spraying doses were reported by open field growers than by greenhouse growers. In
terms of fungicides, the plastic cover in the GH system in addition to the drip irrigation
system helped ensure the leaf area remained dry. However, free water due to
condensation can remain on the leaves, resulting in up to nine hours per day of a wet
leaf period (Gil et al., 2015), maintaining the need for fungicide applications.
Alternative pest control strategies such as biological or ethological control were not
applied in any system.
45
Table 6. Inventory data (mean and standard deviation) for the subsystems considered for the LCA, all are allocated to one ton of tomatoes.
Production Component Material Unit Greenhouse Open field
subsystem Lifespan Mean Standard Lifespan Mean Standard
(years) Deviation (years) Deviation
Infrastructure
Greenhouse Low-density kg 2.5 8.37 6.33 - - -
polyethylene
Sawn timber m3 10 0.025 0.019 - - -
Steel kg 15 1.23 1.38 - - -
Training system Wire kg 15 0.47 0.62 8 0.068 0.043
Polypropylene kg 1 4.81 2.83 1 0.41 0.24

Sawn timber m3 10 2.99E-04 2.49E-04 4 1.82E-04 1.70E-04
Irrigation system Polypropylene kg 5 0.23 0.26 5 0.12 0.16
Polyvinylchloride kg 10 0.065 0.076
Machinery
Soil preparation and Fuel kg - 6.07 4.52 - 2.86 5.47
irrigation
Soil preparation Tractor hours 10 0.22 0.15 0.56 0.87
Irrigation Pump hours 10 3.09 4.55 - -
Fertilization
Organic and inorganic N Total kg - 5.25 3.21 - 4.29 2.83
fertilizers P2O5 kg - 5.41 3.95 - 5.49 3.25
K2O kg - 9.62 7.02 - 6.3 3.63
Plant protection Inorganic pesticides Fungicides kg AI - 0.22 0.38 - 0.75 0.47
products
Insecticides kg AI - 0.067 0.095 - 0.13 0.12
Herbicides kg AI - 6.78E-03 3.21E-03 - 0.022 0.024
46
The emissions for N-fertilizers and pesticides are presented in Table 7. The emissions
of NH3 and N2O to the air were higher for the GH system than the OF systems,
although high variability in the results was observed for both systems. In the GH
system, 25.2% of the nitrogen comes from organic sources and is applied without
incorporation into the soil before transplanting. This fertilization practice favors NH3
losses since organic fertilizers have a higher volatilization potential than other
fertilizers (Bouwman et al. 2002). In the OF system, the application method is similar,
and the proportion of organic sources is lower; however, the urea applications result
in extra emissions and total NH3 emissions similar to those of the GH system. For
N2O, the higher input of nitrogen in the GH system than in the OF system directly led
to increased N2O emissions (Nemecek et al., 2016). On the other hand, the NO-3 losses
in the OF system were higher due to the accelerated water movement along the soil
profile as a consequence of the lower clay content compared to the GH system. Under
protected conditions, the increased topsoil NO-3 content was the result of the higher
N-application through the drip irrigation system on clayey soils. In comparison to the
GH system, in the OF system, the highest emissions into the soil and water were
directly linked to the larger use of pesticides, as shown above. The technological level
of the production system determined the emissions derived from the on-field use of
fertilizers and pesticides. Highly technological agricultural systems and accurate
dosing and timing of agrochemicals reduced their potential emissions, especially for
closed fertigation systems (Chen et al., 2008; Zhang et al., 2013). The two systems
under study represent opposing situations as characterized by the high use of
fertilizers and pesticides per functional unit, leading to an increased risk of emissions
to the environment.
Table 7. Emissions (kg t-1 functional unit) due to the field application of fertilizers
and pesticides in both tomato production systems.
Source Greenhouse Open field
Mean Standard Mean Standard

Deviation Deviation
Fertilization Air NH3 0.17 0.1 0.15 0.11
N2O 0.05 0.04 0.02 0.03
Water NO-3 0.6 0.65 0.73 0.78
Fungicides Soil AI 0.047 0.064 0.14 0.11
Water AI 3.6E-05 5.0E-05 7.5E-05 6.1E-05
Insecticides Soil AI 0.022 0.037 0.073 0.082
Water AI 1.2E-05 1.9E-05 2.4E-05 4.1E-05
Life cycle impact assessment

According to the average comparative LCIA, there is no clear difference in the
environmental performance between systems (Table 8). While the GH system yielded
lower impacts for categories such as AD, AP, EP, OLDP and POP, the OF system
performed better in the remaining categories. Looking at the LCIA results per
production system, the GH fertilization subsystem yielded the highest amount in
seven out of the 10 impact categories under consideration. For the categories AP, EP
47
and GWP, the share of this GH fertilization subsystem was above 90%, while for AD,
POP and TEP, the share was between 49% and 65%. For the OLDP category, the
potential impact was evenly distributed between the machinery, fertilization and pest
management subsystems. Overall, the contribution of the infrastructure subsystem
was low for most impact categories with shares equal to or below 10%.
For the fertilization subsystem, the results for the categories directly associated with
the use of fertilizers such as AP and EP indicated that they contributed up to 67 and
91%, respectively, of the field emissions. Nitrous oxide field emissions also
accounted for an important contribution of 35.5% to GWP. Regarding the pest
management subsystem, a similar situation was observed for the impact of the
pesticide field use that accounted for 76.3 and 96.6% in the FAETP and TEP
categories, respectively.
The fertilization and pest management subsystems were the main contributors to the
environmental impacts for all categories of the OF system. Within the fertilization
subsystem, the N-field emissions accounted for almost the same percentage reported
for the GH system in AP and EP. Of the six categories, the pest management
subsystem contributed the highest proportion, while of the remaining five, the
fertilization subsystem represented the highest share. The field emissions of pesticides
were the top contributor to TEP. Our results confirmed that mitigating field emissions
should be prioritized to improve the environmental performance of these production
systems (Zhang et al., 2013). The LCIA average values mask the wide range of
strategies used by growers in each production system, which in turn increases the
uncertainty of the environmental performance, as will be shown in the next section.
48
Table 8. Average of potential environmental impacts per ton of tomato produced in open field and greenhouse conditions in Colombian
Andes.
System/ Subsystem AD AP EP FAETP GWP HTTP MAETP OLDP POP TEP
(kg Sb eq) (kg SO2 eq) (kg !"#$% eq) (kg 1.4-DB eq) (kg CO2 eq) (kg 1.4-DB eq) (kg 1.4-DB eq) (kg CFC-11 eq) (kg C2H4 eq) (kg 1.4-DB eq)
Greenhouse 0.224 0.525 0.418 11.1 75.9 36.2 1.91E+04 2.82E-06 6.54E-03 0.608
Infrastructure 8.24E-03 5.69E-03 2.93E-03 0.52 1.14 0.809 1.59E+03 7.38E-08 2.45E-04 8.61E-03
Machinery 0.053 0.026 0.014 5.82 2.68 25.5 1.15E+04 8.56E-07 1.49E-03 0.068
Manufacture 0.015 0.016 0.012 5.69 1.86 24.9 1.09E+04 1.14E-07 9.37E-04 0.063
Field use 0.038 9.80E-03 1.42E-03 0.131 0.822 0.622 0.068 7.42E-07 5.56E-04 5.15E-03
Fertilization 0.142 0.478 0.393 1.26 0.697 3.62 3.65E+03 9.88E-07 3.72E-03 0.07
Manufacture 0.142 0.156 0.033 1.26 0.449 3.60 3.65E+03 9.88E-07 3.72E-03 0.07
Field use 0 0.322 0.360 0 0.247 0.02 0 0 0 0
Pest management 0.0201 0.0165 8.57E-03 3.49 2.38 6.29 2.27E+03 8.99E-07 1.08E-03 0.461
Manufacture 0.0201 0.0165 8.57E-03 0.827 2.38E+00 3.87 2.27E+03 8.99E-07 1.08E-03 0.014
Field use 0 0 0 2.67 0 2.42 5.50 0 0 0.447
Open field 0.283 0.555 0.452 8.35 58.1 21.9 1.41E+04 4.61E-06 9.79E-03 0.575
Infrastructure 2.23E-03 1.54E-03 7.92E-04 0.141 0.309 0.219 0.0430 2.00E-08 6.63E-05 2.33E-03
Machinery 0.103 0.0283 5.62E-03 0.896 3.11 3.53 3.19E+03 1.94E-06 1.71E-03 0.019
Manufacture 9.35E-03 4.35E-03 2.16E-03 0.575 1.10 2.01 1.52E+03 1.25E-07 3.48E-04 6.65E-03
Field use 0.094 0.024 3.47E-03 0.321 2.01 1.52 1.67E+03 1.82E-06 1.36E-03 0.013
Fertilization 0.117 0.408 0.422 0.800 47.6 2.46 2.37E+03 6.35E-07 2.42E-03 0.046
Manufacture 0.117 0.116 0.023 0.800 36.0 2.44 2.37E+03 6.35E-07 2.42E-03 0.046
Field use 0 0.292 0.399 0 11.6 0.02 0 0 0 0
Pest management 0.0602 0.117 0.024 6.51 7.06 15.70 0.311 2.01E-06 5.59E-03 0.507
Manufacture 0.0602 0.117 0.024 3.48 7.06 9.40 8.15E+03 2.01E-06 5.59E-03 0.059
Field use 0 0 0 3.03 0 6.30 0.303 0 0 0.448
49

The varying degree of correlation among the environmental indicators (Figure 8)
cannot be overlooked; therefore, a multivariate random sampling procedure was
followed to build a realistic sampling space to properly apply the SMAA outranking
procedure. The degree of association between the environmental indicators varied per
production system, but higher correlations were observed for the pairwise
comparisons of the impact categories of the OF system than those of the GH system.
Similar to what happened with the correlations between input parameters, ignoring
the correlations among impact categories would lead to an under- or overestimation
of the determination of a single indicator for each system. In the case of a comparative
LCA, Groen and Heijungs (2017) state that if a correlation exists, then its inclusion
leads to a well-informed decision about the environmental performance of a system.
Figure 8. Graphical representation of correlations between environmental indicators

for tomato production under greenhouse (a) and open field (b) conditions.
Based on the above information, we fitted statistical distributions to each
environmental indicator per production system but considered the dependence
structure between environmental indicators through the copula approach. Table 9
summarizes the outcome of the fitted distributions, which in all cases resulted in
positively skewed distributions. The graphical comparison between observed and
fitted statistical distributions, presented through their probability density functions, is
included in Appendix 1.
50
Table 9. Properties of the fitted statistical distributions to the environmental indicators distributions estimated for the greenhouse and open
field tomato production systems.
Environmental indicator Greenhouse Open field
Distribution Parameters1 AIC2 BIC3 Distribution Parameters AIC BIC
Abiotic depletion (AD) Log-normal µ = -1.632; s = 0.391 -83.63 -80.36 Log-normal µ = -1.571; s = 0.572 -27.22 -25.04
Acidification potential (AP) Log-normal µ = -0.817; s = 0.549 4.13 7.41 Log-normal µ = -0.775; s = 0.577 8.15 10.33
Eutrophication potential (EP) Log-normal µ = -1.136; s = 0.674 -4.56 -1.28 Log-normal µ = -0.991; s = 0.562 -2.54 -0.36
Fresh water aquatic ecotoxicity Weibull k = 1.746; l = 7.01 204.25 207.52 Gamma a = 4.026; b = 0.793 103.39 105.57
potential (AETP)
Global warming potential (GWP) Log-normal µ = 4.091; s = 0.696 395.21 398.49 Gamma a = 2.736; b = 0.048 216.50 218.69
Human toxicity potential (HTTP) Gamma a = 2.083; b = 0.086 312.22 315.49 Weibull k = 1.707; l = 16.57 156.24 158.43
Marine aquatic ecotoxicity potential Weibull k = 2.049; l = 16744.7 786.74 790.02 Weibull k = 1.712; l = 15036.8 455.62 457.80
(MAETP)
Ozone layer depletion (OLDP) Log-normal µ = -12.97; s = 0.501 -926.30 -923.03 Log-normal µ = -12.614; s = 0.57 -511.17 -64.77
Photochemical oxidation potential Log-normal µ = -5.262; s = 0.519 -337.90 -334.62 Log-normal µ = -4.832; s = 0.525 -174.52 -172.34
(POP)
Terrestrial ecotoxicity potential Gamma a = 1.978; b = 14.533 -80.14 -76.86 Gamma a = 4.521; b = 37.243 -62.97 -60.79
(TETP)
1
Parameters of the distributions: Log-normal: mean (µ) and standard deviation (s); gamma: shape (a) and scale (b); Weibull: shape (k) and
scale (l)
2
AIC: Akaike Information Criterion
3
BIC: Bayesian Information Criterion
51
The copula approach for the greenhouse and open field environmental indicator
datasets resulted in R-vine models consisting of 46 and 45 bivariate copulae,
respectively, requiring 47 parameters each. The R-vine structures for both greenhouse
and open field environmental indicators, including their corresponding pair-copula
families and Kendall’s correlations, are presented in Appendix 2. Each tree shows the
three components of the vine copulae: the tree structure, the copula family for each
edge in the tree structure and the corresponding dependence parameters for each
copula pair.
In both production systems, notably, after the second tree, a path line structure was
shown, which corresponds to a subclass of R-vines known as D-vines. This result
suggests the existence of a natural order among the impact categories (Czado et al.,
2013). Furthermore, in the third tree, the conditional dependencies started to be
captured, with the impact categories GWP, POP, MAETP, AETP and TETP
configuring the conditioning set. The same trend was observed for both production
systems. Starting with the first tree capturing the strongest pairwise dependencies, a
decrease in the magnitude of the pairwise conditional dependencies in the following
trees was observed until independence was reached between the edges of the last tree
(Czado et al., 2013).
Table 10 presents the log-likelihood and information criteria for the two D-vine
models fitted to the environmental indicators of each production system, indicating
similar performances. A slightly better fit was observed for the D-vine model of the
GH system than the OF system since the best AIC and BIC were found for this model.
Table 10. Log-likelihood, Akaike information criterion (AIC) and Bayesian
information criterion (BIC) of the estimated D-vine models for the environmental
indicators of the greenhouse and open field production systems.
Log-Likelihood AIC BIC
Greenhouse 413.73 -733.46 -656.49
Open field 413.73 -716.56 -639.59
Although the theoretical foundations of copulae are complex, their algorithms have
already been developed in existing software packages, making them available and
relatively easy to couple with other methods, such as uncertainty analysis. Within the
LCA context, the use of copulae for uncertainty analysis enables the possibility of
working with a wide range of probability distributions, including the intrinsic
correlation of the input variables or the impact categories.
The final step of the SMAA method considers building a probabilistic rank that
compares the likelihood between alternatives, i.e., GH and OF production systems.
Figure 9 presents the cumulative distribution functions of the rankings for each tomato
production system. In this case, since the CDFs did not cross, we inferred a rank-
ordering with first-order stochastic dominance (Canis et al., 2010). This first-order
dominance means that the OF system prevailed over the GH system since the CDF
for the OF system was smaller, or at most equal, to that of the GH system. In this case,
a decision maker with a completely uncertain weighting preference should classify
the OF tomato production system as having the best environmental performance.
52
Figure 9. Cumulative distribution functions of the rankings for each tomato

production system.
An additional way to consider these results is through the rank acceptability index
(RAI), which assigns an overall probability to each production system based on
counting the rank of each alternative obtained on each run (Prado-Lopez and
Heijungs, 2018). The OF production system yielded a RAI of 57.6%, ranking first in
terms of environmental performance compared with that of the GH production
system, which scored an RAI of 42.4% (Figure 10). The opposite results are presented
for the second position in the rank since there were only two alternatives (i.e.,
production systems) under consideration. Thus, the OF system is more likely to
perform better environmentally than the GH tomato production system mainly
because the GH system employs more inputs per ton of tomato than the OF system.
The better environmental performance of the OF system is related to less intensive
use of infrastructure, machinery and agrochemicals, specifically fertilizers. In the
particular case of pesticides, although its use is greater in OF system, there are no
important differences with respect to pesticide use under the GH.
53
Figure 10. Rank acceptability index for the tomato production systems under
greenhouse and open field conditions.
These results also highlight that the Colombian tomato production system under
protected conditions has not been properly implemented since the intensification
inherent to a GH production system should lead to an optimized use of resources and
simultaneously boost productivity per unit area. The protection provided by the
greenhouse cover in conjunction with appropriate ventilation and auxiliary
mechanisms to limit the insect infestation (anti-insect meshes) should result in less
pesticide application and better environmental performance of the system. With
regard to the application of fertilizers, we hypothesize that a reduction in the quantity
and fractionation based on the plant's demands should be reflected in a better
environmental performance of the system, as will be demonstrated in the chapter
seven.
4.4. Conclusions
The present work demonstrated the application of the SMAA method to compare the
environmental performance of agricultural systems by deriving a single, unbiased
indicator that groups, in a stochastic way, the individual outcome of the categories
selected from the impact assessment method. A typical evaluation method based on
the comparison of averages or median values of the impact categories ignores the
inherent variability in agricultural production systems, as well as the correlation
between the impact categories. In contrast, the approach presented, overcomes these
limitations and improves the accuracy of detecting significant differences in different
LCAs. Additionally, the SMAA method avoids any bias in the weighing procedure,
resulting in an objective selection approach of the most appropriate alternative from
an environmental performance perspective.
To obtain results closer to reality, the dependence structure, considered in the copula
method, needs to be constructed based on local primary data if comparative
54
agricultural LCAs at a regional scale are to be performed. Low-tech and yield-

constrained agricultural systems subject to varying biophysical conditions, such as
those run by smallholders in developing countries, exhibit a wide range of
management practices exacerbating the intra- and inter-regional variability in crop
performance and subsequently the environmental impacts of these systems. Local
information systems for this type of agricultural system are not in place, making the
situation even worse, and bookkeeping practices are not common among the majority
of growers. Therefore, as in the case of Colombian tomatoes, as well as for many
other locally consumed horticultural products, challenges remain related to
constructing proper information systems that collect reliable primary data that are
required not only for environmental assessments but also for of the development of
public policies.
The copula method introduced in this work can be applied to any type of uncertainty
analysis involving a Monte Carlo simulation or multicriteria decision analysis where
the correlation structure needs to be accounted for. The copula method is
computationally inexpensive, and currently, with the wealth of distribution families
available, it is possible to model any type of multivariate joint distributions at higher
dimensions, moving away from the constraints imposed by the multivariate normal
scenario.
The link between the copula method and the SMAA method to model the correlation
structure between environmental indicators to produce an unbiased preference
indicator opens the possibility of evaluating optimization strategies in a more realistic
way, leading to results that can easily be communicated to decision-makers.
Moreover, the evaluation of these optimization strategies through soil-crop modeling-
based scenarios rather than through field experiments or observation allows an
assessment of a broad range of strategies only limited by the restrictions and
assumptions imposed by the mechanistic model under consideration.
55
Chapter 5. A tailor-made crop growth model for the tomato in
Colombia
Published as: Gil, R., Bojacá, C. R., & Schrevens, E. (2017). A tailor-made crop
growth model for the tomato production systems in Colombia. Agronomía
Colombiana, 35(3), 301-313.
Abstract
Potential crop models simulate the plant growth under nonlimiting biophysical
conditions with no other factor than the climate to which the plants are exposed to.
These models may fail to adequately represent the crop performance if they are not
adapted to the local conditions. The particularities of Colombian tomato systems
(greenhouse and open field) demand the recalibration of existing models to make a
more realistic representation of those systems. Therefore, a locally calibrated crop
model was proposed considering both production systems. For this purpose, four on-
farm calibration experiments were carried out, two under greenhouse conditions with
average temperatures of 17.4 and 17.9 ºC in Santa Sofía (Boyacá) and two under open
field conditions in Páramo and San Gil (Santander), with average temperatures of 20.6
and 24.0 ºC, respectively. The crops were commercially managed according to the
local practices. Plant data were collected through destructive measurements carried
out on a fortnightly basis, while climate data were collected for the entire crop growth
cycle. Independent calibration of the dry matter fractions allocated to the plant organs
in function of thermal time resulted in an acceptable model performance. The
calibration of the model under commercial conditions gave a better representation of
the local systems but at the expense of accuracy since on-farm experiments cannot be
controlled as well as those performed in research facilities.
A tailor-made crop growth model for the tomato in Colombia
5.1. Introduction
During the last 40 years, crop system simulations have evolved from a new science
with inadequate computer power to a robust and increasingly accepted science
supported by improved software and computing capabilities (Boote et al., 2010). Over
the last decade, the most significant demand for cropping system models has involved
assessing climate change impacts on agriculture and evaluating mitigation and
adaptation strategies, conducted over different spatial scales and degrees of
agricultural system complexity (Stöckle et al., 2014).
Most crop system models have evolved as compilations of crop and soil models
focusing on modeling a single point in space over time to explore variability in crop
responses to soil, management and weather (Jones et al., 2017). The crop modeling
component simulates phenology and partitioning and integrates processes of carbon
(C), nitrogen (N) and water balance from planting to maturity, providing a final yield
and production as well as daily values of crop components over time (Boote et al.,
2013). This potential yield of an adapted cultivar is solely determined by solar
radiation, temperature, carbon dioxide, and genetic traits that govern the length of the
growing period The light interception by the crop canopy and its conversion to
biomass results in the partition of biomass to the different organs (Grassini et al.,
2015). In this approach, crop growth is not constrained by factors such as water,
nutrients, diseases or pests.
In the developed world, single crop models are well established (Stöckle et al., 2014),
but the development of crop growth models for less developed agricultural production
areas remains an important study subject. In their overview of crop growth and yield
models, Di Paola et al. (2016) showed this unbalanced situation in which most of the
available models have been applied for temperate regions with a couple of references
for subtropical regions in Brazil and Mexico. Moreover, the development of such
models has been focused on staple crops such as cereals, sugar beet and potato.
Tomato is among the most important horticultural crops worldwide. A variety of
tomato growth models have been developed in the past (i.e., Heuvelink, 1999; Jones
et al., 1991; Scholberg et al., 1997; Soto et al., 2014; Valdés-Gómez et al., 2014) with
different levels of complexity and for different purposes. Tomato crop growth models
have been included in decision-support systems such as the world-renowned DSSAT
(Jones et al., 2003) as well as many others (Jizhang et al., 2006; Massa et al., 2013).
Moreover, 3D functional structural models of tomato plants have been developed for
purposes such as optimizing LED lighting to increase light absorption and crop
growth (de Visser et al., 2014).
In Colombia, Cooman (2002) evaluated the feasibility of protected tomato cropping
in the high-altitude tropics by locally calibrating the second version of the TOMGRO
model through controlled experiments on the Bogota Plateau. He modified the model
by reducing the leaf expansion rate at a low temperature and incorporating a direct
effect of temperature on the distribution of dry matter between the vegetative and
generative plant organs. This modified version of the TOMGRO model was later
applied by Bojacá et al. (2009) to evaluate the variability in greenhouse tomato yield
caused by spatial temperature variations.
57
However, most Colombian tomatoes are grown throughout the year in several Andean
Mountain valleys and hills in warmer climates at altitudes below those of the Bogota
Plateau. Tomato is a small-scale business represented by clusters of growers
cultivating tomato under one of the two established systems: open field or greenhouse
production. Under both systems, growers apply suboptimal practices despite the
differences in the demand for resources per unit area (Bojacá et al., 2014, 2013).
As process-based crop models closely reflect the behavior of particular crops, a
locally calibrated growth model is needed for the Colombian tomato systems.
However, this calibration is a highly data-demanding task and specific to the available
data, preventing broader applicability (Robertson et al., 2013). On the other hand,
most calibration experiments are carried out under controlled conditions, which in
some cases are not representative of those observed under natural field conditions
(Craufurd et al., 2013).
Thus, the objective of the present work is to propose a summary tomato crop growth
and development model with the ability to simulate open field and greenhouse
production, calibrated under potential conditions. Although the experiments were
carried out on-farms of local producers, it was guaranteed that the climate was the
main determinant of the plant growth.

Model description
The model structure, presented in Figure 11, is based on earlier crop growth models
(Cooman, 2002; Gil et al., 2017b). The model runs on a daily basis, with the exception
of gross photosynthesis, which runs on an hourly basis. All model calculations are
performed on a per-plant basis. Starting with initial values for all state variables at the
beginning of the simulation, once the climate (air temperature, relative humidity and
global radiation) for the corresponding day is updated, the total dry matter production
is simulated followed by its distribution among the aboveground plant organs and
roots.
58
Figure 11. Schematics of the proposed tomato crop growth model for the open field
and greenhouse production systems, which is a simplification of the TOMGRO model
(Jones et al., 1991).
Dry matter production
The amount of dry matter available for growth was calculated at the end of the day as
the difference between gross photosynthesis and maintenance respiration. Daily gross
photosynthesis resulted from the integration of the photosynthetic rates calculated on
an hourly basis. The photosynthetic rate depends mainly on the photosynthetic active
radiation (PAR) absorbed by the canopy leaf area (LAI), the air temperature and the
CO2 concentration, as modeled by Acock et al. (1978). The model considers
restrictions on the photosynthetic rate due to extreme temperatures and vapor pressure
deficits. The group of equations that govern the photosynthesis process are presented
below.
"%&' (1 − (1 ) × "%&' + 45 × () × ""67
!"# = × +, - =
() (1 − (1 ) × "%&' + 45 × () × ""67 × 8 9':×;&< (12)
× 3600
""67 = (AB+CDE × 0.47) × 4.57 (13)
"%&' = J × LMN × "O"7 × "!CP7 (14)
"O"7 = 8 QR)×(STU9STU;)V (15)
0, \] ^_" < 9ºL
⎧ ^_" − 9, \] 9 ≤ ^_" < 10ºL
⎪
"!CP7 = 1, \] 10 ≤ ^_" < 28ºL (16)
⎨−0.083 × ^_" + 3.33, \] 28 ≤ ^_" < 40ºL
⎪
⎩ 0, \] ^_" ≥ 40ºL
59
where !"# is the hourly gross photosynthesis (μmol CO2 m-2 h-1), "%&' is the
maximum leaf photosynthetic rate (μmol CO2 m-2 s-1), () is the light extinction
coefficient, (1 is the leaf light transmission coefficient, 45 is the leaf quantum
efficiency (μmol CO2 μmol-1 photon), ""67 is the photosynthetic photon flux density
(μmol photons m-2 s-1), ghi is the leaf area index, AB+CDE is the hourly solar radiation
(J m-2 s-1), J is the CO2 use efficiency (μmol CO2 m-2 s-1 ppmv-1), LMN is the carbon
dioxide concentration in air (ppmv), "O"7 is a function that corrects the "%&' for the
air vapor pressure deficit, Lj is a factor used to determine the effect of vapor pressure
deficit on photosynthesis (kPa-1), O"7 is the air vapor pressure deficit (kPa), O"7g
is a factor used to determine the effect of vapor pressure deficit on photosynthesis
(kPa), "!CP7 is a function that corrects the Plmn for suboptimal temperatures and
^_" is the hourly mean temperature (°C).
At the end of each day hourly gross photosynthesis is integrated, and the result is
transformed into the amount of carbohydrates synthesized by the plant at the current
day, following this formula:
Nq
30 × 109t
!"o = p !"# × (17)
"g_2
#rs
Where !"o is the daily gross photosynthesis (g CH20 d-1 per plant), !"# is the hourly
gross photosynthesis (μmol CO2 m-2 h-1), "g_2 is the plant density (plants m-2),
30 × 109t is used to transform the units from μmol to mol, and then from mol to
grams.
Total respiration per plant was represented by the maintenance respiration and the
growth efficiency. Daily maintenance respiration was calculated as a fraction of the
accumulated dry matter in stems, active leaves and growing fruits at a reference
temperature of 20°. Then, the maintenance respiration was corrected for temperature
using a Q10 value. Next, we presented the equations that describe this module.
_uvw = 410x.s×Qy%Tz{|9NxV (18)
× (C_Cg × (7_+ + 7_}) + C_C6 × 7_])
where _uvw is the maintenance respiration per day (g CH2O d-1 per plant), ^_"~Ä is
the daily average temperature (°C), C_Cg is a respiration coefficient for stem and
leaf tissues (g CH2O g-1 DM d-1 per plant), 7_+ is the dry matter in leaves (g DM per
plant), 7_} is the dry matter in stems (g DM per plant), C_C6 is a respiration
coefficient for growing fruits (g CH2O g-1 DM d-1), 7_] is the dry matter in fruits (g
DM per plant) and 20 is the reference temperature for the given respiration
coefficients.
Based on the above, the daily biomass production per plant was calculated using the
following expression:
7_Å = (!"o − _uvw ) × !CP6 (19)
where 7_Å is the total dry matter produced that day (g DM d-1 per plant), !"o is the
daily gross photosynthesis (g CH2O d-1 per plant), _uvw is the maintenance respiration
per day (g CH2O d-1 per plant) and !CP6 is a growth efficiency coefficient (g DM g-
1
CH2O), representing growth respiration.
60
Dry matter distribution

We considered the organs in the model as single units (i.e., big-leaf model approach),
meaning no dry matter distribution occurred among cohorts or sympodial units. The
dry matter synthesized each day is allocated to the plant organs through accumulated
thermal time (ATT, °Cd)-dependent functions. Each function describes the proportion
(on a scale from 0 to 1) of the daily dry matter assigned to the considered organ. The
base temperature at which plant growth starts was set at 10°C (Valdés-Gómez et al.,
2014). A fixed fraction (9%) of the daily dry matter was allocated to the roots. The
equations describing dry matter distribution were as follows:
ÑÇ
7_Ç = 7_Å × É + EÇ á (20)
1 + DÇ × 8 9ÖÜ×&yy
Ñà
7_à = 7_Å × - = (21)
1 + Dà × 8 9Öâ×&yy
7_ä = 7_Å − Q7_Ç + 7_à + 7_ã V (22)
where 7_Å is the total dry matter produced that day (g DM d-1), 7_Ç is the daily dry
matter allocated to leaves (g DM d-1), 7_à is the daily dry matter allocated to fruits
(g DM d-1), 7_ä is the daily dry matter allocated to stems (g DM d-1), 7_ã is the
daily dry matter allocated to roots (g DM d-1), h^^ is the accumulated thermal time
(°Cd) and Då , çå , Ñå and Eå are the parameters that must be fitted for each function, and
the sub index i represents the corresponding plant organ (leaves, stems and fruits).
In the integration step, the daily dry matter allocated to each organ was added to the
one accumulated from the previous days to determine the total dry matter in leaves
(^7_Ç ), fruits (^7_à ), stems (^7_ä ) and roots (^7_ã ). The daily leaf area was
calculated as the total dry matter corresponding to the active leaves multiplied by the
specific leaf area (SLA). Once leaf senescence started, the fraction of dry matter
distributed to these leaves was estimated as a function of the ATT. The proportion of
senescent leaves (APéÇ ) per plant was calculated in the following way for open field
tomatoes:
0, \] h^^ < 677
APéÇ = è1.08 × 109ê × h^^ − 0.73, \] 677 ≤ h^^ ≤ 1600 (23)
1, h^^ > 1600
For tomatoes in the greenhouse, the APéÇ per plant was calculated as follows:
Ñä5í
APéÇ = (24)
1 + Dä5í × 8 9Öìîï×&yy
where, h^^ is the accumulated thermal time (°Cd), Dä5í , çä5í and Ñä5í are the
parameters to be fitted.
On the other hand, ripe fruits per plant ready to be harvested were calculated with the
following function:
Ñãà
6#ñ = ^7_à × - = (25)
1 + Dãà × 8 9Öóâ×&yy
61
where 6#ñ is the dry matter of ripe fruits (g DM per plant); ^7_à is the accumulated
dry matter in fruits throughout the crop cycle (g DM per plant); h^^ is the
accumulated thermal time (°Cd); and Dãà , çãà and Ñãà are the parameters that must be
fitted.
The leaf area (LA) per plant was calculated based on the DM allocated to leaves and
the SLA according to the following expression:
^7_ÄÇ = ^7_Ç × (1 − APéÇ ) (26)
gh = ^7_ÄÇ × Agh (27)
where gh is the plant leaf area (m2 plant-1); ^7_ÄÇ is total dry matter of
photosynthetically active leaves (g DM per plant); Agh is the specific leaf area (m2 g-
1
) estimated as 0.019 and 0.021 for greenhouse and open field tomatoes, respectively;
^7_Ç is total dry matter allocated in leaves (g DM per plant); and APéÇ is the
proportion of senesces for the leaves. Finally, the ATT was calculated as follows:
ío
h^^ = p ^5à (28)

ors
0, \] ^_"~Ä ≤ ^Ö
^5à = ò (29)
^_"~Ä − ^Ö , \] ^_"~Ä > ^Ö
where h^^ is the accumulated thermal time (°Cd), ,E is the number of days of the
growing cycle, ^5à is the daily effective temperature (°C), ^_"~Ä is the daily average
temperature (°C) and ^Ö is the base temperature (10°C). The values of the parameters
included in the previous model equations are shown in Table 11, while the values of
the fitted parameter values (Då , çå , Ñå , and Eå ) for equations (20), (21), (24) and (25),
respectively, are shown in the Results section.
Table 11. Parameters included in the tomato growth model
Meaning Abbreviation Value Units
Air carbon dioxide concentration CO2 405 ppmv
Light extinction coefficient XK 0.58 Dimensionless
Leaf light transmission coefficient XM 0.091 Dimensionless
Leaf quantum efficiency 45 0.0645 μmol CO2 μmol-1 photon
Carbon dioxide use efficiency J 0.0693 μmol CO2 m-2 s-1 ppm-1
Sensitivity to temperature Q10 1.4 Dimensionless
Effect of VPD on photosynthesis VPDL 4.0 kPa
Effect of VPD on photosynthesis CK -0.8 kPa-1
Respiration rate for leaves and stems RMRL 0.015 g CH2O g-1 DM d-1
Respiration rate for fruit RMRF 0.01 g CH2O g-1 DM d-1
Growth efficiency GREF 0.75 g DM g-1 CH2O
62
Crop model calibration
Field experiments
The experimental work was conducted in 2016 in four commercial tomato production
plots, with two plots planted under open field conditions and the other two planted
under greenhouse conditions. The crops were planted and managed according to the
the best practices to ensure no effect of factors as water or nutrient stresses. The crops
were planted in two of the most representative tomato production areas of Colombia.
Alto Ricaurte Province in the department of Boyaca is one of the major tomato
greenhouse production areas in Colombia, while Guanentá Province in the department
of Santander is an important production area for open field vegetables, including
tomato. Table 12 describes the general characteristics of the experiments carried out
to calibrate the proposed crop growth and development model. Next, we present a
general description of the management practices applied in the experimental fields for
both production systems.
The protected experiments were carried out under plastic, naturally ventilated
greenhouses with wooden structures. Plants were grown on a single stem of
indeterminate length by periodically removing side shoots. Plants were trained
following a high wire system, and no fruit pruning was performed. The leaves located
under each of the harvested trusses were removed since they no longer contributed to
plant photosynthesis and were susceptible to infection by fungal diseases. Nutrients
were delivered through a fertigation system with the irrigation water.
For the open field experiments, determinate growth cultivars grew freely without any
leaf or fruit pruning, and the shoots were trained to an elevated wire. Solid fertilization
was carried out throughout the cropping cycle with the amounts and timing defined
by each grower. Under both systems, pest and disease management was entirely with
pesticides and with a spraying schedule defined according to the growers’ criteria.
63
Table 12. General characteristics of the on-farm experiments used to calibrate the tomato crop growth model.
Plot code Production system Location Altitude (masl) Planting date Cycle length Density Cultivar Plot area
(days) (plants m-2) (ha)
GH1 Greenhouse 5º 42’ 26.7’’ N – 73º 2346 28/01/2016 131 3.0 Libertador 0.28
36’ 4.1’’ W
GH2 Greenhouse 5º 44’ 8.0’’ N – 73º 2347 28/03/2016 113 3.9 Roble F1 0.28
36’ 13.1’’ W
OF1 Open field 6º 25’ 15.4’’ N – 73º 1703 27/01/2016 82 1.3 DRD 4.0
11’ 56.7’’ W
OF2 Open field 6º 28’ 55.4’’ N – 73º 1140 27/01/2016 97 1.3 Roble F1 1.0
6’ 54.7’’ W
64
Data collection
We collected the data for the model calibration through a series of destructive
biometrical measurements carried out for each experimental plot. Starting at
transplanting time and on a fortnightly basis, we removed the aerial parts of three
plants from each experimental plot. Under all conditions, the sampled plants were
surrounded by edge plants. Then, we divided the plant into its organs and weighed the
separate organs after being oven dried at 70°C for at least 72 hours. The leaf weight
included the weight of the blades and all petioles.
Once fruit harvest and leaf pruning began, we registered the amount of biomass
removed from the plant, and a sample was taken to determine its dry matter content.
The grower defined the frequency and amount of biomass harvested or removed
according to his/her criteria. We determined the leaf area by taking digital pictures of
all the active leaves present at the moment of the destructive measurement. From the
digital pictures, we extracted the number of pixels representing the leaves and those
of a reference object of known area also included in each picture. To determine the
leaf area, we implemented a decision tree algorithm, which was applied on images
including a reference square. All pictures were taken at the same height through a
fixed mount tripod. We estimated the corresponding leaf area through the relation
between the number of pixels of the reference object and the number of pixels
corresponding to leaf surfaces. This image processing step was carried out with the R
statistical software package (Team R Core, 2017). All the data collected from the fruit
harvest, leaf pruning and leaf area were later integrated on a per plant basis.
The calibration was a function of the grower’s decision to continue with his/her crop.
Therefore, the number of destructive measurements was variable and specific for each
experimental plot. Regularly, greenhouse growers are able to extend a cropping cycle
for a longer period than that of an open field system. For the greenhouse plots, we
were able to carry out ten and nine destructive measurements for GH1 and GH2,
respectively, while for the open field plots, we performed seven and eight destructive
measurements for OF1 and OF2, respectively. In all cases, the destructive
measurements were carried out until the end of the cropping cycle, ensuring that the
complete plant cycle was characterized through these measurements. Table 12
includes the duration of the crop cycle for each experiment.
As global radiation, air temperature and relative humidity are input variables for the
model, we collected these data by placing the required sensors within the experimental
plots. We recorded the hourly weather data using a Vantage Pro2 Weather System
(Davis Instruments, Hayward, CA, USA) for each of the open field experimental
plots. For the greenhouse plots, we installed two copper-constantan thermocouples
linked to a datalogger (Cox-Tracer Junior, Escort DLS, Edison, NJ, USA) to register
dry and wet bulb temperatures. Through the psychrometric relationship between these
two temperatures, we derived the air relative humidity. We placed the thermocouples
inside a ventilated white capsule to avoid the readings being affected by direct
radiation. The global radiation within the greenhouses was measured throughout the
growing period with a pyranometer (Model LI200RX, Campbell Scientific, Inc.,
Logan, UT, USA) placed at 2.5 m above the ground. A weather station (Model
Vantage Pro2, Davis Instruments, Hayward, CA, USA), placed outside the
greenhouses, registered the external hourly climate.
65
Calibration of dry matter partitioning

We focused the calibration of the proposed model on the dry matter partitioning
among organs because this is the key process defining the overall growth and
development of a plant. Thus, for each tomato system, we fitted individual models to
the fractions that define the amount of daily dry matter allocated to leaves, Equation
(20), and fruits, Equation (21), as a function of ATT. The parameters for each model
were estimated through the Nelder-Mead algorithm for derivative-free optimization
(Kelley, 1999) implemented in the dfoptim package (Varadhan and Borchers, 2016)
of the R statistical software package (Team R Core, 2017). We followed the same
procedure for the models that define the fractions of senescent leaves, Equation (24),
and ripe fruits, Equation (25).
A fixed fraction of 9% was allocated to the dry matter of the roots (Gil et al., 2017b).
The fraction of the dry matter allocated to the stems was calculated as the remaining
fraction after discounting those allocated to fruits, leaves and roots.
The statistical analysis comparing the observed field data and the simulated values
included the following statistical criteria: Bias (g DM plant-1), root mean square error
(RMSE, g DM plant-1), and model efficiency (EF, dimensionless). A full description
of these goodness-of-fit measures can be found in Wallach (2006).
5.3. Results and Discussion

Climate conditions
The daily average climate conditions under which the calibration experiments were
carried out are summarized in Figure 12. The climate conditions for the greenhouse
experiments were similar since both greenhouses were located near each other in the
same municipality. However, being planted on different dates resulted in some
climate differences, especially those related to radiation levels. The global radiation
level experienced by plants in the GH1 experiment was higher than that observed in
the GH2 experiment. As the open field experiments were located at a lower altitude,
these plants grew at higher temperatures than those in the GH experiments with
averages above 20°C. The climate of the OF2 experiment had the highest temperature
and radiation levels compared to those of the other three experiments. The lower
radiation levels of the other experiments are explained by the plastic covering in the
greenhouse experiments and by the geographical location of the OF1 experiment.
This open field experiment was located on top of a mountain with permanent cloud
cover observed throughout the data collection period. The climate in OF2 is
characterized by higher radiation levels and temperatures than other experimental
locations. Oppositely to OF1, experiment OF2 was located in a narrow valley near the
Fonce River, and during the data collection period is characterized by sunny days.
With respect to relative humidity, these values were similar in the experiments carried
out for the greenhouse system, but for the open field system, the OF1 experiment had
higher air humidity than that of the OF2.
66
Figure 12. Daily average air temperature (A), relative humidity (B) and solar radiation
(C) during the calibration experiments carried out under greenhouse and open field
conditions.
The temperatures registered in the open field experiments were more suitable for
tomato cropping than those of the greenhouse experiments. Despite the use of plastic
coverings, the average temperatures for the night hours (18:00 – 5:00) were 15.4 and
14.8°C for the GH1 and GH2 experiments, respectively, while for the OF1 and OF2
67
experiments, they were 18.5 and 21.3°C, respectively. During the day hours (6:00 –
17:00), the GH1 experiment had an average temperature of 19.4°C, while the
temperature during the GH2 experiment was warmer, with an average of 21°C. Higher
daily temperatures were observed for the open field experiments than the greenhouse
experiments, with averages of 22.8 and 26.7°C for OF1 and OF2, respectively.
Dry matter distribution calibration
Dry matter allocation to plant organs is a process linked to the total dry matter
accumulation of a plant. Since the daily amount of assimilates produced by the
photosynthetic process is a function of climate conditions and the available leaf area,
the dry matter fraction allocated to leaves defines the daily dry matter produced by a
plant. Therefore, with the calibration of the dry matter distributed to leaves and fruits,
we simultaneously calibrated the total dry matter plant accumulation. The fitted
parameters of the functions defining the fractions of daily dry matter allocated to
leaves, Equation (20); fruits, Equation (21); senescent leaves, Equation (24); and ripe
fruits, Equation (25), as a function of ATT are presented in Table 13.
Table 13. Fitted parameters for dry matter allocation in leaves (DMl) and fruits (DMf),
leaf senescence rate (SENl) and ripening fruit rate (Fhvt).
Fraction Open field Greenhouse
a b c d a b c d
!"# 0.044 -0.005 0.547 0.251 0.006 -0.012 0.377 0.301
!"$ 51.34 0.006 0.521 - 136.23 0.012 0.543 -
%&'# - - - - 3239.37 0.011 0.735 -
()*+ 1083 0.009 0.668 - 1000.00 0.009 0.668 -

0
Because the dry matter distribution fractions were calibrated as a function of ATT, we
present the cumulated ATT of the four experiments. The highest accumulation of ATT
occurred in the OF2 experiment with a value of 1371.5°Cd, followed by that in the
GH1 experience at 972.5°Cd. The GH2 and OF1 experiments had similar ATTs of
888.7 and 885.6°Cd, respectively. Since the dry matter distribution fractions were
calibrated as a function of thermal time, we removed the time effect, allowing a more
general application of these temperature-dependent functions.
The graphical representation of the dry matter partitioning functions of the plant
organs is depicted in Figure 13. The initial calibration procedure considered of unique
dry matter distribution functions for both tomato types. However, the results of this
calibration procedure and the low values of the goodness-of-fit measures indicated
that independent calibration procedures should be followed for each tomato
production system.
68
Figure 13. (A) Greenhouse and (B) open field tomato dry matter partitioning fractions
as a function of thermal time for each of the plant organs.
As stated previously, the fraction allocated to the roots was fixed to 0.09, while for
the aboveground organs, the calibration was carried out for the leaves and fruits
fractions. The stem fraction was calculated as 1 minus the sum of the other fractions.
A marked difference in partitioning functions was found between the production
systems. Under greenhouse conditions, tomatoes showed a greater decline in the dry
matter allocated to the leaves and stems compared to the decline observed for the open
field tomatoes. Even under open field conditions, the plant started allocating a higher
proportion of assimilates to the leaves, and then, the stem fraction increased and
stabilized at a value of 0.2.
The observed behavior of organ fractions was defined by the growing habit of each
tomato type and the way each production system is handled by the growers. Under
open field conditions, tomato cultivars were mostly of determinate growth, and
growers did not apply any shoot pruning. Therefore, these plants had a higher stem
fraction than that of the indeterminate single-stem tomatoes planted under greenhouse
conditions.
After the vegetative growth stage, the photosynthetically active leaf fraction of the
open field plants declined to a minimum at the end of the growing cycle. Most of the
remaining leaves were part of the senescent fraction. On the other hand, a higher and
more constant fraction of active leaves was observed for the greenhouse plants than
for open field plants, a situation that is characteristic of their indeterminate growth
habit.
The fraction allocated to the fruits in the open field tomatoes showed a gentle slope
compared to the trend observed for the greenhouse conditions. However, at the end
of the growth cycle, the fraction of ripe and growing fruits accounted for
approximately half of the dry matter produced by the plant. While the same pattern
was observed for greenhouse tomatoes, in this case, the fruit fraction was stabilized
and remained constant at approximately 1000°Cd ATT. Under both production
systems, it is important to note the fraction of growing fruits that remained on the
plant. As the crop reached the end of its production cycle, the amount of harvested
69
fruit should have been higher than the amount remaining on the plant, especially for
open field tomatoes. Nevertheless, under local conditions, growers did not properly
balance the vegetative and generative growth of the plant or apply proper pollination
and pruning strategies.
Once the dry matter distribution functions for each tomato type were calibrated, we
incorporated them into the model. The observed and simulated total dry matter per
plant and its allocation to the plant organs are presented in Figure 14. In most cases,
the simulated dry matter closely followed the pattern depicted by the observed field
data. The observed data also included not only the average of the sampled plants but
also the standard deviation as a variability measure. Especially for the open field
experiments and in particular for the last destructive measurements, there was an
important variation in the data collected in the field.
Figure 14. Observed and simulated dry matter accumulation and distribution
throughout the plant organs for the calibration experiments carried out under
greenhouse and open field conditions. Vertical bars represent plus and minus one
standard deviations.
Table 14 presents the goodness-of-fit measures selected to establish crop growth
model performance compared to the observed field data performance. As the dry
matter allocation fractions to the plant organs were estimated independently for each
tomato type, we also presented the goodness-of-fit measures per type of production
system. Bias quantifies the average difference between measured and simulated
values, with the best fit indicated when the Bias index is closer to zero. According to
the results for the whole plant and for each organ, we observed a better model fit to
the open field condition since values were closer to zero than those obtained for
greenhouse tomatoes. In most cases, the Bias results were positive, indicating that the
model tended to underestimate values, especially for the fruit dry matter since a higher
Bias value was obtained for this organ and for both systems. The underestimation
70
reported by this index was a common pattern observed in particular for the first
measurement dates (Figure 14). Only the Bias for the total dry matter per plant in the
open field condition was negative, indicating an overall overestimation of the model.
In general, the Bias was close to zero.
Table 14. Goodness-of-fit measures of the simulated dry matter per plant and per
organ by the calibrated tomato crop growth model.
Plant organ Greenhouse system Open field system
Bias (g DM RMSE (g DM EF Bias (g DM RMSE (g DM EF

plant-1) plant-1) plant-1) plant-1)
Plant 13.77 22.68 0.97 -2.46 61.24 0.91
Stem 0.67 10.66 0.57 0.81 19.89 0.82
Leaf 10.0 14.76 0.86 3.69 25.42 0.80
Fruit 14.72 22.51 0.91 11.8 27.09 0.94
The model performance should be evaluated with more than one goodness-of-fit
measure. Therefore, we also included the RMSE as another measure that is commonly
used to check the agreement between measured and simulated results. The RMSE is a
measure of the average difference between simulated and observed values and is
expressed in the same units as the original variable. The highest RMSE was obtained
for the total dry matter per plant of the open field plants. The RMSE for the other plant
organs and for the results of the greenhouse plants yielded comparable RMSE values.
Looking only at the results for the organs, the simulated dry matter allocated to the
fruits had the lowest goodness-of-fit under both production systems.
The other goodness-of-fit criterion included to verify the model performance was the
EF, which is the most widely used distance measure, including upper and lower
bounds (Wallach, 2006). In this case, a model with an EF equal to one indicated a
perfect fit between observed and predicted values. For the present case, the crop
model obtained similar EF values when considering the simulated dry matter per plant
for both production systems. The lowest degree of agreement was observed for the
simulated stem dry matter allocated to the greenhouse plants. For both production
systems, the simulated fruit dry matter yielded a better fit than the one simulated for
the leaves.
Previous modeling efforts applied to Colombian greenhouse tomatoes, such as the one
carried out by Gil et al. (2017b), yielded an RMSE of 4.21 g DM plant-1 for the
simulated total plant dry matter. This potential crop growth model was calibrated
based on experimental crops planted on the Bogota Plateau and carried out under the
best possible management practices without any technical constraints.
The calibration of the present model yielded comparable results to those obtained for
other tomato model calibrations. For instance, Battista et al. (2015) calibrated a
modified version of the TOMGRO model for tomato growing in low-tech Italian
greenhouses. The plant dry matter calibration for three cultivars indicated RMSE
values ranging from 15.4 to 48.5 g plant-1 and EF values between 0.852 and 0.976.
The paper of Fan et al. (2015) described a knowledge- and data-driven modeling
71
approach for simulating the growth of a tomato plant. In that case, the RMSE for the
plant dry matter simulated with different modeling techniques ranged from 20.95 to
35.73 g plant-1. Our results are comparable to these results, except for the plant dry
matter estimated for the open field tomatoes.
It is important to highlight that the on-farm calibration experiments were carried out
with the required rigor from the data collection point of view but were developed
under the current set of management practices applied by most growers in the included
zones. Research at experimental stations often does not reflect the crop yield of
surrounding commercial farms (Leeuwis, 2013). Therefore, we decided to carry out
the model calibration through on-farm experiments but keeping the crop under
optimal conditions respect to nutritional status and pest management.
The yield gap represented the difference between the yield achieved by farmers and
the potential yield (Guilpart et al., 2017). Different yield gaps can be established
depending on the reference point used to evaluate a current yield obtained by local
growers (Tittonell and Gilller, 2013). The first gap was obtained by comparing
potential yields, with no restrictions other than those imposed by climate conditions,
and those currently obtained by local farmers. Potential yields can be calculated based
on models calibrated with data obtained from perfectly controlled conditions. This
gap is narrow in areas where production is characterized by high technological levels
and where factors such as soil fertility and pest and disease pressure do not impose
major restrictions on crop development. Another driving factor to explain model
performance is the variability introduced by genetic factors. Mavromatis et al. (2001)
stated that successful use of crop models in technology transfer requires that
coefficients describing new cultivars be available as soon as the cultivars are
marketed. On the other hand, current market trends, including specialization, have led
to genetic differentiation in contemporary tomato varieties (Sim et al., 2010). While
genetic variation is recognized, we overlooked this factor since the purpose of the
proposed model is to be as generic as possible. Future improvements in model
performance can be achieved by including genetic variation since temperature effects
on crop yield are also recognized as cultivar-dependent (Vanthoor et al., 2011).
5.4. Conclusions
The specificities of cropping systems, such as Colombian tomatoes, demand local
calibration of crop growth models under potential conditions. In contrast to our
original expectations, we found differences in the dry matter distribution to the plant
organs among greenhouse and open field tomatoes; therefore, it was necessary to
derive independent functions to characterize each tomato type. Despite including
these two sets of functions, the crop model was conceived as one entity able to
simulate plant behaviors for both types of tomato.
The tomato model proposed in this study is characterized by a fair compromise
between representativeness and accuracy. The on-farm calibration experiments
entailed a series of challenges and technical issues, commonly addressed in
commercial agriculture, reducing the potential yield achievable by the crop. The
challenge is to ensure an optimal supply of nutrients, so this implies a slight overdose
in fertilization, and a zero tolerance to the presence of pests and diseases. This implies
leaving aside concepts such as the economic threshold.
72
Chapter 6. Soil-plant model validation under Colombian
conditions
In preparation for submission to the journal: Agronomía Colombiana
Abstract
The purpose of the present work was to validate a soil-plant model for both tomato
production systems in Colombia, resulting from the integration of an adapted version
of the tomato model and a previously calibrated soil model. We used a soil model
developed for temperate conditions coupled to a tomato growth model calibrated and
validated locally. The coupling of the tomato model to the soil model also required
the incorporation of functions to represent the plant N-demand as well as the rooting
system growth, since these are dependent on the species. The water interception by
the leaves and runoff modules were omitted for greenhouse production given that the
crop under this system is protected from the direct effect of rain. Additionally, the
evapotranspiration module was changed to the one of Stanghellini, which is
recognized as more appropriate to estimate the evapotranspiration inside greenhouse
crops, especially for tomato. The last significant modification corresponded to change
the soil temperature module since the original model version incorporates an approach
able to simulate the soil temperature only under temperate seasonal climate variations.
The model, which is programmed in the R statistical language, uses as inputs climate
data (solar radiation, air temperature, humidity and wind speed), soil characteristics
(e.g. pH, nitrogen content, bulk density and moisture, lower limit, field capacity and
saturation), crop management practices (e.g. planting date, plant density, cycle
duration), as well as irrigation and fertilization strategies. The field data for the
validation came from experiments carried out in the main greenhouse and open field
tomato production areas of Colombia. The model was validated by comparing the
simulated values against independent observed ones. Three goodness-of-fit measures
were used to evaluate the model fit: Nash-Sutcliffe modeling efficiency (NSE), root-
mean-square error (RSR) and percent bias (PBias). The plant model, soil water
movement and soil temperature showed satisfactory goodness-of-fit. Although the
experimental data demonstrated that the simulated soil nitrogen content follows the
observed tendency, the NSE score showed the poor soil N-module performance.
Based on the above, extra future calibration and validation experiments are needed to
improve the accuracy of the proposed model.
Soil-plant model validation under Colombian conditions
6.1. Introduction
In agriculture, dynamic crop modeling has a wide range of applications, including the
assessment of potential pollutants such as nitrogen (N) emission released into the
environment. Among the most referenced models able to simulate yields and N-
emissions are the DNDC (Giltrap et al., 2010), DAYCENT (Del Grosso et al., 2005)
and CERES-EGC (Drouet et al., 2011). However, crop modeling applications in
horticultural crops are less referenced in the tropics than in temperate regions.
We used the calibrated soil model of Heuts (2018) coupled with a tomato crop model
that had been locally calibrated and validated for both greenhouse and open field
production systems (Gil et al., 2017c). The soil module allowed the simulation of soil
water content (SWC) based on water flow that was modeled as a ‘tipping bucket’
system, in which water moves from upper to deeper layers (Burns, 1974), carrying
ions, such as nitrates, dissolved in the solution. The N-dynamics in the soil were
simulated based on a submodel that integrated processes such as mineralization
(Bradbury et al., 1993; Coleman and Jenkinson, 1996; Jenkinson et al., 1987; Smith
et al., 2010), ammonium adsorption, urea hydrolysis (Jones, 1986), ammonia
volatilization (Génermont and Cellier, 1997; Shaffer et al., 2001), and (de)nitrification
(Bessou et al., 2010). Heuts (2018) carried out the soil model calibration and
validation by applying different N-treatments to a cauliflower-leek rotation
experiment. A list of all soil parameters is available in Heuts (2018). Figure 15 shows
the processes related to N-transformations that are included in the soil-plant model.
Figure 15. Schematic representation of the nitrogen processes considered in the soil
model. Adapted from Heuts (2018).
In the current model, we modified the soil temperature and evapotranspiration
modules. While Heuts (2018) used the temperature model proposed by Jury and
74
Horton (2004), which assumes a seasonal fluctuation throughout the year, we found
this unsuitable for tropical soils. In the present version, we incorporated the soil
temperature model proposed by Lei et al. (2011), which incorporates the asymmetric
oscillation of temperatures throughout the day, estimating the soil temperature under
tropical conditions better. The other change introduced for the greenhouse system was
the evapotranspiration model, changing from the Penman-Monteith FAO56 model
(Allen et al., 2006) to a model from Stanghellini (1987), which is adapted for low
wind speed conditions. The Stanghellini model has a better fit to estimate crop
evapotranspiration under greenhouse conditions (Prenger et al., 2002). The coupling
of the tomato model with the soil model also required the incorporation of functions
to represent plant N-demand as well as rooting system growth since these are
dependent on the species. The leaf water interception and runoff modules were
omitted for greenhouse production given that the crop under this system is protected
from the direct effect of rain.
Except for photosynthesis that occurs hourly, the model simulated the soil-plant
interactions on a daily basis and was driven by climate conditions, soil properties, and
crop management practices. The core modeled processes were crop growth and
development, soil temperature, water movement and N- and C-dynamics (Heuts,
2018). The hourly based climate inputs were solar radiation (SR, W m-2), air
temperature (AT, °C), relative humidity (RH, %) and wind speed (WS, m s-1). The
soil is divided into a number of horizontal layers characterized by their bulk density
(BD, g cm-3), pH, soil organic carbon content (SOC, %), cation exchange capacity
(CEC, cmolc kg-1), texture (clay, loam and sand contents, %), water content (SWC,
cm3 cm-3) at the lower limit (LL), field capacity (FC) and saturation (SAT) points.
Starting values on transplanting day for SWC and mineral nitrogen (Nmin, kg ha-1) are
also required per layer. For daily N-inputs, two main aspects must be defined: first,
the layer(s) to which the N enters. In the case of fertigation or broadcast applications,
N is added to the surface layer (kg N ha-1); however, for incorporated fertilizers, the
entering layers and their proportions should be defined. The second factor is the N-
source; for inorganic fertilizers, the model considers two sources: urea and
ammonium nitrate. For organic fertilizers and crop residues, the carbon content (%)
and the C:N ratio need to be specified. Finally, crop management includes information
about the planting date (Julian day), crop-cycle duration (days), plant density (plants
m-2), initial dry matter per organ (DM, g) and initial leaf area (LA, m2).
The aims of the present work were 1) to integrate an adapted version of the tomato
growth model into the soil model and 2) to validate the soil-plant model under
Colombian greenhouse and open field conditions, focusing on soil water movement
and N-dynamics throughout the crop cycle.

N-dilution curves
In this model, the amount of N demanded by the plant was calculated as a proportion
of the DM allocated to each organ. However, it is well-documented that the
concentration of this element decreases with plant development, a process commonly
referred to as N-dilution (Tei et al., 2002). A dilution curve determines the minimum
N-concentration required to ensure optimal plant development. N-dilution curves are
also called critical N-curves. The dilution curve was calculated using the nitrogen
75
concentrations determined for the aboveground plant organs experimentally

established by Gil et al. (2017c) for the plant model calibration and validation. The
critical nitrogen concentration was determined by fitting segmented linear models per
organ as a function of the accumulated thermal time (ATT).
Root system growth
The maximum rooting depth (",-.//+012+) ) and root-length density (RLD, cm cm-
3
) distribution are the main parameters controlling plant water and nitrogen uptake
(Pedersen et al., 2010). A required step to adapt the soil-plant model to local
conditions was to determine the maximum depth at which roots can be found
depending on the plant development stage as well as its distribution in the soil layers.
To determine this depth, in each production system, we developed a general
characterization of the tomato root system to determine its maximum depth and
distribution. We carried out destructive sampling of the root system during one tomato
cycle per system. Samplings started the second week after transplanting and were
carried out on a fortnightly basis until the end of harvest. On each sample date, we
randomly selected a plant and dug a pit 40 cm away from the stem. Then, we sliced
the soil profile at 5 cm intervals parallel to the bed-line and in the direction of the
stem. Throughout the growing cycle, the crop remained free of weeds to ensure that
the sampled roots corresponded only to those of the tomatoes. The root distribution
was determined through the cut-off points observed on pictures taken for each soil
profile. The soil profile pictures were referenced spatially on an XY plane with the
origin (0, 0) placed at the stem insertion in the soil. A detailed description of the
method used to study the root distribution is described by Vansteenkiste et al. (2014).
From these data, we selected the deepest 5% of points for each sampling and fitted a
logistic function (equation 30) to those points to estimate the maximum root depth as
a function of ATT.
4567
",-.//+012+) = (30)
1+ : ;(=>?@;ABB)⁄DEF#
where ",-.//+012+) is the maximum rooting depth (cm), 4567 is the asymptotic
depth (cm), -7HI represents the ATT at the inflection point, 4JJ is the accumulated
thermal time (°Cd), 5K,L indicates the thermal time that elapses before the roots reach
half the asymptotic depth.
On the other hand, as mentioned in Gil et al. 2017c, the plant model assigned 9% of
the DM generated daily to the roots. However, water- and N-uptake depends on the
distribution of the roots within a soil profile. The J!"N was distributed over the soil
profile with roots present based on a gamma probability distribution function fitted to
the experimental observations of root presence at different depths. To determine the
root lengths, twelve soil samples were taken at four depths: 10, 20, 30 and 40 cm with
a metal cylinder of known volume. Then, individual roots were isolated from the soil
by successive washing and sieving. The isolated roots from each sample were glued
to a sheet of paper, photographed and processed to determine the total root length at
each depth. Root lengths were then estimated following an adaptation of the procedure
described by Kimura et al. (1999). Next, the roots were dried and weighed. Finally,
similar to the calculation of leaf area based on the specific leaf area, the total root
length per layer was estimated based on the J!"N assigned per layer and multiplied
by the specific root length (SRL, cm/g).
76
Validation datasets
The datasets for validation were extracted from the N-trials carried out within
commercial greenhouse (954 m2) and open field (523 m2) plots. From these trials, a
subset of treatments was selected for validation, while the other treatments were used
to calibrate the plant model as described by Gil et al. (2017c). The experiments were
carried out under a completely randomized statistical design with three replicates per
treatment and each replicat being a plant bed-row. Table 15 summarizes the main
characteristics of the experiments, including the DM in each of the organs of the plant
at transplant.
Table 15. Summary of the main crop management parameters used as inputs for the
model.
Parameter Unit Open field Greenhouse
Latitude Decimal Degrees 6°28'55.40"N, 5°42'26.05"N,
Longitude Decimal Degrees 73° 6'54.74"W 73°36'3.97"W
Altitude masl 1140.0 2346.0
Planting date MM/DD/YYYY 01/27/2016 01/28/2016
Plant density Plants m-2 1.3 3.3
Cycle length Days 110.0 131.0
Initial dry matter
Leaves g plant-1 0.179 0.0915
Stems g plant-1 0.063 0.0325
Fruits g plant-1 0.0 0.0
2 -1
Leaf Area m plant 0.005175 0.004983
An automatic weather station (Vantage Pro2, Davis Instruments®) was installed next
to the plots to record the climate variables required by the model. Additionally, inside
the greenhouse, a pyranometer SR sensor (Model LI200RX, Campbell Scientific, Inc.,
Logan, UT, USA) was installed above the plant canopy, and air temperature and RH
were recorded by dry- and wet-bulb copper-constantan (T-type) thermocouples placed
inside a ventilated capsule. The wind speed inside the greenhouse was estimated by a
linear regression that required the external wind speed and the coefficients proposed
by Wang et al. (2000). Concerning the soil, for validation, we assumed a 40 cm
profile, which was divided into four layers of 10 cm each. For each soil layer, the
parameters required by the model were determined following the standard methods in
a soil laboratory. A summary of the soil properties used in model validation is given
in Table 16.
77
Table 16. Properties of the two soil profiles used in the validation process in
greenhouse (GH) and open field (OF) conditions.
Sys Ly LT pH OC BD Nmin CEC CL LO SA LL FC SAT SWC
GH 1 10 6.3 1.26 1.24 81.68 20.5 40.2 30.2 29.6 0.22 0.33 0.35 0.25
2 10 6.3 1.32 1.32 81.68 20.5 40.2 30.2 29.6 0.21 0.32 0.33 0.23
3 10 6.3 1.39 1.40 81.68 20.5 40.2 30.2 29.6 0.17 0.30 0.27 0.19
4 10 5.7 1.55 1.56 49.01 18.0 30.0 38.6 31.4 0.19 0.27 0.30 0.21
5 10 5.7 1.55 1.56 49.01 18.0 30.0 38.6 31.4 0.19 0.27 0.30 0.21
OF 1 10 6.5 1.08 1.46 21.2 11.3 17.1 42.3 40.6 0.15 0.36 0.40 0.14
2 10 6.5 1.08 1.49 21.2 11.3 17.1 42.3 40.6 0.15 0.28 0.32 0.21
3 10 6.5 1.08 1.45 21.2 11.3 17.1 42.3 40.6 0.15 0.28 0.32 0.20
4 10 5.2 1.08 1.36 14.8 11.3 45.2 26.5 28.3 0.12 0.28 0.32 0.19
5 10 5.2 1.08 1.36 14.8 11.3 45.2 26.5 28.3 0.12 0.28 0.32 0.19
Sys: Production system; Ly: Soil layer; LT: layer thickness (cm); OC: organic carbon content (%); BD:
bulk density (g cm-3); Nmin: mineral nitrogen content (kg ha-1); CEC: cation exchange capacity (cmolc kg-
1
); CL, LO and SA: clay, loam and sand content (%), respectively; LL, FC and SAT: lower limit, field
capacity and saturation (cm3 cm-3), respectively; SWC: initial water content (cm3 cm-3).
Nitrogen fertilization and water management

For the greenhouse trial, the average daily irrigation water was 3.1 l m-2, and the total
applied amount during the crop cycle was 416.8 l m-2. The selected fertilization
treatment included the addition of 10.89 t ha-1 of composted manure at transplanting
time, with a C:N ratio of 25 and a carbon content of 22.8%, on a DM basis. During
the production cycle, 399 kg N ha-1 was added through the fertigation system using
inorganic sources such as ammonium nitrate and calcium nitrate. N fertigation began
three weeks after transplanting with weekly additions of 19.2 kg N ha-1 week-1, and
after the tenth week, this value increased to 28.75 kg N ha-1 week-1 until the end of
harvest.
In the case of the open-field trial, the water input came from rainfall, supplemented
with sprinkler irrigation resulting in a total applied water amount of 777 l m-2.
Sprinkler irrigation was used during 41 days of the cropping cycle, with an average
daily discharge of 3.4 l m-2. The crop was fertilized at transplanting with 5.92 t ha-1 of
organic material with a C:N ratio of 13 and a carbon content of 22%. During the cycle,
239 kg N ha-1 was added by localized placement using ammonium nitrate, calcium
nitrate and diammonium phosphate. The N-inputs, both organic and inorganic, were
applied to the upper soil layer without incorporation. These N-doses were defined
based on the results of previous experiments for each of the production systems.
Plant destructive measurements were carried out throughout the crop cycle to
determine the DM allocated to the organs as well as the N-uptake and accumulation
per organ. On a fortnightly basis, six plants were sampled and divided into leaves,
stems and fruits (including flowers). The samples were labeled and oven-dried for 72
78
hours at 70°C. Once dried, the samples were weighed on a precision scale to determine
their DM. Then, a subsample was taken to the laboratory to determine the N-content
through the Kjeldahl method (Bradstreet, 1954). The LA was determined through
digital pictures of the individual leaves taken with a commercial grade digital camera
and placed over a reference gridded surface. The LA was determined based on the
relation between the number of pixels of a predefined area (reference grid) and the
number of pixels corresponding to the leaf surface.
We measured soil temperature (ST), SWC, Nmin, and ammonium (NH+4 ) plus nitrate
(NO-3 ) in the soil. For ST, a platinum-probe wire thermistor (range: -40°C to 65°C;
accuracy: ±0.5°C) programmed to record hourly measurements was employed. The
SWC was indirectly measured using a sensor that converted an electrical resistance
signal into a soil water suction in the range of 0 to 200 centibars (Watermark®,
Irrometer). The sensor measurements were converted to volumetric water content
using a previously fitted soil moisture retention curve (pF-curve) following the
method proposed by Richards (1941). This method relates the soil water retention at
a given negative pressure head to fit the pF curve. The following negative pressure
heads (bars) were evaluated: -0.1, -0.3 (pF 2.54: FC), -1, -3 and -15 (pF 4.2: LL).
Inside the experimental area, four pairs of ST and SWC sensors were installed at each
layer. The soil was sampled for Nmin on a fortnightly basis at a depth of 20 cm
throughout the crop cycle.
Model validation
The first step in validating the model was to perform a mass balance for water and N
in the system to verify the correct model operation. The water balance was carried out
following an adaptation of the hydrological balance equation described by Neitsch et
al. (2005), as shown below:
Q
%OPQ = %OPR + STU? + VW? − V? − WY? − %&F,? − UJF,? − !U? [ (31)

?\]
where %OPQ is the soil water content at the end of the cycle, %OPR is the soil water
content at the beginning of the cycle, ^ represents the crop cycle duration in days, U?
(cm) is the precipitation on the ith day, VW? is the irrigation (cm), V? is the water
intercepted by the foliage (cm) on the ith day, WY? is the surface runoff (cm) on the
ith day, %&F,? represents the actual soil evaporation (cm) on the ith day, UJF,? is the
actual plant transpiration (cm) on the ith day, and !U? represents the deep percolation
(cm) on the ith day. For the N-balance, we followed a similar approach, as shown in
the next equation:
Q
7'Q = 7'R + STH'_,? + `'_,? + "V? − 'a Y? − 'Y=,? − 'a,?

(32)
?\]
− b'cd,? − 'e1FE),? − W'f? [
where, 7'Q is the mineral N-soil content at the end of the cycle; 7'R is the mineral
N-soil content at beginning of the cycle; ^ is the crop cycle duration in days; H'_,? and
`'_,? represent the N-inputs for inorganic and organic fertilizers on the ith day,
respectively; "V? indicates the N-released by the mineralization process on the ith
day; W'f? is the plant root N-uptake (kg N ha-1) on the ith day; b'cd,? is the
79
ammonium volatilization on the ith day; 'Y=,? represents the nitrogen oxides emitted
to the atmosphere (kg N ha-1) on the ith day; 'a Y? is the nitrous oxide (kg N ha-1)
generated during the nitrification and denitrification processes on the ith day; 'a,?
represents the dinitrogen volatilization (kg N ha-1) on the ith day; and 'e1FE),? is the
potential N-losses due to leaching (kg N ha-1) on the ith day.
The model validation aimed at determining the goodness-of-fit of the model using
both qualitative and quantitative criteria, such as those in Heuts (2018). The
agreement between observed and measured values in the figures was used as
qualitative criteria. From the quantitative point of view, a set of goodness-of-fit
statistics was calculated comparing the model estimates with observed data. These
statistics are shown in Table 17.
Table 17. Goodness-of-fit statistics used to validate the model predictions.
Statistic Abbreviation Equation Range Reference
value for
perfect fit
Nash-Sutcliffe NSE ∑Q?\](%? − Y? )a [-¥,1] 1
modelling 1 −
∑Q?\](Y? − Yh)a
efficiency
Root mean RSR i∑Q?\](%? − Y? )a [0, ¥] 0
square error –
observed j∑Q?\](Y? − Yh)a
standard
deviation ratio
Percent bias PBias 100 × ∑Q?\](%? − Y? ) [-100, 100] 0
∑Q?\] Y?
deviation
ratiodeviation
% h
? : is each simulated value; Y? : is each observed value and Y: is the observations average.
ratio)
The variables involved in the model validation were the plant DM accumulation and
distribution, N total uptake and accumulation per organ, ST, SWC, and the soil Nmin
content within the first 20 cm. The model validation was carried out using the R
statistical software package (R Core Team, 2018).

Nitrogen dilution curves
Figure 16 shows the N-concentration change in thermal time as a function of the plant
organ. The critical N-concentration curve was included in the model using a tabulated
function. Therefore, the critical N-concentration for each organ and at each stage of
development was determined by means of linear interpolation. Thus, the model
estimated the daily nitrogen demand by multiplying the DM assigned to each organ
with the corresponding critical N-concentration. The data dispersion in the fruits and
leaves measurements does not allow to establish a difference in N-concentration
between plants planted in OF and under GH conditions. In the case of stems this
difference among plants cropped in each of the systems was evident. These
differences are probably related with the plant growth habits, while under GH system
varieties with an indeterminate growth habit are planted and in OF determinate ones.
80
However, in this model approach we decided to fit a single curve for both systems in
order to simplify the model. This choice causes an overestimation of nitrogen demand
by the stems in plants under GH, and an underestimation for this variable in the case
of OF production.
Figure 16. Nitrogen content in the tomato plant organs as a function of the
accumulated thermal time.
Rooting system growth
According to the adjusted parameters for the logistic functions, for the greenhouse
experiment, the asymptotic depth for the maximum rooting depth was 49.9 cm, while
for the open field experiment, it was 54.4 cm (Figure 17). The inflection point
occurred earlier in the open field experiment than in the greenhouse experiment,
namely, at 155.8°Cd. For the greenhouse experiment, it occurred at 293.1°Cd. This
variation in root growth was mainly due to the different soil types. The slower and
more restricted root growth in the greenhouse experiment was due to the soil clay
texture, while in the open field experiment, the sandy texture of the upper layers
allowed faster development of the roots as well as expansion to greater depths.
81
Figure 17. Maximum rooting depth of tomato plants as a function of accumulated

thermal time determined for greenhouse and open field conditions. The points
represent the observations. The line corresponds to the fitted logistic function.
Figure 18 shows the J!"N distribution as a function of depth modeled by the gamma
distribution function whose parameters change dynamically depending on the plant
developmental stage. As a plant grows, its roots reach a greater depth, and at the same
time, the highest concentration of roots moves to deeper areas. Thus, the RLD was
obtained by summing the root dry matter per soil layer multiplied by a constant SRL
and divided by the layer volume. The RLD evolution throughout the cycle determined
the dynamic interactions between plant and soil in terms of water and N-uptake.
Figure 18. Probability distribution to control the root dry matter allocated to each layer
with root presence as a function of the accumulated thermal time.
Water and nitrogen overall balances
For the open field treatment, the final water balance (right-hand side of Equation 31)
showed a difference of 0.09 cm with respect to SWCn, while for the N-mass balance,
this difference was only 6.52 kg ha-1, as indicated by the balance components
82
presented in Table 18. For the greenhouse treatment, the differences in the final
balances were 0.06 cm for water and 9.53 kg N ha-1 for the N-mass balance. These
negligible differences in the water and N-balances, mostly due to rounding
differences, confirmed appropriate implementation of the soil water and N-modules.
Table 18. Final water and mass balance predictions to confirm the appropriate
implementation of the water and nitrogen soil modules.
Abbreviation Open field Greenhouse
Water-balance (cm)
Final soil water content %OPQ +15.21 +14.25
Initial soil water content %OPR +11.25 +10.96
Precipitation U +63.60 +0.00
Irrigation VW +14.10 +41.68
Interception by foliage V -3.04 -0.00
Superficial runoff WY -3.40 -0.02
Soil evaporation %&F -17.13 -10.97
Plant transpiration UJF -10.81 -26.97
Deep percolation !U -39.45 -0.50
Nitrogen mass-balance (kg ha-1)

Final mineral nitrogen in the soil 7'Q +261.68 +932.65
Initial mineral nitrogen in the soil 7'R +134.13 +561.00
Inputs of inorganic fertilizers 5'_ +239.00 +399.00
Inputs of organic fertilizers `'_ +70.1 +87.0
Nitrogen from mineralization "V +242.60 +231.01
Nitrous oxide air emissions 'a Y -2.84 -0.95
Nitrogen oxides air emissions 'Y= -0.52 -0.76
Dinitrogen released to air 'a -6.13 -0.19
Ammonium volatilization b'cd -0.97 -1.50
Nitrogen leaching 'e1FE) -216.04 -26.26
Plant nitrogen uptake W'f -196.81 -308.42
Plant dry matter and nitrogen

The goodness-of-fit for DM production, N-uptake and its allocation in the plant
organs are shown in the Figure 19. Although the crop model validation was already
published (Gil et al., 2017c), the present results include its coupling with the soil
model, meaning that N- and water-supplies affect crop growth and development. For
the greenhouse experiment, the crop model achieved a good representation of the total
DM production and of the DM allocated to the fruits. However, the model consistently
overestimated the DM allocated to stems and leaves. Regarding plant N-uptake, a
similar trend was observed, with the model adequately representing the total N-uptake
and the N-uptake for the fruits but overestimating the predicted N-uptake for leaves
and stems. In the open field trial, at the end of the cycle, the model underestimated
83
total plant DM accumulation. However, considering the results per organ, only the
DM allocated to fruits showed a similar trend, while the leaf DM was overestimated.
Overestimations of the total N-uptake and N-uptake for leaves and stems were
observed, while the model yielded a slight N-uptake underestimation for fruits.
Figure 19. Observed and simulated dry matter accumulation and partitioning and
nitrogen content of the plant organs for the validation experiments carried out under
greenhouse and open field conditions.
According to the goodness-of-fit measurements, the model showed a satisfactory
performance (NSE ≥ 0.92, RSR ≤ 0.28 and |UyH,5| < 22) for both total DM and
total N-uptake (Table 19). However, we found differences in the model performance
by examining the output for the plant organs. Of the organs, fruits showed the best fit
for both DM accumulation (NSE ≥ 0.95, RSR ≤ 0.22 and |UyH,5| < 9) and N-
allocation (NSE ≥ 0.92, RSR ≤ 0.28 and |UyH,5| < 22). For leaves, the model
overestimated both DM and N-accumulation; however, a higher overestimation
occurred for the stem DM than the other types of DM (Table 19).
Under the open field conditions, the DM was underestimated except for the leaf DM,
which showed the highest deviation from the observed values (Table 19). The stems
showed the least agreement, especially under the greenhouse condition for both DM
accumulation and N-allocation (NSE < −5, RSR > 2.5 and UyH,5 < 110). Although
under the open field condition, the stem DM showed a good adjustment, this did not
84
occur for N-allocation. Table 19 also includes the average DM and N-accumulation
per plant (-̅~ ), in addition to the model predicted values (6).
Table 19. Model performance scores for dry matter accumulation and nitrogen content
for the whole plant and disaggregated per organ.
Greenhouse Open field
Variable -̅~ 6 NSE RSR PBias -̅~ 6 NSE RSR PBias
Dry Matter (g
DM plant-1)
Plant 221.62 249.69 0.96 0.2 13.21 635.40 520.53 0.94 0.24 -10.01
Fruits 158.08 179.79 0.97 0.17 -1.96 406.16 325.00 0.95 0.22 -8.2
Leaves 51.92 92.46 -0.71 1.31 77.44 116.10 194.64 0.7 0.55 27.01
Stem 11.62 42.16 -5.35 2.52 110.99 113.14 128.66 0.97 0.17 -0.65
Nitrogen content
(g N plant-1)
Plant 9.86 8.89 0.97 0.16 -0.83 15.35 15.68 0.92 0.28 21.2
Fruits 6.56 5.63 0.95 0.22 -16.48 11.16 9.58 0.94 0.25 10.23
Leaves 3.05 2.36 0.79 0.46 8.13 2.83 3.81 0.74 0.51 26.95
Stem 0.28 0.9 -7.18 2.86 110.19 1.36 2.28 0.33 0.82 51.56
-̅~ : experimental measurements average, 6: model predictions.
The discrepancy between the experimental measurements and model outputs,

particularly for leaves and stems, could be strongly influenced by the considerable
presence of pests and diseases during the validation experiments. The greenhouse
experiment was affected by whitefly and consequently by sooty mold, despite the
applied control measures to keep the pest population at minimum.
Soil water content
The simulated and observed SWC for both trials are presented in Figure 20. Under
both conditions, the observed 0-10 cm SWC was characterized by high variation.
Under the open field condition, as a consequence of the combined effect of
precipitation and irrigation, a low SWC was observed for all soil layers at
transplanting, following a rapid increase until 8 days after transplanting (DAT).
However, for most of the cropping cycle, all soil layers continued to have high SWC
values, except for the period between 72 and 88 DAT, when a decrease was observed
in the field and predicted by the model. In the greenhouse experiment, one day before
transplanting, irrigation was applied to the soil, increasing the moisture of all layers,
but the model was not able to simulate this situation. Nevertheless, after transplanting,
the observed SWC decreased sharply, while the model predicted a flat constant trend
for all layers.
Under greenhouse conditions, the soil moisture behavior of the two upper layers was
different than that of the deepest layers. While the first two layers had higher moisture
content variations throughout the cycle, the two deeper layers maintained low
85
constant SWC values during a large part of the cycle, followed by a sudden increase
at the end (Figure 20). This situation was well represented by the model. The results
observed for this soil were probably due to its clay texture, historical use and tillage
management. This area was converted from livestock grazing to greenhouse
horticultural production approximately 5 years ago, and the regular soil preparation
consisted of plowing at no more than 20 cm. The tillage decreased the compaction of
the surface layer, and according to Hamza and Anderson (2005), the low observed
and simulated infiltration rates of the deeper layers indicated that these layers were
still compacted.
Additionally, the placement of the sensors with respect to the drip lines contributed
to the observed differences between the measured and predicted SWC. The sensors
placed at depths of 10-20 and 20-30 cm were closer to the drip lines, while the
remaining two sensors were further away from them. This scenario probably
introduced a slight bias in the soil moisture records with respect to the model
predictions given that the soil model followed a one-dimensional approach that
assumed moisture contents were homogeneous within each layer.
Figure 20 also includes the goodness-of-fit statistics used to evaluate model
performance. The model showed a better fit for the layers between 10 and 40 cm depth
under the open field condition compared to the greenhouse situation. However, in
comparison with the other layers, the first layer had a greater underestimation of the
model; nevertheless, the model satisfactorily represented the water content in the soil.
Based on the PBias, we established that the model overestimated the SWC in the first
layer (0-10 cm), while in deeper layers, the trend was to slightly underestimate the
SWC. For the greenhouse trial, the results were different; the predicted SWC with the
lowest agreement with the observations occurred for the deepest layers, exhibiting an
overestimation trend for most of the cropping cycle. In contrast, the predicted SWC
of the second layer showed the best agreement with the observations. The 0-10 and
20-30 cm layers showed good agreement with observations, although the SWC for
both layers were overestimated.
86
Figure 20. Observed and simulated water content of the four soil layers for the open
field and greenhouse trials.
Soil nitrogen
The model capability to predict the soil Nmin-content was evaluated by comparing
the predicted NH+4 and NO-3 in the first 20 cm against the records obtained from the
soil samplings. Figure 21 shows these comparisons in addition to the goodness-of-fit
statistics used to evaluate model agreement. In both the open field and greenhouse
trials, the PBias showed that the model overestimates the soil concentrations of NH+4
and NO-3 . Although the predicted values follow the trend of the experimental
measurements, the NSE and RSR score results showed the limitations of the model
performance (Figure 21). Nevertheless, both NH+4 and NO-3 showed a wide range of
variation for most sampling dates. Although all samples were taken before
fertilization and in a random pattern from the planting bed, it is important to note that
soil N-concentration is strongly influenced by the application method (Zhang et al.,
87
1996), especially in the upper layers. Without a sampling pattern based on the
irrigation lines as in the greenhouse trial, some subsamples were probably collected
closer to the drip than other samples, resulting in high N-concentrations, while other
samples probably were collected farther away with a lower N-concentration. As a
one-dimensional model, the soil model was not designed to account for any horizontal
spatial variation in the predicted variables.
Figure 21. Simulated (lines) and measured (± standard deviations) nitrate and
ammonium contents at a 20 cm depth throughout the experimental cycle. The modeled
nitrogen contents were obtained by adding the predicted values for the first two layers
(0-10 and 10-20).
Under open field conditions, of the factors, ammonium had the highest variation in
nitrogen content, while in the greenhouse system, nitrate measurements had the
highest variation in nitrogen content. In contrast to SWC, soil nitrogen dynamics are
governed by complex interactions of multiple processes, so their calibration and
validation are still a challenge (Heuts, 2018), especially for highly variable production
systems developed on tropical soils.
Soil temperature
The model closely reproduced the observed trend for soil temperature (Figure 22).
Based on the goodness-of-fit measures, the model performance for temperature was
88
good enough for all depths and under the two experimental conditions. According to
the measurements, the soil temperature variations were higher in the upper layers than
in the other layers. These variations were much higher under the open field conditions
than under the greenhouse conditions due to the use of plastic mulch in the greenhouse
and periodic precipitation. Regarding the time trend, the largest differences between
the measured and predicted soil temperatures were observed after the midpoint of the
cropping cycle and for the upper layers. Some authors (Heuts, 2018; Liu et al., 2013;
Sándor and Fodor, 2012) have associated this trend with the fact that these models do
not incorporate the shading effect generated by canopies, which remains a point to be
addressed in the model.
Figure 22. Observed and simulated soil temperature at different depths and the
goodness-of-fit statistics used to evaluate the model performance.
89
6.4. Conclusions
The first step in this work was to adapt the plant model to respond to the N-demand
required to achieve optimum growth and development and to model root growth.
Given that roots penetrate soil layers, they determine the availability of N and water
to meet plant demands. When N and water in the rooting layers fulfill plant demands,
growth and development will be optimal; otherwise, they will be constrained.
In this section, we assessed the soil-plant model performance for DM production and
partitioning, N-allocation to plant organs, soil water content, soil temperature and
mineral soil N-content for the first 20 cm in depth. Although the results for the
performance of those variables varied depending on the production conditions, plant
organ or soil layer, in general, the soil-plant model was able to adequately reproduce
the N and water soil dynamics and their effects on tomato plant growth and
development.
Given that the variables involved in the present validation showed that the model
performed satisfactorily, it is reasonable to assume that the remaining predicted
variables, such as those directly related to nitrogen emissions, should perform equally
well. For this reason, we consider the performance of the soil-plant model to be
satisfactory and suitable for environmental assessments of current and prospective N-
fertilization scenarios.
90
Chapter 7. Evaluation of nitrogen fertilization scenarios
In preparation for submission to the journal: Journal of Cleaner Production
Abstract
Crop modelling and life cycle assessment (LCA) are sophisticated tools widely used
in the agricultural sector. However, combining modelling and LCA for the evaluation
of the environmental performance of prospective scenarios is still in an early stage.
The purpose of this work was to combine crop modeling with LCA but incorporating
the stochastic multi-attribute analysis (SMAA) method into the normalization and
weighting steps allowing an objective environmental assessment of alternative
production strategies. As a proof of concept, we assessed the environmental
performance of the current tomato production systems in Colombia and prospective
scenarios related to nitrogen (N) fertilization. The primary data were extracted
through a locally validated soil-plant model, which was used to simulate the effect of
weather seasonality on tomato production both for open field and greenhouse systems
under three N-fertilization scenarios. Secondary data related to the fertilizer
manufacturing were extracted from the Ecoinvent database (v.2.0). The following
three scenarios will be simulated:
1. The current N-fertilization practices carried out by local growers.
2. Applying a fixed dose of nitrogen.
3. N-fertilization depending on the simulated plant N-demand to achieve
optimal development.
The results showed significant effects of seasonal climatic variability on crop yield
as well as on the environmental impact. This work demonstrated the feasibility of
coupling soil-crop models with life cycle assessment to dynamically determine the
environmental impacts. Both the results of the evaluated scenarios and the feasibility
of the integrated methodology are presented and discussed.
Evaluation of nitrogen fertilization scenarios
7.1. Introduction
The knowledge accumulated in recent decades in areas such as computer science,
applied mathematics, plant physiology, crop science, edaphology, and genetics has
boosted the development of crop modeling (Jones et al., 2017). As a result, crop
modeling is now a mature tool used in a wide range of applications ranging from crop
forecasting (Soler et al., 2007) to climate change effects on food production
(Rosenzweig et al., 2014), although this type of modeling is mostly restricted to
temperate zones. In this work, crop modeling was used to predict plant growth and
development in relation to climate, environmental factors and management practices;
more specifically, the constraints imposed by water and nutrient availability
throughout the crop cycle were investigated.
Currently, a wide range of these models is available, including ORYZA (Bouman and
van Laar, 2006), CERES-N (Quemada and Cabrera, 1995), WOFOST (Van Diepen
et al., 1989), SOYGRO (Wilkerson et al., 1983), and SUCROS2 (Van Keulen et al.,
1992). These models have been developed for specific purposes and are used to
simulate scenarios that are rarely under actual conditions due to their complexity or
high costs in terms of budget or time.
To simulate water and nutrient dynamics, soil processes must be linked to crop
processes. These processes drive the movement of water, the transformation of
organic and inorganic compounds and plant uptake of both water and nutrients
through the rooting system. Within this modeling framework, it is possible to
introduce algorithms representing the potential losses of polluting substances. Models
such as DNDC (Giltrap et al., 2010), CERES-ECG (Dufossé et al., 2013), STICS
(Brisson et al., 1998), EU-Rotate_N (Rahn et al., 2010) and the one proposed by
Heuts (2018) have the ability to simulate crop yields in addition to water dynamics in
the soil-plant system, including greenhouse gas emissions as well as ammonia and
nitrate emissions.
Special attention has been given to nitrogen (N). N is a key element to ensuring proper
yields and is consequently added through external sources; however, the application
of high amounts of N to the soil results in risk to the environment. The most important
N-pollutant molecules emitted from agricultural fields are ammonia (NH3) and nitrous
oxide (N2O), which are released into the air, and nitrate (NO-3 ), which is released into
water bodies (Brentrup et al., 2000). The processes to model soil N-dynamics, and
therefore the derived pollutant losses, include the flow of water out of a system
through leaching and runoff (if open field conditions) and the main N-processes, such
as decomposition/mineralization, ammonium adsorption, urea hydrolysis (if used),
ammonia volatilization, and (de)nitrification.
On the other hand, life cycle assessments (LCAs) of agricultural systems often
estimate on-field N-emissions within the life cycle inventory (LCI) step (Bojacá et
al., 2014; Brentrup et al., 2001). In most related literature, the methods used to
calculate these on-field emissions are static and result in aggregated scores per
production cycle using meta-regression analysis, with N-fertilization as the main
driving factor (Brentrup et al., 2000; Nemecek et al., 2016). These methods include
empirical correction factors to modify the amounts of N-released to the environment
depending on characteristics such as the type of fertilizer, edapho-climatic conditions
and, in some cases, crop characteristics and management practices (Nemecek et al.,
92
2016). As a consequence, these empirical models are easily calculated but have a high
uncertainty due to their generic parametrization.
This work contributes to a short list of studies that have recently been published
regarding the use of dynamic soil-crop models to calculate the emissions in a life cycle
assessment context (Dufossé et al., 2013; Goglio et al., 2018; Heuts, 2018; Liao et al.,
2014; Nitschelm et al., 2018; Perrin et al., 2017). These studies demonstrate the
advantages of crop modeling, such as the dynamic incorporation of local climatic and
soil conditions as well as fertilization and irrigation practices to more realistically
simulate the interactions in a soil-plant system. Goglio et al. (2018) reported good
agreement between the emissions calculated by the DNDC model and those measured
in the field by micrometeorological techniques, thereby demonstrating the validity of
the approach.
Although LCIAs result in a set of environmental indicators, integration of this
information into an overall index is still problematic. Thus, several multidimensional
methods have been applied for LCA normalization and weighting, with optional steps
to improve communication of the results to stakeholders (Zanghelini et al., 2018). The
present work proposes a method for agricultural LCAs by coupling soil-crop model
tools and LCA by integrating the stochastic multi-attribute analysis (SMAA) method
to generate a single dimensionless indicator that allows an unbiased environmental
evaluation of a system or production alternatives. This method was proposed by Prado
and Heijungs (2018); however, in its implementation, they assumed a single
symmetrical distribution for the hypothetical environmental indicators. The first step
in the SMAA method is to carry out Monte Carlo simulations of the environmental
indicators to build the joint probability distributions that later will be used to outrank
the alternatives. To take into account the correlation structure of the impact categories
caused by their linear relationship with primary data (Bojacá and Schrevens, 2010),
the copula method was applied to derive multivariate joint distributions (Czado et al.,
2013; Sklar, 1959) that represent the existing level of correlation among impact
categories. As a proof of concept, we assessed the environmental performance of the
current tomato production systems in Colombia and prospective scenarios related to
N-fertilization.

Goal and scope definition
The present study is a comparative LCA for N-fertilization scenarios in which a
locally calibrated soil-plant model is used for the life cycle inventory to predict the
N-emissions as well as the crop yield. The ISO 14040 guidelines (ISO, 2006a, 2006b)
were followed for this environmental analysis.
We used smallholder tomato production in Colombia as a study case, which was
carried out under open field and greenhouse conditions and has been already described
by Gil et al. (2019b). The impacts due to fertilizer manufacturing plus those from on-
field emissions during a production cycle were considered with a functional unit of
one ton of tomato, with the system boundaries ranging from raw material extraction
to the farm gate (cradle to gate).
93
Life cycle inventory analysis

A soil-crop modeling approach was used to predict the N-emissions as well as the
crop yield. Generally, soil-crop modeling is categorized according to the simulated
processes and factors that constrain crop growth and production. In the first category,
models are able to represent processes such as photosynthesis, respiration, and dry
matter (DM) production and distribution, given a specific set of weather conditions.
In the next category, soil and plant water balances are included in the model to
represent the effect of water supply constraints on crop development. The model used
in this work falls into the third category where the plant growth is conditioned by the
N and soil water supply with respect to plant demands (de Vries, 1980). This type of
model incorporates the main processes of N-cycling in the soil coupled with plant
growth, soil moisture and temperature gradients. Primary data, such as on-field N-
emissions, were obtained from the soil-crop model, while secondary data related to
the fertilizer manufacturing processes were extracted from the Ecoinvent v.2.0
database (Frischknecht et al., 2007).
Soil-crop model background and description
In this work, we used the soil model calibrated by Heuts (2018) coupled with the
tomato model locally calibrated by Gil et al. (2017) for both open field and greenhouse
systems under Colombian conditions. The water movement in the soil occurs based
on the tipping bucket approach proposed by Burns (1974), where the water flows to
deeper layers based on gravitational force, while upward movement is mainly driven
by diffusivity, capillarity and soil evaporation. The water movement between layer
transports ions, such as nitrates, dissolved in the soil solution. The modeled N-soil
processes were mineralization/immobilization (Bradbury et al., 1993; Coleman and
Jenkinson, 1996; Jenkinson et al., 1987; Smith et al., 2010), urea hydrolysis (Jones,
1986), ammonium adsorption/desorption (Li et al., 2006), ammonia volatilization
(Génermont and Cellier, 1997; Shaffer et al., 2001), and (de)nitrification (Bessou et
al., 2010).
The plant model simulated tomato growth and development. Hourly weather
conditions and canopy leaf area determined photosynthetic rates as suggested by
Acock et al. (1978). Daily maintenance respiration was discounted from gross
photosynthesis resulting in net photosynthesis. Net photosynthesis was transformed
to total biomass DM considering growth respiration. Then, the total DM was
partitioned over the plant organs as a function of the development stage. The tomato
model used accumulated thermal time (ATT) to define the phenological development
stages (Gil et al., 2017c; Marcelis et al., 2009). Plant water demand was calculated
based on evapotranspiration, while the N-demand was calculated by multiplying the
critical N-concentration (read from the N-dilution curves) at each development stage
with the daily produced DM. The soil characteristics necessary to simulate water and
N-uptake by plant roots are rooting depth and root length density per soil layer. These
were calculated by the model as a function of the root DM per layer and development
stage. The sink-source relation between N or water supply and the DM available for
growth decreased as did the crop yield. This soil-plant model assumed no incidence
of pests or diseases on crop development and yield. All soil and plant processes were
simulated on a daily basis, except for the photosynthesis that was calculated hourly.
94
N-fertilization scenarios
The soil-plant model was used to evaluate the following three N-fertilization scenarios
for the two production systems:
1. Current N-fertilization practices carried out by local growers (Cr).
2. Application of a fixed dose of nitrogen (Fr).
3. N-fertilization based on the simulated plant N-demand to achieve
optimal development (Op).
Planting densities of 1.1 and 3.0 plants m-2 and cycle lengths of 114 and 166 days
were used for the open field and greenhouse system simulations, respectively. The
phenological and physiological parameters driving the tomato model are presented in
Gil et al. (2017). In terms of the N-fertilization source, we assumed the application of
N (ammonium nitrate), P (phosphorus pentoxide) K (potassium oxide) was 13:6:40
and 15:4:23 for the greenhouse and open field systems, respectively, consistent with
the most used fertilizers by growers as reported by Gil et al. (2019b). The specific N-
fertilization scenarios in terms of timing and dosing are described in the next
paragraph.
Scenario 1 – Current N-fertilization strategy (Cr)
In the first scenario, the N-fertilization strategy currently applied by growers in each
production system was simulated. This scenario was based on the field data collected
by Gil et al. (2019b).
For the greenhouse system, a total of 272 kg N ha-1 was applied per production cycle
with 9.6% of that being applied on the transplanting day, while the remaining amount
was added proportionally on a daily basis until the end of harvest. This daily dose
corresponded to 1.5 kg N ha-1 (0.55% of the total dose) and was applied through
fertigation.
For the open field system, a total dose of 188.6 kg of N ha-1 was applied, distributed
over the crop cycle in eight applications (2, 21, 29, 36, 41, 49, 56, and 65 days after
transplanting) at varying amounts (6.6, 16, 26.4, 28.3, 30.2, 22.6, 22.6, and 35.9 kg N
ha-1, respectively). This is a generalization, since the detailed follow-ups revealed that
producers apply fertilizers with time intervals ranging from 4 days at the beginning
of the cycle to 20 days at the end of the crop.
The irrigation water for the greenhouse system was applied through the fertigation
system at a daily rate of 2.52 l m-2 from the transplanting day until the end of the
cycle, following the standard practice of Colombian growers (Gil et al., 2019b). Open
field production is mainly a rainfed system with occasionally extra irrigation input.
Scenario 2 – Fixed N-fertilization strategy (Fx)
For this scenario, both production systems received a fixed dose of 400 kg N ha-1.
This dose was derived from N-dose-response experiments carried out for open field
and greenhouse tomatoes planted under Colombian conditions (Gil et al., 2017b). This
dose produced the maximum yields for both production systems. The timing and
application schemes of this scenario were the same as those designed for the Cr
scenario. Irrigation strategies for both systems also remained the same for this
scenario.
95
Scenario 3 – Optimum N-fertilization strategy (Op)

The third scenario calculated the potential daily N- and water-demands of the plants,
assuming unrestricted N- and water-availability. This scenario was generated by
running the model twice. In the first run, only the soil-crop model was active, and the
total N in a plant was the sum of the allocated DM to the organs multiplied by their
respective critical N-concentrations. This first run predicted the N-uptake by the
plants without considering the soil processes. In the second run, the soil processes
were included, allowing estimation of the emissions, as was conducted for the first
two scenarios. The N-uptake of the first run was adjusted by applying an efficiency
factor to determine the N-doses for both production systems. This efficiency factor
was calculated from local experimental data as the proportion between the N-uptake
by the plant and the N-applied on the field (Castillo-Hernández, 2008). This efficiency
factor ranged between 0.43 and 0.58, with an average value of 0.56, which was used
to determine the N-doses for the optimum scenario in this study. For the greenhouse
system, the amount of water applied in the second model run was derived from the
crop evapotranspiration results of the first run. For the open field system, rainfall
remained as the sole source of water for the crop.
Climate variability
Since greenhouse and open field tomatoes can be grown year-round in many regions
of Colombia, we accounted for the climate variability throughout the year by running
different simulations with planting dates every week of the year and historical climate
data, resulting in 312 simulations for both production systems. The simulations were
executed using historical weather datasets for each production system. The historical
weather data of Alto Ricaurte (5°39’ N – 73°34’ W, 2070 masl) and Guanentá (6°25’
N – 73°10’ W, 1370 masl) provinces were used for the simulations of the greenhouse
and open field production systems, respectively. These two provinces are among the
main tomato production regions in Colombia for both production systems.
A detailed description of these daily historical series as well as a monthly summary
of the main variables can be found in Gil et al. (2019a) and also in section 2 of this
thesis. The soil-plant model required hourly records of solar radiation (W m-2), air
temperature (°C), relative humidity (%), wind speed (m s-1) and daily precipitation
(cm). Since the available historical datasets were on a daily basis, we applied the
methods proposed by Teh (2006) to reconstruct the required hourly time series. These
methods use sine and cosine functions to represent the cyclic trend of temperature,
relative humidity, wind speed, and radiation during daytime hours. At night,
regression models were used for temperature, while zero values were set for radiation
and for the minimum daily wind speed.
Since the representative year could be calculated for most of the weather variables by
averaging the records over several years, this could not be done for precipitation.
Thus, for precipitation, we followed the approach proposed by Richardson and Wright
(1984), where a first-order Markov chain was used to generate the occurrence of
rainfall on a given day conditioned by the rainfall status of the previous day. When a
rainfall day occurred, the two-parameter gamma distribution was used to generate the
precipitation amount. In this work, maximum likelihood estimators were employed to
estimate the gamma distribution parameters for each month based on the historical
data available (Husak et al., 2007). Precipitation was used for the open field system
96
as the only water input, while for the greenhouse system, this variable was not
included.
For the greenhouse production system, the model calibrated and validated by Díaz-
Guevara (2015) for Colombian conditions was used to estimate the internal
greenhouse climate based on external weather conditions. The model implemented by
Díaz-Guevara (2015) was a set of nonlinear differential equations designed to predict
the air, soil, crop and cover surface temperatures as well as sensible and latent heat
fluxes as a function of radiation. Vapor pressure and CO2 concentration were also
considered in this approach. The model inputs were the external global radiation, wind
speed, relative humidity, air temperature, soil surface layer temperature, and leaf area
index (LAI). The LAI required a prediction of the vapor pressure inside the
greenhouse, which was obtained from the crop growth simulations.
Soil configuration
For all simulations, the soil profile was divided into four layers from 0 to 0.9 m (0-
10, 10-30, 30-60 and 60-90 cm), as suggested by Heuts (2018). An N-flow below 0.9
m was considered irreversible leaching to the aquifer. The initial values for the soil
N-content, pH, organic carbon, and soil texture were averaged from local soil data
from the aforementioned Colombian provinces as reported by Gil et al. (2019b). The
water content at the wilting point (-15 bar), field capacity (-0.3 bar) and saturation (0
bar) were calculated using pedo-transfer functions that require soil texture and organic
matter of each layer as inputs (Saxton et al., 1986). The initial soil water content was
assumed to be 90% of the field capacity for each layer, based on the standard
management practice of irrigating soil before transplanting. The soil parameters and
initial values used in the simulations are presented in Appendix 3.
Nitrogen fertilization emissions
The soil-plant model predicted the on-field emissions of NH3, N2O and NO-3 following
a process-based mechanistic approach as proposed in Heuts (2018). Hence, the NH3
emissions were generated as a result of a first-order kinetic approach for the
ammonium (NH+4 ) pool, which used mineral fertilization and the NH+4 flow from that
mineralization as inputs. In the present work, as in Heuts (2018), the pH of the soil
was assumed to have a relatively low effect on the volatilization of NH3, and this
process was also assumed to occur in the topmost soil layers (0-15 cm). An NH+4 pool
also feeds the nitrification process. N2O emissions are generated by both oxidation in
aerobic conditions of NH+4 into nitrite (NO2) and then from this compound to NO-3 as
well as by the reduction of NO-3 in anaerobic conditions. For both nitrification and
denitrification, the Michaelis-Menten approach was used to represent the enzymatic
kinetics of each process. The NO-3 from fertilizers and the resulting nitrification
process can potentially be lost by leaching as a result of the movement of water from
the superficial soil layers to the deepest layers due to gravity. Although the model also
estimated the losses of nitrogen in the form of NH+4 , these did not exceed 2% of the
total content of this element in the soil layers under consideration. These processes
are dynamically driven by biophysical factors such as fertilization, soil temperature,
water content and plant N-uptake. At the end of each simulation, the emissions were
aggregated from the daily results to represent the total emissions per production cycle.
97
Impact assessment method and interpretation

We applied the CML2001 v.2.05 impact assessment method (Guinée, 2002),
including the following midpoint impact categories: abiotic depletion (AD; kg Sb eq.),
acidification potential (AP; kg SO2 eq.), eutrophication potential (EP; kg UYÄd; eq.),
freshwater aquatic ecotoxicity potential (FAETP; kg 1,4-DB eq.), global warming
potential (GWP; kg CO2 eq.), human toxicity potential (HTTP; kg 1,4-DB eq.),
marine aquatic ecotoxicity potential (MAETP; kg 1,4-DB eq.), photochemical
oxidation potential (POP; kg C2H4 eq.), and terrestrial ecotoxicity potential (TEP; kg
1,4-DB eq.). We allocated the NH3 emissions to AP and NH3, N2O and NO-3 emissions
to EP, N2O emissions to GWP and NH3 emissions to HTTP (Brentrup et al., 2001).
To compare the environmental performance of the prospective scenarios, we carried
out the optional normalization and weighting LCA steps. Given the within variability
of the scenarios caused by including climate variability in the simulations, we
performed the normalization and weighting steps following a stochastic ranking
method, resulting in an unbiased overall environmental impact index.
We applied the SMAA method as proposed by Prado and Heijungs (2018a) to
compare the environmental performance of the considered scenarios. The outranking
procedure of the SMAA method generated an overall ranking of the scenarios while
considering the uncertainty of the environmental performance. The inputs for the
outranking procedure were the probability distributions of the environmental
indicators of the scenarios under consideration. Since each scenario was represented
by 52 simulations, we had the data required to fit statistical distributions to the nine
environmental indicators. The first step was to fit univariate theoretical statistical
distributions to each environmental indicator. Due to the skewed distribution of the
impact categories, we restricted the options to lognormal, gamma and Weibull
probability distributions to fit the simulated results. To fit those probability functions,
we used the package fitdistrplus (Delignette-Muller and Dutang, 2015) from the R
statistical software package (version 3.5.1., R Core Team, 2018). We selected the
probability function with the best fit for each impact category based on the Akaike
and Bayesian information criteria. The correlation structure between the impact
categories was considered by using a random sampling Monte Carlo procedure, which
is a core method of the SMAA method. This step is important because ignoring the
presence of correlations can lead to inaccurate decisions (Groen and Heijungs, 2017).
To build multivariate distributions, the starting point is the recursive decomposition
of a multivariate density into products of conditional densities (Czado, 2010). Starting
with the transformation of the original data to uniform marginal distributions, we
fitted a regular vine (R-vine) copula model to the 9-dimensional data set. The method
followed a joint search of an appropriate R-vine structure, its pair copula families and
an estimation of their parameters as a top down strategy in an automatic way
(Dißmann et al., 2013). In the next step, the families and parameters for all pair of
copulae in the top tree were selected using the smallest AIC. The complete R-vine
specification was built by selecting the strongest pairwise conditional dependencies
first (Czado et al., 2013). This method was applied using the package VineCopula
(Schepsmeier et al., 2018) included in the R statistical software package.
98
Once the R-vine copula model was fitted, we performed the Monte Carlo procedure
for each environmental indicator considering a sample size of 10000. Since the
marginal distributions of the R-vine copula model are uniform, we transformed back
the sampled data to the aforementioned fitted distributions.
Next, we calculated uncertainty-based preference (Pi) and indifference (Qi) thresholds
based on the determined uncertainty of the data. An outranking score was calculated
per pairwise comparison and per environmental indicator, and the net flow was
established as the sum of these scores. Then, we defined the weights for each
environmental indicator based on a stochastic procedure by assigning a distribution
to the weight factor values, resulting in an aggregated overall score that accounted for
all possible value systems (Prado-Lopez and Heijungs, 2018). These overall scores
represented the relative performance of the scenarios under comparison, with a higher
overall score indicating the best environmental performance. Finally, a probability
value per scenario per rank was configured based on ranking the scenario per run. A
schematic overview of the proposed method is presented in Figure 23.
99
Figure 23. A schematic overview of the coupled soil-crop model and LCA to assess the environmental performance of N-fertilization scenarios.
100

Life cycle inventory
Out of the simulations, yield was predicted for both production systems and under
different scenarios considering 52 production cycles per scenario. The kernel
probability distribution for the predicted yield was skewed for all combinations of the
production systems and scenarios (Figure 24). This result was the consequence of
seasonal weather fluctuations on crop growth driven by the weekly transplanting
dates. A higher level of variability was observed for the open field scenarios due to
the exposure of the crop to the elements, particularly rainfall (Appendix 4). Since the
distributions were skewed, we decided to use the median to describe the central trend
of the simulated tomato yield. Irrespective of the scenario, the yield in the greenhouse
system varied between 62.4 and 115.4 t ha-1 with a median yield for the Op scenario
of 111.0 t ha-1, i.e., 48.9% and 61.8% higher than in the Fx and Cr scenarios,
respectively. For the open field system, the trend was similar, but with yields ranging
between 12.2 and 52.3 t ha-1. In the open field scenarios, the optimal scenario showed
a median yield of 28.7 t ha-1 that was 4 and 50% higher than the one obtained for the
Cr and Fx scenarios, respectively.
Figure 24. Kernel probability density functions describing the tomato yield of the
scenarios evaluated in the greenhouse (A) and open field (B) systems.
The results of the simulations were mainly driven by the climate as well as the
irrigation and N-fertilization strategies of the scenarios. While the applied N was
defined before the simulations for the Cr and Fx scenarios, the N-demands for the Op
scenario were predicted by the soil-crop model. In the open field system, the optimal
N-fertilization that matched the N-demand, ranged from 302 to 436 kg N ha-1, while
in the greenhouse system, it ranged from 431 to 498 kg N ha-1. For the greenhouse
system, the water demand was estimated from crop evapotranspiration, and the results
indicated that the plant required between 248 and 342 l m-2 per cycle. Water demand
is a function of development stage, starting at 0.5 l m-2 d-1 for the vegetative stage and
ending at 4.8 l m-2 d-1 when the plant was in the full production stage. In the open field
system, the water input from precipitation varied between 580.3 and 1233.1 mm per
cycle.
101
The planting date had an important effect on yield (Appendix 4). In the case of the
greenhouse production system, the highest yields were obtained when the cycles were
established between the months of February and March, while the lowest yields were
obtained for cycles established between May and November. From a climatic
standpoint, the months in which the best yields were obtained were characterized by
periods of transition from the dry to rainy season, while the lowest yields occurred
when the crop was established in the dry season between June and August. In the case
of the open field production system, the highest yields were observed for the crops
established between October and December. This period was characterized by
decreased rainfall, which was accompanied by an increase in solar radiation. In the
open field production system, high rainfall throughout the year seemed to restrict
production, and only when the rainfall decreased an increased production was
possible. In the greenhouse production system, where the water supply was
controlled, the intense dry season characterized by low relative humidity appeared to
affect tomato yield. Notably, greenhouses do not have climate control systems, which
makes them highly dependent on external climatic conditions.
Nitrogen emissions
The simulated N-emissions also showed a high degree of variation, particularly in the
open field system (Figure 25), also as a result of the seasonal weather fluctuations.
Appendix 5 depicts the dynamics of the N-emissions for each scenario throughout the
52 simulated production cycles. In the open field system, a higher level of variation
occurred, especially for nitrate losses, which depend to a large extent on soil water
inputs and properties such as the sandy texture of the soils (Gil et al., 2019b). A higher
level of N-emissions was predicted in the open field scenarios than in the greenhouse
scenarios. This result was previously reported by Gil et al. (2018) who applied
standard methods such as the emissions factors proposed by Brentrup et al. (2000)
and the regression analyses method of Nemecek et al. (2016). The high emission
variability was also highlighted in previous studies, and this variability was mainly
associated with the great diversity in management practices involving fertilization
carried out by growers (Gil et al., 2018). However, the results presented here include
the seasonal effect of the weather on the predicted N-emissions in a more structured,
mechanistic way.
102
Figure 25. Emissions of ammonia (NH3), nitrous oxide (N2O) and nitrate (NO-3 )
during the crop cycles simulated for each scenario in the greenhouse and open field
systems.
103
Considering the scenarios, we observed an overlap between the Fx and Op scenarios

of the open field system. In quantitative terms, the Fx scenario achieved a greater N-
emissions reduction, although the difference was less than 1% compared to that of the
Op scenario (Table 20). In the case of the greenhouse system, the Op scenario clearly
obtained the greatest reductions in ammonia and nitrous oxide emissions with no
nitrate emissions. Although there was no clear differentiation for the remaining two
scenarios, the Cr scenario had higher emissions (Figure 25). According to the
descriptive statistics presented in Table 20, the Op scenario of the greenhouse system
achieved the highest N-emissions reductions. For this scenario, the N2O emissions
were reduced by 75.5 and 77.9% compared to those of the Cr and Fx scenarios,
respectively. The reductions in NH3 emissions were more moderate, with reductions
of 19.1% and of 16.2% compared with those of the Cr scenario and the Fx scenario,
respectively.
nitrogen fertilizers in the scenarios evaluated for both open field and greenhouse
systems.
Scenario Emission source Greenhouse Open field
Median Standard Deviation Median Standard Deviation
Cr Air emissions
NH3 0.014 1.05E-03 0.037 0.012
N2O 0.123 0.011 0.593 0.159
Water emissions
NO-3 4.55 0.44 19.81 6.51
Fx Air emissions
NH3 0.013 1.02E-03 0.030 9.83E-03
N2O 0.111 0.011 0.419 0.121
Water emissions
NO-3 4.02 0.41 13.72 4.76
Op Air emissions
NH3 0.011 9.23E-04 0.029 9.74E-03
N2O 0.027 1.64E-03 0.424 0.117
Water emissions
NO-3 0.000 0.0000 13.35 4.64
Comparing the results for the Cr scenario with those from previous works in which
standard methods were used to calculate emissions (Brentrup et al., 2000; Nemecek
et al., 2016), important differences can be seen. The predicted NH3 was one order of
magnitude lower than those reported for the same production systems but using the
standard methods (Gil et al., 2018). For N2O and NO-3 emissions, the model results
were higher by one order of magnitude for the greenhouse system than for the open
field system, while the modeled NO-3 emissions of the open field system were two
orders of magnitude higher than those previously reported (Gil et al., 2018). Although
the soil-plant model was calibrated and validated (Gil et al., 2017c; Heuts, 2018), the
simulated scenarios still represent a simplification of the practices carried out by the
growers. For instance, the scenarios did not include the use of organic fertilizers, and
only a single N-inorganic fertilizer product was used for the simulations, which
104
generated, at least partially, the aforementioned differences. The emission estimations

need to be validated with experimental measurements to further calibrate and validate
the required modules of the soil model for tropical conditions.
Aside from the precision of the emissions estimates, the soil-crop model allowed the
estimation of the potential N-emissions associated with fertilization and irrigation
management scenarios under local weather conditions. The intrinsic errors that may
have occurred due to model simplifications will have the same degree of magnitude
for each scenario, allowing us to compare environmental performance among the
scenarios.
The planting date and, consequently, the climatic conditions under which each cycle
was developed also led to fluctuations in N-emissions. For example, Figure 26 shows
the variation in the emissions of NO-3 as a consequence of the planting date. The
figures in Appendix 5 illustrate the variation for the other N-pollutants. Specifically,
in the open field production system, considerable variation can be observed for the
three pollutant types considered. In this system, the three evaluated scenarios showed
similar trends in emissions, although the Cr scenario always shows the highest values.
In the open field, the lowest emissions occurred when crops were established at the
end of the second rainy season, i.e., between the months of October and November.
In the case of the greenhouse production system, the fluctuations in the emissions as
a consequence of the planting date were smaller compared to those observed in the
open field system. In this production system, an important decrease in emissions was
observed in response to the Op scenario. The differences due to the planting date were
associated with the water flowing into the system and the temperature gradient in the
soil. These two factors largely determine the type of processes related to nitrogen
transformation and the speed at which they occur.
Figure 26. Effect of the transplanting date on emissions of nitrate in the greenhouse
and open field systems.
105
Life cycle impact assessment

In the next section, information related to the on-field N-emissions was integrated
with the information related to fertilizer manufacturing to assess the environmental
performance of the scenarios. Figure 27 and Figure 28 show the kernel probability
distributions for the analyzed impact categories of the greenhouse and open field
systems, respectively. In all cases, the probability distributions were skewed, again
describing the weather seasonality effect. The temporal dynamics of the impact
categories for each production system and scenario are shown in Appendix 6. The
comparative LCA results are shown in Table 21, indicating that the Op scenario
showed the best performance for categories such as GWP, HTTP, AP and EP, which
were directly affected by the on-field N-emissions. In contrast, the Cr scenario
performed better in the categories where only the impacts derived from fertilizer
manufacturing were considered.
106
Figure 27. Kernel probability density functions for each impact category in the
scenarios evaluated for the greenhouse system.
107
Figure 28. Probability density functions for each impact category in the scenarios
evaluated for the open field system.
108
Analyzing each production system, we observed that the Op scenario of the

greenhouse system resulted in a reduction for all the categories compared to those of
the Fx scenario, with the highest reductions occurring for EP and GWP with savings
of 99 and 69.9%, respectively. The comparison between the Op and Cr scenarios
showed similar reductions to those observed between the Op and Fx scenarios for the
AP, EP and GWP categories. However, the remaining impact categories in the Op
scenario showed, on average, an increase of 10.6% compared with those of the Cr
scenario. In the open field system, in comparison to the Fx scenario, the Op scenario
showed a slight decrease for all impact categories except for EP. For this category,
the environmental burdens of both scenarios were similar. In contrast to the
expectations, we observed an average increase of 26.8% for the environmental
impacts for the Op scenario in comparison to that of the Cr scenario, except for EP.
For the EP category, in comparison to the Cr scenario, the Op scenario reduced the
impact by 26.5%. The increase in yield achieved in the Op scenario failed to offset
the environmental burdens due to the extra on-field N-emissions and fertilizer
manufacturing.
Based on the results, the planting date clearly affected the environmental performance
of tomato production, especially in the open field production system under the Cr and
Fx scenarios. In this system, the lowest environmental impacts were generated when
the crop was established in October and November, which was when the highest
yields and lowest emissions were obtained, as mentioned above. In the case of the
greenhouse system, the lowest impacts were observed for the cycles established in the
months of February and March, which were the months when the highest yields and
lowest emissions were achieved.
The overall scenario comparison based on point estimates of the impact categories
such as the median or their probability distributions was not straightforward due to
the multivariate nature of the information. This comparative LCA, involving several
impact categories, described a common situation where none of the alternatives stood
out, hindering the possibility of providing direct recommendations to stakeholders.
Therefore, the results of a stochastic single indicator-based method are presented next
to determine the best environmental performance alternative.
109
Table 21. Median values for potential environmental impacts per ton of tomato produced in the scenarios evaluated for open field and
greenhouse conditions in the Colombian Andes.
System/Scenario AD AP EP FAETP GWP HTTP MAETP POP TEP
Greenhouse (kg Sb (kg SO2 eq) (kg PO-34 (kg 1.4-DB (kg CO2 eq) (kg 1.4-DB (kg 1.4-DB eq) (kg C2H4 (kg 1.4-DB
eq) eq) eq) eq) eq) eq)
Cr 0.010 0.037 2.064 2.15E-05 61.421 0.015 1.62E-04 2.05E-04 2.32E-06
Fx 0.014 0.040 1.827 2.91E-05 57.136 0.019 2.20E-04 2.78E-04 3.14E-06
Op 0.011 0.034 0.019 2.39E-05 17.192 0.016 1.81E-04 2.28E-04 2.57E-06
Open field
Cr 0.239 0.280 2.639 5.14E-04 168.937 0.316 3.89E-03 4.92E-03 5.55E-05
Fx 0.340 0.385 1.935 7.31E-04 182.953 0.448 5.53E-03 6.99E-03 7.88E-05
Op 0.313 0.356 1.940 6.73E-04 173.601 0.412 5.09E-03 6.43E-03 7.25E-05
110

As mentioned above, the probability density functions associated with each impact
category were characterized by a high level of variation and skewness. In the open
field system, the average coefficient of variation for all impact categories of the Op
scenario was 31.2%, while for the Fx and Cr scenarios, their coefficients of variation
were on average 29.6% and 29.5%, respectively. In comparison to the open field
system, the greenhouse system had much lower levels of variation, although the
highest levels of variation were observed for the Fx and Cr scenarios with coefficients
of variation averaging 7.7% in both cases, while for the Op scenario, the average
coefficient of variation was 4.7%.
The varying degree of correlation among the impact categories (Appendix 7) was
considered by following a multivariate random sampling procedure to construct a
realistic sampling space per production system per scenario allowing the application
of the SMAA outranking method. The inclusion of the correlations between impact
categories in the analysis mitigated the risk of generating under- or overestimates in
the determination of the indicator for each assessed scenario. In comparative LCAs,
such as the one carried out in this work, the inclusion of the correlation between the
impact categories guarantees to find the alternative with the best environmental
performance (Groen and Heijungs, 2017). The correlation among impact categories
was achieved through the copula method, and its output is summarized in Table 22.
For all the scenarios, the copula method yielded a subclass of R-vines known as D-
vines. Multivariate distribution construction with dependent marginals using copulas
is of interest in fields where dependence between variables is the rule, e.g., financial
risk analysis (Ane and Kharoubi, 2003). However, its use in environmental
assessments such as agricultural LCAs that yield a correlated group of environmental
indicators remains unexplored. The main advantages of the copula method for
uncertainty analysis in LCAs involve the possibility of including actual correlations
between impact categories and the possibility to work with a wide range of probability
distributions moving away from the normal distribution, if needed. The comparison
between the observed and simulated probability density functions for the assessed
scenarios are included in Appendix 8.
Table 22. Number of parameters, log-likelihood, Akaike information criterion (AIC)
and Bayesian information criterion (BIC) of the 36 bivariate copula estimated D-vine
models for the environmental indicators of the scenarios evaluated in greenhouse and
open field systems.
System/Scenario Parameters Log-Likelihood AIC BIC
Greenhouse
Cr 21 5100.65 -10141.31 -10082.77
Fx 21 4296.76 -8531.52 -8471.03
Op 21 4761.87 -9451.75 -9381.50
Open field
Cr 22 7007.00 -13942.01 -13871.76
Fx 23 7473.54 -14871.08 -14796.94
Op 25 7186.53 -14289.07 -14207.12
111
The last step, after the SMAA method, was the construction of the probabilistic rank
that allowed the comparison of the likelihood between the scenarios for both systems.
The cumulative distribution of the rankings for each considered scenario is shown in
Figure 29. The ranking order of the scenarios showed a first-order stochastic
dominance (Canis et al., 2010), meaning that in comparison to the other scenarios, the
Op scenario of the greenhouse system had a higher probability of ranking better, while
the opposite situation was observed for the open field Fx scenario. In this study, a
better ranking means better environmental performance; therefore, comparisons can
be made between inter- and intra-production system scenarios. For instance, in the
open field system, the alternative Fx and Op scenarios did not improve the
environmental performance of the current system, while in the greenhouse system,
these alternative scenarios showed the expected improvements.
Figure 29. Cumulative distribution functions of the rankings for each nitrogen
fertilization scenario.
Another way to consider the results of the SMAA method is through the rank
acceptability index (RAI). This index assigns an overall probability value to each
evaluated scenario based on the count of the ranking for each alternative obtained in
each run (Prado-Lopez and Heijungs, 2018). The RAI values allowed a
straightforward classification of the scenarios, as shown in Figure 30. According to
the RAI, the two scenarios with the best environmental performance were the Op and
Fx scenarios of the greenhouse system. The Cr scenario of the open field system
showed the highest probability of ranking last, i.e., the scenario with the lowest
environmental performance.
112
Figure 30. Rank acceptability index for each of the scenarios evaluated regarding
nitrogen fertilization.
The alternative Fx scenario demostrates what commonly occurs in the agricultural
sector at the local level, where fixed doses of inputs such as fertilizers are
recommended to growers. However, in the case of N-fertilization, this approach
overlooked the dependence on the timing and dosing of fertilizers and irrigation
throughout the production cycle. In Belgium, Van Loon (2018) concluded that
although vegetable production and nitrate emissions can be matched by an appropriate
fertilization strategy, ultimately, the weather conditions determine both aspects.
Therefore, in open field systems, the uncertain climatic conditions, particularly
rainfall, largely define the risk of nitrate losses in a production system (Van Loon,
2018).
On the basis of these findings, the greenhouse production system, despite its low
technological level, had the best prospects for environmental improvement. The
pathway to improve the environmental performance of crops managed by
smallholders involves implementing practices aimed at satisfying crop nutrient
demands by considering both local climate and soil characteristics. For instance, the
simulation results revealed the possibility of reducing the current volume of irrigation
water applied by growers without harming production but reducing the risk of nitrate
leaching into groundwater. The results of Gil et al. (2019b) showed that 467.5 l m-2
are currently used for a greenhouse tomato production cycle, but the Op scenario
reduced this amount to 300 l m-2 while simultaneously increasing yield. Water saving
in agriculture is recognized as one of the greatest challenges (Hamdy et al., 2003),
and in the case of systems such as greenhouse tomatoes, these savings can be
associated with lower potential emissions of harmful pollutants to the environment.
In contrast, the open field production system had greater limitations in terms of
improving its environmental performance due to its direct exposure to the elements,
113
for instance to thunderstorms and high precipitation events that favor nitrate leaching
towards the aquifer. In the open field system, the climatic conditions of each cycle,
determined by the transplanting date, were the driving factors of the achieved yield
(Appendix 4) and the generated emissions (Appendix 5). Although the climatic
datasets were generated from historical data in the present work, the model can also
be used to evaluate the effect of prospective climate (change) scenarios. In the case
of the greenhouse system, the effect of modifications such as the inclusion of active
climate control systems can be evaluated in the future. Although this work focuses on
technical aspects, this approach could be coupled with participatory research methods
in which the community can weight environmental indicators based on local interests,
such as reducing the pollution risk to water bodies.
Favorable climatic conditions in tropical regions allow the establishment of crops
throughout the year. The results of the simulations showed an important effect of the
planting date on the environmental performance of the production. This effect
occurred because the climatic conditions of each cycle had a direct effect on
photosynthesis and thus on plant growth and development. In addition, climatic
conditions, e.g., precipitation in the open field system, affected the soil nitrogen
dynamics and thus the N-emissions and plant water availability. However, in the case
of the greenhouse system, where precipitation did not affect the crop directly, other
climatic variables (e.g., relative humidity) affected plant development and thus tomato
production as well as the environmental performance. In an ideal scenario, producers
should adjust fertilization practices according to the climatic conditions that occur
during a cycle. However, this type of approach requires constant monitoring of
climatic conditions, which does not occur in smallholder-based production systems in
countries such as Colombia.
The integration of soil-crop models with an environmental assessment of agricultural
systems allows an analysis of the dynamics of emissions in time and their impacts
throughout the production cycle. This approach helps to identify critical periods in
time where actions should be carried out to address pollution risks. In the case study
addressed in the present work, adequate management of fertilization and irrigation
can minimize the risk of NO-3 emissions in the greenhouse production system and
consequently reduce the impact on the environment. The proposed method can be
applied to multiyear crop rotations or perennial crops where the dynamic estimation
of emissions can be more informative because of the long cropping season, especially
in the tropics.
One of the main objections to using soil-crop models to calculate on-field emissions
is that they require a large amount of input data as well as the calibration of many
parameters, which may lead to a high uncertainty in the results (Nemecek et al., 2016).
However, currently, soil-crop models are well-developed and reliable tools that are
complex enough to adequately represent the main biogeochemical processes involved
in agricultural production (Brisson et al., 1998; Dufossé et al., 2013; Giltrap et al.,
2010; Rahn et al., 2010). On the other hand, the availability of climate and soil data,
which are the main input constraints for using these models, is increasing worldwide
through open access web-based platforms. The modular structure of soil-crop models
allows inclusion or improvement of the modules representing current or additional
processes as long as the knowledge is generated in related fields of study. For instance,
modules modeling the fate of other potentially polluting agrochemicals such as
114
phosphorus or the effects of farming practices such as mulching can be included in

the future. The flexibility of soil-crop models allows them to be reused or adapted to
represent local conditions depicting hypothetical scenarios before their field
implementation (Jones et al., 2017).
7.4. Conclusions
This work demonstrated the feasibility of coupling soil-crop models with a life cycle
assessment to dynamically determine the environmental impacts of prospective
scenarios as an alternative to the standard evaluation of alternatives through field
experimentation. Once the model is calibrated and validated locally, regional
environmental assessments of production strategies are possible with a high degree of
confidence given the current soil-crop modeling development. Given the dynamic
nature of the model, it was possible to analyze in detail the effect of management
practices, e.g., fertilization, not only from a production perspective but also from an
environmental impact perspective. In the present work, the proposed method was
applied to a short-season crop; however, this method can also be applied to assess the
environmental performance of rotation schemes and annual or perennial crops. Being
able to gain information on the dynamics of pollutant emissions during a cropping
cycle, particularly for crop rotations or long-cycle crops, will help identify key points
in time where actions can be taken to reduce those emissions.
We must acknowledge that the soil modules defining N-emissions, such as NO-3
leaching, NH3 volatilization, and N2O losses, were not directly calibrated due to
missing data. The environmental assessment of production strategies by coupling crop
modeling and an LCA provides an opportunity to evaluate an abundance of scenarios
that, depending on the impact assessment method, can result in the interpretation stage
being an overwhelming task. To overcome this situation, we propose carrying out the
normalization and weighting stages through the SMAA method while recognizing the
correlation structure among the impact categories. Thus, we incorporated the copula
method, which constructs multivariate joint probability distributions based on relaxed
sample-size assumptions and in a computationally inexpensive way. Overall, the
SMAA method produced a stochastic ranking of the alternatives under consideration
without bias or subjective criteria. Thus, the outcome of the SMAA method can be
easily communicated to stakeholders or policy-makers.
115
Chapter 8. General conclusions
To characterize the open field and greenhouse tomato production systems in terms of
biophysical conditions and management practices carried out by Colombian growers
with special emphasis on nitrogen fertilization.
The present work began with a characterization of tomato production systems in open
field and greenhouse in Colombia. In Colombia tomato production takes place in
heterogeneous soils that are characterized by low levels of fertility, leading to
excessive applications of fertilizers. Due to favorable tropical climatic conditions, the
cultivation of tomatoes occurs throughout the year. Plantations are established in
small inter-Andean valleys or on mountain slopes, with great diversity in soil
properties, such as texture, minerals and organic matter content, which introduces an
additional source of variation in this agricultural production system. Finally, both
production systems are characterized by low technological levels and the lack of
adequate extension services, resulting in a wide range of management practices. The
characterization of these systems is rarely carried out for smallholders and therefore
constitutes valuable information about their current status, generating
recommendations to improve these systems considering that they produce most of the
locally consumed vegetables.
Both production systems intensively use agrochemicals, especially in relation to
fertilization. However, the results demonstrated that in comparison to the open field
system, the greenhouse system more intensively used fertilizer. In addition, the results
showed that yields per unit area were higher in the greenhouse production system than
in the open field system. Thus, the question arises as to whether, from an
environmental point of view, the higher production that occurred in the greenhouse
system effectively compensates for the larger use of agrochemicals and the use of
more complex and expensive infrastructure.
To compare the environmental performance of the open field versus the greenhouse
production system by applying a standard LCA based on a single indicator that
incorporates the intrinsic variation in each system and the correlation structure
between the impact categories.
A standard LCA was used to evaluate and compare the environmental performance
of each of these systems. A single environmental impact indicator was developed,
integrating all the impact categories. This indicator was an improvement proposed for
the LCA normalization and weighting stage due to the capability to include the
intrinsic variation in the production systems together with the existing correlation
structure among the impact categories. For this purpose the copula method was
introduced. During the development of this work, the stochastic multi-attribute
analysis (SMAA) method was successfully implemented to compare the
environmental performance of each production system. In relation to this last aspect.
This approach ensured that the SMAA could generate unbiased indicators that would
contribute to the identification of the most environmentally sound production systems.
Based on the indicator, it was determined that the open field production system had
better environmental performance than the greenhouse production system.
General conclusions
After developing a method to construct a single objective indicator to evaluate the

overall environmental performance of the production systems, the next step was to
modify the static character of the LCAs. Thus, the second step involved developing a
tomato model with the ability to dynamically simulate the growth and development
of tomato plants under management practices carried out by local producers.
To develop a climate-based tomato model with the ability to simulate tomato growth
and development based on the growing patterns and management practices applied
by local farmers.
The model adopted the same structure as TOMGRO, but in terms of the outputs, it
was simplified to provide the necessary outputs for coupling with the LCA. The model
represents the dry matter (DM) production and its allocation to the different plant
organs. Hourly, the photosynthetic rate was determined for the canopy leaf area and
at the end of each day, gross photosynthesis was calculated as the sum of the hourly
photosynthetic assimilates. Daily maintenance respiration was calculated based on the
accumulated DM in stems, active leaves and growing fruits. DM availability for
growth was calculated at the end of the day as the daily gross photosynthesis minus
daily total respiration. The total DM is allocated to the plant organs in function of the
development stage using the accumulated thermal time (ATT) approach. This model
was calibrated for the local conditions of each production system. As a result, the
model was able to simulate the potential yields as a function of the current
technological level of each system and the restrictions imposed by the local climate
conditions. The model required specifying whether the crop was grown in open field
or greenhouse because although photosynthesis and respiration processes work the
same way, the functions that determine dry matter partitioning depend on the cropping
system.
To validate a soil-plant model for both tomato production systems in Colombia,
resulting from the integration of an adapted version of the tomato model and a
previously calibrated soil model.
With the calibrated plant model, the next step was to integrate it with the soil model,
previously calibrated for Belgian conditions. The adaptation consisted of adding
modules to the soil-crop models to estimate evapotranspiration, N-uptake and root
growth. The model was thus able to estimate the growth requirements of water and N
as well as plant root growth and uptake of these inputs from the soil layers. The plant
model was coupled to a soil model previously calibrated and validated for Belgian
horticultural production conditions (Heuts, 2018). The resulting soil-plant model was
validated for Colombian conditions by evaluating production variables related to plant
growth and development, as well as the soil water and N-dynamics. The result of the
validation process was acceptable, looking at the goodness of fit statistics. Thus, a
soil-plant model capable of simulating both the nitrogen- and water-processes that
occur in the soil as well as their interaction with the growth and development of a
tomato crop was obtained. The soil-plant model allowed the simulation of variables
such as soil temperature, water flow through the system and the dynamic
transformation of C and N and also included the potential emissions of pollutants such
as NH3, N2O and NO-3 . In addition, the model allowed the estimation of the impact on
the yield of deficient supplies of water and/or N.
117
General conclusions
To objectively evaluate a set of nitrogen fertilization scenarios as alternatives to

improve the current environmental performance of the tomato production systems in
Colombia.
The final step of the present work was to use the soil-plant model to evaluate the effect
of several scenarios in relation to the use of N-fertilizers and to incorporate the outputs
within the LCI. Thus, the LCI was given a dynamic character depending on the
climatic conditions and the nitrogen fertilization strategies used in terms of dosing
and timing. This dynamic character also allowed the LCA to incorporate inherent
biophysical variations in agricultural systems, such as climatic conditions and soil
properties. The effects of N-fertilization practices, in terms of dosing and timing, on
environmental performance were also evaluated.
As a result of the information on the problematic fertilization practices carried out by
local growers, the soil-plant model was used together with the LCA to simulate
alternative scenarios to improve the environmental performance of tomato production
systems. The results of the LCA were used to construct a single objective indicator
using the SMAA method. Scenarios were evaluated for 52 planting dates, spread over
the year, to represent the initiation of a production cycle every week of the year. Thus,
we determined that in comparison to the open field system, the greenhouse production
system has greater potential to improve its environmental performance. The best
environmental performance was achieved by the scenario in which the N-supply
matched the plant demand and this supply was fractionated throughout the production
cycle. The simulated scenarios also showed the effect of planting date on the
environmental performance of tomato production. In the case of the open field
production system, the best environmental performance was achieved when the crops
were established by the end of the rainy season in the months of October and
November. The cycles established in this period showed a decrease in N-emissions
and a simultaneous yield increase. In the greenhouse production system, the crops
established between February and March at the beginning of the first rainy season
showed the best environmental performance. These results are important because
although tomato cultivation in the Colombian Andes can be established at any time
of the year, the effect of the planting date on crop productivity and environmental
performance had not been analyzed prior to this study.
Integrating soil-crop models into the LCA framework, as proposed in this thesis, adds
dynamic character to agricultural environmental assessments that are typically
characterized by a general static nature. Soil-crop modeling to predict nitrogen
emissions and crop yields given climate, soil and crop management data in relation to
nitrogen fertilization impact, disrupts the static character of LCAs developed in the
agricultural context. Under the assumption that the model is properly calibrated and
validated for local conditions, this modeling approach confers a predictive and
process-based character to the LCA, that can be used to evaluate prospective
scenarios. The present work was limited to the evaluation of scenarios related to N-
fertilization; however, the structure of soil-crop models makes it possible to evaluate
scenarios related to climate variability or spatial soil variability. Based on the above,
we conclude that the combined use of soil-crop modeling and LCA produces a
synergistic effect necessary to address the current environmental challenges of both
large- and small-scale agriculture around the world.
118
General conclusions
Useful recommendations for local growers, derived from our results, are to leave aside
traditional practices such as application of fixed doses of fertilizers cycle after
production cycle and begin to adjust the fertilization using soil analysis results as a
reference. Fertilizer application should also be fractionated so that applications are
matching plant demands.
In the greenhouse system, it could be proposed that fractionation be done on a daily
basis using the fertigation system. This necessarily implies investment in new
technologies (irrigation controllers) and an increase in the producers' level of
knowledge for the preparation of nutrient stock solutions. Furthermore, considering
that tomato production under greenhouse is subject to external climatic conditions, as
demonstrated by simulations, another step that producers must take is to improve the
current greenhouse characteristics. A general recommendation is to increase control
over the climate within greenhouses, for example, by improving ventilation during
daylight hours, and closing them completely at night to retain heat. In recent years
new greenhouse designs specifically adapted to local conditions have been proposed
and evaluated (Villagrán et al., 2019), which are characterized by significantly
improved ventilation, and which could improve conditions for tomato production.
These recommendations are aimed to increase the tomato production profitability and
also to reduce the risk of negative impacts on the environment.
For the open field growers, direct exposure to ever more unpredictable weather
conditions restricts the usefulness of some recommendations. However, in this system
a more frequent fractionation in the application of fertilizers is also recommended (at
least once a week). If the weather presents a regular behavior it would be
recommended to synchronize the planting with the periods in which the conditions
are more favorable to obtain higher yields with lower negative environmental risks,
which in the case of the study area used in the present work corresponds to the period
between October and December. Although this recommendation ignore the market
conditions such as the prices.
119
Chapter 9. Future works
The results of this work allowed the coupling of the LCA method with soil-crop
modeling to dynamically predict emissions due to the use of nitrogen fertilizers and
their corresponding environmental impact. Additionally, it was possible to construct
a global indicator based on the set of impact categories resulting from the LCA, which
takes into account the existing correlation between them as well as their inherent level
of variation. However, as a result of the analyses carried out throughout the work, four
topics that could be the focus of future work. The first corresponds to subjects related
to work that will lead to methodological improvements, the second focuses on
applications in which the proposed model can be used, the third focuses on the need
for quality input data to use the model, and the last is how transfer the results to the
local growers.
From the methodological point of view, it is recommended to prioritize the calibration
of the parameters that control the biogeochemical cycles of nitrogen and carbon in the
soil. After this calibration, efforts should focus on the validation of nitrogen emissions
through field measurements. This work demands the use of specialized methods and
equipment; therefore, these activities could be developed in coordination with
institutions or groups with experience on the subject. From the plant modelling
viewpoint, it is necessary to increase the volume of data available to improve aspects
of the model, such as the dry matter compartment functions. A critical aspect that
needs to be addressed in future research is related to determining the expansion
patterns of the soil rooting volume (3D), as well as the root distribution within this
volume. This topic is of concern at the local level, and it has been recognized as
needed in the models developed by Van Loon (2018) for temperate regions in
developed countries. These methodological improvements will induce higher
precision of the prediction of the state variables of the model. From the standpoint of
the methodology applied for the construction of a single indicator based on the results
of the LCA, future work may explore the effect of assigning different relative weights
to each impact category, defining priorities by local communities. The SMAA method
would allow this modification, and it would not require major modifications to the
overall proposed framework.
From the standpoint of the model's application, there are a wide range of scenarios
that can be evaluated to design strategies to improve current production systems. The
model can be used to evaluate the long-term effect of fertilization strategies that, in
addition to inorganic fertilizers, also include the use of different sources of organic
matter. In relation to water use, the effect of strategies such as the use of deficit
irrigation on crop yields and simultaneously on the environmental performance of the
system can be evaluated. The model can be used to focus on the analysis of different
climate change scenarios. In this case, the model is a suitable tool to anticipate future
scenarios and to design and evaluate strategies that minimize the impacts on
production. In the specific case of greenhouse production, the models also enable the
evaluation of the effect of modifications to the infrastructure that will lead to better
control of climatic variables such as maintaining daily temperature and humidity
oscillations in ranges closer to the optimal range for plant development.
Future works
The lack of input data, especially regarding soil properties and climate variables, is a
major constraint for using this model under local conditions. We recommend that
future work focuses on the generation or systematization of climatic and edaphic
information to serve as a basis for the use of these models. Some of the information
required by the model is recorded, at least partially, by government agencies, but
accessibility to such information is not an easy process even when digital platforms
provide access to data in real time. Obtaining soil data involves an extensive effort
since the model requires variables that describe the flow of water through the soil
profile. Thus, future work should focus on the development or adaptation of methods
that allow the generation of these data from existing archives or their generation for
the principal agricultural areas of the country.
The method of successfully transferring the results of this and other research to a
smallholder is also a remainder of future work. Surely it is necessary to leave aside
traditional strategies focused on the results divulgation and involve directly the
growers in the research processes to ensure that they appropriate the new knowledge.
This effort will require interdisciplinary groups that include professional agronomists
as well as sociologists to help understand and overcome the reasons that block the
adoption of new technologies.
121
References
Aas, K., Czado, C., Frigessi, A., Bakken, H., 2009. Pair-copula constructions of
multiple dependence. Insurance: Mathematics and Economics 44 (2), 182–198.
Acock, B., Charles-Edwards, D.A., Fitter, D.J., Hand, D.W., Ludwig, L., Warren
Wilson, J., Withers, A.C., 1978. The contribution of leaves from different levels
within a tomato crop to canopy net photosynthesis: an experimental examination
of two canopy models. J. Exp. Bot. 29, 815–827.
Agronet, 2018. Statistics for the Agricultural Sector [WWW Document]. Agronet -
Colomb. Minist. Agric. Rural Dev. URL http://www.agronet.gov.co (accessed
3.8.17).
Allen, R.G., Pruitt, W.O., Wright, J.L., Howell, T.A., Ventura, F., Snyder, R., Itenfisu,
D., Steduto, P., Berengena, J., Yrisarry, J.B., Smith, M., Pereira, L.S., Raes, D.,
Perrier, A., Alves, I., Walter, I., Elliott, R., 2006. A recommendation on
standardized surface resistance for hourly calculation of reference ETo by the
FAO56 Penman-Monteith method. Agric. Water Manag. 81, 1–22.
Ane, T., Kharoubi, C., 2003. Dependence structure and risk measure. J. Bus. 76, 411–
438.
Angstrom, A., 1924. Solar and terrestrial radiation. Report to the international
commission for solar research on actinometric investigations of solar and
atmospheric radiation. Q. J. R. Meteorol. Soc. 50, 121–126.
Ardila-Villamizar, H.L., Fajardo-Rojas, M.A., Ardila-Jaimes, O.A., Millán-Delgado,
O., 2011. Plan Prospectivo Guanentá 2025. Gobernación de Santander y
Fundación Universitaria de San Gil.
Aujla, M.S., Thind, H.S., Buttar, G.S., 2007. Fruit yield and water use efficiency of
eggplant (Solanum melongema L.) as influenced by different quantities of
nitrogen and water applied through drip and furrow irrigation. Sci. Hortic.
(Amsterdam). 112, 142–148.
Battista, P., Rapi, B., Raschi, A., Romani, M., Massa, D., Carmassi, G., Diara, C.,
Incrocci, L., Pardossi, A., 2015. Modified TOMGRO outputs as guide factors to
estimate evapotranspiration and water use efficiency of three tomato fresh
cultivars, grown in a low-tech Italian glasshouse, in: VIII International
Symposium on Irrigation of Horticultural Crops 1150. pp. 39–46.
Bessou, C., Mary, B., Léonard, J., Roussel, M., Gréhan, E., Gabrielle, B., 2010.
Modelling soil compaction impacts on nitrous oxide emissions in arable fields.
Eur. J. Soil Sci. 61, 348–363.
Bojacá, C.R., Arias, L.A., Ahumada, D.A., Casilimas, H.A., Schrevens, E., 2013.
Evaluation of pesticide residues in open field and greenhouse tomatoes from
Colombia. Food Control 30, 400–403.
Bojacá, C.R., García, S.J., Schrevens, E., 2012. Investigating the technical
sustainability of farming systems with correlational biplots. Int. J. Sustain. Dev.
World Ecol. 19, 361–368.
Bojacá, C.R., Gil, R., Gómez, S., Cooman, A., Schrevens, E., 2009. Analysis of
greenhouse air temperature distribution using geostatistical methods. Trans.
References
ASABE 52.
Bojacá, C.R., Schrevens, E., 2010. Parameter uncertainty in LCA: stochastic sampling
under correlation. Int. J. Life Cycle Assess. 15, 238–246.
Bojacá, C.R., Wyckhuys, K.A.G., Schrevens, E., 2014. Life cycle assessment of
Colombian greenhouse tomato production based on farmer-level survey data. J.
Clean. Prod. 69, 26–33.
Boote, K.J., Jones, J.W., Hoogenboom, G., White, J.W., 2010. The role of crop
systems simulation in agriculture and environment. Int. J. Agric. Environ. Inf.
Syst. 1, 41–54.
Boote, K.J., Jones, J.W., White, J.W., Asseng, S., Lizaso, J.I., 2013. Putting
mechanisms into crop production models. Plant. Cell Environ. 36, 1658–1672.
Boulard, T., Raeppel, C., Brun, R., Lecompte, F., Hayer, F., Carmassi, G., Gaillard,
G., 2011. Environmental impact of greenhouse tomato production in France.
Agron. Sustain. Dev. 31, 757–777.
Bouman, B.A.M., van Laar, H.H., 2006. Description and evaluation of the rice growth
model ORYZA2000 under nitrogen-limited conditions. Agric. Syst. 87, 249–273.
Bradbury, N.J., Whitmore, A.P., Hart, P.B.S., Jenkinson, D.S., 1993. Modelling the
fate of nitrogen in crop and soil in the years following application of 15 N-labelled
fertilizer to winter wheat. J. Agric. Sci. 121, 363–379.
Bradstreet, R.B., 1954. Kjeldahl method for organic nitrogen. Anal. Chem. 26, 185–
187.
Bremner, J.M., Keeney, D.R., 1966. Determination and Isotope-Ratio Analysis of
Different Forms of Nitrogen in Soils: 3. Exchangeable Ammonium, Nitrate, and
Nitrite by Extraction-Distillation Methods1. Soil Sci. Soc. Am. J. 30, 577–582.
Brentrup, F., Küsters, J., Kuhlmann, H., Lammel, J., 2001. Application of the Life
Cycle Assessment methodology to agricultural production: an example of sugar
beet production with different forms of nitrogen fertilisers. Eur. J. Agron. 14, 221–
233.
Brentrup, F., Küsters, J., Lammel, J., Kuhlmann, H., 2000. Methods to estimate on-
field nitrogen emissions from crop production as an input to LCA studies in the
agricultural sector. Int. J. Life Cycle Assess. 5, 349–357.
Brisson, N., Mary, B., Ripoche, D., Jeuffroy, M.H., Ruget, F., Nicoullaud, B., Gate,
P., Devienne-Barret, F., Antonioletti, R., Durr, C., Richard, G., Beaudoin, N.,
Recous, S., Tayot, X., Plenet, D., Cellier, P., Machet, J.M., Meynard, J.M.,
Delécolle, R., 1998. STICS: a generic model for the simulation of crops and their
water and nitrogen balances. I. Theory and parameterization applied to wheat and
corn. Agrinomie 18, 311–346.
Burns, I.G., 1974. A model for predicting the redistribution of salts applied to fallow
soils after excess rainfall or evaporation. J. Soil Sci. 25, 165–178.
Canis, L., Linkov, I., Seager, T.P., 2010. Application of stochastic multiattribute
analysis to assessment of single walled carbon nanotube synthesis processes.
Environ. Sci. Technol. 44, 8704–8711.
Carrera, E., Arévalo, D., Forero, J., Molina, C., 1975. Estudio general de suelos de la
provincia de Ricaurte y municipio de Samacá (departamento de Boyacá).
República de Colombia, Ministerio de Hacienda y Crédito Público, Instituto
123
References
Geográfico Agustín Codazzi, Subdirección Agrológica.

Castillo-Hernández, M.A., 2008. Modelo del efecto del suministro diferencial de
nitrógeno sobre el crecimiento potencial del cultivo de tomate adaptado a
condiciones colombianas. MSc thesis in Modeling and Simulation. Universidad
Jorge Tadeo Lozano.
Chen, D., Suter, H., Islam, A., Edis, R., Freney, J.R., Walker, C.N., 2008. Prospects
of improving efficiency of fertiliser nitrogen in Australian agriculture: a review of
enhanced efficiency fertilisers. Soil Res. 46, 289–301.
Chenoune, R., Belhouchette, H., Gomez, S., Capillon, A., 2016. NJAS - Wageningen
Journal of Life Sciences Assessing the diversity of smallholder rice farms
production strategies in Sierra Leone 76, 7–19.
Coleman, K., Jenkinson, D.S., 1996. RothC-26.3-A Model for the turnover of carbon
in soil, in: Evaluation of Soil Organic Matter Models. Springer, pp. 237–246.
Cooman, A., 2002. Feasibility of protected Tomato cropping in the high altitude
tropics using statistical and system dynamic models for plant growth and
development. PhD Thesis. KU Leuven.
Craufurd, P.Q., Vadez, V., Jagadish, S.V.K., Prasad, P.V.V., Zaman-Allah, M., 2013.
Crop science experiments designed to inform crop modeling. Agric. For.
Meteorol. 170, 8–18.
Czado, C., 2010. Pair-Copula Constructions of Multivariate Copulas, in: Jaworski, P.,
Härdle, W., Rychlik, T. (Eds.), Copula Theory and Its Applications. Berlin, pp.
93–109.
Czado, C., Jeske, S., Hofmann, M., 2013. Selection strategies for regular vine copulae.
J. la Société Française Stat. 154, 174–191.
Danaher, P.J., Smith, M.S., 2011. Modeling Multivariate Distributions Using
Copulas: Applications in Marketing. Mark. Sci. 30, 4–21.
de Visser, P.H.B., Buck-Sorlin, G.H., van der Heijden, G.W.A.M., 2014. Optimizing
illumination in the greenhouse using a 3D model of tomato and a ray tracer. Front.
Plant Sci. 5, 48.
de Vries, F.W.T.P., 1980. Simulation models of growth of crops, particularly under
nutrient stress, in: Physiological Aspects of Crop Productivity, Proc. 15th Coll.
Int. Potash Inst., Wageningen. pp. 213–226.
Del Grosso, S.J., Mosier, A.R., Parton, W.J., Ojima, D.S., 2005. DAYCENT model
analysis of past and contemporary soil N2O and net greenhouse gas flux for major
crops in the USA. Soil Tillage Res. 83, 9–24.
Delignette-Muller, M.L., Dutang, C., 2015. fitdistrplus: An R Package for fitting
distributions. J. Stat. Softw. 64, 1–34.
Di Paola, A., Valentini, R., Santini, M., 2016. An overview of available crop growth
and yield models for studies and assessments in agriculture. J. Sci. Food Agric.
96, 709–714.
Dias, L.C., Passeira, C., Malça, J., Freire, F., 2016. Integrating life-cycle assessment
and multi-criteria decision analysis to compare alternative biodiesel chains. Ann.
Oper. Res. 1–16.
Díaz-Guevara, D.C., 2015. Modelado y simulación del microclima en un invernadero
para el estudio de métodos de calefacción pasivos adaptables a las zonas climáticas
124
References
de Colombia. Universidad Nacional de Colombia–Sede Bogotá.

Dißmann, J., Brechmann, E.C., Czado, C., Kurowicka, D., 2013. Selecting and
estimating regular vine copulae and application to financial returns. Comput. Stat.
Data Anal. 59, 52–69.
Domingues, A.R., Marques, P., Garcia, R., Freire, F., Dias, L.C., 2015. Applying
Multi-Criteria Decision Analysis to the Life-Cycle Assessment of vehicles. J.
Clean. Prod. 107, 749–759.
Drouet, J.-L., Capian, N., Fiorelli, J.-L., Blanfort, V., Capitaine, M., Duretz, S.,
Gabrielle, B., Martin, R., Lardy, R., Cellier, P., Soussana, J.-F., 2011. Sensitivity
analysis for models of greenhouse gas emissions at farm level. Case study of N2O
emissions simulated by the CERES-EGC model. Environ. Pollut. 159, 3156–
3161.
Dufossé, K., Gabrielle, B., Drouet, J.-L., Bessou, C., 2013. Using Agroecosystem
Modeling to Improve the Estimates of N2O Emissions in the Life-Cycle
Assessment of Biofuels. Waste and Biomass Valorization 4, 593–606.
Durante, F., & Sempi, C. (2010). Copula theory: an introduction. In Copula theory
and its applications (pp. 3-31). Springer, Berlin, Heidelberg.
Ecoinvent, 2007. Ecoinvent Data v. 2.0: Ecoinvent Reports No. 1–25. Swiss Center
for Life Cycle Inventory: Dübendorf, Switzerland.
Emmenegger, M., Reinhard, J., Zah, R., 2009. Sustainability Quick Check for
Biofuels–intermediate background report. With Contrib. from T. Ziep, R.
Weichbrodt, Prof. Dr. V. Wohlgemuth, FHTW Berlin A, Roches, R. Freiermuth
Knuchel, Dr. G Gaillard, Agroscope Reckenholz-Tänikon, Düendorf,
Switzerland.
Fan, X.-R., Kang, M.-Z., Heuvelink, E., de Reffye, P., Hu, B.-G., 2015. A knowledge-
and-data-driven modeling approach for simulating plant growth: A case study on
tomato growth. Ecol. Modell. 312, 363–373.
Filzmoser, P., Garrett, R.G., Reimann, C., 2005. Multivariate outlier detection in
exploration geochemistry. Comput. Geosci. 31, 579–587.
Frere, M., Popov, G.F., 1979. Agrometeorological crop monitoring and forecasting.
FAO Plant Protection Paper, No. 17, UN/FAO, Rome, 63 pp.
Friis-Hansen, E., 2004. Concepts and experiences with demand driven advisory
services: Review of recent literature with examples from Tanzania. DIIS Working
paper.
Frischknecht, R., Jungbluth, N., Althaus, H.J., Hischier, R., Doka, G., Dones, R.,
Heck, T., Hellweg, S., Wernet, G., Nemecek, T., 2007. Overview and
methodology. Data v2. 0. Ecoinvent report No. 1, Swiss Centre for LCI, Düendorf,
Switzerland.
Génermont, S., Cellier, P., 1997. A mechanistic model for estimating ammonia
volatilization from slurry applied to bare soil. Agric. For. Meteorol. 88, 145–167.
Gil, R., Bojacá, C.R., Rodríguez, M.A., 2015. Adaptation of a leaf wetness duration
model for tomato under Colombian greenhouse conditions. Agron. Colomb. 33,
11–19.
Gil, R., Bojacá, C.R., Schrevens, E., 2019a. Datasets of the environmental factors and
management practices of the smallholder tomato production systems in the
125
References
Colombian Andes. Data Br. 24, 103844.

Gil, R., Bojacá, C.R., Schrevens, E., 2019b. Understanding the heterogeneity of
smallholder production systems in the Andean tropics – The case of Colombian
tomato growers. NJAS - Wageningen J. Life Sci. 88, 1–9.
Gil, R., Bojacá, C.R., Schrevens, E., 2018. Uncertainty on nitrogen emission data of
smallholder greenhouse tomato growers in the Andean tropics of Colombia, Acta
Horticulturae.
Gil, R., Bojacá, C.R., Schrevens, E., 2017a. Uncertainty of the Agricultural Grey
Water Footprint Based on High Resolution Primary Data. Water Resour. Manag.
31.
Gil, R., Bojacá, C.R., Schrevens, E., 2017b. Environmental savings in tomato
production under optimal agrochemicals management: a modeling approach. Acta
Hortic. 1154, 137–144.
Gil, R., Bojacá, C.R., Schrevens, E., 2017c. A tailor-made crop growth model for the
tomato production systems in Colombia. Agron. Colomb. 35, 301–313.
Giltrap, D.L., Li, C., Saggar, S., 2010. DNDC: A process-based model of greenhouse
gas fluxes from agricultural soils. Agric. Ecosyst. Environ. 136, 292–300.
Gnanadesikan, R., 2011. Methods for Statistical Data Analysis of Multivariate
Observations., Wiley Series in Probability and Statistics. Wiley Series in
Probability and Statistics.
Goglio, P., Smith, W.N., Grant, B.B., Desjardins, R.L., Gao, X., Hanis, K., Tenuta,
M., Campbell, C.A., McConkey, B.G., Nemecek, T., Burgess, P.J., Williams,
A.G., 2018. A comparison of methods to quantify greenhouse gas emissions of
cropping systems in LCA. J. Clean. Prod. 172, 4010–4017.
Grassini, P., van Bussel, L.G.J., Van Wart, J., Wolf, J., Claessens, L., Yang, H.,
Boogaard, H., de Groot, H., van Ittersum, M.K., Cassman, K.G., 2015. How good
is good enough? Data requirements for reliable crop yield simulations and yield-
gap analysis. F. Crop. Res. 177, 49–63.
Greco, S., Ishizaka, A., Matarazzo, B., Torrisi, G., 2017. Regional Studies Stochastic
multi-attribute acceptability analysis (SMAA): an application to the ranking of
Italian regions Stochastic multi-attribute acceptability analysis (SMAA): an
application to the ranking of Italian regions.
Groen, E.A., Heijungs, R., 2017. Ignoring correlation in uncertainty and sensitivity
analysis in life cycle assessment: what is the risk? Environ. Impact Assess. Rev.
62, 98–109.
Gruda, N., 2005. Impact of Environmental Factors on Product Quality of Greenhouse
Vegetables for Fresh Consumption. CRC. Crit. Rev. Plant Sci. 24, 227–247.
Guerrero, A.M., Torres, A., 2016. Estimación de la oferta y demanda hídrica como
base para la planificación de su uso en la producción de tomate en sistemas bajo
invernadero y a campo abierto. MSc thesis in Environmental Sciences.
Universidad Jorge Tadeo Lozano.
Guilpart, N., Grassini, P., Sadras, V.O., Timsina, J., Cassman, K.G., 2017. Estimating
yield gaps at the cropping system level. F. Crop. Res. 206, 21–32.
Guinée, J.B., 2002. Handbook on life cycle assessment operational guide to the ISO
standards. Int. J. Life Cycle Assess. 7, 311.
126
References
Haileslassie, A., Priess, J., Veldkamp, E., Teketay, D., Lesschen, J.P., 2005.
Assessment of soil nutrient depletion and its spatial variability on smallholders’
mixed farming systems in Ethiopia using partial versus full nutrient balances.
Agric. Ecosyst. Environ. 108, 1–16.
Hamdy, A., Ragab, R., Scarascia‐Mugnozza, E., 2003. Coping with water scarcity:
water saving and increasing water productivity. Irrig. Drain. J. Int. Comm. Irrig.
Drain. 52, 3–20.
Hamza, M.A., Anderson, W.K., 2005. Soil compaction in cropping systems: A review
of the nature, causes and possible solutions. Soil Tillage Res. 82, 121–145.
Hasler, K., Bröring, S., Omta, S.W.F., Olfs, H.-W., 2015. Life cycle assessment
(LCA) of different fertilizer product types. Eur. J. Agron. 69, 41–51.
Hatirli, S.A., Ozkan, B., Fert, C., 2006. Energy inputs and crop yield relationship in
greenhouse tomato production. Renew. Energy 31, 427–438.
Havlikova, M., Kroeze, C., 2006. Evaluation of methods for quantifying agricultural
emissions of air, water and soil pollutants. Sci. Total Environ. 372, 133–147.
Hayer, F., Gaillard, G., 2010. Aquatic and terrestrial ecotoxicity as well as human
toxicity characterisation factors for pesticide emissions to soil according to the
methods USES-LCA and EDIP. Agroscope Reckenholz-Tänikon ART, Zürich,
Switz.
Heuts, R., 2018. Technical sustainability of biological production systems. A model
based life cycle assessment of fertilizer management. Ph.D. Thesis. Katholieke
Universiteit Leuven.
Heuts, R., Van Loon, J., Schrevens, E., 2012. Investigating the probability distribution
between global warming and acidification of the pesticide management in
greenhouse tomato production: a regional LCA for the Boyacá department,
Colombia.
Heuvelink, E., 1999. Evaluation of a Dynamic Simulation Model for Tomato Crop
Growth and Development. Ann. Bot. 83, 413–422.
Huijbregts, M.A.J., Breedveld, L., Huppes, G., de Koning, A., van Oers, L., Suh, S.,
2003. Normalisation figures for environmental life-cycle assessment. J. Clean.
Prod. 11, 737–748.
Husak, G.J., Michaelsen, J., Funk, C., 2007. Use of the gamma distribution to
represent monthly rainfall in Africa for drought monitoring applications. Int. J.
Climatol. 27, 935–944.
International Organization for Standardization (ISO), 2006a. Environmental
management—Life cycle assessment—Requirements and Guidelines. ISO14044,
Geneva.
International Organization for Standardization (ISO), 2006b. Environmental
Management—Life Cycle Assessment—Principles and Framework. ISO14040,
Geneva.
Islam, S., Editors, G., Sabev Varbanov, P., Jaromír Klemeš, J., Rafidah Wan Alwi,
S., Yow Yong, J., Liu, X., Ponnambalam, S.G., Loong Lam, H., 2015. An
overview on life cycle inventory leads to green manufacturing: Methods and
modifications A Green Framework for Manufacturing : Product, Process and
System levels View project Multi Robot Formation View project An Overview on
127
References
Life Cycle Inventory Leads .

Jenkinson, D.S., Hart, P.B.S., Rayner, J.H., Parry, L.C., 1987. Modelling the turnover
of organic matter in long-term experiments at Rothamsted. Rothamsted Exp. Stn.
Bulletin, 15 (1987), pp. 1-8
Jiménez, L.A., 2003. Estudio general de suelos y zonificación de suelos en el
departamento de Santander. República de Colombia, Instituto Geográfico Agustín
Codazzi, Subdirección Agrológica.
Jizhang, W., Pingping, L., Hanping, M., 2006. Decision support system for
greenhouse environment management based on crop growth and control cost.
Trans. Chinese Soc. Agric. Eng. 9, 33.
Joe, H., 1997. Multivariate models and dependence concepts. Chapman & Hall.
Jones, C.A., 1986. CERES-Maize; a simulation model of maize growth and
development. Texas University Press.
Jones, J.., Hoogenboom, G., Porter, C.., Boote, K.., Batchelor, W.., Hunt, L..,
Wilkens, P.., Singh, U., Gijsman, A.., Ritchie, J.., 2003. The DSSAT cropping
system model. Eur. J. Agron. 18,
Jones, James W, Antle, J.M., Basso, B., Boote, K.J., Conant, R.T., Foster, I., Godfray,
H.C.J., Herrero, M., Howitt, R.E., Janssen, S., Keating, B.A., Munoz-Carpena, R.,
Porter, C.H., Rosenzweig, C., Wheeler, T.R., 2017. Toward a new generation of
agricultural system data, models, and knowledge products: State of agricultural
systems science. Agric. Syst. 155, 269–288.
Jones, James W., Antle, J.M., Basso, B., Boote, K.J., Conant, R.T., Foster, I., Godfray,
H.C.J., Herrero, M., Howitt, R.E., Janssen, S., Keating, B.A., Munoz-Carpena, R.,
Porter, C.H., Rosenzweig, C., Wheeler, T.R., 2017. Brief history of agricultural
systems modeling. Agric. Syst. 155, 240–254.
Jones, J.W., Dayan, E., Allen, L.H., Van Keulen, H., Challa, H., 1991. A dynamic
tomato growth and yield model (TOMGRO). Trans. ASAE 34, 0663–0672.
Jury, W.A., Horton, R., 2004. Soil physics. John Wiley & Sons.
Kägi, T., Dinkel, F., Frischknecht, R., Humbert, S., Lindberg, J., De Mester, S.,
Ponsioen, T., Sala, S., Schenker, U.W., 2016. Session “Midpoint, endpoint or
single score for decision-making?”—SETAC Europe 25th Annual Meeting, May
5th, 2015. Int. J. Life Cycle Assess. 21, 129–132.
Kalbar, P.P., Birkved, M., Nygaard, S.E., Hauschild, M., 2017. Weighting and
Aggregation in Life Cycle Assessment: Do Present Aggregated Single Scores
Provide Correct Decision Support? J. Ind. Ecol. 21, 1591–1600.
Kayalar, D.E., Küçüközmen, C.C., Selcuk-Kestel, A.S., 2017. The impact of crude
oil prices on financial market indicators: copula approach. Energy Econ. 61, 162–
173.
Kelley, C.T., 1999. Iterative methods for optimization. SIAM Press, Philadelphia.
Kimura, K., Kikuchi, S., Yamasaki, S., 1999. Accurate root length measurement by
image analysis. Plant Soil 216, 117–127.
Kupper, T., Bonjour, C., Zaucker, F., Achermann, B., Menzi, H., Cordovil, C.,
Ferreira, L., 2010. Agrammon: An internet based model for the estimation of
ammonia emissions, in: 14th RAMIRAN International Conference. pp. 334–337.
128
References
Kurowicka, D., Cooke, R., 2006. Uncertainty analysis with high dimensional
dependence modelling. Wiley, Chichester.
Kwong, K.F.K., Bholah, A., Volcy, L., Pynee, K., 2002. Nitrogen and phosphorus
transport by surface runoff from a silty clay loam soil under sugarcane in the
humid tropical environment of Mauritius. Agric. Ecosyst. Environ. 91, 147–157.
Lahdelma, R., Hokkanen, J., Salminen, P., 1998. SMAA - Stochastic multiobjective
acceptability analysis. Eur. J. Oper. Res. 106, 137–143.
Leeuwis, C., 2013. Communication for rural innovation: rethinking agricultural
extension. John Wiley & Sons.
Lei, S., Daniels, J.L., Bian, Z., Wainaina, N., 2011. Improved soil temperature
modeling. Environ. Earth Sci. 62, 1123–1130.
Li, C., Farahbakhshazad, N., Jaynes, D.B., Dinnes, D.L., Salas, W., McLaughlin, D.,
2006. Modeling nitrate leaching with a biogeochemical model modified based on
observations in a row-crop field in Iowa. Ecol. Modell. 196, 116–130.
Liao, W.J., van der Werf, H.M.G., Salmon-Monviola, J., 2014. Modelling of nitrogen
releases in life cycle assessment of crop production., in: Proceedings of the 9th
International Conference on Life Cycle Assessment in the Agri-Food Sector (LCA
Food 2014), San Francisco, California, USA, 8-10 October, 2014. American
Center for Life Cycle Assessment, pp. 718–724.
Licker, R., Johnston, M., Foley, J.A., Barford, C., Kucharik, C.J., Monfreda, C.,
Ramankutty, N., 2010. Mind the gap: how do climate and agricultural
management explain the ‘yield gap’ of croplands around the world? Glob. Ecol.
Biogeogr. 19, 769–782.
Liu, S., Yang, J.Y., Zhang, X.Y., Drury, C.F., Reynolds, W.D., Hoogenboom, G.,
2013. Modelling crop yield, soil water content and soil temperature for a soybean–
maize rotation under conventional and conservation tillage systems in Northeast
China. Agric. Water Manag. 123, 32–44.
Luo, L., Qin, L., Wang, Y., Wang, Q., 2016. Environmentally-friendly agricultural
practices and their acceptance by smallholder farmers in China—A case study in
Xinxiang County, Henan Province. Sci. Total Environ. 571, 737–743.
Marcelis, L.F.M., Elings, A., De Visser, P.H.B., Heuvelink, E., 2009. Simulating
growth and development of tomato crop. Acta Hortic. 821, 101–110.
Massa, D., Incrocci, L., Pardossi, A., Delli Paoli, P., Battilani, A., 2013. Application
of a decision support system for increasing economic and environmental
sustainability of processing tomato cultivated in mediterranean climate. Acta
Hortic. 51–58.
Mavromatis, T., Boote, K.J., Jones, J.W., Irmak, A., Shinde, D., Hoogenboom, G.,
2001. Developing Genetic Coefficients for Crop Simulation Models with Data
from Crop Performance Trials Florida Agric. Exp. Stn., J. Series No R-07163.
Crop Sci. 41, 40–51.
Meucci, A., 2011. A short, comprehensive, practical guide to copulas. Risk Prof.
2011, 22–27.
Myllyviita, T., Leskinen, P., Seppälä, J., 2014. Impact of normalisation, elicitation
technique and background information on panel weighting results in life cycle
assessment. Int. J. Life Cycle Assess. 19, 377–386.
129
References
National Oceanic Atmospheric Administration, 2016. Cold & Warm Episodes by

Season [WWW Document]. URL
http://origin.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ONI_v5.
php
Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., King, K.W., 2005. Soil and
water assessment tool theoretical documentation version 2005. Grassland. Soil
Water Res. Lab. Agric. Res. Serv. Blackl. Res. Center, Texas Agric. Exp. Station.
Texas.
Nelson, D.B., 1990. ARCH models as diffusion approximations. J. Econom. 45, 7–
38.
Nemecek, T., Schnetzer, J., Reinhard, J., 2016. Updated and harmonised greenhouse
gas emissions for crop inventories. Int. J. Life Cycle Assess. 21, 1361–1378.
Nielsen, P.H., Nielsen, A.M., Weidema, B.P., Dalgaard, R., Halberg, N., 2003. LCA
food data base. Representative inventory for resource use and environmental
impact of Danish farms and food products.
Nitschelm, L., Parnaudeau, V., Vertès, F., van der Werf, H.M.G., Corson, M.S.,
Viaud, V., Aubin, J., Walter, C., 2018. Improving Estimates of Nitrogen
Emissions for Life Cycle Assessment of Cropping Systems at the Scale of an
Agricultural Territory. Environ. Sci. Technol. 52, 1330–1338.
Payen, S., Basset-Mens, C., Perret, S., 2015. LCA of local and imported tomato: an
energy and water trade-off. J. Clean. Prod. 87, 139–148.
Pedersen, A., Zhang, K., Thorup-Kristensen, K., Jensen, L.S., 2010. Modelling
diverse root density dynamics and deep nitrogen uptake—A simple approach.
Plant Soil 326, 493–510.
Perrin, A., Basset-Mens, C., Gabrielle, B., 2014. Life cycle assessment of vegetable
products: a review focusing on cropping systems diversity and the estimation of
field emissions. Int. J. Life Cycle Assess. 19, 1247–1263.
Perrin, A., Basset-Mens, C., Huat, J., Gabrielle, B., 2017. The variability of field
emissions is critical to assessing the environmental impacts of vegetables: A Benin
case-study. J. Clean. Prod. 153, 104–113.
Pishgar-Komleh, S.H., Akram, A., Keyhani, A., Raei, M., Elshout, P.M.F.,
Huijbregts, M.A.J., van Zelm, R., 2017. Variability in the carbon footprint of
open-field tomato production in Iran - A case study of Alborz and East-Azerbaijan
provinces. J. Clean. Prod. 142, 1510–1517.
Pizzol, M., Laurent, A., Sala, S., Weidema, B., Verones, F., Koffler, C., 2017.
Normalisation and weighting in life cycle assessment: quo vadis? Int. J. Life Cycle
Assess. 22, 853–866.
Ponisio, L., Ehrlich, P., Ponisio, L.C., Ehrlich, P.R., 2016. Diversification, Yield and
a New Agricultural Revolution: Problems and Prospects. Sustainability 8, 1118.
Prado-Lopez, V., Heijungs, R., 2018. Implementation of Stochastic Multi Attribute
Analysis (SMAA) in Comparative Environmental Assessments. Environ. Model.
Softw. Forthcomin, 223–231.
Prado-Lopez, V., Seager, T.P., Chester, M., Laurin, L., Bernardo, M., Tylock, S.,
2014. Stochastic multi-attribute analysis (SMAA) as an interpretation method for
comparative life-cycle assessment (LCA). Int. J. Life Cycle Assess. 19, 405–416.
130
References
Prenger, J.J., Fynn, R.P., Hansen, R.C., 2002. A comparison of four

evapotranspiration models in a greenhouse environment. Trans. ASAE 45, 1779.
Presidency of the Republic of Colombia, 2016. Acuerdo Final Para la Terminación
del Conflicto y la Construcción de una Paz Estable y Duradera [WWW
Document]. URL http://www.altocomisionadoparalapaz.gov.co/ (accessed
3.8.17).
Quemada, M., Cabrera, M.L., 1995. CERES-N model predictions of nitrogen
mineralized from cover crop residues. Soil Sci. Soc. Am. J. 59, 1059–1065.
R Core Team, R.F. for S.C., 2018. R: A Language and Environment for Statistical
Computing.
Rahn, C.R., Zhang, K., Lillywhite, R., Ramos, C., Doltra, J., de Paz, J.M., Riley, H.,
Fink, M., Nendel, C., Thorup-Kristensen, K., Pedersen, A., Piro, F., Venezia, A.,
Firth, C., Schmutz, U., Rayns, F., Strohmeyer, K., 2010. EU-Rotate_N–a decision
support system–to predict environmental and economic consequences of the
management of nitrogen fertiliser in crop rotations. Eur. J. Hortic. Sci. 75, 20–32.
Reeb, C.W., Venditti, R., Gonzalez, R., Kelley, S., 2016. Environmental LCA and
Financial Analysis to Evaluate the Feasibility of Bio-based Sugar Feedstock
Biomass Supply Globally: Part 2. Application of Multi-Criteria Decision-Making
Analysis as a Method for Biomass Feedstock Comparisons. BioResources 11,
6062–6084.
Reza, B., Sadiq, R., Hewage, K., 2014. Emergy-based life cycle assessment (Em-
LCA) of multi-unit and single-family residential buildings in Canada. Int. J.
Sustain. Built Environ. 3, 207–224.
Richards, L.A., 1941. A pressure-membrane extraction apparatus for soil solution.
Soil Sci 51, 377–386.
Richardson, C.W., Wright, D.A., 1984. WGEN: A model for generating daily weather
variables. US Department of Agriculture, Agricultural Research Service
Washington, DC.
Robertson, R., Nelson, G., Thomas, T., Rosegrant, M., 2013. Incorporating Process-
Based Crop Simulation Models into Global Economic Analyses. Am. J. Agric.
Econ. 95, 228–235.
Rosenzweig, C., Elliott, J., Deryng, D., Ruane, A.C., Müller, C., Arneth, A., Boote,
K.J., Folberth, C., Glotter, M., Khabarov, N., Neumann, K., Piontek, F., Pugh,
T.A.M., Schmid, E., Stehfest, E., Yang, H., Jones, J.W., 2014. Assessing
agricultural risks of climate change in the 21st century in a global gridded crop
model intercomparison. Proc. Natl. Acad. Sci. 111, 3268–3273.
Ruben, R., Pender, J., 2004. Rural diversity and heterogeneity in less-favoured areas:
The quest for policy targeting. Food Policy 29, 303–320.
https://doi.org/10.1016/j.foodpol.2004.07.004
Sánchez-Navarro, D., Lis-Gutiérrez, J.., Campo-Robledo, J., Herrera-Saavedra, J.P.,
2013. Estudio sobre fertilizantes en Colombia.
Sándor, R., Fodor, N., 2012. Simulation of soil temperature dynamics with models
using different concepts. Sci. World J. 2012.
Saxton, K.E., Rawls, W., Romberger, J.S., Papendick, R.I., 1986. Estimating
generalized soil-water characteristics from texture 1. Soil Sci. Soc. Am. J. 50,
131
References
1031–1036.
Schepsmeier, U., Stoeber, J., Brechmann, E.C., Graeler, B., Nagler, T., Erhardt, T.,
2018. VineCopula: Statistical inference of vine copulas. R Packag. version 1.
Scholberg, J.M.S., Boote, K.J., Jones, J.W., McNeal, B.L., 1997. Adaptation of the
CROPGRO model to simulate the growth of field-grown tomato. Springer,
Dordrecht, pp. 135–151.
Shaffer, M.J., Ma, L., Hansen, S., 2001. Modeling carbon and nitrogen dynamics for
soil management. CRC Press.
Sim, S.-C., Robbins, M.D., Deynze, A. Van, Michel, A.P., Francis, D.M., 2010.
Population structure and genetic differentiation associated with breeding history
and selection in tomato (Solanum lycopersicum L.). Heredity (Edinb). 106, 927.
Sklar, M., 1959. Fonctions de repartition a n dimensions et leurs marges. Publ. Inst.
Stat. Univ. Paris 8, 229–231.
Smith, J., Gottschalk, P., Bellarby, J., Richards, M., Nayak, D., Coleman, K., Hillier,
J., Flynn, H., Wattenbach, M., Aitkenhead, M., 2010. Model to Estimate Carbon
in Organic Soils–Sequestration and Emissions (ECOSSE). Carbon N. Y. 44, 1–
73.
Soler, C.M.T., Sentelhas, P.C., Hoogenboom, G., 2007. Application of the CSM-
CERES-Maize model for planting date evaluation and yield forecasting for maize
grown off-season in a subtropical environment. Eur. J. Agron. 27, 165–177.
Sonneveld, C., & Voogt, W. (2009). Nutrient Management of Soil Grown Crops. In
Plant Nutrition of Greenhouse Crops. 363-391. Springer, Dordrecht, Netherlands.
Soto, F., Gallardo, M., Giménez, C., Peña-Fleitas, T., Thompson, R.B., 2014.
Simulation of tomato growth, water and N dynamics using the EU-Rotate_N
model in Mediterranean greenhouses with drip irrigation and fertigation. Agric.
Water Manag. 132, 46–59.
Stanghellini, C., 1987. Transpiration of greenhouse crops: an aid to climate
management.
Stöckle, C.O., Kemanian, A.R., Nelson, R.L., Adam, J.C., Sommer, R., Carlson, B.,
2014. CropSyst model evolution: From field to regional to global scales and from
research to decision support systems. Environ. Model. Softw. 62, 361–369.
Team R Core, 2017. R: A language and environment for statistical computing.
Teh, C.B.S., 2006. Introduction to mathematical modeling of crop growth: How the
equations are derived and assembled into a computer model. BrownWalker Press,
Boca Raton, Florida.
Tei, F., Benincasa, P., Guiducci, M., 2002. Critical nitrogen concentration in
processing tomato. Eur. J. Agron. 18, 45–55.
Tittonell, P., Gilller, K., 2013. When yield gaps are poverty traps: The paradigm of
ecological intensification in African smallholder agriculture. F. Crop. Res. 143,
76–90.
Tittonell, P., Muriuki, A., Shepherd, K.D., Mugendi, D., Kaizzi, K.C., Okeyo, J.,
Verchot, L., Coe, R., Vanlauwe, B., 2010. The diversity of rural livelihoods and
their influence on soil fertility in agricultural systems of East Africa - A typology
of smallholder farms. Agric. Syst. 103, 83–97.
132
References
Tittonell, P., Vanlauwe, B., Corbeels, M., Giller, K.E., 2008. Yield gaps, nutrient use
efficiencies and response to fertilisers by maize across heterogeneous smallholder
farms of western Kenya. Plant Soil 313, 19–37.
Tittonell, P., Vanlauwe, B., de Ridder, N., Giller, K.E., 2007. Heterogeneity of crop
productivity and resource use efficiency within smallholder Kenyan farms: Soil
fertility gradients or management intensity gradients? Agric. Syst. 94, 376–390.
Tittonell, P., Vanlauwe, B., Leffelaar, P.A., Rowe, E.C., Giller, K.E., 2005. Exploring
diversity in soil fertility management of smallholder farms in western Kenya.
Agric. Ecosyst. Environ. 110, 149–165.
Torrellas, M., Antón, A., López, J.C., Baeza, E.J., Parra, J.P., Muñoz, P., Montero,
J.I., 2012. LCA of a tomato crop in a multi-Tunnel greenhouse in Almeria. Int. J.
Life Cycle Assess. 17, 863–875.
Tsang, M.P., Bates, M.E., Madison, M., Linkov, I., 2014. Benefits and risks of
emerging technologies: Integrating life cycle assessment and decision analysis to
assess lumber treatment alternatives. Environ. Sci. Technol. 48, 11543–11550.
Valdés-Gómez, H., Gary, C., Brisson, N., Matus, F., 2014. Modelling indeterminate
development, dry matter partitioning and the effect of nitrogen supply in tomato
with the generic STICS crop–soil model. Sci. Hortic. (Amsterdam). 175, 44–56.
Van Diepen, C.A. van, Wolf, J., Van Keulen, H., Rappoldt, C., 1989. WOFOST: a
simulation model of crop production. Soil use Manag. 5, 16–24.
Van Kernebeek, H.R.J., Oosting, S.J., Van Ittersum, M.K., Bikker, P., De Boer,
I.J.M., 2016. Saving land to feed a growing population: consequences for
consumption of crop and livestock products. Int. J. Life Cycle Assess. 21, 677–
687.
Van Keulen, H., Goudriaan, J., Stroosnijder, L., Lantinga, E.A., Van Laar, H.H.,
1992. Crop growth model for water-limited conditions (SUCROS2)., in:
Simulation of Crop Growth for Potential and Water-Limited Production Situations
(as Applied to Spring Wheat). pp. 27–72.
Van Loon, J., 2018. Model-based approaches for fertilizer recommendations and their
environmental impact due to nutrient leaching under variable weather conditions
in open field horticultural production. Ph.D. Thesis. KU Leuven.
Van Zelm, R., Huijbregts, M.A.J., van de Meent, D., 2009. USES-LCA 2.0—a global
nested multi-media fate, exposure, and effects model.
Vanlauwe, B., Chianu, J., Giller, K.E., Merckx, R., Mokwunye, U., Pypers, P.,
Shepherd, K.D., Smaling, E.M.A., Woomer, P.L., Sanginga, N., 2010. Integrated
soil fertility management: operational definition and consequences for
implementation and dissemination. Outlook Agric. 39, 17–24.
Vansteenkiste, J., Van Loon, J., Garré, S., Pagès, L., Schrevens, E., & Diels, J. (2014).
Estimating the parameters of a 3-D root distribution function from root
observations with the trench profile method: case study with simulated and field-
observed root data. Plant and soil, 375(1-2), 75-88.
Vanthoor, B.H.E., de Visser, P.H.B., Stanghellini, C., van Henten, E.J., 2011. A
methodology for model-based greenhouse design: Part 2, description and
validation of a tomato yield model. Biosyst. Eng. 110, 378–395.
Varadhan, R., Borchers, H.W., 2016. dfoptim: Derivative-free optimization. R
133
References
package version 2016.7-1.

Villagrán, E.A., Baeza-Romero, E.J., Bojacá, C.R. 2019. Transient CFD analysis of
the natural ventilation of three types of greenhouses used for agricultural
production in a tropical mountain climate. Biosyst. Eng. 188, 288–304.
Wall, P.C., 2007. Tailoring Conservation Agriculture to the Needs of Small Farmers
in Developing Countries. J. Crop Improv. 19, 137–155.
https://doi.org/10.1300/J411v19n01_07
Wallach, D., 2006. Evaluating crop models, in: Wallach, D., Makowski, D., Jones,
J.W. (Eds.), Working with Dynamic Crop Models. Elsevier, Amsterdam, pp. 11–
53.
Wang, S., Boulard, T., Haxaire, R., 2000. Measurement and analysis of air speed
distribution in a naturally ventilated greenhouse, in: International Conference and
British-Israeli Workshop on Greenhouse Techniques towards the 3rd Millennium
534. pp. 277–284.
Wang, X., Zou, C., Zhang, Y., Shi, X., Liu, J., Fan, S., Liu, Y., Du, Y., Zhao, Q., Tan,
Y., Wu, C., Chen, X., 2018. Environmental impacts of pepper (Capsicum annuum
L) production affected by nutrient management: A case study in southwest China.
J. Clean. Prod. 171, 934–943.
Weber, C., McCann, L., 2015. Adoption of Nitrogen-Efficient Technologies by U.S.
Corn Farmers. J. Environ. Qual. 44, 391–401.
Wilkerson, G.G., Jones, J.W., Boote, K.J., Ingram, K.T., Mishoe, J.W., 1983.
Modeling soybean growth for crop management. Trans. ASAE 26, 63–73.
Willigen, P. de, 2000. An analysis of the calculation of leaching and denitrification
losses as practised in the NUTMON approach. Rapp. Res. Int.
Xia, Y., Yan, X., 2011. Life-cycle evaluation of nitrogen-use in rice-farming systems:
implications for economically-optimal nitrogen rates. Biogeosciences 8, 3159–
3168.
Yemefack, M., Rossiter, D.G., Njomgang, R., 2005. Multi-scale characterization of
soil variability within an agricultural landscape mosaic system in southern
Cameroon. Geoderma 125, 117–143.
Zanghelini, G.M., Cherubini, E., Soares, S.R., 2018. How Multi-Criteria Decision
Analysis (MCDA) is aiding Life Cycle Assessment (LCA) in results
interpretation. J. Clean. Prod. 172, 609–622.
Zhang, M., Alva, A.K., Li, Y.C., Calvert, D. V, 1996. Root distribution of grapefruit
trees under dry granular broadcast vs. fertigation method. Plant Soil 183, 79–84.
Zhang, W., Dou, Z., He, P., Ju, X.-T., Powlson, D., Chadwick, D., Norse, D., Lu, Y.-
L., Zhang, Y., Wu, L., 2013. New technologies reduce greenhouse gas emissions
from nitrogenous fertilizer in China. Proc. Natl. Acad. Sci. 110, 8375–8380.
134
Appendix 1
Appendix 1. Observed and simulated probability density functions for the environmental indicators of the two tomato production systems under
consideration: greenhouse and open field.
Greenhouse production system
135
Appendix 1
Open field production system
136
Appendix 2
Appendix 2. Tree structure showing the edge and node labels for the ten-dimensional R-vine fitted to the environmental indicators for each production
system. Edge labels indicate the pair-copula family abbreviations (1) and Kendal t correlation coefficient. At each node label, the environmental indicators
(2)
conditioned set is presented before the ‘;’ and the conditioning set is shown after the ‘;’.
Greenhouse production system
137
Appendix 2
138
Appendix 2
Open field production system
139
Appendix 2
140
Appendix 2
(1) (2)
Pair-copula family abbreviations Environmental indicators abbreviations
BB8_90: rotated BB8 copula (90 degrees) AD: abiotic depletion
C90: rotated Clayton copula (90 degrees) AP: acidification potential
C270: rotated Clayton copula (270 degrees) EP: eutrophication potential
F: Frank copula FAETP: fresh-water aquatic ecotoxicity potential
G: Gumbel copula GWP: global warming potential
G90: rotated Gumbel copula (90 degrees) HTTP: human toxicity potential
G270: rotated Gumbel copula (270 degrees) MAETP: marine aquatic ecotoxicity potential
I: independence copula OLDP: ozone layer depletion potential
J90: rotated Joe copula (90 degrees) POP: photochemical oxidation potential
J270: rotated Joe copula (270 degrees) TEP: terrestrial ecotoxicity potential
N: Gaussian copula
SBB1: Survival BB1 copula
SBB8: Survival BB8 copula
SG: Survival Gumbel copula
SJ: Survival Joe copula
t: Student t copula
Tawn: Tawn type 1 copula
Tawn90: rotated Tawn type 1 copula (90 degrees)
Tawn2: Tawn type 2 copula
Tawn2_90: rotated Tawn type 2 copula (90 degrees)
Tawn2_270: rotated Tawn type 2 copula (270 degrees)
141
Appendix 3
Appendix 3. Soil profile characteristics used to evaluate the scenarios in the two production systems.
System Layer LT pH OC BD Nmin CEC CL LO SA LL FC SAT SWC
Greenhouse 1 10 6.3 1.26 1.42 82 20.5 40.2 30.2 29.6 0.23 0.33 0.35 0.30
2 20 6.3 1.32 1.32 82 20.5 40.2 30.2 29.6 0.21 0.32 0.33 0.28
3 30 5.7 0.50 1.54 49 20.5 50.2 24.2 25.6 0.19 0.27 0.30 0.24
4 30 5.7 0.10 1.65 19 18.0 50.2 23.6 26.2 0.17 0.27 0.30 0.24
Open field 1 10 6.5 1.08 1.46 21.2 11.3 17.1 42.3 40.6 0.15 0.28 0.40 0.32
2 20 6.5 1.08 1.49 21.2 11.3 17.1 42.3 40.6 0.15 0.28 0.32 0.25
3 30 6.5 0.80 1.45 21.2 11.3 17.1 42.3 40.6 0.15 0.28 0.32 0.25
4 30 5.2 0.10 1.36 14.8 11.3 45.2 26.5 28.3 0.12 0.28 0.32 0.25
Where, LT is the soil layer thickness (cm), OC is the organic carbon content (%); BD is the soil bulk density (g cm-3), Nmin is the mineral nitrogen
content (kg ha-1); CEC is cation exchange capacity (cmolc kg-1); CL, LO and SA are the clay, loam and sand content (%), respectively; LL, FC and
SAT are the lower limit, field capacity and saturation (cm3 cm-3), respectively; and, SWC is the initial soil water content (cm3 cm-3).
142
Appendix 4
Appendix 4. Temporal dynamic of tomato yield according to the simulations carried out for the scenarios evaluates in the greenhouse
(A) and open field (B) systems
143
Appendix 5
Appendix 5. Temporal dynamics of nitrogen emission of ammonia (NH3) and

nitrous oxide (N2O) realised to environment for each tomato simulated cycle
in greenhouse and open field.
144
Appendix 6
Appendix 6. Temporal dynamics of the impact categories for each production

system and scenario.
145
Appendix 6
146
Appendix 6
147
Appendix 6
148
Appendix 6
149
Appendix 6
150
Appendix 7
Appendix 7. Graphical representation of correlations between environmental indicators for tomato production under greenhouse
scenarios Cr (a), Fx (b)and Op (c) and open field Cr (d), Fx (e) and Op (f).
151
Appendix 8
Appendix 8. Observed and simulated probability density functions for the environmental indicators of the nitrogen scenarios
evaluated in the two tomato production systems
Greenhouse system, Current N fertilization strategy (Cr)
152
Appendix 8
Greenhouse system, Fixed N-fertilization strategy (Fx)
153
Appendix 8
Greenhouse system, Optimum N-fertilization strategy (Op)
154
Appendix 8
Open field system, Current N-fertilization strategy (Cr)
155
Appendix 8
Open field system, Fixed N-fertilization strategy (Fx)
156
Appendix 8
Open field system, Optimum N-fertilization strategy (Op)
157
Curriculum vitae and list of publications
Rodrigo Gil currently works at the Facultad de Ciencias Naturales e Ingeniería,
Universidad de Bogotá Jorge Tadeo Lozano as Associate professor. He teaches
courses of statistics and research seminars to undergraduate and graduate students.
His main research interests are crop modeling, environmental impact derived from
the agricultural sector and horticultural production systems under protected
conditions.
Degrees
2006 B.Sc. in Agronomy. Faculty of Agronomy, National University of Colombia,
Bogotá. Dissertation title: Etiología y distribución espacio-temporal de la
deformación de hojas de la espinaca (Spinacia oleracea L.) en Cota (Cundinamarca).
[Etiology and spatiotemporal distribution of the deformation in spinach leaves
(Spinacia oleracea L.) in Cota (Cundinamarca)]
2011 M.Sc. in Geomatics. Faculty of Agronomy, National University of Colombia,
Bogotá. Dissertation title: Metodología para el estudio de la variabilidad espacio-
temporal caso de estudio: temperatura dentro de invernaderos naturalmente
ventilados. [Methodology for the study of spatiotemporal variability, case study:
temperature inside naturally ventilated greenhouses].
List of publications
International Journals
Gil, R., Bojacá, C.R., Schrevens, E., 2019. Datasets of the environmental factors and
management practices of the smallholder tomato production systems in the
Colombian Andes. Data Br. 103844.
Gil, R., Bojacá, C.R., Schrevens, E., 2019b. Understanding the heterogeneity of
smallholder production systems in the Andean tropics – The case of Colombian
tomato growers. NJAS - Wageningen J. Life Sci. 88, 1–9.
Gil, R., Bojacá, C.R., Schrevens, E., 2017c. Uncertainty of the Agricultural Grey
Water Footprint Based on High Resolution Primary Data. Water Resour. Manag. ,
31(11), 3389-3400.
Escobar, A., Gil, R., Bojacá, C.R., Jiménez, J. 2012. Modeling the eggs development
of the pest Clavipalpus ursinus (Blanchard) (Coleoptera: Melolonthidae) using a
temperature dependent approach. Insect Science. 19 (6) 657–665.
Bojacá, C. R., Casilimas, H. A., Gil, R., & Schrevens, E. (2012). Extending the input–
output energy balance methodology in agriculture through cluster analysis. Energy,
47(1), 465-470.
Bojacá, C. R., Wyckhuys, K. A., Gil, R., Jiménez, J., & Schrevens, E. (2010).
Sustainability aspects of vegetable production in the peri-urban environment of
Bogotá, Colombia. International Journal of Sustainable Development & World
Ecology, 17(6), 487-498.
Bojacá, C. R., Gil, R., Gómez, S., Cooman, A., & Schrevens, E. (2009). Analysis of
greenhouse air temperature distribution using geostatistical methods. Transactions of
the ASABE, 52(3), 957-968.
Bojacá, C. R., Gil, R., & Cooman, A. (2009). Use of geostatistical and crop growth
modelling to assess the variability of greenhouse tomato yield caused by spatial
temperature variations. Computers and electronics in agriculture, 65(2), 219-227.
National Journal
Gil, R., Bojacá, C.R., Schrevens, E., 2017a. A tailor-made crop growth model for the
tomato production systems in Colombia. Agron. Colomb. 35, 301–313.
Gil, R., Bojacá, C. R., & Rodríguez, M. Á.. 2015. Adaptation of a leaf wetness
duration model for tomato under Colombian greenhouse conditions. Agronomía
Colombiana, 33(1), 11-19.
Villagrán, E. A., Gil, R., Acuña, J. F., & Bojacá, C. R. (2012). Optimization of
ventilation and its effect on the microclimate of a colombian multispan greenhouse.
Agronomía Colombiana, 30(2), 282-288.
Gil, R., Luque, N. Y., & Bojacá, C. R. (2011). Effect of aluminized screens on
greenhouse tomato (Solanum lycopersicum L.) production systems at the high tropical
conditions. Agronomía Colombiana, 29(3), 423-431.
Rodríguez, M., Plaza, G., Gil, R., Chaves, B., & Jiménez, J. (2010). Propuesta de
manejo de arvenses en el cultivo de espinaca (Spinacea oleracea L.) basada en
aspectos agronómicos y socioeconómicos. [Proposal for weed control in spinach
crops (Spinacia oleracea L.) based on agronomic and socioeconomic aspects].
Agronomía Colombiana, 28(1).
Niño, N. E., Espinosa, L., Gil, R., Menza, G., & Jiménez, J. A. (2009). Enfermedades
de la espinaca (Spinacia oleracea L.) en Cota (Cundinamarca) y control del mildeo
velloso (Peronospora farinosa, Byford). [Diseases of spinach (Spinacia oleracea L.)
in Cota (Cundinamarca) and downy mildew management (Peronospora farinosa,
Byford)]. Revista Colombiana de Ciencias Hortícolas, 3(2), 161-174.
Gil, R., Smith, A., Chaves, B., Wyckhuys, K., Forero, C., & Jiménez, J. (2009).
Combined efficacy assessment of soil solarization and bio-fungicides for management
of Sclerotinia spp. in lettuce (Lactuca sativa L.). Agronomía Colombiana, 27(2), 193-
201.
Rodríguez, M., Plaza, G., Gil, R., Chaves, B., & Jiménez, J. (2008). Reconocimiento
y fluctuación poblacional arvense en el cultivo de espinaca (Spinacea oleracea L.)
para el municipio de Cota, Cundinamarca. [Recognition and population dynamics of
weeds in spinach crop (Spinacea oleracea L.) for the municipality of Cota,
Cundinamarca]. Agronomía Colombiana, 26(1), 87-96.
Gil, R., Carrillo, D., & Jiménez, J. G. (2007). Determinación de las principales plagas
de la espinaca (Spinacia oleracea) en Cota, Colombia. [Determination of the main
pests of spinach (Spinacia oleracea) in Cota, Colombia]. Revista Colombiana de
entomología, 33(2), 124-129.
Proceedings
Gil, R., Guerrero, A., Casas-Diaz, A.V., Bojacá, C.R. and Schrevens, E. (2019).
Availability of water resources for tomato production in Colombia: a basin
modeling approach. Acta Hortic. 1253, 429-436
Gil, R., Bojacá, C.R., Schrevens, E., 2018. Uncertainty on nitrogen emission data of
smallholder greenhouse tomato growers in the Andean tropics of Colombia. Acta
Hort. 1205, 865-870.
Gil, R., Bojacá, C.R., Schrevens, E., 2017. Environmental savings in tomato
production under optimal agrochemicals management: A modeling approach. Acta
Hort. 1154, 137-144.
159
Molina, J. F., Gil, R., Bojacá, C., Gómez, F., & Franco, H. (2014). Automatic
detection of early blight infection on tomato crops using a color based classification
strategy. In 2014 XIX Symposium on Image, Signal Processing and Artificial Vision
(pp. 1-5). IEEE.
Baquero, D., Molina, J., Gil, R., Bojacá, C., Franco, H., & Gómez, F. (2014). An
image retrieval system for tomato disease assessment. In 2014 XIX Symposium on
Image, Signal Processing and Artificial Vision (pp. 1-5). IEEE.
Molina, J. F., Gil, R., Bojacá, C., Díaz, G., & Franco, H. (2013). Color and size image
dataset normalization protocol for natural image classification: A case study in tomato
crop pathologies. In Symposium of Signals, Images and Artificial Vision-2013:
STSIVA-2013 (pp. 1-5). IEEE.
Gil, R., Bojacá, C.R., Schrevens, E. and Suay, R. 2012. Analysis of air temperature
distribution inside a cold store by means of geostatistical methods. Acta Hort. 945.
29-37.
Gil, R., Bojacá, C.R., Casilimas, H., Schrevens, E., Suay, R. 2011. Assessment of
sidewall and roof vents opening configurations to improve airflow inside
greenhouses. Acta Hort. 952. 141-146.
Bojacá, C.R., Gil, R., Casilimas, H., Arias, L.A., Schrevens, E. 2012.Modelling the
environmental impact of pesticides sprayed on greenhouse tomatoes: a regional case
study in Colombia. Acta Hort. 957. 61-68.
Gil, R., Bojacá, C. R., & Schrevens, E. (2009). Suitability evaluation of four methods
to estimate leaf wetness duration in a greenhouse rose crop. Acta Hort. 893. 797-804.
Bojacá, C. R., Gil, R., & Schrevens, E. (2009). The greenhouse effect in the high
tropics of Colombia: a modeling approach. Acta Hort. 893. 791-796.
160

Tesis Rodrigo Gil

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tesis Rodrigo Gil

Uploaded by

Copyright:

Available Formats

Rodrigo

NITROGEN EMISSIONS OF TOMATO PRODUCTION IN COLOMBIA

NITROGEN EMISSIONS OF TOMATO PRODUCTION IN COLOMBIA

Prof. Dr. C.R. Bojacá degree of Doctor of Bioscience

Prof. Dr. C.R. Bojacá degree of Doctor of Bioscience

Members of the Examination

© 2019 Rodrigo GIL

estimated the environmental burdens of the GH tomato production system based on

considerations, it has been suggested that uncertainty in the N-emissions estimates

Table 1. Metadata of collected information in open field and greenhouse production

More specific subject Soil fertility, crop management, fertilization strategies

Type of data Tables, figures and text ﬁles.

Data accessibility Data are attached to the article

Value of the Data

inputs and outputs of production systems lacking a proper record keeping

Figure 1. Overview of the greenhouse tomato production system in the province of

Figure 3. Elevation profile of a cross-section (southwest-northeast direction) of the

2.3. Experimental Design, Materials, and Methods

/.1 = w(D1 − Μ){ )* |' (D1 − Μ) for : = 1, … ,  (2)

where Μ is the multivariate arithmetic mean (centroid) and )* is the covariance

3.2. Materials and methods

A preliminary data inspection evidenced unusually high values in some samples,

Nitrate (NO-3 ) mg kg-1 5.13 3.90 94 4.44 2.95 108 0.40

Exchangeable mg kg-1 158.05 127.70 72 219.10 199.98 63 <0.05

pH in 5.09 4.80 15 5.93 5.65 17 <0.05

ECssa dS m-1 0.15 0.11 79 0.65 0.34 115 <0.05

Figure 7. Scaled PCA-biplot showing the variation and correlational structure of

Agronomic efficiency of fertilizer (AEF) use

level of knowledge of grower communities (Wall, 2007). To improve the farmer’s

The C: 12 ⟶ 1 function represents a d-dimensional copula only if the following

Goal and scope definition

Nitrogen fertilization emissions

observations in addition to rainfall measurements from a weather station that was

4.3. Results and discussion

Polypropylene kg 1 4.81 2.83 1 0.41 0.24

Mean Standard Mean Standard

Fertilization Air NH3 0.17 0.1 0.15 0.11

N2O 0.05 0.04 0.02 0.03

Water NO-3 0.6 0.65 0.73 0.78

Fungicides Soil AI 0.047 0.064 0.14 0.11

Water AI 3.6E-05 5.0E-05 7.5E-05 6.1E-05

Insecticides Soil AI 0.022 0.037 0.073 0.082

Water AI 1.2E-05 1.9E-05 2.4E-05 4.1E-05

Life cycle impact assessment

Normalization and weighting

Figure 8. Graphical representation of correlations between environmental indicators

Distribution Parameters1 AIC2 BIC3 Distribution Parameters AIC BIC

Greenhouse 413.73 -733.46 -656.49

Open field 413.73 -716.56 -639.59

Figure 9. Cumulative distribution functions of the rankings for each tomato

agricultural LCAs at a regional scale are to be performed. Low-tech and yield-

5.2. Materials and methods

Dry matter distribution

h^^ = p ^5à (28)

Crop model calibration

Calibration of dry matter partitioning

5.3. Results and Discussion

!"# 0.044 -0.005 0.547 0.251 0.006 -0.012 0.377 0.301

!"$ 51.34 0.006 0.521 - 136.23 0.012 0.543 -

%&'# - - - - 3239.37 0.011 0.735 -

()*+ 1083 0.009 0.668 - 1000.00 0.009 0.668 -

Bias (g DM RMSE (g DM EF Bias (g DM RMSE (g DM EF

Plant 13.77 22.68 0.97 -2.46 61.24 0.91

Stem 0.67 10.66 0.57 0.81 19.89 0.82

/.1 = w(D1 − Μ){ )* |' (D1 − Μ) for : = 1, … , (2)

Variable -̅~ 6 NSE RSR PBias -̅~ 6 NSE RSR PBias

-̅~ : experimental measurements average, 6: model predictions.