83782 MEUNIER 2019 Archivage

Optimal design of photovoltaic water pumping systems
for rural communities – a technical, economic and social

approach
Simon Meunier
To cite this version:

Simon Meunier. Optimal design of photovoltaic water pumping systems for rural communities – a
technical, economic and social approach. Electric power. Université Paris Saclay (COmUE), 2019.
English. �NNT : 2019SACLS440�. �tel-02414400�
HAL Id: tel-02414400

https://theses.hal.science/tel-02414400v1
Submitted on 16 Dec 2019
HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Optimal design of photovoltaic
NNT : 2019SACLS440
water pumping systems for rural

communities – a technical,
economic and social approach
Thèse de doctorat de l'Université Paris-Saclay
préparée à l’Université Paris-Sud
Ecole doctorale n°575 Electrical, optical, bio-physics and engineering, EOBE

Spécialité de doctorat: Génie Electrique
Thèse présentée et soutenue à Gif-sur-Yvette, le 06/12/2019, par
Simon Meunier
Composition du Jury :
Anne Migan-Dubois
Professeur, Université Paris-Sud (GeePs) Présidente du jury
Mohamed Machmoum
Professeur, Université de Nantes (IREENA) Rapporteur
Stéphane Brisset
Professeur, École Centrale de Lille (L2EP) Rapporteur
Frédéric Wurtz
Directeur de recherche, CNRS (G2ELAB) Examinateur
Philippe Malbranche
Directeur de l’INES (Institut National de l'Énergie Solaire) Examinateur
Judith Cherni
Research Fellow and Lecturer, Imperial College London (CEP) Examinatrice (Co-encadrante)
Claude Marchand
Professeur, Université Paris-Sud (GeePs) Directeur de thèse
Loïc Quéval
Enseignant-chercheur, CentraleSupélec (GeePs) Invité (Co-encadrant)
Encadrement :
Philippe Dessante
Professeur, CentraleSupélec (GeePs) Co-encadrant
Lionel Vido
Maître de conférences, Université de Cergy-Pontoise (SATIE) Co-encadrant
2
Acknowledgements
I would like to thank Mohamed Machmoum and Stéphane Brisset for carefully reviewing this work and for
their very interesting feedbacks. I also thank Anne Migan-Dubois, Frédéric Wurtz, Philippe Malbranche and
Judith Cherni for examining this work and for their very valuable remarks.
I wish to express my sincere gratitude to my five PhD advisors. I thank Loïc Quéval for his commitment,
his daily availability, his constructive suggestions and his professionalism. He pushed me to do my best
during these last three years. I thank my PhD director Claude Marchand who was present and active
throughout the thesis. His commitment and support allowed me to carry out my work in the best conditions.
I also would like to thank Judith Cherni for her continuous support and her guidance on socio-economic
aspects. I thank Philippe Dessante for his very constructive feedbacks and guidance throughout the whole
PhD thesis. I thank Lionel Vido for his never-ending enthusiasm and knowledge filled inputs.
I am extremely grateful to Bernard Multon, Arouna Darga, Peter Kitanidis and Dale Manning for their
selfless help and their extremely valuable inputs on systemic approaches, industrial aspects, hydrology and
econometrics. They helped me to grow as a person and as a researcher. I thank Cedric Koeniguer and Bruno
Darracq for their precious guidance on teaching and pedagogical aspects.
I would like to thank Matthias Heinrich, Thomas Vezin and Severin Darga for their great help on data
collection, original scientific contributions as well as for all the exciting and fascinating discussions that we
had. I thank all the students that I had the opportunity to advise during this PhD work for their cheerfulness
and high quality contributions. Special mention to Vitali Caplain, Elvire de la Fresnaye and Vincent Roy
for their exceptional investment. I thank all the 280 crowdfunders of the Turning Sun Into Water project,
the NGO Respublica and the company DargaTech for making the installation of the photovoltaic water
pump in Gogma possible. I thank the inhabitants of Gogma and Basile Darga for their warm welcome and
a life-changing experience.
I thank all the teachers and researchers of the laboratory GeePs and of the energy department for their
welcome and for all the technical and friendly discussions. I do not forget the technical and administrative
team, especially Damien Huchet and Richard Beljio and Sophie Goncalvez, for their daily help.
It is now time to thank all my doctoral friends. We had a lot of fun and I know it is only the beginning of
our friendship. A special mention to my “co-bureau” Mokrane Bala with whom we shared amazing moments
and listened to great Algerian songs. I also thank my great friends from the Tennis Club CentraleSupelec,
with whom we visited the whole Essonne. Special mention to Jose de Freitas for his unconditional support.
My thoughts are also with my friends in France. I thank them for their support and encouragement during
this adventure. I am really glad to have them in my life.
I thank my family and my parents. I am truly grateful for their trust and support during all these years.
Finally, I thank my girlfriend Marine for all the special moments that we spent together and the
unconditional support during difficult times.
3
Table of contents
List of figures ................................................................................................................................................ 7
List of tables .................................................................................................................................................. 9

Abbreviations .............................................................................................................................................. 10
Nomenclature .............................................................................................................................................. 11
Conversion rate............................................................................................................................................ 13
Introduction ................................................................................................................................................. 15
Chapter I Literature review ..................................................................................................................... 17
I.1 The water-energy nexus in rural areas of sub-Saharan Africa .................................................... 18

I.1.1 Domestic water access ............................................................................................................. 18
I.1.2 Electricity access ..................................................................................................................... 19
I.2 Electrified water pumping technologies for off-grid areas .......................................................... 20

I.3 Conventional PVWPS design ...................................................................................................... 22
I.3.1 Architecture ............................................................................................................................. 22
I.3.2 Position .................................................................................................................................... 22

I.3.3 Sizing ....................................................................................................................................... 24
I.4 Literature gaps and research objectives ....................................................................................... 25

Chapter II Experimental setup .............................................................................................................. 28
II.1 Case study village and PVWPS ................................................................................................... 29

II.1.1 The village of Gogma, Burkina Faso....................................................................................... 29
II.1.2 PVWPS .................................................................................................................................... 31
II.2 Data collection ............................................................................................................................. 37

II.2.1 On-field observations .............................................................................................................. 39
II.2.2 GIS mapping............................................................................................................................ 39

II.2.3 Geophysical measures ............................................................................................................. 40
II.2.4 Account books of water sources .............................................................................................. 41

II.2.5 Household surveys................................................................................................................... 41
II.2.6 Pumping test ............................................................................................................................ 43

II.2.7 Water quality data.................................................................................................................... 44
II.2.8 Monitoring of the PVWPS of Gogma ..................................................................................... 45
II.3 Partial conclusion ........................................................................................................................ 49

Chapter III Interdisciplinary model ........................................................................................................ 51
4
III.1 Overview ..................................................................................................................................... 52
III.2 Demand model ............................................................................................................................ 54
III.2.1 Determination of the water sources where the households wish to go after installation of the
PVWPS ……………………………………………………………………………………………...54
III.2.2 Determination of the water demand profile at the PVWPS ................................................. 60
III.3 Technical model .......................................................................................................................... 63

III.3.1 Energy conversion model – presentation and application to the current PVWPS of Gogma
……………………………………………………………………………………………...64
III.3.2 Energy conversion model – generalization ......................................................................... 72
III.3.3 Beneficiaries identification model ....................................................................................... 76
III.4 Impact model ............................................................................................................................... 78

III.4.1 Indicators identification and ranking ................................................................................... 78
III.4.2 Quantification of the indicators for each household ............................................................ 81

III.4.3 Socio-economic impact at the scale of the village............................................................... 82
III.5 Economic model .......................................................................................................................... 84

III.5.1 Model................................................................................................................................... 84
III.5.2 Results ................................................................................................................................. 87
III.6 Partial conclusion ........................................................................................................................ 89

Chapter IV Optimal design..................................................................................................................... 91
IV.1 Optimisation problem .................................................................................................................. 92

IV.2 Analysis of a reference result ...................................................................................................... 92
IV.2.1 Mono-objective optimisations results .................................................................................. 93
IV.2.2 Bi-objective optimisation results ......................................................................................... 96
IV.3 Influence of the error in the demand model output ................................................................... 104
IV.4 Influence of the expression of the socio-economic impact function ......................................... 106
IV.5 Influence of the groundwater parameters .................................................................................. 108
IV.6 Proposition of an improved procedure for the design and installation of PVWPS ................... 111
IV.6.1 Procedure ........................................................................................................................... 111
IV.6.2 Case study.......................................................................................................................... 113
IV.7 Partial conclusion ...................................................................................................................... 116

Conclusion ................................................................................................................................................. 118
List of Publications .................................................................................................................................... 122
Appendix ................................................................................................................................................... 125
Appendix A. Household survey ............................................................................................................ 125
5
Appendix B. Motor-pump model polynomial coefficients – case of the SQFlex 5A-7 ........................ 133
Appendix C. Satellite climatic data ....................................................................................................... 134
Appendix D. Economic survey for local companies ............................................................................. 135
Appendix E. Range of variation of groundwater parameters ................................................................ 139
Extended summary in French .................................................................................................................... 141
Chapitre I : Revue de littérature ............................................................................................................ 141

Chapitre II : Dispositif expérimental ..................................................................................................... 142
Chapitre III : Modèle interdisciplinaire ................................................................................................. 146
Chapitre IV : Conception optimale........................................................................................................ 148
References ................................................................................................................................................. 156
6
List of figures
Figure I-1 – Share of rural population with access to improved water sources, 2015. ............................... 18
Figure I-2 – Share of rural population with access to electricity, 2015 ....................................................... 19
Figure I-3 – Schematic layout of PVWPS with battery storage (left) and water tank storage (right) ......... 22
Figure II-1 – Location of Burkina Faso. .................................................................................................... 29
Figure II-2 – Centre-East region in Burkina Faso. ...................................................................................... 29
Figure II-3 – Household gatherings. ............................................................................................................ 30
Figure II-4 – Open well. .............................................................................................................................. 30
Figure II-5 – Hand pump. ............................................................................................................................ 30
Figure II-6 – Conventional procedure for the design and installation of a PVWPS. .................................. 33
Figure II-7 – Architecture of the PVWPS. .................................................................................................. 35
Figure II-8 – Picture of the PVWPS of Gogma. .......................................................................................... 35
Figure II-9 – Overview of the PVWPS. .................................................................................................... 35
Figure II-10 – GIS mapping of Gogma. ...................................................................................................... 40
Figure II-11 – Water level in the borehole measured during the step pumping tests. ................................. 44
Figure II-12 – Water level in the borehole measured during the long pumping test ................................... 44
Figure II-13 – Data logger developed for monitoring the PVWPS of Gogma. ........................................... 46
Figure II-14 – Example of data collected by the data logger. ..................................................................... 48
Figure III-1 – Block diagram of the interdisciplinary model. ................................................................... 52
Figure III-2 – Block diagram of the demand model. ................................................................................... 54
Figure III-3 – Demand flow rate profile simulated by the demand model and measured by the data logger
(i.e. flow meter data) – dry season. ............................................................................................................. 61
Figure III-4 – Demand flow rate profile simulated by the demand model and measured by the data logger
(i.e. flow meter data) – wet season. ............................................................................................................. 61
Figure III-5 – Block diagram of the technical model. ................................................................................. 63
Figure III-6 – Block diagram of the energy conversion model. .................................................................. 64
Figure III-7– Definition of the heights. ....................................................................................................... 65
Figure III-8 – Controller model. .................................................................................................................. 67
Figure III-9 – Model of the motor-pump SQFlex 5A-7. ............................................................................. 68
Figure III-10 – Drawdown measured during the pumping tests and simulated by the model. .................... 69
Figure III-11 – Measured model inputs ....................................................................................................... 71
Figure III-12 – Water level in the tank measured and simulated (model output)........................................ 72
Figure III-13 – Base radius 𝑅𝑡 and height 𝐻𝑡, 𝑡 chosen for different tank volumes 𝑉𝑡. ............................. 73
Figure III-14 – Irradiance on the plane of the PV array from satellite data. ............................................... 75
7
Figure III-15 – Ambient temperature from satellite data. ........................................................................... 75
Figure III-16 – Simulated water demand for the position of the current PVWPS....................................... 75
Figure III-17 – Block diagram of the beneficiaries identification model. ................................................... 76
Figure III-18 – Oversized PVWPS. ............................................................................................................. 76
Figure III-19 – Undersized PVWPS. ........................................................................................................... 76
Figure III-20 – Block diagram of the impact model. ................................................................................... 78
Figure III-21 – Theory of change – effect of going to a water source......................................................... 79
Figure III-22 – Block diagram of the economic model. .............................................................................. 84
Figure III-23 – Capital cost of the PV modules as a function of the PV modules peak power. .................. 86
Figure III-24 – Capital cost of the tank as a function of the tank volume. .................................................. 86
Figure IV-1 – Optimal positions of the PVWPS. ........................................................................................ 94
Figure IV-2 – Points of the Pareto front for each of the motor-pump references and deduced final Pareto
front. ............................................................................................................................................................ 97
Figure IV-3 – Pareto front. .......................................................................................................................... 98
Figure IV-4 – Variation of 𝑃𝑝𝑣, 𝑝 and 𝑉𝑡 as a function of the 𝑆𝐸𝐼. ........................................................... 98
Figure IV-5 – Variation of 𝑀𝑃 as a function of the 𝑆𝐸𝐼. ............................................................................ 98
Figure IV-6 – Variation of 𝐿𝑎𝑡 and 𝐿𝑜𝑛 as a function the 𝑆𝐸𝐼. .................................................................. 99
Figure IV-7 – Values of the direct indicators variation along the Pareto front. ...................................... 100
Figure IV-8 – Number of PVWPS consumers along the Pareto front and sources where these consumers
used to go before the installation of the PVWPS. ................................................................................... 101
Figure IV-9 – Capital and operational costs along the Pareto front. ......................................................... 102
Figure IV-10 – Vertical break at a 𝑆𝐸𝐼 of ~0.08. ...................................................................................... 102
Figure IV-11 – Block diagram considered for investigating the influence of the error in the demand model
output. ........................................................................................................................................................ 104
Figure IV-12 – Influence of the error in the demand model output on the Pareto front............................ 105
Figure IV-13 – Pareto fronts obtained from the different expressions of the 𝑆𝐸𝐼 function. ..................... 106
Figure IV-14 – Pareto fronts obtained for different values of the static water level 𝐻𝑏, 𝑠........................ 108
Figure IV-15 – Pareto fronts obtained for different values of the aquifer losses coefficient 𝜅0. .............. 109
Figure IV-16– Pareto fronts obtained for different values of the borehole losses coefficient 𝜇0. ............ 109
Figure IV-17 – Procedure of application of the developed optimisation methodology. ........................... 112
Figure IV-18 – Pareto fronts obtained through the proposed procedure of design and installation of PVWPS.
................................................................................................................................................................... 114
8
List of tables
Table I-1 – Comparison between diesel and photovoltaic energy for water extraction. ............................. 20
Table I-2 – GHG emissions from PVWPS and their GHG emissions mitigation potential. ....................... 20
Table I-3 – PVWPS optimal sizing in the literature. ................................................................................... 24
Table II-1 – Features of the current PVWPS of Gogma.............................................................................. 36
Table II-2 – Summary of collected data. ..................................................................................................... 38
Table II-3 – Water sources costs. ................................................................................................................ 41
Table II-4 – Themes covered by the household survey. .............................................................................. 42
Table II-5 – Results of bacteriological tests performed by Laboratoire Aina ............................................. 44
Table II-6 – Data monitored by the data logger. ......................................................................................... 46
Table III-1 – Quantification of predictors in the case of Gogma. ............................................................... 55
Table III-2 – Multicollinearity matrix for Gogma. ...................................................................................... 57
Table III-3 – Values of the regression coefficients and of the prediction rate before installation of the
PVWPS – case of Gogma. .......................................................................................................................... 58
Table III-4 – Four possibilities for the prediction result at the current PVWPS of Gogma. ....................... 59
Table III-5 – Prediction success rates for several combinations of predictors. ........................................... 59
Table III-6 – Daily integrals of the simulated and measured demand profiles for both seasons................. 61
Table III-7 – Parameters of the energy conversion model and value for the current PVWPS of Gogma. .. 70
Table III-8 – Causal links between indicators of the theory of change. ...................................................... 80
Table III-9 – Direct indicators. .................................................................................................................... 81
Table III-10 – Lifetime of the PVWPS and of its components. .................................................................. 85
Table III-11 – Fixed capital cost breakdown. .............................................................................................. 86
Table III-12 – Capital, operational and life-cycle costs - current PVWPS of Gogma. .............................. 88
Table IV-1 – Boundary of the optimisation variables – Case of Gogma. ................................................... 93
Table IV-2 – Results of the mono-objective optimisation which aims at maximising the 𝑆𝐸𝐼. ................. 94
Table IV-3 – Demanders and consumers of the PVWPS and impact indicators......................................... 95
Table IV-4 – Results of the mono-objective optimisation which aims at minimising the 𝐿𝐶𝐶, with the
constraint 𝑆𝐸𝐼 > 0. ....................................................................................................................................... 96
Table IV-5 – Considered expressions of the 𝑆𝐸𝐼 function. ....................................................................... 106
Table IV-6 – Value of the groundwater parameters for the sensitivity analysis – case of Gogma. .......... 108
Table IV-7 – Range of variation of the optimal position of the PVWPS for different values of the
groundwater parameters. ........................................................................................................................... 110
9
Abbreviations
ISAG Institut Superieur d’Application des Géosciences
GIS Geographic information system
GHG Greenhouse gases
PA Pipe assembly
PV Photovoltaic
PVWPS Photovoltaic water pumping system
STC Standard test conditions
std Standard deviation
WASH Water, sanitation and hygiene
10
Nomenclature
𝐴𝑂𝐼(𝑡) Angle of incidence between the sun’s rays and the PV array (rad)
𝑏(𝑡) Triggering signal from the controller
𝐶𝐴𝑃𝐸𝑋 Capital cost ($)
𝑑𝑟 Discount rate (%)
𝐺𝑑ℎ (𝑡) Diffuse horizontal irradiance (W/m2)
𝐺𝑑𝑛 (𝑡) Direct normal irradiance (W/m2)
𝐺𝑔ℎ (𝑡) Global horizontal irradiance (W/m2)
𝐺𝑝𝑣 (𝑡) Irradiance on the plane of the PV array (W/m2)
𝐻𝑏 (𝑡) Height between the ground level and the water level in the borehole (m)
𝐻𝑏,𝑠 Height between the ground level and the static water level in the borehole (m)
𝐻𝑏,𝑑 (𝑡) Height between the static water level and the water level when there is pumping (drawdown) (m)
𝐻𝑡,𝑏 Height between the ground level and the bottom of the tank (m)
𝐻𝑡,𝑖 Height between the bottom of the tank and the level at which water enters the tank (m)
𝐻𝑡,𝑟 Height between the bottom of the tank and the restart level (m)
𝐻𝑡,𝑠 Height between the bottom of the tank and the stop level (m)
𝐻𝑡,𝑡 Height between the bottom of the tank and the top of the tank (m)
𝐻𝑝𝑎1 (𝑡) Additional head due to pressure losses in pipe assembly (PA) 1 (m)
𝐻𝑡 (𝑡) Height between the bottom of the tank and the water level in the tank (water level in the tank) (m)
ℎ Household
𝒉 All the households of the village
𝒉𝒔𝒖𝒓 Households surveyed
𝒉𝒔𝒃𝒆𝒇 Water sources where the households go before installation of the PVWPS
𝒉𝒔∗𝒂𝒇𝒕 Water sources where the households wish to go after installation of the PVWPS
𝒉𝒔𝒂𝒇𝒕 Water sources where the households effectively go after installation of the PVWPS
𝐼𝑝𝑣 (𝑡) Current from the PV array (A)
𝑘𝑚,𝑛 Coefficients for the fitting of the motor-pump characteristic
𝐿𝑎𝑡 Latitude
𝐿𝑜𝑛 Longitude
𝐿𝐶𝐶 Life-cycle cost
𝑀𝑃 Motor-pump reference
𝑁𝑂𝐶𝑇 Nominal operating cell temperature (°C)
𝑛𝑣 Number of households in the village
𝑛𝑑 Number of households that wish to go to the PVWPS (number of demanders of the PVWPS)
𝑛𝑐 Number of households that consume water at the PVWPS (number of consumers of the PVWPS)
𝑛𝑐,ℎ𝑝 Number of households that consume water at the PVWPS and used to consume water at a hand pump
before installation of the PVWPS
𝑛𝑐,𝑜𝑤 Number of households that consume water at the PVWPS and used to consume water at an open well
before installation of the PVWPS
𝑂𝑃𝐸𝑋 Operational cost ($)
𝑃𝑝𝑣,𝑝 PV array peak power (Wp)
𝑃𝑝𝑣 (𝑡) Power produced by the PV array (W)
11
𝑃𝑚𝑎𝑥 (𝑀𝑃) Maximum power input to the motor-pump reference 𝑀𝑃
𝑄𝑑 (𝑡) Water demand flow rate (m3/s)
𝑄𝑐 (𝑡) Water consumption flow rate (m3/s)
𝑄𝑝 (𝑡) Pump flow rate (m3/s)
𝑄𝑝,𝑚𝑎𝑥 Maximum flow rate that can be pumped from a borehole (m3/s)
𝑟 Rate of satisfaction of the water demand
𝑠 Water source
𝑆𝑡 Area of the base of the tank (m2)
𝑆𝐸𝐼 Socio-economic impact
𝑡 Time
𝑇𝑎 (𝑡) Ambient temperature (°C)
𝑇𝑝𝑣 (𝑡) PV modules temperature (°C)
𝑇𝐷𝐻(𝑡) Total dynamic head (m)
𝑉𝑝𝑣 (𝑡) Voltage of the PV array (V)
𝑉𝑡 Tank volume (m3)
𝑉𝑐 Volume consumed at the PVWPS (m3)
𝑉𝑑 Volume demanded at the PVWPS (m3)
𝑤𝑖 Weighting coefficient of the socio-economic impact function
𝑦𝑂𝑃𝐸𝑋 Yearly operational cost ($)
𝛼 Coefficient relating two impact indicators
𝛽𝑖 Regression coefficient of the demand model
𝑠𝑡𝑐 Efficiency of the PV modules in standard test conditions (STC) (%)
𝜙𝑏𝑒𝑓 Prediction rate before installation of the PVWPS
𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆 Prediction rate at the PVWPS
𝛾 Coefficient of loss on the maximum power related to PV modules temperature (°C-1)
ℐ Impact indicator
ℐ𝑑 Distance impact indicator
ℐ𝑑𝑖𝑎 Diarrhea impact indicator
ℐ𝑒𝑒 Extraction easiness impact indicator
ℐ𝑤𝑐 Water cost impact indicator
ℐ𝑤𝑞 Water quality impact indicator
𝑛 Aquifer losses coefficients (s/m2)
𝜆 Azimuth of the PV array (rad)
µ𝑛 Borehole losses coefficients (s2/m5)
 Pipe pressure losses coefficient (s2/m5)
𝜃 Tilt of the PV array (rad)
𝜋𝑖 Predictor of the water demand
𝜋𝑑 Distance predictor
𝜋𝑒𝑒 Extraction easiness predictor
𝜋𝑝𝑤𝑞 Perceived water quality predictor
𝜋𝑡𝑐𝑤 Time collecting water predictor
𝜋𝑤𝑐 Water cost predictor
12
Π Number of predictors of the water demand
𝜌 Probability that a given household attends a given water source
𝜎 Season (dry or wet)
𝜁 Albedo of the surrounding environment
Conversion rate
All the prices are expressed in US dollars ($) and the rates retained are the ones on the 23/08/2018:
1 XOF (currency in Burkina Faso) = $0.00177
1 € = $1.16
13
Introduction
14
Introduction
In sub-Saharan Africa, more than 300 million people do not have access to potable water sources and most
of them live in rural areas. Photovoltaic water pumping systems (PVWPS) can improve access to potable
water, especially in rural villages that are not connected to the electricity grid.
The objective of this PhD thesis is to propose a new methodology for the optimal design of PVWPS for
domestic water access in rural villages. The proposed methodology aims at determining the sizings of the
PVWPS and its positions in the village that maximize the positive impact of the system on socio-economic
development (e.g. use of water of better quality, decrease in the distance to collect water) and minimize its
life-cycle cost. The first main originality is the introduction of the position of the PVWPS in the village as
an optimisation variable. This is particularly relevant given that many rural villages in sub-Saharan Africa
are extended and that households of the same village often have an uneven access to potable water sources.
Therefore there may be positions of the PVWPS that are more favourable for the village as a whole. The
second main originality is the inclusion of the socio-economic impact as an objective function of the
optimisation. Indeed, institutions that finance these systems aim at maximising their positive socio-
economic impact.
This methodology was developed in collaboration with researchers from various disciplines, i.e., electrical
engineering (GeePs and SATIE laboratories), environmental policy (Imperial College London),
econometrics (Colorado State University) and hydrology (Stanford University), and in association with the
company DargaTech based in Burkina Faso. Besides, this methodology is applied to a rural village in
Burkina Faso, where technical and socio-economic data have been collected since September 2017.
Chapter I presents the literature review. Chapter II describes the case study village and the experimental
data collected. Chapter III presents the interdisciplinary model. Chapter IV presents the formulation of the
optimisation problem and the results and proposes an improved procedure for the design and installation of
PVWPS.
15
Chapter I Literature review
16
Chapter I Literature review
In the first section of this chapter, we discuss the low access to improved domestic water sources and the
low grid coverage in sub-Saharan Africa. In the second section, we compare the main energy sources for
pumping domestic water in off-grid areas. In the third section, we present the conventional way to design
PVWPS. In the fourth section, we detail the main gaps that had been identified in the literature reviewed in
the previous sections and relate those to the objectives of this PhD thesis.
17
Chapter I. Literature review
I.1 The water-energy nexus in rural areas of sub-Saharan Africa

I.1.1 Domestic water access
The most common alternatives for water extraction in poor rural areas of developing countries are gathering
water from an open well with a bucket and a rope, and hand pumps [1]. An open well is a hole sunk by hand
in the ground, about 5-10 meters deep, used to collect water by throwing a bucket into it [1]. A hand pump
is set on a drilled deep borehole and therefore allows to extract groundwater [1].
The water that is extracted from open wells with a bucket and a rope is not potable notably because it is
exposed to contamination through mud [2]. This is the main reason why these sources are categorized as
“unimproved”. Figure I-1 shows that these unimproved sources are nevertheless the only ones available to
a large share of the population in sub-Saharan Africa. Drinking water from unimproved water sources is
responsible for diseases such as diarrhea and trachoma [3, 4]. For instance, according to [4], in Cameroon
children living in a household with no access to potable water are 1.3 times more likely to get diarrheal
diseases than those living in households with an easy access to potable water.
Figure I-1 – Share of rural population with access to improved water sources, 2015 [5].
Contrarily to open wells, hand pumps provide potable water as they extract water from aquifers and they
are sealed to prevent contamination [2]. However, like for open wells, the pumped flow rate of hand pumps
is limited by human strength, water extraction is hard and time consuming, which is responsible for
significant queuing times [6, 7, 8]. Finally, hand pumps do not allow to reach deep aquifers (typically
groundwater levels deeper than 50 m [9]) and require regular maintenance due to moving parts [2].
In comparison to the alternatives described above, electrified water pumping systems appear as more
promising for providing water for domestic use, i.e., drinking, cooking, personal hygiene and laundry.
18
Indeed, despite their higher initial cost [1], they allow to reach deeper aquifers and they provide higher flow
rates. Consequently the queuing time at the system is reduced, which allows to free time so to enable people
to undertake other activities [10, 11]. In addition, electrified water pumping systems allow to lower the
physical hardness of water collection [10, 11]. Finally, they can be a stepping stone toward the installation
of piping systems that deliver water to households individually [12, 13]. Nevertheless, as electrified
pumping systems provide the opportunity to extract larger volumes of water than hand pumps, increased
attention has to be paid to the effect of pumping on groundwater resources in order to preserve their
sustainability [9].
I.1.2 Electricity access

Figure I-2 presents the share of the population of each country with access to electricity in 2015. A close
observation to Figure I-1 and Figure I-2 points out that, in general, countries where access to improved water
sources is the lowest are also the ones where the electricity access is deficient. In addition, national grid
extension in the affected countries has proved to be too slow to reach remote rural areas in the near future
[14, 15]. Other energy sources, such as off-grid solar energy systems, are a viable alternative to provide
energy services in rural areas [16], and in particular electrified water pumping.
0 – 43 %
43 – 91 %
91 – 100 %
Figure I-2 – Share of rural population with access to electricity, 2015 [17].
19
I.2 Electrified water pumping technologies for off-grid areas

The most often used energy sources for providing electrified water pumping in off-grid areas are
photovoltaic energy and diesel [18]. Table I-1 compares photovoltaic water pumping systems (PVWPS) to
diesel water pumping systems.
Table I-1 – Comparison between diesel and photovoltaic energy for water extraction.
Diesel pumping system PVWPS
Storage Not required Electrical or water storage required
Capital cost Low [1, 19] High [1, 11, 19]
Operation cost High [19] Low [20, 21]
Lifetime Short [19] Long (typically 20 years) [19, 22, 23]
Greenhouse gases High [24, 25] Low [24, 25]
emissions
Maintenance Frequent maintenance required [1, 19] Reduced maintenance needs [19, 26, 27]
Local impact Noise, toxic fumes [28, 29]
Reliability of Intermittent supply in fuels in many Variability of solar resource [30]
supply regions [19]
Despite their higher capital cost [1, 11, 19], PVWPS have become competitive in comparison to diesel
pumps in off-grid rural areas in terms of life cycle cost [31, 32]. In several cases, they are even more
economically viable [19, 33]. However, the high capital cost still represents a challenge for financing
PVWPS [1, 11, 19].
Regarding greenhouse gases (GHG) emissions, life cycle analyses were carried out to evaluate the GHG
emissions from PVWPS and the reduction in GHG emissions achieved when replacing diesel pumps by
PVWPS. Some studies consider only CO2 and other studies also take into account the other greenhouse
gases (e.g. CH4). In the latter case, results are given in kg of CO2 equivalent (CO2eq). In Table I-2, we
present the results of these studies in terms of emissions per kW, as 1 kW is a typical size for a PVWPS for
domestic water access [34, 35].
Table I-2 – GHG emissions from PVWPS and their GHG emissions mitigation potential.
Reference Location Size of the system PVWPS emissions Emissions reduction for the replacement
considered (kW) of a diesel pump by a PVWPS
[25] China 3.4 294 kgCO2/kW/year NC
[24] Algeria 1 NC 4 kgCO2/kW/year
[36] India 1.8 NC 1160 kgCO2/kW/year
[37] Bangladesh 2 NC 500 kgCO2eq/kW/year
[38] Saudi Arabia 20 NC 1200 kgCO2eq/kW/year
NC: not considered.
The figures in Table I-2 can be compared to other figures of merit such as the average yearly GHG emissions
per capita in the OECD countries: 11700 kgCO2eq/capita/year [39]. We observe that, for domestic water
access, the impact of the implementation of new PVWPS and of the replacement of diesel pumps by PVWPS
in terms of GHG emissions is restricted. This is mainly due to the fact that a low power PVWPS (typically
1 kW) can allow to deliver a very high value energy service, i.e. water provision for dozen of households
[34]. We will therefore not consider GHG emissions from PVWPS in this PhD thesis.
Another key advantage of PVWPS in comparison to diesel pumps is their long lifetime [19, 22, 23] and their
reduced maintenance needs [19, 26, 27], which is particularly important in rural areas [40]. Moreover,
20
PVWPS do not emit toxic fumes [29] and are less noisy [28], reducing their effect on the local operator.
PVWPS are therefore a good candidate for development projects, as considered in this thesis. Nonetheless,
thanks to their quicker installation [19], diesel pumps may be a more suitable for emergency and temporary
projects, following earthquakes for instance.
21
I.3 Conventional PVWPS design

We define the design of a PVWPS for domestic water access in a rural village as the determination of its:
 Architecture, i.e., the choice and the disposition of the components.

 Position, i.e., the geographical location in the village.
 Sizing, i.e., the size of the components.
I.3.1 Architecture
In order to be able to fulfil the water needs and to deal with the variability of the climatic conditions, a
storage component is always required in PVWPS. The two mains possibilities for providing storage are:
storing electrical energy produced by the PV array in batteries, or storing pumped water in a tank [31]
(see Figure I-3). However, batteries wear out quickly in the harsh conditions considered (dust, temperature)
and therefore have to be replaced frequently [41]. When batteries are effectively replaced, this increases the
operation cost [42]. In addition and more importantly, batteries are not always replaced in these isolated
areas, which jeopardizes the sustainability of the whole PVWPS [41]. As the architecture with water tank is
more reliable than the one with battery, it is the most commonly used for providing domestic water in
isolated areas [31, 41]. In this thesis, the architecture with water tank will thus be considered.
Figure I-3 – Schematic layout of PVWPS with battery storage (left) and water tank storage (right)
– adapted from [31, 43, 44], MPPT: maximum power point tracker.
I.3.2 Position
Setting the position of a PVWPS, and of a water source in general, can be separated into two main phases
[45]:
 The proposition of a position by decision makers. This is referred to as ‘positioning’ in this thesis.
 The validation or invalidation of the proposed position through geophysical, hydrological and water
quality analyses. This is referred to as ‘position validation’. If the position proposed initially is
invalidated, the decision maker suggests another position (i.e. back to the positioning phase).
22
Positioning
The positioning of PVWPS and of water sources in general is a relatively undocumented topic and no
scientific article on the subject was encountered. The few documents of the grey literature on this topic
mention that the position may be proposed by the financing institution and/or village authorities [46, 47]
and that the following elements should be considered when proposing a position:
1. The position should be away from potential contaminations sites (e.g. latrines, burial sites,
municipal garbage dump) [46].
2. The position should be safe [46].
3. Reports on local geophysical studies and on the characteristics of existing wells of the village should
be examined, if there are any [45].
4. The position should be accessible by the households [46].
Regarding element 3, it is difficult to implement in isolated areas of developing countries due to the scarcity
of geophysical and hydrology data [46, 48]. In addition, an element that is not mentioned is that the new
water point should be installed in the vicinity of households which have the worst access to water (e.g.
households that only have access to open wells).
Finally, we observe that, no support tool is provided for the application of the mentioned elements and
therefore for the positioning of the water source. As a consequence, the decision maker may decide on a
position that is not the most favourable for the village as a whole. This non-optimal choice may be due to
the inability of the decision maker to grasp the whole situation of the village or may be done on purpose by
the decision maker, who may not be acting in the interest of the whole village. A support tool may help the
decision maker to decide on the optimal position for the PVWPS. It may also allow a separate institution to
verify that the position proposed by the decision maker is the most favourable for the whole village.
Position validation
Firstly, geophysical studies are performed to detect the presence of water around the proposed position,
noted (𝐿𝑎𝑡1 , 𝐿𝑜𝑛1 ) [45]. If geophysical studies suggest the presence of water at a position (𝐿𝑎𝑡𝑆 , 𝐿𝑜𝑛𝑆 ),
next to (𝐿𝑎𝑡1 , 𝐿𝑜𝑛1 ), a borehole is drilled at the position (𝐿𝑎𝑡𝑆 , 𝐿𝑜𝑛𝑆 ). If water is indeed encountered at
(𝐿𝑎𝑡𝑆 , 𝐿𝑜𝑛𝑆 ) while drilling, three tests are then performed [45, 49]:
 pumping tests which consist in pumping water at different flow rates while monitoring the water
level in the borehole. The purpose of these tests is to determine the maximum flow rate that can be
extracted from the borehole.
 physico-chemical tests which consist in measuring the physico-chemical parameters of the water,
such as pH, temperature, alkalinity, electrical conductivity, arsenic concentration.
 bacteriological tests which consist in measuring the bacteriological parameters of the water such as
the concentration of thermotolerant coliform and of faecal streptococci.
If the decision maker considers that the maximum pumped flow rate, determined from the pumping tests, is
sufficient to meet the needs of the inhabitants and that the water quality is satisfying, the position is validated
and the PVWPS is installed at (𝐿𝑎𝑡𝑆 , 𝐿𝑜𝑛𝑆 ).
23
I.3.3 Sizing
Several studies have been performed to determine the optimal sizing of PVWPS for domestic water access.
We have summarized the objective function(s), variables and constraint(s) considered in these optimisations
in Table I-3.
Table I-3 – PVWPS optimal sizing in the literature.
Reference Objective function(s) Variables Constraint(s)
[50] Life-cycle cost, PV array peak power, None
probability of not fulfilling the water consumption tank volume
probability of not fulfilling the water consumption, tank volume
excess in pumped water in comparison to the consumption
probability of not fulfilling the water consumption tank volume
[53] Probability of not fulfilling the water consumption PV array peak power, None
tank volume
We observe that previous studies have mostly aimed at minimising the life-cycle cost of the PVWPS [50,
51, 52] and at minimising the probability of not fulfilling the water consumption of the inhabitants [50, 51,
52, 53]. The variables considered are the peak power of the PV array and the volume of the water tank [50,
51, 52, 53]. To our best knowledge, no study considers the motor-pump reference as an optimisation
variable. It may be interesting to do so as the motor-pump is at the centre of the energy conversion chain
(see Figure I-3). Therefore, from now on, we define “sizing” a PVWPS as determining the peak power of
the PV array, the motor-pump reference and the tank volume.
In addition, it is noticeable that the position of the PVWPS in the village is not taken into account in the
above studies on sizing [50, 51, 52, 53]. It is considered that the PVWPS should meet the water consumption
of the whole village, no matter what is its position. This approach may be valid for small villages where all
the inhabitants go to only one water source. However, a large share of sub-Saharan Africa rural villages are
extended (area of several km2) and the inhabitants of the same village go to different water sources [54, 55].
For these villages, it is rational thus to assume that the water demand, i.e. the load curve, to the newly
installed PVWPS depends on its position in the village. This should be reflected in the sizing, which varies
with the load curve at the PVWPS, and should therefore depend on the position of the PVWPS in the village.
24
I.4 Literature gaps and research objectives

We see in section I.3.3 that the positioning and the sizing of the PVWPS should be coupled. This means that
the water demand, i.e. the load curve, at a new PVWPS should be predicted, for any position in the village,
and this predicted water demand should be used for sizing the PVWPS. Some econometrics models have
investigated the water demand in rural villages and the first idea could be to use the output of these models
as an input for sizing the PVWPS. For instance, [56] studied the determinants of the demand for different
water sources (e.g. public sources, private wells) in Sri Lanka and similar studies were performed in Kenya
[55], Honduras [57] and Philippines [58]. However, to our knowledge, current studies do not go as far as
predicting the load curve at water sources. In addition, they focus on existing sources and they do not look
at adding a new source, that may in addition be of a different type (e.g. adding a PVWPS in a village where
there are only open wells). These elements highlight that considering the position of the PVWPS in the
village also requires to adapt existing water demand models.
We also observe that none of the studies on the optimal design of PVWPS seek to maximize the positive
impact on socio-economic development (e.g. use of water of higher quality, decrease in distance to collect
water), while it is the main objective of the institutions that finance these systems [59, 60]. Some studies
have been working on the evaluation of the socio-economic impact of energy systems. For instance, [61]
developed an approach and model which compares several energy technologies for rural electrification over
a wide range of criteria and forecasts social, human, financial and environmental impacts of energy supply
on the population. In addition, [62] proposed a model to evaluate energy planning options (e.g. combined
heat and power plants, energy saving) over technical, economic and social criteria (e.g. consistence of
installation and maintenance requirements with local technical know-how, cost of primary energy saved).
However, we did not find a similar methodology for investigating the socio-economic impact of electrified
water pumping systems, like PVWPS. In addition and more importantly, the above articles on energy
systems do not relate the design variables of the system (e.g. position, size of each component) to the values
of the considered criteria. This prevents to find the optimal design, i.e. the optimal value of the design
variables, vis-à-vis the socio-economic criteria. This also highlights the need to develop models that relate
design variables to the socio-economic impact of the system.
The aim of this PhD thesis is to develop a methodology to determine the PVWPS sizings and positions in a
village that both maximize the positive impact on socio-economic development and minimize the lifecycle
cost of the PVWPS. This is fulfilled in three main steps:
 Chapter II. A PVWPS is designed and installed in the conventional way in a rural village of sub-
Saharan Africa. On the one hand, this allows to understand in detail the current situation regarding
PVWPS and to design a methodology that builds on this current situation. On the other hand, this
allows to collect data to apply the developed methodology and to validate the proposed models.
 Chapter III. We build an interdisciplinary model that links the sizing and position of the PVWPS
to its socio-economic impact and its life-cycle cost. The interdisciplinary model is composed of 4
sub-models: demand, technical, impact and economic.
25
 Chapter IV. We define an optimisation problem to determine the PVWPS sizings and positions
that maximize the positive impact on socio-economic development and minimize the life-cycle cost
of the PVWPS, and then we present the results.
To summarize, in Chapter II, we learn from the design and installation of a PVWPS in the conventional
way. Then, in Chapter III and Chapter IV, we use the knowledge acquired to propose an improved design
and installation methodology (i.e. present how we could have designed and installed the conventionally set
PVWPS more optimally).
26
Chapter II Experimental setup
27
Chapter II Experimental setup
In the frame of this PhD thesis, we raised funds and organized the installation of a PVWPS in the rural
village of Gogma in Burkina Faso. The PVWPS was designed and installed in 2017, following the
conventional way described in section I.3. The first reason for installing this PVWPS was to improve our
understanding of the current situation regarding PVWPS and to propose a methodology for the optimal
design of PVWPS (see Chapter III and Chapter IV) that builds on this current situation. Other reasons were
to gather data for applying the proposed methodology and to be able to compare model results to
experimental measurements.
In section II.1, we present the characteristics of the village of Gogma and the PVWPS installed and in section
II.2 we describe the data collected.
28
Chapter II. Experimental setup
II.1 Case study village and PVWPS

II.1.1 The village of Gogma, Burkina Faso
Burkina Faso is a Sub-Saharan country of West Africa (Figure II-1) with a population of 19 million and a
Human Development Index of 0.42, the 7th lowest in the world in 2017 [63]. The rural village of Gogma
(GPS coordinates: 𝐿𝑎𝑡: 11.73°; 𝐿𝑜𝑛: - 0.58°) is a village of ~2 km×2 km located in the “Centre-East” region
of Burkina Faso (see Figure II-2). The closest town to Gogma is the town of Garango which is 15 km away.
Figure II-1 – Location of Burkina Faso. Figure II-2 – Centre-East region in Burkina Faso.
Source: Wikipedia. Source: Wikipedia.
The village counts with 1100 inhabitants who live in 125 households. The 125 households are themselves
grouped into 41 “household gatherings” such that shown in Figure II-3. The vast majority of people in
Gogma work in agriculture and live on less than $1/capita/day. Households do not have access to electricity.
We have identified 4 types of domestic water uses: drinking, cooking, personal hygiene and laundry. Water
for these uses is collected from 22 sources divided into 3 categories: 16 open wells from which water is
extracted with a bucket and a rope (see Figure II-4), 5 hand pumps (see Figure II-5) and 1 PVWPS
(Figure II-8 and Figure II-9).
29
Figure II-3 – Household gatherings.
Figure II-4 – Open well. Figure II-5 – Hand pump.
30
II.1.2 PVWPS
II.1.2.1 Design and installation
The installation of the PVWPS in Gogma followed up on the demand from the local authorities of Gogma
to the NGO Respublica [64]. The local authorities asked for the installation of an improved water source in
an area of Gogma where only open wells were available. Funding for the PVWPS was provided through a
donation from Respublica and a crowdfunding organized by the association Eau Fil du Soleil [65]. Eau Fil
du Soleil was created by several members of our research group in 2017 and aims at promoting science
initiatives in relation to PVWPS in developing regions [66]. The design and installation of the PVWPS was
coordinated by the company DargaTech [67], based in Ouagadougou and specialised in photovoltaic
systems. The data that were collected during the design and the installation of the PVWPS are presented in
section II.2.
The design and installation of the PVWPS were performed according to the following steps:
Step 1, beginning of September 2017: the local authorities looked for potential areas to avoid (e.g. burial
sites and unsafe areas, see section I.3.2), i.e. areas where the PVWPS should not be installed. No particular
area to avoid was encountered.
Step 2, beginning of September 2017: the local authorities of Gogma highlighted the position where they
would like the PVWPS to be installed.
Step 3, 18 September 2017: geophysical measurements, performed by the company Institut Superieur
d’Application des Géosciences (ISAG), highlighted a suitable position to drill (𝐿𝑜𝑛 = -0.5722° and
𝐿𝑎𝑡 = 11.7244°) next to the position proposed by the local authorities.
Step 4, 3 November 2017: a borehole was drilled by the company Sogedaf. The drilling machine was rent
to SAIRA international. Water was found at -14 m (water strike level). Then, due to the pressure of the
ground over the aquifer, the water level in the borehole went up to -6 m (static water level). The good
execution of the drilling was controlled by a third party called a “drilling controller”.
Step 5, 5 – 8 November 2017: 4 step pumping tests and 1 long pumping test were performed by Sogedaf.
Sogedaf highlighted that the maximum flow rate that can be withdrawn from the borehole 𝑄𝑝,𝑚𝑎𝑥 is equal
to 1.8 10-3 m3/s. The pumping tests were also controlled by the “drilling controller”. Physico-chemical tests
were performed by the laboratory Aina. They showed that the water is suitable for drinking in terms of
physico-chemical quality.
Step 6, November 2017: the PVWPS was sized by the company DargaTech. The peak power of the PV
array and the tank volume were determined through analytical calculations. The motor-pump was selected
from the Grundfos catalogue [68], notably because of the high quality of Grundfos motor-pumps. The PV
modules and the tank were bought in Ouagadougou and the motor-pump was imported from France.
Step 7, December 2017 – January 2018: the PVWPS was installed by DargaTech in three phases:
(a) 2 - 8 December 2017: civil engineering was performed to lay the foundations of the water tank,
build the borehole head and the fountain. 4 employees from DargaTech and 2 inhabitants of the
village participated in this civil engineering work.
31
(b) 8 January 2018: the water tank was installed by 5 employees from DargaTech.
(c) 14 January 2018: the PV array and the motor-pump were installed by 4 employees of
DargaTech and the welder from Garango. A signpost was also put in place next to the fountain
with the rules of use of the PVWPS (e.g. do not do laundry next to the fountain, do not wash
motor bikes next to the fountain)
Step 8, 14 January 2018: the PVWPS was opened for consumption to the inhabitants.
Step 9, 27 February 2018: bacteriological tests were performed by the laboratory Aina at the PVWPS.
They showed that the water is suitable for drinking in terms of bacteriological quality. Based on physico-
chemical (step 5) and bacteriological tests results, the water at the PVWPS is therefore potable.
Step 10, 27 February 2018: following the bacteriological tests results, the PVWPS was kept open for
consumption by local authorities of Gogma.
The analysis of the different steps and a semi-structured interview with Arouna Darga, the CEO of
DargaTech, allowed to determine the procedure for the design and installation of PVWPS which is presented
in Figure II-6. According to this interview, this procedure is standard in Burkina Faso.
It is interesting to observe that the bacteriological test is not performed at step 5, at the same time as physico-
chemical tests, but after the installation of the PVWPS. Indeed, Arouna Darga explained that the components
of the PVWPS (e.g. pipes, tank, fountain) may also be source of bacteriological contamination which
explains why the bacteriological test is performed after the installation of the PVWPS. In addition, Arouna
Darga specified that these bacteriological tests are not always performed or that they may be performed a
few months after the opening of the PVWPS (as in the case of the PVWPS of Gogma). Besides, according
to Arouna Darga, negative bacteriological water quality tests (i.e. non potable water) are extremely rare for
sealed boreholes in rural areas. He also added that, if the bacteriological tests were to be negative, the
opening of the PVWPS to consumption would depend on the situation in the village. For instance, if there
are only non-potable sources (e.g. open wells) in the village, the PVWPS may still be opened.
32
1
Identify areas to avoid (e.g. burial sites, unsafe areas).
2
The decision maker proposes a position ( ), outside the areas to avoid, around
which to start looking for water.
2b
The decision maker proposes
3 another position ( , ),
Perform geophysical measures along several profiles in a square of ~350 350 m around
which is outside the
( ).
~350 350 m square around
( ) and the areas to
avoid.
Do geophysical measures indicate a suitable
No
position ( to drill within the square?
Yes
4
Drill a borehole at ( ).
Is there water flowing out of the borehole? No
Yes
5
Perform pumping tests.
Determine the maximum flow rate that can be pumped thanks to the pumping tests.
Perform physico-chemical tests on the water flowing out of the borehole.
Does the physico-chemical water quality

No
meet the regulation for domestic consumption?
Yes
Does the decision maker consider that it is worth

No
installing a PVWPS at the position ( )?
Yes
6
Size the PVWPS: determine the PV array peak power, the suitable motor-pump and the
tank volume.
7
Install the PV array, motor-pump, tank and fountain.
8
Open the PVWPS for consumption to the inhabitants.
9
Perform bacteriological tests on the water flowing out of the fountain
Does the decision maker consider that the

No
bacteriological water quality is satisfying?
Yes
10 Positioning step
The PVWPS is kept open Sizing step
Figure II-6 – Conventional procedure for the design and installation of a PVWPS.
II.1.2.2 Technical description
The architecture of the PVWPS is presented in Figure II-7. The components that are encompassed in this
architecture are:
33
 A PV array.
 A motor-pump with a maximum power point tracking (MPPT) controlled converter, which tracks
the best operating point of the PV array. This whole set is immersed in the borehole and is called
“motor-pump” in the rest of the manuscript.
 A controller which starts and stops the motor-pump according to two set points of the water level
in the tank, which is obtained by a float switch.
 A water tank.
 A pipe assembly PA1 which links the motor-pump to the tank.
 A fountain at which inhabitants collect water by using 3 taps.
 A pipe assembly PA2 which links the tank to the fountain.
The water collected at the fountain is used for 4 types of domestic uses: drinking, cooking, personal hygiene
and laundry. Most of the users take water back home for drinking, cooking and personal hygiene and do the
laundry next to the fountain.
Figure II-8 and Figure II-9 show pictures of this PVWPS. A video of the village and of the PVWPS is also
available at the following link: https://youtu.be/VrjM0edKVsI. Table II-1 summarizes the features of the
PVWPS. In the rest of the thesis, we refer to the PVWPS that was installed as the “current PVWPS of
Gogma”.
34
Figure II-7 – Architecture of the PVWPS. Figure II-8 – Picture of the PVWPS of Gogma.
Figure II-9 – Overview of the PVWPS.

On the left: the fountain where the inhabitants collect water.
On the right: the PV array and the water tank surrounded by a wire netting.
35
Table II-1 – Features of the current PVWPS of Gogma.

PV array: 3 multicrystalline silicon modules in series
Surface of the PV array 3.9 m2
Tilt 𝛩 0.19 rad (11°)
Azimuth λ π rad (180°)
Borehole
Distance from the ground level to the bottom of the borehole 56 m
Interior diameter of the borehole 0.11 m
Motor-pump: Grundfos SQFlex 5A-7 [69]
Distance from the ground level to the motor-pump 30 m
Maximum power input 𝑃𝑚𝑎𝑥 (𝑀𝑃) 1400 W
Maximum pumping height 50 m
Maximum pump flow rate 2.5 10-3 m3/s
Tank: cylindrical made of steel
Height between the ground level and the bottom of the tank 𝐻𝑡,𝑏 4.2 m
Volume of the cylinder 𝑉𝑡 11.4 m3
Base surface of the cylinder 𝑆𝑡 3.3 m2
Height of the cylinder 𝐻𝑡,𝑡 3.5 m
Controller: Grundfos CU 200 [70]
Pipe assembly PA1: 4 pipes of different diameters and materials
Total length 47 m
Fountain
Number of taps 3
Pipe assembly PA2: 1 pipe
Total length 21 m
36
II.2 Data collection

In Gogma, since September 2017, we have been collecting data from the following disciplines: energy
systems, social sciences, geography, hydrology and geophysics. The collected data are summarized in
Table II-2 and further detailed in the following sections. In Table II-2, we also specify:
 The data that have been collected during the procedure of design an installation of the PVWPS
(labelled as ‘Conventional procedure’) and the ones that have been collected in addition,
specifically for research purposes (labelled as ‘Research’).
 If data were collected by our research group or through specialized companies based in Burkina
Faso.
 The collection period, i.e. the total period over which data have been collected (note that the data
are not necessarily collected full-time during the collection period).
 The collection time, i.e. the time required to collect the data in number of work days of one data
collector (we consider that the collector works 8 hours per day).
 The collection cost. When the data collection was performed by a company, we use the bill from
the company. When then data collection was performed by our research group, we estimate data
collection cost by multiplying the collection time by the daily cost of a data collector, estimated to
$160/day.
 The sub-model(s) where data is (are) used. Indeed, the interdisciplinary model presented in Chapter
III, is composed of 4 sub-models (demand, technical, impact and economic) and the collected data
may be used in one or several of these sub-models as input and/or for experimental validation.
The considerations on the collection period, time and cost allow to quantify the investment in time and
money required to collect each type of data.
37
Table II-2 – Summary of collected data.

Type Description Collected by Collection period Data collection cost Sub-model
(Collection time for where data
1 data collector) is used (see
Chapter III)
On-field Observations on living Research 18 Sept 2017 – $1600 Demand,
observations conditions and water access. group Present Impact
Research (10 days)
GIS GPS positions of the Research 18 Sept 2017 – 10 $480 Demand,
(geographic households, water sources, group Nov 2017 Impact
information important points of the village (3 days)
system) (shops, mosques, church, …).
mapping Research
Geophysical Detection of the presence of Institut 18 Sept 2017 $386
measures water along 4 profiles of Superieur (1 day)
150 m long by using d’Application
electromagnetic methods. des
Conventional procedure Géosciences
(ISAG) [71]
Account The account book of a source Research 03 Oct 2017 – 20 $160 Demand,
books of specifies the households that group June 2018 Impact
water attend this source and the cost (1 day)
sources for collecting water at the
source.
Research
Household Survey on the living Survey before Survey before $3200 Demand,
surveys conditions and water access. installation: installation: Impact
Two surveys are performed: Research 03 Oct 2017 – 21
one before the installation of group Oct 2017 (10 days)
the PVWPS and one after.
Research Survey after Survey after
installation: installation:
company Best 27 Oct 2018 to 6
Sigma Nov 2018 (10 days)
Pumping 4 step pumping tests, of 2 Sogedaf 05 Nov 2017 – 09 4 step pumping tests: Technical
tests hours each, at 4 different Nov 2017 $511
pump flow rates. 1 long (3 days)
pumping test of 47 hours at a Long pumping test:
given flow rate. $2050
Conventional procedure
Water 1 physico-chemical test at the Laboratoire 7 Nov 2017 – 27 1 physico-chemical Impact
quality tests PVWPS. Aina [72] Feb 2018 test at the PVWPS =
1 bacteriological test at the (2 days) $141
PVWPS.
Conventional procedure 1 bacteriological test
at the PVWPS = $21
21 bacteriological tests for the
other 21 water sources for 21 bacteriological
domestic water in Gogma. tests for the other
Research sources = $441
PVWPS Continuous technical data Research 14 Jan 2018 – Hardware and Demand,
monitoring collection on the operation of group Present installation1 = $2000 Technical
the PVWPS. (10 days2)
Research Data collection and
processing = $1000
1
The development time and cost of the data logger is not included. The development time is estimated to 3 months
and the development cost to $50000.
38
II.2.1 On-field observations

In total 8 months were spent in Gogma by members of our research group, split at different moments during
the last 3 years. This allowed us to perform 250 on-field observations. For each observation, the following
elements were specified:
 Content.
 Is it something that was observed with the eyes or is it something that was said by someone. For the
latter case, the name of the person was written down.
 Location.
 Day and time of the day.
The following key information were gathered through on-field observations:
 Households do not need to pay to collect water at open wells.

 Most of the inhabitants perceive the water at sealed boreholes (hand pumps and PVWPS) as potable
and perceive the water at open wells as non-potable.
 Inhabitants perceive water extraction at PVWPS as easy.
 Most of the households collect water in the morning and in the evening but it seems that they collect
more water in the evening than in the morning.
 Inhabitants sometimes use their bike to carry water.
 In the household surveys, inhabitants reported to use ~20 L/capita/day for personal hygiene.
According to our on-field observations, we think that it is overestimated.
 Inhabitants sometimes make several return journeys to collect water.
II.2.2 GIS mapping

The GPS coordinates of the households, water sources and important locations in the village have been
collected. We went to each location and used the mobile application “GPS Satellite” [73]. The coordinates
obtained are represented on the satellite picture in Figure II-10.
39
Figure II-10 – GIS mapping of Gogma.

Data collected between the 18 September 2017 and the 10 November 2017.
II.2.3 Geophysical measures

The geophysical study was realized by the Institut Superieur d’Application des Géosciences (ISAG) by
using the very low-frequency electromagnetic method [74]. Geophysical measurements were performed
along 4 profiles of 150 m each, which corresponds to scanning a square of ~350×350 m. These profiles are
represented in Figure II-10. Measurements along profiles 1, 2 and 4 did not suggest the presences of water
along those profiles. On profile 3, there is a specific position that suggested the presence of water. It is at
this position that the borehole for the current PVWPS was drilled.
40
Typically, performing measurements in a square of 350×350 m (i.e. 0.12 km2) takes half a day and costs
~$500. It would therefore be very costly and time consuming to perform geophysical measurements over an
entire village of several square kilometres (e.g. Gogma has an area of 4 km2). In addition, it is important to
have in mind that these geophysical measurements only provide information about the level of the top of
the aquifer, which is not equal to the static water level in the borehole that will be drilled in the case of a
confined aquifer [49]. Besides, geophysical measurements do not provide information about the response
of the water level in the borehole to water pumping nor about the water quality [49]. This explains why
geophysical measurements and groundwater exploration are performed only after the positioning of the
PVWPS, i.e. after a position is proposed by the decision maker (see Figure II-6).
II.2.4 Account books of water sources

Account books of the 5 hand pumps and of the PVWPS were accessed by our research group. There is one
account book for each hand pump and one account book for the PVWPS. The account book specifies which
households go to this source and the cost to collect water at this source. We also remind that we determined
that open wells are free of charge through on-field observations (section II.2.1). The cost for each source is
given in Table II-3. The hand pumps numbers correspond to the ones presented in Figure II-10.
It is interesting to note that households have to pay annually for hand pumps and monthly for the PVWPS.
In addition, we observe that the yearly amount paid for water at hand pumps and at the PVWPS does not
depend on the quantity of water consumed.
Table II-3 – Water sources costs.

Source type/number Price
Open wells $0/year
Hand pump 1 $2.1/year
PVWPS $0.9/month i.e. $10.8/year
Data collected between 3 October 2017 and 20 June 2018.
II.2.5 Household surveys

II.2.5.1 Description
Two rounds of household structured surveys were performed. The first round took place in October 2017,
before the installation of the PVWPS. The second round took place in October/November 2018, after the
installation of the PVWPS. The households surveyed were the same for both survey rounds. 88 households
were selected randomly to be surveyed from the 125 households in Gogma2. The main points covered by
the survey are presented in Table II-4 and all the questions of the survey are available in Appendix A. The
survey duration was about 45 minutes. The survey was designed drawing on existing surveys on water and
2
90 households were in fact surveyed but 2 are not considered because survey answers for these households are not
complete, notably regarding their choice of water source.
41
electricity access in developing countries (e.g. survey of the SURE-DSS tool [61]). Surveying a large share
of the households in Gogma (~70%) and selecting these households randomly allowed us to survey both
households that started going to the PVWPS after its installation, and others that did not. Answers to the
second round of surveys showed that 22 of the 88 surveyed households go to the PVWPS both during the
dry and the wet season.
In this PhD thesis, we use answers to the first survey round as inputs for the demand and impact models (see
sections III.2 and III.4) and answers to the second survey round to compare the results of the demand model
to experimental measurements (see sections III.2). In the future, comparing the results of the first and second
survey rounds may allow us to quantify some local impacts of the installation of the PVWPS of Gogma.
This is notably with this in mind that we performed both survey rounds at the same moment of the year
(October/November).
Table II-4 – Themes covered by the household survey.
Theme Description Survey section
(see Appendix A)
Time and position Time of survey and GPS position of the surveyed household. a
Household Age, gender, level of education and economic activity of all household a
demographic situation members.
Economic assets - Number of houses owned and materials of the floor, walls and roof of a
each house.
- Number of phones.
- Number of motor bikes and bicycles.
Access to services Access to internet, electricity, gas cooking. a
Agricultural activities Number of fields, quantities of the different crops grown, use of crops b
grown (self-consumption or selling in the market).
Livestock Number and type of the animals owned. b
Health List of diseases and symptoms contracted by each member of the c
household in the last month, and associated medical expenses.
Women’s schedule Time allocated for their different activities of the day. d
Safety Perception of the safety in the village (robbers, snakes …) e
Water use For the dry season and the wet season: f
- source where water is collected.
- daily quantity of water collected for each water use.
- time at which water is collected.
Water sources For each water source used by the household: f
- availability of water at the source during the dry season and the wet
season.
- perception about the water quality at the source
- perception about the difficulty to extract water at the source .
- cost of using the source.
- time spent queuing and extracting water.
Income and sparings Income from different sources of revenues (crop and animal sales, sales at g
shop, wages earned, money received from government, money received
from family), and amount of money spared.
First survey round (before installation of the PVWPS): 3 October 2017 to 21 October 2017, second survey round (after
the installation of the PVWPS): 27 October 2018 to 6 November 2018.
II.2.5.2 Data cleaning
The answers to questions on ‘water use’ and ‘water sources’ (section F of the survey) will be the most used
in this thesis. We observe that ~30% of the households go to different sources between the dry season and
the wet season. This is mostly due to the fact that some open wells are not available during the dry season
because they are dry. Therefore, some households go to an open well during the wet season but have to go
42
to a different water source during the dry season. Seasonality will thus be considered in the interdisciplinary
model (see Chapter III).
In addition, ~15% of the households go to different sources for different uses for a given season. For
instance, for the dry season the household may go to two sources for drinking, cooking and hygiene but may
go to a third one for laundry. In this case, we consider the source(s) used for drinking as the destination(s),
as drinking is the most critical use in terms of water quality. Finally, ~15% of the households go to two
sources for drinking, instead of one, for a given season. In this case, we select randomly one of both sources
as the final destination. We made these assumptions, and therefore did not consider that households may
use different sources for different uses and that they may use different sources for drinking in the
interdisciplinary model (and more specifically in the demand model) because we did not understand the
motivation behind these behaviours of the households. This could be the object of future work.
II.2.6 Pumping test

Two types of pumping tests were performed at the PVWPS just after the borehole drilling, in November
2017, by the company Sogedaf.
Firstly, 4 step pumping tests were performed for 4 different flow rates (2.8 10-4, 8.3 10-4, 1.7 10-3 and
1.8 10-3 m3/s). For each step pumping test, the water level in the borehole 𝐻𝑏 was monitored, with a water
level meter [75], during one hour of pumping at a given flow rate and during one additional hour when there
was no pumping (recovery phase). Figure II-11 presents the water level in the borehole 𝐻𝑏 measured during
the 4 step pumping tests. Secondly, a 47 h pumping test was performed. The water level in the borehole
𝐻𝑏 was monitored during 36 hours of pumping at a flow rate of 1.8 10-3 m3/s and during an additional
11 hours when there was no pumping (recovery phase). The results are given in Figure II-12.
On these pumping tests we observe that the static water level, i.e. the water level in the borehole when there
is no pumping, is equal to -6 m at this time of the year. The decrease of 𝐻𝑏 from this static water level while
pumping corresponds to the dynamic behaviour of the water level in the borehole. Following these
measurements, Sogedaf highlighted that the maximum flow rate that can be pumped from the borehole
𝑄𝑝,𝑚𝑎𝑥 is equal to 1.8 10-3 m3/s. However, Sogedaf did not explain the choice of this value for 𝑄𝑝,𝑚𝑎𝑥 and
did not model the response of the water level in the borehole 𝐻𝑏 to pumping.
43
-4 3
2.8 10 m /s
-4 3
8.3 10 m /s
-3 3
1.7 10 m /s
-3 3
1.8 10 m /s
Figure II-11 – Water level in the borehole measured Figure II-12 – Water level in the borehole measured
during the step pumping tests. during the long pumping test
Test performed on 5 November 2017 by Sogedaf. (pump flow rate: 1.8 10 m3/s).
Test performed on 8 and 9 November 2017 by Sogedaf.
II.2.7 Water quality data

All the water quality tests were performed by the Laboratoire Aina [72].
II.2.7.1 Physico-chemical tests
The physico-chemical tests were performed only at the PVWPS. They took place during the pumping tests
on 7 November 2017. Laboratoire Aina measured the physico-chemical parameters of the water amongst
which the pH, temperature, alkalinity, electrical conductivity, arsenic concentration. These measurements
were compared to the physico-chemical quality standards for drinking water in Burkina Faso. The physico-
chemical quality of water fulfilled the requirements for consumption.
II.2.7.2 Bacteriological tests
Bacteriological tests on all the open wells and hand pumps were performed on 14 November 2017 and the
test on the PVWPS was performed on 27 February 2018. For each source, Laboratoire Aina measured the
concentration of total coliform, thermotolerant coliform and faecal streptococci. The results of the water
quality tests are summarized in Table II-5.
Table II-5 – Results of bacteriological tests performed by Laboratoire Aina [72].

Number of Concentration of total Concentration of thermotolerant Concentration of faecal
Source
sources coliform (CFU/100mL) coliform (CFU/100mL) streptococci (CFU/100mL)
Open Mean: 441; Std: 482; Mean: 222; Std: 311; Mean: 377; Std:264;
16
well Min: 0; Max: 1688 Min: 0; Max: 800 Min: 44; Max: 816
Hand Mean: 0; Std: 0; Mean: 0; Std: 0; Mean: 0; Std:0;
5
pump Min: 0; Max: 0 Min: 0; Max: 0 Min: 0; Max: 0
Mean: 0; Std: 0; Mean: 0; Std: 0; Mean: 0; Std: 0;
PVWPS 1
Min: 0; Max: 0 Min: 0; Max: 0 Min: 0; Max: 0
Std: standard deviation. Numbers higher than zeros are in red.
Tests performed on 14 November 2017 and on 27 February 2018.
According to international recommendations, in drinking water, there should be 0 CFU/100mL total
coliform [76], 0 CFU/100mL thermotolerant coliform [76] and 0 CFU/100mL faecal streptococci [77].
Therefore, according to Table II-5, water from all open wells is not suitable for drinking in terms of
bacteriological quality and water from all hand pumps and from the PVWPS is suitable (see Table II-5).
44
Therefore, the bacteriological tests show that the water at open wells is not potable, which was expected
based on the literature (see section I.1.1).
We did not perform physico-chemical tests at the hand pumps of the village for budgetary reasons (~$140
for each test, see Table II-2) and because these physico-chemical tests were already performed when the
hand pumps were installed and they showed that the water is suitable for drinking in terms of physico-
chemical quality (otherwise the hand pumps could not have been opened to consumption). Consequently,
in Gogma, the water from the hand pumps and from the PVWPS is potable, which was expected based on
the literature (see section I.1.1).
The comparison of the water quality tests results to the water quality perceived by the households, which is
obtained through the surveys (see section II.2.5) and on-field observations (see section II.2.1), shows that
households have a good perception of water quality: in general, they know that the water at open wells is
not potable (i.e. low water quality) and that the water at sealed boreholes (hand pumps and PVWPS) is
potable (i.e. high water quality).
II.2.8 Monitoring of the PVWPS of Gogma

The quantities measured in the frame of the monitoring of the PVWPS of Gogma are presented in Table II-6.
In this table, the pump flow rate 𝑄𝑝 is the flow rate extracted from the borehole and is measured thanks to
a flow meter set on PA1 (see Figure II-7). The consumption flow rate 𝑄𝑐 corresponds to the water collected
by the inhabitants at the fountain. This flow rate is the amount of water retrieved from the water tank and is
measured by setting up a flow meter on PA2.
Most of the quantities of Table II-6 have been collected since January 2018 thanks to a data logger that we
developed. The data logger is powered by external PV modules (different from the ones of the PVWPS) and
the recorded data are collected by using a USB stick. This data logger was conceived with the idea of
minimizing its cost in order to encourage its use for monitoring other PV water pumping installations. The
architecture and a picture of the data logger are shown in Figure II-13. The data logger permits to collect
data with a time step of ~2.2 s and the recording rate is equal to the frequency of acquisition. In February
2019, an independent hydrostatic pressure sensor was added to measure the water level in the borehole 𝐻𝑏 .
The water level in the borehole is measured with a time step of 1 minute. For convenience, all the measured
data were rescaled to an equally spaced temporal resolution of 1 min by nearest interpolation for this PhD
thesis.
45
Table II-6 – Data monitored by the data logger.

Measured data Type of sensor Model of Location of Frequency Sensor Period of
sensor sensor (Samples/min) accuracy data
collection
Irradiance on the Calibrated Solems RG100 On the plane 27 ± 10% 14 Jan 2018
plane of the PV photovoltaic cell of the PV - present
modules (𝐺𝑝𝑣 ) & Data logger modules
Ambient Platinum RS Pro In the 27 ± 0.05% 14 Jan 2018
temperature (𝑇𝑎 ) resistance PT1000 shadow, next - present
thermometer & to the PV
Data logger modules
PV modules Platinum RS Pro On the back 27 ± 0.05% 14 Jan 2018
temperature (𝑇𝑝𝑣 ) resistance PT1000 of a PV - present
thermometer & module
Data logger
Voltage of the PV Voltage LEM LV 25P At the output 27 ± 0.9% 14 Jan 2018
array (𝑉𝑝𝑣 ) transducer & of the PV - present
Data logger array
Current from the Current LEM LA 55P At the output 27 ± 0.9% 14 Jan 2018
PV array (𝐼𝑝𝑣 ) transducer & of the PV - present
Data logger array
Pump flow rate Turbine flow HaiHuiLai On PA1 27 ± 5% 14 Jan 2018
(𝑄𝑝 ) sensor & Data YF-DN40 - present
logger
Consumption flow Turbine flow HaiHuiLai On PA2 27 ± 5% 14 Jan 2018
rate (𝑄𝑐 ) sensor & Data YF-DN40 - present
logger
Water level in the Hydrostatic DCX-22 SG In the 1 ± 0.1% 8 Feb 2019 -
borehole (𝐻𝑏 ) pressure sensor borehole present
Data collected from 14 January 2018.
Figure II-13 – Data logger developed for monitoring the PVWPS of Gogma.
Figure II-14 presents an example of data collected by the data logger on the 19 February 2019. The
interruptions in the pump flow rate profile (at 9h43 and 13h55) correspond to the moments at which the
water level in the tank 𝐻𝑡 has reached the stop level 𝐻𝑡,𝑠 , which means that the tank is full. When water
pumping is interrupted, the PV modules are in open circuit. As the water level in the tank can be directly
deduced from the pump and the consumption flow rate, which are both measured, it is also considered as
measured. Moreover, in order to avoid a shift of this measured water level in the tank, which may come
46
from the uncertainty on the flow meters measurements, we reset the measured water height to 𝐻𝑡,𝑠 each time
the tank is full. During the whole record history of the PVWPS, the water level in the tank has always
remained higher than 0 m. This means that the sizing of the PVWPS is such that all the households that wish
to go to the PVWPS are able to collect water there. The PVWPS is said to be ‘oversized’.
To our best knowledge, it is the first time that experimental measurements have been collected on a PVWPS
for domestic water access in a rural village and on a PVWPS located in sub-Saharan Africa [78], and we
have the objective of monitoring the PVWPS of Gogma during its whole lifetime. This unique database can
therefore help to study the performance and the sustainability of photovoltaic water pumping in sub-Saharan
Africa and for domestic water access.
More generally, this database can also be used for other studies related to photovoltaic systems in rural
villages in sub-Saharan Africa. For instance, we used it for the detection of cleaning interventions on
photovoltaic modules by using machine learning algorithms [79]. In addition, the monitoring of the PVWPS
and the development of the data logger was at the origin of the “Axo” project, which focuses on the potential
of big data for PVWPS in rural areas [80]. Finally, we built a second data logger, that is installed on a
photovoltaic water pumping pilot at the University of Paris-Saclay. The pilot and the associated data logger
are notably used for teaching projects on energy conversion, testing motor-pumps from manufacturers and
training students that will undertake humanitarian work in developing countries.
47
Figure II-14 – Example of data collected by the data logger.
48
II.3 Partial conclusion

In this chapter, firstly, we presented the rural village of Gogma in Burkina Faso where we installed a
PVWPS. We highlighted the complexity of the global water situation in Gogma where 125 households
collect water at 22 sources. We detailed the conventional procedure that was followed for installing the
PVWPS and we provided a technical description of the system (architecture, components).
Secondly, we described the experimental data that we have been collecting in Gogma. The following data
were collected:
 On-field observations were performed during the 8 months of fieldwork of members of our research
group.
 The GPS coordinates of the households, water sources and important points of the village were
gathered.
 Geophysical measures were performed, before the installation of the PVWPS, in order to detect the
presence of water.
 The cost to collect water at each source was determined through the account books of the sources.
 88 randomly selected households were surveyed both before and after the installation of the
PVWPS.
 Pumping tests were undertaken to quantify the effect of water pumping on groundwater resources.
 Bacteriological analyses were performed for all water sources of the village in order to quantify
their quality.
 The operation of the PVWPS has been monitored continuously since January 2018 thanks to a data
logger that we have built and installed.
49
Chapter III Interdisciplinary model
50
Chapter III Interdisciplinary model
In this PhD thesis, we aim at performing an optimisation to determine the sizings and the positions of the
PVWPS that minimize its life-cycle cost and maximise its positive socio-economic impact. To do so, the
first step is to build a model that relates the optimisation variables, which are associated to the sizing and
the position of the PVWPS, to the objective functions of the optimisation, which are the life-cycle cost of
the PVWPS and its socio-economic impact. The proposed interdisciplinary model is presented in this
chapter. Then, in Chapter IV, we will show how this model is used for the optimisation.
In section III.1, we present an overview of the interdisciplinary model and introduce the different sub-
models that compose it. Then, from section III.2 to section III.5 we detail the sub-models: the demand model
in section III.2, the technical model in section III.3, the impact model in section III.4 and the economic
model in section III.5.
51
Chapter III. Interdisciplinary model
III.1 Overview
The block diagram of the interdisciplinary model is presented in Figure III-1. The 4 sub-models (demand,
technical, impact and economic) are shown on this diagram.
Interdisciplinary model
Life-cycle
Sizing Economic
cost
model
Climatic data
Groundwater Technical
parameters model
, Socio-economic
Impact
impact
model
Position
Demand
model
Optimisation variables
Households behaviour
Objective functions
Figure III-1 – Block diagram of the interdisciplinary model.

𝒉𝒔𝒃𝒆𝒇 : water sources where the households go before the installation of the PVWPS, 𝒉𝒔∗𝒂𝒇𝒕 : water sources where the
households wish to go after installation of the PVWPS,𝑄𝑑 : water demand at the PVWPS, 𝒉𝒔𝒂𝒇𝒕 : water sources where
the households effectively go after installation of the PVWPS, 𝑄𝑐 : water consumption at the PVWPS.
The inputs of the interdisciplinary model can be grouped into 4 categories:
1. Optimisation variables. We distinguish two types of optimisation variables: (1) the sizing variables
of the PVWPS and (2) the position of the PVWPS in the village, which is given by its longitude
𝐿𝑜𝑛 and its latitude 𝐿𝑎𝑡. These sizing variables include the peak power of the PV array in standard
test conditions (STC) 𝑃𝑝𝑣,𝑝 , the reference of motor-pump 𝑀𝑃 and the volume of the water tank 𝑉𝑡 .
These particular three sizing variables are chosen based on the existing literature (see section I.3.3)
and on a sensitivity analysis that we performed in article [81], which showed that these three
variables have the highest influence on the operation of PVWPS.
2. Climatic data. Climatic data are composed of the irradiance on the plane of the PV array 𝐺𝑝𝑣 and
the ambient temperature 𝑇𝑎 .
3. Groundwater parameters. The groundwater parameters are the static water level in the borehole
𝐻𝑏,𝑠 (i.e. the water level in the borehole when there is no pumping) and the aquifer losses coefficient
𝜅0 and the borehole losses coefficient 𝜇0 . 𝜅0 and 𝜇0 characterize the dynamic behaviour of the water
level in the borehole. The parameters 𝐻𝑏,𝑠 , 𝜅0 and 𝜇0 depend on the groundwater resources in the
considered village.
4. Water sources where the households go before installation of the PVWPS: 𝒉𝒔𝒃𝒆𝒇. More specifically,
𝒉𝒔𝒃𝒆𝒇 is a vector, of length the number of households in the village 𝑛𝑣 , which specifies, for each
household, the water source where it goes before the installation of the PVWPS.
52
The outputs of the interdisciplinary model are the two objective functions of the optimisation: the life-cycle
cost 𝐿𝐶𝐶 of the PVWPS and its socio-economic impact 𝑆𝐸𝐼.
In Figure III-1, 𝒉𝒔∗𝒂𝒇𝒕 is a vector of length 𝑛𝑣 which indicates, for each household, the water source where
it wishes to go after the installation of the PVWPS and 𝑄𝑑 (𝑡) is the corresponding water demand at the
PVWPS. 𝒉𝒔𝒂𝒇𝒕 is a vector of length 𝑛𝑣 which indicates, for each household, the water source where it
effectively goes after the installation of the PVWPS and 𝑄𝑐 (𝑡) is the corresponding water consumption. If
the PVWPS is ‘oversized’, all the households that wish to go to the PVWPS can go there (𝒉𝒔𝒂𝒇𝒕 = 𝒉𝒔∗𝒂𝒇𝒕 )
and the water consumption is equal to the water demand (𝑄𝑐 (𝑡) = 𝑄𝑑 (𝑡)). If the PVWPS is ‘undersized’,
all the households that wish to go to the PVWPS cannot go there (𝒉𝒔𝒂𝒇𝒕 ≠ 𝒉𝒔∗𝒂𝒇𝒕 ) and the water consumption
is not equal to the water demand, (𝑄𝑐 (𝑡) ≠ 𝑄𝑑 (𝑡)).
Regarding the four sub-models:
 The demand model is an econometric model which predicts the water sources where the households
wish to go after installation of the PVWPS from the position of the PVWPS in the village and the
water sources where the households go before installation of the PVWPS.
 The technical model allows to identify the households that can effectively go to the PVWPS
amongst the households that wish to go to the PVWPS. This model considers the climatic
conditions, the groundwater parameters and the sizing of the PVWPS. It simulates the different
stages of the energy conversion chain within the PVWPS.
 The impact model permits to evaluate the socio-economic impact 𝑆𝐸𝐼 associated with the changes
in water sources between before and after the installation of the PVWPS.
 The economic model allows to determine the life-cycle cost 𝐿𝐶𝐶 of the PVWPS from the values of
the sizing variables. The model takes into account the capital and operational costs and the time
value of money.
These four sub-models are further detailed in the following sections.
For the case of Gogma, some open wells are not available during the dry season because they are dry.
Therefore, some households go to an open well during the wet season but have to go to a different water
source during the dry season (see section II.2.5.2). Consequently, the vectors 𝒉𝒔𝒃𝒆𝒇, 𝒉𝒔∗𝒂𝒇𝒕 and 𝒉𝒔𝒂𝒇𝒕 are
not the same between the dry season and the wet season. In addition, the demand, technical and impact
models are evaluated once for the dry season and once for the wet season. For the application of the model
to Gogma, we consider that the dry season lasts from 1 December 2017 to 30 June 2018 and that the wet
season lasts from 1 July 2018 to 30 November 2018.
53
III.2 Demand model

The block diagram of the demand model is shown in Figure III-2. During the first step (section III.2.1) we
determine, for each household, the water source where it wishes to go after the installation of the PVWPS
𝒉𝒔∗𝒂𝒇𝒕 from the position of the PVWPS in the village (𝐿𝑎𝑡, 𝐿𝑜𝑛) and the water source choice of each
household before installation of the PVWPS 𝒉𝒔𝒃𝒆𝒇. During the second step (section III.2.2), we determine
the evolution of the water demand 𝑄𝑑 at the PVWPS from 𝒉𝒔∗𝒂𝒇𝒕 . Our demand model builds on the related
MSc thesis project undertaken by Vitali Caplain in 2018 [82]. To the best of our knowledge, it is the first
model that predicts the water demand profile at water sources of a rural village, considers the inclusion of a
new type of water source and accounts for the seasonality in the water demand.
Demand model
Position Determination of
, the water sources
Determination of
where the
the water demand
households wish to
profile at the
go after the
PVWPS
installation of the
PVWPS
Figure III-2 – Block diagram of the demand model.

𝒉𝒔𝒃𝒆𝒇 : water sources where the households go before the installation of the PVWPS, 𝒉𝒔∗𝒂𝒇𝒕 : water sources where the
households wish to go after the installation of the PVWPS, 𝑄𝑑 : water demand at the PVWPS.
III.2.1 Determination of the water sources where the households wish to go after installation of
the PVWPS
In section III.2.1.1, we use the data collected in the household survey regarding water source choice before
the installation of the PVWPS 𝒉𝒔𝒃𝒆𝒇 to build an econometric model that predicts the choice of water source
of each household. In section III.2.1.2, this model will then be applied to predict the water sources where
the households wish to go after the installation of the PVWPS 𝒉𝒔∗𝒂𝒇𝒕 .
III.2.1.1 Building the regression using the water source choice before installation of the PVWPS
In this section, we consider only the surveyed households 𝒉𝒔𝒖𝒓 and the water sources that were there before
the installation of the PVWPS (i.e. 16 open wells and 5 hand pumps for Gogma).
The probability 𝜌(ℎ, 𝑠) that the household ℎ goes to the water source 𝑠 is obtained by a linear regression:
𝜌(ℎ, 𝑠) = 𝛽0 + ∑ 𝛽𝑖 ⋅ 𝜋𝑖 (ℎ, 𝑠) (1)

𝑖
where 𝛽𝑖 are the regression coefficients and 𝜋𝑖 are the predictors of the water source choice (𝜋𝑖 (ℎ, 𝑠) is the
value of the predictor 𝜋𝑖 for the household ℎ and the source 𝑠). The main advantage of the linear regression
is that the regression coefficients 𝛽𝑖 obtained are easy to interpret. In addition, we also tried a logistic
regression without significant improvement in the prediction accuracy [82].
54
Firstly, we identify the potential predictors 𝜋𝑖 of the water source choice through a literature review:
 Distance household-source (𝜋𝑑 ). The distance between the household and the water source is
negatively correlated to the likelihood of choosing the water source [83].
 Water cost (𝜋𝑤𝑐 ). The lower the cost of water at a source, the higher the will to use the source [84].
 Perceived water quality (𝜋𝑝𝑤𝑞 ). The lower the perceived quality of the water from the source, the
lower the will to use the source [84].
 Extraction easiness (𝜋𝑒𝑒 ). People are more likely to choose a water source if water collection at this
source is easy [85].
 Time collecting water (𝜋𝑡𝑐𝑤 ). The time collecting water is equal to the sum of the time spent queuing
and of time spent extracting water. The time collecting water at a given source is negatively
correlated to the likelihood of choosing the source [83].
Secondly, each predictor is quantified for the considered village. The methods of quantification of the
predictors for Gogma are presented in Table III-1. Thanks to the data collected through GIS mapping
(sections II.2.2) and water sources account books (section II.2.4) and household surveys (section II.2.5), it
is possible to compute the value of each predictor 𝜋𝑖 for each household ℎ and water source 𝑠: 𝜋𝑖 (ℎ, 𝑠).
Table III-1 – Quantification of predictors in the case of Gogma.

Predictor Method of quantification Possible values
Distance For each household, we calculated the straight line distance Euclidean distance in
household-source between the household and the water sources of Gogma from GPS meters
(𝜋𝑑 ) coordinates. The choice of the straight line distance is justified in
[86].
Water cost (𝜋𝑤𝑐 ) The cost for collecting water at each source was obtained through Monthly cost in $
account books and on-field observations. For a given source 𝑠, the
water cost is the same for all households ℎ.
Perceived water In the survey, each household was asked about how it perceives the 0 (high perceived quality)
quality (𝜋𝑝𝑤𝑞 ) water quality at the source it uses. It could choose between: 0 (high – 1 (low perceived quality)
quality) and 1 (low quality). However, we do not know how the
households perceives water quality at a source where it does not go.
We extrapolate missing data by looking at the answers of the
households that go to this source (we take the average of their
answers).
Extraction In the survey, each household was asked about the arduousness to 0 (not arduous) –
easiness (𝜋𝑒𝑒 ) collect water at the source it uses. It could choose between: 0 (not 2 (very arduous)
arduous), 1 (arduous), 2 (very arduous). Missing data are
extrapolated in the same way as for the perceived water quality.
Time collecting In the survey, each household was asked about the time it takes to Time in minutes
water (𝜋𝑡𝑐𝑤 ) collect water at the source it uses. Missing data are extrapolated in
the same way as for the perceived water quality.
Thirdly, we select the predictors to include in the model. We remind that the final objective and the
originality of the demand model is to predict which households wish to go to a water source of a new type,
i.e. the PVWPS (see section I.4). According to this objective, we propose to add an original step in the
design of the demand model. This step consists in analysing the risks of using some predictors, knowing
that they will be used to predict demand at a source of a new type. In the case of Gogma, we found the
following risks associated with using the predictors 𝜋𝑡𝑐𝑤 , 𝜋𝑤𝑐 and 𝜋𝑒𝑒 :
55
 For the predictor 𝜋𝑡𝑐𝑤 (time collecting water), it is difficult to assume an accurate value for the
time collecting water at the PVWPS. Indeed, this value depends on many unknown factors such as
the attendance at the PVWPS and the management scheme adopted at the PVWPS.
 For the predictor 𝜋𝑤𝑐 (water cost) and the predictor 𝜋𝑒𝑒 (extraction easiness), the value of the
predictor for the PVWPS is expected to be out of the range of the values of the predictor for the
water sources available before the installation of the PVWPS. More specifically, for the predictor
𝜋𝑤𝑐 , water collection at open wells is free and it costs ~$1/year at hand pumps, but it costs ~$11/year
at the PVWPS (see section II.2.4). For the predictor 𝜋𝑒𝑒 , extraction is ‘very hard’ at open wells
(𝜋𝑒𝑒 ~2) and ‘hard’ at hand pumps (𝜋𝑒𝑒 ~1) but it is expected to be ‘easy’ at the PVWPS (𝜋𝑒𝑒 ~0)
(see section II.2.1). Consequently, as the determination of the regression coefficients 𝛽𝑖 is
performed only with the water sources available before the installation of the PVWPS, there is a
large risk of error when using these 𝛽𝑖 coefficients to predict the households that wish to go to the
PVWPS. In other words, we do not know how the households would react to a source for which
water collection would cost $11/year or for which water extraction would be ‘easy’ (𝜋𝑒𝑒 ~0),
because this type of source was not considered for the identification of the 𝛽𝑖 coefficients of the
model.
On the contrary, for the predictor 𝜋𝑝𝑤𝑞 (perceived water quality), we can reliably assume that the perceived
water quality at the PVWPS will be high (~0). Indeed, according to on-field observations, inhabitants have
a high perception of the water quality at sealed boreholes like the PVWPS (see section II.2.1). Besides, the
values of 𝜋𝑝𝑤𝑞 used for the identification are within 0 and 1, as 𝜋𝑝𝑤𝑞 ~1 for open wells and 𝜋𝑝𝑤𝑞 ~0 for
hand pumps. Thus, the perceived water quality at the PVWPS (𝜋𝑝𝑤𝑞 ~0) is within the range of perceived
water quality used for the identification (𝜋𝑝𝑤𝑞 between 0 and 1).
We also analyse the multicollinearity between predictors. Indeed, if two predictors are correlated, they may
not be both required in the prediction model and therefore it may be possible not to use some of the
predictors, which are associated with some risk when the PVWPS will be added to the available sources.
We compute the multicollinearity matrix for each season by using the values of the predictors for each
association household/source. The results are given in Table III-2. The small differences in correlation
coefficients between the dry and wet seasons come from the fact that some open wells are not available
during the dry season. According to [87], two predictors are strongly correlated if the correlation coefficient
between these two predictors is not within -0.4 and 0.4. In Table III-2, we write in red the values of the
correlation coefficients that are not within -0.4 and 0.4.
56
Table III-2 – Multicollinearity matrix for Gogma.

Distance Water cost Perceived water Extraction Time collecting
household-source (𝜋𝑤𝑐 ) quality easiness water
(𝜋𝑑 ) (𝜋𝑝𝑤𝑞 ) (𝜋𝑒𝑒 ) (𝜋𝑡𝑐𝑤 )
Distance household-source (𝜋𝑑 )
Water cost (𝜋𝑤𝑐 ) Dry: 0.14
Wet: 0.13
Perceived water quality (𝜋𝑝𝑤𝑞 ) Dry: -0.09 Dry: -0.90
Wet: -0.08 Wet: -0.91
Extraction easiness (𝜋𝑒𝑒 ) Dry: -0.01 Dry: -0.61 Dry: 0.65
Wet: -0.03 Wet: -0.70 Wet: 0.76
Time collecting water (𝜋𝑡𝑐𝑤 ) Dry: -0.15 Dry: 0.43 Dry: -0.57 Dry: -0.41
Wet: -0.12 Wet: 0.49 Wet: -0.63 Wet: -0.44
In red: values of the correlation coefficients that are not within -0.4 and 0.4.
Results indicate that the predictor 𝜋𝑑 (distance household-source) is not strongly correlated with any other
predictor. However, we observe that there are signs of correlation between the predictors 𝜋𝑤𝑐 (water cost),
𝜋𝑝𝑤𝑞 (perceived water quality), 𝜋𝑒𝑒 (extraction easiness), and 𝜋𝑡𝑐𝑤 (time collecting water). Therefore, given
the risk analysis performed above, we propose to consider the predictors 𝜋𝑑 and 𝜋𝑝𝑤𝑞 , instead of other
possible combinations of predictors (e.g. 𝜋𝑑 /𝜋𝑤𝑐 , 𝜋𝑑 /𝜋𝑒𝑒 …). Equation (1) therefore becomes:
𝜌(ℎ, 𝑠) = 𝛽0 + 𝛽𝑑 ⋅ 𝜋𝑑 (ℎ, 𝑠) + 𝛽𝑝𝑤𝑞 ⋅ 𝜋𝑝𝑤𝑞 (ℎ, 𝑠) (2)
where 𝜋𝑑 (ℎ, 𝑠) is the distance between the household ℎ and the source 𝑠 and 𝜋𝑝𝑤𝑞 (ℎ, 𝑠) is the perception
of household ℎ of the water quality at the source 𝑠. 𝛽𝑑 and 𝛽𝑝𝑤𝑞 are the regression coefficients associated
to the 𝜋𝑑 and 𝜋𝑝𝑤𝑞 predictors respectively.
Fourthly, we identify the 𝛽𝑖 coefficients. For this identification, the values of the probability 𝜌(ℎ, 𝑠) are
needed (see equation (1)). For this purpose, we use the data collected on the households 𝒉𝒔𝒖𝒓 (i.e.
households part of the survey pool). Indeed, thanks to these data we know the source to which each
household goes. We therefore deduce the ‘measured’ values of 𝜌(ℎ, 𝑠) as following: for the household ℎ,
for the source 𝑠 where it goes we set 𝜌(ℎ, 𝑠) = 1 and, for all the other water sources we set 𝜌(ℎ, 𝑠) = 0. We
identify the 𝛽𝑖 coefficients by using the data of 70% of the households 𝒉𝒔𝒖𝒓 . The 70% of households are
selected randomly. Then, we predict the source choice of the remaining 30% of the households thanks to
the values of the predictors and the identified 𝛽𝑖 coefficients (see equation (1)). The predicted source choice
of the household ℎ is the one that maximizes 𝜌(ℎ, 𝑠). The predictions are compared to the real source choice
of the households before the installation of the PVWPS. For each household, either the regression predicts
the real source choice and it is a success; or it does not predict the real source choice and it is a failure.
Dividing the number of successes, i.e. when the prediction matches the real source choice, by the number
of predictions gives us the ‘prediction rate before the installation of the PVWPS’ 𝜙𝑏𝑒𝑓 . For instance, if we
were to randomly select the source choice of Gogma’s households, the prediction rate 𝜙𝑏𝑒𝑓 would be equal
to 1/21=5%, as there were 21 sources before the installation of the PVWPS.
In the case of Gogma, we identify the regression coefficients and we compute the prediction rate 𝜙𝑏𝑒𝑓
separately for the dry season and the wet season, as the source choice depends on the season (see
57
section III.1). Additionally, 𝒉𝒔𝒖𝒓 contains 88 households for Gogma. Therefore, the values of the 𝛽𝑖
coefficients and of 𝜙𝑏𝑒𝑓 depend on the 70% of the households selected for identifying the 𝛽𝑖 coefficients
and of the 30% of the households selected for computing 𝜙𝑏𝑒𝑓 . To mitigate this problem, we perform the
following sequence 1000 times:
 Select 70% of the 88 households randomly. Determine the values of 𝛽0 , 𝛽𝑑 and 𝛽𝑝𝑤𝑞 with the
data from these 70% of households.
 Use the remaining 30% of household to compute the prediction rate 𝜙𝑏𝑒𝑓 .
Then, the final value of 𝛽0 is the average of the 1000 values of 𝛽0 obtained (same for 𝛽𝑑 and 𝛽𝑝𝑤𝑞 ). The
final prediction rate is the average of the 1000 prediction rates obtained. The final values encountered for
𝛽0 , 𝛽𝑑 and 𝛽𝑝𝑤𝑞 and for the prediction rate 𝜙𝑏𝑒𝑓 are given in Table III-3. In addition, for all of the 1000
repetitions, the p-values of the ‘distance household-source’ predictor and of the ‘perceived water quality’
predictor are smaller than 0.01. This shows that both predictors have statistical significance [88].
Table III-3 – Values of the regression coefficients and of the prediction rate before installation of the PVWPS
– case of Gogma.
season 𝛽0 𝛽𝑑 (m )
-1
𝛽𝑝𝑤𝑞 Prediction rate before installation of the
PVWPS 𝜙𝑏𝑒𝑓
dry 0.28 -0.00015 -0.13 62%
wet 0.24 -0.00013 -0.09 48%
The coefficient 𝛽𝑑 is negative. This means that the higher the distance 𝜋𝑑 (ℎ, 𝑠) between the household ℎ
and the source 𝑠, the lower the probability 𝜌(ℎ, 𝑠) that the household ℎ chooses the source 𝑠. The coefficient
𝛽𝑝𝑤𝑞 is also negative. This means that the worse the perception that the household ℎ has of the water quality
at the source 𝑠 (𝜋𝑝𝑤𝑞 (ℎ, 𝑠) close to 1), the lower the probability 𝜌(ℎ, 𝑠). These results on the distance and
the perceived water quality are consistent with the literature.
III.2.1.2 Integrating the PVWPS
In order to be able to predict the choice of water source of all the households of the village after the
installation of the PVWPS 𝒉𝒔∗𝒂𝒇𝒕 , we have to integrate the PVWPS and the households that are not part of
𝒉𝒔𝒖𝒓 (i.e. not part of the survey pool).
We first focus on the integration of the PVWPS. We need to be able to compute the value of each predictor
𝜋𝑖 for an household ℎ and the PVWPS: 𝜋𝑖 (ℎ, 𝑃𝑉𝑊𝑃𝑆). For Gogma, the distance between the PVWPS,
located at the position (𝐿𝑎𝑡, 𝐿𝑜𝑛), and any household ℎ of the village can be computed from the GPS
coordinates (see section II.2.2). In addition, as justified in section III.2.1.1, we consider that all the
households perceive the water quality at the PVWPS as high: 𝜋𝑝𝑤𝑞 (ℎ, 𝑃𝑉𝑊𝑃𝑆) = 0, ∀ℎ.
Consequently, thanks to the application of equation (2), we can predict the source where each household
wishes to go, for any position (𝐿𝑎𝑡, 𝐿𝑜𝑛) of the PVWPS in the village. For the households that are predicted
not to wish to go to the PVWPS, we do not keep the prediction of the model in 𝒉𝒔∗𝒂𝒇𝒕 . Instead we consider
that these households remain using the same water source where they used to go before the installation of
the PVWPS.
58
It is possible to propose a first validation of the prediction of the households that wish to go to the PVWPS
thanks to the available data. Indeed, for the position of the current PVWPS (see Figure II-10), we can predict
the households that wish to go to the PVWPS and we also know the households that in reality go to the
PVWPS, thanks to the account book of the PVWPS (see section II.2.4) and the household surveys (see
section II.2.5). Moreover, as the current PVWPS is oversized, all the households that wish to go there are
able to do so (see section II.2.8). In order to compare the prediction to the reality we distinguish four cases
presented in Table III-4.
Table III-4 – Four possibilities for the prediction result at the current PVWPS of Gogma.
Prediction from the model Reality as per surveys and account books Success or
failure of the
prediction
Household wishes to go to the PVWPS Household wishes to go to the PVWPS Success
Household wishes to go to the PVWPS Household does not wish to go to the PVWPS Failure
Household does not wish to go to the PVWPS Household wishes to go to the PVWPS Failure
Household does not wish to go to the PVWPS Household does not wish to go to the PVWPS Not relevant
As the current PVWPS is oversized, all the households that wish to go there are able to do so.
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑢𝑐𝑐𝑒𝑠𝑠𝑒𝑠
We define the prediction rate at the PVWPS as: 𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆 = 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑢𝑐𝑐𝑒𝑠𝑠𝑒𝑠 + 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑎𝑖𝑙𝑢𝑟𝑒𝑠. We
obtain a prediction rate at the PVWPS 𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆 of 91% for the dry season and of 91% for the wet season.
This means that we predict with more than 90% accuracy, the households that wish to go to the PVWPS.
We can expect the accuracy of the model to be similar for other positions of the PVWPS in the village.
Indeed, the identification of the 𝛽𝑖 regression coefficients was performed by using data in the whole village
and not only near the position of the current PVWPS. The prediction rate at the PVWPS 𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆 is the
most relevant metric for our study because it influences the interdisciplinary model output 𝑆𝐸𝐼 (socio-
economic impact), contrarily to 𝜙𝑏𝑒𝑓 .
Finally, in Table III-5, we also present the results of the prediction rates 𝜙𝑏𝑒𝑓 and 𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆 for the other
sets of predictors that we could have selected (see section III.2.1.1). For the combinations of predictors
𝜋𝑑 /𝜋𝑤𝑐 , 𝜋𝑑 /𝜋𝑒𝑒 and 𝜋𝑑 /𝜋𝑡𝑐𝑤 , the values of 𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆 are lower than the values of 𝜙𝑏𝑒𝑓 . This result is
aligned with the choice of the combination 𝜋𝑑 /𝜋𝑝𝑤𝑞 instead of the combinations 𝜋𝑑 /𝜋𝑤𝑐 , 𝜋𝑑 /𝜋𝑒𝑒 and
𝜋𝑑 /𝜋𝑡𝑐𝑤 that was made in section III.2.1.1.
Table III-5 – Prediction success rates for several combinations of predictors.

Predictors considered Prediction success rate before the Prediction success rate at
installation of the PVWPS 𝜙𝑏𝑒𝑓 the PVWPS 𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆
Distance household-source / Water cost Dry: 52% Dry: 25%
(𝜋𝑑 /𝜋𝑤𝑐 ) Wet: 45% Wet: 25%
Distance household-source / Perceived water quality Dry: 62% Dry: 91%
(𝜋𝑑 /𝜋𝑝𝑤𝑞 ) Wet: 48% Wet: 91%
Distance household-source / Extraction easiness Dry: 64% Dry: 36%
(𝜋𝑑 /𝜋𝑒𝑒 ) Wet: 57% Wet: 37%
Distance household-source / Time collecting water Dry: 49% Dry: 0%
(𝜋𝑑 /𝜋𝑡𝑐𝑤 ) Wet: 48% Wet: 0%
We now focus on the integration of the households that are not part of 𝒉𝒔𝒖𝒗 (i.e. not part of the survey pool).
We use the GPS coordinates of these households. For each of these households, we suppose that it behaves
59
in the same way as the household of 𝒉𝒔𝒖𝒗 that is the closest geographically, both before and after the
installation of the PVWPS. We also consider that it has the same values of 𝜌(ℎ, 𝑠).
III.2.2 Determination of the water demand profile at the PVWPS

In the previous section, we predicted the households that wish to go to the PVWPS. In this section, we
determine the water demand profile at the PVWPS, i.e. the evolution of the water demand flow rate 𝑄𝑑 at
the PVWPS over the day.
For this purpose, we use the information on the daily water quantity collected and the time at which water
was collected before the installation of the PVWPS, that were gathered through the survey. In addition we
make the following assumptions:
 Each household that wishes to go to the PVWPS collects the same daily water quantity as it used to
at its previous water source. This assumption is based on the fact that no piping system is installed
along with the PVWPS and that households therefore still have to walk to the fountain of the
PVWPS to collect water.
 Each household that wishes to go to the PVWPS collects water at the same time as it used to at its
previous water source.
We acknowledge that these assumptions merit further investigation in future works, notably because the
habits of the inhabitants may evolve with the installation and over the lifetime of the PVWPS.
For the position of the current PVWPS of Gogma, we simulate the water demand curve at the PVWPS for
the dry season (in blue in Figure III-3) and for the wet season (in blue in Figure III-4). The difference in the
simulated demand curve between the dry season and the wet season is due to the following factors. Firstly,
the vector 𝒉𝒔∗𝒂𝒇𝒕 varies from one season to another. Secondly, the time at which water is collected, which is
reported in the surveys, also varies from one season to another for certain households.
For the position of the current PVWPS of Gogma, we can compare the simulated demand curve to the
measures of the demand flow rate performed by the data logger of the current PVWPS3 (see section II.2.8).
For a given season, we determine the hourly averaged daily demand profile from the measured demand flow
rate as following: for each hour 𝑖 ∈ [1, 24], the average demand flow rate for this hour is the average of the
flow rates measured at this hour for all the days of the season4. The measured daily average demand profile
for the dry season is represented in red in Figure III-3 and the one for the wet season is in Figure III-4. In
Table III-6 we also give the daily integrals of the simulated and measured demand profiles for both seasons.
Results indicate that the daily water quantity is overestimated by a factor of 1.52 to 2.05.
3
Actually, it is the consumption flow rate 𝑄𝑐 that is measured. However, as the current PVWPS of Gogma is oversized,
the demand flow rate and the consumption flow rate are equal (𝑄𝑑 = 𝑄𝑐 ).
4
For the dry season, which last from 1 December 2017 to 30 June 2018, we use the data from the monitoring system
acquired between 14 January 2018 and 30 June 2018. Indeed, we started acquiring data with the monitoring system on
14 January 2018.
60
Figure III-3 – Demand flow rate profile simulated by Figure III-4 – Demand flow rate profile simulated by
the demand model and measured by the data logger the demand model and measured by the data logger
(i.e. flow meter data) – dry season. (i.e. flow meter data) – wet season.
Table III-6 – Daily integrals of the simulated and measured demand profiles for both seasons.
Season Daily integral of the simulated Daily integral of the measured
water demand profile (m3) water demand profile (m3)
Dry 13.1 8.6
Wet 13.1 6.4
The differences between the simulated and measured demand (see Figure III-3, Figure III-4 and Table III-6)
may be explained by the following elements:
 The error on the prediction success rate at the PVWPS 𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆 , which is about 10% (see
section III.2.1.2).
 Some households may not accurately estimate the daily water quantity that they use. According to
our on-field observations (see section II.2.1), we think that households notably overestimated the
water quantity used for personal hygiene, which represents ~50% of the domestic water use in the
survey answers. In its literature review of water demand models, [54] highlighted that there are
significant errors on households water consumption reported in surveys.
 Some households may not have reported accurately the times at which they collect water in the
survey. Indeed, we observed through on-field observations that inhabitants of Gogma rarely wear
watch and Ho et al. [86] reported that inhabitants who do not wear a watch often do not know the
time accurately.
 Some households may also collect a share of the water at other sources, and notably water for
personal hygiene at open wells. This may especially be the case during the wet season as all open
wells are available and there is more water in open wells during this season. This could help to
explain why the measured daily water demand at the PVWPS is lower during the wet season than
during the dry season (see Table III-6).
 In the survey, we asked people about the quantity of water collected and the time at which they
collected water. We then split equally the quantity of water collected between these times. In
Gogma, the majority of households reported that they collect water in the morning and in the
evening. This notably explains the two spikes in the simulated water demand in Figure III-3 and
Figure III-4. However, when comparing the simulated water demand to the measured one, a possible
61
hypothesis may be that in reality people collect more water in the evening than in the morning. This
was also suggested by on-field observations (see section II.2.1)
 Users of the PVWPS may have changed their habits and they may not use the PVWPS in the same
way as they used their previous water source.
In the future, we will investigate the influence of these elements by completing the household survey and
applying the demand model in other villages. We will also continue to compare model results, which are
obtained from household survey answers, to flow meter data. Indeed, this comparison is original and
relevant, in our opinion.
This will also allow us to gain more experience on this type of demand model. Indeed, as it is the first model
that predicts the attendance at a source of a new type and load curves at sources, there is no point of
comparison for the moment. Besides, in section IV.3, we investigate the effect of the difference between the
simulated and measured demand on the optimisation results.
62
III.3 Technical model

The block diagram of the technical model is shown in Figure III-5. The energy conversion model allows to
determine the evolution of the water level in the tank 𝐻𝑡 from the sizing variables, the groundwater
parameters, the climatic data and the water demand at the PVWPS 𝑄𝑑 . The beneficiaries identification
model allows to determine the water consumption at the PVWPS 𝑄𝑐 and the water sources where the
households effectively go after the installation of the PVWPS 𝒉𝒔𝒂𝒇𝒕 from the water level in the tank 𝐻𝑡 , the
water demand at the PVWPS 𝑄𝑑 and the water sources where the households wish to go after the installation
of the PVWPS 𝒉𝒔∗𝒂𝒇𝒕 .
In section III.3.1, we present the energy conversion model, apply it and validate it for the current PVWPS
of Gogma. In section III.3.2, we present the generalization of the energy conversion model. In section III.3.3,
we present the beneficiaries identification model.
Technical model
Sizing
Water level in tank
Groundwater
Beneficiaries
parameters Energy conversion
identification
, model
model
Climatic data
,
Figure III-5 – Block diagram of the technical model.

𝒉𝒔∗𝒂𝒇𝒕 : water sources where the households wish to go after installation of the PVWPS, 𝑄𝑑 : water demand at the
PVWPS, 𝒉𝒔𝒂𝒇𝒕 : water sources where the households effectively go after installation of the PVWPS, 𝑄𝑐 : water
consumption at the PVWPS.
63
III.3.1 Energy conversion model – presentation and application to the current PVWPS of Gogma
The block diagram of the energy conversion model is presented in Figure III-6. For clarity reasons, in this
diagram, we only present on the arrows the time dependent quantities. The sizing variables and the
groundwater parameters are inside the sub-models blocks and they are in purple and in blue respectively.
Energy conversion
model Controller
[ , ]
Hydraulic
System
[
PV array
, , , ]
with MPPT control
Thermal
[ ]
Climatic
Motor-pump Tank Water level in tank
data
[ ] [ ( ),
,
+: Parameters identified
Electrical *: Monitored by data
[ ] logger
Optimisation variables,
Demand at the Sizing
PVWPS Groundwater
parameters
Figure III-6 – Block diagram of the energy conversion model.
In Figure III-6, we also present within square brackets, the parameters encompassed in each sub-model. We
can distinguish three types of parameters:
 The groundwater parameters (𝐻𝑏,𝑠 , 𝜅0 and 𝜇0 ) which are in blue in Figure III-6.
 The parameters that are also optimisation variables, such as 𝑃𝑝𝑣,𝑝 , or that are dependent on
optimisation variables. This is for instance the case of the height between the bottom of the tank
and the stop level 𝐻𝑡,𝑠 which depends on the tank volume 𝑉𝑡 .
 The remaining the parameters such as the nominal operating temperature cell 𝑁𝑂𝐶𝑇.
In Figure III-7, we represent the heights that feature in the energy conversion model.
It is important to note that the inclusion of the water demand flow rate at the PVWPS 𝑄𝑑 as an input of the
energy conversion chain model is an original contribution. It allows to model the instantaneous operation of
PVWPS which include a tank and a controller that stops and restarts the motor-pump depending on the water
level in the tank [78], which are commonly used for domestic water access. It also permits to link the demand
model to the technical one (see Figure III-1).
64
Figure III-7– Definition of the heights.

The height datum is set at the ground level so 𝐻𝑏,𝑠 and 𝐻𝑏,𝑑 are negative.
III.3.1.1 Sub-models
In this section, we detail the sub-models of the energy conversion model. We also give the values of the
parameters for the current PVWPS of Gogma. Some of the parameters were measured directly on the
PVWPS or obtained from the literature. The other parameters are identified by performing regressions using
the data acquired by the data logger (see section II.2.8) from 12 February 2018 to 18 February 2018
(identification set). The square of the multiple correlation coefficient R2 is used to estimate the accuracy of
identifications. We indicate the data acquired by the data logger and the parameters identified in Figure III-6.
III.3.1.1.1 Photovoltaic array
Thermal model
The PV array thermal model computes the temperature of the PV modules 𝑇𝑝𝑣 using:
𝑁𝑂𝐶𝑇 − 20
𝑇𝑝𝑣 (𝑡) = 𝑇𝑎 (𝑡) + 𝐺𝑝𝑣 (𝑡) (3)
800
where 𝑇𝑎 is the ambient temperature and 𝐺𝑝𝑣 is the irradiance on the plane of the PV array. The 𝑁𝑂𝐶𝑇
(nominal operating cell temperature), generally specified by the manufacturer, corresponds to the
temperature of the open circuited modules under the following conditions: irradiance on the plane of the
modules of 800 W/m2, ambient temperature of 20°C, wind velocity of 1 m/s and the modules are mounted
such that their back side is open.
For the current PVWPS of Gogma, the 𝑁𝑂𝐶𝑇 was not given in the datasheet of the PV modules. The 𝑁𝑂𝐶𝑇
is therefore determined by identification. To do so, we perform a regression using equation (3) and the
65
experimental data of 𝑇𝑝𝑣 , 𝑇𝑎 and 𝐺𝑝𝑣 from the identification set. The identification yields to a 𝑁𝑂𝐶𝑇 of
32 °C (R2 = 0.98). This is low compared to usual values for multicrystalline silicon PV modules, which are
in the order of 43 °C to 45 °C [89]. This can be explained by a wind speed higher than 1 m/s and which is
not taken into account in the thermal model. Indeed, we installed a complete weather station at the PVWPS
in February 2019 in addition to existing sensors, and we measured an average wind speed of 6.6 m/s between
February 2019 and August 2019. Besides, the PV modules are placed at 2 m above ground which favours
cooling by convection.
Through a sensitivity analysis, we showed that the thermal parameters of PV modules (𝑁𝑂𝐶𝑇 and 𝛽) have
a small impact on the output of the energy conversion model and on the optimal sizing of PVWPS [81].
Consequently, it is not indispensable to model the thermal behaviour of PV modules and, in this case, the
ambient temperature 𝑇𝑎 is not required as an input to the energy conversion model.
Electrical model
For the PV array electrical model, considering that the maximum power point tracking of the PV array is
correctly performed by the converter of the motor-pump, a simplified model is used [90, 91]:
𝐺𝑝𝑣 (𝑡)
𝑃𝑝𝑣 (𝑡) = 𝑃𝑝𝑣,𝑝 (1 + 𝛾 (𝑇𝑝𝑣 (𝑡) − 25)) 𝑏(𝑡) (4)
𝐺0
where 𝑃𝑝𝑣 is the PV array output power (i.e. input power to the motor-pump), 𝐺0 is the reference irradiance
(1000 W/m2), 𝑃𝑝𝑣,𝑝 is the peak power of the PV array in standard test conditions (STC), 𝛾 is the coefficient
of loss on the maximum power related to modules temperature and 𝑏 is the controller trigger signal. The
signal 𝑏 allows to transfers (𝑏 = 1) or not (𝑏 = 0) the power of the PV array to the motor-pump depending
on the water level in the tank 𝐻𝑡 (see section III.3.1.1.2).
For the PVWPS of Gogma, the soiling losses are neglected as the modules are cleaned at least twice a month
by the person of the village who is responsible of maintaining the PVWPS. In addition, the PV modules
were bought in Burkina Faso and the only documentation that was provided was a tag on the back of the
modules. 𝛾 was not given in the tag and is thus taken equal to -0.4%/°C, which is the standard value for
multicrystalline silicon modules [90]. According to the tag, the total peak power of the PV array in STC
𝑃𝑝𝑣,𝑝 is equal to 750 Wp. Nevertheless, since in these regions the tag is not always reliable, it was preferred
to determine the value of 𝑃𝑝𝑣,𝑝 by identification. A value of 620 Wp was obtained (R2 = 0.96).
III.3.1.1.2 Controller
The controller is modelled by a switch which transfers (𝑏 = 1) or not (𝑏 = 0) the power of the PV array to
the motor-pump depending on the water level in the tank 𝐻𝑡 . The value of 𝑏 is governed by the hysteresis
function presented in Figure III-8, where 𝐻𝑡,𝑠 is the water level in the tank for which the motor-pump stops
(the tank is full) and 𝐻𝑡,𝑟 is the level for which it restarts.
66
Figure III-8 – Controller model.
For the current PVWPS of Gogma, we measured 𝐻𝑡,𝑠 = 3.3 m and 𝐻𝑡,𝑟 = 3.0 m.
III.3.1.1.3 Tank
As the tank is sealed, it is considered that no water leaves the tank by evaporation. The height of water in
the tank 𝐻𝑡 is thus expressed as:
𝐻𝑡 (𝑡0 ) = 𝐻𝑡,𝑠
𝑡 𝑄 (𝜏) − 𝑄𝑑 (𝜏)
𝑝 (5)
𝐻𝑡 (𝑡) = max (0, 𝐻𝑡 (𝑡0 ) + ∫ 𝑑𝜏)
𝑡0 𝑆𝑡
where 𝑄𝑝 is the pump flow rate, 𝑄𝑑 is the water demand flow rate and 𝑆𝑡 is the cylindrical tank base area5.
The model is initialized at a time 𝑡0 at which the tank is full and the water level in the tank is therefore equal
to the stop level 𝐻𝑡,𝑠 .
For the current PVWPS of Gogma, we measured 𝑆𝑡 = 3.3 m2.
III.3.1.1.4 Motor-pump
The characteristic of the motor-pump reference 𝑀𝑃 provided by the manufacturer is used to build the motor-
pump model. This characteristic links the total dynamic head 𝑇𝐷𝐻, the pump flow rate 𝑄𝑝 and the input
power 𝑃𝑝𝑣 . The points of the characteristic for which 𝑄𝑝 > 0 are fitted by a polynomial 𝑃𝑎 (the coefficients
of 𝑃𝑎 are noted 𝑘𝑚,𝑛 ). The pump flow rate is thus given by:
𝑄𝑝 (𝑡) = max (0, 𝑃𝑎 (𝑃𝑝𝑣 (𝑡), 𝑇𝐷𝐻(𝑡))) (6)
The motor-pump SQFlex 5A-7 [69] is used in the current PVWPS in Gogma. We use a 4th order polynomial
𝑃𝑎4,4 for fitting the characteristic of this motor-pump. The coefficients 𝑘𝑚,𝑛 of the polynomial are given in
Appendix B. Figure III-9 presents the points from the datasheet and the surface obtained by fitting. The
fitting is very accurate (R2 = 1.00).
𝑡 𝑄𝑝 (𝜏)−𝑄𝑑 (𝜏)
5
Taking the maximum between ‘0’ and the second member ‘𝐻𝑡 (𝑡0 ) + ∫𝑡 𝑑𝜏’ allows to make sure that the
0 𝑆𝑡
water level in the tank 𝐻𝑡 never becomes negative when the PVWPS is undersized (see section III.3.2.3).
67
Figure III-9 – Model of the motor-pump SQFlex 5A-7.
III.3.1.1.5 Hydraulic system
The model of the hydraulic system allows to compute the total dynamic head 𝑇𝐷𝐻 between the motor-pump
and the tank. The total dynamic head 𝑇𝐷𝐻 between the motor-pump and the tank is given by [92]:
𝑇𝐷𝐻(𝑡) = − (𝐻𝑏,𝑠 + 𝐻𝑏,𝑑 (𝑡)) + 𝐻𝑡,𝑏 + 𝐻𝑡,𝑖 + 𝐻𝑝𝑎1 (𝑡), ∀𝑃𝑝𝑣 (7)
where 𝐻𝑏,𝑠 is the static water level (height between the ground level and the water level in the borehole
when there is no pumping; 𝐻𝑏,𝑠 <0), 𝐻𝑏,𝑑 is the drawdown (height between the static water level and the
water level in the borehole when there is pumping; 𝐻𝑏,𝑑 <0), 𝐻𝑡,𝑏 is the height between the ground level and
the bottom of the tank, 𝐻𝑡,𝑖 is the height between the bottom of the tank and the tank input (independent of
the water level in the tank) and 𝐻𝑝𝑎1 is the additional head due to pressure losses in pipe assembly 1. These
heights are shown in Figure III-7. 𝐻𝑏,𝑑 and 𝐻𝑝𝑎1 can be expressed as function of 𝑄𝑝 only (see equation (8)
and (9)) and equation (7) holds for any value of 𝑃𝑝𝑣 . Equation (7) therefore represents the equation of a
surface versus variables 𝑄𝑝 and 𝑃𝑝𝑣 , which is called the hydraulic system surface.
For the PVWPS of Gogma, we measured 𝐻𝑡,𝑏 = 4.2 m and 𝐻𝑡,𝑖 = 3.4 m.
Drawdown model
It is considered that the drawdown 𝐻𝑏,𝑑 depends on the pump flow rate 𝑄𝑝 and of its square 𝑄𝑝2 [93] and
that the drawdown at a time 𝑡 also depends on the flow rates at previous times [94]. Based on this, we
propose the following expression for the drawdown 𝐻𝑏,𝑑 :
𝑁 𝑁
𝐻𝑏,𝑑 (𝑡) = − ∑ 𝜅𝑛 𝑄𝑝 (𝑡 − 𝑛 Δ𝑇) − ∑ 𝜇𝑛 𝑄𝑝 (𝑡 − 𝑛 Δ𝑇)2 (8)

𝑛=0 𝑛=0
68
where 𝜅𝑛 are coefficients that represent the aquifer losses, 𝜇𝑛 are coefficients that represent the borehole
losses and Δ𝑇 is a time difference. The values chosen for 𝑁 and Δ𝑇 depend on the data availability, the
accuracy required for the model and the speed of the drawdown response.
For the current PVWPS of Gogma, 𝜅𝑛 and 𝜇𝑛 were identified from the step drawdown tests that were
performed before the installation of the PVWPS, at the moment of the borehole drilling in November 2017
(see section II.2.6). Regressions of 𝐻𝑏,𝑑 against 𝑄𝑝 and 𝑡 with a Δ𝑇 of 10 minutes for different values of 𝑁
were tried [78, 94]. In the end, a model with only 𝜅0 and 𝜇0 was selected. Indeed the aquifer response is fast
and the statistical significance of the following coefficients (𝜅1 , 𝜇1 , 𝜅2 , 𝜇2 …) is low. The values obtained
for 𝜅0 and 𝜇0 are:
𝜅0 = 2.0 103 m−2 s; 𝜇0 = 5.8 105 m−5 s 2 (R2 = 0.97)
Figure III-10 presents the drawdown measured during the 4 step pumping tests and the drawdown simulated
by the model.
Figure III-10 – Drawdown measured during the pumping tests and simulated by the model.
As we have demonstrated in [95], the main advantage of the developed data-driven drawdown model is that,
contrarily to existing models, it is not based on assumptions that are rarely met (e.g. homogenous and
isotropic aquifer) and can be applied for all types of aquifers.
Pipe model
The additional head due to losses in pipe assembly 1 𝐻𝑝𝑎1 evolves quadratically with the pump flow rate
𝑄𝑝 [96]:
𝐻𝑝𝑎1 (𝑡) = 𝜈𝑄𝑝 (𝑡)2 (9)
where  is a constant. These losses occur along the length of the pipes and at junctions (elbows, curvatures,
diameter changes between pipes and pipe output) [96].
In the case of the current PVWPS of Gogma,  and 𝐻𝑏,𝑠 are determined by identification by using
equations (7) and (9) and data acquired by the monitoring system. As 𝑇𝐷𝐻 is not measured directly, it is
obtained from 𝑃𝑝𝑣 and 𝑄𝑝 by fitting the motor-pump characteristic with another 4th order polynomial 𝑃𝑏4,4
(see Appendix B for the values of the coefficients):
69
𝑇𝐷𝐻(𝑡) = 𝑃𝑏4,4 (𝑃𝑝𝑣 (𝑡), 𝑄𝑝 (𝑡)) , ∀𝑄𝑝 > 0 (R2 = 1.00) (10)
By combining equation (7) and (10), the operating points of the PVWPS, which correspond to the
intersection of the motor-pump surface and of the hydraulic system surface, can be found:
𝑃𝑏4,4 (𝑃𝑝𝑣 (𝑡), 𝑄𝑝 (𝑡)) = −(𝐻𝑏,𝑠 − 𝜅0 𝑄𝑝 (𝑡) − 𝜇0 𝑄𝑝 (𝑡)2 ) + 𝐻𝑡,𝑏 + 𝐻𝑡,𝑖 + 𝜈𝑄𝑝 (𝑡)2 (11)
The values of 𝐻𝑏,𝑠 and  are identified by using the measured 𝑃𝑝𝑣 and 𝑄𝑝 from the identification set:
𝐻𝑏,𝑠 = − 4.9 m; 𝜈 = 4.9 106 m−5 s2 (R2 = 0.84)

The value of the static water level identified (-4.9 m) is close to the one measured during the step drawdown
tests (see section II.2.6), which is of -6 m. The difference (1.1 m) may be due to measurement error, seasonal
change, model inaccuracy or a combination of these factors. The value of 𝜈 is greater than usual values for
this type of system, which corresponds to significant head losses due to the pipe assembly 1 (PA1) [96].
For installations for which pumping tests are not available, it is possible to determine the coefficients
(𝜅𝑛 , 𝜇𝑛 , 𝐻𝑏,𝑠 , 𝜈) from the intersection of the motor-pump surface and the hydraulic system surface only
(equation (11) for the PVWPS of Gogma). However, as 𝜇0 and 𝜈 are both coefficients for 𝑄𝑝 (𝑡)2 , only the
sum of 𝜇0 and 𝜈 is obtained and not the individual values of the coefficients. This prevents to separate the
contributions of the drawdown and the pipe losses to the total dynamic head. It would therefore not be
possible to know if the drawdown goes below the level of the motor-pump.
III.3.1.1.6 Summary of the parameters value for the current PVWPS of Gogma
We summarise the parameters of the energy conversion model and their value for the current PVWPS of
Gogma in Table III-7.
Table III-7 – Parameters of the energy conversion model and value for the current PVWPS of Gogma.
Symbol Description Value for the current
PVWPS of Gogma
𝑁𝑂𝐶𝑇 Nominal operating cell temperature 32 °C
𝛾 Coefficient of loss due to PV modules temperature -0.4%/°C
𝑃𝑝𝑣,𝑝 Peak power of the PV modules in standard test conditions (STC) 620 Wp
𝑘𝑚,𝑛 Coefficients of the polynomial which fit the characteristic of the motor-pump see Appendix B
reference 𝑀𝑃
𝐻𝑡,𝑠 Height between the bottom of the tank and the stop level 3.3 m
𝐻𝑡,𝑟 Height between the bottom of the tank and the restart level 3.0 m
𝑆𝑡 Area of the base of the cylindrical tank 3.3 m2
𝐻𝑡,𝑖 Height between the bottom of the tank and the level at which water enters the 3.4 m
tank
𝐻𝑡,𝑏 Height between the ground level and the bottom of the tank 4.2 m
𝐻𝑏,𝑠 Height between the ground level and the static water level in the borehole -4.9 m
𝜅0 Aquifer losses coefficient 2.0 103 m-2 s
𝜇0 Borehole losses coefficients 5.8 105 m-5 s2
𝜈 Pipe pressure losses coefficient 4.9 106 m-5 s2
70
III.3.1.2 Experimental validation
In this section, we detail the experimental validation of the energy conversion model for the current PVWPS
of Gogma. The validation is performed by using data from different periods than the identification set, called
validation sets. The model validation is detailed for the first validation set which lasts from 19 February
2018 to 21 February 2018. Moreover, the results of the model validation are also given for two other
validation sets of two weeks. The first one lasts from 16 May to 29 May 2018 and takes place during the
dry season. The second one lasts from 29 July to 11 August 2018 and takes place during the wet season. To
validate the model, we compare the model output, the water level in the tank 𝐻𝑡 , to the measures of the data
logger for each validation set.
Figure III-11 present the irradiance on the plane of the PV array 𝐺𝑝𝑣 and the ambient temperature 𝑇𝑎 and
the demand flow rate6 𝑄𝑑 measured by the data logger for the validation set which last from 19 February to
21 February.
(a) (b)
(c)
Figure III-11 – Measured model inputs (a) irradiance (b) ambient temperature (c) demand flow rate.
The computations for the model for the February validation set start on 19 February at 9h20 as the motor-
pump has just stopped at this moment and the water level in the tank has just reached the stop level (see
6
Actually, it is the consumption flow rate 𝑄𝑐 that is measured. However, as the current PVWPS of Gogma is oversized,
the demand flow rate and the consumption flow rate are equal (𝑄𝑑 = 𝑄𝑐 ).
71
section III.3.1.1.3). Figure III-12 compares the measured water level in the tank 𝐻𝑡 to the one obtained from
the model.
Figure III-12 – Water level in the tank measured and simulated (model output).
The root mean square error 𝑅𝑀𝑆𝐸 between the measured and modelled height in the water tank is of 0.03 m.
The 𝑅𝑀𝑆𝐸 values are normalized by the height between the bottom of the tank and the stop level 𝐻𝑡,𝑠 , which
is of 3.3 m, in order to obtain the normalized root mean square errors (𝑁𝑅𝑀𝑆𝐸). For the validation period
of February, the model allows to predict the water level in the tank with 1.0% error.
The same calculations were carried out for the two-week validation sets in May and July-August 2018. The
𝑁𝑅𝑀𝑆𝐸 for the May period is of 4.8%. The 𝑁𝑅𝑀𝑆𝐸 for the July-August period is 3.8%. Moreover, for all
3 validation periods, the absolute value of the difference between the instantaneous measured and modelled
water heights in the tank does not durably increase along the validation period.
III.3.2 Energy conversion model – generalization

III.3.2.1 Parameters values
In order to generalize the PVWPS model, we need to determine the values of the parameters for other
PVWPS than the current PVWPS of Gogma. We remind that the values of the parameters for the current
PVWPS are given in Table III-7.
Firstly, for any PVWPS, we consider that the values of the parameters 𝑁𝑂𝐶𝑇, 𝛾, 𝜈 are the same as the ones
for the current PVWPS of Gogma (see Table III-7).
Secondly, the peak power of the PV array 𝑃𝑝𝑣,𝑝 is both an optimisation variable and a parameter of the
model which is used in equation (4). The energy conversion model can therefore be evaluated for any value
of 𝑃𝑝𝑣,𝑝 .
Thirdly, we can digitize the characteristic of any motor-pump reference 𝑀𝑃 and determine the associated
fitting coefficients 𝑘𝑚,𝑛 , in the same way as we did for the motor-pump SQFlex 5A-7 in section III.3.1.1.4.
For this thesis, in addition to the SQFlex 5A-7 [69], we digitize the characteristic of the motor-pumps
SQFlex 8A-5 [97], SQFlex 2.5-2 [98], SQFlex 1.2-2 [99], SQFlex 0.6-2 [100], SQFlex 11A-3 [98], SQFlex
8A-3 [99] and SQFlex 5A-3 [100].
72
Fourthly, we determine the values of the parameters 𝐻𝑡,𝑖 , 𝐻𝑡,𝑠 , 𝐻𝑡,𝑟 and 𝑆𝑡 from the value of the tank volume
𝑉𝑡 . For this purpose, we collected data from several tank manufacturers regarding the base radius 𝑅𝑡 and the
height of the tank 𝐻𝑡,𝑡 chosen for different tank volumes 𝑉𝑡 . The results are presented in Figure III-13.
Figure III-13 – Base radius 𝑅𝑡 and height 𝐻𝑡,𝑡 chosen for different tank volumes 𝑉𝑡 .
We fit the collected data presented in Figure III-13 and we obtain the expression of 𝐻𝑡,𝑡 as a function of 𝑅𝑡 :
𝐻𝑡,𝑡 = 2.49 𝑅𝑡 (R2 = 0.61) (12)
By using equation (12), 𝑆𝑡 = 𝜋𝑅𝑡2 and 𝑉𝑡 = 𝐻𝑡,𝑡 ⋅ 𝑆𝑡 we obtain:

2
𝑉𝑡 3 (13)
𝑆𝑡 = 𝜋 ( )
2.49 𝜋
1
𝑉𝑡 3 (14)
𝐻𝑡,𝑡 = 2.49 ( )
2.49 𝜋
Following discussions with PVWPS installers, we also set 𝐻𝑡,𝑖 , 𝐻𝑡,𝑟 and 𝐻𝑡,𝑠 from 𝐻𝑡,𝑡 as following:
𝐻𝑡,𝑖 = 𝐻𝑡,𝑡 − 0.1 m (15)
𝐻𝑡,𝑠 = 𝐻𝑡,𝑡 − 0.2 m (16)
𝐻𝑡,𝑟 = 𝐻𝑡,𝑡 − 0.5 m (17)
According to the definition of 𝐻𝑡,𝑟 (equation (17)), 𝐻𝑡,𝑡 has to be strictly higher than 0.5 m. When
reinjecting this minimum value for 𝐻𝑡,𝑡 into equation (14), we obtain a lower boundary for the tank volume
𝑉𝑡 of 0.1 m3.
The only remaining parameters of the energy conversion model are the groundwater parameters (𝐻𝑏,𝑠 , 𝜅0
and 𝜇0 ). These parameters depend on the position of the PVWPS in the village (𝐿𝑎𝑡, 𝐿𝑜𝑛). However, as
presented in section II.2.3, it is not possible to determine the groundwater parameters from the position,
without drilling a borehole and performing pumping tests (which is expensive: ~$104 for drilling a borehole
as shown in section III.5.1.2 and ~$2.5 103 for performing pumping tests as shown in section II.2). In section
IV.5, we will present the range of variation of the groundwater parameters and the influence of the
73
uncertainty on groundwater parameters on optimisation results. Until then, for any position in the village,
the groundwater parameters are supposed equal to their values for the current PVWPS of Gogma (see
Table III-7).
III.3.2.2 Model inputs
In order to provide results for the energy conversion model for different positions of the PVWPS (𝐿𝑎𝑡, 𝐿𝑜𝑛)
in the village, we use the water demand 𝑄𝑑 simulated by the demand model (see section III.2.2). We assume
that the daily evolution of the water demand is the same for every day of the season. For the dry season, it
is given by the blue curve in Figure III-3. For the wet season, it is given by the blue curve in Figure III-4.
The simulated water demand 𝑄𝑑 , is obtained with a temporal resolution of 1 hour as it comes from survey
responses (see section III.2.2). We therefore also use a temporal resolution of 1 hour for input climatic data,
instead of 1 minute. In [104] and [105], we have shown that using a temporal resolution of 1 hour instead
of 1 minute for water demand data and climatic data has a low impact on the energy conversion model
output. Besides, using a temporal resolution of 1 hour allows to reduce computing time.
Finally, we use satellite climatic data as input from this point onward. Indeed, we showed that the loss of
accuracy associated to the use of satellite data instead of measured ones does not exceed 3% [78].
Additionally, the use of satellite data permits the implementation of the energy conversion model in
geographic areas where no local climatic measurements are available. In Appendix C, we present how the
ambient temperature 𝑇𝑎 and the irradiance on the plane of the PV array 𝐺𝑝𝑣 are obtained from satellite
databases. For the case of Gogma, the satellite climatic data are the same for any position of the PVWPS in
the village. Indeed, the whole village, which is of area ~2×2 km, is contained within one pixel of the satellite
database, which is of area ~5×5 km [106]. We note that the average daily irradiance from satellite data are
similar for the dry season (5.76 kWh/m2) and the wet season (5.54 kWh/m2). In addition, the average
temperature from satellite data is 28.7 °C for the dry season and 26.2 °C for the wet season.
The satellite irradiance and temperature data for the period from 1 to 2 December 2017 (first two days of
the dry season) are shown in Figure III-14 and Figure III-15. The simulated water demand for the same
period and for the position of the current PVWPS is shown in Figure III-16.
74
Figure III-14 – Irradiance on the plane of the PV array Figure III-15 – Ambient temperature from satellite
from satellite data. data.
Figure III-16 – Simulated water demand for the

position of the current PVWPS
III.3.2.3 Model results for undersized and oversized PVWPS
We now simulate the energy conversion model for two sets of values of the sizing variables. This will allow
us to illustrate the notion of oversized and undersized PVWPS.
Firstly, we simulate the energy conversion model for the period from 1 to 2 December 2017 for the following
sizing of the PVWPS: 𝑃𝑝𝑣,𝑝 = 550 Wp, 𝑀𝑃 = SQFlex 5A-7, 𝑉𝑡 = 17 m3. The evolution of the water level in
the tank 𝐻𝑡 , the energy conversion model output, is presented in Figure III-18a. We observe that the water
level in the tank 𝐻𝑡 never reaches 0 m, i.e. the tank is never empty. Additionally, when we simulate 𝐻𝑡
during the whole dry and wet seasons, we observe that it always remains higher than 0 m (not shown here).
Consequently, the entire water demand 𝑄𝑑 is fulfilled. In this case, the PVWPS is said to be ‘oversized’.
Secondly, we simulate the energy conversion model for the following sizing of the PVWPS: 𝑃𝑝𝑣,𝑝 = 100 Wp,
𝑀𝑃 = SQFlex 5A-7, 𝑉𝑡 = 8 m3. The evolution of the water level in the tank 𝐻𝑡 for this sizing is presented in
Figure III-19a. We observe that, for several periods of time when there is water demand at the system
(𝑄𝑑 > 0 in Figure III-16), the water level in the tank 𝐻𝑡 is equal to 0 m, i.e. the tank is empty. This is notably
the case in the evening on 1 December and in the morning and in the evening on 2 December. This means
that the water demand during these periods of time is not fulfilled. In this case, the PVWPS is said to be
‘undersized’.
75
III.3.3 Beneficiaries identification model

Figure III-17 shows the block diagram of the beneficiaries identification model.
Beneficiaries identification model
Determination of the sources

Determination of the
where the households effectively
water consumption
go after the installation
flow rate
of the PVWPS
Figure III-17 – Block diagram of the beneficiaries identification model.

𝑄𝑑 : water demand at the PVWPS, 𝐻𝑡 : water level in the tank, 𝒉𝒔∗𝒂𝒇𝒕 : water sources where the households wish to go
after installation of the PVWPS, 𝑄𝑐 : water consumption at the PVWPS, 𝒉𝒔𝒂𝒇𝒕 : water sources where the households
effectively go after installation of the PVWPS.
III.3.3.1 Determination of the water consumption flow rate

We propose the following formula to determine the water consumption flow rate 𝑄𝑐 from the water demand
flow rate 𝑄𝑑 and the water level in the tank 𝐻𝑡 :
𝑄𝑑 (𝑡) if 𝐻𝑡 (𝑡) > 0
𝑄𝑐 (𝑡) = { (18)
0 if 𝐻𝑡 (𝑡) = 0
In Figure III-18b and Figure III-19b we plot the consumption profiles obtained for the oversized and
undersized systems considered in section III.3.2.3. We also superimpose the demand profile on these figures.
(a) (a)
(b) (b)
Figure III-18 – Oversized PVWPS. (a) Water level in Figure III-19 – Undersized PVWPS. (a) Water level in
the tank (b) Demand and consumption flow rates. the tank (b) Demand and consumption flow rates.
76
III.3.3.2 Determination of the sources where the households effectively go after installation of the PVWPS
For each season (𝜎 ∈ {dry, wet}), we determine the volume demanded 𝑉𝑑𝜎 and the volume consumed 𝑉𝑐𝜎
from the water demand profile and the water consumption profile as following:
𝑉𝑑𝜎 = ∫ 𝑄𝑑 (𝑡)𝑑𝑡 (19)

𝜎
𝑉𝑐𝜎 = ∫ 𝑄𝑐 (𝑡)𝑑𝑡 (20)

𝜎
We can therefore define the rate of satisfaction of the water demand 𝑟 𝜎 as:
𝑉𝑐𝜎
𝑟𝜎 = (21)
𝑉𝑑𝜎
Then, the number of households that effectively go to the PVWPS (i.e. number of consumers of the PVWPS)
𝑛𝑐𝜎 is determined from the number of households that wish to go to the PVWPS (i.e. number of demanders
of the PVWPS) 𝑛𝑑𝜎 by:
𝑛𝑐𝜎 = round(𝑟 𝜎 ⋅ 𝑛𝑑𝜎 ) (22)
In order to select the households that actually benefit from the PVWPS, we take the 𝑛𝑐𝜎 households that have
the highest probability to go to the PVWPS 𝜌(ℎ, 𝑃𝑉𝑊𝑃𝑆) amongst the 𝑛𝑑𝜎 households that wish to go to the
PVWPS. In addition, we consider that the 𝑛𝑑𝜎 - 𝑛𝑐𝜎 households that are not selected remain at the water source
where they used to go before the installation of the PVWPS. This allows to deduce the vector 𝒉𝒔𝒂𝒇𝒕 from
the vector 𝒉𝒔∗𝒂𝒇𝒕 .
𝑑𝑟𝑦 𝑑𝑟𝑦
For instance, for the dry season and the current position of the PVWPS, 𝑉𝑑 = 2778 m3 and 𝑛𝑑 = 36. For
𝑑𝑟𝑦 𝑑𝑟𝑦 𝑑𝑟𝑦 𝑑𝑟𝑦 𝑑𝑟𝑦
the oversized PVWPS, 𝑉𝑐 = 𝑉𝑑 and all the 𝑛𝑑 households can attend the PVWPS (𝑛𝑐 = 𝑛𝑑 ). For
𝑑𝑟𝑦 𝑑𝑟𝑦
the undersized PVWPS, 𝑉𝑐 = 427 m3 and only 𝑛𝑐 = 6 households can attend the PVWPS.
This is a first proposal to determine the number of consumers from the number of demanders in the case of
an undersized system. This proposal could of course be refined in the future. It may notably be interesting
to study households’ behaviour in existing cases where PVWPS are undersized.
77
III.4 Impact model

Thanks to the demand and technical models, we can predict the water sources where the households
effectively go after the installation of the PVWPS 𝒉𝒔𝒂𝒇𝒕 . The impact model computes the socio-economic
impact 𝑆𝐸𝐼 associated to the change in water sources of the households between before (𝒉𝒔𝒃𝒆𝒇) and after
(𝒉𝒔𝒂𝒇𝒕 ) the installation of the PVWPS (see Figure III-20). The socio-economic impact is quantified thanks
to several indicators.
Socio-economic
Impact
impact
model
Figure III-20 – Block diagram of the impact model.
III.4.1 Indicators identification and ranking

This section aims at identifying and ranking indicators that are related to the water source attended by a
given household. To do so, we use the theory of change, which is a common tool in the impact evaluation
literature [107, 108]. The proposed theory of change (see Figure III-21) presents the consequences of going
to a given water source and each consequence is associated to an indicator. The proposed theory of change
allows to have a better understanding of the socio-economic repercussions of investments in water
infrastructures in rural sub-Saharan villages.
We distinguished two types of indicators: direct and indirect ones. Direct indicators correspond to attributes
of the household-source couple (ℎ, 𝑠). When an household changes of source, this is immediately associated
with a change in these attributes. A change in these attributes may then trigger a change in other indicators,
which are called 1st order indirect indicators. 1st order indirect indicator may themselves influence other
indirect indicators, called 2nd order indirect indicators, so on and so forth.
Causal links between the indicators are represented by arrows in Figure III-21. In Table III-8, we present
justifications for some of these causal links based on the literature. For the other causal links in Figure III-21,
no references were encountered but we still represented these relations because they seem logical to us.
Investigating these potential relations may represent future directions of research. For this purpose, similar
methodologies as the one that we used in [109] to investigate the link between electricity access and
willingness to pay for improved water access may be considered. Finally, the causal link between two
indicators may be modelled by a coefficient 𝛼 [110, 111]. This coefficient represents the magnitude to which
a change in the value of the first indicator will impact the second indicator. For some of the causal links, we
found results of studies that can allow to deduce a value for the coefficient 𝛼. This is specified in Table III-8
by the “(𝛼)” sign.
Finally, it is important to note that, the more indirect the indicator, the higher the influence of external factors
on this indicator [107]. As a result, changes in indirect indicators may be more difficult to attribute to the
change in water source than changes in direct indicators.
78
1st order 2nd order 3rd order

Direct
indirect indirect indirect
Water Health
WASH diseases
quality expenses
) )
Back pain
Time
Income
- sick
Extraction
- collecting water
easiness
)
- walking to the
Choice of the water source
source by the
School
household
attendance
Water
cost
)
Distance
household - Strength of evidence in literature
water source No evidence
) Water quantity Evidence available
consumed Value of the coefficient available
Figure III-21 – Theory of change – effect of going to a water source. 𝛼: causal link coefficient.
WASH diseases: diseases related to water, sanitation and hygiene.
79
Table III-8 – Causal links between indicators of the theory of change.

Causal Input Output indicator Justification from the literature
link indicator
n°
1 Water WASH diseases  (𝛼1 ) The percentage of reduction in diarrhea from water quality
quality improvements is comprised between 12% and 47% [110].
 (𝛼1 ) Water quality interventions trigger a 42% relative reduction in child
diarrhea morbidity on average [112].
 (𝛼1 ) A better access to safe water sources and better hygiene practices can
decrease trachoma morbidity by 27% [113].
2 Extraction Back pain
easiness
3 Extraction Time collecting
easiness water
4 Distance Time walking to  In Mozambique, Cairncross and Cuff [114] showed that women living in
household - the water source a village with a centrally located water source spent 106 minutes less to
water source collect water per day than women living in a village who relied on a
distant source.
 In Benin, Gross et al. [115] found that reduced walking distances from
improved water access contributed to allow women to save 35 minutes
per day.
5 Distance Water quantity  The higher the distance to collect water, the lower the water quantity
household - consumed consumed. According to [116], if the distance household-water source is:
water source o <250 m, the daily water consumption typically varies between
15 and 50 L/capita/day
o 250-1000 m, the daily water consumption typically varies
between 10 and 30 L/capita/day
o >1000 m, the daily water consumption typically varies between
5 and 15 L/capita/day
6 WASH Health expenses
diseases
7 WASH Time sick  Improving water, sanitation and hygiene services in schools can help to
diseases reduce the number of school days missed because of diarrhea [117].
8 Back pain Health expenses
9 Back pain Time sick
10 Time Income  (𝛼10 ) In rural Kenya, Whittington et al. [111] computed that the value of
savings the time saved from buying water to a vendor instead of collecting it was
$0.38/hour.
 (𝛼10 ) Churchill et al. [118] considered $0.125/hour as the value of the
time of poor people.
 The percentage loss in gross domestic product due to diseases and
productivity losses linked to water and sanitation is of 5% in developing
countries [119].
 The Dutch ministry of foreign affairs raises awareness on the fact that the
time saved from improved water access is often spent undertaking other
unpaid work such as collecting firewood or unpaid agricultural labour. As
a consequence, time savings may not allow a significant increase in
income [120].
11 Time School  (𝛼11 ) In Ghana, reducing water collection time of 15 minutes allowed to
savings attendance increase girls school attendance by 8 to 12% [121].
 (𝛼11 ) In Yemen, improved water supplies permitted to increase the
percentage of girls enrolled in schools of 4 to 8% [120].
 In Tanzania, one third of water users groups reported that improved
access to safe water supplies allowed to increase the time girls spend
studying and school attendance [120].
 During the time when parents or siblings are sick, children skip class to
take care of them. In addition, during the time when the teacher is sick,
classes get cancelled [117].
80
III.4.2 Quantification of the indicators for each household

III.4.2.1 Direct indicators
When household ℎ attends a source 𝑠, we can compute the associated value of each direct indicator. The
direct indicators, their methods of quantification and their range of variation in the case of Gogma are
presented in Table III-9. Note that the lower the indicator, the better the living conditions of the household
ℎ.
Table III-9 – Direct indicators.

Indicator Quantification methods Range of variation in
Gogma
ℐ𝑖,𝑚𝑖𝑛 to ℐ𝑖,𝑚𝑎𝑥
Water Based on the water quality tests performed, we determined that water from 0 (potable water) to
quality open wells is not potable and that water from hand pumps and the PVWPS is 1 (non-potable water)
ℐ𝑤𝑞 potable (see section II.2.7). We define the water quality variable as equal to 0
is the water is potable and 1 if the water is not potable. Note that, for a given
source 𝑠, the water quality is the same for all households ℎ.
Extraction For the PVWPS, we assume that water extraction at this source is ‘not arduous’ 0 (not arduous) to
easiness for all households, following on-field observations (see section II.2.1). 2 (very arduous)
ℐ𝑒𝑒 For the other sources, we know how households of the survey pool 𝒉𝒔𝒖𝒓
perceive water extraction at these sources (see Table III-1). In addition, we
assumed that each household, that is not in 𝒉𝒔𝒖𝒓 , goes to the same source as the
household of 𝒉𝒔𝒖𝒓 that is the closest geographically (see section III.2.1.2).
Thus, we also assume that it has the same perception of the extraction easiness
at the source.
Water cost The cost for collecting water at each source was obtained through account $0/month to
ℐ𝑤𝑐 books and on-field observations. For a given source 𝑠, the water cost is the $0.89/month
same for all households ℎ.
Distance We calculate the straight line distance between the household ℎ and the source 0 m to 983 m 7
household- 𝑠 from GPS coordinates.

source
ℐ𝑑
The value of the direct indicator 𝑖 when the household ℎ chooses the source 𝑠 is noted ℐ𝑖 (ℎ, 𝑠). Besides,
before the installation of the PVWPS, the household ℎ attends the source 𝑠𝑏𝑒𝑓 and, after the installation of
the PVWPS, the household ℎ attends the source 𝑠𝑎𝑓𝑡 (it is of course possible to have 𝑠𝑎𝑓𝑡 = 𝑠𝑏𝑒𝑓 ). We define
the variation of the direct indicator 𝑖 between before and after installation of the PVWPS, Δℐ𝑖 (ℎ), by:
Δℐ𝑖 (ℎ) = − (ℐ𝑖 (ℎ, 𝑠𝑎𝑓𝑡 ) − ℐ𝑖 (ℎ, 𝑠𝑏𝑒𝑓 )) (23)
We added the ‘minus’ sign at the beginning of equation (23) so that an improvement of the living conditions
of the household ℎ, i.e. ℐ𝑖 (ℎ, 𝑠𝑎𝑓𝑡 ) < ℐ𝑖 (ℎ, 𝑠𝑏𝑒𝑓 ), corresponds to a positive value of the indicator variation,
i.e. Δℐ𝑖 (ℎ) > 0.
7
The distance household-source depends on the prediction of the demand model. In order to determine the maximum
that this distance can reach (ℐ𝑑,𝑚𝑎𝑥 ), we run 1000 times the demand model for random positions of the PVWPS in the
village. We keep the maximum distance that a household is willing to travel to collect water, according to the model.
81
III.4.2.2 Indirect indicators
The variation of the 1st order indirect indicators is deduced from the variation of the direct indicators thanks
to the coefficients 𝛼. The variation of the 2nd order indirect indicators is obtained from the variation of the
1st order indirect indicators in the same way, and so on for the 3rd order indirect indicators.
The first method to determine the value of a coefficient 𝛼 is to extrapolate it from the literature. For instance,
for the coefficient 𝛼1 , according to [110], the percentage of reduction in diarrhea from water quality
improvements is comprised between 12% and 47%. Therefore we consider a value of 𝛼1 equal to the mean
between 12% and 47%, i.e. 35%. We also assume that a water quality improvement for the household ℎ
corresponds to switching from a non-potable source to a potable one (Δℐ𝑤𝑞 (ℎ) = 1) and that the percentage
of reduction in diarrhea corresponds to the reduction of the percentage of household members affected by
diarrhea, noted Δℐ𝑑𝑖𝑎 (ℎ). Δℐ𝑑𝑖𝑎 (ℎ) is thus given by:
Δℐ𝑑𝑖𝑎 (ℎ) = 𝛼1 ⋅ Δℐ𝑤𝑞 (ℎ) (24)
The first problem associated with this literature-based method is that studies are not available for all of the
coefficients 𝛼. For instance, for the theory of change of Figure III-21, studies were found for only 3
coefficients 𝛼 (𝛼1 , 𝛼10 , 𝛼11 ). In addition, we observe that only 𝛼1 can be used in our case as using 𝛼10 and
𝛼11 would require the knowledge of 𝛼3 , 𝛼4 , 𝛼7 and 𝛼9 . The second problem is that the value used for the
coefficient is not specific to the considered case study and was determined by using data from another case
study. Indeed, the value of the coefficient may vary from one case study to another due to the fact that the
socio-economic conditions are not the same.
The second method to determine the value of a coefficient 𝛼 is to perform a regression between the two
indicators that are related by the coefficient 𝛼, using the data on the indicators collected in the village. For
instance, for determining 𝛼1 , one could perform a regression between the quality of the water source used
and the percentage of household members affected by diarrhea. The main advantage of this method is that
the value of 𝛼1 obtained is more accurate because it is specific to the considered case study. However, there
are two main challenges for the application of this method. Firstly, it requires a complex analysis notably
for identifying the control variables to include in the regression model and for demonstrating causality and
not only correlation [122]. Secondly, for several coefficients, data that can be collected in one village only
(~100 households) may not represent a large enough sample to accurately determine the coefficient [123].
In future works, it may be interesting to study which 𝛼 coefficients may be determined with data from one
village only, by using the household surveys performed in Gogma.
Following these considerations, in this thesis, we consider only the 4 direct indicators and the 1st order
indirect indicator “WASH diseases”. The “WASH diseases” indicator is itself restricted to “diarrhea”, which
is quantified from the “water quality” indicator using the coefficient 𝛼1 from the literature (see
equation (24)), taken equal to 35%.
III.4.3 Socio-economic impact at the scale of the village

The normalised variation of the indicator 𝑖, between before and after the installation of the PVWPS, at the
scale of the village Δℐ𝑖 (𝒉) is given by:
82
∑𝑛𝑘=1
𝑣
Δℐ𝑖 (ℎ𝑘 )
Δℐ𝑖 (𝒉) = (25)
𝑛𝑣 ⋅ (ℐ𝑖,𝑚𝑎𝑥 − ℐ𝑖,𝑚𝑖𝑛 )
where 𝒉 is a vector containing all the households of the village, 𝑛𝑣 is the number of households in the
village, ℐ𝑖,𝑚𝑖𝑛 and ℐ𝑖,𝑚𝑎𝑥 are the minimum and maximum possible values of the indicator 𝑖 at the scale of
the household. The minimum and maximum values of the direct indicators are given in Table III-9. The
minimum of the diarrhea indicator is 0 and the maximum is 𝛼1 (0.35). Thanks to the normalisation by the
denominator, for each indicator 𝑖, Δℐ𝑖 (𝒉) is a quantity without unit comprised between 0 and 1.
As presented in section III.1, the vectors 𝒉𝒔𝒃𝒆𝒇 and 𝒉𝒔𝒂𝒇𝒕 , which are the inputs of the impact model, are not
the same between the dry season and the wet season. Therefore, the variation of the indicators for each
household Δℐ𝑖 (ℎ) (equations (23) and (24)) and the variation of the indicators for the whole village Δℐ𝑖 (𝒉)
are computed for each season (𝜎 ∈ {dry, wet}). We define the socio-economic impact for the season 𝜎,
𝑆𝐸𝐼 𝜎 , as the weighted sum of the normalized variations of the indicators at the scale of the village:
𝑆𝐸𝐼 𝜎 = ∑ 𝑤𝑖 ⋅ Δℐ𝑖𝜎 (𝒉) (26)

𝑖
The weights 𝑤𝑖 are set by the decision maker depending on the indicators that he wants to favour. The socio-
economic impact for the whole year 𝑆𝐸𝐼 is therefore equal to:
7 5
𝑆𝐸𝐼 = 𝑆𝐸𝐼 𝑑𝑟𝑦 + 𝑆𝐸𝐼 𝑤𝑒𝑡 (27)
12 12
7 5
The coefficients 12 and 12 come from the fact that, in Gogma, the dry season lasts for 7 months and the wet
season lasts for 5 months. Overall, we compare the values of the indicators for the dry season that follows
the installation of the PVWPS to the ones for the dry season that precedes the installation of the PVWPS
(same for the wet season). We do not consider the evolution of the indicators over the whole lifetime of the
PVWPS (for instance the water cost ℐ𝑤𝑐 at the PVWPS may change over the years). This will be the object
of future work.
83
III.5 Economic model

The economic model determines the life-cycle cost 𝐿𝐶𝐶 of the PVWPS for different values of the sizing
variables 𝑃𝑝𝑣,𝑝 , 𝑀𝑃, 𝑉𝑡 (see Figure III-22). The model is presented in section III.5.1 and results are given in
section III.5.2. Our economic model builds on the related MSc thesis project undertaken by Elvire Andre de
la Fresnaye in 2018 [18].
Sizing Economic Life-cycle cost

model
Figure III-22 – Block diagram of the economic model.
III.5.1 Model
The life-cycle cost of the PVWPS is given by [50]:
𝐿𝐶𝐶 = 𝐶𝐴𝑃𝐸𝑋 + 𝑂𝑃𝐸𝑋 (28)
where 𝐶𝐴𝑃𝐸𝑋 is the capital cost and 𝑂𝑃𝐸𝑋 is the total discounted operational cost over the lifetime of the
PVWPS. 𝑂𝑃𝐸𝑋 is given by:
𝐿
𝑦𝑂𝑃𝐸𝑋(𝑗)
𝑂𝑃𝐸𝑋 = ∑ (29)
(1 + 𝑑𝑟)𝑗
𝑗=1
where 𝑦𝑂𝑃𝐸𝑋(𝑗) is the yearly operational cost for year 𝑗, 𝑑𝑟 is the discount rate and 𝐿 is the lifetime of the
PVWPS. It is important to note that the lifetimes of all the components of the PVWPS also have to be
determined in order to compute 𝑦𝑂𝑃𝐸𝑋(𝑗).
In section III.5.1.1, we determine the lifetimes of the PVWPS and of its components. The capital cost and
the operational cost are determined in sections III.5.1.2 and III.5.1.3 respectively.
III.5.1.1 Lifetime
Data on the lifetime of the PVWPS and of its different components were collected through a literature review
and through surveys with members of companies specialised in photovoltaic energy and PVWPS based in
Burkina Faso (survey in Appendix D). The collected data are presented in Table III-10. In this table, we also
specify the values selected in this thesis.
84
Table III-10 – Lifetime of the PVWPS and of its components.

Selected
Lifetime from the literature and surveys (years) lifetime
(years)
 10 [124]
PVWPS  20 [20, 24, 125, 126, 127, 128, 129] 20
 25 [34]
Borehole  50 (Survey, A. Darga) (Survey, F. Lingani) 50
 20 [20, 110]
PV modules  25 [50] (Survey, F. Lingani) 25
 30 (Survey, G. J. Balima)
PV modules structure  50 (Survey, G. J. Balima) 50
Water level controller, float switch  5 (Survey, A. Darga) 5
Motor-pump  10 [34, 49, 128, 20] (Survey, F. Lingani) (Survey, G. J. Balima) 10
 20 [20]
Tank 30
 50 (Survey, F. Lingani)
Fountain  30 (Survey, A. Darga) 30
Fountain taps  2 (Survey, A. Darga) 2
Pipes  50 (Survey, G. J. Balima) 50
Wire fence  20 (Survey, A. Darga) 20
 10 [20]
Cables  25 (Survey, F. Lingani) 25
 30 (Survey, G. J. Balima)
Signpost  20 (Survey, A. Darga) 20
III.5.1.2 Capital cost
The capital cost is equal to:
𝐶𝐴𝑃𝐸𝑋 = 𝐶𝐴𝑃𝐸𝑋𝑝𝑣 + 𝐶𝐴𝑃𝐸𝑋𝑀𝑃 + 𝐶𝐴𝑃𝐸𝑋𝑡 + 𝐶𝐴𝑃𝐸𝑋𝑓 (30)
where 𝐶𝐴𝑃𝐸𝑋𝑝𝑣 is the PV modules capital cost, 𝐶𝐴𝑃𝐸𝑋𝑀𝑃 is the motor-pump capital cost, 𝐶𝐴𝑃𝐸𝑋𝑡 is the
tank capital cost and 𝐶𝐴𝑃𝐸𝑋𝑓 is the fixed capital cost. 𝐶𝐴𝑃𝐸𝑋𝑝𝑣 , 𝐶𝐴𝑃𝐸𝑋𝑀𝑃 and 𝐶𝐴𝑃𝐸𝑋𝑡 depend on the
sizing variables of the PVWPS (𝑃𝑝𝑣,𝑝 , 𝑀𝑃, 𝑉𝑡 ), contrarily to 𝐶𝐴𝑃𝐸𝑋𝑓 .
The capital cost 𝐶𝐴𝑃𝐸𝑋𝑀𝑃 of each of the 8 references of motor-pumps 𝑀𝑃 that we consider is available at
[130]. Several local companies in Burkina Faso order their motor-pumps from Off-grid Europe. It is
interesting to observe that, despite the fact that the 8 motor-pumps do not have the same performances, they
all cost the same price: $2.2 103.
Then, we determine the capital cost of the PV modules 𝐶𝐴𝑃𝐸𝑋𝑝𝑣 as a function of the PV modules peak
power 𝑃𝑝𝑣,𝑝 and the capital cost of the steel tank 𝐶𝐴𝑃𝐸𝑋𝑡 as a function of the tank volume 𝑉𝑡 . For this
purpose, we collected data on the capital cost of PV modules and of steel tanks for different PV modules
peak powers and different tank volumes with the following methods:
 We included questions on the capital cost of PV modules and tanks in the survey to local companies
(see Appendix D).
 Some local companies gave us quotations on PVWPS that they installed in the past.
 We asked several local companies to make us a quotation for installing a PVWPS in the village of
Bidiga, which is situated at ~10 km from Gogma. For each company, we changed the system
85
specifications in terms of total dynamic head and daily water consumption in order to obtain
quotations for PVWPS of different sizes.
The results of the data collection are presented in Figure III-23 for the PV modules and in Figure III-24 for
the tank.
Figure III-23 – Capital cost of the PV modules as a Figure III-24 – Capital cost of the tank as a function of
function of the PV modules peak power. the tank volume.
We fit the collected data presented in Figure III-23 and we obtain the expression of 𝐶𝐴𝑃𝐸𝑋𝑝𝑣 as a function
of 𝑃𝑝𝑣,𝑝 :
𝐶𝐴𝑃𝐸𝑋𝑝𝑣 = 0.79 𝑃𝑝𝑣,𝑝 (R2 = 0.64) (31)
We do the same for the data of Figure III-24 for the tank:
𝐶𝐴𝑃𝐸𝑋𝑡 = 6.2 102 𝑉𝑡 + 5.2 103 (R2 = 0.78) (32)
Finally, regarding the fixed capital cost 𝐶𝐴𝑃𝐸𝑋𝑓 , we use the data for the current PVWPS of Gogma which
are given in Table III-11. These information were provided by the NGOs Respublica and Eau Fil du Soleil
which financed the PVWPS of Gogma.
Table III-11 – Fixed capital cost breakdown.

Component Cost ($)
Borehole 8.3 103
PV modules structure 1.0 102
Water level controller, float switch 4.2 102
Fountain 1.3 103
Fountain taps 5.0 101
Pipes 1.9 103
Wire fence 1.1 103
Other material (cables, glue, locks, signpost) 7.2 102
Installation workforce 1.4 103
Total fixed capital cost 𝑪𝑨𝑷𝑬𝑿𝒇 1.5 104
In reality, there is a part of the borehole cost that depends on the depth of the borehole and therefore on the
groundwater resources where the borehole is drilled. Consequently, the borehole cost varies with the
86
position (𝐿𝑎𝑡, 𝐿𝑜𝑛) of the PVWPS in the village. However, we cannot take this variation into account
because, as explained in section II.2.3, it is not possible to determine the groundwater resources for all
positions in a village.
III.5.1.3 Operational cost
The yearly operational cost is composed of a yearly replacement cost 𝑦𝑂𝑃𝐸𝑋𝑟𝑒 and of a yearly maintenance
cost 𝑦𝑂𝑃𝐸𝑋𝑚𝑎 . We therefore have:
𝑦𝑂𝑃𝐸𝑋(𝑗) = 𝑦𝑂𝑃𝐸𝑋𝑟𝑒 (𝑗) + 𝑦𝑂𝑃𝐸𝑋𝑚𝑎 (𝑗) (33)
Firstly we estimate the yearly replacement cost 𝑦𝑂𝑃𝐸𝑋𝑟𝑒 . According to the lifetime of the installed
components presented in Table III-10, the only components that are replaced during the 20 years lifetime of
the PVWPS are:
 the motor-pump 𝑀𝑃 which is replaced every 10 years

 the fountain taps which are replaced every 2 years.
 the water level controller and the associated float switch which are replaced every 5 years
As a consequence, the yearly replacement cost 𝑦𝑂𝑃𝐸𝑋𝑟𝑒 for year 𝑗 is:
𝐶𝐴𝑃𝐸𝑋𝑓𝑜𝑢𝑛𝑡𝑎𝑖𝑛 𝑡𝑎𝑝𝑠 if 𝑗 ∈ [2, 4, 6, 8, 12, 14, 16, 18]

𝐶𝐴𝑃𝐸𝑋𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑙𝑒𝑟 & 𝑠𝑤𝑖𝑡𝑐ℎ if 𝑗 ∈ [5, 15]
𝑦𝑂𝑃𝐸𝑋𝑟𝑒 (𝑗) = (34)
𝐶𝐴𝑃𝐸𝑋𝑀𝑃 + 𝐶𝐴𝑃𝐸𝑋𝑓𝑜𝑢𝑛𝑡𝑎𝑖𝑛 𝑡𝑎𝑝𝑠 + 𝐶𝐴𝑃𝐸𝑋𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑙𝑒𝑟 & 𝑠𝑤𝑖𝑡𝑐ℎ if 𝑗 ∈ 10
{ 0 otherwise
where 𝐶𝐴𝑃𝐸𝑋𝑓𝑜𝑢𝑛𝑡𝑎𝑖𝑛 𝑡𝑎𝑝𝑠 is the capital cost of the fittings and taps for the fountain and
𝐶𝐴𝑃𝐸𝑋𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑙𝑒𝑟 & 𝑠𝑤𝑖𝑡𝑐ℎ is the capital cost of the water level controller and the float switch.
Secondly we estimate the yearly maintenance cost 𝑦𝑂𝑃𝐸𝑋𝑚𝑎 . Data on the maintenance cost of PVWPS are
scarce because PVWPS have been deployed in a large scale only in recent years [131, 132]. [50] considers
a value of 1% of the capital cost for the yearly maintenance cost. Consequently, the variable maintenance
cost 𝑦𝑂𝑃𝐸𝑋𝑚𝑎 for year 𝑗 is:
𝐶𝐴𝑃𝐸𝑋(𝑗)
𝑦𝑂𝑃𝐸𝑋𝑚𝑎 (𝑗) = (35)
100
III.5.2 Results
We compute the capital, operational, and life-cycle costs for the sizing of the current PVWPS of Gogma
(𝑃𝑝𝑣,𝑝 = 620 Wp, 𝑀𝑃 = SQFlex 5A-7, 𝑉𝑡 = 11.4 m3). The discount rate 𝑑𝑟 considered for Burkina Faso is
5.6% [18]. The results for the current PVWPS of Gogma are given in Table III-12.
87
Table III-12 – Capital, operational and life-cycle costs - current PVWPS of Gogma.
Cost Value ($)
𝑪𝑨𝑷𝑬𝑿 3.0 104
𝐶𝐴𝑃𝐸𝑋𝑝𝑣 4.9 102
𝐶𝐴𝑃𝐸𝑋𝑀𝑃 2.2 103
2
𝐶𝐴𝑃𝐸𝑋𝑡 – variable part (6.2 10 𝑉𝑡 ) 7.1 103
𝐶𝐴𝑃𝐸𝑋𝑡 – fixed part 5.2 103
𝐶𝐴𝑃𝐸𝑋𝑓 1.5 104
𝑶𝑷𝑬𝑿 6.0 103
𝑳𝑪𝑪 3.6 104
We observe that the 𝐶𝐴𝑃𝐸𝑋 is 5 times higher than the 𝑂𝑃𝐸𝑋. This prevalence of the 𝐶𝐴𝑃𝐸𝑋 over the 𝑂𝑃𝐸𝑋
is characteristic of PVWPS [31]. In addition, the sum of 𝐶𝐴𝑃𝐸𝑋𝑝𝑣 , 𝐶𝐴𝑃𝐸𝑋𝑀𝑃 and the variable part of
𝐶𝐴𝑃𝐸𝑋𝑡 is equal to $9.8 103, i.e. one third of the 𝐶𝐴𝑃𝐸𝑋. This shows that the margin of cost reduction that
can be achieved through the sizing variables (𝑃𝑝𝑣,𝑝 , 𝑀𝑃 and 𝑉𝑡 ) is moderate.
88
III.6 Partial conclusion

In this chapter, we described the proposed interdisciplinary model. This model links the sizing and the
position of the PVWPS to its life-cycle cost and socio-economic impact.
The interdisciplinary model is composed of 4 sub-models:
 The demand model predicts the water demand at the PVWPS (i.e. the households that wish to go to
the PVWPS) for any position of the PVWPS in the village.
 The technical model allows to identify the households that can effectively go to the PVWPS
amongst the households that wish to go to the PVWPS.
 The impact model evaluates the socio-economic impact associated with the changes in water
sources of the households between before and after the installation of the PVWPS.
 The economic model allows to determine the life-cycle cost of the PVWPS depending on its sizing.
Besides, we presented how the experimental data that we collected in Chapter II are integrated into each of
these 4 sub-models and we compared the outputs of the demand and technical models to experimental
measurements.
89
Chapter IV Optimal design
90
Chapter IV Optimal design
In Chapter III, we proposed an interdisciplinary model which links the sizing and the position of the PVWPS
to its life-cycle cost and socio-economic impact. In section IV.1, we define an optimisation problem which
aims at determining the sizings and the positions of the PVWPS which minimize its life-cycle cost and
maximize its socio-economic impact. In section IV.2, we analyse an optimisation result for the case of
Gogma, obtained for a socio-economic impact 𝑆𝐸𝐼 which considers all the direct indicators with weights all
equal to one, and supposing that the groundwater resources are the same for all positions in Gogma. In
section IV.3, we investigate the influence of the error in the demand model output on the optimisation
results. In section IV.4, we study the influence of the indicators selected in the 𝑆𝐸𝐼 function on the
optimisation results. In section IV.5, we investigate the effect of the groundwater resources on the
optimisation results. Finally, in section IV.6, we propose an improved procedure for the design and
installation of PVWPS, that includes the optimisation methodology developed.
91
Chapter IV. Optimal design
IV.1 Optimisation problem

The two objective functions of the optimisation are the life-cycle cost 𝐿𝐶𝐶 of the PVWPS, that we want to
minimise, and its socio-economic impact 𝑆𝐸𝐼, that we want to maximise. The variables of the optimisation
are the peak power of the PV array 𝑃𝑝𝑣,𝑝 , the motor-pump reference 𝑀𝑃, the tank volume 𝑉𝑡 , the latitude
𝐿𝑎𝑡 and the longitude 𝐿𝑜𝑛 of the PVWPS. The motor-pump reference 𝑀𝑃 is the only discrete variable and
we remind that we have digitized the characteristic curves of 8 references of motor-pumps (see section
III.3.2.1).
We add the constraint that, for each motor-pump reference 𝑀𝑃, the total dynamic head 𝑇𝐷𝐻 must remain
smaller than the maximum pumping height 𝐻𝑚𝑎𝑥 (𝑀𝑃) specified in the datasheet of the motor-pump:
𝑇𝐷𝐻(𝑡) < 𝐻𝑚𝑎𝑥 (𝑀𝑃), ∀𝑡 (36)
We also remind that we allow for undersized systems and the whole water demand 𝑄𝑑 not to be fulfilled
(see section III.3.3). In order not to consider undersized systems, the constraint that the water level in the
tank remains higher than 0 m (𝐻𝑡 (𝑡) > 0, ∀𝑡) would need to be added.
Once the position of the PVWPS is set and the borehole is drilled, the maximum flow rate that can be
pumped 𝑄𝑝,𝑚𝑎𝑥 can be determined from pumping tests. At this moment, we propose to set the constraint
that the pump flow rate 𝑄𝑝 must remain lower than 𝑄𝑝,𝑚𝑎𝑥 . For sections IV.2 to IV.5, the position is not set
so we do not implement this constraint. In section IV.6 though, we propose a real procedure for the design
and installation of PVWPS and we therefore show the implementation of this constraint.
We use a bi-objective differential evolution algorithm [133], which is a stochastic algorithm, to solve the
optimisation problem. We chose a stochastic algorithm because the optimisation problem is non-linear. We
perform one optimisation for each motor-pump reference 𝑀𝑃, and we therefore obtain one Pareto front for
each motor-pump reference. We then draw the final Pareto front by going through the best points of the
Pareto fronts of the motor-pump references.
IV.2 Analysis of a reference result

In this section, we perform a first optimisation for the case study of Gogma.
We consider the four direct indicators in the 𝑆𝐸𝐼 and we take all the weighting coefficients equal to 1. The
𝑆𝐸𝐼 is therefore given by:
7 𝑑𝑟𝑦 𝑑𝑟𝑦 𝑑𝑟𝑦 𝑑𝑟𝑦
𝑆𝐸𝐼 = 12
(Δℐ𝑤𝑞 (𝒉) + Δℐ𝑒𝑒 (𝒉) + Δℐ𝑤𝑐 (𝒉) + Δℐ𝑑 (𝒉)) +
5 (37)
12
𝑤𝑒𝑡 (𝒉)
(Δℐ𝑤𝑞 + 𝑤𝑒𝑡 (𝒉)
Δℐ𝑒𝑒 𝑤𝑒𝑡 (𝒉)
+ Δℐ𝑤𝑐 + Δℐ𝑑𝑤𝑒𝑡 (𝒉))
We consider the boundaries presented in Table IV-1 for the optimisation variables. The boundaries on 𝐿𝑎𝑡
and 𝐿𝑜𝑛 correspond to the dimensions of the village of Gogma. There are no areas to avoid in Gogma. If
they were, we would have set the constraint that the position of the PVWPS (𝐿𝑎𝑡, 𝐿𝑜𝑛) cannot be in these
areas. In section IV.6, we show how to implement a constraint on the PVWPS position in the optimisation.
Finally, we remind that the justification for the lower boundary of 𝑉𝑡 is given in section III.3.2.1.
92
Table IV-1 – Boundary of the optimisation variables – Case of Gogma.

Variable Discrete/ Value for the current Lower Upper
Continuous PVWPS of Gogma boundary boundary
Sizing
Peak power PV array 𝑃𝑝𝑣,𝑝 Continuous 620 Wp 0 Wp 104 Wp
Motor-pump reference 𝑀𝑃 Discrete SQFlex 5A-7 NA NA
Tank volume 𝑉𝑡 Continuous 11.4 m3 0.1 m3 15 m3
Position
Latitude 𝐿𝑎𝑡 Continuous 11.7244 ° 11.720 ° 11.738 °
Latitude 𝐿𝑜𝑛 Continuous -0.5722 ° -0.585 ° -0.567 °
NA: not applicable
In addition, we suppose that the values of the groundwater parameters are the same for all positions of the
PVWPS in the village. They are taken equal to the values for the current PVWPS (see Table III-7).
IV.2.1 Mono-objective optimisations results

Before performing the bi-objective optimisation, we perform two mono-objective optimisations: for the first
one we aim at maximising the 𝑆𝐸𝐼 and for the second one we aim at minimising the 𝐿𝐶𝐶. This allows to
verify that the two objective functions are contradictory and to better understand the influence of each
objective function.
We also use the differential evolution algorithm of [133] and perform one mono-objective optimisation for
each motor-pump reference 𝑀𝑃. We keep the best result from all the motor-pump references as final result
of the mono-objective optimisation. For each mono-objective optimisation (i.e. for each motor-pump
reference), we consider 150 individuals and the optimisation stops when the value of the objective function
has not changed for 20 generations. The starting population is generated randomly. The total computing
time for the mono-objective optimisation, when the 8 motor-pump references are considered, is ~1 h with a
HP EliteBook 840 G3 (processor: Intel(R) Core(TM) i7-6600U CPU @ 2.60GHz 2.70 GHz; RAM: 8 Go)
[134].
IV.2.1.1 Objective function: SEI
We perform the mono-objective optimisation with the 𝑆𝐸𝐼 as objective function. The maximum value of
the 𝑆𝐸𝐼 encountered is 0.105. The optimal values of the variables are given in Table IV-2. The optimal
position of the PVWPS is represented by a green star in Figure IV-1. We then compute the 𝐿𝐶𝐶 for the
optimal values of the variables. It is equal to $3.50 104. The 𝑆𝐸𝐼 of 0.105 corresponds to the highest impact
that can be encountered.
93
Table IV-2 – Results of the mono-objective optimisation which aims at maximising the 𝑆𝐸𝐼.
𝑃𝑝𝑣,𝑝 2.7 103 Wp
𝑀𝑃 SQFlex 2.5-2
𝑉𝑡 7.9 m3
𝐿𝑎𝑡 11.7248 °
𝐿𝑜𝑛 - 0.5709 °
𝐿𝐶𝐶 $3.50 104
𝑆𝐸𝐼 0.105
Figure IV-1 – Optimal positions of the PVWPS.
In Table IV-3, we provide more information about the quantities that are related to the value of the 𝑆𝐸𝐼.
94
Table IV-3 – Demanders and consumers of the PVWPS and impact indicators.
Dry Wet
Number of households that wish to go to the PVWPS (number of demanders) 𝑛𝑑 39 39
Number of households that effectively go to the PVWPS (number of consumers) 𝑛𝑐 36 38
Number of consumers that used to go to an open well before the installation of the PVWPS 𝑛𝑐,𝑜𝑤 18 33
Number of consumers that used to go to a hand pump before the installation of the PVWPS 𝑛𝑐,ℎ𝑝 18 5
Variation of the water quality indicator Δℐ𝑤𝑞 (𝒉) 0.14 0.26
Variation of the extraction easiness indicator Δℐ𝑒𝑒 (𝒉) 0.15 0.22
Variation of the water cost indicator Δℐ𝑤𝑐 (𝒉) -0.28 -0.30
Variation of the distance indicator Δℐ𝑑 (𝒉) 0.04 -0.01
Socio-economic impact 𝑆𝐸𝐼 0.06 0.17
It is interesting to observe that all the households that wish to go to the PVWPS are not able to go there
(𝑛𝑐 ≠ 𝑛𝑑 ). This is related to the fact that some of the indicators in the chosen 𝑆𝐸𝐼 function are contradictory
(as explained in the next paragraph). The small difference between the number of PVWPS consumers during
the dry season (𝑛𝑐 = 36) and the wet season (𝑛𝑐 = 38) is due to the difference in the water demand 𝑄𝑑 ,
irradiance 𝐺𝑝𝑣 and ambient temperature 𝑇𝑎 between both seasons. Results also indicate that most consumers
of the PVWPS used to go to open wells before the installation of the PVWPS. Indeed, as open wells are
sources where the water quality is the worst and where extraction is the most difficult, the households that
used to go to open wells are targeted in priority by the optimisation. This also explains the choice of the
optimal position (𝐿𝑎𝑡 = 11.7248°, 𝐿𝑜𝑛 = -0.5709°). Indeed, the household’s density is relatively high around
this position and the only neighbouring sources are open wells (see Figure IV-1).
Regarding the values of the impact indicators:

𝑑𝑟𝑦
 We observe an improvement in the water quality indicator for both seasons (Δℐ𝑤𝑞 (𝒉) > 0 and
𝑤𝑒𝑡 (𝒉)
Δℐ𝑤𝑞 > 0) which is explained by the fact that households which switch from an open well to
𝑑𝑟𝑦
the PVWPS see an improvement in water quality. 18 households (𝑛𝑐,𝑜𝑤 ) see an improvement in
𝑤𝑒𝑡
water quality during the dry season and 33 (𝑛𝑐,𝑜𝑤 ) during the wet season.
𝑑𝑟𝑦 𝑤𝑒𝑡 (𝒉)
 The extraction easiness is also improved for both seasons (Δℐ𝑒𝑒 (𝒉) > 0 and Δℐ𝑒𝑒 > 0) as
switching from an open well or a hand pump to the PVWPS is associated to an easier water
𝑑𝑟𝑦
extraction. 36 households (𝑛𝑐 ) see an improvement in extraction easiness during the dry season
and 38 (𝑛𝑐𝑤𝑒𝑡 ) during the wet season. However, it is important to keep in mind that the improvement
of the extraction easiness indicator is higher when an household switches from an open well to the
PVWPS (Δℐ𝑒𝑒 (ℎ)~2) than when an household switches from a hand pump to the PVWPS
(Δℐ𝑒𝑒 (ℎ)~1).
𝑑𝑟𝑦 𝑤𝑒𝑡 (𝒉)
 The water cost has increased for both seasons (Δℐ𝑤𝑐 (𝒉) < 0 and Δℐ𝑤𝑐 < 0) because collecting
water at the PVWPS is more expensive than at other water sources (section II.2.4). 36 households
𝑑𝑟𝑦
(𝑛𝑐 ) see an increase in water cost during the dry season and 38 (𝑛𝑐𝑤𝑒𝑡 ) during the wet season.
𝑑𝑟𝑦
 The variation of the distance indicator is slightly positive during the dry season (Δℐ𝑑 (𝒉) = 0.04)
and slightly negative during the wet season (Δℐ𝑑𝑤𝑒𝑡 (𝒉)= -0.01). On average, for the dry season,
PVWPS consumers see their distance to the water source reduced of 42 m with the installation of
the PVWPS. For the wet season, they see their distance to the water source increased of 13 m.
95
Indeed, the share of PVWPS consumers that used to go to open wells is higher for the wet season
than for the dry season and open wells are in general very close to the households (see Figure IV-1).
We see that the 𝑆𝐸𝐼 𝑑𝑟𝑦 (0.06) is lower than the 𝑆𝐸𝐼 𝑤𝑒𝑡 (0.17). As seen previously, this is due to the fact
𝑤𝑒𝑡
that more households used to go to open wells during the wet season than during the dry one (𝑛𝑐,𝑜𝑤 = 33
𝑑𝑟𝑦
and 𝑛𝑐,𝑜𝑤 = 18), triggering higher variation in the water quality and extraction easiness indicators (which
are not compensated by the decrease in the variations of the water cost and distance indicators).
IV.2.1.2 Objective function: LCC
We perform the mono-objective optimisation with the 𝐿𝐶𝐶 as objective function. We add the constraint the
the socio-economic impact 𝑆𝐸𝐼 has to be strictly higher than 0, to force the existence of a system. The
minimum value of the 𝐿𝐶𝐶 encountered is $2.72 104. This means that installing a PVWPS at least costs
$2.72 104. The associated values of the variables are given in Table IV-4. We then compute the 𝑆𝐸𝐼 for
these values of the variables. It is equal to 0.001. For these values of the variables, one household consumes
water at the PVWPS and only during the wet season. The encountered system does not have a real
application because it has a very low 𝑆𝐸𝐼 and already a significant cost 𝐿𝐶𝐶, which is due to the high share
of the 𝐿𝐶𝐶 that does not depend on the values of the sizing variables (see section III.5).
Table IV-4 – Results of the mono-objective optimisation which aims at minimising the 𝐿𝐶𝐶,
with the constraint 𝑆𝐸𝐼 > 0.
𝑃𝑝𝑣,𝑝 27 Wp
𝑀𝑃 SQFlex 2.5-2
𝑉𝑡 0.1 m3
𝐿𝑎𝑡 - 0.5735 °
𝐿𝑜𝑛 11.7263 °
𝐿𝐶𝐶 $2.72 104
𝑆𝐸𝐼 0.001
IV.2.1.3 Comparing the results of the mono-objective optimisations
We now compare the results of the mono-objective optimisations. The optimisation which aims at
maximising the 𝑆𝐸𝐼 yields to a 𝑆𝐸𝐼 of 0.105 and to a 𝐿𝐶𝐶 of $3.50 104. The optimisation which aims at
minimising the 𝐿𝐶𝐶 yields to a 𝐿𝐶𝐶 of $2.72 104 and a 𝑆𝐸𝐼 of 0.001. We observe that aiming to increase
the socio-economic impact 𝑆𝐸𝐼 is associated to an increase in life-cycle cost 𝐿𝐶𝐶. Reciprocally, aiming to
decrease the 𝐿𝐶𝐶 is associated to a decrease in 𝑆𝐸𝐼. Therefore, the minimisation of the 𝐿𝐶𝐶 and the
maximisation of the 𝑆𝐸𝐼 are two contradictory objectives, which justifies the need for bi-objective
optimisation. We also observe in Table IV-2 and Table IV-4 that the minimisation of the 𝐿𝐶𝐶 tends to lower
the values of the PV array peak power 𝑃𝑝𝑣,𝑝 and of the tank volume 𝑉𝑡 and that the maximisation of the 𝑆𝐸𝐼
tends to increase the values of these variables.
IV.2.2 Bi-objective optimisation results

As presented in section IV.1, we perform a bi-objective optimisation for each of the 8 motor-pump
references 𝑀𝑃. Before each bi-objective optimisation, we always perform the mono-objective optimisation
with the 𝑆𝐸𝐼 as objective function. Then, we integrate the optimal values of the variables from the mono-
96
objective optimisation in the starting population of the bi-objective one. If we do not do this, we notice that
the bi-objective optimisation may not find the maximum 𝑆𝐸𝐼 of the mono-objective optimisation. We also
generate 60% of the starting population of the bi-objective optimisation from the optimal values of the
variables obtained from the mono-objective optimisation: we consider smaller sizings and randomly chosen
nearby positions (within a radius of ~100 m). The remaining 40% of the starting population are generated
randomly. We do not perform the mono-objective optimisation with the 𝐿𝐶𝐶 as objective function because
the points of the Pareto front with very low 𝑆𝐸𝐼 (<0.02) are not relevant (see section IV.2.1.2).
For each bi-objective optimisation (i.e. for each motor-pump reference), we consider 150 individuals and
the optimisation stops after 150 generations. The total computing time for one bi-objective optimisation,
when the 8 motor-pump references are considered, is ~2 h with a HP EliteBook 840 G3 [134]. When we
add the time to run the mono-objective optimisation (~1 h), we obtain a total time of ~3 h. We chose these
values of the optimisation parameters (e.g. number of individuals, number of generations) in order to keep
the optimisation time below half a work day (~4 h) and facilitate the implementation of the optimisation
methodology in the field (see section IV.6).
In Figure IV-2, we show the Pareto fronts of the 8 motor-pump references and the final Pareto front that is
obtained by going through the best points of these 8 Pareto fronts.
Figure IV-2 – Points of the Pareto front for each of the motor-pump references and deduced final Pareto front.
In Figure IV-3, we plot only the final Pareto front. We also add the results of the mono-objective
optimisations (see section IV.2.1) and the values of the objective functions for the current PVWPS. For
comparison, we also plot in pink the Pareto front obtained for 1000 individuals and 1800 generations, instead
of 150 individuals and 150 generations. We observe a small difference between the fronts obtained with 150
individuals and 150 generations (in blue in Figure IV-3) and the one obtained with 1000 individuals and
1800 generations (in pink in Figure IV-3). In the rest of the thesis, we only consider results obtained with
150 individuals and 150 generations.
97
Figure IV-4, Figure IV-5 and Figure IV-6 show the values of the optimisation variables along the Pareto
front. We also add the values of the variables for the current PVWPS on these figures.
Figure IV-3 – Pareto front.
Current PVWPS
Figure IV-4 – Variation of 𝑃𝑝𝑣,𝑝 and 𝑉𝑡 as a function of the 𝑆𝐸𝐼.
Figure IV-5 – Variation of 𝑀𝑃 as a function of the 𝑆𝐸𝐼.
98
Current PVWPS
Figure IV-6 – Variation of 𝐿𝑎𝑡 and 𝐿𝑜𝑛 as a function the 𝑆𝐸𝐼.
Figure IV-3 is used to determine (1) the maximum socio-economic impact 𝑆𝐸𝐼 that can be expected from a
PVWPS of cost 𝐿𝐶𝐶 and (2) the minimum cost 𝐿𝐶𝐶 to achieve a given socio-economic impact 𝑆𝐸𝐼.
Logically, we observe that larger costs lead to more significant positive socio-economic impacts. Results
also indicate that the difference in 𝐿𝐶𝐶 between the point of minimum 𝑆𝐸𝐼 and the point of maximum 𝑆𝐸𝐼
is $7.9 103, which represents only 29% of the 𝐿𝐶𝐶 of the point of minimum 𝑆𝐸𝐼 ($2.72 104). This suggests
that significant economies of scale can be made and this goes in favour of selecting points with high values
of 𝑆𝐸𝐼.
We also observe in Figure IV-4 that the optimal values of the peak power of the PV array 𝑃𝑝𝑣,𝑝 and of the
volume of the tank 𝑉𝑡 generally increase with the socio-economic impact 𝑆𝐸𝐼. Indeed, generally speaking,
larger systems increase the number of beneficiaries and thus the socio-economic impact. Regarding the
choice of motor-pump 𝑀𝑃 (Figure IV-5), the optimisation favours references 5A-7, 2.5-2, 1.2-2 and 0.6-2.
Finally, we observe a small variation in the position of the PVWPS along the Pareto front (Figure IV-6).
We can therefore identify a zone of the village (𝐿𝑎𝑡 ∈ [11.723°, 11.725°] × 𝐿𝑜𝑛 ∈ [-0.573°, -0.571°]) where
the installation of the PVWPS would be optimal. We have represented this zone with a green rectangle in
Figure IV-1. As for the optimal position for the highest 𝑆𝐸𝐼 (see section IV.2.1), this zone is selected because
of the relatively high household density and the unavailability of hand pumps in the vicinity.
We can also compare the values of the objective functions and of the variables for the current PVWPS to
the optimal results. In Figure IV-3, we observe that the cost 𝐿𝐶𝐶 of the current PVWPS ($3.55 104) may
have allowed to reach the highest possible 𝑆𝐸𝐼 of 0.105, instead of the 𝑆𝐸𝐼 of 0.068 of the current PVWPS.
Results also indicate that the socio-economic impact of the current PVWPS (0.068) was obtained for a 𝐿𝐶𝐶
of $3.55 104, whereas, according to the Pareto front, it could have been obtained for a 𝐿𝐶𝐶 of ~$2.78 104.
This suggests that the application of the methodology may have allowed to save ~$7.7 103 in Gogma, for
the expression of the 𝑆𝐸𝐼 function considered in this section. For the 𝑆𝐸𝐼 of the current PVWPS (0.068),
Figure IV-4 indicates that the current PV array peak power 𝑃𝑝𝑣,𝑝 is close to the optimal one and that the
current tank volume 𝑉𝑡 is much larger than the optimal one. Figure IV-5 suggests that the SQFlex 2.5-2
motor-pump may be more adapted than the current one (SQFlex 5A-7). Indeed, for the closest point of the
99
Pareto front (𝑆𝐸𝐼 = 0.070), the optimal motor-pump is the SQFlex 2.5-2. Finally, Figure IV-6 shows that
the current position of the PVWPS is close to the optimal position.
As shown by equation (37), we consider all the direct indicators with all weights equal to one in the socio-
economic impact 𝑆𝐸𝐼 for this example. We now study the results for the direct indicators along the Pareto
𝑦
front. For this purpose, we define the yearly variation of the indicator 𝑖, Δℐ𝑖 (𝒉), as:
𝑦 7
𝑑𝑟𝑦 5
Δℐ𝑖 (𝒉) = 12 Δℐ𝑖 (𝒉) + 12 Δℐ𝑖𝑤𝑒𝑡 (𝒉) (38)
Figure IV-7 presents the values of the yearly variation of the 4 direct indicators along the Pareto front. In
order to have a good understanding of Figure IV-7, we also compute, for each point of the Pareto front, the
𝑦
average number of consumers of the PVWPS over the year 𝑛𝑐 , and the number of these consumers that used
𝑦 𝑦
to go to open wells 𝑛𝑐,𝑜𝑤 and to hand pumps 𝑛𝑐,ℎ𝑝 before installation of the PVWPS. Results are given in
𝑦 𝑦 𝑦
Figure IV-8. The values of 𝑛𝑐 , 𝑛𝑐,𝑜𝑤 , 𝑛𝑐,ℎ𝑝 , which correspond to the whole year, are obtained from seasonal
𝑑𝑟𝑦 𝑑𝑟𝑦 𝑑𝑟𝑦 𝑤𝑒𝑡 5 7
values (𝑛𝑐 , 𝑛𝑐,𝑜𝑤 , 𝑛𝑐,ℎ𝑝 and 𝑛𝑐𝑤𝑒𝑡 , 𝑛𝑐,𝑜𝑤
𝑤𝑒𝑡
, 𝑛𝑐,ℎ𝑝 ) by using the 12 and 12 factors, similarly to what is done
in equation (38).
𝑦
Δℐ𝑤𝑞 (𝒉)
𝑦
Δℐ𝑒𝑒 (𝒉)
𝑦
Δℐ𝑤𝑐 (𝒉)
𝑦
Δℐ𝑑 (𝒉)
Figure IV-7 – Values of the direct indicators variation along the Pareto front.
𝑦 𝑦
𝛥ℐ𝑤𝑞 (𝒉): variation of the water quality indicator, 𝛥ℐ𝑒𝑒 (𝒉): variation of the extraction easiness indicator,
𝑦 𝑦
𝛥ℐ𝑤𝑐 (𝒉): variation of the water cost indicator, 𝛥ℐ𝑑 (𝒉): variation of the distance indicator.
100
Figure IV-8 – Number of PVWPS consumers along the Pareto front and sources where these consumers used to go
before the installation of the PVWPS.
𝑦 𝑦
𝑛𝑐 : number of PVWPS consumers, 𝑛𝑐,𝑜𝑤 : number of PVWPS consumers that used to go to an open well,
𝑦
𝑛𝑐,ℎ𝑝 : number of PVWPS consumers that used to go to a hand pump.
We now analyse the variation of the impact indicators along the Pareto front (see Figure IV-7):
𝑦
 The higher the 𝑆𝐸𝐼, the higher the yearly variation of the water quality indicator Δℐ𝑤𝑞 (𝒉). Indeed,
we observe in Figure IV-8 that the larger the PVWPS, the more households can switch from open
wells, where water is not potable, to the PVWPS.
𝑦
 The higher the 𝑆𝐸𝐼, the higher the yearly variation of the extraction easiness indicator Δℐ𝑒𝑒 (𝒉).
Indeed, we observe in Figure IV-8 that the larger the PVWPS, the more households can switch
from hand pumps and open wells, where water extraction is arduous, to the PVWPS.
𝑦
 The higher the 𝑆𝐸𝐼, the lower the yearly variation of the water cost indicator Δℐwc (𝒉). Indeed, as
the PVWPS is more expensive than other water sources, the more households switch to the
PVWPS, the higher the expenses for accessing water at the scale of the village.
𝑦
 The yearly variation of the distance indicator Δℐ𝑑 (𝒉) remains close to 0 along the Pareto front. As
explained in section IV.2.1, this is mainly due to the fact that open wells are close to the households
(see Figure IV-1).
As presented in section III.5, the life-cycle cost 𝐿𝐶𝐶 is composed of the capital cost 𝐶𝐴𝑃𝐸𝑋 and the
operational cost 𝑂𝑃𝐸𝑋. Figure IV-9 presents the 𝐶𝐴𝑃𝐸𝑋 and the 𝑂𝑃𝐸𝑋 along the Pareto front. We observe
that the 𝑂𝑃𝐸𝑋 remains nearly constant along the Pareto front, which is related to the fact that the 𝑂𝑃𝐸𝑋
only slightly depends on the values of the optimisation variables (see section III.5).
101
Figure IV-9 – Capital and operational costs along the Pareto front.
Finally, we observe in Figure IV-3 that the Pareto front is not smooth. Notably, at a couple of instances,
there is a strong increase in the 𝐿𝐶𝐶 for a small increase of the 𝑆𝐸𝐼 between two points of the Pareto front,
which we call a “vertical break”. We here explain the most significant vertical break, for which the 𝐿𝐶𝐶
increases from $2.85 104 to $3.14 104 for an increase of the 𝑆𝐸𝐼 of only 0.079 to 0.081 (see Figure IV-10).
𝑦
In Figure IV-10, we also represent the number of PVWPS consumers 𝑛𝑐 and the ratio between the number
of PVWPS consumers that used to go to an open well and the number of PVWPS consumers that used to
𝑦
𝑛𝑐,𝑜𝑤
go to a hand pump 𝑦 .
𝑛𝑐,ℎ𝑝
Pareto
𝑦
𝑛𝑐
𝑦
𝑛𝑐,𝑜𝑤
𝑦
𝑛𝑐,ℎ𝑝
Figure IV-10 – Vertical break at a 𝑆𝐸𝐼 of ~0.08.

𝑦 𝑦
𝑛𝑐 : number of PVWPS consumers, 𝑛𝑐,𝑜𝑤 : number of PVWPS consumers that used to go to an open well,
𝑦
𝑛𝑐,ℎ𝑝 : number of PVWPS consumers that used to go to a hand pump.
𝑦
At the moment of the vertical break, the total number of PVWPS consumers 𝑛𝑐 increases from 17 to 30,
which is associated to an increase in the size of the system and more specifically of the tank volume (see
Figure IV-4). This thus explains the high increase in the 𝐿𝐶𝐶. In addition, before the vertical break the ratio
𝑦
𝑛𝑐,𝑜𝑤
𝑦 is higher than 4.5 and after the vertical break this ratio is lower than 2. Besides, the positive impact is
𝑛𝑐,ℎ𝑝
102
higher when an household switches from an open well to the PVWPS than when it switches from a hand
pump to the PVWPS, thanks to higher improvements in water quality and extraction easiness (see
𝑦
𝑛𝑐,𝑜𝑤
section IV.2.1.1). Therefore, the drop of the ratio 𝑦 during the vertical break explains why the high
𝑛𝑐,ℎ𝑝
increase in the 𝐿𝐶𝐶 does not go hand in hand with a high increase of the 𝑆𝐸𝐼. To summarize, to increase the
𝑆𝐸𝐼, we see that there is no other possibility than to increase the size and therefore the cost 𝐿𝐶𝐶 of the
𝑦
𝑛𝑐,𝑜𝑤
system but this cannot be done by maintaining a very high ratio 𝑦 (> 4.5), i.e. by attracting in the greatest
𝑛𝑐,ℎ𝑝
majority households that used to go to open wells. We also observe a vertical break for a 𝑆𝐸𝐼 of ~0.08 for
the Pareto front obtained for 1000 individuals and 1800 generations (in pink in Figure IV-3). The same
explanations apply for this break as for the one observed on the Pareto front obtained with 150 individuals
and 150 generations.
103
IV.3 Influence of the error in the demand model output

In section III.2, we observed that, for the position of the current PVWPS, there is inaccuracy in the demand
model output (𝒉𝒔∗𝒂𝒇𝒕 , 𝑄𝑑 ). Indeed, there is ~10% error in the prediction of the households that wish to go to
the PVWPS (see value of 𝜙𝑎𝑓𝑡,𝑃𝑉𝑊𝑃𝑆 in section III.2.1.2). In addition, there are important differences
between the simulated and the measured demand curve at the PVWPS 𝑄𝑑 (see Figure III-3 and Figure III-4).
We remind that the simulated demand curve was obtained from the demand model with survey data as input
and that the measured demand curve was obtained from the data acquired by the monitoring system of the
current PVWPS (see section III.2.2).
In this section, we investigate the influence of the error in the demand model output on the optimisation
results. To do so, we consider the block diagram presented in Figure IV-11. We perform a first optimisation
with the measured 𝒉𝒔∗𝒂𝒇𝒕 and 𝑄𝑑 . We perform a second optimisation with the simulated 𝒉𝒔∗𝒂𝒇𝒕 and 𝑄𝑑 for
the position of the current PVWPS. As shown in Figure IV-11, the only variables for this optimisation are
the sizing variables. Indeed, we remind that the measured output of the demand model is available only for
the position of the current PVWPS, and thus the position cannot be integrated as an optimisation variable.
Interdisciplinary model
Life-cycle
Sizing Economic
cost
model
Climatic data
Groundwater Technical
parameters model
, Socio-economic
Impact
impact
model
Optimisation variables
Households behaviour
Objective functions
Figure IV-11 – Block diagram considered for investigating the influence of the error in the demand model output.
The results of the optimisations are presented in Figure IV-12. We observe that the maximum 𝑆𝐸𝐼 for the
simulated demand model output (0.073) is different from the one for the measured demand model output
(0.085). This is notably due to the error in the prediction of the households that wish to go to the PVWPS
(~10%). Results also indicate that, for a given 𝑆𝐸𝐼, the 𝐿𝐶𝐶 associated to the simulated demand model
output is always higher than the 𝐿𝐶𝐶 associated to the measured demand model output. This is due to the
fact that the simulated daily water quantity demanded is up to 2.05 times larger than the measured one (see
Table III-6).
104
Figure IV-12 – Influence of the error in the demand model output on the Pareto front.
If an error in the demand estimation or an evolution of the demand with time is observed once the PVWPS
has been installed, the PVWPS installer may adjust the peak power of the PV array 𝑃𝑝𝑣,𝑝 and choose another
motor-pump reference 𝑀𝑃. For this purpose, he can do another optimisation with 𝑃𝑝𝑣,𝑝 and 𝑀𝑃 as only
optimisation variables.
Finally, it is interesting to compare the Pareto front corresponding to the simulated demand output (in blue
in Figure IV-12), which is obtained for the PVWPS fixed at the position of the current PVWPS, to the Pareto
front of Figure IV-3, for which the position of the PVWPS is an optimisation variable which can vary in the
whole village. We observe that the maximum socio-economic impact 𝑆𝐸𝐼 that can be reached when the
position is fixed is 0.073, while, when the position is not fixed, the maximum 𝑆𝐸𝐼 is 0.105. This highlights
that the conventional approach, which consists in setting the position before performing any optimisation
(see Figure II-6), may reduce the maximum 𝑆𝐸𝐼 reachable.
105
IV.4 Influence of the expression of the socio-economic impact function

As presented in section III.4.3, the weighting coefficients 𝑤𝑖 of the socio-economic impact 𝑆𝐸𝐼 function are
chosen by the decision maker depending on the indicators that he wants to favour. In this section, we
investigate the influence of that choice on the optimisation results. We consider the expressions of the 𝑆𝐸𝐼
function given in Table IV-5.
Table IV-5 – Considered expressions of the 𝑆𝐸𝐼 function.

𝑦 𝑦 𝑦 𝑦
𝑆𝐸𝐼1 Δℐ𝑤𝑞 (𝒉) + Δℐ𝑒𝑒 (𝒉) + Δℐ𝑤𝑐 (𝒉) + Δℐ𝑑 (𝒉)
𝑦
𝑆𝐸𝐼2 Δℐ𝑤𝑞 (𝒉)
𝑦
𝑆𝐸𝐼3 Δℐ𝑒𝑒 (𝒉)
𝑦
𝑆𝐸𝐼4 Δℐ𝑤𝑐 (𝒉)
𝑦
𝑆𝐸𝐼5 Δℐ𝑑 (𝒉)
𝑦
𝑆𝐸𝐼6 Δℐ𝑑𝑖𝑎 (𝒉)
𝑦 𝑦
Δℐ𝑤𝑞 (𝒉): variation of the water quality indicator, Δℐ𝑒𝑒 (𝒉): variation of the extraction easiness indicator,
𝑦 𝑦
Δℐ𝑤𝑐 (𝒉): variation of the water cost indicator, Δℐ𝑑 (𝒉): variation of the distance indicator,
𝑦
Δℐ𝑑𝑖𝑎 (𝒉): variation of the diarrhea indicator.
We perform an optimisation for each expression of the 𝑆𝐸𝐼 function and the resulting Pareto fronts are
presented in Figure IV-13.
Figure IV-13 – Pareto fronts obtained from the different expressions of the 𝑆𝐸𝐼 function.
We observe that the expression of the 𝑆𝐸𝐼 function has a large influence on the Pareto front. We now analyse
the Pareto front obtained for each expression of the 𝑆𝐸𝐼 function:
𝑦 𝑦 𝑦 𝑦
 The Pareto front obtained for 𝑆𝐸𝐼1, i.e. Δℐ𝑤𝑞 (𝒉) + Δℐ𝑒𝑒 (𝒉) + Δℐ𝑤𝑐 (𝒉) + Δℐ𝑑 (𝒉), is the same as
the one presented in section IV.2.2. Indeed, the expression of the 𝑆𝐸𝐼 function and the values of the
parameters are the same.
𝑦
 The Pareto front obtained for 𝑆𝐸𝐼2 , i.e. Δℐ𝑤𝑞 (𝒉), reaches high values of the socio-economic impact
𝑆𝐸𝐼 (up to 0.20). For the different points of the front, we looked at the sources where consumers of
the PVWPS used to go before installation of the PVWPS (not shown here). Logically, we observe
that the optimisation targets in priority households that used to go open wells. Indeed, when an
household switches from an open well to the PVWPS, there is an increment in water quality.
106
𝑦
 The Pareto front obtained for 𝑆𝐸𝐼3 , i.e. Δℐ𝑒𝑒 (𝒉), reaches high values of the 𝑆𝐸𝐼 (up to 0.20). When
looking at the sources where consumers of the PVWPS used to go before installation of the PVWPS
(not shown here), we see that the optimisation targets both households that used to go to open wells
and that used to go to hand pumps. Indeed, when an household switches from an open well or a
hand pump to the PVWPS, there is an increment in extraction easiness. In addition, for the point
with the highest 𝑆𝐸𝐼 (0.20), all the households that wish to go to the PVWPS are able to go there
𝑦 𝑦
(𝑛𝑐 = 𝑛𝑑 = 41). Indeed, 𝑆𝐸𝐼3 is maximum when as many households as possible switch from their
previous source (open well or hand pump) to the PVWPS.
𝑦
 There is no Pareto front for 𝑆𝐸𝐼4, i.e. Δℐ𝑤𝑐 (𝒉). Indeed, as presented in section II.2.4, collecting
water at the PVWPS is more expensive than at any other water source of the village.
𝑦
 The Pareto front obtained for 𝑆𝐸𝐼5 , i.e. Δℐ𝑑 (𝒉), does not reach high values of the 𝑆𝐸𝐼 (maximum
reached of 0.04). Indeed, due to the high density of hand pumps and open wells in the village (see
Figure IV-1), most of households were not travelling a long way to collect water before the
installation of the PVWPS. This leaves small room for the improvements that can be provided by
the optimisation.
𝑦
 The Pareto front obtained for 𝑆𝐸𝐼6 , i.e. Δℐ𝑑𝑖𝑎 (𝒉), does not reach very high values of the 𝑆𝐸𝐼
(maximum reached of 0.07). It is interesting to observe that the maximum reached by 𝑆𝐸𝐼6 is equal
to 𝛼1 (0.35) times the maximum reached by 𝑆𝐸𝐼2 (0.20). This is consistent with the fact that the
variation of the diarrhea indicator is obtained by multiplying the variation of the water quality
indicator by the coefficient 𝛼1 (see equation (24)). In addition, as for 𝑆𝐸𝐼2 , the households that used
to go open wells are targeted in priority by the optimisation.
107
IV.5 Influence of the groundwater parameters

As presented in sections II.2.3 and III.3.1.1.5, it is not possible to determine the values of the groundwater
parameters (𝐻𝑏,𝑠 , 𝜅0 and 𝜇0 ) for all positions (𝐿𝑎𝑡, 𝐿𝑜𝑛) in a village. Indeed, as shown in section III.3.1.1.5,
these parameters can be known only once a borehole has been drilled and pumping tests have been
performed. At the scale of a village, we can only provide a range of variation (lower and upper boundary)
of these parameters. In Appendix E, we show how to determine this range and deduce the range of variation
of these parameters for the case of Gogma.
In order to quantify the effect of this uncertainty on groundwater parameters, we study here the influence of
the values of the groundwater parameters on the optimisation results. For each parameter, we consider the
lower boundary, the upper boundary and the value for the current PVWPS of Gogma (see Table IV-6).
Table IV-6 – Value of the groundwater parameters for the sensitivity analysis – case of Gogma.
Groundwater parameter Symbol Lower boundary Current PVWPS Upper boundary
Height between the ground level and the 𝐻𝑏,𝑠 0m -4.9 m -25 m
static water level in the borehole
Aquifer losses coefficient 𝜅0 0 m-2 s 2.0 103 m-2 s 6.9 103 m-2 s
Borehole losses coefficient 𝜇0 0m s -5 2 5
5.8 10 m s-5 2
1.9 106 m-5 s2
First of all, we perform an optimisation for each value of 𝐻𝑏,𝑠 in Table IV-6, and keeping 𝜅0 and 𝜇0 equal
to their value for the current PVWPS of Gogma. The results are presented in Figure IV-14. We then proceed
in the same way for 𝜅0 and 𝜇0 as for 𝐻𝑏,𝑠 . The results for 𝜅0 and 𝜇0 are given in Figure IV-15 and
Figure IV-16 respectively.
Figure IV-14 – Pareto fronts obtained for different values of the static water level 𝐻𝑏,𝑠 .
108
Figure IV-15 – Pareto fronts obtained for different values of the aquifer losses coefficient 𝜅0 .
Figure IV-16– Pareto fronts obtained for different values of the borehole losses coefficient 𝜇0 .
Regarding the influence of static water level 𝐻𝑏,𝑠 on the Pareto front (see Figure IV-14), we observe that
the highest 𝑆𝐸𝐼 reached is the same for all three values of 𝐻𝑏,𝑠 (0.105). However, in most cases, for a given
𝑆𝐸𝐼, the lower the value of 𝐻𝑏,𝑠 , the higher the 𝐿𝐶𝐶. For instance, for a given 𝑆𝐸𝐼, the 𝐿𝐶𝐶 for 𝐻𝑏,𝑠 = -25 m
is always higher than the 𝐿𝐶𝐶 for 𝐻𝑏,𝑠 = 0 m (the difference between both 𝐿𝐶𝐶 reaches up to $2.3 103).
Regarding the aquifer loss coefficient 𝜅0 (Figure IV-15) and the borehole loss coefficient 𝜇0 (Figure IV-16),
we do not observe a specific influence of these parameters on the Pareto front.
We performed a similar analysis in [95]. In this article, we considered a mono-objective optimisation to

determine the sizing of the PVWPS that allows to minimize the 𝐿𝐶𝐶 while fulfilling the water demand at
the current PVWPS. We performed this optimisation for several values of the groundwater parameters and
we also found that 𝐻𝑏,𝑠 more strongly influences the optimisation results than 𝜅0 and 𝜇0 . All these results
regarding the influence of groundwater parameters, are due to the fact that 𝐻𝑏,𝑠 has a larger influence on the
water level in the borehole 𝐻𝑏 than 𝜅0 and 𝜇0 [95]. Even though 𝜅0 and 𝜇0 have a low influence on the
Pareto front, it remains essential to also measure these groundwater parameters through pumping tests, once
the borehole has been drilled at a given position. Indeed, it permits to refine the optimal sizing of the PVWPS
109
and more importantly to make sure that the sizing chosen for the PVWPS does not threaten the sustainability
of groundwater resources.
We also looked at the optimal latitude 𝐿𝑎𝑡 and longitude 𝐿𝑜𝑛 along the Pareto fronts of Figure IV-14,
Figure IV-15 and Figure IV-16. In Table IV-7, we give, for each Pareto front, the minimum and maximum
reached by the optimal latitude and longitude.
Table IV-7 – Range of variation of the optimal position of the PVWPS for different values of the groundwater
parameters.
Groundwater Value Range of variation of the optimal latitude and of the optimal longitude:
parameter [range 𝐿𝑎𝑡 in °]×[range 𝐿𝑜𝑛 in °]
0m [11.723 to 11.726] × [-0.573 to -0.567]
𝐻𝑏,𝑠 -4.9 m [11.723 to 11.726] × [-0.573 to -0.571]
-25 m [11.723 to 11.725] × [-0.573 to -0.571]
0 m-2 s [11.723 to 11.725] × [-0.573 to -0.570]
𝜅0 2.0 103 m-2 s [11.723 to 11.726] × [-0.573 to -0.571]
6.9 103 m-2 s [11.724 to 11.725] × [-0.573 to -0.571]
0 m-5 s2 [11.722 to 11.725] × [-0.573 to -0.570]
𝜇0 5.8 105 m-5 s2 [11.723 to 11.726] × [-0.573 to -0.571]
1.9 106 m-5 s2 [11.724 to 11.725] × [-0.574 to -0.571]
For each groundwater parameter, we observe that, for any value of the parameter, the optimal position of
the PVWPS remains in the same area of the village ([11.722 to 11.726] × [-0.574 to -0.567]). Therefore, the
values of the groundwater parameters do not have a significant influence on the optimal position of the
PVWPS. This may lead us to think that the uncertainty on groundwater parameters is not so detrimental for
the positioning of the PVWPS. However, it is important to keep in mind that the variation of the drilling
cost with the position (𝐿𝑎𝑡, 𝐿𝑜𝑛) of the PVWPS cannot be considered (see section III.5.1.2).
110
IV.6 Proposition of an improved procedure for the design and installation of PVWPS
The proposed improved procedure for the design and installation of PVWPS is presented in Figure IV-17.
It is based on the conventional procedure (see Figure II-6) and includes the optimisation methodology
developed. In section IV.6.1, we discuss the main changes from the conventional procedure (in purple in
Figure IV-17). In section IV.6.2, we show the application of the methodology for a case study and we go
through all the steps.
IV.6.1 Procedure
In order to evaluate the demand and impact models (see sections III.2 and III.4), data must be collected in
the village. This is the purpose of step 1. The GPS coordinates of all the households and sources must be
gathered as well as the water cost at all sources. In addition, the source choice 𝒉𝒔𝒃𝒆𝒇 of at least 70% of the
households of the village is recorded as well as their perception of water quality and extraction easiness at
the sources they use. Regarding, the factual water quality of the water sources, it seems possible to assume
that water from open wells is not potable and that water from hand pumps is potable, instead of performing
water quality tests for all sources of the village. Indeed, this assumption is in accordance with the literature
(see section I.1.1), has proven to be accurate for all sources in Gogma (see section II.2.7.2), and allows to
save the cost of water quality tests (see Table II-2). According to section II.2, we estimate that the data
collected at step 1 can be gathered in 5 days by one collector for a cost of $800 in Gogma. When we compare
this number to the potential savings from the application of the methodology in Gogma (~$7.7 103, see
section IV.2.2), the application of the improved procedure appears to be economically viable.
A first round of bi-objective optimisation is performed at steps 2 and 3, which allows to propose a first
position (𝐿𝑎𝑡1 , 𝐿𝑜𝑛1 ), around which groundwater resources will be investigated. For this optimisation, the
weights of the 𝑆𝐸𝐼 function are set by the decision maker depending on the indicators that he wants to
favour. Besides, the constraint that the PVWPS cannot be located in the areas to avoid (e.g. burial sites,
unsafe areas) is set. Finally, the groundwater parameters are set to their upper boundary (see Table IV-6)
for all positions (𝐿𝑎𝑡, 𝐿𝑜𝑛) in the village, which helps to avoid costs higher than expected when the borehole
is effectively drilled. Thanks to this first round of optimisation, the decision maker can now count on a
support tool which helps him to identify a potential position, around which to install the PVWPS (𝐿𝑎𝑡1 ,
𝐿𝑜𝑛1). This position maximises the positive socio-economic impact 𝑆𝐸𝐼 on the village while minimising
the cost of the system 𝐿𝐶𝐶. We remind that in the conventional situation the decision maker could only rely
on his intuition for this positioning step (see section I.3.2).
During the steps 4 to 6, the borehole is drilled, the water quality is tested and the groundwater parameters
(𝐻𝑏,𝑠 , 𝜅0 and 𝜇0 ) and the maximum flow rate that can be pumped 𝑄𝑝,𝑚𝑎𝑥 are determined through pumping
tests (see sections II.2.6 and III.3.1.1.5).
111
1
Gather GPS coordinates of all households, existing water sources and areas to avoid (e.g.
burial sites, unsafe areas). Collect the water cost at existing sources.
Record the source choice of at least 70 % of the households and their perceived water
quality and extraction easiness at the sources they use.
2
The decision maker sets the weights between the indicators of the function. Set the
constraint that the PVWPS cannot be installed in areas to avoid. Set the groundwater
parameters ( , ) to their upper boundaries for all positions ( , ) in the village.
Perform the optimization.
3
The decision maker selects an optimal design. The corresponding position is noted
( ).
2b
Add the constraint that the
4 PVWPS cannot be installed in
Perform geophysical measures along several profiles in a square of ~350 350 m around
a square of ~350 350 m
( ).
around ( ). Perform
the optimization to look for a
position ( , ).
Do geophysical measures indicate a suitable
No
position ( to drill within the square?
Yes
5
Drill a borehole at ( ).
Is there water flowing out of the borehole? No
Yes
6
Perform pumping tests. Determine the groundwater parameters ( , and ) and the
maximum flow rate that can be pumped thanks to the pumping tests. Perform
physico-chemical and bacteriological tests on the water flowing out of the borehole.
Does the physico-chemical water quality

meet the regulation for domestic consumption?
No
Does the decision maker consider that
the bacteriological water quality is satisfying?
Yes
7
Perform the optimization with the values of the groundwater parameters determined and
with the position of the PVWPS fixed at ( ). The constraint that the pumped flow
rate must remain lower than is set.
Does the decision maker consider that it is worth

No
installing a PVWPS at the position ( )?
Yes
8
The decision maker selects an optimal sizing (i.e. PV array peak power , motor-pump
reference and tank volume .
9
Install the PV array, motor-pump, tank and fountain.
10
Perform bacteriological tests on the water flowing out from the fountain.
Does the decision maker consider that

the bacteriological water quality is satisfying? No
Changes from conventional procedure
Yes Positioning step
11 Sizing step
Open the PVWPS for consumption to the inhabitants. Path taken during the case study
Figure IV-17 – Procedure of application of the developed optimisation methodology.
112
The second round of optimisation is then performed at steps 7 and 8, which allows to determine the values
of the sizing variables. For this optimisation, the position is fixed at the position of the borehole and the
constraint that the pump flow rate 𝑄𝑝 must remain lower than 𝑄𝑝,𝑚𝑎𝑥 is set (𝑄𝑝 (𝑡) < 𝑄𝑝,𝑚𝑎𝑥 , ∀𝑡), in order
to preserve the sustainability of groundwater resources. It is important to note that the determination of the
groundwater parameters (𝐻𝑏,𝑠 , 𝜅0 and 𝜇0 ) is key to ensure that the above mentioned constraint is respected,
as these parameters are involved in the computation of the pump flow rate 𝑄𝑝 (see section III.3.1.1.5).
Overall, this second round of optimisation helps the PVWPS installation company and the decision maker
to determine the sizing of the PVWPS that maximize the positive socio-economic impact 𝑆𝐸𝐼 and minimize
the life-cycle cost 𝐿𝐶𝐶, while preserving groundwater resources.
Finally, we also propose some modifications regarding bacteriological tests. Firstly, we propose to add
bacteriological tests at the same time as physico-chemical tests, at step 6. Indeed, bacteriological tests cost
only ~$20 (see section II.2.7.2) and they will provide information about the bacteriological quality of the
groundwater. If the decision maker is not satisfied with the quality of the groundwater, he may then decide
to change the position of the PVWPS before further investments. Secondly, we recommend that the second
bacteriological tests are performed before the opening of the PVWPS for consumption to the inhabitants
(see steps 10 and 11).
IV.6.2 Case study

In this section, we apply the proposed procedure to the village of Gogma. However, the decision maker
choices and the results of the geophysical studies, of the pumping tests and of the water quality analyses
performed at the PVWPS are fictive. The path considered for this case study is represented in blue in
Figure IV-17.
Step 1: The required data are collected (GPS coordinates, sources cost, source choice …). We also remind
that there are no areas to avoid in Gogma (see section II.1.2.1).
Step 2: We suppose that the decision maker considers the expression of the socio-economic impact 𝑆𝐸𝐼,
which features all the direct impact indicators with all weights equal to 1, as in section IV.2. The
groundwater parameters (𝐻𝑏,𝑠 , 𝜅0 and 𝜇0 ) are set to their upper boundary (see Table IV-6) for all positions
of the village. We perform the optimisation and we obtain the Pareto front in blue in Figure IV-18.
Step 3: We suppose that all the points of the Pareto front are within the budget of the decision maker and
that he decides to consider the point of maximum 𝑆𝐸𝐼: 𝑆𝐸𝐼 = 0.105 and 𝐿𝐶𝐶 = $3.68 104. The corresponding
position of the PVWPS is 𝐿𝑎𝑡1 = 11.72480° and 𝐿𝑜𝑛1 = -0.57081°.
Step 4: The geophysical study is performed in a square of 350×350 m around the identified position. The
coordinates of the scanned square are 𝐿𝑎𝑡 ∈ [11.72300°, 11.72650°] × 𝐿𝑜𝑛 ∈ [-0.57256°,-0.56906°]. We
suppose that no suitable position to drill is encountered in the square.
Step 2b: We add the constraint that the position of the PVWPS cannot be in the scanned square and we run
the optimisation again. The new Pareto front obtained is presented in red in Figure IV-18. Note that, if there
would have been areas to avoid (e.g. burial sites, unsafe areas), they would have been considered with the
same type of constraint on the PVWPS position.
113
Step 3: Due to the constraint on the position, the maximum 𝑆𝐸𝐼 that can be reached has dropped to 0.078.
We suppose that the decision maker selects this point of maximum 𝑆𝐸𝐼. The corresponding position of the
PVWPS is 𝐿𝑎𝑡2 = 11.72480° and 𝐿𝑜𝑛2 = -0.57263°.
Step 4: New geophysical studies are performed, and we suppose that this time a suitable position to drill is
encountered at the position 𝐿𝑎𝑡𝑆 = 11.72500° and 𝐿𝑜𝑛𝑆 = -0.57290°.
Step 5: The borehole is drilled and water is flowing out from the borehole.
Step 6: Pumping tests are performed. According to the pumping tests, the maximum flow rate that can be
pumped is 𝑄𝑝,𝑚𝑎𝑥 = 1.5 10-3 m3/s and the values of the groundwater parameters are 𝐻𝑏,𝑠 = 17 m,
𝜅0 = 1.5 103 m-2 s and 𝜇0 = 5.0 105 m-5 s2. Physico-chemical and bacteriological tests are performed on the
water from the borehole. We suppose that the water is found to be suitable for drinking in terms of physico-
chemical and bacteriological quality.
Step 7: The optimisation is run to determine the sizing of the PVWPS. For this optimisation, the values of
𝐿𝑎𝑡 and 𝐿𝑜𝑛 are set to the position of the borehole that has been drilled (𝐿𝑎𝑡𝑆 , 𝐿𝑜𝑛𝑆 ), we use the values of
the groundwater parameters that have been determined from the pumping tests and we set the constraint that
the pump flow rate 𝑄𝑝 must remain lower than 𝑄𝑝,𝑚𝑎𝑥 . We obtain the Pareto front presented in yellow in
Figure IV-18.
Step 8: We suppose that the decision maker chooses the point of highest 𝑆𝐸𝐼 (0.077) which corresponds to
the following values of the sizing variables: 𝑃𝑝𝑣,𝑝 = 1.0 103 Wp, 𝑀𝑃 = SQFlex 2.5-2 and 𝑉𝑡 = 1.3 m3.
Step 9: The PV array, motor-pump, tank and fountain are installed.
Step 10: Bacteriological tests are performed on the water from the fountain and we suppose that the water
is found to be suitable for drinking in terms of bacteriological quality. According to the results of the
physico-chemical tests of step 6 and of the bacteriological tests of step 10, the water is potable.
Step 11: The PVWPS is opened for consumption to the inhabitants.
Figure IV-18 – Pareto fronts obtained through the proposed procedure of design and installation of PVWPS.
114
The procedure therefore appears as viable and should be tested in reality for the design of new PVWPS for
domestic water access. The implementation of the procedure should be performed in collaboration with
decision makers and local companies. With this in mind, the development of an interface for facilitating the
use of the procedure by these stakeholders may be useful. The interface should notably permit to easily load
data acquired by GIS mapping and by geophysical and hydrological measuring devices, as these data are
required for the application of the procedure.
115
IV.7 Partial conclusion

In this chapter, we first presented the optimisation problem. The objective functions are the life-cycle cost
of the PVWPS, that we want to minimise, and its socio-economic impact, that we want to maximise. The
optimisation variables are the PV array peak power, the motor-pump reference, the tank volume and the
PVWPS latitude and longitude. The constraints are related to the operation of the motor-pump and to the
preservation of groundwater resources. We then detailed and interpreted a reference optimisation result. In
addition, we performed several analyses to understand the influence of the error in the demand model output,
of the expression of the socio-economic impact function and of the groundwater parameters on the
optimisation results. Finally, we proposed and described an improved procedure for the design and
installation of PVWPS. It is based on the conventional procedure and includes the optimisation methodology
developed.
116
Conclusion
117
Conclusion
Overview
Improving water access remains one of the most significant challenges in rural sub-Saharan Africa.
Photovoltaic water pumping systems (PVWPS) have proved to be an interesting solution to improve water
access in off-grid rural areas. Our literature review highlighted that the conventional approach for the
optimal design of PVWPS aims at determining the sizing of the PVWPS that minimizes its cost while
maximising the share of the water needs of the inhabitants that is fulfilled. We identified two main
drawbacks of this approach. Firstly, the positioning of the PVWPS is not considered in the optimal design
and is performed in an arbitrary way by the decision maker, without any support tool. Secondly, the
conventional approach does not aim at maximising the positive socio-economic impact of the PVWPS (e.g.
use of water of a higher quality, reduced distance to collect water), although it is the main objective of
institutions and governments which finance these systems. This also prevents from targeting the inhabitants
of the village that have the worst access to water.
The aim of this PhD thesis was therefore to bridge these two gaps by developing a methodology that allows
to determine the PVWPS sizings and positions in a village that both minimize the life-cycle cost of the
PVWPS and maximize its positive socio-economic impact.
In order to determine the conventional procedure for the design and installation of PVWPS, to be able to
apply the proposed methodology with actual data and to compare model results to experimental
measurements, a PVWPS was designed and installed in the conventional way in the rural off-grid village of
Gogma in Burkina Faso. We performed the GIS mapping of the village; we acquired cost and water quality
data on the 22 water sources of the village; we performed 88 household surveys before the installation of
the PVWPS and another 88 after the installation. Besides, hydrological measurements were performed for
the installed PVWPS and a data logger was developed and installed to monitor the PVWPS continuously.
To this date, 20 months of technical data have been collected by the monitoring system on the PVWPS
(>2 107 data points), which forms a unique database for the study of such systems.
Then, we developed an interdisciplinary model which links the sizing and the position of the PVWPS to its
socio-economic impact and its life-cycle cost. This interdisciplinary model is composed of four sub-models:
demand, technical, impact and economic model. We presented these four sub-models and demonstrated
their application with data from Gogma. The parameters of the demand and technical models were identified
using local data and their results were compared to experimental measurements.
Finally, we defined an optimisation problem which aims at determining the sizings and the positions of the
PVWPS which minimize the life-cycle cost of the PVWPS and maximize its socio-economic impact. We
started by detailing a reference optimisation result. Then, we studied the influence of the error on the demand
model output, of the definition of the socio-economic impact by the decision maker and of the groundwater
resources on the optimisation results. Finally, we proposed an improved procedure for the design and
installation of PVWPS, which is based on the conventional procedure and includes the optimisation
methodology developed.
118
Contributions
The contributions of this PhD thesis can be split into methodological contributions, that may apply to other
types of systems than water systems, contributions to the field of water access, and technical contributions
on PVWPS.
The two main methodological contributions are:
 The inclusion of the position of the system as an optimisation variable. This is notably done through
the development of the demand model that simulates the demand for the service provided by the
system depending on its position.
 The consideration of the socio-economic impact as an objective function of the optimisation. This
is done through the development of the demand, technical and impact models that link the
optimisation variables to the socio-economic impact.
The two main contributions for water access are:
 The development of a demand model that takes into account seasonality, considers the inclusion of
a source of a new type, and predicts the load curve at the future water source. The load curve
predicted by the demand model can then be used for the technical sizing of the future water system.
 The development of a theory of change regarding the repercussions of choosing a given water
source, which goes along with the identification and organisation of relevant references of the
literature on the subject. This helps to better understand the consequences of investing in improved
water sources.
The six main technical contributions on PVWPS are:
 The development of a data logger and its use for the continuous monitoring of the PVWPS of
Gogma. To our knowledge, it is the first time that a PVWPS for domestic water access has been
monitored and that data on a PVWPS in rural sub-Saharan Africa have been collected. This permits
a more precise forecast of the performance and sustainability of PVWPS for this water use and in
this region, where they are particularly useful.
 The inclusion of the water demand at the PVWPS as an input of the energy conversion chain model.
This allows to model the instantaneous operation of PVWPS which include a tank and a controller
that stops and restarts the motor-pump depending on the water level in the tank.
 The development and the experimental validation of a data-driven borehole model. Contrarily to
previous models, the developed model is not based on assumptions that are rarely met (e.g.
homogenous and isotropic aquifer) and can be applied for all types of aquifers.
 The demonstration that irradiance data from a satellite database can be used instead of data from a
local sensor for modelling and optimising PVWPS. This can favour the implementation of PVWPS
in areas where no local irradiance measurements are available, which is often the case in sub-
Saharan Africa.
 The demonstration that a time step of 1 hour can be used for modelling and optimising PVWPS,
instead of smaller time steps (e.g. 1 minute). This notably allows to reduce computing time for the
optimisation.
119
 The demonstration that is not indispensable to consider the thermal behaviour of PV modules and
the evolution of the ambient temperature 𝑇𝑎 when modelling PVWPS for domestic water access.
Perspectives
The first perspective concerns the enhancement of the interdisciplinary model. There is room for improving
the four sub-models:
 For the demand model, the evolution of the demand along the lifetime of the system should be
integrated. Indeed, the habits of the inhabitants may evolve over the lifetime of system (~20 years)
and/or following technical modifications in the system. For instance, if a piping system that reaches
each house individually is installed, the water quantity demanded and the time at which people use
water may vary along time, as people get used to living with more accessible water.
 For the technical model, ageing of the PVWPS components and the evolution of groundwater
resources with time, and notably with climate change, could be quantified for the site of Gogma.
This objective is aligned with the aim of monitoring continuously the PVWPS of Gogma along its
whole lifetime.
 For the impact model, the quantification of missing 𝛼 coefficients, which are coefficients that relate
two impact indicators, should be performed. Although the determination of all 𝛼 coefficients of the
theory of change for a given case study does not appear realistic, performing studies that specifically
aim at determining one 𝛼 coefficient is helpful as the obtained coefficient may be used in other case
studies. Besides, in this thesis we only consider socio-economic impacts between before and after
installation of the PVWPS. In the future, the evolution of the socio-economic impacts over the
whole lifetime of the system should be integrated.
 For the economic model, the main scientific obstacle is the forecast of drilling costs depending on
the position. Despite the key importance of forecasting drillings costs and more generally
groundwater resources, it has remained a very challenging problem for decades in rural sub-Saharan
Africa.
The second perspective is the application of the methodology to other case studies involving PVWPS. This
will allow to see how the identified parameters of each sub-model and the results of the optimisation vary
from one case study to another. This will go hand in hand with the development of an interface for
facilitating the application of the proposed methodology.
The third perspective is the use of the methodology for assessing the suitability of technically original
solutions for water pumping (e.g. motor-pumps ‘Saurea’ and ‘Futurepump’). It may therefore be used as a
tool to guide governments in the financing of research and development efforts.
The fourth perspective is the application of the methodology to other types of systems. According to the
specificities of the methodology, systems for which the demand for the service provided depends on the
system’s position and that have a significant socio-economic impact should be privileged. In the frame of
rural areas of developing countries, community mills powered by photovoltaic energy where the inhabitants
bring their grains to make flour may be a particularly interesting application. Indeed, depending on the mill’s
position different farmers may be targeted and there are significant socio-economic impacts associated with
120
being able to sell transformed agricultural products. In the frame of developed countries, an application may
be the positioning of public charging stations for electrical vehicles (e.g. cars, bikes, scooters). Indeed, the
demand at the charging station depends on its position and socio-economic impacts may include time loss
to go to the charging station and facilitated access for people with reduced mobility.
121
List of Publications
Peer-reviewed journal articles – published/accepted
1. S. Meunier, M. Heinrich, L. Queval, J.A. Cherni, L. Vido, A. Darga, Ph. Dessante, B. Multon, P.K.
Kitanidis, C. Marchand, "A validated model of a photovoltaic water pumping system for off-grid rural
communities", Applied Energy, vol 241, pp. 580-591, 2019,
DOI: https://doi.org/10.1016/j.apenergy.2019.03.035
2. S. Meunier, D. T. Manning, L. Queval, J.A. Cherni, Ph. Dessante, D. Zimmerle "Determinants of the
marginal willingness to pay for domestic water and irrigation in partially electrified villages: the case
of Rwanda", International Journal of Sustainable Development & World Ecology, vol 26, no 6, pp. 547-
559, 2019, DOI: https://doi.org/10.1080/13504509.2019.1626780
3. C.O. Peralta P., G.T.T. Vieira, S. Meunier, R.J. Vale, M.B.C. Salles, B.S. Carmo, "Evaluation of the
CO2 emissions reduction potential of Li-ion batteries in ship power systems", Energies, vol 12, no 3, art
375, 2019, DOI: https://doi.org/10.3390/en12030375
4. T. Vezin, S. Meunier, L. Queval, J.A. Cherni, L. Vido, A. Darga, Ph. Dessante, P.K. Kitanidis,
C. Marchand, "Borehole water level model for photovoltaic water pumping systems”, Applied Energy,
2019, DOI: https://doi.org/10.1016/j.apenergy.2019.114080
Peer-reviewed journal articles – submitted
5. M. Heinrich, S. Meunier, A. Samé, L. Quéval, A. Darga, L. Oukhellou, B. Multon, "Detection of

cleaning interventions on photovoltaic modules with machine learning”
6. S. Meunier, L. Queval, A. Darga, Ph. Dessante, C. Marchand, M. Heinrich, J.A. Cherni, E.A. de la
Fresnaye, L. Vido, B. Multon, P.K. Kitanidis, "Modelling and optimal sizing of photovoltaic water
pumping systems – Sensitivity analysis"
Peer-reviewed international conference papers – published/accepted
7. S. Meunier, L. Queval, M. Heinrich, E.A. de la Fresnaye, J.A. Cherni, L. Vido, A. Darga, Ph. Dessante,
B. Multon, P.K. Kitanidis, C. Marchand, "Effect of irradiance data on the optimal sizing of photovoltaic
water pumping systems," 46th IEEE Photovoltaic Specialists Conference (PVSC), Chicago (IL), United
States, June 2019 – Nominated for the best student paper award
8. S. Meunier, L. Queval, A. Darga, Ph. Dessante, C. Marchand, M. Heinrich, J.A. Cherni, E.A. de la
Fresnaye, L. Vido, B. Multon, P.K. Kitanidis, "Modelling and optimal sizing of photovoltaic water
pumping systems – Sensitivity analysis," 14th International Conference on Ecological Vehicles and
Renewable Energies (EVER), Monaco, Monaco, May 2019,
DOI: https://doi.org/10.1109/EVER.2019.8813580 – Award of the best paper on renewable energies,
Eligible for re-submission to the IEEE Transactions on Industry Applications
9. S. Meunier, L. Queval, A. Darga, Ph. Dessante, C. Marchand, M. Heinrich, J.A. Cherni, L. Vido, B.
Multon, "Influence of the temporal resolution of the water consumption profile on photovoltaic water
122
pumping systems modelling and sizing," 7th International Conference on Renewable Energy Research
and Application (ICRERA), Paris, France, Oct. 2018,
DOI: https://doi.org/10.1109/ICRERA.2018.8566828
Peer-reviewed national conference papers – published
10. S. Meunier, M. Heinrich, L. Queval, J.A. Cherni, L. Vido, A. Darga, Ph. Dessante, B. Multon, P.K.
Kitanidis, C. Marchand, “Conception optimale des systèmes photovoltaïques de pompage d’eau en sites
isolés avec prise en compte des aspects socio-économiques,” Conf. des Jeunes Chercheurs en Génie
Électrique (JCGE), Ile d'Oleron, France, June 2019,
URL : http://seeds.cnrs.fr/jcge-seeds-2019/264996.pdf
11. S. Meunier, M. Heinrich, L. Queval, A. Darga, J.A. Cherni, L. Vido, Ph. Dessante, B. Multon, C.
Marchand, “Étude multidisciplinaire d’un système de pompage photovoltaïque dans une communauté́
rurale du Burkina Faso”, Journées Nationales du Photovoltaïque (JNPV), Dourdan, December 2018,
URL: https://jnpv2018.geeps.centralesupelec.fr/archives/11-
autres/JNPV_Meunier_Simon_Pompage_PV-Simon_Meunier_GeePs.pdf
12. A. N’Dour, S. Meunier, A. Kaboré , M. Heinrich, L. Queval, A. Darga, “Modules photovoltaïques

contrefaits en vente sur le marché Africain : détection et caractérisation électrique d’échantillons issus
du marché Burkinabé”, Journées Nationales du Photovoltaïque (JNPV), Dourdan, December 2018,
URL: https://jnpv2018.geeps.centralesupelec.fr/archives/11-autres/JNPV_Arouna_28092018-
Arouna_DARGA_GeePs-Centralesupelec.pdf
13. S. Meunier, M. Heinrich, J.A. Cherni, L. Queval, Ph. Dessante, L. Vido, A. Darga, B. Multon C.
Marchand, “Modélisation et validation expérimentale d’un système de pompage photovoltaïque dans
une communauté rurale isolée du Burkina Faso,” Symposium de Génie Electrique (SGE), Nancy,
France, July 2018, URL: http://actes.sge-conf.fr/2018/articles/article_182692.pdf
14. S. Meunier, "Optimisation d’un système de pompage photovoltaïque pour les communautés rurales des
pays en voie de développement", Conf. des Jeunes Chercheurs en Génie Électrique (JCGE), Arras,
France, June 2017, URL: https://hal.archives-ouvertes.fr/hal-01629662
123
Appendix
124
Appendix
Appendix A. Household survey

Preliminary remark: The layout of the original survey has been adapted here to save space. The survey has been
translated from French.
Text explaining the survey, to read completely to the surveyee
Hello,
I am here on behalf of French and English laboratories to research water use in rural Burkina Faso. You have been
selected to answer our survey.
Your participation is completely optional, however your responses will help us understand living condition in Burkina
Faso.
We will be asking you personal information about your family and household. Do not worry, survey results are strictly
confidential. Your answers will be anonymized. When the information is summarized for your village, you may see
the results if you are interested. We will follow up with you in 1 year to see how the answers to the survey change
over time.
Please feel free to speak with all other members of your household to answer to the questions of the survey. If questions
are unclear, let me know and I can explain further.
Do you agree to participate to this survey?
 Yes
 No
Thank you for your time,
University of Paris-Saclay / Imperial College London
Thank you for participating in this research!
If you have any questions about this project, you can contact the project coordinator at any time at:
simon.meunier@centralesupelec.fr.
Would you like a written version of these contacts?
a. Household Characteristics
Surveyor. Date and time of the survey: ________________________
Surveyor. GPS coordinates of the household: Lat = _________; Lon = ___________
S1. Are there specific problems in the village that worry people in your household?
 Yes, detail:________________________
 No
125
Give the following description for each members of the household:
S2. S3. S4. S5. Does S6. S7. S8. S9. S10. S11. S12. S13.
N° Name Present he/she live Position in the Observation of Gender Age Occupation Highest Does Does
member during here household children. Do level of he he
survey? usually or they have the education know know
he/she is following validated how how
just characteristics? to to
passing? read? write?
1 Yes 1. Usually 1. Head of the 1. Distended Male/ 1. Agriculture Yes Yes
No 2. Just household belly Female 2. Livestock No No
passing 2. 1st wife of the 2. Fair hair 3. Building
head 3. Oedema on 4. Manufacture
3. 2nd wife of the face or 5. Transformation
head limbs 6. Trade
4. 3rd wife of the 7. Service
head 8. Student
5. 4th wife the 9. Housewife
head 10. Other:
6. Son/daughter
of the head
7. Father/mother
of the head
8. Brother/Sister
of the head
9. Other:
Same
question
for the
other
members
of the
household
Comments:
S14. How many houses do you own? _______________________________
Observation. Characteristics of the houses owned by the household:

House n° S15. Material of the walls S16. Material of the roof S17. Material of the floor
1 1. Clay 1. Straw 1. Clay
2. Concrete 2. Sheet metal 2. Concrete
3. Other:_________ 3. Other:_______ 3. Straw Carpet
4. Woven Carpet
5. Other
Same questions for other
houses
Comments:
S18. Is there anything that your household need? __________________________________________
126
Does the household have access to the following services?
Service Answer
S19. Lighting  Yes, with kerosene
 Yes, with electricity
 Yes, other: _____
 No
S20. Food and water heating  Yes, with wood
 Yes, with gas
 Yes, other: _____
 No
S21. Do people of your household sleep under a mosquito  Yes, everybody
netting?  Yes, but only _____
 No
S22. How many phones does your household own?  Yes
 No
S23. Is the phone network fine in your houses?  Yes
 No
S24. Do people of your household have access to the  Yes, on the phone
internet?  Yes, at a shop with a computer
 Yes, other:______
 No
S25. How many bikes does your household own?  Yes
 No
S26. How many motorbikes does your household own?  Yes
 No
Comments:
b. Agriculture and livestock

S27. In how many fields do people of your household work in? _________________________
For each of the fields in which people of the household work in:
Field n° S28. Is it the field of S29. What is the S30. How much of each S31. How much of this
the household, does size of the field? product is produced from product has the household
the household rent it, each field yearly? sold in the village, outside
or do people of the (considering only non-lost of the village and
household work in it products and including consumed last month?
as contractual? products gathered before
harvest)
1 1. Own ____ha/m2/paces ______ kg/tons/bags of ___% sold in village
2. Rent /Football fields ______ ___% sold outside village
3. Contractual ___% self-consumption
______ kg/tons/bags of ___% sold in village
______ ___% sold outside village
___% self-consumption
Same questions for
other fields
Comments
Surveyor: If bags used, how much does a bag weigh? _________ kg per bag
S32. Do members of your household borrow fertilizer?
 Yes, every year
 Yes, it happened some times
 No
127
For each type of animal owned by the household:
Type of animal S33. Number of this animal S34. How many of this animal has the
household sold in the village, outside
the village and consumed last month?
1. Chicken ___ sold in village
2. Turkey ___ sold outside village
3. Hens with eggs ___ self-consumption
4. Hens without eggs
5. Rabbit
6. Goat
7. Sheep
8. Pig
9. Beef
10. Cows with milk
11. Cows without milk
12. Donkey
13. Dog
14. Other
Same questions for other types of ___ sold in village
animals ___ sold outside village
___ self-consumption
Comments:
c. Diseases and access to health

For every member of the household that was sick during the last 30 days:
N° S35. Which S36. Which S37. Is S38. For S39. What did S40. Where S41. How S42. How
member disease? symptoms? he still how long you do to heal is the person much did the much did
sick? was he him? who you consultation the drugs
sick/has visited to to heal him to heal
he been heal him? cost? him cost?
sick?
1 1. Diarrhea 1. Stomatchache 1. Go to doctor 1. Sanogho
2. Malaria 2. Liquid stool 2. Go to 2. Garango
3. Typhoid fever 3. Blood in the pharmacist 3. Tenkodogo
4. Yellow fever stool 3. Go to 4. Komina
5. Amibiase 4. Vomiting marabout 5. Other
6. Bilharziose 5. Sweating 4. Self-medicine
7. Polio 6. Headache and advice
8. Meningitis 7. Fever from
9. Hepatitis A 8. Rapid loss of family
10. Hepatitis E weight 5. Nothing
11. Dengue 9. Dehydration 6. Other:
12. Cholera 10. Convulsion
13. Onchocerciasis 11. Blindness
14. Other: 12. itching
13. Other:
Same
questions
for the
other
members
of the
household
Comments:
128
d. Time use
How much time did women of your household, who are not at school anymore, have spent doing the following
activities yesterday?
Surveyor: Get an average answer from all the women together.
Activity S43. Time spent
Work in the field
Productive activity: transformation
Cook for the household
Collect water for drinking and cooking
Wash clothes
Collect water for washing yourself + wash yourself
Take care of children
Take care of sick people and sick children
Sleep (night + nap)
Free time
Other: ____________
Other: ____________
Comment:
e. Safety
S44. Do members of your household feel safe in the village?
 Yes
 No, afraid of robbers
 No, afraid of snakes
 No, detail __________________________________________
f. Household water use
S45. What is the main problem that members of your household encounter regarding water?
______________________________________________________
129
Use S46. S47. S48. S49. S50. S51. S52. S53. S54. S55. S56.
From During What is What is At what Is the How Do Do you treat Are you Are you
which which the water the name time of water long members water? satisfied satisfied
source months quantity of the the day for this does it who with the with the
does the of the used? members do you use take to collect water water
water year do who collect collected each water use quality? quantity ?
come you go collect water for at the person water at
from? to this water for this use? same who the source
source? this use? time as collects or bring it
water water back
for for this home?
another use?
use?
(specify
the use)
Drinking Surveyor: __ 1. 6-10h 1. At the 1. Filter with 1. Yes 1. Yes
Show on Barrique/ 2. 10-14h source plastic 2. No, 2. No,
map if Buckets/ 3. 14-18h 2. Bring 2. Make it detail detail
necessary L 4. 18-22h home boil _____ _____
5. 22-6h 3. Add
chlorine
4. No
5. Other:
___
Drinking
(if
another
source is
used)
Same
questions
for other
uses:
cooking,
personal
hygiene,
laundry,
animal
drinking
Surveyor: If barrique used, how many liters does a barrique contain? _________ Liters per barrique
Surveyor: If bucket used, how many liters does a bucket contain _________ Liters per Bucket
130
Answer the questions for each of the following sources:
Water source S57. How long did it S58. Do you find it S59. Does your back S60. Does the source
(fill only for sources take you to queue at arduous to take out hurt when you extract dry up during certain
used by the members the source and extract water from the source? water from the source? moments of the year ?
of the household) water from the source
yesterday?
Open well 1 ______________ min 1. Not arduous 1. Does not hurt Yes/No
2. Arduous 2. Hurts
3. Very arduous 3. Hurts a lot If yes, during which
Detail: ________ Details: __________ months? ______
If yes, why does it dry

up?______
Open well 2
same questions for

other water sources
for domestic use: hand
pump, PVWPS
Comments
How much did you spend for the following things the last month:
S61. Money spent (XOF)
Buying water to water vendors that come to your home
Buying water at the store
Buying water at the open well 1
Same question for open well 2, hand pump 1, hand pump 2, PVWPS
Paying for the maintenance of open well 1
Same question for open well 2, hand pump 1, hand pump 2, PVWPS
Comments:
g. Income and sparings

Surveyor: The following questions are only for the head of the household. At this point, separate him from the rest of
the household members and remind him that the data collected are strictly anonymous.
Head only: How much did your household cashed in during the previous month from the following sources:
Category S62. Money earned (XOF)
Crop sales
Milk/eggs sales
Animal/meat sales
Trade, sell at shop
Wages earned
Government payments/pensions
Money received from family
Other
Comments:
S63. Head only: What is the amount of current household’s savings? Consider money from household and money
invested in other places but not the value of real estate. _______________________XOF
S64. Head only: Is some money of the household saved at a cooperative and/or at a bank?
 Bank
 Cooperative
 Bank and cooperative
 No
131
S65. Head only: Have you ever borrowed money from a cooperative or a bank?
 Bank
 Cooperative
 Bank and cooperative
 No
S66. Head only: Do you participate to a tontine8?

 Yes
 No
Surveyor: The following question is only for the women of the household who are not at school anymore. At this point,
separate them from the rest of the household members and remind them that the data collected are strictly anonymous.
S67. Women of the household who are not at school anymore: How many of you participate to a tontine? ____,
details: _____________
h. Closing the survey
S68. Date and time of the end of the survey: _______________________________________
Thank you for your participation!
8
voluntary system of group savings
132
Appendix B. Motor-pump model polynomial coefficients – case of the SQFlex 5A-7
Polynomial 𝑸𝒑 = 𝑷𝟒,𝟒
𝒂 (𝑷𝒑𝒗 , 𝑻𝑫𝑯) 𝑻𝑫𝑯 = 𝑷𝟒,𝟒
𝒃 (𝑷𝒑𝒗 , 𝑸𝒑 )
Formula 𝑚+𝑛=4 𝑚+𝑛=4

𝑚 𝑛
𝑄𝑝 = ∑ 𝑘𝑚,𝑛 𝑃𝑝𝑣 𝑇𝐷𝐻 𝑇𝐷𝐻 = ∑ 𝑙𝑚,𝑛 𝑃𝑝𝑣 𝑚 𝑄𝑝𝑛
𝑚+𝑛=0 𝑚+𝑛=0
Coefficients Index Value Unit Index Value Unit
k0,0 2.3 10-4 m3 s-1 l0,0 4.2 m
k1,0 6.5 10-6 m3 s-1 l1,0 1.5 10-1 m W-1
k0,1 -5.2 10-5 m2 s-1 l0,1 -3.4 104 m-2 s
k2,0 -7.3 10-9 m3 s-1 W-2 l2,0 -6.1 10-5 m W-2
k1,1 - 2.9 10-8 m2 s-1 W-1 l1,1 -1.4 102 m-2 s W-1
k0,2 8.3 10-8 m s-1 l0,2 5.4 107 m-5 s2
k3,0 3.6 10-12 m3 s-1 W-3 l3,0 5.2 10-8 m W-3
k2,1 1.1 10-10 m2 s-1 W-2 l2,1 -1.5 10-2 m-2 s W-2
k1,2 -2.0 10-9 m s-1 W-1 l1,2 9.4 104 m-5 s2 W-1
k0,3 2.8 10-8 s-1 l0,3 -4.1 1010 m-8 s3
k4,0 -6.7 10-16 m3 s-1 W-4 l4,0 -2.1 10-11 m W-4
k3,1 -3.9 10-14 m2 s-1 W-3 l3,1 1.8 10-5 m-2 s W-3
k2,2 1.5 10-13 m s-1 W-2 l2,2 -15 m-5 s2 W-2
k1,3 1.0 10-11 s-1 W-1 l1,3 -1.3 107 m-8 s3 W-1
k0,4 -2.6 10-10 m-1 s-1 l0,4 7.8 1012 m-11 s4
133
Appendix C. Satellite climatic data
Regarding satellite ambient temperature data, the Modern-Era Retrospective analysis for Research and
Applications, Version 2 (MERRA-2), provides the ambient temperature with a temporal resolution of
1 minute [135]. The data are available, for every location of the world, since 1980 and are regularly updated
with one month of delay.
Regarding satellite irradiance data, the Copernicus Atmosphere Monitoring Service provides the direct
normal irradiance 𝐺𝑑𝑛 , the global horizontal one 𝐺𝑔ℎ and the diffuse horizontal one 𝐺𝑑ℎ for the actual
weather conditions with a temporal resolution of 1 minute [106]. The data are available, for locations
between -66° and 66° in both latitude and longitude, since 2004 and are regularly updated with two days of
delay. The irradiance on the plane of the PV array 𝐺𝑝𝑣 , can then be deduced from satellite data by [136]:
1 − 𝑐𝑜𝑠(𝜃) 1 + 𝑐𝑜𝑠(𝜃)
𝐺𝑝𝑣 (𝑡) = 𝐺𝑑𝑛 (𝑡) 𝑐𝑜𝑠(𝐴𝑂𝐼(𝑡, 𝜃, 𝜆)) + 𝐺𝑔ℎ (𝑡) 𝜁 + 𝐺𝑑ℎ (𝑡) (39)
2 2
where 𝜁 is the albedo of the surrounding environment, 𝜃 and 𝜆 are the tilt and azimuth of the PV array
respectively, 𝐴𝑂𝐼 is the angle of incidence between the sun’s rays and the PV array. The 𝐴𝑂𝐼 is computed
by using the MATLAB library PVLIB developed by the Sandia National Laboratories [137]. For the
PVWPS of Gogma, according to the observation of the local environment a value of 0.25 is considered for
the albedo 𝜁 [91].
In Figure C.1 and Figure C.2, we compare the ambient temperature 𝑇𝑎 and the irradiance 𝐺𝑝𝑣 obtained from
satellite data and from local measurements, for the period between 19 February to 21 February 2018, The
correlation between local measurements and satellite data is fair for ambient temperature (R2 = 0.83) and
good for irradiance (R2 = 0.96).
Figure C-1 – Comparison of the ambient temperature Figure C-2 – Comparison of the irradiance on the
from local measurements and from satellite data. plane of the PV array from local measurements and
from satellite data.
134
Appendix D. Economic survey for local companies
Preliminary remark: The survey contained questions on solar pumps (i.e. PVWPS) and hand pumps. Here, we write
down only the questions on solar pumps as it is the focus of this thesis.
Preliminary remark: The layout of the original survey has been adapted here to save space. The survey has been
translated from French.
Text explaining the survey, to read completely to the surveyee
Date:
Name:
Company/Institution:
Email:
I am a student at Imperial College London and I am currently carrying a research project on the economic analysis and
life cycle cost of solar pumps and hand pumps in Burkina Faso. You have been chosen to participate to this interview
because of your knowledge on solar pumps and/or hand pumps. This survey includes oral and written questions.
Questions refer to solar pumps and hand pumps and, depending on your expertise, you may be asked only about a part
of the survey.
Are you selling:
 Hand pumps
 Solar pumps
Do you do maintenance for:
 Hand pumps
 Solar pumps
Preventive maintenance for a pumping system consists in verifying regularly that the pumping system is working
correctly and possibly performing some minor maintenance operations such as greasing some parts. This intervention
is not the consequence of a specific request of the village due to a dysfunction of the pumping system.
Do you do preventive maintenance for:
 Hand pumps
 Solar pumps
Do you do repairs for:
 Hand pumps
 Solar pumps
Remarks:
- If you do not have a precise idea of the costs, please do not answer
- If one of the question is not clear enough, please do not hesitate to ask for clarifications.
- Your answers to the questionnaire will be anonymized and only used for the research project.
a. Written survey
1. How much do you charge for the following components of a solar pump?
135
Components Price
Solar panels
- Give power
- Give brand and reference if possible
Connection cables
- Give length
- Give cross section
- Give materials
Motor-pump
- Give power, flow rate and maximum pumping height if possible
- Give brand and reference if possible
Reservoir
- Give materials
- Give volume
Borehole
- Give depth
- Give diameter
Support structure for solar panels
Pipes
- Give materials
- Give diameter
- Give length
Standpipes
2. What lifetime do you consider for the following components of a solar pump?
Components Lifespan
Solar panels
- If you draw a distinction according to the
brand, please mention it.
Connection cables
material, please mention it.
Pump
brand, please mention it.
Reservoir
Borehole
depth or diameter, please mention it.
Support structure for solar panels
Pipes
Standpipes
b. Oral survey
1. On average, how many years will a solar pump last?
2. What reasons can lead to the closing or abandonment of a solar pump?
3. Do you offer a guarantee for the solar pumps installed?

a. If yes, for how long is it effective?
b. If yes, what does it cover?
136
4. According to you, are the prices of solar systems components:
 Very low
 Low
 Normal
 High
 Very high
Comments:
5. How much do you charge for the installation of a solar pump with a reservoir (transport and workforce included,
not considering the cost of the components)?
Distance between the solar pump and the office of your company Cost
50 km
200 km
400 km
b.1. Preventive maintenance

1. In general, who is responsible for the preventive maintenance of solar pumps?
2. In general, who pays for the preventive maintenance of solar pumps?
3. Which preventive maintenance scheme do you usually use for solar pumps?
 A package
 A single payment
If package :
o Is it:
 A monthly fee
 An annual fee
 A multiyear fee. Specify the number of years:
o How much do you charge for it?

50 km
200 km
400 km
If single payment for each intervention:

50 km
200 km
400 km
4. What is, on average, the frequency of intervention for preventive maintenance on a solar pump?
137
5. According to you, are the costs of preventive maintenance interventions (transport included, without the cost of
the components) for solar pumps:
 Very low
 Low
 Normal
 High
 Very high
Comments:
b.2. Repairs
1. In general, who is responsible for the repairs of solar pumps?
2. In general, who pays for the repairs of solar pumps?
3. Which repairs scheme do you usually use for solar pumps?

 A package
 A single payment
If package:
o Is it:
 A monthly fee
 An annual fee
 A multiyear fee. Specify the number of years:

50 km
200 km
400 km
If single payment for each intervention:

50 km
200 km
400 km
4. What is, on average, the frequency of intervention for repairs on a solar pump?
5. According to you, are the costs of repairs (transport included, without the cost of the components) for solar pumps:
 Very low
 Low
 Normal
 High
 Very high
Comments:
Thank you for your participation!
138
Appendix E. Range of variation of groundwater parameters
There are 3 groundwater parameters: the height between the ground level and the static water level in the
borehole 𝐻𝑏,𝑠 , the aquifer losses coefficient 𝜅0 and the borehole losses coefficient 𝜇0 . In this section, we
present how to determine the range of variation of these parameters for a given location in Africa, without
drilling a borehole and performing pumping tests.
Regarding 𝐻𝑏,𝑠 , MacDonald et al. [9] provided a map with the range of variation of 𝐻𝑏,𝑠 for all locations of
Africa (see Figure E.1). We added the location of Gogma on this map. Gogma is at the limit between the
category “0 to -7 m” and the category “-7 m to -25 m”. Thus, we consider that 𝐻𝑏,𝑠 is comprised between 0
and -25 m in Gogma.
Figure E-1 – Range of variation of 𝐻𝑏,𝑠 in Africa [9].
Regarding 𝜅0 and 𝜇0 , no map is available. However, we encountered values of these parameters in studies
performed worldwide [78, 93, 138, 139]: 72 values were found for 𝜅0 and 𝜇0 . For 𝜅0 , the minimum and
maximum values encountered are 0.5 m-2 s and 2.3 103 m-2 s, respectively. For 𝜇0 , the minimum and
maximum values encountered are 13 m-5 s2 and 6.2 105 m-5 s2, respectively. From this and after discussions
with hydrologist Prof. Peter K. Kitanidis, we set the following ranges of variation for 𝜅0 and 𝜇0 :
 𝜅0 ∈ [0, 6.9 103] m-2 s

 𝜇0 ∈ [0, 1.9 106] m-5 s2
139
Extended summary in French
140
Extended summary in French
Chapitre I : Revue de littérature

En Afrique subsaharienne, les alternatives les plus communes pour l’accès à l’eau pour l’usage domestique
dans les zones pauvres et non connectées au réseau électrique sont les puits ouverts, desquels l’eau est
extraite avec un seau et une corde, et les pompes à main. L’eau des puits ouverts n’est pas potable car ils
sont exposés à la contamination extérieure. En revanche, l’eau des pompes à main est potable. Cependant,
pour les pompes à main, tout comme pour les puits ouverts, l’extraction d’eau est difficile, le débit
d’extraction est limité par la force humaine et ces pompes ne permettent pas d’atteindre les ressources en
eau profondes (> 50 m). Les technologies de pompage d’eau électrifiées permettent de remédier à ces
difficultés. Les deux principales sources d’énergie pour l’électrification des pompes sont l’énergie
photovoltaïque et le diesel. Les systèmes de pompage photovoltaïques (PVWPS) sont compétitifs
économiquement avec les pompes diesel et ont une durée de vie plus longue, des besoins en maintenance
plus réduits et n’émettent pas de fumées toxiques. Nous avons de plus mis en évidence que les émissions de
gaz à effet de serre des PVWPS pour l’accès à l’eau domestique sont faibles, notamment car un PVWPS de
faible puissance (~1 kW) fournit un service à haute valeur ajoutée (i.e. fournir de l’eau à des dizaines de
ménages). Cela explique pourquoi nous ne nous intéresserons pas aux émissions de gaz à effet de serre des
PVWPS dans cette thèse.
La conception d’un PVWPS pour un village rural consiste à déterminer son architecture, sa position dans le
village et son dimensionnement. Conventionnellement, ces éléments sont déterminés de la manière
suivante :
 Architecture : le stockage grâce à un réservoir d’eau est privilégié par rapport aux batteries, pour
des raisons de fiabilité qui sont cruciales dans ces zones rurales.
 Position : le décideur propose une position dans le village en suivant quelques lignes directrices.
 Dimensionnement : les articles scientifiques actuels visent à déterminer le dimensionnement du
système qui minimise le coût sur cycle de vie et qui maximise la satisfaction de la demande en eau
des habitants.
Conformément à la littérature, l’architecture avec le réservoir d’eau nous paraît aussi la plus adaptée et nous
considérons donc cette architecture dans cette thèse. Cependant, concernant la position et le
dimensionnement, nous avons identifié les manquements suivants de l’approche conventionnelle :
 Le positionnement est un sujet très peu abordé dans la littérature et aucun outil de support n’est
fourni au décideur pour le positionnement du PVWPS.
 La position et le dimensionnement ne sont pas couplés, c’est-à-dire que les travaux sur le
dimensionnement des PVWPS ne prennent pas en compte le fait que la demande en eau au PVWPS
dépend de sa position dans le village. En effet, un grande partie des villages d’Afrique subsaharienne
sont très étendus (plusieurs km2) et les conditions d’accès à l’eau ne sont pas les mêmes dans toutes
les zones du village. Ainsi, l’affluence à un nouveau point d’eau (ici le PVWPS) dépendra de sa
position dans le village.
141
 L’approche conventionnelle pour la conception des PVWPS ne considère pas l’impact sur le
développement socio-économique (e.g. utilisation d’une eau de meilleure qualité, diminution de la
distance à parcourir pour la collecte d’eau), alors que la maximisation de cet impact positif est
l’objectif principal des institutions qui financent les PVWPS. Cela empêche de plus de cibler les
habitants du village qui ont l’accès à l’eau le plus défavorable.
L’objectif de cette thèse est donc de développer une méthodologie de conception optimale des PVWPS qui
permette de déterminer les dimensionnements du PVWPS et ses positions dans le village qui maximisent
l’impact positif sur le développement socio-économique et minimisent le coût sur cycle de vie du PVWPS.
Cet objectif est atteint en trois étapes principales:
 Chapitre II. Un PVWPS est conçu et installé de manière conventionnelle dans un village rural
d’Afrique subsaharienne. D’une part, cela permet de comprendre en détail la situation actuelle
concernant les PVWPS et de mettre en place une méthodologie qui se base sur cette situation.
D’autre part, cela permet de collecter des données pour appliquer la méthodologie développée et
pour valider les modèles proposés.
 Chapitre III. Nous construisons un modèle interdisciplinaire liant le dimensionnement et la
position du PVWPS à son impact socio-économique et à son coût du cycle de vie. Le modèle
interdisciplinaire est composé de 4 sous-modèles : demande, technique, impact et économique.
 Chapitre IV. Nous définissons un problème d'optimisation pour déterminer les dimensionnements
et les positions du PVWPS qui maximisent l'impact positif sur le développement socio-économique
et minimisent le coût du cycle de vie du PVWPS puis nous présentons les résultats.
Pour résumer, dans le chapitre II, nous apprenons de la conception et de l’installation d’un PVWPS de
manière conventionnelle. Ensuite, dans les chapitres III et IV, nous utilisons les connaissances acquises pour
proposer une méthodologie améliorée de conception et d’installation des PVWPS (i.e. nous déterminons
comment nous aurions pu concevoir et installer le PVWPS mis en place conventionnellement de manière
plus optimale).
Ce travail a été effectué au sein d’une équipe interdisciplinaire composée de chercheurs en génie électrique
(laboratoires GeePs et SATIE), politique environnementale (Imperial College London), économétrie
(Colorado State University) et hydrologie (Stanford University) en collaboration avec l’entreprise
Burkinabé DargaTech, spécialisée dans les systèmes d’énergie solaire.
Chapitre II : Dispositif expérimental

Village d’étude
Le village de Gogma (latitude 11.73°, longitude -0.58°) compte 1100 habitants répartis dans 125 ménages.
Les ménages sont eux-mêmes regroupés en 41 concessions. La grande majorité des habitants travaillent
dans le domaine de l’agriculture et vivent avec un revenu inférieur à 1 $/jour. Les ménages n’ont pas accès
à l’électricité. Nous avons identifié 4 types d’usages domestiques de l’eau : boire, cuisiner, se laver et laver
les vêtements. L’eau pour ces usages est collectée auprès de 22 points d’eau répartis en 3 catégories : 16
puits ouverts desquels l’eau est extraite avec un seau et une corde, 5 pompes à main et 1 PVWPS.
142
Conception, installation et description du PVWPS
Nous avons suivi les différentes étapes de la conception et de l’installation du PVWPS qui a été coordonnée
par l’entreprise Burkinabé DargaTech. L’analyse de ces étapes et les discussions avec les membres de
DargaTech nous ont permis de mettre en évidence la procédure conventionnelle d’installation des PVWPS
au Burkina Faso. Elle est détaillée sur la figure 1. Cette procédure a duré 5 mois à Gogma entre fin 2017 et
début 2018 et le PVWPS a été ouvert à la consommation en janvier 2018.
143
1
Identifier les zones à éviter (e.g. cimetières, zones non sûres)
2
Le décideur propose une position ( ), en dehors des zones à éviter, autour de
laquelle il faut chercher de l’eau. 2b
Le décideur propose une
3 autre position ( , ),
Effectuer des mesures géophysiques le long de plusieurs profils dans un carré de
qui est à l’extérieur du carré
~350 350 m autour de ( ).
de ~350 350 m autour de
( ) et à l’extérieur
des zones à éviter.
Est-ce que les mesures géophysiques indiquent
Non
une position ( appropriée pour forer
dans le carré ?
Oui
4
Effectuer le forage à ( ).
Est-ce qu’il y a de l’eau qui sort du forage ? Non
Oui
5
Effectuer les tests de pompage.
Déterminer le débit maximum qui peut être pompé grâce aux tests de pompage.
Effectuer des test physico-chimiques sur l’eau qui sort du forage.
Est-ce que la qualité physico-chimique de l’eau est

Non
conforme à la régulation pour l’usage domestique ?
Oui
Est-ce que le décideur considère que cela vaut la peine

d’installer un PVWPS à la position ( )? Non
Oui
6
Dimensionner le PVWPS: déterminer la puissance crête totale des modules PV, la
référence de motopompe adaptée et le volume du réservoir.
7
Installer les modules PV, la motopompe, le réservoir et la fontaine.
8
Ouvrir le PVWPS à la consommation pour les habitants.
9
Effectuer des tests bactériologiques sur l’eau qui sort de la fontaine.
Est-ce que le décideur considère que la qualité Non

bactériologique de l’eau est satisfaisante ?
Oui
10 Étapes de positionnement
Le PVWPS est maintenu ouvert Étape de dimensionnement
figure 1 – Procédure conventionnelle pour la conception et l’installation de PVWPS.
La figure 2 présente l’architecture du PVWPS installé et la figure 3 montre le système. Une vidéo du village
et du PVWPS est aussi disponible au lien suivant : https://youtu.be/VrjM0edKVsI. Ce système comprend
144
des modules PV polycristallins pour une puissance crête totale de 620 Wc, une motopompe Grundfos SQFlex
5A-7 et un réservoir d’eau cylindrique en acier de 11.4 m3. Le contrôleur régule l’énergie fournie par les
modules PV à la motopompe, selon le niveau d'eau dans le réservoir, obtenu par un interrupteur à flotteur.
L'eau est collectée à la fontaine par les habitants.
Réservoir
Tête de Modules
forage PV Fontaine
figure 2 – Architecture du PVWPS. figure 3 – Photo du PVWPS.
Collecte de données
Voici les données principales qui ont été collectées à Gogma :
 Les coordonnées GPS des ménages, des sources d’eau et des points importants du village ont été
relevées et sont représentées sur la photo satellite de la figure 4.
 Des mesures géophysiques ont été effectuées, avant la mise en place du PVWPS actuel, visant à
détecter l’éventuelle présence d’eau (voir figure 4).
 Nous avons déterminé le coût pour collecter de l’eau à chaque source grâce aux carnets de comptes
des sources.
 Des enquêtes ménage socio-économiques ont été effectuées, avant et après l’ouverture à la
consommation du PVWPS, auprès de 88 ménages tirés au sort parmi les 125 ménages du village.
Ces enquêtes ont notamment permis de connaître le choix de source de chaque ménage avant et
après l’installation du PVWPS, ainsi que de savoir comment les ménages perçoivent la qualité de
l’eau et la facilité d’extraction aux sources qu’ils utilisent. Les données recueillies pour les 88
ménages intérrogés sont extrapolées aux 37 ménages du village qui n’ont pas été interrogés,
notamment grâce à la connaissance de la position GPS de ces 37 ménages.
 Des tests de pompage ont été réalisés pour le PVWPS actuel. Ils ont permis de déterminer le débit
maximum 𝑄𝑝,𝑚𝑎𝑥 qui peut être extrait du forage sans mettre en péril les ressources en eau.
 Des analyses bactériologiques ont été effectuées pour toutes les sources d’eau afin de quantifier leur
qualité.
 Le PVWPS est monitoré en continu depuis janvier 2018 grâce à un système de collection de données
autonome que nous avons nous-mêmes conçu et installé. Les grandeurs collectées sont l’irradiance,
la température ambiante, la tension et le courant des modules PV, le débit pompé, le débit collecté
à la fontaine et le niveau d’eau dans le forage. Les données sont collectées avec un pas de temps de
145
2 s. À notre connaissance, c’est la première fois qu’un PVWPS pour l’accès à l’eau domestique est
monitoré et que des données ont été collectées sur un PVWPS en Afrique subsaharienne rurale.
figure 4 – Cartographie GPS du village de Gogma.
Chapitre III : Modèle interdisciplinaire

Le modèle interdisciplinaire lie le dimensionnement et la position du PVWPS à son impact socio-
économique et à son coût du cycle de vie. Le synoptique du modèle est représenté sur la figure 5.
146
Modèle interdisciplinaire
Coût
Dimensionnement Modèle
sur cycle de vie
Economique
Données
climatiques
Paramètres Modèle Consommation

hydrodynamiques Technique d’eau
, , Impact
Modèle
socio-économique
Impact
Position
Modèle Demande
Demande en eau
Variables
Choix de source des
d’optimisation
ménages avant
Comportement des
installation du
ménages
PVWPS
Fonctions objectifs
figure 5 – Synoptique du modèle interdisciplinaire.
Entrées
Les entrées du modèle interdisciplinaire peuvent être regroupées en 4 catégories :
1. Variables d’optimisation. Nous distinguons deux types de variables d'optimisation: (1) les variables
de dimensionnement du PVWPS et (2) la position du PVWPS dans le village, qui est donnée par sa
longitude 𝐿𝑜𝑛 et sa latitude 𝐿𝑎𝑡. Les variables de dimensionnement incluent la puissance crête totale
des modules PV dans les conditions de test standard (STC) 𝑃𝑝𝑣,𝑝 , la référence de la motopompe 𝑀𝑃
et le volume du réservoir d'eau 𝑉𝑡 . Nous avons numérisé les courbes caractéristiques de 8 références
de motopompes.
2. Données climatiques. Ce sont les données d’irradiance et de température ambiante.
3. Paramètres hydrodynamiques: le niveau d’eau statique dans le forage 𝐻𝑏,𝑠 (i.e. le niveau d’eau
lorsqu’il n’y a pas de pompage), le coefficient de pertes dues à l’aquifère 𝜅0 et le coefficient de
pertes dues au forage 𝜇0 . Ces paramètres hydrodynamiques dépendent des ressources en eau. Ils
sont déterminés après avoir fait un forage puis des tests de pompage. Par exemple, ils ont été
déterminés pour le PVWPS actuel de Gogma grâce aux tests de pompage effectués (voir Chapitre
II). Sans faire de forage et de tests de pompage, il est seulement possible de donner une plage de
variation de ces paramètres (voir Appendix E).
4. Le choix de source des ménages avant installation du PVWPS.
Sorties
Les sorties du modèle interdisciplinaire sont le coût du PVWPS sur cycle de vie 𝐿𝐶𝐶 et son impact socio-
économique 𝑆𝐸𝐼.
Sous-modèles
Le modèle de demande est un modèle d’économétrie qui permet de prévoir le choix de source d’eau effectué
par chaque ménage à partir:
147
 des distances entre le ménage et les différentes sources d’eau ;
 de la perception du ménage de la qualité de l’eau des différentes sources.
Nous pouvons donc, pour chaque position du PVWPS, prévoir quels ménages quitteraient leur ancienne
source d’eau au profit du PVWPS. Cela permet d’obtenir la demande en eau au PVWPS. Les paramètres de
ce modèle sont identifiés grâce au choix de source des ménages et à leur perception et distance aux sources
avant installation du PVWPS (données collectées grâce aux enquêtes ménage).
Le modèle technique permet de déterminer le pourcentage de la demande en eau au PVWPS qui est satisfait,
c’est-à-dire la consommation d’eau, et d’identifier les ménages qui bénéficient effectivement du PVWPS.
Les entrées du modèle sont les données climatiques, les paramètres hydrodynamiques et le
dimensionnement du PVWPS. Ce modèle prend en compte les différentes étapes de la conversion d’énergie
au sein du PVWPS et la réponse de la nappe phréatique au pompage d’eau. Nous avons validé ce modèle
pour le système actuel de Gogma grâce aux données acquises par le système de collection de données. Nous
évaluons ce modèle pour une année avec un pas de temps de 1h. Grâce aux modèles de demande et
technique, nous pouvons donc prédire quels ménages vont effectivement quitter leur ancienne source d’eau
pour le PVWPS.
Le modèle d’impact permet alors d’évaluer l’impact socio-économique 𝑆𝐸𝐼 associé à ces changements de
sources d’eau. L’impact socio-économique est évalué grâce à des indicateurs. Nous avons séparé ces
indicateurs en deux catégories :
 Indicateurs directs. Ils résultent directement du changement de source d’eau. Ils sont au nombre de
4 : la qualité de l’eau utilisée, la facilité pour extraire l’eau, le prix à payer pour utiliser l’eau et la
distance à parcourir pour collecter l’eau.
 Indicateurs indirects. Ils résultent des indicateurs directs. Ces indicateurs considèrent notamment la
prévalence des maladies hydriques (diarrhée, trachome), les gains de temps, les dépenses de santé
et l’accès à l’éducation.
Nous quantifions tout d’abord l’effet des changements de sources d’eau sur les indicateurs directs, puis
l’effet de la variation des indicateurs directs sur la valeur des indicateurs indirects. L’impact socio-
économique est égal à la somme pondérée des valeurs des indicateurs directs et indirects, normalisées au
préalable. Les coefficients de pondération dépendent du choix du décideur politique, selon les indicateurs
qu’il veut favoriser.
Le modèle économique permet de déterminer le coût du PVWPS sur cycle de vie 𝐿𝐶𝐶 à partir de son
dimensionnement (valeurs de 𝑃𝑝𝑣,𝑝 , 𝑀𝑃 et 𝑉𝑡 ). Le modèle prend en compte les coûts d’investissement et de
fonctionnement et l’actualisation de la monnaie. Les données utilisées pour ce modèle ont été collectées
auprès d’entreprises Burkinabès.
Chapitre IV : Conception optimale

Problème d’optimisation
Les 2 fonctions objectifs de l’optimisation sont le coût sur cycle de vie 𝐿𝐶𝐶 du PVWPS, que nous voulons
minimiser, et l’impact socio-économique 𝑆𝐸𝐼, que nous voulons maximiser. Les variables de l’optimisation
148
sont la puissance de crête totale des modules PV 𝑃𝑝𝑣,𝑝 , la référence de motopompe 𝑀𝑃, le volume du
réservoir d'eau 𝑉𝑡 , la latitude 𝐿𝑎𝑡 et la longitude 𝐿𝑜𝑛 du PVWPS. La référence de motopompe 𝑀𝑃 est la
seule variable discrète. Nous effectuons une optimisation pour chaque référence de motopompe 𝑀𝑃 et nous
obtenons donc un front de Pareto pour chaque référence de motopompe. Nous déterminons alors le front de
Pareto final en passant par les meilleurs points des fronts de Pareto associés aux références de motopompes.
Nous ne mettons pas de contrainte sur la satisfaction de la demande en eau et nous considérons donc aussi
les systèmes sous-dimensionnés, pour lesquels la consommation d’eau est inférieure à la demande en eau.
Nous utilisons un algorithme d’évolution différentielle bi-objectif, qui est un algorithme stochastique pour
l’optimisation. Nous avons choisi un algorithme stochastique car le problème d’optimisation est non-
linéaire.
Analyse d’un résultat de référence
Nous présentons ici un exemple de résultat d’optimisation pour lequel l’impact socio-économique considère
les 4 indicateurs directs avec des poids tous identiques. Nous supposons de plus que, pour toutes les positions
de Gogma, les valeurs des paramètres hydrodynamiques sont les mêmes que les valeurs déterminées pour
le PVWPS actuel. La figure 6 montre le front Pareto obtenu.
figure 6 – Front de Pareto.
La figure 7, la figure 8 et la figure 9 illustrent l’évolution des variables d’optimisation le long du front de
Pareto. Nous présentons aussi sur ces figures les valeurs des fonctions objectifs et des variables pour le
système actuel de Gogma.
149
PVWPS actuel
figure 7 – Variation de 𝑃𝑝𝑣,𝑝 et 𝑉𝑡 en fonction de 𝑆𝐸𝐼.
figure 8 – Variation de 𝑀𝑃 en fonction de 𝑆𝐸𝐼.
PVWPS actuel
figure 9 – Variation de 𝐿𝑎𝑡 et 𝐿𝑜𝑛 en fonction de 𝑆𝐸𝐼.
150
La figure 6 permet de déterminer (1) l’impact socio-économique 𝑆𝐸𝐼 maximum envisageable pour un
PVWPS de coût 𝐿𝐶𝐶 et (2) la coût minimal 𝐿𝐶𝐶 pour atteindre un impact socio-économique 𝑆𝐸𝐼 donné. Il
apparaît logiquement que des dépenses plus importantes conduisent à des impacts socio-économiques
positifs plus significatifs. Les résultats indiquent aussi que la différence de 𝐿𝐶𝐶 entre le point de minimum
𝑆𝐸𝐼 et le point de maximum 𝑆𝐸𝐼 est 7.9 103 $, ce qui représente 29% du coût 𝐿𝐶𝐶 du point de minimum
𝑆𝐸𝐼 (2.72 104 $). Cela suggère que, dans la mesure du possible, il est préférable de choisir des points du
front avec des valeurs de 𝑆𝐸𝐼 élevées.
Nous observons de plus sur la figure 7 que les valeurs optimales de la puissance crête totale des modules
PV 𝑃𝑝𝑣,𝑝 et du volume du réservoir 𝑉𝑡 augmentent avec l’impact socio-économique 𝑆𝐸𝐼. En effet, de manière
générale, des systèmes de plus grande taille permettent d’accroître le nombre de bénéficiaires et donc
l’impact socio-économique. Concernant le choix de motopompe (figure 8), l’optimisation privilégie les
références 5A-7, 2.5-2, 1.2-2 et 0.6-2. Enfin, nous observons une faible variation de la position du PVWPS
le long du front de Pareto (figure 9). Cela permet d’identifier une zone du village (𝐿𝑎𝑡 ∈ [11.723°, 11.725°]
× 𝐿𝑜𝑛 ∈ [-0.573°, -0.571°]) où l’installation du PVWPS serait optimale. Nous avons représenté cette zone
par un rectangle vert sur la figure 4. Nous observons sur la figure 4, que la densité de ménages est
relativement élevée dans cette zone et que les seules sources d’eau disponibles sont des puits ouverts, qui
sont des sources de mauvaise qualité (eau non potable) et pour lesquelles l’extraction d’eau est très
fastidieuse. Ainsi, dans le cas où un PVWPS est installé dans cette zone, les ménages proches du PVWPS
peuvent alors quitter leur puits ouvert pour le PVWPS. Cela conduit à un fort impact socio-économique et
explique notamment le choix de cette zone par l’optimisation.
Nous pouvons aussi comparer les valeurs des fonctions objectifs et des variables pour le système actuel aux
résultats optimaux. Sur la figure 6, nous observons que le coût du PVWPS actuel (3.55 104 $) aurait pu
permettre d’atteindre l’impact socio-économique 𝑆𝐸𝐼 le plus élevé (0.105), à la place du 𝑆𝐸𝐼 du PVWPS
actuel (0.068). En outre, il apparaît que l’impact socio-économique du système actuel (0.11) a été obtenu
pour un coût 𝐿𝐶𝐶 de 3.55 104 $ alors que, d’après le front de Pareto, il aurait pu être obtenu pour un coût de
seulement ~2.78 104 $. Cela suggère que l’application de la méthodologie aurait pu permettre d’économiser
~7.7 103 $ à Gogma, pour l’expression du 𝑆𝐸𝐼 considérée ici. Pour le 𝑆𝐸𝐼 du PVWPS actuel (0.068), la
figure 7 indique que la puissance crête totale des modules PV 𝑃𝑝𝑣,𝑝 actuelle est proche de la puissance
optimale mais que le volume du réservoir 𝑉𝑡 actuel est bien plus large que le volume optimal. La figure 8
suggère que la motopompe SQFlex 2.5-2 serait plus adaptée que la motopompe actuelle (SQFlex 5A-7). En
effet, pour le point du front de Pareto le plus proche (𝑆𝐸𝐼 = 0.070), la motopompe optimale est la SQFlex
2.5-2. Enfin, la figure 9 montre que la position actuelle du PVWPS est proche de la position optimale.
Nous avons en outre évalué l’influence de l’erreur en sortie du modèle de demande, de la définition de
l’impact socio-économique par le décideur politique et des ressources en eau sur les résultats d’optimisation.
Proposition de procédure améliorée pour la conception et l’installation des PVWPS
La procédure améliorée proposée est présentée sur la figure 10. Elle est basée sur la procédure
conventionnelle (voir figure 1) et inclut la méthodologie d’optimisation développée.
151
1
Collecter les coordonnées GPS de tous les ménages, sources d’eau et zones à éviter (e.g.
cimetières, zones non sûres). Collecter le coût de l’eau aux sources.
Enregistrer le choix de source d’au moins 70 % des ménages et leurs perceptions de la
qualité de l’eau et de la facilité d’extraction aux sources qu’ils utilisent.
2
Le décideur fixe les poids entre les indicateurs de la fonction .
Fixer la contrainte que le PVWPS ne peut pas être installé dans les zones à éviter
Fixer les paramètres hydrodynamiques ( , ) à la borne supérieure de leur intervalle
de variation pour toutes les positions ( , ) dans le village. Effectuer l’optimisation.
3
Le décideur sélectionne un design optimal.
La position correspondante est notée ( ). 2b
Ajouter la contrainte que le
4 PVWPS ne peut pas être installé
Effectuer des mesures géophysiques le long de plusieurs profils dans un carré de dans un carré de ~350 350 m
~350 350 m autour de ( ). autour de ( ). Effectuer
l’optimisation pour rechercher
une position ( , ).
Est-ce que les mesures géophysiques indiquent
Non
une position ( appropriée pour forer
dans le carré ?
Oui
5
Effectuer le forage à ( ).
Est-ce qu’il y a de l’eau qui sort du forage ? Non
Oui
6
Effectuer les tests de pompage. Déterminer les paramètres hydrodynamiques ( , , )
et le débit maximum qui peut être pompé grâce aux tests de pompage. Effectuer des
tests physico-chimiques et bactériologiques sur l’eau qui sort du forage.
Est-ce que la qualité physico-chimique de l’eau est

conforme à la régulation pour l’usage domestique ? Non
Est-ce que le décideur considère que la qualité
Oui
7
Effectuer l’optimisation avec les valeurs des paramètres hydrodynamiques déterminées et
la position du PVWPS fixée à ( ). Mettre la contrainte que le débit pompé doit
rester inférieur à .
Est-ce que le décideur considère que cela vaut la peine

Non
d’installer un PVWPS à la position ( )?
Oui
8
Le décideur sélection un dimensionnement optimal (i.e. puissance crête totale des modules
PV , référence de motopompe et volume du réservoir .
9
Installer les modules PV, la motopompe, le réservoir et la fontaine.
10
Effectuer des tests bactériologiques sur l’eau qui sort de la fontaine.
Est-ce que le décideur considère que la qualité

Non
Changements par rapport à la
Oui procédure conventionnelle
11 Étape de positionnement
Ouvrir le PVWPS à la consommation pour les habitants. Étape de dimensionnement
figure 10 – Procédure améliorée pour la conception et l’installation des PVWPS.
152
Nous présentons ici les différences entre la procédure conventionnelle (voir figure 1) et la procédure
améliorée proposée (voir figure 10).
Pour évaluer les modèles de demande et d’impact, il est nécessaire de collecter des données dans le village.
C’est l’objet de l’étape 1. Les coordonnées GPS de tous les ménages et sources doivent être recueillies ainsi
que le coût de l’eau à toutes les sources. Il faut aussi recueillir le choix de source d’au moins 70% des
ménages du village ainsi que leur perception de la qualité de l’eau et de la facilité d’extraction aux sources
qu'ils utilisent. Nous estimons que cette collecte de données coûte environ ~800 $ pour un village comme
Gogma. Lorsque nous comparons ce coût avec les économies potentielles liées à l’application de la
méthodologie développée à Gogma (~7.7 103 $), cela suggère que l’application de la procédure améliorée
est économiquement viable.
Une première optimisation bi-objectif est effectuée aux étapes 2 et 3, ce qui permet de proposer une première
position (𝐿𝑎𝑡1 , 𝐿𝑜𝑛1 ), autour de laquelle les ressources en eau seront examinées. Pour cette optimisation,
les poids de la fonction 𝑆𝐸𝐼 sont définis par le décideur en fonction des indicateurs qu’il souhaite privilégier.
Enfin, les paramètres hydrodynamiques sont pris égaux à la limite supérieure de leur intervalle de variation
pour toutes les positions (𝐿𝑎𝑡, 𝐿𝑜𝑛) du village, ce qui aide à éviter des coûts plus élevés que prévu lors du
forage et du dimensionnement du PVWPS. Grâce à cette première phase d’optimisation, le décideur peut
désormais compter sur un outil lui permettant d’identifier une position potentielle (𝐿𝑎𝑡1 , 𝐿𝑜𝑛1) autour de
laquelle installer le PVWPS. Cette position maximise l'impact socio-économique positif 𝑆𝐸𝐼 sur le village
tout en minimisant le coût sur cycle de vie du système 𝐿𝐶𝐶. Nous rappelons que dans la situation classique,
le décideur ne pouvait compter que sur son intuition pour cette étape de positionnement.
Au cours des étapes 4 à 6, le forage est effectué, la qualité de l'eau est testée et les paramètres
hydrodynamiques (𝐻𝑏,𝑠 , 𝜅0 et 𝜇0 ) et le débit maximal pouvant être pompé 𝑄𝑝,𝑚𝑎𝑥 sont déterminés grâce
aux essais de pompage.
Une seconde optimisation bi-objectif est ensuite effectuée aux étapes 7 et 8, ce qui permet de déterminer les
valeurs des variables de dimensionnement. Pour cette optimisation, la position est fixée à la position du
forage et la contrainte que le débit de la pompe doit rester inférieur à 𝑄𝑝,𝑚𝑎𝑥 est définie, afin de préserver
les ressources en eau. Il est important de noter que la détermination des paramètres hydrodynamiques (𝐻𝑏,𝑠 ,
𝜅0 et 𝜇0 ) est essentielle pour garantir le respect de la contrainte mentionnée précédemment car ces
paramètres sont nécessaires pour calculer le débit pompé. Dans l’ensemble, cette deuxième phase
d’optimisation aide la société qui installe le PVWPS et le décideur à déterminer le dimensionnement du
PVWPS qui maximise l’impact socio-économique positif 𝑆𝐸𝐼 et minimise le coût du cycle de vie 𝐿𝐶𝐶, tout
en préservant les ressources en eau.
Enfin, nous proposons également quelques modifications concernant les tests bactériologiques.
Premièrement, nous proposons d’ajouter des tests bactériologiques en même temps que les tests physico-
chimiques, à l’étape 6. En effet, les tests bactériologiques ne coûtent que ~20 $ (voir section II.2.7.2) et
fournissent des informations sur la qualité bactériologique de l’eau du forage. Si le décideur n'est pas satisfait
par la qualité de l’eau du forage, il peut alors décider de modifier la position du PVWPS avant d’effectuer
153
de nouveaux investissements. Deuxièmement, nous recommandons que les seconds tests bactériologiques
soient toujours effectués avant l'ouverture du PVWPS à la consommation (voir étapes 10 et 11).
154
References
155
References
[1] F. Carlevaro, C. Gonzalez, Costing improved water supply systems for low-income
communities: A practical manual, 1st ed, IWA Publishing, 2015.
[2] WaterAid, “Handpumps Technical Brief,” 2013. [Online]. Available:
https://washmatters.wateraid.org/sites/g/files/jkxoof256/files/technical-brief-
handpumps_1.pdf [accessed 10 Oct 2018].
[3] G. F. White, D. J. Bradley, A. U. White, “Drawers of water: Domestic water use in East
Africa,” Bulletin World Health Organization, vol. 80, no. 1, pp. 63-73, 1972.
[4] J. Ntouda, F. Sikodf, M. Ibrahim, I. Abba, “Access to drinking water and health of
populations in Sub-Saharan Africa,” Comptes Rendus Biologies, vol. 336, no. 5–6, pp.
305–309, 2013.
[5] H. Ritchie, M. Roser, “Water Use and Sanitation,” 2015. [Online]. Available:
https://ourworldindata.org/water-use-sanitation [accessed 20 Oct 2018].
[6] Agence Française de Développement (AFD) and Programme Solidarité Eau, “Réalisation
et gestion des forages équipés d’une pompe à motricité humaine en Afrique
subsaharienne”,
2011. [Online]. Available:
https://www.pseau.org/outils/ouvrages/afd_realisation_et_gestion_des_forages_equipes_
d_une_pompe_a_motricite_humaine_en_afrique_subsaharienne_2011.pdf [accessed 10
Oct 2018].
[7] J. Bartram, S. Cairncross, “Hygiene, sanitation, and water: forgotten foundations of
Health”, PLoS Med, vol. 7, no. 11, 2010.
[8] World health organization (WHO) and United Nations Children's Fund (UNICEF),
“Progress on Sanitation and Drinking Water - Joint Monitoring Programme on Water
Supply and Sanitation,” 2010, [Online]. Available:
http://www.who.int/water_sanitation_health/monitoring/9789241563956/en/ [accessed 10
Oct 2018].
[9] A. M. MacDonald, H. C. Bonsor, B. É. Ó. Dochartaigh, R. G. Taylor, “Quantitative maps
of groundwater resources in Africa”, Environmental Research Letters, vol. 7, no. 2, 2012
[10] S. Cairncross, V. Valdmanis, Disease Control Priorities in Developing Countries – Chap.
41 Water supply, sanitation and hygiene promotion, 2nd ed, New York: Oxford University
Press, 2006.
[11] J. Smet, C. van Wijk, “Small Community Water Supplies,” Technical Paper Series 40 -
IRC International Water and Sanitation Centre, 2002.
[12] J. M. Puerta, “Study on the Performance and Sustainability of Water and Sanitation
Initiatives in Rural Areas,” Inter-American development bank, 2016. [Online]. Available:
https://publications.iadb.org/en/study-performance-and-sustainability-water-and-
sanitation-initiatives-rural-areas-drinking-water [accessed 10 Oct 2018].
[13] N. H. Ravindranath, H. I. Somashekar, S. Dasappa, C. J. Reddy, “Sustainable biomass
power for rural India: Case study of biomass gasifier for village electrification,” Current
Science, vol. 87, no. 7, pp. 932–941, 2004.
[14] J. P. Murenzi, T. S. Ustun, “The case for microgrids in electrifying Sub-Saharan Africa,”
in 6th International Renewable Energy Congress (IREC), Sousse, Tunisia, 2015.
[15] H. Ahlborg, L. Hammar, “Drivers and barriers to rural electrification in Tanzania and
Mozambique–Grid-extension, off-grid, and renewable energy technologies,” Renewable
Energy, vol. 61, pp. 117–124, 2014.
156
[16] A. K. Akella, R. P. Saini, M. P. Sharma, “Social, economical and environmental impacts
of renewable energy systems,” Renewable Energy, vol. 34, no. 2, pp. 390–396, 2009.
[17] World Bank, “Access to electricity, rural (% of rural population),” 2016. [Online].
Available: https://data.worldbank.org/indicator/EG.ELC.ACCS.RU.ZS [accessed 17 Dec
2018].
[18] E. André de La Fresnaye, “A financial and technical assessment of solar versus hand water
pumping for off-grid area – the case of Burkina Faso”, MSc thesis, Imperial College
London, UK, 2018.
[19] W. Ratterman, J. Cohen, A. Garwood, “Green Empowerment Solar Pumping Systems
(SPS) Introductory and Feasibility Guide,” 2003. [Online]. Available:
https://sswm.info/sites/default/files/reference_attachments/RATTERMAN%20et%20al%
202003%20Solar%20Pumping%20Systems.pdf [accessed 17 Dec 2018].
[20] I. Odeh, Y. G. Yohanis, B. Norton. “Economic viability of photovoltaic water pumping
systems”, Solar Energy, vol. 80, no. 7, 850–860, 2006.
[21] M. Abu-Aligah, “Design of Photovoltaic Water Pumping System and Compare it with
Diesel Powered Pump.,” Jordan Journal of Mechanical and Industrial Engineering, vol.
5, no. 3, 2011.
[22] B. Singh, A. K. Mishra, R. Kumar, “Solar Powered Water Pumping System Employing
Switched Reluctance Motor Drive”, IEEE Transactions on Industry Applications, vol. 52,
no. 5, pp. 3949-3957, 2016.
[23] M. I. Chergui, M. O. Benaissa, “Strategy photovoltaic pumping system in scattered area,”
in 4th International Conference on Renewable Energy Research and Applications
(ICRERA), pp. 283–286, Palermo, Italy, 2015.
[24] S. Ould-Amrouche, D. Rekioua, A. Hamidat, “Modelling photovoltaic water pumping
systems and evaluation of their CO2 emissions mitigation potential,” Applied Energy, vol.
87, no. 11, pp. 3451–3459, 2010.
[25] J. Yang, A. Olsson, J. Yan, B. Chen, “A hybrid life-cycle assessment of CO2 emissions of
a PV water pumping system in China,” Energy Procedia, vol. 61, pp. 2871–2875, 2014.
[26] J. V. Mapurunga Caracas, G. De Carvalho Farias, L. F. Moreira Teixeira, L. A. De Souza
Ribeiro, “Implementation of a High-Efficiency, High-Lifetime, and Low-Cost Converter
for an Autonomous Photovoltaic Water Pumping System,” IEEE Transactions on Industry
Applications, vol. 50, no. 1, pp. 631–641, 2014.
[27] S. K. Jha, “Application of solar photovoltaic system in Oman–Overview of technology,
opportunities and challenges,” International Journal of Renewable Energy Research, vol.
3, no. 2, pp. 331–340, 2013.
[28] A. M. Armanuos, A. Negm, A. H. M. H. El Tahan, “Life Cycle Assessment of Diesel Fuel
and Solar Pumps in Operation Stage for Rice Cultivation in Tanta, Nile Delta, Egypt,”
Procedia Technology, vol. 22, pp. 478–485, 2016.
[29] P. R. Mishra, R. Panguloori, N. Udupa, D. Mitra, “Economic evaluation of solar hybrid
DC grid for petrol pump stations,” in Annual IEEE India Conference (INDICON), 2013.
[30] R. J. Chilundo, U. S. Mahanjane, D. Neves, “Design and Performance of Photovoltaic
Water Pumping Systems: Comprehensive Review towards a Renewable Strategy for
Mozambique,” Journal of Power and Energy Engineering, vol. 6, no. 7, p. 32, 2018.
[31] S. S. Chandel, M. Nagaraju Naik, R. Chandel, “Review of solar photovoltaic water
pumping system technology for irrigation and community drinking water supplies,”
Renewable and Sustainable Energy Reviews, vol. 49, pp. 1084–1099, 2015.
[32] R. Foster, A. Cota, “Solar Water Pumping Advances and Comparative Economics,” Energy
Procedia, vol. 57, pp. 1431–1436, 2014.
157
[33] R. Parajuli, G. R. Pokharel, P. A. Ostergaard, “A comparison of diesel, biodiesel and solar
PV-based water pumping systems in the context of rural Nepal,” International Journal of
Sustainable Energy, 2013.
[34] Z. Girma, “Techno-economic analysis of photovoltaic pumping system for rural water
supply in Ethiopia,” International Journal of Sustainable Energy, vol. 36, no. 3, pp. 277–
295, 2017.
[35] A. Allouhi, M. S. Buker, H. El-houari, A. Boharb, M. Benzakour Amine, T. Kousksou, A.
Jamil, “PV water pumping systems for domestic uses in remote areas: Sizing process,
simulation and economic evaluation,” Renewable Energy, vol. 132, pp. 798–812, 2019.
[36] A. Kumar, T. C. Kandpal, “Potential and cost of CO2 emissions mitigation by using solar
photovoltaic pumps in India,” International Journal of Sustainable Energy, vol. 26, no. 3,
pp. 159–166, 2007.
[37] M. N. I. Sarkar, H. R. Ghosh, “Techno-economic analysis and challenges of solar powered
pumps dissemination in Bangladesh”, Sustainable Energy Technologies and Assessments,
vol. 20, pp. 33-46, 2017.
[38] S. Rehman, A. Z. Sahin, “Performance comparison of diesel and solar photovoltaic power
systems for water pumping in Saudi Arabia,” International Journal of Green Energy, vol.
12, no. 7, pp. 702–713, 2015.
[39] Organisation for Economic Co-operation and Development (OECD), “Greenhouse gas
emissions,” 2017. [Online]. Available:
https://stats.oecd.org/Index.aspx?DataSetCode=AIR_GHG [accessed 20 Aug 2019].
[40] R. Sen, S. C. Bhattacharyya, “Off-grid electricity generation with renewable energy
technologies in India: An application of HOMER,” Renewable Energy, vol. 62, pp. 388–
398, 2014.
[41] K. Meah, S. Fletcher, S. Ula, “Solar photovoltaic water pumping for remote locations,”
Renewable and Sustainable Energy Reviews, vol. 12, no. 2, pp. 472–487, 2008.
[42] A. S. Mundada, K. K. Shah, J. M. Pearce, “Levelized cost of electricity for solar
photovoltaic, battery and cogen hybrid systems,” Renewable and Sustainable Energy
Reviews, vol. 57, pp. 692–703, 2016.
[43] V. C. Sontake, V. R. Kalamkar, “Solar photovoltaic water pumping system - A
comprehensive review,” Renewable and Sustainable Energy Reviews, vol. 59, pp. 1038–
1067, 2016.
[44] B. D. Vick, B. A. Neal, “Analysis of off-grid hybrid wind turbine/solar PV water pumping
systems,” Solar Energy, vol. 86, no. 5, pp. 1197–1207, 2012.
[45] E. G. Wagner, J. N. Lanoix, “Water supply for rural areas and small communities,” World
Health Organization (WHO), 1959. [Online]. Available:
https://www.who.int/water_sanitation_health/dwq/monograph42.pdf [accessed 19 Dec
2018].
[46] A single drop for safe water, Oxfam, Kadtundaya foundation inc., MTB, “Hand pump and
sanitation facilities design manual,” 2011. [Online]. Available:
https://www.singledrop.org/wp-
content/uploads/2010/12/Handpump_and_Sanitation_Facilities_Design_Manual.pdf
[accessed 20 Aug 2018].
[47] A. C. Telmo, “A Water Supply and Sanitation Study of the Village of Gouansolo in Mali,
West Africa,” MSc Thesis, Michigan Technological University, MI: USA, 2002.
158
[48] The United Nations Refugee Agency (UNHCR), “The geophysical exploration for
groundwater at the Kakuma refugee camp and the proposed Kaloeyei refugee camp in
Turkana county, Kenya,” 2016. [Online]. Available:
https://seg.org/Portals/0/SEG/About%20SEG/GWB/Projects/Geophysical%20Exploratio
n%20GW%20Kakuma_Kalobeyei%20Turkana%20County-
Rev%200%20Final%20Report%20No%20financials.pdf [accessed 15 May 2018].
[49] A. R. Freeze, J. A. Cherry, Groundwater, Pearson Publishing, 1979.
[50] Y. Bakelli, A. Hadj Arab, B. Azoui, “Optimal sizing of photovoltaic pumping system with
water tank storage using LPSP concept”, Solar Energy, vol. 85, no. 2, pp. 288-294, 2011.
[51] D. H. Muhsen, A. B. Ghazali, T. Khatib, “Multiobjective differential evolution algorithm-
based sizing of a standalone photovoltaic water pumping system,” Energy Conversion and
Management, vol. 118, pp. 32–43, 2016.
[52] D. H. Muhsen, A. B. Ghazali, T. Khatib, I. A. Abed, E. M. Natsheh, “Sizing of a standalone
photovoltaic water pumping system using a multi-objective evolutionary algorithm,”
Energy, vol. 109, pp. 961–973, 2016.
[53] A. Hamidat, B. Benyoucef, “Systematic procedures for sizing photovoltaic pumping
system, using water tank storage,” Energy Policy, vol. 37, no. 4, pp. 1489–1501, 2009.
[54] C. Nauges, D. Whittington, “Estimation of water demand in developing countries: An
overview,” World Bank Research Observer, vol. 25, no. 2, pp. 263–294, 2009.
[55] X. Mu, D. Whittington, J. Briscoe, “Modeling village water demand behavior: A discrete
choice approach,” Water Resources Research, vol. 26, no. 4, pp. 521-529, 1990.
[56] C. Nauges, C. Van Den Berg, “Perception of health risk and averting behavior: An analysis
of household water consumption in Southwest Sri Lanka,” Toulouse School of Economics
(TSE) Working Papers 09-139, 2009.
[57] C. Nauges, J. Strand, “Estimation of non-tap water demand in Central American cities,”
Resource and Energy Economics, vol. 29, no. 3, pp. 165–182, 2007.
[58] T. H. Persson, “Household choice of drinking–water source in the Philippines,” Asian
Economic Journal, vol. 16, no. 4, pp. 303–316, 2002.
[59] Deutsche gesellschaft für internationale zusammenarbeit (GIZ), “Access to Water and
Sanitation in Sub-Saharan Africa.” 2016. [Online]. Available:
http://www.oecd.org/water/GIZ_2018_Access_Study_Part%20I_Synthesis_Report.pdf
[60] African Development Bank (AfDB), “Improve the quality of life for the people of Africa -
Annual Development Effectiveness Review 2016,” 2016. [Online]. Available:
https://www.afdb.org/fileadmin/uploads/afdb/Documents/Development_Effectiveness_R
eview_2016/ADER_2016_-_07_Chapter_5__En_.pdf [accessed 20 July 2018].
[61] J. A. Cherni, I. Dyner, F. Henao, P. Jaramillo, R. Smith, R. O. Font, “Energy supply for
sustainable rural livelihoods. A multi-criteria decision-support system,” Energy Policy,
vol. 35, no. 3, pp. 1493–1504, 2007.
[62] M. Beccali, M. Cellura, D. Ardente, “Decision making in energy planning: the ELECTRE
multicriteria analysis approach compared to a FUZZY-SETS methodology,” Energy
Conversion and Management, vol. 39, no. 16-18, pp. 1869-1881, 1998.
[63] United Nations Development Programme (UNDP), “Human Development Report 2016:
Human Development for Everyone,” 2016. [Online]. Available:
https://www.un-ilibrary.org/economic-and-social-development/human-development-
report-2016_b6186701-en [accessed 20 Aug 2018].
[64] Respublica, “Respublica – ONG Burkina Faso,” 2019. [Online]. Available:
http://www.respublica.asso.fr/ong-humanitaire-au-burkina-faso [accessed 12 Aug 2019].
159
[65] Eau Fil du Soleil, “Gogma - Crowdfunding,” 2017. [Online]. Available:
https://www.kisskissbankbank.com/fr/projects/turning-sun-into-water/tabs/description
[66] Eau Fil du Soleil, “Eau fil du soleil - website,” 2019. [Online]. Available:
https://eaufildusoleil.blogspot.com/ [accessed 15 Aug 2019].
[67] DargaTech SARL, “DargaTech,” 2019. [Online]. Available:
https://www.facebook.com/dargatech/ [accessed 15 Aug 2019].
[68] Grundfos, “Grundfos - main page,” 2019. [Online]. Available:
https://www.grundfos.com/ [accessed 10 Aug 2019].
[69] Grundfos, “Performance curve of SQFlex 5A-7 motor-pump,” 2019, [Online]. Available:
https://product-selection.grundfos.com/product-detail.product-
detail.html?custid=GMA&productnumber=95027342&qcid=421916400 [accessed 10
Aug 2019].
[70] Grundfos, “CU 200 controller,” 2019, [Online]. Available:
detail.html?custid=GMA&productnumber=96625360 [accessed 10 Aug 2019].
[71] Institut Supérieur d’Application des Géosciences - ISAG, “ISAG - Website,” 2019.
[Online]. Available: https://www.facebook.com/ISAG.Geosciences/ [accessed 05 March
2019].
[72] Laboratoire Aina, “Water quality tests in Gogma performed for the NGO Eau Fil du
Soleil,” 2018.
[73] Windows, “GPS Satellite.” 2017, [Online]. Available:
https://www.microsoft.com/fr-fr/p/gps-
satellite/9wzdncrfj0dg?activetab=pivot%3Aoverviewtab [accessed 07 Oct 2018].
[74] United States Environmental Protection Agency (EPA), “Frequency Domain
Electromagnetic Methods,” 2016. [Online]. Available:
https://archive.epa.gov/esd/archive-
geophysics/web/html/frequency_domain_electromagnetic_methods.html [accessed 22
Nov 2018].
[75] Solinst, “Water Level Meter 101 Datasheet”, 2016. [Online]. Available:
https://www.solinst.com/products/level-measurement-devices/101-water-level-
meter/datasheet/ [accessed 12 May 2019].
[76] Massachusetts Institute of Technology, “Safe water for one billion people,” 2008. [Online].
Available: http://web.mit.edu/watsan/methods_waterqualitystandards.html [accessed 01
Oct 2018].
[77] The Pennsylvania State University, “Groundwater Surveillance In the Vicinity of
Mahoning County Landfills, 2006-2008,” 2008. [Online]. Available:
https://www.mahoninghealth.org/wp-
content/uploads/file/Documents/Documents/SpecialReports/Groundwater%20Surveillanc
e%20around%20Mahoning%20County%20Landfills%202006-2008.pdf [accessed 05 Dec
2018].
[78] S. Meunier, M. Heinrich, L. Queval, J. A. Cherni, L. Vido, A. Darga, Ph. Dessante, B.
Multon, P. K. Kitanidis and C. Marchand, "A validated model of a photovoltaic water
pumping system for off-grid rural communities", Applied Energy, vol 241, pp. 580-591,
2019.
[79] M. Heinrich, S. Meunier, A. Samé, L. Quéval, A. Darga, L. Oukhellou, B. Multon,
“Detection of cleaning interventions on photovoltaic modules with machine learning” –
submitted
160
[80] Axo Energy, “Website – Axo Energy,” 2019. [Online]. Available:
https://axoenergy.blogspot.com/?fbclid=IwAR0yZtbFxO40k-
kkS6n8Wf4MCFZLl6Iwiy79rbZku7CVHQkXlX4PAjopR5M [accessed 12 July 2019].
[81] S. Meunier, L. Queval, M. Heinrich, J. A. Cherni, L. Vido, A. Darga, Ph. Dessante, B.
Multon, P. K. Kitanidis and C. Marchand, "Modelling and optimal sizing of photovoltaic
water pumping systems – Sensitivity analysis," in 14th International Conference on
Ecological Vehicles and Renewable Energies (EVER), Monaco, Monaco, 2019.
[82] V. Caplain, “Water demand and organization and payment schemes for solar pumping
systems in emerging countries: a case study in Burkina Faso,” MSc Thesis, Imperial
College London, UK, 2018.
[83] C. Boone, P. Glick, D. E. Sahn, “Household Water Supply Choice and Time Allocated to
Water Collection: Evidence from Madagascar,” vol. 47, no. 12, pp. 1826-1850, 2011.
[84] The World Bank Water Demand Research Team, “The demand for water in rural areas:
determinants and policy implications,” The World Bank Research Observer, vol.8, no. 1,
pp. 47–70, 1993.
[85] J.-A. L. Geere, P. R. Hunter, P. Jagals, “Domestic water carrying and its implications for
health: a review and mixed methods pilot study in Limpopo Province, South Africa,”
Environmental Health, vol. 9, no. 1, pp. 52, 2010.
[86] J. C. Ho, K. C. Russel, J. Davis, “The challenge of global water access
monitoring: Evaluating straight-line distance versus self-reported travel time among
rural households in Mozambique,”, Journal of Water and Health, vol. 12, no. 1, pp. 173–
183, 2014.
[87] The Pennsylvania State University, “Detecting Multicollinearity Using Variance Inflation
Factors,” 2018. [Online]. Available:
https://newonlinecourses.science.psu.edu/stat462/node/180/ [accessed 05 Dec 2018].
[88] R. M. Craparo, Significance level, N. J. Salkind (Ed.), Encyclopedia of Measurement and
Statistics 3, Thousand Oaks, CA: SAGE Publications, 2007.
[89] M. Koehl, M. Heck, S. Wiesmeier, J. Wirth, “Modeling of the nominal operating cell
temperature based on outdoor weathering,” Solar Energy Materials and Solar Cells, vol.
95, no. 7, pp. 1638–1646, 2011.
[90] S. Ponce-Alcántara, J. P. Connolly, G. Sánchez, J. M. Míguez, V. Hoffmann, R. Ordás, “A
Statistical Analysis of the Temperature Coefficients of Industrial Silicon Solar Cells,”
Energy Procedia, vol. 55, pp. 578–588, 2014.
[91] A. McEvoy, T. Markvart, L. Castañer, Practical Handbook of Photovoltaics - Fundamental
and Applications, 2nd ed, Academic Press, 2011.
[92] P. E. Campana, H. Li, J. Yan, “Dynamic modelling of a PV pumping system with special
consideration on water demand,” Applied Energy, vol. 112, pp. 635–645, Dec. 2013.
[93] W. H. Bierschenk, “Determining well efficiency by multiple step-drawdown tests,”
International Association of Scientific Hydrology, 1963. [Online]. Available:
http://hydrologie.org/redbooks/a064/064043.pdf [accessed 05 Oct 2018].
[94] T. E. Reilly, L. Franke, G. D. Bennett, “The principle of superposition and its application
in ground-water hydraulics,” US Geological Survey, Open-File Report 84-459, 1987.
[Online]. Available:
https://pubs.usgs.gov/of/1984/0459/report.pdf [accessed 05 Oct 2018].
[95] T. Vezin, S. Meunier, L. Queval, J. A. Cherni, L. Vido, A. Darga, Ph. Dessante, P. K.
Kitanidis, C. Marchand, "Borehole water level model for photovoltaic water pumping
systems”, Applied Energy, 2019.
[96] B. R. Munson, D. F. Young, T. H. Okiishi, Fundamentals of Fluid Mechanics, 5th ed.,
Wiley, 2006.
161
[97] Grundfos, “Performance curve of the SQFlex 8A-5 motor-pump,” 2019. [Online].
Available:
Aug 2019].
[98] Grundfos, “Performance curve of the SQFlex 2.5-2 motor-pump,” 2019. [Online].
Available:
Aug 2019].
Available:
Aug 2019].
Available:
detail.html?from_suid=156826518846804965939417405789&hits=3&productnumber=9
5027324&searchstring=95027324&qcid=638338263 [accessed 4 Aug 2019].
Available:
Aug 2019].
[102] Grundfos, “Perfromance curve of the SQFlex 8A-3 motor-pump,” 2019. [Online].
Available:
Aug 2019].
Available:
detail.html?from_suid=15682666844790044808622003218224&hits=1&productnumber
=95027338&searchstring=95027338&qcid=638358897 [accessed 13 Aug 2019].
[104] S. Meunier, L. Queval, A. Darga, Ph. Dessante, C. Marchand, M. Heinrich, J. A. Cherni,
L. Vido, and B. Multon, "Influence of the temporal resolution of the water consumption
profile on photovoltaic water pumping systems modelling and sizing," 7th International
Conference on Renewable Energy Research and Application (ICRERA), pp. 494-499,
Paris, France, 2018.
[105] S. Meunier, L. Queval, M. Heinrich, E. A. de la Fresnaye, J. A. Cherni, L. Vido, A. Darga,
Ph. Dessante, B. Multon, P. K. Kitanidis, and C. Marchand, "Effect of irradiance data on
the optimal sizing of photovoltaic water pumping systems," 46th IEEE Photovoltaic
Specialists Conference (PVSC), Chicago (IL), United States, June 2019.
[106] European Commission, “Copernicus Atmosphere Monitoring Service (CAMS),” 2017,
http://www.soda-pro.com/web-services/radiation/cams-radiation-service [accessed 01 Oct
2018].
[107] J. Mayne, “Useful theory of change models,” Canadian Journal of Program Evaluation,
vol. 30, no. 2, 2015.
162
[108] A. A. Anderson, “The community builder’s approach to Theory of Change,” The Aspen
Institute, 2009, [Online]. Available:
http://www.theoryofchange.org/pdf/TOC_fac_guide.pdf [accessed 15 Jan 2018].
[109] S. Meunier, D. T. Manning, L. Queval, J.A. Cherni, Ph. Dessante, and D. Zimmerle
"Determinants of the marginal willingness to pay for domestic water and irrigation in
partially electrified villages: the case of Rwanda", International Journal of Sustainable
Development & World Ecology, vol 26, no. 6, pp. 547-559, 2019.
[110] L. Fewtrell, R. B. Kaufmann, D. Kay, W. Enanoria, L. Haller, J. M. Jr. Colford, “Water,
Sanitation, and Hygiene Interventions to Reduce Diarrhoea in Less Developed Countries:
A Systematic Review and Meta-Analysis,” Lancet Infectious Diseases, vol. 5, no. 1, pp.
42-52, 2005.
[111] D. Whittington, X. Mu, R. Roche, “Calculating the value of time spent collecting water:
Some estimates for Ukunda, Kenya,” World Development, vol. 18, no. 2, pp. 269–280,
1990.
[112] H. Waddington, B. Sinlstveit, H. White, L. Fewtrell, “Water, Sanitation and Hygiene
Interventions to Combat Childhood Diarrhea in Developing Countries,” International
Initiative for Impact Evaluation, 2009. [Online]. Available:
https://www.3ieimpact.org/sites/default/files/2019-01/sr1-WASH.pdf [accessed 15 Dec
2018].
[113] World Health Organization (WHO), “Water, sanitation and hygiene links to health,” 2004.
http://www.who.int/water_sanitation_health/publications/facts2004/en/ [accessed 15 Dec
2017].
[114] S. Cairncross, J. L. Cuff, “Water use and health in Mueda, Mozambique,” Transactions of
The Royal Society of Tropical Medicine and Hygiene, vol. 81, no. 1, pp. 51-54, 1987.
[115] E. Gross, I. Günther, Y. Schipper, “Women: Walking and Waiting for Water The Time
Value of Public Water Supply,” Courant Research Centre: Poverty, Equity and Growth-
Discussion Papers, no. 134, 2013.
[116] H. Batteson, K. Davey, R. Shaw, “Guidance manual on water supply and
sanitation programmes”, Technical report, WEDC Loudhborough University, 1998.
[Online]. Available: http://www.rural-water-supply.net/_ressources/documents/default/1-
491-2-1366101476.pdf [accessed 15 Dec 2017].
[117] The water project, “A good education begins with access to safe water and proper
sanitation,” 2017. [Online]. Available: https://thewaterproject.org/why-water/education.
[118] A. A. Churchill, D. de Ferranti, R. Roche, C. Tager, A. A. Walter, and A. Yazer, “Rural
water supply and sanitation: time for a change.,” World Bank Discussion Paper, no. 18,
1987.
[119] United Nations Development Programme (UNDP), “Human Development Report,” 2006.
[Online]. Available: http://hdr.undp.org/en/reports/global/hdr2006/ [accessed 29 Dec
2018].
[120] Policy and Operations Evaluation Department, Ministry of Foreign Affairs, the
Netherlands, “Rural water and sanitation - Assessing impacts,” 2012. [Online]. Available:
https://www.oecd.org/dac/evaluation/Evaluation%20insights%20WASH%20final%20dra
ft.pdf [accessed 17 May 2018].
[121] United Nations (UN), “The Millennium Development Goals Report.” 2015. [Online].
Available:
https://www.un.org/millenniumgoals/2015_MDG_Report/pdf/MDG%202015%20rev%2
0(July%201).pdf [accessed 17 May 2018].
163
[122] J. M. Rohrer, “Thinking clearly about correlations and causation: Graphical causal models
for observational data,” Advances in Methods and Practices in Psychological Science, vol.
1, no. 1, pp. 27–42, 2018.
[123] J. A. List, S. Sadoff, M. Wagner, “So you want to run an experiment, now what? Some
simple rules of thumb for optimal experimental design,” Experimental Economics, vol. 14,
no. 4, p. 439, 2011.
[124] Marie Tapiwa Bossyns, “Appropriate Renewable Energy for Water Pumping in Rural
Mozambique,” MSc Thesis, Universiteit Gent, 2013.
[125] C. Zhang, P. E. Campana, J. Yang, C. Yu, J. Yan, “Economic assessment of photovoltaic
water pumping integration with dairy milk production,” Energy Conversion and
Management, vol. 177, pp. 750–764, 2018.
[126] E. Phiri, P. Rowley, R. E. Blanchard, “The feasibility of solar water-pumping in a rural
village of Malawi,” in 1st International Conference on Solar Energy Solutions for
Electricity and Water Supply in Rural Area, Cairo, Egypt, 2015.
[127] M. Montorfano, D. Sbarbaro, L. Moran, “Economic and Technical Evaluation of Solar-
Assisted Water Pump Stations for Mining Applications: A Case of Study,” IEEE
Transactions on Industry Applications, vol. 52, no. 5, pp. 4454–4459, 2016.
[128] B. Bouzidi, “New sizing method of PV water pumping systems”, Sustainable Energy
Technologies and Assessments, vol. 4, pp. 1‑10, 2013.
[129] I. Yahyaoui, A. Atieh, A. Serna, F. Tadeo, “Sensitivity analysis for photovoltaic water
pumping systems: Energetic and economic studies,” Energy Conversion and Management,
vol. 135, pp. 402–415, 2017.
[130] Off-grid Europe, “Off-grid Europe - Grundfos SQ Flex,” 2017. [Online]. Available:
https://www.off-grid-europe.com/accessories/pumps/grundfos/sq-flex [accessed 23 May
2018].
[131] B. Bora, K. Yadav, M. Bangar, A. Kumar, O. S. Sastry, “Analysis of temperature effect on
optimum sizing of solar photovoltaic water pumping system,” in 42nd IEEE Photovoltaic
Specialist Conference (PVSC), New Orleans, USA (LA), 2015.
[132] A. A. Ghoneim, “Design optimization of photovoltaic powered water pumping systems,”
Energy Conversion and Management, vol. 47, no. 11–12, pp. 1449–1463, 2006.
[133] K. Price, R. M. Storn, J. A. Lampinen, Differential evolution: a practical approach to
global optimization (natural computing series), Secaucus, NJ, USA: Springer-Verlag New
York, Inc., 2005.
[134] Hewlett-Packard (HP), “HP EliteBook 840 G3 - datasheet,” 2018. [Online]. Available:
http://www8.hp.com/h20195/V2/GetPDF.aspx/4aa6-3222eee [accessed 2 Sept 2019].
[135] National Aeronautics and Space Administration (NASA), “Modern-Era Retrospective
analysis for Research and Applications, Version 2,” 2017, [Online]. Available:
https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/ [accessed 01 Oct 2018].
[136] L. Fraas, L. Partain, Solar cells and their applications, 2nd ed., Wiley, 2010.
[137] Sandia National Laboratories,”PV_LIB Toolbox”, 2018, [Online]. Available:
https://pvpmc.sandia.gov/applications/pv_lib-toolbox/ [accessed 02 Dec 2018].
[138] I. Nazrul, A. Zahangir, M. Minhaz, “Determination of Well Loss and Aquifer Loss of New
Construction Deep Water Well at Artesian Aquifer in Khulna City, Bangladesh”, Recent
Trends in Civil Engineering & Technology, vol. 3, no. 1, pp. 8-13, 2013.
[139] D. P. Dufresne, “Developing Better Regional Groundwater Flow Models with Effective
Use of Step-Drawdown Test Results”, Florida Water Resources Journal, vol. 63, no. 2, pp.
36-40., 2011.
164
Titre : Conception optimale des systèmes photovoltaïques de pompage d’eau pour les communautés
rurales – une approche technique, économique et sociale
Mots clés : Modélisation interdisciplinaire, Analyse multi-critères, Conception optimale,

Planification des infrastructures, Système photovoltaïque de pompage d’eau
Résumé : Les systèmes photovoltaïques de La première originalité principale de ce travail est

pompage d’eau (PVWPS) sont une solution la modélisation du lien entre la conception du
intéressante pour améliorer l’accès à l’eau dans PVWPS et son impact socio-économique, ce qui
les communautés rurales des pays en voie de permet d’inclure l’impact socio-économique
développement. Cette thèse développe une comme fonction objectif de l’optimisation. La
méthodologie de conception optimale des seconde originalité principale est l’intégration de
PVWPS pour l’accès à l’eau domestique basée la position géographique du PVWPS dans le
sur une approche interdisciplinaire. L’objectif est village comme variable d’optimisation, en plus
de déterminer les dimensionnements du PVWPS du dimensionnement du système. Cette
et ses positions géographiques dans le village qui méthodologie pourrait également être appliquée à
maximisent l’impact positif du système sur le la mise en place d'autres types de systèmes, tels
développement socio-économique et minimisent que les moulins communaux alimentés par
son coût sur cycle de vie. Cette méthodologie est l’énergie photovoltaïque dans les zones isolées ou
appliquée au cas d’un village rural du Burkina les bornes de recharges publiques pour les
Faso, où nous avons collecté des données véhicules électriques dans les villes.
techniques et sociaux-économiques depuis 2 ans.
Title : Optimal design of photovoltaic water pumping systems for rural communities – a technical,
economic and social approach
Keywords : Interdisciplinary modelling, Multi-criteria analysis, Optimal design,

Infrastructure planning, Photovoltaic water pumping system
Abstract : Photovoltaic water pumping systems The first main originality of this work is the
(PVWPS) are an interesting solution to improve modelling of the link between the design of a
access to water in rural communities of PVWPS and its socio-economic impact, which
developing countries. This thesis develops a allows to include the socio-economic impact as
methodology for the optimal design of PVWPS an objective function of the optimisation. The
for domestic consumption based on an second main originality is the inclusion of the
interdisciplinary approach. The objective is to geographical position of the PVWPS in the
determine the sizings of the PVWPS and its village as an optimisation variable, in addition to
geographical positions in the village that the sizing of the system. There is potential for
maximise the positive impact of the system on applying the proposed methodology for the set-
socio-economic development and minimise its up of other types of systems such as community
life-cycle cost. This methodology is applied to mills powered by photovoltaic energy in isolated
the case of a rural village in Burkina Faso, where areas and public charging points for electrical
we have been collecting technical and socio- vehicles in cities.
economic data for 2 years.
Université Paris-Saclay
Espace Technologique / Immeuble Discovery
Route de l’Orme aux Merisiers RD 128 / 91190 Saint-Aubin, France

83782 MEUNIER 2019 Archivage

Uploaded by

Copyright:

Available Formats

83782 MEUNIER 2019 Archivage

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

83782 MEUNIER 2019 Archivage

Uploaded by

Copyright:

Available Formats

Optimal design of photovoltaic water pumping systems

for rural communities – a technical, economic and social

To cite this version:

HAL Id: tel-02414400

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

water pumping systems for rural

Ecole doctorale n°575 Electrical, optical, bio-physics and engineering, EOBE

Thèse présentée et soutenue à Gif-sur-Yvette, le 06/12/2019, par

List of figures ................................................................................................................................................ 7

List of tables .................................................................................................................................................. 9

Chapter I Literature review ..................................................................................................................... 17

I.1 The water-energy nexus in rural areas of sub-Saharan Africa .................................................... 18

I.1.2 Electricity access ..................................................................................................................... 19

I.2 Electrified water pumping technologies for off-grid areas .......................................................... 20

I.3.2 Position .................................................................................................................................... 22

I.4 Literature gaps and research objectives ....................................................................................... 25

II.1 Case study village and PVWPS ................................................................................................... 29

II.1.2 PVWPS .................................................................................................................................... 31

II.2 Data collection ............................................................................................................................. 37

II.2.2 GIS mapping............................................................................................................................ 39

II.2.4 Account books of water sources .............................................................................................. 41

II.2.6 Pumping test ............................................................................................................................ 43

II.2.8 Monitoring of the PVWPS of Gogma ..................................................................................... 45

II.3 Partial conclusion ........................................................................................................................ 49

III.2.2 Determination of the water demand profile at the PVWPS ................................................. 60

III.3 Technical model .......................................................................................................................... 63

III.3.2 Energy conversion model – generalization ......................................................................... 72

III.3.3 Beneficiaries identification model ....................................................................................... 76

III.4 Impact model ............................................................................................................................... 78

III.4.2 Quantification of the indicators for each household ............................................................ 81

III.5 Economic model .......................................................................................................................... 84

III.5.2 Results ................................................................................................................................. 87

III.6 Partial conclusion ........................................................................................................................ 89

IV.1 Optimisation problem .................................................................................................................. 92

IV.2.2 Bi-objective optimisation results ......................................................................................... 96

IV.6.2 Case study.......................................................................................................................... 113

IV.7 Partial conclusion ...................................................................................................................... 116

List of Publications .................................................................................................................................... 122

Appendix ................................................................................................................................................... 125

Appendix A. Household survey ............................................................................................................ 125

Chapitre I : Revue de littérature ............................................................................................................ 141

1 XOF (currency in Burkina Faso) = $0.00177

I.1 The water-energy nexus in rural areas of sub-Saharan Africa

I.1.2 Electricity access

I.2 Electrified water pumping technologies for off-grid areas

I.3 Conventional PVWPS design

 Architecture, i.e., the choice and the disposition of the components.

I.4 Literature gaps and research objectives

II.1 Case study village and PVWPS

Figure II-3 – Household gatherings.

Figure II-4 – Open well. Figure II-5 – Hand pump.

Is there water flowing out of the borehole? No

Does the physico-chemical water quality

Does the decision maker consider that it is worth

Does the decision maker consider that the

II.1.2.2 Technical description

Figure II-9 – Overview of the PVWPS.

Table II-1 – Features of the current PVWPS of Gogma.

II.2 Data collection

Table II-2 – Summary of collected data.