Benchmarking water productivity
in agriculture and the scope for
Sander J. Zwart
Benchmarking water productivity in agriculture and
the scope for improvement
remote sensing modelling from field to global scale
ter verkrijging van de graad van doctor
aan de Technische Universiteit Delft,
op gezag van de Rector Magnificus Prof. ir. K.C.A.M. Luyben,
voorzitter van het College voor Promoties,
in het openbaar te verdedigen
op woensdag 26 mei 2010 om 12.30 uur
Sander Jaap ZWART
Ingenieur in tropisch landgebruik
Master in geo-informatiekunde
geboren te Smallingerland
Dit proefschrift is goedgekeurd door de promotor:
Prof. dr. W.G.M. Bastiaanssen
Samenstelling promotiecommissie:
Rector Magnificus, voorzitter
Prof. dr. W.G.M. Bastiaanssen, Technische Universiteit Delft, promotor
Prof. dr. M. Menenti, Technische Universiteit Delft
Prof. dr. ir. A.Y. Hoekstra, Universiteit Twente
Prof. dr. ir. N.C. van de Giesen, Technische Universiteit Delft
Prof. dr. ir. P. van der Zaag, UNESCO-IHE - Technische Universiteit Delft
Prof. dr. ir. R. Rabbinge, Wageningen Universiteit
Dr. P. Steduto, Food and Agriculture Organisation of the United Nations
Prof. dr. ir. H.H.G. Savenije, Technische Universiteit Delft, reservelid
Title: Benchmarking water productivity in agriculture and the scope for improvement
- remote sensing modelling from field to global scale
Keywords: water productivity, remote sensing, global modelling, wheat, cotton, rice,
ISBN: 978-90-6562-237-2
Published by: VSSD
Address: Leeghwaterstraat 42, 2628 CA Delft, The Netherlands
Telephone: +31 15 2782124
E-mail: [email protected]
Since my youth I have been intrigued by maps. When I was still in elementary school
I bought a map of the north of Canada from my first pocket money on the annual fair.
I studied this map for days and was impressed by the scale of the country and the fact
that many villages were accessible by water or by air only. Luckily, my dream of
becoming a forester in this lonely and remote area vanished, but my interest for maps
remained. Geography was my favourite subject in high school and after starting my
studies Irrigation and Water Engineering at Wageningen University I immediately
started to follow courses in GIS and Remote Sensing. I still remember the moment
that I was convinced to continue in remote sensing when a guest lecturer showed
colourful maps of soil water content derived from radar images. Although I did not
immediately think in terms of applied products for operational water management, I
became aware of the value of this new and promising technique. I studied and used
this till the end of my studies at Wageningen University.
Exactly eight years ago in May 2002 I had a job interview with Wim Bastiaanssen at
WaterWatch. He asked me whether I would be interested in pursuing a PhD alongside
my regular job. It would not entail more than writing four scientific articles - two per
year – bind them and be finished! It was at the end of the day, after a bottle of beer,
and supported by Wim’s positivism I did not hesitate to say yes to his proposal. I am
not sure whether I would have said the same if I would have known what would come
ahead. However, at the moment of writing these acknowledgements I am extremely
relieved, happy and proud that I have finalised this work. Wim, I want to thank you
deeply for your continuous positive encouragements, energy and enthusiasm during
the eight years I worked on this thesis. Your innovative ideas and your good eye to
address issues in science that become important few years later, have given me a
flying start in my career. Something I am very grateful for and I hope there will be
many possibilities for future cooperation.
Next, I would like to thank my colleagues at WaterWatch for providing technical
support where possible and for showing interest in my work. In particular, I am very
grateful to Henk Pelgrum with whom I have cooperated since I started with
WaterWatch. Your mathematical knowledge and programming skills together with
your background in remote sensing are invaluable. Without your help, the method to
determine crop seasons from the NDVI curves - a major component in the application
of my models - would not have been possible. During the course of my work, David
Molden and Charlotte de Fraiture from the International Water Management Institute
have shown their interest in my work of mapping water productivity at a global scale.
I have thereafter visited Colombo twice for a period of one month to work on the last
two papers. The open atmosphere at the institute and the possibility to critically
discuss the models and results have been a great motivation to continue and finish the
Several months ago an article was published in a national newspaper entitled ‘How’s
your PhD going?’ about people pursuing a PhD in their spare time besides their
regular job. The estimated average time frame to finish was between six and ten years,
whereas most people indicated that it was a lonely process. I have not experienced
that differently. Although I never put myself any deadline, I have continuously felt the
unconscious pressure to work on my PhD, especially in the last two years. This has
negatively impacted holidays, visits to family and friends and not least my
relationship. I am glad that this period therefore will soon come to end and it is time
to look ahead. I want to thank everybody who has shown her or his interest in my
work in the last years by asking how my PhD matters were going. Moving to another
country for the coming two years does not make it easier to catch up lost time, but
everybody is most welcome to visit me in Benin!
Finally, I would like to express deep appreciation to my parents for supporting me to
where I am now. In my early years you have always guided me softly in the right
directions and you have always created the environment to continue expanding my
knowledge. I do remember the moment that you showed me the amateurish looking
brochures for the studies Tropical Land Use at Wageningen Agricultural University.
Although not fully convinced at the start, I am extremely grateful that you ordered
these one day as it certainly supported me to put myself on the world map!
Due to the rapid growth in world population, the pressure on water resources is
increasing. In the future less water will be available for agricultural production due to
competition with the industrial and domestic sectors, while at the same time food
production must be increased to feed the growing population. It is inevitable that the
production per unit water consumed, the water productivity, must be increased to
meet this challenge. Till start of this research, little was known on the current levels of
water productivity in agriculture. Information is outdated or measured values are
made in small experimental plots that are not representative for the situation in
farmer’s fields.
This research will therefore focus on the benchmarking of physical water productivity
and gaining a better understanding of the spatial variations and the scope for
improvement. The major goal of this research was to benchmark water productivity
values globally and at various scales (field level, system level and global level). A
review of the literature sources that provide measurements of water productivity was
conducted to assess plausible ranges of water productivity levels for wheat, maize,
cotton and rice. Remote sensing and modelling were the major tools applied for this
work to assess the spatial variation of water productivity of wheat at system and
global level, and to provide a first explanation for the differences that are found.
The first step was to establish a water productivity database for the four major crops
in the world, namely wheat, rice, cotton and maize. Results from field experiments
that were reported in international literature in the recent 25 years were synthesized in
a data base to provide up-to-date ranges of feasible water productivity values. The
ranges found were higher than those reported some 25 years earlier in the FAO33
publication by Doorenbos and Kassam (1979). For example, this research provides a
plausible range of 0.6 - 1.7 kg m-3 for the water productivity of wheat (with an
average of 1.1 kg m-3), whereas a much lower range of 0.8 - 1.0 was provided by
FAO33 in the 1970’s. Also for the other three researched crops it was found that the
water productivity values in FAO33 are on the conservative side. This might partially
be related to the development of crops that are able to produce higher yields and to
improved soil fertility and water management.
Spatial information on water use, crop production and water productivity will play a
vital role for water managers to assess where scarce water resources are wasted and
where in a given region the water productivity can be improved. A methodology has
been developed to quantify spatial variation of crop yield, evapotranspiration and
water productivity using the SEBAL algorithm and high and low resolution satellite
images. SEBAL-based actual evapotranspiration estimates were validated over an
irrigated, wheat dominated area in the Yaqui Valley, Mexico and proved to be
accurate (8.8% difference for 110 days). Estimated average wheat yields in Yaqui
Valley of 5.5 ton ha-1 were well within the range of measured yields reported in the
literature. Area average water productivity in the Yaqui Valley was 1.37 kg m-3 and
could be considered to be high as compared to other irrigated systems around the
world where the same methodology was applied. A higher average value was found in
Egypt’s Nile Delta (1.52 kg m-3), Kings County (CA), USA (1.44 kg m-3) and in
Oldambt, The Netherlands (1.39 kg m-3). The spatial variability of water productivity
within low productivity systems (CV=0.33) is higher than in high productivity
systems (CV=0.05) because water supply in the former case is uncertain and farming
conditions are sub-optimal. The high CV found in areas with low water productivity
indicates that there is considerable scope for improvement. The average scope for
improvement in eight systems was 14%, indicating that 14% reduction in water
consumption can be achieved while maintaining the same yield.
The WATer PROductivity (WATPRO) model was developed to assess water
productivity of wheat on a global scale. WATPRO is based on remote sensing-derived
input data sets and can be applied at local to global scales. The model is a
combination of Monteith’s theoretical framework for dry matter production in plants
and an energy balance model to assess actual evapotranspiration. It is shown that by
combining both approaches, the evaporative fraction and the atmospheric
transmissivity, two parameters which are usually difficult to estimate spatially, can be
omitted. Water productivity can then be assessed from four spatial variables:
broadband surface albedo, the vegetation index NDVI, the extraterrestrial radiation
and air temperature.
The WATPRO model was applied at 39 locations where water productivity was
measured under experimental conditions. The correlation between measured and
modelled water productivity was low, and this can be mainly attributed to differences
in scales and in the experimental and modelling periods. A comparison with
measurements from farmer’s fields in areas surrounded by other wheat fields located
in Sirsa District, NW India, showed an improved correlation. Although not a
validation, a comparison with SEBAL-derived water productivity in the same region
in India proved that WATPRO can spatially predict water productivity with the same
spatial variation.
WATPRO was applied with global data sets of the NDVI and surface albedo to
benchmark water productivity of wheat for the beginning of this millennium. Time
profiles of the NDVI were used to determine the growing season from crop emergence
to harvest on a pixel basis. The WATPRO results were compared with modelling
information by Liu et al. (2007) who applied the GEPIC model at a global scale to
map water productivity, and by Chapagain and Hoekstra (2004) who used FAO
statistics to determine water productivity per country. A comparison with Liu et al.
showed a good correlation for most countries, but the correlation with the results by
Chapagain and Hoekstra was less obvious. It was found that water productivity varies
from approximately 0.2 to 1.8 kg of harvestable wheat per cubic metre of water
consumed. From the 10 largest producers of wheat, France and Germany score the
highest country average water productivity of 1.42 and 1.35 kg m-3 respectively.
The global patterns of the water productivity map were compared with global data
sets of precipitation and reference evapotranspiration to determine the impact of
climate and of water availability reflected by precipitation. It appeared that the highest
levels of water productivity are to be expected in temperate climates with high
precipitation. Due to its non-linear relationship with precipitation, it is expected that
large gains in water productivity can be made with rain water harvesting or
supplemental irrigation in dry areas with low seasonal precipitation. Investing in rain
water harvesting techniques and/or systems for supplemental irrigation, in
combination with improved agronomic management and the use of fertilizers, may
give a significant boost to the productive use of water resources within a basin. A full
understanding of the spatial patterns by country or river basin will support decisions
on where to invest and what measures to take to make agriculture more water
Als gevolg van een snel toenemende wereldbevolking, neemt de druk op de
beschikbare water bronnen toe. In de toekomst zal minder water beschikbaar zijn voor
landbouwproductie als gevolg van competitie met andere sectoren, zoals de industrie
en met huishoudelijk gebruik. Tegelijkertijd zal de productie moeten worden
verhoogd om de groeiende wereldbevolking van voldoende voedsel te blijven
voorzien. Het is derhalve onvermijdelijk dat de productie per eenheid waterverbruik,
ofwel de waterproductiviteit, zal moeten worden verbeterd. Voor aanvang van deze
studie was weinig bekend van het huidige niveau van water productiviteit in de
landbouw. Informatie was verouderd of de gemeten waarden waren slechts
representatief voor kleine experimentele opzetten en niet voor de actuele situatie in de
percelen van boeren.
Dit onderzoek heeft zich gericht op het vastleggen van de fysieke water productiviteit
en op het verkrijgen van een beter begrip van de ruimtelijk patronen en de ruimte om
de water productiviteit te verbeteren. Het hoofddoel van dit onderzoek was om de
water productiviteit mondiaal vast te leggen op veldschaal, systeem niveau en globale
schaal. Een overzicht wordt gegeven van literatuurbronnen die experimentele water
productiviteitsmetingen rapporteren met als doel voor de gewassen tarwe, maïs,
katoen en rijst een aannemelijke bandbreedte van de waterproductiviteit aan te geven.
Aardobservatie en modellen waren de belangrijkste methoden die zijn gebruikt binnen
dit werk om de ruimtelijke variatie van tarwe op systeem niveau en globale schaal in
beeld te brengen, en om een eerste interpretatie te geven voor de verschillen die
werden gevonden.
Het eerste onderdeel van dit onderzoek was het opzetten van een data base met
metingen van water productiviteit van de vier belangrijkste gewassen in de wereld, te
weten tarwe, maïs, katoen en rijst. Experimentele resultaten die in de afgelopen 25
jaar werden gerapporteerd in de internationale literatuur werden samengevat in een
data base met als doel actuele waarden van de water productiviteit vast te leggen.
Deze waarden waren hoger dan degenen die 25 jaar eerder werden gerapporteerd in de
FAO33 publicatie door Doorenbos and Kassam (1979). Dit onderzoek leverde
bijvoorbeeld voor tarwe een aannemelijk bandbreedte van water productiviteit op van
0,6 tot 1,7 kg m-3 (met een gemiddelde van 1,1 kg m-3), terwijl eind jaren 70 in de
FAO33 publicatie veel lagere waarden worden gerapporteerd (0,8-1,0 kg m-3). Ook
voor de andere drie onderzochte gewassen werd waargenomen dat de waarden in de
FAO33 publicatie lager waren dan degene gevonden in deze studie. Dit zou
gedeeltelijk verklaard kunnen worden door de ontwikkeling van verbeterde gewassen
die betere opbrengsten kunnen leveren en door verbeterd management ten aanzien van
bodemvruchtbaarheid en irrigatie.
Ruimtelijke informatie over water gebruik, gewasproductie en water productiviteit is
van wezenlijk belang voor water managers om vast te kunnen stellen waar schaars
beschikbaar water wordt verspild en waar binnen een gebied de water productiviteit
kan worden verbeterd. Een methodologie werd ontwikkeld om de gewasopbrengsten,
de verdamping en de water productiviteit te kwantificeren met de SEBAL algoritme
toegepast op satellietbeelden met hoge en lage resolutie. De op SEBAL gebaseerde
schattingen van de verdamping werden naar tevredenheid gevalideerd in de
geïrrigeerde Yaqui Vallei in noordwest Mexico, een gebied waar voornamelijk tarwe
wordt verbouwd. Het verschil bedroeg 8.8% voor 110 dagen van het groeizoen. De
geschatte tarwe opbrengsten bedroegen 5.5 ton per hectare en waren daarmee vrijwel
gelijk aan gerapporteerde opbrengsten in de literatuur. De gemiddelde water
productiviteit in de Yaqui Vallei bedroeg 1,37 kg m-3 en dit wordt als relatief hoog
beschouwd in vergelijking tot geïrrigeerde systemen in andere landen wereldwijd
waar dezelfde methodologie werd toegepast. Een gemiddeld hogere waarde werd
echter gevonden in de Nijl Delta in Egypte (1,52 kg m-3), Kings County (Californië,
VS) (1,44 kg m-3) en in Oldambt, Nederland (1,39 kg m-3). De ruimtelijk variëteit van
water productiviteit, gemeten aan de hand van de Coëfficiënt van Variatie (CV), is
hoog binnen systemen met een gemiddelde lage water productiviteit. Dit kan worden
verklaard door onzekere beschikbaarheid van irrigatie water en door suboptimale
landbouw omstandigheden. De hoge coëfficiënt van variatie in deze gebieden in
vergelijking tot gebieden met een hoge water productiviteit geeft aan dat er
aanzienlijke mogelijkheden zijn voor verbetering. De gemiddelde gelegenheid tot
verbetering in de 8 systemen bedroeg 14%, hetgeen aangeeft dat 14% water kan
worden bespaard zonder dat de opbrengsten lager worden.
Het WATer PRODuctiviteit (WAPTRO) model werd ontwikkeld om de water
productiviteit op mondiale schaal te modelleren en in kaart te brengen. De
belangrijkste invoergegevens van WATPRO zijn afkomstige van globale
aardobservatie data sets en het model kan daarom op globale schaal worden toegepast.
Het model is een combinatie van het door Monteith opgezette theoretische raamwerk
om droge massa productie in planten te berekenen, en een energie balans model om de
actuele verdamping te schatten. Het werd aangetoond dat, door beide benaderingen te
combineren en een aantal versimpelingen toe te passen, twee parameters die moeilijk
ruimtelijk in kaart kunnen worden gebracht, kunnen worden weggelaten. Hierdoor
blijkt dat de water productiviteit kan worden geschat op basis van vier ruimtelijke
variabelen: de breedband oppervlaktereflectie, de vegetatie index NDVI, de
extraterrestriale straling en de luchttemperatuur.
Het WATPRO model werd allereerst toegepast voor 39 locaties waar de water
productiviteit van tarwe werd gemeten onder experimentele condities. De correlatie
tussen de gemeten en gemodelleerde water productiviteit was laag, en dit werd
voornamelijk toegeschreven aan verschillen in de schaal (veld versus pixel) en in de
periode waarin werd gemeten en de periode die werd gemodelleerd. Een vergelijking
met metingen in velden van boeren die werden omringt door andere tarwevelden in
het Sirsa district in noordwest India, toonde een betere correlatie. Alhoewel het niet
een validatie betrof, toonde een vergelijking met SEBAL berekende water
productiviteit voor dezelfde regio in India aan dat WATPRO dezelfde ruimtelijk
spreiding in water productiviteit kan berekenen.
WATPRO werd vervolgens toegepast op mondiale schaal met behulp van globale data
sets van de NDVI en de oppervlakte reflectie om de gemiddelde condities van de
water productiviteit van tarwe vast te leggen aan het begin van dit millennium.
Tijdprofielen van de NDVI werden gebruikt om het groeiseizoen van de opkomst van
het gewas tot aan de oogst te bepalen voor iedere pixel. De WATPRO resultaten
werden vergeleken met de gemodelleerde resultaten van Liu et al. (2007) die het
GEPIC model toepasten voor tarwe op globale schaal, en met Chapagain and
Hoekstra (2004) die FAO statistieken gebruikten om de water productiviteit van tarwe
per land te bepalen. De vergelijking met Liu et al. liet een goede vergelijking zien
voor de meeste landen, terwijl de relatie met de Chapagain and Hoekstra minder goed
was. Voorts werd gevonden dat de water productiviteit varieert van 0,2 tot 1,6
kilogram geoogste tarwe per kubieke meter verbruikt water. Van de 10 grootste
producenten van tarwe, scoorden Frankrijk en Duitsland, de gemiddeld hoogste water
productiviteit van respectievelijk 1,42 and 1,35 kg m-3.
De mondiale patronen op de kaart van de water productiviteit werden vergeleken met
globale data sets van neerslag en referentie verdamping. Het doel hiervan was het
bepalen wat de invloed van het klimaat en de beschikbaarheid van water is op het
niveau van de water productiviteit. Zo bleek dat de hoogste waarden van water
productiviteit verwacht kunnen worden in gematigde klimaten met hoge neerslag. Als
gevolg van de non-lineaire relatie met de totale neerslag tijdens het seizoen, worden
de grootste verbeteringen kunnen worden verwacht in aride gebieden met lage
neerslag. Investeringen dienen te worden gemaakt in technieken die kunnen worden
toegepast voor het in-situ vasthouden van regenwater of het toedienen van
supplementair irrigatiewater, in combinatie met verbeterd agronomisch management
en het gebruik van (kunst)mest. Hierdoor kan een aanzienlijk verbetering worden
bewerkstelligd ten aanzien van de water productiviteit binnen stroomgebieden of
irrigatiesystemen. Informatie over de ruimtelijke patronen van de water productiviteit
binnen landen of stroomgebieden kan beslissingen ondersteunen over waar
investeringen kunnen worden gemaakt en welke maatregelen dienen te worden
geïmplementeerd om de productiviteit van water in de landbouw te verbeteren.
List of symbols
absorbed photosynthetical active radiation (MJ m-2)
above ground dry mater production (g m-2)
actual evapotranspiration (mm day-1)
maximum evapotranspiration (mm day-1)
APAR/PAR fraction (-)
soil heat flux (W m-2)
sensible heat flux (W m-2)
harvest index (-)
applied irrigation water (mm)
crop yield response factors (-)
normalized difference vegetation index (-)
photosynthetical active radiation (MJ m-2)
aerodynamic resistance to water vapour transport (s m-1)
bulk surface resistance (s m-1)
minimum bulk surface resistance (s m-1)
net radiation (W m-2)
slope of the saturated vapour pressure curve (mbar K-1)
extraterrestrial radiation (W m-2)
incoming shortwave radiation (W m-2)
actual transpiration (mm day-1)
maximum transpiration (mm day-1)
monthly average air temperature (oC)
Tmon for period with maximum leaf area index (oC)
water scalar (-)
water productivity (kg m-3)
harvestable/marketable yield (kg m-2)
maximum harvestable/marketable yield (kg m-2)
broadband surface albedo (-)
light use efficiency (g MJ-1)
ε max
maximum light use efficiency (g MJ-1)
θ grain
grain water content fraction (-)
latent heat flux (W m-2)
evaporative fraction (-)
τ SW
atmospheric transmissivity (-)
psychometric constant (mbar K-1)
stomatal response to ambient temperature (-)
fraction (-)
Acknowledgements................................................................................... iii
Summary .....................................................................................................v
Samenvatting............................................................................................. ix
List of symbols........................................................................................ xiii
Contents ....................................................................................................xv
Introduction ..........................................................................................1
Food and water in a changing world..............................................................1
The scientific approach on food and water ....................................................2
The contribution of this research ...................................................................6
Thesis outline .................................................................................................7
Review of measured water productivity values for irrigated wheat,
rice, cotton and maize...........................................................................9
2.1 Introduction....................................................................................................9
2.2 Results..........................................................................................................11
2.3 Discussion ....................................................................................................15
2.4 Conclusions..................................................................................................20
Appendix 1: Summarized water productivity values from literature ...................21
SEBAL for detecting spatial variation of water productivity and
scope for improvement in eight irrigated wheat systems...................23
WATPRO: A remote sensing based model for mapping water
productivity of wheat..........................................................................39
Materials and methods .................................................................................24
Results and discussion .................................................................................29
Water productivity model ............................................................................41
Model sensitivity and performance..............................................................45
Results and discussion .................................................................................47
A global benchmark map of water productivity for rainfed and
irrigated wheat ....................................................................................57
5.1 Introduction..................................................................................................57
5.2 Materials and methods .................................................................................59
5.3 Results and discussion .................................................................................66
5.4 Conclusions..................................................................................................73
Appendix 1: Harvest index, Hi .............................................................................76
Appendix 2: Country average WPET .....................................................................77
Discussion and conclusions................................................................79
The current levels of water productivity......................................................79
Defining the scope for improvement ...........................................................82
Explaining the variation and options for improvement ...............................83
Final considerations .....................................................................................87
Curriculum Vitae ....................................................................................101
1 Introduction
1.1 Food and water in a changing world
In recent decades the world’s human population has shown tremendous growth from
an estimated 2.5 billion in the 1950’s to approximately 6.7 billion to date. The
continents where the majority of this growth has taken place include Asia, Africa and
Latin America. This growth is expected to continue: it is predicted that the global
population may even reach 9.1 billion by 2050 (United Nations Population Division,
2009). Although the growth rates have diminished in many countries, strong growth
will continue in developing countries located in sub-Saharan Africa,. A major
challenge for the coming years is to provide a secure food supply to all newcomers. It
is believed that currently around 850 million people are already undernourished, and
the demand for food is growing. Due to increased welfare, people are changing to
more nutritious diets, and therefore the demand for food is growing even faster than
the growth in population. Food security is at stake and international organisation such
as the Food and Agriculture Organization (FAO), the World Bank and the United
Nations are calling for action. Recently the FAO argued that global food production
will have to increase 70 percent for an additional 2.3 billion people by 2050 (FAO,
2009). Moreover, the effects of climate change, such as rising temperatures and more
erratic rainfall patterns, and the recent focus on biofuel production both represent
major risks for long-term food security and water availability (De Fraiture et al.,
2008). The latter issue emerged, for example, in 2008 when food prices sharply
increased by 50%, partially as a result of the competing demands for agricultural
lands for biofuels, resulting in protests and riots throughout the world.
By 1798, it was already predicted by British economist Malthus that the world would
face a food crisis. In his theory on hunger, Malthus predicted an exponential growth
of the population, whereas the food supply would only grow arithmetically (Malthus,
1798). At a certain point in history, population growth would outpace food
production, and the world would be swept by hunger and stricken by wars. Malthus'
theory has, however, yet to come true. So far, on a global scale, the world's population
has increased at a tremendous rate as predicted by Malthus, but technological
advances have likewise increased food production. These technological advances
have included the widespread use of artificial fertilizers, breeding efforts to develop
high yielding hybrid crop varieties, large scale development of irrigation systems, and
the increasing use of machines in the production of food. This so called “Green
Revolution”, which started in the 1950s in Mexico and rolled out around the globe in
the decades thereafter, was a massive, coordinated effort to transfer these latest
advances in agricultural technology from developed countries to the developing
world. It resulted in a strong increase in food production. Today this process continues
in the genetic modification of crops to reduce the risk of failure, though it is not
believed that this development will impact water productivity significantly. The
success of the Green Revolution in the past decades depended on ample amounts of
fresh water and arable land, both of which are now in short supply. Agricultural lands
are degrading due to salinization and erosion, and urbanization also claims fertile
lands. Most mega-cities are located on alluvial plains with highly fertile soils which
are considered best suited to agricultural production.
That fresh water resources are not infinite is clearly demonstrated in river basins
where, through increased water withdrawals for the expansion of irrigated agricultural
areas, rivers fail to reach the sea, i.e. closed basins. Typical issues in such closed
basins are environmental degradation (water quality reduction, loss of biodiversity),
declining ground water tables, intrusion of seawater in estuaries and aquifers, and
deterioration of the ecological state of wetlands (Molle et al., 2010). River discharges
have dropped significantly in many basins, and insufficient water is available to meet
the competing demands from various other users. Industries and the tourist sector are
demanding more water, and growing populations require more water for domestic use.
The production of food in agricultural systems, whether in rainfed or irrigated areas,
takes water from the system that is not available for later reuse. Water disappears into
the air through evaporation from the surface and transpiration from plants. It is
estimated that approximately 80% of the global evapotranspiration budget comes
from rainfed areas, whereas the remaining 20% comes from irrigated agriculture (De
Fraiture and Wichelns, 2010). To supply water to agricultural fields for the
evapotranspiration process, water is diverted from rivers, pumped from groundwater
reservoirs, or harvested from the rain. Excess water infiltrates the soil and returns to
the system where it may be available for reuse (Perry, 2007). It is estimated that
globally agriculture accounts for approximately 70% of total water diversions
(Comprehensive Assessment of Water Management in Agriculture, 2007).
In the context of a changing climate, a growing population, an affected ecology and
increasing competition for water, it is therefore unlikely that agriculture can secure a
larger share of the already highly exploited fresh water resources. With the limits of
the Green Revolution being reached, and the fresh water resources unsustainably
exploited, international research and development organizations are opting to increase
the productivity of water in agriculture to sustain and improve food security for the
coming generations. This strategy is more popularly stated: to produce more crop per
drop (Kijne et al., 2003). In a broader sense, increasing the productivity of water
means getting more value from each drop of water. Water may be used for growing
crops, but also for cultivating fish, keeping livestock or for forestry. With agriculture
being the largest consumer of water, the largest gains in water production are
expected to be made in this sector. Questions that are raised are whether we can save
water for other users while maintaining food production, or whether we can increase
food production from the same amount of water (Postel, 1998).
1.2 The scientific approach on food and water
The relationship between agricultural production and water consumption through
evapotranspiration is complex. It is affected by numerous growing conditions, such as
climate, agronomic practices, soil type and fertility, and crosses scales varying from
individual plants to farmer fields, river basins, nations and the global level. Since the
1900’s the food production-water consumption relationship has been investigated by
scientists from different backgrounds and with different interests. As a result of these
different points of views by scientists or engineers, and the different scales of
application, many definitions of water productivity exist in scientific publications.
The water consumption and the production parts of the water productivity function are
therefore defined in several ways (Molden et al., 2003).
Plant physiologists and breeders have analysed photosynthesis or dry matter
production in relation to the plant’s transpiration, which can be considered the true
water consumption for production. However, at field level it is inevitable that water is
also lost though the evaporation process from soil. Soil and crop scientists therefore
commonly define evapotranspiration as water consumption, and express crop
production as harvestable yields of grains or fruits, for example. At farm level,
farmers aim at maximizing or optimizing the agricultural output, defined as total
harvestable yield or economic profit. Agricultural engineers and economists define
water productivity at farm level in terms of economic benefit in relation to
evapotranspiration or irrigation water supply. Similarly irrigation engineers consider
water deliveries, or water depletion and the available water at the irrigation system
level to evaluate the economic benefit of water diversions. Numerous irrigation
indicators are available to evaluate the system’s conveyance, distribution and
applications of water to fields. These relate the total crop water use or the beneficial
crop water use to water availability from irrigation water diversions and/or from
(effective) precipitation (Bos et al., 2005).
In the beginning of the 20th century agricultural scientists from the United States
started to look at the relationship between water use and dry matter production.
Calculation of evapotranspiration in field experiments proved to be quite unreliable
since certain components of the water balance could not be determined at all, or could
only be estimated roughly. Most experiments at that time were conducted in pots, and
by covering the soil surface, transpiration could be determined with greater certainty.
Pioneering work was conducted by Briggs and Shantz (1913) who determined for
lucerne a transpiration ratio, defined as the amount of water required to grow a certain
dry weight of crop. One of the conclusions drawn by many and summarized by De
Wit (1958), based on a synthesis of experimental results, was that solar radiation
played a dominant role in determining the levels of both yield and transpiration,
especially when water is non-limiting. Similar conclusions were drawn by Stanhill
(1960) who plotted linear relations between cumulative dry matter production and
cumulative evapotranspiration of grass grown at different latitudes. The highest
slopes, and thus the highest water use efficiencies, were found in locations at higher
latitudes (Denmark, Netherlands, England), and the lowest ones in Israel and
With the development of new and better equipment, such as climate-controlled glass
houses and electronic equipment, more accurate measurements could be carried out.
Bierhuizen and Slatyer (1965) conducted experiments on cotton leaves where
airstreams with fixed temperature, humidity and CO2 concentrations were passed
through a leaf chamber. Photosynthesis and transpiration were measured as the
difference in CO2 and water vapour concentrations of air before and after passing
through the leaf chamber. Using this experimental setup, the transpiration efficiency
under different levels of air temperature, wind speed, CO2 concentration and light
intensities could be determined with higher accuracy. They were the first to claim and
prove that transpiration and photosynthesis (and thus the transpiration efficiency)
were more controlled by evaporative demand from the air, expressed as the vapour
pressure deficit, than by radiation regimes or by latitude as claimed by De Wit (1958)
or Stanhill (1960). This conclusion was later confirmed in a thorough review by
Tanner and Sinclair (1983) who defined the water productivity relation as the
transpiration efficiency which is the reciprocal of the transpiration ratio.
With the Green Revolution at its peak, numerous programmes were set up at
universities and national research organizations to determine the optimal growing
conditions for maximizing crop yields in farmer’s fields. Whereas most experimental
results from the first half of the 20th century originated from the western countries, the
focus shifted to the developing countries in the later decades. International research
organizations were established with large campuses to develop new crop varieties,
make them available to the local farmers, and to provide optimal irrigation and
fertilizer application strategies applicable to local conditions. Examples are the
International Maize and Wheat Improvement Center (CIMMYT) in Mexico, the
International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) in India,
and the International Rice Research Institute (IRRI) in the Philippines.
With water resources being abundantly available in most new or expanding irrigation
systems, research focussed on maximizing crop yields for farmers by meeting the
maximum crop water demands. Several models were developed that describe the
relation between crop production and water use, with the purpose of determining the
effect of crop water stress on yields. For example Hanks (1974) linearly relates yields
(Yact) to transpiration (Tact), with the maximum attainable yields (Ymax) under
maximum transpiration (Tmax):
 Yact   Tact 
 = 
 Ymax   Tmax 
FAO research paper 33 on yield response to water (Doorenbos and Kassam, 1979)
provided a simple method to assess the impact of crop water stress on yield reduction
for more than 25 crops. Water stress is determined as the difference between the
actual evapotranspiration (ETact) and the evapotranspiration when crop requirements
are met (ETmax). These are linearly related to crop yield (Yact) under certain conditions,
and maximum yields (Ymax) under optimal conditions (Stewart et al., 1977):
Y 
ETact 
1 − act  = k y 1 −
max 
max 
where ky is the crop-dependent empirical yield response factor. A major drawback of
both frameworks is the need to estimate ETmax, Tmax and Ymax. These are difficult to
estimate under actual field conditions and agronomic management conditions. In
many publications, ETmax is considered equal to the potential evapotranspiration
(ETpot), whereas Ymax is usually considered as the maximum yield obtained in an
experimental set-up. This FAO33 method, which assumes a linear relationship
between ETact and Yact, is, however, still commonly applied in irrigation design and
operation, and referred to in scientific literature (see e.g Istanbulluoglu, 2009; Abou
Kheira, 2009; Payero et al., 2009). The FAO33 publication was, however, the first to
provide average water productivity values, defined as ‘water utilization efficiency for
harvested yield’ or Yact divided by ETact , for more than 25 different crops.
In recent decades water productivity (or water use efficiency) has shifted from being a
by-product of the developed strategies for maximizing crop yields per unit land, to a
means to express the efficiency of plants or farmers to use scarce water resources for
food production. Growing pressure on fresh water resources gave a new direction to
the water productivity concept. Purposely reducing irrigation water applications and
stressing crops to achieve higher water productivity (deficit irrigation) was introduced
as a means of saving water (Fereres and Soriano, 2007). During particular growth
periods, crop water stress is induced, thereby reducing final crop yields, but also
reducing evapotranspiration and increasing water productivity. This is an entirely
different approach to earlier high intensity irrigation strategies based on entirely
meeting crop water demands and keeping the root zone wet. Other agronomic
management options are also adopted to increase water productivity, such as using
plastic or mulch soil covers, optimizing planting distance, adjusting planting date and
growing season, soil tillage, and optimal fertilization rates.
As a response to these changing strategies, the FAO introduced the AquaCrop
toolbox* in 2009 as a revision of the previously mentioned FAO33 publication.
AquaCrop is a crop water productivity model that simulates the yield response to
water and that is particularly suited to function under water scarce conditions (Steduto
et al., 2009). The model simulates biomass production, which is converted into fresh
crop yields using a harvest index that is adjusted for water and heat stress during the
growing season. As shown by Steduto et al. (2007), biomass production shows a
remarkable linear relation with crop transpiration when normalised for the climatic
demand quantified by the reference evapotranspiration (preferred to the vapour
pressure deficit). This theory is used to assess climate adjusted water productivities.
Moreover, in line with the recent focus on beneficial and non-beneficial water use,
this model allows separation of beneficial transpiration from non-beneficial
Recent attention by scientists to water shortages and improving water productivity has
increasingly focussed on river basins, rather than on fields, farms or irrigation
districts. By considering river basins, policies, measures and interventions for water
savings at more local level can be placed in a wider context that accounts for impacts
on downstream water users, such as wetlands, irrigation systems or industries A
saving in diverted irrigation water at field level may not necessarily lead to more
water being available at basin level, since excess applied water usually returns to the
system for reuse. Exceptions occur when water quality deteriorates, or when water
flows to the sea or to evaporation ponds. These are called the sinks in a system
(Seckler, 1996). Measures for saving water in agriculture should therefore focus on
reducing the real losses in a basin, such as evapotranspiration from agricultural lands,
rather than on saving water diverted from rivers that can be later reused. This is
described by Keller and Keller (1995) as dry and wet water savings, where the latter is
a real water saving where more water becomes available to other users. At basin level,
water can thus be saved by reducing water losses to sinks or by reducing the pollution
of water. Measures are advocated where the output of agriculture per unit of water
evapotranspired is promoted, or in other words where the water productivity is
increased (Comprehensive Assessment of Water Management in Agriculture, 2007).
1.3 The contribution of this research
The importance of the water productivity concept for food security in a world where
water resources are rapidly being exhausted, has been outlined in the previous two
paragraphs. An uncountable number of publications has been published in which
experimental results on yields, water consumption and/or water productivity are
reported. These studies are conducted in small fields representing local conditions in
specific years under varying seasonal weather conditions. The impact of weather
conditions on water productivity levels, in particular the evaporative demand defined
by either the vapour pressure deficit or the reference evapotranspiration, is stressed by
many authors. However, the plausible ranges of water productivity on a global level
are poorly understood. The latest overview for various crop was presented in the
FAO33 publication, but it is realised that in recent decades crops have been improved,
and it is likely that the reported ranges have altered in a positive way. Moreover the
question can be raised whether a system is performing well with respect to water
productivity or whether improvements can still be achieved. This is particularly of
interest to governments, donors and water managers, who would like to know whether
a system is performing well, and whether there is scope for improvement. Farmers
also realize that lower water use in crops contributes to their own sustainability.
Measurements within a system are location dependent and may not represent the
overall system performance. Remote sensing, the use of satellite images and spatial
modelling have been tested and applied over the last thirty years to assess
evapotranspiration and yields. Time series analysis of satellite images allows accurate
spatial estimation of yields and evapotranspiration over large areas and at high spatial
detail. This information will be used to assess and benchmark water productivity and
the scope for improvement in various systems. On a global level such spatial
information is not available, but relevant to understanding where crops can be grown
most efficiently with regards to their climate, and where improvements are still
This research will therefore focus on the benchmarking of physical water
productivity, i.e. the amount of agricultural production (harvestable yield) that can be
attained per unit of water evapotranspired, rather than on the economic water
productivity where the production term is replaced by a revenue or profit. The major
goal of this research is to benchmark water productivity values globally and at various
scales (field level, system level and global level), which at the start of this thesis was
unavailable. Remote sensing and modelling are the major tools applied for this work
to assess the spatial variation of water productivity of wheat at system and global
level, and to provide a first explanation for the differences that are found. A review of
the literature sources that provide measurements of water productivity is conducted to
assess plausible ranges of water productivity levels for wheat, maize, cotton and rice.
1.4 Thesis outline
The research in this thesis follows the steps in scale that were described before. The
next chapter provides a literature review of experiments where water productivity of
wheat, rice, cotton and maize, the major staple crops, were measured. These results
were consolidated in a database, and current benchmark levels of water productivity
for the four crops were established. An initial explanation of the large variation found
in the experimental results was provided by linking water productivity to climate
conditions and to variations in the applied amounts of irrigation water and fertilizers.
The large variation also underlines the necessity of mapping water productivity at a
regional scale, but at high resolution. The third chapter provides a methodology that
allows mapping of water productivity using the SEBAL algorithm applied to low and
high resolution satellite imagery. This methodology was tested and validated in the
wheat dominated Yaqui Valley region in north-western Mexico, and thereafter applied
in seven other wheat regions in the Netherlands, Pakistan, India, China, Egypt and the
United States. A statistical method based on the coefficient of variation was used to
determine the scope for improvement in water productivity with the purpose of
quantifying the potential for water savings. Finally, the thesis moves up to the global
level in chapters 4 and 5 with the purpose of benchmarking the water productivity of
wheat globally at a high spatial resolution with the use of remote sensing data sets.
The WATer PROductivity (WATPRO) model is developed, based on the same
principles underlying the evapotranspiration and biomass production modules in
SEBAL. The development of the model, the inherent simplifications and assumption
that were required, and the validation are outlined in chapter 4. Chapter 5 describes
the application of the WATPRO model at a global scale to benchmark wheat water
productivity. The resulting map was compared with earlier water productivity
modelling efforts from two different sources. Variations in water productivity were
attributed to variations in seasonal precipitation from standard TRMM products, and
to differences in seasonal reference evapotranspiration. This thesis ends with
conclusions on water productivity benchmark values at different scales synthesized
from the previous chapters.
2 Review of measured water
productivity values for irrigated
wheat, rice, cotton and maize*
2.1 Introduction
With a rapidly growing world population, the pressure on limited fresh water
resources increases. Irrigated agriculture is the largest water-consuming sector and it
faces competing demands from other sectors, such as the industrial and the domestic
sectors. With an increasing population and less water available for agricultural
production, the food security for future generations is at stake. The agricultural sector
faces the challenge to produce more food with less water by increasing crop water
productivity (see Kijne et al., 2003a for a review). A higher water productivity results
in either the same production from less water resources, or a higher production from
the same water resources, so this is of direct benefit for other water users. In this study
water productivity (WPET in kg m-3), which is originally referred to in literature as
'water use efficiency', is defined as the marketable crop yield over actual
(kg m-3)
where Yact is the actual marketable crop yield (kg ha-1) and ETact is the seasonal crop
water consumption by actual evapotranspiration (m3 ha-1). When considering this
relation from a physical point of view, one should consider transpiration only. The
portioning of evapotranspiration in evaporation and transpiration in field experiments
is, however, difficult and therefore not a practical solution. Moreover, evaporation is
always a component related to crop specific growth, tillage and water management
practices, and this water is no longer available for other usage or reuse in the basin.
Since evapotranspiration is based on root water uptake, supplies from rainfall,
irrigation and capillary rise are integrated.
This chapter is published as “SJ Zwart, WGM Bastiaanssen, 2004. Review of measured crop water
productivity values for irrigated wheat, rice, cotton and maize. Agricultural Water Management 69(2),
pp. 115-133”.
Despite that water productivity is a key element in longer-term and strategic water
resources planning, the actual and practically feasible values are hardly understood.
The most complete international work so far is compiled by Doorenbos and Kassam
(1979), who used crop yield response factors (ky) for relating ETact to Yact. The
problem with the standard 'FAO33-approach' is that the maximum yield ought to be
known, which differs for given cultural practices. This implies that Yact = f(ky, Ypot,
ETact, ETmax) is not straightforward, although it is often applied in absence of
alternative expressions.
Kijne et al. (2003b) provide several strategies for enhancement of water productivity
by integrating varietal improvement and better resources management at plant level,
field level and agro-climatic level. Examples of options and practices that can be
taken are: increasing the harvest index, improving drought tolerance and salinity
tolerance (plant level), applying deficit irrigation, adjusting the planting dates and
tillage to reduce evaporation and to increase infiltration (field level), water reuse and
spatial analysis for maximum production and minimum ETact (agro-ecological level),
to mention a few.
Due to agronomical research (e.g. plant breeding) and improved land and water
management practices, water productivity has increased during the years. For example
Grismer (2002) conducted a study on water productivity values for irrigated cotton in
Arizona and California and concluded that water productivity values exceed the range
given by Doorenbos and Kassam (1979) in many cases. In rice production water
productivity increased due to shorter growing periods (Tuong, 1999) and due to
increase in the ratio of photosynthesis to transpiration (Peng et al., 1998). It is likely
that water productivity for other crops has changed significantly as well.
Various studies have researched water use and yield relationship of specific crops, on
specific locations, with specific cultural and water management practices. The current
investigation summarizes the results of field experiments that have been conducted
over the last 25 years and tries to find a range of plausible values for four major staple
crops: wheat (Triticum aestivum L.), rice (Oryza sativa L.), cotton (Gossypium spp.)
and maize (Zea mays L.). The second objective of this paper is to find some first order
explanatory variables for the global scale water productivity differences found.
Database and terminology
A database is established with water productivity data collected from field
experiments that were reported in the international literature, conference proceedings
and technical reports. The majority of field experiments was conducted at
experimental stations under varying growing conditions, including variations in
climate, irrigation, fertilization, soils, cultural practices, etc. As the purpose of this
research is to find plausible water productivity ranges under farm management
conditions, all measured water productivity values of an experiment are included in
the database.
To be included in the database, the results of the experiments should provide
minimally the total seasonal measured actual evapotranspiration (ETact), the method
applied to determine ETact and the crop yield, Yact. Most studies do not measure ETact
– 10 –
and use the potential evapotranspiration (ETpot) instead. These studies are not
incorporated into the database and, hence, not used and discussed in this paper.
Results from greenhouse experiments, pot experiments and water balance simulation
models were excluded. Also, experiments based on the reference evapotranspiration
method (Allen et al., 1998) has not been regarded as being suitable for the current
review; evapotranspiration is not measured, but estimated.
Lysimeters are a common instrument for determining ETact. The soil water balance
methods that monitor soil water content during the growing season by measurements
of gravimetric soil moisture, or by neutron scattering equipment (neutron probes) or
by Time-Domain-Reflectometry (TDR), is also often used. Micro-meteorological in
situ flux measurement techniques, such as the Bowen ratio and eddy-correlation
methods are not common for agronomical studies (they are mainly used for micrometeorological and climate studies in which yield is not reported).
Yield is defined as the marketable part of the total above ground biomass production;
for wheat, maize and rice total grain yield is considered, and for cotton the total lint
yield and/or total seed yield. Unfortunately, very few sources give the moisture
content at which the yield was measured, which inevitably means an error exists in the
final results. Siddique et al. (1990) investigated water productivity of old and new
wheat cultivars and found that older cultivars have lower water productivity values
due to lower harvest index. No significant difference in total biomass production
between the old and new cultivars was found. For example in rice production water
productivity increased throughout the years due to developments in the new plants
types with a higher ratio of photosynthesis to transpiration and due to a decrease in
growth period (Peng et al., 1998; Tuong, 1999). Thus, experiments with results older
than approximately 25 years are excluded to minimize the influence of older varieties
with lower harvest index and longer growth period.
The results of experiments were first re-organized into a crop-wise database, that
includes latitude/longitude, country, location, ETact, Yact, biomass production, harvest
index, experimental year(s) and reference. Some of the references cited provide the
results of each field experiments, while others give averages for e.g. each
experimental year or each management strategy applied. Each value, whether it is
reported as an average of more experiments or a unique value for one experiment, is
considered as one value in the database.
2.2 Results
An overview of the contents of the database is given in Table 2-1, while appendix 1
depicts all results by crop and by source. A total of 84 publications was included. For
wheat, 28 data sources across 13 countries on 5 continents were analysed. Data on
rice is with 13 sources across 8 countries remarkably less. Many studies on rice
production and water use were found to focus on irrigation water inputs, while few
consider actual evapotranspiration (ETact). For cotton, 16 experiments conducted in 9
different countries were found, while maize had 27 sources in 10 different countries
on 4 continents. Research on water productivity of maize is concentrated mainly in
– 11 –
the USA (9 sources) and China (7 sources). Although the literature search was
conducted in the Spanish and French language as well, few publications that meet the
minimal data demands for all four crops could be found for the African, Latin
American and European continents. Unfortunately many publications focus on either
determination of crop water use or crop yields, whereas others only consider irrigation
water applied.
Table 2-1: Summary of the database.
# publications
# continents
# countries*
Water productivity
Figure 2-1a-d depict the frequency distribution histograms of wheat, rice, cotton and
maize. For the purpose to exclude extreme values, the water productivity range is
determined by taking the 5 and 95 percentiles of the cumulative frequency
distribution. The results are presented in Table 2-2.
Wheat has the largest number of experimental points (n = 412) and the WPET range is
between 0.6 and 1.7 kg m-3. Doorenbos and Kassam (1979) give a lower range of 0.8
– 1.0 kg m-3 (see Table 2-2). The maximum values are found by Jin et al. (1999) in
China: application of manure led to higher production and straw mulching improved
soil water and soil temperature conditions. WPET for the experiment with straw
mulching was 2.67 kg m-3 and 2.41 kg m-3 for a combination of straw mulching and
manure. ETact in the winter season was tempered to 268 and 236 mm respectively,
while yields were relatively high with 7,150 and 5,707 kg ha-1 (see Figure 2-2a).
WPET of rice ranges between 0.6 and 1.6 kg m-3 (Figure 2-1b). Tuong and Bouman
(2003) give a very similar range of 0.4 –1.6 kg m-3 for lowland rice conditions. The
maximum WPET value of 1.1 kg m-3 for rice given by Doorenbos and Kassam (1979)
(Table 2-2) is exceeded in six out of thirteen data sources. The WPET range of rice is
similar to wheat ; the shape of the frequency distribution of rice is not as smooth as
for wheat because less points are available. The maximum values go up to 2.20 kg m-3
and were measured in China on alternate wetting and drying rice plots (Dong et al.,
2001). Rice grain yields of over ten tons per hectare were amongst the highest
measured, whereas ETact was on the lower side with 465 mm (Figure 2-2b).
WPET values of cotton lint yield range from 0.14 to 0.33 kg m-3. The maximum values
exceed 0.35 kg m-3 and are found by Jin et al. (1999) and Saranga et al. (1998) in
China and Israel, respectively. Jin et al. (1999) conducted experiments in which
cotton was planted in furrows and the soil covered with plastic leaving holes for
infiltration near the plants, thus reducing soil evaporation and improving soil water
status of the root zone. Saranga et al. (1998) measured average lint yield values of
1,300 kg ha-1 in a field trial with deficit irrigation, while seasonal ETact was very low
with 390 mm (see Figure 2-2c). Howell et al. (1984) measured similar values (0.33 kg
m-3) in an experiment with high frequency trickle irrigation and reduced water deficits
– 12 –
management for narrow row cotton in California (USA). Lint yield was more than
2,000 kg m-3, while seasonal ETact was relatively low (617 mm). The range for cotton
seed yield is with 0.41-0.95 kg m-3 higher than the range given in FAO33 (0.4-0.6 kg
m-3). In Argentina maximum values were measured exceeding 1.0 kg m-3 in
experiments where water was applied during critical periods such as pre-seeding and
flowering (Prieto and Angueira, 1999). Cotton seed yields did not differ compared to
other treatments, though ETact was lower (447-495 mm – see also Figure 2-2c).
Finally, maize WPET values were measured ranging from 0.22 up to a maximum of
3.99 kg m-3 (Figure 2-1d) which exhibits a large range of variation (CV=0.38). In 67
per cent of the publications the maximum value of the source exceeds the value of 1.6
kg m-3 provided by FAO33. The WPET range of 1.1-2.7 kg m-3 for maize, a C4-crop, is
significantly higher than wheat, rice and cotton, which are C3-crops. The maximum
values were measured by Kang et al. (2000b) in a combination of alternate furrow
irrigation and deficit irrigation experiments under Chinese conditions: low amounts of
irrigation water were alternately applied to one of the two neighbouring furrows. ETact
was with 226 mm very low, whereas grain yield was still 9,058 kg ha-1 (Figure 2-2d).
kg m-3
kg m-3
kg m-3
kg m-3
kg m-3
kg m-3
kg m-3
0.8 – 1.0
0.6 – 1.7
0.7 – 1.1
0.6 – 1.6
Cotton (seed yield)
0.4 – 0.6 0.41 – 0.95
Cotton (lint yield)
not given 0.14 – 0.33
0.8 – 1.6
1.1 – 2.7
* defined as the 5 and 95 percentiles of the entire range
WPET -range*
(this research)
WPET -range
Table 2-2: Water productivity (WPET) benchmark values per unit of water depletion according to
"FAO33" (Doorenbos and Kassam, 1979), WPET ranges according to this study, the maximum,
minimum, mean and median WPET values and the standard deviation (SD) and coefficient of variation
(CV) of the data sets by crop.
– 13 –
b: rice (n = 105)
– 14 –
water productivity, WP ET (kg m-3)
a: wheat (n = 412)
water productivity, WP ET (kg m-3)
frequency (no. of experiments)
frequency (no. of experiments)
– 15 –
water productivity, WP ET (kg m-3)
d: maize (n =233)
Figure 2-1: Frequency of water productivity (WPET) per unit water depletion for wheat, rice, cotton
and maize.
2.3 Discussion
In Figure 2-2a-d, the yield is plotted against the ETact for each of the four crops. All
four graphs show that the Yact-ETact relation is not as straightforward as often is
assumed: r-squared values are low: cottonlint has the highest correlation (r2=0.39),
> 1.000
frequency (no. of experiments)
frequency (no. of experiments)
lint yield
seed yield
water productivity, WP ET (kg m )
c: cotton (nseed = 126, nlint = 66)
followed by wheat (r2=0.35), maize (r2=0.33), cottonseed (r2=0.19) and rice (r2=0.09).
The lesson learnt here is that Yact (ETact) functions are only locally valid and cannot be
used in macro-scale planning of agricultural water management. A broad range in
water productivity values for all four crops exists (see Table 2-2), which is caused by
the many factors that influence the soil-plant-water relationship. In a search for first
order explanations for the wide ranges in water productivity, only three aspects are
discussed here: climate, irrigation water management and soil management.
R2 = 0.35
yield, Yact (kg ha )
yield, Yact (kg ha )
R2 = 0.09
actual evapotranspiration, ET act (mm)
a: wheat
actual evapotranspiration, ET act (mm)
b: rice
R2 = 0.33
R2 = 0.40
yield, Yact (kg ha )
R2 = 0.19
yield, Yact (kg ha )
actual evapotranspiration, ET act (mm)
c: cotton
actual evapotranspiration, ET act (mm)
d: maize
Figure 2-2: yield-evapotranspiration relations of wheat, rice, cotton and maize.
De Wit (1958) was among the first to describe the photosynthesis-transpiration
relationship. Bierhuizen and Slayter (1965) researched the influence of climatic
– 16 –
parameters on this relationship and found a proportionally inverse relation (reviewed
and confirmed by Tanner and Sinclair in 1983) between vapour pressure deficit of the
air and water productivity. Similar results were found by Stanhill (1960) for pastures
grown at different latitudes. As the vapour pressure deficit generally decreases when
moving away from the equator, water productivity is expected to increase with
increasing latitude. This proposition was tested for the current dataset: for each
experimental site (defined as each unique geographic location), the maximum water
productivity of each crop is plotted against the latitude value of the experimental site.
The maximum value is being taken to approach the optimal growing conditions with
respect to soil fertility management and irrigation water application at a certain
location. The result, depicted in Figure 2-3, confirms that water productivity decreases
with lower latitude. It also shows that the highest water productivity values occur
between 30 and 40 degrees latitude where a factor 2 to 3 difference in water
productivity of wheat, rice and maize is detected when compared to areas between 1020 degrees.
maximum water producitivity, WPET (kg m )
cotton - seed
cotton - lint
latitude (decimal degrees)
Figure 2-3: Relation between latitude and maximum water productivity (WPET) value per unit water
depletion per location and per crop (both northern and southern latitude are considered positive).
Many examples from literature describe the influence of irrigation water management
on water productivity (e.g. Oktem et al., 2003; Zhang et al., 1998; Yazar et al., 2002a;
Kang et al., 2000a; Sharma et al., 1990). Deficit irrigation practices have been
researched to quantify the effect on yield and to find optimum water productivity
values. In Figure 2-4a and b, water productivity of wheat and maize are plotted
against the net amount of irrigation water applied in various experiments. It was found
that without irrigation water productivity in rainfed systems is low, but that water
productivity rapidly increases when a little irrigation water is applied. According to
the database, optimum values for water productivity are reached at approximately 150
– 17 –
and 280 mm of irrigation water applied for wheat and maize respectively (in addition
to rainfall). Figure 2-4 demonstrates how water productivity can be increased while
simultaneously saving water by reduced irrigations. A maximum water productivity
will often not coincide with farmers' interests, whose aim is a maximum land
productivity or economic profitability. It requires a shift in irrigation science,
irrigation water management and basin water allocation to move away from
'maximum irrigation-maximum yield' strategies to 'less irrigation-maximum water
productivity' policies. Besides the total amount of irrigation water applied, the timing
of irrigation is important. Water stress during different growth stages affect water
productivity differently; lower water productivity was measured in cotton experiments
where water stress occurred during vegetative and early bud formation periods. Gentle
stress during yield formation did not affect yield production, but reduced vegetative
growth and would thus improve water productivity (Prieto and Angueira, 1999).
WPET (kg m )
maize - Turkey Oktem et al. 2002
maize - India Mishra et al. 2001
wheat - USA Al Kaisi et al. 1997
wheat - China Zhang et al. 1999
seasonal applied irrigation water, I (mm)
Figure 2-4: Relation between amount of irrigation water applied (I) and measured water productivity
(WPET) per unit water depletion for four wheat and maize experiments.
The relationship between irrigation and water productivity in rice is not the same as
found for wheat and maize. In rice cultivation, instead of traditional continuous
flooding, other water management strategies, such as alternate wetting and drying
(intermittent irrigation) and saturated soil culture, were researched. Analysis of
alternate wetting and drying experiments in India by Mishra et al. (1990) shows that,
although irrigation water is saved, there is no significant improvement in water
productivity, which remains between 0.80 and 0.99 kg m-3 (n=24). For this specific
study in India, the ETact was not reduced because irrigation application was in excess
of ETact. Dong et al. 2001 found similar results and concluded that there was no
significant difference between continuous flooding and alternate wetting and drying
experiments; ten year average ETact and water productivity amounted 590 and 591
mm and 1.49 and 1.58 kg m-3 for continuous flooding and intermittent irrigation
experiments, respectively. On the other hand, Shi et al. (2003) measured in lysimeter
– 18 –
experiments higher water productivity values for intermittent irrigation experiments
(2.0 kg m-3) compared with continuous flooding (1.6 kg m-3), whereas yields were
only 200 kg ha-1 lower). Moreover, ETact in the intermittent experiment (347 mm) was
22 per cent lower compared to continuous flooding. For the sake of clarity, Seckler
(1996) distinguishes "dry" and "wet" water savings: reduction in ETact is a wet saving
because the evapotranspired water is lost for future use in the basin. On the other hand
irrigation water savings are dry savings as the water may be recycled within the basin
for future use (unless it is polluted). As is shown by the results from Mishra et al.
(1990) and Dong et al. (2001) intermittent irrigation is merely an example of a dry
water saving as ETact is hardly affected by reduced supplies.
Hatfield et al. (2001) reviewed the effects of soil management on water productivity
by modification of the soil surface, such as tillage and mulching, and by improvement
in soil nutrient status by adding nitrogen and/or phosphorus. A modification of the
soil surface changes the processes of ETact and is often found to be positively related
to water productivity. Nutrients indirectly affect the physiological efficiency of the
plant. In Figure 2-5 the nitrogen rate is plotted against the water productivity of wheat
during studies in Niger, Syria and Uruguay. Water productivity increases when
nitrogen is applied and reaches an optimum at a rate of approximately 150 kg ha-1. On
the other hand Corbeels et al. (1998) and Fernandez et al. (1996) did not measure
significant differences when N fertilization was applied. Combined nutrient and
irrigation supply levels are more commonly researched (e.g. Li et al., 2001; Pandey et
al., 2001; Oweis et al., 2000; Zima Szalokine and Szaloki, 2002). Optimum values for
amount nutrient and irrigation water application can be found to maximize water
WPET (kg m-3)
Caviglia & Sadras 2001 - Uruguay
Pandey et al. 2001 - Niger
Oweis et al. 2000 - Syria
seasonal applied N (kg ha )
Figure 2-5: Relation between amount of nitrogen applied (N) and measured water productivity (WPET)
per unit water depletion for wheat from experiments in three different countries.
– 19 –
2.4 Conclusions
The water productivity ranges for the four crops investigated are large as indicated by
the high CV of 28–40% and are a logical consequence of the low correlation between
ETact and crop yield (r2=0.09 to 0.39). This variability was mainly ascribed to 1)
climate, 2) irrigation water management, and 3) soil (fertility) management, although
more explanatory variables prevail. The climatic belt between 30 to 40 degrees
latitude was found to be favourable for agriculture with regard to water productivity
and this is likely to be related to vapour pressure deficit. In areas with marginal soils,
application of fertilizer offers large possibilities for improvement of water
productivity. The increase in water productivity is highest if small amounts of
nitrogen (<80 kg ha-1) are applied. Deficit irrigation practices were found to improve
water productivity, sometimes even by more than 200 %. Plants are more efficient
with water when they are stressed. It is therefore tentatively concluded that to achieve
optimum water productivity in water short regions, it is wise to irrigate wheat and
maize with less water as recommended for attaining maximized yields.
In rice cultivation the increase of water productivity when less water was applied
could not be confirmed from the database; during many of the alternate wetting and
drying and continuous flooding experiments there was no significant difference in
water productivity. Water savings in rice are therefore a 'dry saving', because
consumptive use is not or little affected. The wide ranges in water productivity found
suggest that agricultural production can be maintained with 20-40% less water
resources provided that new water management practices are adopted.
– 20 –
Appendix 1: Summarized water productivity values from
Parana, Argentina
Merredin, Australia
Merredin & Mullewa, Australia
Benerpota, Bangladesh
Quzhou, China
Xifeng, China
Wangtong, China
Gansu, China
Luancheng, China
Yucheng, China
Beijing, China
various locations, China
Luancheng, China
West Bengal, India
Pantnagar, India
Uttar Pradesh, India
Karnal, India
Pantnagar, India
Gilat, Israel
Meknes, Morocco
Sidi El Aydi, Morocco
Konni, Niger
Faisalabad, Pakistan
Tel Hadya, Syria
Cukurova, Turkey
Yellow Jacket (CO), USA
Grand Valley (CO), USA
Tashkent, Uzbekistan
kg m-3
0.55 – 1.49
0.56 – 1.14
0.55 – 1.65
0.52 – 1.34
1.38 – 1.95
0.65 – 1.21
1.49 – 2.67
0.58 – 1.45
1.07 – 1.29
0.88 – 1.16
0.92 – 1.55
0.85 – 1.86
1.28 – 1.82
1.11 – 1.29
0.86 – 1.31
0.48 – 0.71
0.27 – 0.82
1.06 – 1.23
0.60 – 1.60
0.11 – 1.15
0.32 – 1.06
0.42 – 0.93
0.70 – 2.19
0.48 – 1.10
1.33 – 1.45
0.47 – 1.08
1.53 – 2.42
0.44 – 1.02
kg m-3
n experimental
1998 – 99
Caviglia & Sadras 2001
Siddique et al. 1990
1991 – 95
Regan et al. 1997
1988 – 92
Rahman et al. 1995
1988 – 89
Deju & Jingwen 1993
1988 – 91
Fengrui et al. 2000
1995 – 96
Jin et al. 1999
Li et al. 2001
1984 – 96
Wang et al. 2001
1986 – 90
Xianqun 1996
1991 – 95
Zhang et al. 1998
1982 – 95
Zhang et al. 1999
1998 – 00
Zhang et al. 2003
1989 – 91 Bandyopadhyay & Mallick, 2003
1983 – 85
Mishra et al. 1995
1993 – 94
Sharma et al. 2001
1986 – 88
Sharma et al. 1990
1979 – 85
Singh & Chauhan 1996
1977 – 87
Amir et al. 1991
1993 – 95
Corbeels et al. 1998
1995 – 99
Mrabet 2002
1996 – 98
Pandey et al. 2001
1991 – 94
Waheed et al. 1999
1991 – 96
Oweis et al. 2000
1991 – 92
Sezen and Yazar, 1996
1993 – 94
Al-Kaisi et al. 1997
1988 – 89
Kruse et al. 1991
2001 – 02
Kamilov et al. 2002
kg m-3
0.70 – 0.75
1.04 – 2.20
1.63 – 2.04
0.80 – 0.99
0.46 – 0.82
0.55 – 0.67
0.87 – 1.46
0.48 – 0.62
0.50 – 0.79
1.39 – 1.61
0.53 – 0.64
1.37 – 1.44
0.88 – 1.34
kg m-3
n experimental
1997 – 98
1991 – 00
1983 – 84
1979 – 83
1996 – 97
1988 – 94
1991 – 9-2
1989 – 91
1979 – 80
Echuca, Australia
Zhanghe, China
Nanchang, China
Pantnagar, India
Raipur, India
New Delhi, India
Punjab, India
Muda, Malaysia
Kadawa, Nigeria
Luzon, Philppines
N’Diaye, Pont-Gendarme, Senegal
Beaumont (TX), USA
Belle Glade (FL), USA
– 21 –
Bethune et al. 2001
Dong et al. 2001
Shi et al. 2003
Mishra et al. 1990
Sastri et al. 1985
Singh et al. 2002
Singh et al. 2001
Cabangon et al. 2002
Nwadukwe & Chude 1998
Bhuiyan et al. 1995
Raes et al. 1992
Roel et al. 1999
Shih et al. 1983
Santiago, Argentina*
various locations, Australia**
Wangtong, China**
Yucheng, China**
Be’eri, Israel**
Faisalabad, Pakistan*
Cordoba, Spain*
Bornova-Izmir, Turkey*
Cukurova, Turkey*
Harran Plain, Turkey**
Wellman (TX), USA**
Halfway (TX), USA**
Five points (CA), USA**
Five points (CA), USA**
Maricopa (AZ), USA**
Tashkent, Uzbekistan**
* seed yield
** lint yield
kg m-3
0.50 – 1.27
0.22 – 0.29
0.20 – 0.37
0.22 – 0.35
0.38 – 0.58
0.45 – 0.71
0.38 – 0.48
0.38 – 0.84
0.50 – 0.74
0.10 – 0.17
0.14 – 0.19
0.14 – 0.24
0.22 – 0.33
0.13 – 0.16
0.54 – 1.70
kg m-3
n experimental
1990 – 95
1996 – 99
1994 – 97
1991 – 94
1985 – 86
1993 – 94
1994 – 95
1992 – 95
1982 – 83
1993 – 94
2000 – 01
kg m-3
1.84 – 2.79
1.12 – 1.33
1.26 – 2.31
1.49 – 2.67
1.36 – 1.65
2.11 – 3.37
2.14 – 3.99
1.63 – 2.22
1.55 – 1.84
1.28 – 2.44
1.17 – 1.74
1.36 – 1.89
2.34 – 2.88
1.50 – 2.16
1.94 – 2.25
1.04 – 1.38
0.22 – 1.25
1.12 – 1.39
0.89 – 1.55
1.47 – 1.74
0.83 – 1.61
1.34 – 3.26
0.36 – 1.57
2.03 – 2.86
1.26 – 1.54
1.13 – 1.68
kg m-3
n experimental
1991 – 95
1989 – 91
1995 – 96
1996 – 97
1997 – 98
1978 – 95
1993 – 95
1991 – 94
1992 – 93
1998 – 99
1993 – 94
1989 – 94
1994 –97
1998 – 99
1989 – 91
1994 – 95
Prieto and Angueira 1999
Tennakoon & Milroy, 2003
Jin et al. 1999
Xianqun 1996
Saranga et al. 1998
Waheed et al. 1999
Orgaz et al. 1992
Anac et al. 1999
Ertek and Kanber 2001
Yazar et al. 2002b
Baumhardt & Lascano 1999
Bordovsky & Lyle 1996
Howell et al. 1984
Howell et al. 1987
Hunsaker et al. 1998
Kamilov et al. 2003
Azul, Argentina
Guaira, Brazil
Xifeng, China
Wangtong, China
Changwu, China
Changwu, China
Gansu province, China
Yucheng, China
Luancheng, China
Szarvas, Hungary
Pantnagar, India
Tal Amara, Lebanon
Fundulea, Romania
Sevilla, Spain
Harran plain, Turkey
Harran plain, Turkey
Cukurova, Turkey
Bushland (TX), USA
Bushland (TX), USA
Bushland (TX), USA
Garden City (KS), USA
Blacksburg (VA), USA
Carolina Bays (SC), USA
Oakes (ND), USA
Bushland (TX), USA
Bushland (TX), USA
– 22 –
Navarro Dujmovich et al. 1996
Libardi et al. 1999
Fengrui et al. 2000
Jin et al. 1999
Liu & Li 1995
Kang et al. 2000a
Kang et al. 2000b
Xianqun 1996
Zhang et al. 2003
Zima and Szaloki 2002
Mishra et al. 2001
Karam et al. 2003
Cracium and Cracium 1999
Fernandez et al. 1996
Yazar et al. 2002a
Oktem et al. 2003
Gencoglan and Yazar 1999
Evett et al. 1996
Howell et al. 1995
Howell et al. 1996
Norwood 2000
Roygard et al. 2002
Sadler et al. 2000
Steele et al. 1994
Tolk et al. 1999
Yazar et al. 1999
3 SEBAL for detecting spatial variation
of water productivity and scope for
improvement in eight irrigated wheat
3.1 Introduction
Due to the rapid growth in world population, the pressure on water resources is
increasing (Rijsberman, 2006). In the future less water will be available for
agricultural production due to competition with the industrial and domestic sectors,
while at the same time food production must be increased to feed the growing
population. In systems where water is becoming the limiting factor, agricultural
production should be expressed per unit of water consumed instead of production
expressed per unit land. It is inevitable that the production per unit water consumed,
the water productivity, must be increased to meet this challenge (see e.g. Kijne et al.,
2003; Molden et al., 2007). Spatial information on water use, crop production and
water productivity will play a vital role for water managers to assess where scarce
water resources are wasted and where in a given region the water productivity can be
Currently, information on water productivity is often only available from experiments
on a single field, so that results are limited to the local (environmental) conditions that
can vary from year-to-year and to the specific soil, crop and water management
practices . Crop water consumption cannot be routinely measured, and this hampers
the introduction of the concept of crop water productivity per unit of water depletion
in policy making and water management. Although water productivity is gaining
importance for evaluating good management practices, there is no standardized
framework that aids the calculations. Remote sensing can help to quantify water
productivity spatially and for large areas.
The Surface Energy Balance Algorithm for Land (SEBAL) (Bastiaanssen et al., 2002;
Bastiaanssen et al., 2005) is a robust remote sensing model that can be applied to
This chapter is published as “SJ Zwart, WGM Bastiaanssen, 2007. SEBAL for detecting spatial
variation of water productivity and scope for improvement in eight irrigated wheat systems.
Agricultural Water Management 89(3), pp. 287-296”.
– 23 –
estimate the different components of the energy balance of the earth surface and thus
also actual evapotranspiration (ETact). This model was extended to produce estimates
of crop biomass production, so that crop yield, water use and water productivity can
be obtained in an integrated way. Remote sensing in combination with crop
production models has been acknowledged to be a powerful tool for estimating crop
yields at various spatial scales: within fields, between fields and on a regional scale
(e.g. Moran et al., 1995; Jongschaap, 2006).
The purpose of this paper is to show the conceptual framework for calculating crop
yields, ETact and water productivity with the SEBAL algorithm and to validate the
results with data from field measurements. Although it is beyond the scope of this
paper to explain the full theory behind SEBAL, an overview will be given in the next
section. The Yaqui Valley, an area in north-western Mexico and dominated by
irrigated wheat, is selected as a case study (Lobell et al., 2005). Beside the validation
in Yaqui Valley, a model validation will be performed based on the results of biomass
production, wheat yield and water productivity measurements and modelling results in
Sirsa, India (Van Dam et al., 2006). The same modelling framework was applied in
irrigated wheat systems in Pakistan, China, Egypt, India, the Netherlands and
California. To show the scope of improvement, the coefficient of variation (CV) of
WPET in the different systems was analysed and related to locally achievable
maximum wheat yields.
3.2 Materials and methods
Research area
The Yaqui River coastal plain is a highly productive agricultural region in the state of
Sonora, north-western Mexico, situated adjacent to the Gulf of California (see Figure
3-1). As being the original centre for the Green Revolution for wheat in Mexico, the
basin has rapidly grown to over 225,000 cultivated hectares. The primary source for
irrigation are two reservoirs, whose water is distributed through a network of canals.
In addition, approximately 700 extraction wells discharge directly to the distribution
canals. The area is dominated by wheat cultivation in the winter period, which was
reported to be 85 per cent of the total cropped area (Lobell et al., 2003). Wheat is
sown in late November – early December and is harvested in late April – early May.
Farmers usually irrigate four to five times within the season and apply around 250 kg
N per hectare.
– 24 –
February 26
January 25
April 14
decade 10 - 12
decades 5 - 9
decades 1 - 4
Normalised Difference Vegetation Index, NDVI (-)
Figure 3-1: The Yaqui irrigation district in Mexico as seen on a false-colour Landsat image acquired
on February 26 (2000). Green vegetation appears red on the image.
1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3
Figure 3-2: Average NDVI from SPOT-VEGETATION sensor* for all 1 km pixels encompassed in the
Yaqui irrigation district to depict the timing and length of the growing season. The bold part of the line
indicates the length of the growing season that is analysed in this study. The dates in the figure are the
Landsat acquisition dates that were used for the spatial disaggregation and time integration of the
corresponding decades.
For more information on SPOT-VEGETATION and free data availability go to
– 25 –
Methodological framework
The SEBAL model was applied for the 1999-2000 winter season and although the
growing season of wheat runs from November to April/May, the analysis of water
productivity relates to January 1 and April 30 (see Figure 3-2). This period is used to
calculate total cumulative evapotranspiration (ETact) and above ground biomass
production at high resolution with the SEBAL algorithm (Bastiaanssen and Ali,
2003). The SEBAL model calculates ETact (or the latent heat flux, λE) for each image
pixel from the energy balance equation:
λE = Rn − G0 − H
where Rn is the net radiation (W m-2), G0 is the soil heat flux (W m-2) and H is the
sensible heat flux (W m-2). Rn is computed from satellite-measured broadband surface
albedo, vegetation index and surface temperature, along with ground measurements of
global radiation. G0 is estimated as a fraction of Rn, surface temperature and
vegetation index. H is estimated from surface temperature, surface roughness, and
wind speed (see Figure 3-3). An essential component is the solution of extreme values
for H, prior to its pixel-to-pixel computations. The extreme values agree with H=0 for
water surfaces and H=Rn- G0 for desert surfaces.
digital elevation
land use
soil heat flux
net radiation
bulk surface
heat flux
momentum flux
sensible heat at extreme
Figure 3-3: Schematic overview of the SEBAL model.
Above ground biomass production on a single image acquisition day (DMday) is
calculated by the SEBAL model as:
– 26 –
DM day
 r
sa + ϕ 1 + s , min 
rav 
= 0.48S EXO
τ sw fε max Φ h
rs 
sa + ϕ 1 + 
 rav 
(kg ha-1 d-1)
where S EXO
is the incoming shortwave radiation at the top of the atmosphere (W m-2),
τ SW is the atmospheric transmissivity (-), ε max is the maximum light use efficiency (g
MJ-1), Φh is the stomatal response to ambient temperature (-), sa the slope of the
saturated vapour pressure curve (mbar K-1), ϕ the psychometric constant (mbar K-1),
rs the bulk surface resistance (s m-1), rs,min the minimum bulk surface resistance (s m-1)
and rav is the aerodynamic resistance to water vapour transport (s m-1). The resistances
rs, rs,min and rav are routinely solved in the latent heat flux. The APAR/PAR fraction, (), can directly be estimated from the NDVI (e.g. Hatfield et al., 1984). In this study we
f = −0.161 + 1.257 NDVI
The SEBAL model was applied to satellite images from both the National Oceanic
and Atmospheric Administration - Advanced Very High Resolution Radiometer
(NOAA-AVHRR) and Landsat satellite images. NOAA-AVHRR images are
characterized by a relatively high temporal resolution (once a day), but a low spatial
resolution of approximately 1 kilometre. Landsat images have a high spatial
resolution of 30 meter, but a low temporal resolution of 16 days. An analysis of the
growing season solely based on Landsat images is practically impossible as the
chance of almost all acquisitions during the season being cloud-free is very low in
most wheat areas. On the other hand an analysis using only NOAA-AVHRR images
would not give sufficient spatial detail. Therefore the advantages of both sensors are
combined in this methodological framework where high and low resolution products
are integrated to calculate total seasonal evapotranspiration and biomass production at
field level.
Table 3-1: NOAA-AVHRR and Landsat image acquisition dates and related decades for the
January 1 – February 10
February 11 – March 31
April 1 – April 30
January 4, 12 & 28 (2000)
February 10 (2000)
March 12, 16 & 20 (2000)
February 19 & 23 (2000)
April 1, 10 & 26 (2000)
January 25 (2000)
February 26 (2000)
April 24 (2000)
Twelve cloud free NOAA-AVHRR images were selected (see Table 3-1) and
processed using average meteorological inputs for the twelve corresponding decades
and the bio-physical properties derived for the NOAA-AVHRR overpass days. These
are surface albedo, fPAR, emissivity, evaporative fraction, surface roughness and bulk
surface resistance. Measurements of daily air temperature, wind speed, relative
humidity and radiation were acquired from a weather station located inside a wheat
– 27 –
field (27º14'23"N, 109º51'20"W). These low resolution 10-daily totals of ETact and
biomass production were then disaggregated to field level by means of high resolution
SEBAL generated ETact and biomass production maps of three Landsat TM scenes
(path/row: 34/41) that were acquired during the growing season on January 25,
February 26 and April 24. It is indirectly assumed that conditions on the individual
high resolution images are representative for the corresponding period that is
considered. The following data fusion method was applied: 10-daily NOAA ETact
results were split into three periods that correspond with one of the Landsat dates (see
Table 3-1). For each of these periods the NOAA-AVHRR-based ETact was summed
and an average ETact was calculated for the entire Yaqui irrigation district ETNOAA,i .
These average values were then multiplied with a relative ETact map that was created
by dividing the Landsat ETact map (ETETM , j ) with an average value of the Yaqui
irrigation district ETETM , j . In this way the spatial patterns are taken from Landsat
and the accumulated values from NOAA-AVHRR. The total seasonal ETact (ETseas)
was then acquired by summing the downscaled products of the three periods (i):
i =3 
ETseas = ∑  ETNOAA,i ⋅ ETM , j
ETM , j 
i =1 
(m3 ha-1)
Where i is the NOAA period and j is the corresponding Landsat image. The same
method to calculate ETseas is applied to calculate seasonal biomass production
(DMseas) from NOAA-AVHRR and Landsat biomass production maps:
i =3
DM ETM , j
DM seas = ∑  DM NOAA,i ⋅
i =1 
(kg ha-1)
(kg m-3)
Water productivity (WPET) is defined as:
H ⋅ DM seas
= i
where Yact is the total marketable grain yield (kg ha-1) and ETseas the seasonal water
use by evapotranspiration (m3 ha-1). Grain yield is obtained by multiplying
accumulated seasonal above ground biomass production (DMseas) with a specific
harvest index (Hi) for wheat. Wheat pixels were discriminated on the three Landsat
images using an unsupervised classification (iso-clustering) with no ground truthing
or field validation.
The ETact calculations were validated with eddy-correlation measurements of the
surface energy balance at one site located inside a large wheat field in Yaqui Valley
(27º14'23"N, 109º51'20"W), which was carried out from January 8 to April 29 (2000)
(Hoedjes et al., 2002; Garatuza-Payan and Watts, 2005). Wheat grain yield was
– 28 –
compared with field measurements and regional statistics from Yaqui Valley and with
field measurements from Sirsa, India.
The same methodology was applied to other seven irrigation systems across the
world. The number of high and low resolution satellite images that were used in these
studies depended on the length of the growing season and the availability of cloud
free images. In the cases of China, Pakistan, United States and the Netherlands, wheat
was classified using an unsupervised classification with no field control data. The
unsupervised wheat classification for Egypt and India was performed with field
control measurements. Although for none of the eight areas the accuracy of the wheat
classification was assessed, general accuracy figures for similar studies are between
80 and 90% (for an overview of classification performance of agricultural crops see
Bastiaanssen, 1998). To depict the scope for improvement in each system, the
coefficient of variation (CV) for WPET was analysed. In systems with a low CV, WPET
is considered homogeneous and optimal. The CV in these systems is a target value for
systems with a high CV. While maintaining the maximum value found within each
system, the WPET values were linearly forced to higher values, so that the CV
decreased to the target value. The scope of improvement was quantified by comparing
average WPET with the new average target WPET.
3.3 Results and discussion
Evapotranspiration validation
The measured ETact fluxes were first compared with the twelve ETact intervals
estimated by the SEBAL model of the corresponding 1 by 1 kilometre NOAAAVHRR pixel. As eddy-correlation measurements started on January 8, no value for
decade 1 was given; for comparison purpose ETact in decade1 was set equal to the
SEBAL estimate of decade 1. Figure 3-4 shows a good agreement between the
cumulative measured (eddy correlation) ETact per decade and the SEBAL estimated
cumulative ETact of the corresponding single NOAA-AVHRR-pixel. During February
and early March, SEBAL ETact is slightly higher, but runs equal with measured ETact
between decade 8 and 11. Cumulative measured ETact is low in decade 12 (9.4 mm),
but SEBAL ETact produces a high value for the same period: 45.6 mm, a difference of
36 mm. Without calibration or parameter tweaking, the accumulated ETact from
SEBAL was 410 mm for 121 growing days. The eddy-correlation measurements
revealed 363 mm, hence a difference of 13% if only the NOAA-AVHRR pixels are
The daily measured ETact was also compared with the SEBAL Landsat ETact on four
different dates during the growing season. Two out of four days show good
agreement: the difference between measured and estimated daily ETact is 0.0 mm on
January 24 and 0.4 mm on February 26 (see Figure 3-5). On the Landsat image of
February 18 some haze was present above a large part of the Yaqui irrigation district
including the measurement site. Due to haze an area appears colder on the satellite
image than it actually is, causing higher SEBAL ETact values. It was therefore decided
not to use this image in the final ETact map. On April 24 SEBAL ETact is 1.0 mm
higher than the measured value. This difference for the Landsat day appears consistent
– 29 –
with the NOAA-AVHRR results: April 24 is in decade 12 for NOAA-AVHRR and
both estimates are higher than the measured values.
The measured seasonal ETact is considerably lower than the SEBAL estimated ETact of
the integrated NOAA-Landsat product: 364 vs. 410 mm respectively (see Figure 3-5).
However, it should be taken into account that the SEBAL accumulated ETact estimates
cover 12 decades, starting from January 1 until April 30, whereas measurements were
being made between January 8 and April 29. Extrapolation of the daily measured
ETact show that approximately 10 mm ETact must be added to the total seasonal ETact
in order to consider the same twelve decades. Finally, it can be concluded that the
difference between measured and estimated ETact is entirely caused by the
discrepancy in decade 12. No satisfactory explanation for this higher estimated value
could be found, but it results in an overestimation of seasonal ETact of 8.8 per cent.
Wilson et al. (2002) compared the eddy-correlation measurements across 22
FLUXNET sites with different vegetation cover for 50 site-years. There was a general
lack of energy balance closure and mean imbalance was around 20 per cent for all
sites. A difference of 8.8 per cent between SEBAL estimated ETact and eddy
correlation measured ETact is well within the range of this measurement error.
cumulative actual evapotranspiration, ETact (mm)
The average SEBAL-based ETact of wheat in the entire Yaqui irrigation district (more
than 190,000 ha) equals 403 mm (σ~37 mm). The coefficient of variation (CV) of
0.09 indicates an extreme homogeneity in spatial water consumption patterns. A
detailed map of the spatial variation of ETact is given in Figure 3-6A.
Figure 3-4: Cumulative measured eddy-correlation ETact and SEBAL estimated cumulative ETact from
the corresponding 1x1 km NOAA pixel.
– 30 –
eddy-correlation measurements
SEBAL (Landsat)
eddy-correlation measurements
SEBAL (NOAA-Landsat integration)
Jan 1 - Apr 30
Jan 8 - Apr 29
Jan 24, 2000
Feb 26, 2000
Apr 24, 2000
actual evapotranspiration, ETact (mm season -1)
actual evapotranspiration, ETact (mm day-1)
Figure 3-5: Measured ETact plotted against SEBAL daily ETact of three Landsat acquisition dates and
seasonal ETact of the integrated Landsat-NOAA ETact map computed according to equation 4.
Biomass production and grain yield
Ideally, estimated wheat yields are validated with farmer reported yields in individual
fields that are large enough to be distinguished separately on the wheat yield map.
Unfortunately such measurements were not available for the 1999-2000 season that
was analysed, and a spatial validation could not be organized. However, three other
sources were consulted (see Table 3-2 and compared with the SEBAL yields.
Average estimated above ground biomass production for wheat in Yaqui irrigation
district between January 1 and April 30 (2000) is 14.9 ton ha-1 (σ~ 2.3). This figure
matches well with Del Blanco et al. (2000) who conducted 24 trials on wheat in
Yaqui in 1999-2000 and found an average biomass production of 15.2 ton ha-1
(σ~2.3). Moreover they measured an average grain yield of 6.4 ton ha-1 (at 0%
moisture), resulting in an average harvest index (Hi) of 0.42 (σ~0.040). Yield
measurements were made in small plots, each with an area of 4.2 m2, where wheat
was grown under optimal growing conditions regarding weed control, irrigation
management, soil fertility, etc. This may not be representative for the actual
conditions in farmers’ fields and average wheat yields for the entire Yaqui Valley are
expected to be lower.
During the 2000-2001 winter season, a farmer survey was conducted that reported
wheat yields from 80 different fields in the Yaqui Valley. Average yield in these
fields equalled 6.3 ton ha-1 (at 12% moisture content) or 5.5 ton ha-1 (at 0% moisture
content, σ~0.73) (Lobell et al., 2002). According to the official statistics, 191,281
hectares were cultivated with wheat in 1999-2000 and a total yield of 1,082,542 tons
was reported (Lobell et al., 2003). This results in an average yield of 5.7 ton ha-1 at
12% moisture, being equivalent to 5.0 ton ha-1 when it is dried. A Hi of 0.37 at 0%
moisture for wheat in Yaqui is reported by Lobell et al. (2003) and if this figure is
applied to derive the wheat yields from the SEBAL biomass production map for
– 31 –
Yaqui, an average wheat yield of 5.5 ton ha-1 (σ~0.87) is reached (at 0% moisture)
(see Figure 3-6B). It can be summarized that on average SEBAL yields are 0.5 ton ha1
higher when compared to the official statistics, but a comparison with farmer
reported yields shows exact correspondence, both in area average figures, as well in
the spreading of the yield data. It must be noted that the Hi is an important factor to
determine final yields from biomass production maps. Better knowledge on the
relation between biomass production and Hi will improve the accuracy of yield maps.
Table 3-2: Wheat yields in Yaqui Valley.
average yield
(ton ha-1)
Yaqui irrigation district
Ciudad Obregón
Yaqui irrigation district
* corrected for 12% moisture content
official statistics, 1999-2000
24 trial plots, 1999-2000
80 fields, farmer reported, 2000-2001
Lobell et al., 2003
Del Blanco et al., 2000
Lobell et al., 2002
SEBAL estimated wheat yields in Yaqui Valley depict a large range varying between
3 and 8 ton ha-1. Large variations in ETact are also observed, ranging from 250 to 450
mm. High yields can only be obtained when ETact is high as well. This agrees with the
fact that high yielding varieties from the Green Revolution consume relatively larger
amounts of water.
Based on the above described results of yield and seasonal ETact the water
productivity of wheat was calculated spatially using equation 5. The average WPET of
wheat for the entire Yaqui irrigation district equals 1.37 kg m-3 (σ~0.16, Table 3-3).
As can be observed from Figure 3-6c and Figure 3-7, the range of WPET in Yaqui lies
between 0.9 to 1.8 kg m-3. Apparently WPET not only varies within the irrigation
system, but also strongly between fields. This indicates that, besides climatology and
regional soil physical properties and hydrological conditions, farm management, such
as irrigation amount and timing, fertilization, weeding, choice of seed variety, crop
rotation, etc., play an important role in the level of WPET that is reached. With respect
to spatial variations of wheat yield in Yaqui Valley, Lobell et al. (2002) concluded
that management differences were more important than soil type and climate
– 32 –
a: ETseas (mm)
b: Yact (ton ha-1)
c: WPET (kg m-3)
Figure 3-6: Detailed maps of seasonal evapotranspiration (mm), yield (ton ha-1) and water productivity
(kg m-3) of large wheat fields in Yaqui irrigation district. The rectangular pattern is the result of 1 by 1
mile blocks into which the area originally was divided.
In Sirsa district, located in the western part of Haryana state (India), wheat yields
were measured during the 2001-2002 Rabi season in 24 farmers plots (Van Dam et
al., 2006). Average measured yield was 4.6 ton ha-1 (σ~1.3), while SEBAL yields
were slightly lower: 4.0 ton ha-1 (σ~0.8). Agricultural fields in Sirsa are small and
individual fields cannot be recognised on the Landsat images. This means that
SEBAL yield values may be a mixture of the actual field, some surrounding fields and
non-agricultural objects such as roads, houses and ditches. A slightly lower SEBAL
wheat yield can thus be expected for such kind of landscape.
The calibrated SWAP-WOFOST simulation model for soil moisture flow, crop ETact,
carbon assimilation and crop growth, was used to estimate water productivity for
wheat, in Sirsa district at field level and regional level (Singh et al., 2006; Van Dam et
al., 2006). Simulated WPET at field level averaged 1.32 kg m-3 in 24 fields, while
distributed modelling techniques for all fields in Sirsa District yielded average
regional WPET for wheat of 1.20 kg m-3. These values are close to the area average of
SEBAL modelled values for WPET in Sirsa district which equalled 1.22 kg m-3.
Modelled wheat yields by SWAP-WOFOST averaged 4.4 ton ha-1 with an associated
ETact of 388 mm; both values are higher (4.3 and 7.0 % respectively) than SEBAL
wheat yield and ETact, but the water productivity is similar.
In the province of Zeeland, the Netherlands spring wheat yields of nine farms were
predicted for the 2005 season using SEBAL (unpublished). The results were validated
using measured yields from nine agricultural fields. On average measured yields were
– 33 –
3.5% higher than the remote sensing estimates. The consistency in yield predictions
across three different continents reveals that SEBAL formulation is acceptable and
good enough for further water productivity analysis.
Water productivity and scope for improvement
SEBAL was also applied to quantify water productivity for wheat dominated areas in
China, India, Pakistan, Egypt, the Netherlands and California. In all studies SEBAL
was used to estimate ETact, yield and WPET according to the methodology described
before. Table 3-3 summarizes the results of these studies and Figure 3-8 depicts the
frequency histograms of WPET for each system. Large differences exist in WPET
levels: the highest values occur in the Nile Delta, Egypt (1.52 kg m-3), Kings County
(CA), USA (1.44 kg m-3) and Oldambt, The Netherlands (1.39 kg m-3). The lowest
average value was found in Sindh Province, Pakistan: 0.54 kg m-3. Water productions
in Pakistan are almost three times lower than in Egypt. This can be ascribed to both
low yields and high ETact levels. The correlation figures in Table 3-3 reveal that in all
systems, except Sindh, the differences in WPET can be better explained by yield levels
than by ETact. The implication of this finding is that agronomical practices are more
important for increasing crop water productivity than water management practices.
Zachariasse (1974) concluded that farm management practices have more influence
on yield than physical system properties. Average ETact for all systems varies between
355 and 467 mm while yield levels range from 2.5 to 5.4 ton ha-1. The range of
average WPET detected by satellites goes from 0.54 kg m-3 in Sindh to 1.52 kg m-3 in
Egypt. The consequence of this observation is that better crop growth conditions and
alert crop management resulting in higher yields should be regarded as the major
vehicle to improve crop water productivity.
A plausible range of WPET values for wheat is 0.6-1.7 kg m-3 and average WPET, based
on 82 literature sources with measurements conducted in the last 25 years, is 1.09 kg
m-3 (Zwart and Bastiaanssen, 2004). The average WPET, based on the eight areas in
this study, is 1.11 kg m-3. This comparison shows that the global data sets of Figure
3-8 display a good representation.
In this study the coefficient of variation (CV) was calculated for the population of
WPET in each of the eight irrigation systems (see Table 3-3). Spatial spreading of
WPET is instructive for possible water savings in given irrigation systems. A low CV
value, such as in the Nile Delta, Kings County, Oldambt and Sirsa, indicates that the
distribution of WPET within the system is homogeneous and that there is little scope
for improvement. On the other hand the systems in Linxian County, Hebei Province
and Sindh Province depict a high CV of 0.33, 0.33 and 0.22 respectively, and
consequently increases in the regional average WPET can be achieved more easily in
these systems than in systems with low CV. Moreover, Figure 3-7 shows that the
average WPET decreases with increasing CV and this suggests that heterogeneity is
responsible for low crop water production performance. A similar finding was made
by Thiruvengadachari et al. (1997) who noticed CV-values of 0.30 in low
productivity wheat fields in Haryana, India. It could be explained that heterogeneity is
responsible for low yields and hence low crop water productivity.
– 34 –
coefficient of variation for WPET
y = -0.25x + 0.44
R2 = 0.68
average WP ET for each system(kg m )
Figure 3-7: Average WPET for each wheat system plotted against the coefficient of variation of WPET.
average WPET
(kg m-3)
max WPET
(kg m-3)
WPET – ETact
WPET – yield
average Yact
(ton ha-1)
average ETact
Nile Delta, Egypt
Kings County (CA), USA
Oldambt, The Netherlands
Yaqui Valley, Mexico
Sirsa, India
Linxian County, China
Hebei Province, China
Sindh Province, Pakistan
Hi (-)
Table 3-3: Key figures of the eight wheat systems that were analysed. The harvest index (Hi), average
ETact, yield, WPET, maximum WPET, the coefficient of variation (CV) for WPET, and the correlation (ρ)
between WPET and yield, and WPET and ETact. The standard deviation (σ) for average ETact, yield, and
WPET is given between brackets. Maximum WPET is defined as the 98% percentile.
355 (20)
395 (19)
372 (36)
403 (37)
361 (16)
436 (35)
380 (50)
467 (82)
5.4 (0.44)
5.7 (0.59)
5.2 (0.64)
5.5 (0.90)
4.4 (0.33)
3.8 (1.42)
2.5 (0.96)
2.5 (0.77)
1.52 (0.09)
1.44 (0.11)
1.39 (0.07)
1.37 (0.16)
1.22 (0.06)
0.85 (0.28)
0.64 (0.21)
0.54 (0.11)
– 35 –
China - Hebei province (2002)
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
frequency (%)
India - Sirsa district (2002)
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
United States - Kings County, CA (2002)
Mexico - Yaqui Valley (2000)
Egypt - Nile delta (2003)
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
Pakistan - Sindh province (1999)
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
China - Linxian county (2002)
Netherlands - Oldambt (1995)
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
water productivity, WPET (kg m-3)
Figure 3-8: Frequency histograms of wheat water productivity (defined as the yield divided by the
water use by evapotranspiration) in eight wheat systems in seven different countries.
To quantify the scope of improvement, the CV of all systems were forced to the
lowest CV value found across the eight systems (0.05 in Oldambt and Sirsa), while
maintaining the maximum value for WPET that can be obtained under practical
circumstances. The 98% percentile is considered representative for opportunistic
– 36 –
WPET values. This is visualized in the frequency histogram for WPET in the Yaqui
Valley (see Figure 3-9). Pixels on the lower side (0.9-1.0 kg m-3) are forced to values
between approximately 1.3 and 1.4 kg m-3 to meet the criterium of a CV equalling
0.05, while maintaining the 98% percentile value of 1.69 kg m-3. Fields or areas with
significant scope for improvement are now spatially recognized. Such information
gives water managers the opportunity to allocate plots with inefficient water use.
Extension officers can direct their services to specific farms where crop water
productivity is low. Water policy makers can also release sanctions to encourage
farmers to increase their profit from scarce water resources.
CV=0.12 avg=1.37 kg m-3
frequency (%)
improved WPET CV=0.05 avg=1.54 kg m-3
water productivity, WP ET (kg/m )
Figure 3-9: The original frequency distribution for WPET in Yaqui, Mexico and the improved WPET
that was forced to CV of 0.05 while maintaining the maximum value.
Table 3-4: Improved average WPET for eight irrigation systems at a coefficient of variation (CV) of
0.05. The percentage wise increase is given between brackets.
Nile Delta, Egypt
Kings County (CA), USA
Oldambt, The Netherlands
Yaqui Valley, Mexico
Sirsa, India
Linxian County, China
Hebei Province, China
Sindh Province, Pakistan
improved average WPET
(kg m-3)
1.54 (+1%)
1.47 (+2%)
1.39 (0%)
1.54 (+12%)
1.22 (0%)
1.16 (+35%)
0.87 (+34%)
0.66 (+24%)
1.23 (+14%)
– 37 –
In Yaqui Valley WPET increases on average 12% from 1.37 to 1.54 kg m-3, which
implies that increasing wheat yield by 12% at the same ETact is feasible. It can also be
argued that the same yield can be obtained from 12% saving in water consumption. At
for instance 40% overall irrigation water losses, 12% reduction of ETact will translate
into 17% less water diversions. Similar calculations were made for the other systems
and are summarized in Table 3-4. The largest scope for improvement was found in
Linxian County and Hebei Province where WPET can be increased maximally by 34
and 35% if all interventions are effective. Also in Sindh Province the ETact can be
reduced by 24% leading to substantial water savings in this area. On average for the
eight systems WPET can be increased 14%. The first approximation of possible global
water savings on irrigated wheat is thus 14% (consumption) and 20% (diversion),
provided that 40% of diverted water is lost between the off take and root zone water
3.4 Conclusions
The methodology proposed in this study proved to be accurate in estimating seasonal
ETact and yields of wheat at field level. The eddy-correlation measurements of ETact
correlated well with SEBAL estimated ETact, and wheat yields were well within the
range of yields that were measured in Yaqui Valley and Sirsa District. The analysis of
WPET in the different irrigation systems showed that large variations exist in the levels
of WPET that are achieved. The scope for improvement is largest in the systems where
WPET is lowest (WPET<0.8 kg m-3) and CV is highest (CV>0.3). The average increase
in WPET is 14%, thus 14% of the water resources can be saved in consumption of
irrigated wheat at maintained level of crop production which is equivalent to 20%
saving in water diversion.
It is concluded that by applying this methodology, valuable information is generated
for water resource planners and irrigation managers on the spatial distribution of
water productivity. Using these maps, areas with low WPET can be detected and by
relating them to spatial information on soil type, groundwater table, quality of
irrigation water, farm management, etc. the underlying causes can be searched for.
The most encouraging leap forward is that water productivity can not only be
quantified for different areas (hence a comprehensive picture on wheat and water), but
that also within region-variations can be detected. The latter is essential for the sake of
demonstration that it is realistic to obtain a higher utilization of water resources under
local environmental conditions. Although remote at itself sensing cannot provide the
underlying reasons for variations, it can be used to identify farm plots with good and
poor management practices. This is fundamental information for extension services,
dissemination and farmers-help-farmers teach programs. Since water productivity is
more tightly coupled to yield than to ETact, increasing low yields seems the most
straightforward solution to increase water productivity.
– 38 –
4 WATPRO: A remote sensing based
model for mapping water productivity
of wheat*
4.1 Introduction
The agricultural sector will require more water in the near future to provide more
food, fibre and fuels (Kijne et al, 2003; Molden et al., 2007). Underlying reasons are a
growing population that needs more food, a change in diet due to increased prosperity,
and a recent focus on biofuels. While more water is needed for agriculture, the
demand for water from other users - such as the industry, tourism and domestic
sectors - is also increasing. This competition for water is resulting in a situation where
the irrigation sector is unlikely to be allocated more water in the future in water-scarce
areas. Moreover, climate change adds to the already existing pressures on water
resources. It is within this global setting that the agricultural sector, being the largest
water consumer, must adapt itself and make water use more productive to sustain food
production for the coming generations. The challenge is to produce more food and
animal feed from the same amount of water (Molden et al., 2009).
The productivity of water (WPET) can be expressed by the harvestable grain yield per
unit of water consumption by evapotranspiration (Yact / ETact). It has the advantage of
expressing the water unit of the equation as a true consumption; the
evapotranspiration process withdraws water from soil into the atmosphere, water that
is no longer available for downstream users (Molden and Sakthivadivel, 1999). Based
on a review of water productivity measurements in literature from recent years, Zwart
and Bastiaanssen (2004) presented plausible ranges of water productivity that can be
expected globally for major crops based on carefully selected field experiments.
These ranges, however, display a large variation as a result of growing conditions that
vary from year-to-year and from location-to-location. On a global scale, differences in
the physical water productivity are expected based on climatic variation. Bierhuizen
and Slatyer (1965) reported an inverse linear relationship between vapour pressure
deficit and water productivity. Steduto et al. (2007) and Sadras and Angus (2006)
demonstrated an inverse linear relationship between water productivity and reference
This chapter is in press with Agricultural Water Management as “Zwart, S.J., W.G.M. Bastiaanssen,
C. de Fraiture, D.J. Molden, 2009. WATPRO: a remote sensing based model for mapping water
productivity of wheat”.
– 39 –
evapotranspiration. Considering that temperate climates have lower vapour pressure
deficits and reference evapotranspiration than arid climates, crops may be grown with
less water resources in the former climates.
There is a need for mapping and benchmarking water productivity to support a global
understanding of where crops are grown with least water consumption, and where
there is greatest scope for improvement. In addition to that, Zwart and Bastiaanssen
(2007) proposed a methodology to assess water productivity variability within regions
to identify plots with poor and with good water management practices. Their
methodology allows analysis of spatial variation at field scale, and to quantify the
scope for improvement within an irrigation system or region during a specific year.
As water productivity is related to weather conditions, the level of water productivity
may vary strongly from year to year, even if crop and water management are similar
in both years. Other examples of regional water productivity mapping based on
remote sensing and modelling are given by Bastiaanssen et al. (1999), McVicar et al.
(2004), Van Dam et al. (2006), Immerzeel et al. (2008), Anderson et al. (2008), Mo et
al. (2009) and Zwart and Leclert (2010). Although calibrated deterministic models
allow scenario analysis, their major disadvantage is the detailed input that is required
and which is usually not available spatially. On a global scale, Liu et al. (2007)
modelled wheat yields and potential evapotranspiration to calculate water
productivity. It is known, however, that water productivity rarely meets the potential
values of evapotranspiration or maximum crop yields. Moreover, crop modelling
requires spatial estimates of management data on irrigation, drainage, soil tillage and
fertilization, which are not straightforward to describe, while remote sensing on the
other hand provides a direct measurement of water productivity. Models are therefore
considered less suitable to benchmark water productivity than remote sensing
techniques since agronomic management data are usually not available at the higher
To the authors knowledge, there exists no analysis that provides benchmark values for
water productivity globally on the basis of crop and water measurements under actual
field conditions. The objective of this paper is therefore to develop a simplified
remote sensing based model of the water productivity of wheat (a major food crop
globally), hereafter named WATPRO, with low input data requirements and
applicable at high spatial resolution. The point of departure is that the model should
directly estimate water productivity, instead of estimating crop yield and
evapotranspiration separately. Simplifications in the model’s parameterization and
input data preparation were introduced in order to provide a robust model that can be
applied from local to global scales to assess water productivity with minimum data
requirements. The model was tested using a data set of measured water productivity
values that was reported in the literature and summarized by Zwart and Bastiaanssen
(2004). During a field campaign in Sirsa District, India, the water productivity was
measured in farmer’s fields and assessed at regional scale with remote sensing (Van
Dam and Malik, 2003). Their results will also be used to validate the WATPRO
model. The WATPRO model was applied on global remote sensing data sets to obtain
a global benchmark map of water productivity for wheat (Zwart et al., 2010b).
– 40 –
4.2 Water productivity model
In this study, water productivity (WPET) is defined as the ratio of harvestable wheat
yield (Yact) divided by the accumulated amount of water consumed by
evapotranspiration during the growing season (ETact):
t =h
∑ ET
t =e
(kg m-3)
(t )
where Yact is the harvestable wheat grain yield at 14% moisture (kg ha-1) and ETact is
the accumulated actual evapotranspiration computed for time period t for the growing
season that runs from crop establishment (t=e) to harvest (t=h).
Wheat yield and dry matter production
Wheat grain yield can be approximated as the product of the accumulated dry matter
production (DM) during the growing season multiplied by the harvest index (Hi),
which is the fraction of above-ground dry biomass comprising grain, and a correction
for the fraction of water present in the grain ( θ grain ):
t =h
Yact = H i /(1 − θ grain ) ⋅ ∑ DM (t ) (g m-2)
t =e
Monteith (1972) developed a theoretical framework that relates the absorbed
photosynthetically active radiation (APAR) and the light use efficiency to plant dry
matter production:
DM = APAR ⋅ ε
(g m-2)
where DM is the above ground dry matter production (MJ m-2) and ε is the light use
efficiency (g MJ-1). APAR is calculated from:
APAR = f ⋅ χ ⋅τ SW ⋅ S exo
⋅ 0.0864 (MJ m-2 day-1)
In this equation, the incoming shortwave radiation (0.3-3.0 µm) is calculated by
in W m-2) during the growing
multiplying the average extraterrestrial radiation ( S EXO
season with a dimensionless atmospheric transmissivity ( τ SW ). Only part of the
incoming solar radiation can be absorbed by plants for photosynthetic processes. The
fraction between the so-called photosynthetically active radiation (PAR, 0.4-0.7 µm)
is given by χ . PAR describes the amount of energy that is available for
and S EXO
photosynthesis if the leaves intercept all radiation. APAR will be much lower than
PAR as leaves transmit and reflect solar radiation. APAR is a function of the canopy
reflectance at the upper side of the canopy, the amount transmitted through the
– 41 –
canopy, and the amount that is reflected back to the canopy by the soil. These
processes can be summarized in the APAR/PAR fraction f . Several authors have
reported a linear relation between the NDVI and f (e.g. Hatfield et al., 1984; Asrar et
al., 1992)
f = a ⋅ NDVI + b (-)
Where both a and b are empirical values that can be determined for different crops.
The light use efficiency ε for wheat is a constant that does not vary if environmental
conditions are non-limiting (Monteith, 1972). However, both temperature and water
availability can have a significant impact on the ε . Field et al. (1995) used the
ε = ε max ⋅ W ⋅ T1 ⋅ T2 (g MJ-1)
where ε max is the maximum light use efficiency (g MJ-1), T1 and T2 are scalars for
temperature effects on the light use efficiency. W is a water scalar that is defined as
actual over potential evapotranspiration. In this study the evaporative fraction ( Λ ),
which is the fraction between the energy actually used for evapotranspiration and the
net available energy, is used instead, following Bastiaanssen and Ali (2003). T1 and T2
are provided by mechanistic models that attempt to capture two aspects of the
regulation of plant growth by temperature. Both scalars can be computed from the
monthly average temperature (Tmon) and the monthly average temperature during the
month of maximum leaf area index (Topt) (Field et al., 1995):
T1 = 0.8 + 0.02Topt − 0.0005Topt
T2 =
1 + exp(0.2Topt − 10 − Tmon )
1 + exp{0.3(− Topt − 10 + Tmon )}
T1 provides an optimal temperature for plant production and reduces production for
extreme cold or hot climates, where T2 assumes that vegetation acclimates to the sitespecific seasonal optimal temperature trajectory. Plant production is reduced by
temperatures lower and higher outside the month of optimal production. Both scalars
were, like the model proposed in this research, developed for global application and
they were designed without distinguishing between different ecosystems or crops.
Although the scalars were not validated seperately, the obtained light use efficiencies
for various ecosystems were compared with light use efficiencies reported in other
sources with satisfactory results.
Eq. 4.2 can now be rewritten as the summation of the dry matter production per period
(t) between emergence and harvest of the crop, multiplied by the harvest index and
corrected for the grain water content:
– 42 –
t =h
Yact = H i /(1 − θ grain ) ⋅ ∑ (a ⋅ NDVI + b ) ⋅ χ ⋅ S EXO
⋅ τ SW ⋅ ε max ⋅ Λ ⋅ T1 ⋅ T2 (t ) ⋅ 0.864
t =e
(kg ha-1)
Actual evapotranspiration
Actual evapotranspiration (ETact) is an important component of the energy balance in
cropped land:
Rn = G0 + H + λE
(W m-2)
where Rn is net radiation (W m-2), G0 is the soil heat flux (W m-2), H is the sensible
heat flux (W m-2) and λE is the latent heat flux that is associated with the actual
evapotranspiration. The energy balance can be rewritten to:
λE = Λ ⋅ (Rn − G0 ) (W m-2)
where Λ is the dimensionless evaporative fraction and Rn − G0 equals the net
available energy available for evapotranspiration. On time scales of one day or more,
the soil heat flux can often be ignored and thus the latent heat flux λE is a function of
Λ and Rn only. The net radiation is the sum the net shortwave and net longwave
radiation at a location. De Bruin and Stricker (2000) evaluated three different methods
for estimating the net radiation. It was found that a simple expression, that calculates
) and the extrafrom the surface albedo ( α ), the incoming shortwave radiation ( S IN
terrestrial radiation ( S EXO
), provided the best results. This equation is written as:
Rn = (1 − α ) ⋅ S EXO
⋅τ SW − 135 ⋅τ SW
(W m-2)
where S IN
has been replaced by the atmospheric transmissivity ( τ SW ). For
daily time steps, the latent heat flux of Eq. 4.11, being closely associated with the
actual evapotranspiration, can be written as the product of evaportaitve fraction and
the net radiation described in Eq. 4.12:
t =h
{ (
ETact = ∑ Λ ⋅ (1 − α ) ⋅ S EXO
⋅ τ SW − 135 ⋅ τ SW (t ) ⋅ 0.35 (m3 ha-1)
t =e
where 0.35 is a conversion factor to express ETact in m3 ha-1 from W m-2. This implies
that ETact can be estimated from four spatial parameters; the evaporative fraction,
surface albedo, incoming terrestrial radiation and atmospheric transmissivity.
– 43 –
Water productivity
After substituting equations 4.9 and 4.13 into equation 4.1, the water productivity
(WPET) can be written as:
t =h
H i ⋅ ∑ (a ⋅ NDVI + b ) ⋅ χ ⋅ S EXO
⋅ τ SW ⋅ ε max ⋅ Λ ⋅ T1 ⋅ T2 (t ) ⋅ 0.864
t =e
(1 − θ )⋅ ∑ {Λ ⋅ ((1 − α ) ⋅ S
t =h
t =e
(kg m-3)
⋅ τ SW − 135 ⋅ τ SW (t ) ⋅ 0.35
It is hereby assumed that both the production and evapotranspiration term are applied
for similar time steps (t). If spatial and temporal varying input parameters are
averaged for the growing season (e to h), equation 4.14 can be rewritten as:
)⋅ Λ ⋅ ((1 − α )⋅ S
H i ⋅ a ⋅ NDVI + b ⋅ χ ⋅ S EXO
⋅ τ SW ⋅ ε max ⋅ Λ ⋅ T1 ⋅ T2 ⋅ 0864
(1 − θ
⋅ τ SW − 135 ⋅ τ SW ⋅ 0.35
(kg m-3)
It is recognized that the average of a fraction is not equal to the ratio of the product of
the mean values of the variables. However, it was found for a hypothetical test set that
this assumption generates differences less than 5% with no consistent under or over
estimation of water productivity. The use of seasonal average input parametes is a
simplification that allows to ommit the seasonal average atmospheric transmissivity
and the evaporative fraction so that Eq. 4.15 can be simplified into:
H i ⋅ a ⋅ NDVI + b ⋅ χ ⋅ S EXO
⋅ ε max ⋅ T1 ⋅ T2 ⋅ 0864
(1 − θ )⋅ ((1 − α )⋅ S
− 135 ⋅ 0.35
(kg m-3)
This implies that water productivity (WPET) can be estimated from four spatial
variables: 1) broadband surface albedo α , 2) vegetation index NDVI, 3) the
, and 4) air temperature (Tmon) to compute T1 and T2.
extraterrestrial radiation S EXO
Both NDVI and α are typical remote sensing parameters which can be derived at
different spatial and temporal resolution from narrow band satellite measurements.
Model parameters PAR / S EXO
fraction χ and the maximum light use efficiency ε max
can be held globally constant. This practically implies that all key variables can be
obtained from routine satellite measurements.
The innovation of Eq. 4.16 is that both evaporative fraction ( Λ ) and atmospheric
transmissivity ( τ SW ) can be omitted. This is a significant advantage since both these
parameters are difficult to measure or estimate in the space and time domain. The
evaporative fraction has been recognized as being the most difficult parameter of the
– 44 –
surface energy balance to describe (Shuttleworth, 1989). This analysis, however, does
not provide separate estimates of yield and actual evapotranspiration.
Following Moran et al. (1995), the PAR / S EXO
fraction χ is set at 0.48. Monteith
(1972) concluded that the maximum light use efficiency is constant and if crops are
not short of water, it will only vary between C3 and C4 crops. Bastiaanssen and Ali
(2003) presented an overview of measured values of ε for wheat crops, and based on
their review they propose ε max to be 2.5 g MJ-1 for wheat. Myneni and Williams
(1994) analysed the relationship between the NDVI and the APAR/PAR fraction f
under varying conditions. They found that a scale-invariant linear relationship exists
for backgrounds of moderate brightness (i.e. not bright sandy or dark peaty soils) if
NDVI measurements are corrected for atmospheric and bidirectional effects. Several
authors have reported linear relationships for different crops. The average a and b of
two spring wheat experiments (Asrar et al., 1984 and Hatfield et al., 1984) will be
used in the application of the model (1.23 and -0.149 respectively).
4.3 Model sensitivity and performance
To obtain a better understanding of the behaviour of the WATPRO model, a
sensitivity analysis was performed. WATPRO was applied four times to calculate
WPET , each time varying one of the inputs and keeping the others constant.
Representative global seasonal average values of the four spatially variable inputs
were set at 0.38 (NDVI), 0.16 ( α ), 350 W m-2 ( S EXO
) and 16oC (Topt). For each
parameter a range was set, which was based on the minimum and maximum values of
the seasonal averages, which were derived from the application of the WATPRO
model for wheat at a global scale (Zwart et al, 2010b).
Two different sources were consulted to test the model’s performance: 1) field
measurements of water productivity reported in the literature and incorporated in a
data base by Zwart and Bastiaanssen (2004), and 2) spatially measured, modelled and
validated water productivity reported by Van Dam and Malik (2003). The data base
by Zwart and Bastiaanssen was updated with eleven more recent publications and
now contains results from field experiments on wheat yields and water use by
evapotranspiration that were conducted on 39 different locations globally. The
measurements were conducted in experimental fields where crop growth and water
use were monitored under different growing conditions, such as varying irrigation
water management (amount and timing of application) and fertilizer management
(amount N applied). All sources, except one, provide the results from identical
experiments repeated during more than one season. For all of the included
experiments, multi-seasonal average evapotranspiration, wheat grain yield, water
productivity, geographic location and sowing and harvest times were calculated. If
provided, the measured harvest indices were also included in the data base (see Table
The WATPRO model (Eq. 4.17) was applied at the 39 locations (see Figure 4-1) and
the modelled water productivity values were compared with the measured ones. As
– 45 –
described in the previous section, six out of ten input parameters were taken from
literature sources. For the remaining (NDVI, surface albedo, air temperature) global
data sets were used. Few publications provide air temperature and therefore these
were extracted from the climate grids of the Water and Climate Atlas from the
International Water Management Institute (IWMI) derived from the University of
East Anglia climate grids (New et al., 1999). The SPOT-Vegetation sensor provided
10-daily NDVI maps since April 1998. An average NDVI time series was calculated to
represent a normal growing season to be used as input for the model. The average
NDVI for each of the 36 10-daily periods was calculated from the ten years (19982008) that were available. Seasonal extremes in the average NDVI were minimized by
excluding the minimum and maximum NDVI value at each location. As the
geographic position and start and end of the growing season of the experiment were
provided by the literature source, the seasonal NDVI values for each experimental
location could be extracted from the maps. A similar multi-year average global data
set was available for the broadband surface albedo: spatially complete global spectral
surface albedos (Moody et al., 2005). This data set contains 23 cloud-free and gapfilled surface albedo images that are averaged from the period 2000 to 2004. Each
image is representative of a 16-day time period. The spatial resolution is 1 arcminute,
which is equal to approximately 1.6 km at the equator. The surface albedo images are
public domain and available from the MODIS Atmosphere ftp-site*.
Figure 4-1: Location of the 39 experimental sites where water productivity was measured and for
which water productivity was modelled. The points are superimposed on the wheat fraction map that
shows the spatial dominance of wheat across the world (Leff et al., 2004) , and the degree to which the
sites represent global wheat growing areas.
WATPRO was applied to 10-day time steps using the SPOT-VGT 10-daily period
system as a basis. In this system each month is divided in three periods, where the
first period runs from day 1 to 10, the second from day 11 to 20, and the third period
runs from day 21 to the end of the month. With the provided geographic location of
the 39 experimental locations, and the start and end dates of the growing season, the
– 46 –
seasonal average values of the NDVI, broadband surface albedo, extraterrestrial
radiation and air temperature can be extracted from the maps. If provided by the
literature source, the water productivity model included a harvest index measured in
the experiments. However, not all provided a measured harvest index, and for those
cases the average harvest index value from the other experiments, being 0.35, was
applied. The reported water productivities in this paper assume a 14% grain moisture
In an extensive study in Sirsa District (Haryana, India) the water productivity of
wheat, amongst other crops, was studied (Van Dam and Malik, 2003). Five sites were
intensively monitored during the 2001-02 agricultural season in terms of irrigation
supply, crop development, soil moisture and salinity (Singh et al., 2006). The sites
were located in an alluvial plain with wheat being the major winter crop. The sites
were selected to have different crop rotation (wheat-rice or wheat-cotton), irrigation
water supply (canal, pump or a combination), and soil and ground water conditions.
The fields were managed by farmers under actual fields conditions, and this situation
therefore differed from traditional experimental sites where yields, ETact and WPET are
measured in small fields on experimental stations and under optimal conditions. The
measured water productivity in the fields was compared with WAPTRO modelling
results of the pixel in which the fields were located. For the application of WATPRO,
the NDVI and surface albedo were taken from the SPOT-VGT 1998-2008 and the
MODIS 2000-2004 average data sets respectively. An analysis of the NDVI time
series revealed that the NDVI depicted little variation from year-to-year and therefore
in this specific case, differences are not expected to arise from the use of the average
data sets. The low variation between years is possibly caused by the fact that the
wheat areas are largely irrigated from ground water and they are thus not dependent
on water availability from precipitation or irrigation.
During this study, WPET was also assessed spatially using satellite imagery and the
SEBAL algorithm. Both wheat yields and seasonal actual evapotranspiration were
quantified using NOAA-AVHRR satellite images at a resolution of 1 by 1 kilometre.
For the application in Sirsa District the modelling period was fixed from November 1
to April 30 (Van Dam and Malik, 2003), whereas wheat yields were obtained with a
fixed Hi of 0.39 and a θ grain equalling 11%. The SEBAL WPET map was compared
and validated using the same field measurements of yields and ETact. The WATPRO
model was applied to the same region, using the same Hi, θ plant and modelling period.
The original surface albedo and NDVI imagery were not available, and therefore these
were substituted with the MODIS surface albedo and SPOT-VGT NDVI data sets as
described before. Since the WATPRO model is derived from the SEBAL principles,
the comparison is not a validation of Eq. 16. It may show, however, whether
WATPRO, with its simplifications and assumptions, can spatially predict WPET
similar to SEBAL.
4.4 Results and discussion
The sensitivity of WATPRO towards the four spatial input parameters is depicted in
Figure 4-2. If the global average values are applied, the average WPET amounts 1.06
– 47 –
kg m-3. It can be seen that WPET is least sensitive to air temperature during maximum
crop development (Topt); values higher than 16oC have almost no impact whereas at
lower temperatures WPET is maximally reduced by 17% (compared to the average
1.06 kg m-3). Both surface albedo and NDVI have a positive linear relationship with
WPET , though the impact of the surface albedo is much smaller (maximally 18%) than
the impact of NDVI on WPET (81%). Finally, at values of seasonal average
extraterrestrial radiation lower than the average 350 W m-2, WPET levels are 57%
lower. However, this difference is smaller (16%) for values higher than the global
average of 350 W m-2. This analysis reveals that WPET of wheat can be expected to be
high if NDVI is high and radiation is low.
WPET (kg m )
surface albedo (-)
WPET (kg m )
extraterrestrial radiation (W m )
NDVI (-)
WPET (kg m )
WPET (kg m )
T opt ( C)
Figure 4-2: Sensitivity of WATPRO to variation in the four spatial inputs: (a) NDVI, (b) surface
albedo, (c) extraterrestrial radiation and (d) temperature during maximum crop development. The
white bullets indicate the average value of the parameters.
The NDVI and surface albedo values were extracted from the 39 experimental
locations where water productivity was measured. Wheat experiments are usually
conducted in small fields located on experimental farms or field stations, whereas the
NDVI values are an average of an area of approximately 1 by 1 kilometre. The time
profiles of the NDVI and surface albedo were plotted and checked for consistency
with the average sowing and harvest dates during the experiments. As was shown in
the sensitivity analysis, the seasonal average NDVI has a strong impact on the final
levels of WPET. In 19 locations the inputs were found to be unsuitable to apply
WATPRO for two reasons: first of all, a mismatch was found between the growing
season that was reported for the experiments and the establishment of the crop and the
harvest data that could be derived from the profiles. Secondly, the NDVI profile does
not represent a typical wheat season. The NDVI profile of the NDVI and surface
albedo are plotted for two locations (Figure 4-3). The first location (Mallee, Australia)
shows an NDVI profile that is representative of wheat, whereas the second (Yellow
– 48 –
Jacket, USA) shows interference by a summer crop between the first period in July
and the first period of August. The latter was therefore excluded from further
comparison with the WATPRO model. The remote sensing derived NDVI was in
those cases not representative of the experimental sites and conditions and these were
excluded from further analysis. The remaining 20 locations were situated in an area
surrounded and dominated by other wheat fields and considered suitable for
comparison with modelled water productivity.
NDVI: Mallee
NDVI: Yellow Jacket
α: Mallee
α: Yellow Jacket
surface albedo, α (-)
NDVI (-)
1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3
Figure 4-3: Examples of time profiles of the NDVI and surface albedo ( α ) on two selected locations:
Mallee, Australia (Latta and O'Leary, 2003) and Yellow Jacket, USA (Al-Kaisi et al., 1997). The values
were extracted from 1 by 1 kilometre SPOT-VGT using the reported dates of the growing season.
Figure 4-4 depicts the measured versus modelled crop water productivity in 20
experimental locations. The average measured water productivity is 1.00 kg m-3 and
the average WATPRO modelled water productivity is similar at 0.99 kg m-3. Also the
ranges are very similar: measured values are between 0.52 and 1.42 kg m-3, whereas
modelled values vary between 0.54 and 1.54 kg m-3. There is, however, only a weak
positive correlation between modelled and measured values (r2 equals 0.15). The
comparison between experiment measurements and modelling results shows that
differences of more than 50% occur (see Table 4-1), although there is no consistent
overestimation or underestimation of WPET. In Mullewa, Australia, modelling results
are 56% higher, but in Merredin, Australia (also reported by Regan et al., 1997) there
is no difference between the average of the experiments and the modelled WPET (both
1.32 kg m-3).
– 49 –
y = 0.39x + 0.52
R2 = 0.15
WP ET measured (kg m-3)
Figure 4-4: Measured versus modelled water productivity (WPET in kg m-3). Each point represents the
average of the measured and modelled WPET at one experimental location. In each point a wide range
of measured values is included that varies from experiments under rainfed to full irrigation conditions,
from no fertilization to high-fertilization rates, and from year to year. The minimum and the maximum
measured water productivity at a location are indicated by the horizontal bars.
Beside the model performance, the low correlation and, in certain cases, large
differences between measured and modelled WPET, can be attributed to three major
reasons that are related to model input data and scale. Firstly, the modelled water
productivity is calculated with inputs (NDVI, surface albedo and air temperature) that
are averaged over several years, whereas the measurements were conducted during
specific years (Table 4-1); hence, the periods of consideration are not identical. It is
for example known that water productivity is influenced by weather conditions during
the growing season, which may vary strongly from year to year. Secondly, in 7 of the
20 experiments, the harvest indices were not measured. An average of the harvest
indices of the remaining experiments was used instead, and for individual experiments
this can result in an offset of up to 19% (Table 4-1). Thirdly, the WATPRO water
productivity model is applied to 1 by 1 kilometre pixels. The experiments are
conducted at only a fraction of the 1 square kilometre and it is therefore unclear how
representative the experiments are of the surrounding 1 by 1 kilometre of land.
Usually, in field experiments the timing of irrigation and the irrigation water and
fertilizer quantities are varied to study the effects on crop yield, water consumption
and water productivity. The average of all experiments in one study site was
calculated and assumed to be representative of the varying management and field
conditions in the surrounding 1 by 1 kilometre over which water productivity is
modelled. Despite all the shortcomings, it can be seen from Figure 4-4 that in 12 out
of 20 experiments, modelled WPET is within the range of measured minimum and
maximum WPET (i.e. 60%).
– 50 –
– 51 –
the total number of experimental years is between brackets
<<3;[email protected]?/97/
"/[email protected]?/97/
"3??327;[email protected]?/97/
"B993D/[email protected]?/97/
+/[email protected]?/97/
3;3?=<A//;59/[email protected]
79/[email protected]?/39
[email protected]/;2)'
[email protected]/;2)'
(/@683;A)[email protected]/;
"[email protected]?/
/B:6/?2A/;2<;[email protected]
3E=3?7:3;A/9F3/[email protected]
/C5 :/E
/C5 :/E
Table 4-1: Reported values of water productivity (WPET) and harvest index (Hi) for wheat, and the modelled water productivity. The experiments were first
summarized and reported in Zwart and Bastiaanssen (2004). Experimental results reported in literature sources after 2003 were added to this data base.
The comparison of field measured versus WATPRO modelled WPET in Sirsa District,
India is depicted in Figure 4-6. One out of five sites was located outside the area
classified as wheat (see Figure 4-5) and could not be compared with the modelling
results. The remaining four sites, however, showed acceptable results. Measured
values varied from 1.22 to 1.56 kg m-3 and their differences from modelled WPET
were small, varying from only 1.6 to 6.9%. During the same study by Van Dam and
Malik (2003), the water productivity of wheat, rice and cotton was assessed spatially
using the SEBAL algorithm. The water productivity map of wheat is depicted in
Figure 4-5A. This map was compared with the WATPRO map for wheat (Figure
4-5B). Since both the SEBAL and the WATPRO models are based on the same
principles, a comparison will only show whether the simplifications and assumptions
in the WATPRO model are justified.
Figure 4-5: Water productivity (kg m-3) in Sirsa District, Haryana, India, modelled and validated in
2001-02 (a) and WATPRO WPET during this study (b). The white line indicates the administrative
boundaries of Sirsa District.
– 52 –
y = 0.98x + 0.22
WPET WATPRO (kg m -3 )
y = 1.24x - 0.36
WP ET (Van Dam and Malik, 2003) (kg m-3)
Figure 4-6: Relationship between SEBAL water productivity (◊), measured water productivity (□) (Van
Dam and Malik, 2003) and WATPRO WPET. The SEBAL map (Figure 4-5A) was divided into discrete
classes of 0.01 mm. For each class the average WPET of the corresponding pixels in the WATPRO
WPET map (Figure 4-5B) was calculated.
A visual interpretation of the WPET maps reveals that both models show very similar
patterns despite the difference in model complexity and the source of satellite data
used. An exception is the upper North West corner of Sirsa District where WATPRO
predicts significantly higher values. As is shown in Figure 4-4, WATPRO WPET is on
average 22% higher than SEBAL WPET. This holds for the entire spectrum between
the 5 and 95% percentiles (slope of 0.98). This analysis shows that WATPRO can
reproduce the same variation as SEBAL does, but it is unclear what causes the higher
values predicted by WATPRO. Since the offset is 22% for the entire range, an
explanation could be related to the different NDVI and surface albedo data sets that
were used. The calibration of the NOAA-AVHRR may have resulted in lower NDVI
and surface albedo values. Moreover only 6 images - 1 per month - were used to cover
the wheat season, whereas in the WATPRO application 3 images per month were
used. The good correlation with the field measurements and the strong correlation
spatially with the SEBAL results provide confidence that the WATPRO model is able
to predict the spatial variation of WPET from limited data resources. This is essential
for preparing local action plans that aim to improve water productivity.
4.5 Conclusions
The WATPRO model that was developed to estimate water productivity can be
applied globally with simple input data, and provides a new approach to water
productivity modelling. Spatial inputs are entirely derived from global remote sensing
data sets of NDVI and surface albedo. The growing season of wheat was determined
accurately from the same NDVI data set. Modelling of water balance components
mostly requires detailed inputs, including actual evapotranspiration and crop
production which can only be estimated with a high degree of uncertainty. Growing
seasons are usually obtained from data sets that provide country-wide averages. The
WATPRO model on the other hand considers the lumped result of all the hydrological
– 53 –
and crop production processes, which are captured in remote sensing data sets, to
estimate water productivity directly and without complex procedures. Although the
simplicity of the model and its inputs provides significant advantages over traditional
modelling, the remote sensing approach does not allow e.g. scenario analysis such as
the impact of climate change or changing water availability on water productivity. By
definition, the use of remote sensing allows only the investigation of past events.
Another drawback to the WATPRO model is its dependency on the harvest index.
Traditional models include modules to estimate the harvest index based on daily
calculations of water and or nutrient stress functions throughout the crop season. Such
models, if applied to field studies, are calibrated against measured yields, or, when
applied for regional studies, are calibrated against census data, , which is not possible
for a model that directly estimates the water productivity. One recommendation for
improvement of the WATPRO model that could enhance the results would be to
include a module to estimate the harvest index from remote sensing data sets. For
example, Moriondo et al. (2007) assumed that negative environmental effects on the
repartition processes were reflected in the NDVI values during pre-anthesis to postanthesis stages. This was captured in a simple mechanistic model that was applied for
wheat yield estimation for two provinces in Italy using the NDVI curves. Fereres and
Soriano (2007) derived simple relationships between biomass production and the
harvest index, as well between post-anthesis transpiration and the harvest index, for
various crops under deficit irrigation practices. The application of such models on a
global scale requires, however, more investigation.
The validation of a model that uses remote sensing data as inputs is complex. Since
the model directly provides WPET and not the yields and evapotranspiration
separately, the model cannot be validated with measured or reported yields and
evapotranspiration. Moreover, there are only limited literature sources available to
validate the modelled WPET. Since none of these sources provide measurements of
both the water productivity and the inputs to the model (e.g. NDVI and surface albedo,
but also light use efficiency or APAR), alternatives had to be used. A fixed maximum
light use efficiency and grain water content were used, and average NDVI and surface
albedo were taken from global remote sensing data sets. Consequently errors are
introduced since 1) the scales of the experimental measurements and the remote
sensing data sets do not match, 2) the years in which the experiments were conducted
differ from the years for which the remote sensing data sets are representative, and 3)
fixed constants are used for parameters that may vary among experiments under
actual conditions. The low correlation that was found between measured and
modelled WPET is therefore not surprising. It is encouraging, however, that the
average values and the ranges of measured and modelled WPET of all locations are
similar. The comparison with measured WPET in farmer’s fields surrounded by other
wheat fields in Sirsa District overcame the scale issue. Over a range of WPET as low as
1.22 to as high as 1.56 kg m-3, values were similar to the WATPRO modelling results
with differences of only 1.6 to 6.9%. The WATPRO model is based on the SEBAL
procedures to estimate biomass production and actual evapotranspiration. The SEBAL
model has been satisfactory validated for various crops and vegetation types under
different environmental conditions (see Bastiaanssen et al. (2005) for an overview of
validation studies, and Zwart and Bastiaanssen (2007) for a validation of the biomass
component). The good spatial correlation with the SEBAL model that was found for
– 54 –
the case study in India also provided confidence that the WATPRO-derived WPET
map was accurate. Ideally, WATPRO results should, however, be spatially validated
with validated maps of water productivity from different sources. This is done in a
global application of WATPRO, where modelling results from two different sources
are compared with global data sets on water productivity (Zwart et al., 2010b).
– 55 –
– 56 –
5 A global benchmark map of water
productivity for rainfed and irrigated
5.1 Introduction
Ensuring food security for a growing population in a world with a changing climate is
a major challenge for the coming decades. Reducing malnutrition and meeting the
food requirements for the projected additional 2-3 billion people, a growth mainly
taking place in developing countries, demands major investments in agriculture.
Alongside with food security comes the challenge to provide agriculture with
sufficient water resources which are required for the advocated increase in production
(Molden et al., 2007). It is estimated that by 2050 an additional 5,600 km3 of
evapotranspired water per year is required to meet the food demands, if no gains in
water productivity are made (Falkenmark and Rockstrom, 2004). Water requirements
are defined here as the vapour flow, or evapotranspiration, that is associated with
plant production and that diffuses into the atmosphere. The physical water
productivity, defined as the crop yield divided by the total water depletion through
evapotranspiration, is a performance indicator to determine whether systems use their
resources efficiently or not (Molden and Sakthivadivel, 1999; Bastiaanssen et al.,
In farmer’s fields, the level of water productivity obtained is determined by many
factors which include management of irrigation water (H. Zhang et al., 1998; Geerts
and Raes, 2009) and fertilizers (Caviglia and Sadras, 2001), selection of crop variety
(Siddique et al., 1990), soil tillage (Mrabet, 2002), mulching (Huang et al., 2005),
planting distance (Giunta and Motzo, 2004), and environmental conditions which
include soil type, water quality (Nangia et al., 2008) and weather conditions (Sadras
and Angus, 2006), amongst others. Farmers often make economics-based decisions
on, for example, the use of fertilizers or modern seeds, or the timing and quantities of
irrigation water applied to his field, which determine the actual water productivity that
is attained. In a related study, a comparison of eight wheat systems worldwide was
conducted to analyse the spatial variation of water productivitities using remote
This chapter is in press with Agricultural Water Management as “Zwart, S.J., W.G.M. Bastiaanssen,
C. de Fraiture, D.J. Molden, 2009. A global benchmark map of water productivity for rainfed and
irrigated wheat”.
– 57 –
sensing derived yields and evapotranspiration at field level (Zwart and Bastiaanssen,
2007). This study revealed that the scope for improvement was highest in areas where
the crop yields were low, and where the spatial variation in water productivity was
The spatial distribution of water productivity at a global scale is, however, poorly
understood. Current knowledge is limited to syntheses of reported water productivites
in scientific literature (e.g. Zwart and Bastiaanssen (2004) for irrigated maize, cotton,
rice and wheat, Sadras and Angus (2006) for wheat under Mediterranean conditions,
and Bouman et al. (2007) for rice). As noted before, these results are highly specific
to the location, the crop, water and soil management practices, and to the years in
which the experiments were conducted. They can therefore only provide plausible
ranges for water productivty that can be expected in an area. A first attempt to provide
a global picture of the spatial distribution of water productivity at a reasonable
resolution was performed by Liu et al. (2007) who used the GEPIC model to simulate
wheat yields and to estimate seasonal evapotranspiration to derive water productivity.
The performance of these models, however, depends largely on good quality input
parameters, such as soil physical characterisitcs, plant parameters, and management
data on irrigation and fertilizers, which may often not be available or accurate at a
global scale. In the absence of these data, general statistics of the FAO have been
Globally, there is a diverse range of economic and environmental conditions where
gains in water productivity can be achieved. It is believed that on a global level there
is considerable scope for improvement in the physical water productivity, but not
everywhere (Molden et al., 2009). Raising water productivity shows most promise in
areas where water productivity is low and the so-called green water resources are
largely unused. Due to the non-linear relationship between yield and water
productivity, areas with low yields show the highest potential increase in water
productivity (Rockstrom et al., 2007). Based on a literature study of field-measured
water productivity, Sadras and Angus (2006) concluded, while focussing on wheat
grown in dry climates, that water productivity of wheat in Australia, USA, China and
the Mediterranean basin was 32 to 44% lower than a maximally attainable values of
2.2 kg m-3. These measurements, however, were conducted in small experimental
plots that may not reflect the actual conditions in agricultural fields. It was
acknowledged by the authors that grain yields are generally overestimated in small
experimental plots.
Global benchmark values and regional statistics on water productivity are
fundamental to understanding where a gap in water productivity exisits, where
systems perform well, and where improvements are still possible. A global water
productivity map may also serve as a start to spatially analyse and explain the
underlying reasons for a gap in water productivity, and where and how action should
be taken to improve the productivity of water. Moreover, global information on water
productivity allows uniform intercomparison of river basins, countries or irrigation
systems, and it could provide a basis for discussions on virtual water trade for global
water saving (De Fraiture et al., 2004; Chapagain et al., 2006).
– 58 –
The purpose of this paper is to develop a global map of the water productivity of
wheat, as wheat is the major staple crop in the world. This map will provide
benchmark levels of actual water productivity under average conditions at the
beginning of this millennium. WATPRO, the water productivity model developed by
Zwart et al. (2010a), will be applied at a global scale and at high resolution of
approximately 1 by 1 kilometre. The WATPRO model estimates water productivity
directly from remote sensing measurements, instead of first determining yields and
water use by evapotranspiration separately. Moreover, it uses remote sensing data
products as input. Traditional models require detailed information on e.g. on-farm
management factors (such as the quantity of irrigation water and fertilizer, and their
quantities and timing of application), soil characteristics, meteorological
measurements, etc. as input to simulate wheat yields, to estimate seasonal
evapotranspiration and assess water productivity (see e.g. Singh et al., 2006;
Immerzeel et al., 2008). These inputs are difficult to obtain and WATPRO basically
circumvents these problems by estimating water productivity directly.
A major challenge in the application of the water productivity model is to determine
on a pixel-by-pixel basis where and when wheat is growing. An innovative approach
is presented to determine the crop establishment and harvest date by mathematically
describing the NDVI time profiles of vegetation. The derived harvest dates are
compared with statistics from reported harvest dates by the FAO (1978) and the
USDA. The information on the start, end and length of the growing season, together
with an existing global wheat dominance map, is used to identify wheat dominant
areas. This paper compares the results with global water productivity modelling
information by Liu et al. (2007) for which they used the GEPIC model, and with the
country-level virtual water content, being the inverse of water productivity, reported
by Chapagain and Hoekstra (2004). The virtual water content was derived by
combining official country statistics of wheat yields from FAOSTAT with the crop
water consumption from AQUASTAT. A first attempt to explain the spatial patterns
of the derived water productivities by WATPRO is made by relating it to seasonal
totals of precipitation and reference evapotranspiration.
5.2 Materials and methods
Water productivity model
Zwart et al. (2010a) developed the WATPRO model to estimate water productivity
(WPET) using remote sensing derived products as inputs. WATPRO directly estimates
WPET and does not need to solve crop yields (Yact) or evapotranspiration (ETact)
explicitly. The model combines the framework of Monteith (1972), that calculates dry
matter production (DM) as a function from the absorbed photosynthetically active
radiation (APAR) and the light use efficiency of the plant ( ε max ), with an energy
balance approach to estimate the latent heat flux ( λE ) that is converted into ETact.
APAR is the total energy that can potentially be absorbed by a plant for photosynthetic
processes (PAR). The fraction of PAR to APAR, f, is linearly related to the NDVI (e.g
Hatfield et al., 1984). PAR is the fraction of the total solar energy ( S IN
) that reaches
the surface of the earth, which is a product of the extraterrestrial radiation ( S EXO
), the
– 59 –
atmospheric transmissivity ( τ SW ) and the PAR/ S IN
fraction, χ . Under actual field
conditions, the maximum light use efficiency ( ε max ) is reduced by water stress and air
temperature (Tair) stress. Two temperature reduction functions (T1, T2) are introduced
to describe the effects of temperature on ε max (Field et al., 1995), while the water
stress scalar is replaced by the evaporative fraction following Bastiaanssen and Ali
(2003). DM is now a function of five spatially distributed inputs, and two inputs that
can be held globally constant:
DM = f NDVI ,τ SW , S EXO
, Tair , Λ, χ , ε max
Actual evapotranspiration (ETact), being equal to the latent heat flux term of the
surface energy balance, can be expressed as the net available energy multiplied by the
evaporative fraction ( Λ ), where Λ describes the energy partitioning. The net available
energy is the net radiation (Rn) minus the soil heat flux. However, over longer peiods
the soil heat flux is negliable compared to the net radiation and it may therefore be
ignored. De Bruin and Stricker (2000) developed an empirircal relationship to
estimate Rn, as a function of S EXO
, τ SW and the broadband surface albedo ( α ). This
implies that ETact can now be estimated as a function of four spatially variable terms:
ETact = f Λ, α ,τ SW , S EXO
It was shown by Zwart et al. (2010a) that in both the production and the
evapotranspiration terms of the water productivity equation, the atmospheric
transmissivity and the evaporative fraction, parameters which are difficult to estimate
spatially and in time, can be ommitted. With seasonal averages as input, WPET can
now be calculated as:
H i ⋅ a ⋅ NDVI + b ⋅ χ ⋅ S EXO
⋅ ε max ⋅ T1 ⋅ T2 ⋅ 0864
(1 − θ )⋅ ((1 − α )⋅ S
− 135 ⋅ 0.35
(kg m-3)
where Hi is the harvest index (-), a and b are dimensionless empirical parameters that
linearly relate PAR to APAR through the NDVI, and θ grain represents the grain
moisture content at harvest, which is kept constant at 0.14 for this study. Eq. 5.3 will
hereafter be referred to as the WATPRO model. For quality reasons, wheat is usually
harvested when this moisture level is reached. The modelling period is the growing
season running from the moment of establishment (t=e) to harvest (t=h). This implies
that the water productivity can be estimated from only four spatial variables: NDVI,
, Tair and α in association with a few constants. Both NDVI and α can be
obtained from standard remote sensing products. Reference is made to Zwart et al.
(2010a), for a full description of the water productivity model and how it is derived
and validated.
– 60 –
– 61 –
@=/A7/[email protected]<9BA7<;
A3:=<?/[email protected]<9BA7<; [email protected];A/A7C3=3?7<2
/C3?/53 4?<: +% 2/A/0/@3 .D/?A /;2 /@A7//;@@3;
'%$(*( @F;[email protected] 2/79F =?<[email protected] /C/79/093 4?<:
2/A3<41?<[email protected]/[email protected]:3;A/;26/[email protected]?7C324<[email protected]@AB2F
"$' 1<;A7;B<[email protected] @B?4/13 /9032< =?<2B1A "<<2F /C/79/0934?<:4A=:<[email protected]/A:<@
[email protected]
[email protected]/C/79/0934?<:6AA=DDD
1/91B9/A324<?3/[email protected];5993;
Table 5-1: Values and sources of input parameters for benchmarking water productivity model (Eq. 5.3)
Global application
The model was applied using 10-day time steps (three per month), in conjunction with
SPOT-VGT NDVI maps. The modelling period was determined using the SPOT-VGT
NDVI time profiles; this method, as well as the method to detect wheat dominant
pixels and areas, is outlined later in this section. An overview of the inputs that were
used in this study to derive the wheat water productivity benchmark map is provided
in Table 5-1. ε max is set at 2.5 g MJ-1 following Bastiaanssen and Ali (2003), whereas
a value of 0.48 is accepted for χ (Moran et al., 1995). The empirical parameters a
and b were measured in wheat experiments reported by Asrar et al. (1992) and
Hatfield et al. (1984), and the average a and b of both experiments, 1.23 and -0.149
respectively, were used for the application of the model. The average extraterrestrial
radiation for each 10 day period is calculated using the middle Julian day of each 10day period following standardized procedures on radiation as described by Allen et al.
Harvest index
The harvest index is a crop-specific parameter that defines the fraction of grain, the
economically valuable part of the plant, to the total above-ground biomass production.
For wheat it is considered to be a fairly stable feature within a given climate zone
unless the plant suffers severe stress from low nutrients, temperature (cold or heat) or
water deficits (see Hay (1995) for a review of the harvest index for various crops). In
plant modelling, environmental stress indicators are used, during specific crop stages,
to reduce, the potential harvest index (see e.g. Raes et al. (2009) for a description of
FAO’s AquaCrop model), or to reduce the speed of grain filling after anthesis
(Fletcher and Jamieson, 2009). In recent decennia the harvest index has been shown to
have improved significantly over time due to plant breeding, and variations are seen
between old and new varieties (e.g. Hay, 1995 and Sinclair, 1998). On a global scale
the use of a single potential harvest index therefore does not seem valid. The
implementation of an increasing harvest index after anthesis requires a precise
knowledge of the start and the length of the phenological stages, which is, at this
stage, not possible to derive from remote sensing imagery at a global scale.
There is no known global database of the harvest index that can be used to compare
values commonly measured in temperate climates or in dry climates. As an
alternative, a literature survey was conducted to find the range of harvest indices that
have been commonly measured. The results are summarized in Appendix 1. An
average harvest index of 0.35 was found, based on 15 experiments globally, but there
is a large range varying from 0.22 (Corbeels et al., 1998) to values higher than 0.50
(Amir et al, 1991). It is likely that in countries with temperate climates, the harvest
index will generally be higher due to lower water and nutrient stress, than in countries
in Mediterranean, semi-arid and arid climate zones, where water shortages and heat
stress occur more often. For the global application, the average harvest index value of
0.35 was used as a fixed constant, though it is realized that this may reduce spatial
variation in the final water productivity levels on the benchmark map.
– 62 –
Air temperature
The mean monthly air temperature values (Tmon) that are necessary in the two
reduction functions for maximum light use efficiency (T1, T2) are derived from the
climate grids of the Water and Climate Atlas from the International Water
Management Institute (IWMI) which are based on the University of East Anglia
climate grids (New et al., 1999). These grids are based on measurements between
1961 and 1990 from over 30,000 meteorological stations. The air temperature during
maximum crop development (Topt), which is required to calculate T1 and T2 (Field et
al., 1995), was determined by selecting the period where SPOT-VGT NDVI during
the growing season was maximum. This assumes that maximum photosynthesis
occurs at favourable temperature conditions.
Surface albedo
The broadband surface albedo was obtained from a global data set of spatially
complete spectral surface albedos (Moody et al., 2005) for all surface types and actual
management conditions. This data set contains cloud-free and gap-filled surface
albedo images recorded every 16 days over a 5-year period from 2000 to 2004. The
spatial resolution of this data set is 1 arc-minute, which is equal to approximately 1.6
km at the equator. The surface albedo images are public domain and available from
the MODIS Atmosphere ftp-site*. The five years that are available are averaged to
obtain a representative average surface albedo input data set. The 16-day images are
linearly interpolated to the 10-daily periods to serve as input for the model.
There are two remote sensing systems that provide global NDVI time series
operationally at an acceptable resolution. SPOT-Vegetation (SPOT-VGT) provides
global cloud free NDVI composite products since April 1998 having a spatial
resolution of approximately 1 km at the equator and a temporal resolution of 10 days.
MODIS/Terra provides 16-daily NDVI products at 250 metre resolution since April
2000. Both sources are public domain and can be downloaded through data portals on
the internet. The MODIS/Terra provides the most detailed data, but in terms of disk
resources and computation requirements, and the lower temporal resolution, the
SPOT-VGT NDVI time series is preferred. An average NDVI is calculated for each of
the 36 10-daily periods from the ten years (April 1998 to March 2008) that were
available. In order to minimise seasonal extremes in the NDVI averages, the
minimum and maximum NDVI value at each location are excluded. The implication
is that the resulting water productivity maps will display longer-term average results.
– 63 –
NDVI (-)
modelling period
modelled (Eq. 4)
1 2
3 1
2 3
2 3 1 2
1 2
3 1
2 3 1
2 3 1
2 3
1 2
1 2 3
Figure 5-1: Algebraic function (Eq. 5.4) used to describe the curves of the NDVI time profiles. A data
set from Yaqui, Mexico (Zwart and Bastiaanssen, 2007) has been used for this example.
Where and when is wheat growing
Wheat is cultivated in a large variety of climates, varying from temperate (e.g.
northwest Europe) to semi-arid (e.g. north Africa), which results in large variation in
the length and timing of the growing season. Country-wise information on cropping
seasons of various crops is available from FAO (1978). The AQUASTAT database
mostly focuses on irrigated crops in dry climates, whereas the Crop Explorer system*
from the United States Department of Agriculture (USDA) provides cropping season
information for selected countries and from different sources. Within larger countries,
such as the USA, China and Australia, the growing season may vary significantly as is
shown by the latter source. As an alternative to these sources, the cropping season of
wheat was determined in this study by using the annual time profiles of SPOT-VGT
NDVI images. The major advantage is that instead of a country-wise average, the
growing season becomes a pixel-wise input to the water productivity model. An
example of a typical NDVI profile of a wheat-dominated area is depicted in Figure
5-1; from bare soil to the establishment of the crop the NDVI values are low, but an
increase is shown after that. Thereafter, the NDVI curve reaches its highest values
during the vegetative stage. During the ripening period the crop becomes yellowish
and the NDVI values drop until the moment it is harvested. An algebraic function was
used, similar to Fischer (1994), which describes the NDVI profile mathematically:
NDVI = a +
 J −d 
 J −g
1 + exp −
 1 + exp −
c 
f 
– 64 –
The function was applied on a global scale for each pixel to match the NDVI profiles
accurately. The Julian day (J) of crop establishment and harvest as well as the length
of the season the NDVI curve can be extracted from the model’s parameterization
results (a to g). A full description of this methodology and the application at a global
scale is given by Zwart et al. (2010c), whereas the section below outlines some of the
major findings.
Figure 5-2: Relationship between harvest dates that were reported by countries (FAO, 1978) and the
estimated country-averaged harvest dates from this study (in Julian days).
As depicted in Figure 5-2 a good correlation (r2=0.93) was found over a large range
between the modelled harvest dates of wheat and the average harvest dates that were
reported by countries following surveys from FAO (1978) (in Julian days, J). Wheat
harvest takes place as early as the start of April in India (J=92) to mid August in
Sweden and Denmark (J=233) in the northern hemisphere, and in the southern
hemisphere from early December (J=335) in South-Africa to mid January (J=16) in
Chile. The largest differences, between 31 and 41 days between modelled and
reported harvest dates, are found for Argentina, Uruguay, South Africa, Afghanistan
and Italy, though there is no systematic under- or overestimation. The average total
length of the growing season, defined as the vegetative period from crop
establishment to harvest, is 160 days (n=89, σ=20.3). This does not include dormancy
periods during the winter season.
This function, together with the WATPRO model, can be applied only in rainfed or
irrigated wheat dominant areas where the interference with other crops is minimal or
absent. Mixed agricultural areas show NDVI time profiles where the values remains
high at the end of the wheat season due to influence of other crops that have peak
NDVI values during this period. By using threshold values on the global maps of the
start, end and length of a vegetative period, wheat dominant areas were extracted.
– 65 –
Non-wheat areas were removed from the global wheat dominance map (Leff et al.,
2004) and non-agricultural areas were excluded manually using GoogleEarth. This
generates more reliable statistics that can be used for water productivity improvement
5.3 Results and discussion
The WATPRO model (Eq. 5.3) was applied on a global scale using the input data
described in Table 5-1 and the crop establishment and harvest dates that were derived
from the NDVI time profiles. The results are shown in Figure 5-3, which depicts the
high resolution water productivity benchmark map for wheat, applicable to the
beginning of this millennium, and in Table 5-2, which summarizes average water
productivity values for the ten major wheat producing countries. There is large global
variation in water productivity, but also within countries. The average global value is
0.86 kg m-3, though WPET may reach values up to 1.80 kg m-3. The global range of
WPET, defined by the 5 and 95% percentiles, is 0.2-1.5 kg m-3, which is slightly lower
than the range of 0.6-1.7 kg m-3 which is based on 412 experiments in small plots
reported in 28 literature sources and summarised by Zwart and Bastiaanssen (2004). It
must be noted here that, since the model was applied on wheat dominant areas only,
these results do not encompass all wheat cultivated regions globally. Agricultural
wheat areas in Ukraine, Russia, Kazakhstan, Canada and USA are expected to be
larger (Leff et al., 2004) than the areas depicted in Figure 5-3.
– 66 –
– 67 –
Figure 5-3: Water productivity benchmark values for water productivity of wheat (in kg m-3).
The highest values of WPET are found in the European temperate climate zone where
country averaged WPET in for example Ireland, United Kingdom, Germany and
France amounts to 1.45, 1.36, 1.35 and 1.42 kg m-3 respectively. Similar values can
also be found in large areas of the Indo-Gangetic Basin in India and Pakistan, in
irrigated wheat systems in Australia, and in wheat areas along the Yellow River in
China. The average WPET in these countries is, however, much lower: Pakistan (0.80),
India (1.06), Australia (1.12) and China (0.82 kg m-3). In North Africa, WPET appears
to be strongly correlated with water availability from irrigation and/or precipitation.
Water abundant regions or systems in e.g. north Tunisia and Egypt depict values
between 1.2 and 1.6 kg m-3, but in rainfed systems with low precipitation rates, values
between 0.4 and 0.8 kg m-3 are more common. Similar patterns can be seen in
countries with Mediterranean and semi-arid climates in the Middle East, Turkey, Iran
and Central-Asia, where WPET was shown to be generally low. On the North
American continent, low WPET values (0.4-0.7 kg m-3) are found in the Canadian
states and the northern states of the USA, which could be possibly related to extensive
farming practices and low seasonal rainfall. When moving south within the wheat
belt, WPET steadily increases to 1.0 to 1.3 kg m-3 in the most southern states. The
average water productivity value of 0.78 kg m-3 is higher than the average value of
0.61 kg m-3 for the North American Plains given by Angus and Sadras (2006) (Figure
5-5). Along the Pacific coast in Oregon, in California’s San Joaquin Valley, or in the
irrigated Yaqui Valley in Mexico, WPET values are generally higher between 1.2 and
1.5 kg m-3. Similar levels of WPET were found by Zwart and Bastiaanssen (2007) for
the Yaqui Valley (1.37) and for Kings County (1.44 kg m-3) in the southern San
Joaquin Valley.
Two different sources were consulted to test the performance of the WATPRO model
at a global scale. First of all, Liu et al. (2007), hereafter abbreviated as LIU, published
a water productivity map for wheat based on modelling efforts with the GEPIC
model. The model relies on generalized country statistics on fertilizer use, water
application, cropping calendars, soil characteristics, etc., which are mostly provided
by the FAO through its FAOSTAT and AQUASTAT portals. Wheat yields were
simulated at a 30 arc-minutes resolution (approximately 50 kilometres at the equator)
and calibrated extensively using country reported wheat yields from FAOSTAT for a
period between 1995 and 2004. The reference evapotranspiration was estimated using
the Hargreaves method and is based on extraterrestrial radiation and air temperature
only (Hargreaves and Samani, 1985). Actual evapotranspiration is the sum of the
modelled evaporation and transpiration, which are computed separately following a
method that is similar to Ritchie (1972). The simulated water productivity for wheat
was reported as a map at 30 arc-minutes resolution and as national averages.
Secondly, Chapagain and Hoekstra (2004), hereafter abbreviated as C&H, calculated
actual evapotranspiration using the FAO method (Allen et al., 1998). This method is
based on multiplying a reference evapotranspiration by a crop-specific crop-factor,
which varies during the season, to estimate the whole-season crop evapotranspiration.
It was acknowledged by the authors that this method may overestimate
evapotranspiration since no crop stress is considered. General cropping calendars, also
provided by AQUASTAT, were used to define the modelling period. Crop yields
were directly obtained from FAOSTAT statistics reported by the member countries.
Since both LIU and C&H use the FAOSTAT country statistics on wheat yields to
– 68 –
calibrate yield simulations (LIU) or to calculate water productivity directly (C&H),
their difference in water productivities will mainly be the result of the different
methods that were used to calculate seasonal actual evapotranspiration.
R2 = 0.64
Liu et al. , 2007
Chapagain and Hoekstra, 2004
R2 = 0.18
Chapagain and Hoekstra, 2004
R2 = 0.25
Liu et al., 2007
Figure 5-4: Performance of (a) WATPRO versus the GEPIC model (Liu et al., 2007), (b) WATPRO
versus the method by Chapagain and Hoekstra (2004), and (c) the GEPIC model against the method of
Chapagain and Hoekstra. The water productivity benchmark map was aggregated to country level to
allow comparison. The country average WPET (kg m-3) of the ten major wheat producing countries is
– 69 –
Table 5-2: Summary of water productivity (WPET in kg m-3) of the ten major wheat producing countries
estimated in this study (WATPRO) and by Liu et al. (2007), LIU, and Chapagain and Hoekstra (2004),
C&H. The percentage difference between WATPRO and LIU and C&H is shown between brackets.
(kg m-3)
averages from FAOSTAT between 2000 and 2007
(kg m-3)
(kg m-3)
Table 5-2 summarizes average water productivity for the ten major wheat producing
countries modelled in this study, and by LIU and C&H. The results for all countries
are depicted in Appendix 2. The modelled average WPET for wheat in this study was
0.93 kg m-3, whereas LIU was on average 2.2% lower (0.91 kg m-3), and C&H 11%
lower (0.83 kg m-3). The latter is consistent with an anticipated overestimation of
evapotranspiration, which reduces water productivity estimates. On a country level, a
good correlation is found with LIU (r2=0.64; see Figure 5-4); in six out of ten
countries, the difference between both models is equal to or less than ten percent.
Major differences are, however, found for Australia (-42%) and Canada (34%). For
Australia, both LIU and C&H provide a lower average WPET of 0.65 and 0.63 kg m-3
respectively. Sadras and Angus (2006), who defined southeast Australia as a megaenvironment for benchmarking water productivity, provide a range of 0.3-1.7, and an
average WPET of 0.99 kg m-3 using five literature sources with experimental
measurements (see also Figure 5-5). Other authors, not cited in the study by Sadras
and Angus, provide similar ranges of measured water productivity for wheat in
southwest Australia: 0.56-1.14 kg m-3 (Siddique et al., 1990) and 0.55-1.65 kg m-3
(Regan et al., 1997). More recently Hochman et al. (2009) modelled water
productivity for 334 fields in various regions in Australia and average values per
region ranged from 0.83 to 1.04 kg m-3. Hence, the estimates for the Australian
continent in this study seem reasonable. For Canada, LIU provides a 34% higher
WPET of 0.86, whereas the results of this study and C&H are very similar with only a
5% difference. On the other hand Sadras and Angus (2006) show even lower values of
0.61 kg m-3 for the North American Plains, which includes Canada (Figure 5-5). It is
uncertain what causes these differences, but in the case of Canada, the areas that are
compared are significantly different since, due to the interference of other crops, only
a small wheat dominant area could be extracted (see Figure 5-3). The correlation
between WATPRO’s results for the 10 major wheat countries and C&H is very low
(r2=0.18) and values are on average 11% lower for C&H. Also the relation between
LIU and C&H is low (r2=0.25). Only in Germany and Turkey was a good agreement
(less than 10% difference) obtained among all three models.
– 70 –
North American
China Loess Plateau
Mediterranean basin
SE Australia
Angus and Sadras (2006)
water productivity (kg m )
Figure 5-5: Water productivity for four “mega-environments” from field experimental results in
various literature sources (summarized and reported by Angus and Sadras, 2006), and water
productivity estimated by WATPRO. The black horizontal bars indicate the standard deviation, while
the average values are shown in white numerals.
The large variations in water productivity may offer significant scope for
improvement in many locations in the world. In order to be able to define where
improvements are possible and what measures ought to be taken to create higher
levels of water productivity, the factors that determine the current levels of water
productivity must be understood. Plant growth and water use are driven by
environmental conditions that can or cannot be controlled by humans, and include
carbon availability from the air (Hsiao and Bradford, 1983), plant water availability
(Tanner and Sinclair, 1983), soil fertility (Nangia et al., 2008) and climate (De Wit,
1958;), amongst others. In this study, two variables, the climate and the water
availability, are discussed. Several authors have pointed out the conservative
behaviour of water productivity to the atmospheric demands which can be represented
by the vapour pressure deficit, D (Bierhuizen en Slatyer, 1965; Tanner and Sinclair,
1983). Studies in Argentina (Abbate et al., 2004) and in South Eastern Australia
(Rodriguez and Sadras, 2007) showed a non-linear inverse negative relation between
D and WPET. More recently, it was argued that D should be replaced by the
accumulated reference evapotranspiration, ET0, in order to normalize for the climate
(see Asseng and Hsiao (2000) and Steduto and Albrizio (2005)). The latter was tested
in this study by comparing WPET with the seasonal accumulated ETo. Monthly ETo
maps of the Climate and Water Atlas of the International Water Management Institute
(IWMI) were used, together with the maps of the start and end of the growing season
to calculate the seasonal ETo so that the same period for the growing season is
considered. The seasonal ETo map was divided into zones of 10 mm steps. For each of
these zones, the average WPET of the corresponding pixels of the WPET map was
calculated. The result, depicted in Figure 5-6, shows the inverse negative relation
between WPET and ETo. In areas with seasonal ETo lower than 300-450 mm,
associated with the temperate climates (e.g. France, United Kingdom), average WPET
is fairly constant between 1.1 to 1.2 kg m-3. Between 450 and 750 mm, the average
WPET decreases almost linearly by 46% until 0.59 kg m-3. In the arid climates where
– 71 –
atmospheric demands during the wheat growing season are even higher than 750 mm
(such as in regions in North Africa or Central Asia), average levels of WPET vary
between 0.52 and 0.60 kg m-3. The reference evapotranspiration is a climatic factor
that cannot be controlled by farmers unless growing seasons are adjusted or when
crops are grown in climate controlled greenhouses. The figure does show however,
that climatic or weather conditions affect the levels of water productivity
average water productivity, WPET (kg m-3)
Seasonal reference evapotranspiration, ET 0 (mm)
Figure 5-6: The water productivity versus the seasonal reference evapotranspiration (ET0).
In a similar way, the effect of water availability on WPET was analysed using spatial
precipitation patterns from the Tropical Rainfall Measurement Mission (TRMM). It is
acknowledged that by substituting plant water availability with precipitation, several
processes are ignored such as the initial soil water content at the start of the season,
groundwater use through capillary rise, irrigation and run-off. However, since
precipitation is in many regions the only source of water, it serves as a first step to
understanding global variations in water productivity related to the availability of
water. Seasonal precipitation was obtained from monthly calibrated precipitation
products from TRMM that are freely available from NASA*. The maps cover the
globe between 50 degrees latitude north and south, thereby omitting wheat growing
areas in the analysis that are located in e.g. Great Britain, Ireland, Denmark and
Sweden. A representative year was created by averaging the monthly maps for the
years 2000 to 2008, and computing the total precipitation during the wheat season.
– 72 –
average WPET (kg m-3)
seasonal precipitation (mm)
Figure 5-7. The water productivity versus the seasonal precipitation from calibrated monthly products
provided by Tropical Rainfall Measurement Mission (TRMM).
As can be seen from Figure 5-7, precipitation strongly impacts the levels of water
productivity of wheat. In areas where precipitation is lower than approximately 130
mm, average water productivity levels are higher than 1.0 kg m-3. These are usually
irrigated systems in arid climates. From 130 to approximately 300 mm of seasonal
precipitation, the average WPET strongly increases from 0.6 to 1.0 kg m-3. After 300
mm, average WPET continues to increase, but with a lower slope. After 600 mm
seasonal precipitation, the trend shows mixed results, probably caused by the smaller
selection of wheat areas with these high precipitation rates. The graph shows the
importance of water availability for water productivity. Major improvements can be
made in low rainfall zones, where seasonal precipitation is lower than 300 mm.
Similar conclusions were drawn by Rockstrom et al. (2007) who found that major
improvements can be achieved in rainfed agriculture with low yields. Large gains in
water productivity can also be achieved with supplemental irrigation. Zwart and
Bastiaanssen (2004) showed for several experiments that water productivity is low
under rainfed conditions, but it may increase sharply if supplemental irrigation is
applied. At full irrigation or over-supply of irrigation water, the water productivity
remains constant or it even decreases under certain conditions. Under such conditions,
deficit irrigation practices are a promising option (Fereres and Soriano, 2004),
although the link between water stress and water productivity varies among crops
(Geerts and Raes, 2009).
5.4 Conclusions
The approach used in this study to assess water productivity is entirely new. The
WATPRO model, designed to use only spatial inputs from remote sensing data, was
applied on a global scale at high resolution. Remote sensing data sets that were
averaged over multiple years to represent the normal situation at the beginning of the
millennium, were used to benchmark water productivity of wheat. The growing
season of wheat was derived from the NDVI profiles on a pixel-by-pixel basis. The
– 73 –
approach thereby differs from traditional modelling studies, which require inputs on
climatology, soils and agronomic practices that are usually not available at the spatial
detail for which they are applied. WATPRO is based on the actual situation in the
field, which is captured in the surface albedo and the NDVI. It has therefore much
lower data requirements compared to traditional models. A major constraint in the
application is that WATPRO could be applied only for wheat dominant areas and not
for areas with mixed agricultural patterns. A mixture of crops makes it impossible to
map the harvest date accurately and the NDVI data is then mixed with different NDVI
signatures from surrounding lands. The conversion of dry matter production into the
harvestable yield component, through the grain moisture content and the harvest
index, provides a source of error. There are no known databases for the harvest index
and there is no generally accepted relationship between total biomass production and
the harvest index. In this study a fixed harvest index of 0.35 was used, based on a
review of 15 experiments globally, though it was shown that the majority of the
harvest index values varies from 0.25 to 0.45.
It was found that the spatial variation of water productivity of wheat at a global scale
is high, both between and within countries. Values range from as low as 0.2 up to 1.8
kg m-3, indicating that there is a large scope for improvement. As a first step to
understanding this spatial variation, the water productivity map was linked to global
data sets of rainfall and reference evapotranspiration to analyse water availability and
climatic demand. The analysis shows that the highest levels of water productivity are
achieved in areas with high seasonal rainfall totals and low seasonal reference
evapotranspiration. However, in drier climates with higher reference
evapotranspiration and lower rainfall totals, water productivity levels decreased
rapidly unless irrigation was applied. To gain a full understanding of the current levels
of water productivity that were found in this study, and to define the scope for
improvement for regions or countries, the analysis should be expanded to include
other parameters. Soil fertility and water availability are probably the most dominant
factors that affect water productivity, although various agronomic practices (selection
of variety, planting distance, etc.) will also have an impact. Soil fertility is linked to
soil type and the application of fertilizers. The FAO holds a global database of
fertilizer use by country and also a global map of major soil types, which could both
be linked to the water productivity benchmark map. Precipitation was used in this
study to analyse the impact of water availability on water productivity. Although a
major part of the global wheat production is cultivated under rainfed conditions, the
impact of irrigation on water productivity was clearly visible, and a future analysis
should therefore distinguish between rainfed and irrigated agriculture. This will also
improve understanding of the impact of measures like supplemental irrigation and
water harvesting on water productivity in different regions under varying
environmental conditions. A major advantage of the WATPRO/remote sensing
modelling approach is that the factors affecting water productivity are not input to the
model, thereby allowing a fully independent analysis. By changing the harvest index,
and by updating the constants a, b, ε max and θ grain , WATPRO can also be applied to
other crops. Finally, the WATPRO results facilitate the planning of food production in
relation to limited water resources for agriculture. It provides insights into virtual
water trade and food security issues. A thorough understanding of the complex
processes that lead to certain levels of water productivity will assist policy and
– 74 –
decision makers to define priority areas, to set goals for improvement, and to define
and justify the type of investment or measures that are made to make agriculture more
productive under increasing pressure on fresh water resources.
– 75 –
Appendix 1: Harvest index, Hi
Depicted below is the relationship between field measured wheat yield and dry matter
production with data from 18 different locations under varying regional conditions
(climate, soil) and varying local conditions (varying irrigation and/or fertilizer
management, wheat variety, etc.). See the table below for an overview of the
locations, the years of measurement and the literature source. One point represents the
average yield and dry matter production from more than one experiment with the
same irrigation and/or fertilizer management but during different years with varying
weather conditions). A total of 196 points is included, but as one point represents the
average of multiple year experiments, this graph may reflect the results of
measurements from more than 500 individual experimental plots. The frequency
histogram of all measurements is also depicted below: values vary from 0.20 to 0.52,
whereas the average harvest index of all experiments amounts to 0.35.
y = 0.45 x
y = 0.35 x
R = 0.93
y = 0.25 x
New Delhi
Tel Hadya
harvest index
experimental years
– 76 –
Caviglia and Sadras, 2001
Regan et al., 1997
Regan et al., 1997
Rahman, 1995
Ali et al., 2007
J. Zhang et al., 1998
Huang et al., 2005
Wang et al., 2001
Zhang et al, 2006
Aggarwal et al., 1986
Amir et al., 1991
Corbeels et al., 1998
Mrabet, 2002
H. Zhang et al., 1998
Sezen and Yazar, 2006
dry matter production (kg/ha)
grain yield (kg/ha)
Appendix 2: Country average WPET
South Africa
– 77 –
(kg m-3)
(kg m-3)
(kg m-3)
(kg m-3)
(kg m-3)
(kg m-3)
Russian Federation
United Kingdom
Czech Republic
Saudi Arabia
Modelled wheat water productivity (kg m-3) from this study (WATPRO) and the
GEPIC model (LIU, Liu et al., 2007), and from a combination of FAOSTAT and
AQUASTAT national statistics (C&H, Chapagain and Hoekstra, 2004). The 64
countries are sorted in descending order according to the total national average
production between 2000 and 2007 (in Mega-tonnes).
– 78 –
6 Discussion and conclusions
6.1 Introduction
The major goal of this research was to benchmark the physical water productivity in
agriculture at various spatial scales. Benchmarking should thus provide values of
water productivity that are currently attained. It forms a starting point for future
analysis and by understanding the limiting and constraining factors for higher water
productivities, the scope for improvement can be defined at e.g. regional or national
level. The benchmarking of water productivity was achieved by 1) using the existing
literature for a field scale analysis, by 2) using an established model (SEBAL) to
determine regional variation, and by 3) developing a model (WATPRO) to benchmark
water productivity at a global level. These results were analysed and related to
climatic conditions (evaporative atmospheric demand), soil fertility (the use of
nitrogen), and water availability (irrigation and precipitation). This yielded
recommendations for future actions on where and how to promote and improve the
productive use of fresh water resources.
6.2 The current levels of water productivity
Field scale
The first step was to establish a water productivity database for the four major crops
in the world, namely wheat, rice, cotton and maize. Results from field experiments
reported in the international literature over the last 25 years were consolidated in a
database to provide up-to-date ranges of feasible water productivity values. The most
researched crop is wheat, followed by maize, rice and cotton. The ranges found were
higher than those reported some 25 years earlier in the FAO33 publication by
Doorenbos and Kassam (1979). For example, this research provides a plausible range
of 0.6 - 1.7 kg m-3 for water productivity of wheat (with an average of 1.1 kg m-3),
whereas a much lower range of 0.8 - 1.0 was provided by FAO33 in the 1970’s. For
the other three researched crops, it was also found that the water productivity values
in FAO33 are on the conservative side. This might partially be related to the
development of crops that are able to produce higher yields, and to improved soil
fertility and water management.
However, caution must be exercised when interpreting these findings. First of all, the
results encompass data from trials that are conducted in rather small experimental
fields under optimal conditions. It was noted before by Davidson (1962) that grain
yields are usually overestimated in small plots when compared to farmer’s fields.
Sadras and Angus (2006) detected a difference of 22% between water productivity
measured in large fields in comparison to experimental plots. Secondly, the methods
– 79 –
that are used to estimate actual evapotranspiration are usually based on a water
balance approach. The different terms of the water balance and the soil moisture
storage difference are measured or estimated, and actual evapotranspiration is
estimated as a residual term. The interaction between the root zone and groundwater
is difficult to measure and often ignored. Also, the process of rainfall interception is
often disregarded (Savenije, 2004). Thirdly, from several experiments that are
included in the database, the aim was to evaluate the impact of measures to increase
water productivity such as mulching or plastic covers for example, to reduce
evaporation from the soil (e.g. Baumhardt and Jones, 2002). Such results may
therefore not reflect the general agronomic practices by farmers in a region, but
represent only an optimal water productivity that is feasible under specific
management practices. Finally, it can be noticed that the results are mainly conducted
in drier areas where water is a scarce resource. Almost no experimental results are
reported from the temperate areas with abundant water supply, although several
authors have stressed the importance of the influence of climate on water productivity
(Bierhuizen and Slatyer, 1965; Tanner and Sinclair, 1983; Abbate et al., 2005;
Steduto et al., 2007).
Regional scale
The large differences in water productivity at the local scale, and the fact that only
little information is available for temperate climates, were addressed by analysing
eight global agricultural regions where wheat is dominant. The SEBAL model was
applied to assess water productivity spatially with a high spatial resolution of 30 by 30
metres. This study revealed that the system-wide average water productivity in five
out of the eight systems was higher than the global average of 1.1 kg m-3. This
included diverse regions such as the Nile Delta in Egypt (1.52 kg m-3), Oldambt
region in the Netherlands (1.39 kg m-3) and the irrigated Yaqui Valley in northwestern Mexico (1.37 kg m-3). The highest values, 1.6 kg m-3 and higher, were found
in the irrigated systems of the Nile Delta and the Yaqui Valley. Water productivity of
wheat in Sindh Province in Pakistan and Hebei Province in China showed the lowest
average values of 0.54 and 0.64 kg m-3 respectively. The average of the eight systems
was, however, equal to the global average of 1.1 kg m-3.
Another important finding is that water productivity appears to be tightly related to
crop yields rather than water consumption. System average seasonal actual
evapotranspiration varied between 355 (-11%) and 467 mm (+19% from the average
of the eight systems of 400 mm), whereas average crop yields vary from 2.5 (-43%)
up to 5.7 tonnes (+52% from the average of 4.4 tonnes). It was shown by Steduto et
al. (2007) that the relation between biomass production and transpiration is fairly
linear when normalised for atmospheric demand. This implies that in systems with
high yields, the total water consumption is more beneficial; as a result of the linear
transpiration-biomass production relationship, the beneficial transpiration in the high
production systems is higher. Since our results indicate that total water consumption is
fairly constant, the fraction of the non-beneficial soil evaporation must be lower.
Therefore, interventions aiming at improving yields may positively impact the
beneficial water consumption.
– 80 –
Global scale
The SEBAL approach allows objective spatial analysis and comparison of crop
systems at high resolution. The major drawback, however, is that specific years are
analysed. It is well known that weather conditions during a growing season, or water
availability, can vary greatly from year to year, and therefore have a strong impact on
the values that are attained. The WATPRO model, which can be considered a
simplified derivative of SEBAL, was developed, tested and applied at a global scale
using remote sensing data sets comprising multiple-year averages of surface albedo
and NDVI. The input data sets represented the average conditions at the beginning of
this millennium, thereby allowing a global benchmark study with the same conditions
for each pixel that was analysed. This global analysis revealed that the highest water
productivity values at national level are obtained in West European countries such as
Ireland (1.45), the United Kingdom (1.36), Germany (1.35) and France (1.42 kg m-3).
It is, however, not uncommon for irrigated wheat in drier countries to attain water
productivity values that are similar to those mentioned for Western Europe, though on
average the water productivity in such water scarce countries is lower due to the low
values associated with rainfed agriculture. Examples of major wheat growers in dry
regions are Pakistan (0.80), India (1.06), Australia (1.12), Egypt, (1.22) and China
(0.82 kg m-3). Regions where water productivity is very low include Central-Asia, the
Middle-East (with the exception of Egypt and Israel), Iran, Turkey and North-Africa,
where country averages range from approximately 0.4 to 0.7 kg m-3. The global
average water productivity value amounts to 0.86 kg of wheat grain produced per
cubic metre of water evapotranspired. This is lower than the global average value in
prior studies reported in the literature, but is well within the range of 0.8 to 1.0 kg m-3
that was given by FAO33.
When estimating water productivity, a significant source of error may be introduced
when converting the seasonal biomass production into harvestable yields by means of
a crop specific harvest index. For the global (wheat) application, a fixed harvest index
of 0.35 was used, which was based on an average of results found in 18 experiments
from different regions of the Earth. The range of harvest index (0.25-0.45) is,
however, large.. For a study of eight different wheat systems (chapter 3), average
harvest indices varying from 0.36 to 0.40, were applied. These harvest indices were
based on research reported in the literature for the specific regions. The harvest index
is not fixed, however, but spatially variable due to a large number of influences which
are related to the plant’s growth. Water stress during the phenological periods of
flowering and grain-filling affects the numbers of grains being formed and their
accumulating weight (Passioura, 2005). However, crop variety, nutrient status and
planting distance, all closely related to the farmer’s agronomic decisions, also affect
the fractions of total biomass and the weight of the grains. A consequence of the
spatial variation in the harvest index, as well as the absolute values of the harvest
index, the results from the application of the SEBAL and WATPRO models at
regional and global scale respectively, may be higher. The work by Fereres and
Soriano (2007) provides some quantified relations between total biomass production
and the harvest index, for varying degrees of water supply and vapour pressure
deficits. However, these relations appear to be highly variable between crops. Hence,
more research is required to explain variations in harvest index.
– 81 –
The need to improve water productivity to safeguard future food production and
provide water to other users of scarce freshwater resources has been outlined
extensively in chapter 1. Appropriate measures need to be designed and implemented
that focus on real water savings in agriculture while maintaining production, and on
improving crop productivity while increasing water productivity. This requires major
investments in the agricultural sector that should target areas where improvements are
actually feasible. The focus should be on areas where a gap in water productivity
exists. This raises two major questions that need to be answered. First of all, what is
the actual (benchmark) and the potential water productivity in a specific location, and
scope for improvement? Secondly, what are the underlying reasons for the gap in
water productivity?
6.3 Defining the scope for improvement
This work has shown that large variations in measured and modelled water
productivity exist between experiments at various locations globally, between fields
within specific regions, and between countries globally. It is uncertain what the
potential water productivity is, and whether such a value can be attained at every
location. The field scale water productivity database showed maximum measured
values of 2.6 kg m-3 for wheat, though the 98% percentile of the frequency histogram
yielded a value of 2.2 kg m-3. Angus and Van Herwaarden (2001) used a frontier
concept on a scatter plot of measured evapotranspiration and wheat grain yields in
rainfed areas of Australia. Based on data published in 1984, they found a maximum
water productivity of 2.0 kg m-3. In a similar approach to determine maximum water
productivity using more recent data from Mediterranean climates in South-eastern
Australia, China’s Loess Plateau, the Mediterranean Basin and the North American
Great Plains, Sadras and Angus (2006) found a maximum value of 2.2 kg m-3. For
each of the four “mega-environments” the authors also define the scope for
improvement to reach the attainable maximum values in each region, which ranged
from 32 to 44%. Their findings are based on experimental results from a limited
number of sites for each mega-environment, and it is therefore questionable how
realistic these ranges are. It is, however, unlikely that all farmers can attain the
potential water productivity for many reasons which include spatial variation in
environmental conditions (climate, soil) within a region (e.g. Abbate et al, 2005;
Rodriguez and Sadras, 2007; Sadras and Rodriguez, 2007), and the discrepancy
between farmers objectives of attaining maximum yields or economic profit instead of
maximum water productivity. For example, Oweis et al. (1998) showed that under
Mediterranean conditions the maximum water productivity is attained with suboptimal wheat yields.
A different (remote sensing and modelling) approach was used in this study to
understand where improvements in water productivity are feasible and what the scope
for improvement is. The magnitude of variation within a system, defined by the
coefficient of variation (CV), and the maximum value, defined as the 98% value of
the frequency histogram, needs to be known. The frequency distribution of modelled
water productivity was analysed and a percentage potential increase was defined by
linearly forcing the coefficient of variation to 5% while maintaining the maximum
value found (chapter 4). This implicitly assumes that 1) within a system some
– 82 –
variation continues to exist due to varying environmental conditions, and 2) the
maximally attainable value is defined under actual conditions in a specific year, rather
than by a hypothetical value under optimal, but uncontrollable environmental
conditions. The scope for improvement was then analysed for specific years and under
actual conditions. This study revealed that only a limited scope for improvement is
possible in the irrigated wheat systems investigated in India, Egypt, USA and the
Netherlands, whereas the largest gains could be made in Pakistan (+24%) and in both
systems in China (+34% and +35% respectively). These eight systems depict,
however, only a fraction of the total wheat growing areas, and they are not necessarily
representative of wheat grown in specific countries or agro-climatic zones. The
analysis with the WATPRO model did show that there is a large range in water
productivity globally, and that scope for improvement may be assumed, but not
everywhere. This requires a better understanding of the range in current water
productivity levels attained in given regions, countries or river basins.
6.4 Explaining the variation and options for improvement
The underlying reasons for variations in water productivity were discussed and
outlined extensively in the previous chapters. Controllable agronomic choices by
farmers such as the choice of a crop variety (drought resistance, harvest index, length
of growing season), planting distance, nutrient availability (fertilization), water
availability (quantity and timing of irrigations, water harvesting techniques), soil
tillage (ploughing, mulching, weeding), etc. all affect the water productivity attained
in the field. This results in a high spatial variation which was shown in the review of
reported experimental results, and the regional analysis of wheat systems. Noncontrollable environmental factors that impact on water productivity include climate
(evaporative demand from the atmosphere) and soils. Two controllable factors
(nutrients and water availability) and one uncontrollable factor (climate) were
addressed to explain the variations that were found, and that could lead to measures or
recommendations that promote the productive use of water.
Atmospheric demand
The impact of evaporative demand from the atmosphere is proven by many authors
such as Bierhuizen and Slatyer (1965) who related transpiration efficiency to vapour
pressure deficit, and Steduto et al. (2007) who related vapour pressure deficit and
reference evapotranspiration to biomass transpiration efficiency for sorghum,
sunflower, wheat and chickpea. They concluded that reference evapotranspiration
provides a better relationship than vapour pressure deficit. At a regional or continental
scale, water productivity was shown by Abbate et al. (2005) to be related to climate
conditions for irrigated and rainfed wheat experiments across different locations in
Argentina, and by Rodriguez and Sadras (2007) for wheat grown in various locations
in Australia. In chapter 2 of this study, the maximum water productivity for each
experimental location stored in the database was shown to have a weak, positive
correlation with latitude, which was explained by the more temperate conditions in
latitudes further away from the equator (see Figure 2-3).
– 83 –
This study is the first to relate water productivity to seasonal reference
evapotranspiration at a global scale (see Figure 5-6). In temperate climates where
seasonal reference evapotranspiration is lower than 400 mm, water productivity is on
average twice as high as in arid areas, where reference evapotranspiration exceeds 750
mm. Reference evapotranspiration is a factor that has to be considered when farming
wheat, since simple and affordable measures for farmers to reduce the total seasonal
reference evapotranspiration do not exist. Measures that can be taken are 1) changing
the agricultural growing season to a period with less evaporative demand (e.g. by
changing from summer to winter crops or by earlier sowing), and 2) selecting modern
varieties with a shorter growing season to reduce the total reference
evapotranspiration (Passioura, 2006).
Soil fertility and fertilizers
Soil fertility is a controllable factor that impacts water productivity. It was shown that
the increased use of fertilizers has a positive impact on water productivity. Water
productivity was plotted against the total applied nitrogen (N) over the growing
season. Three sources provided experimental results from Niger (Pandey et al., 2001),
Uruguay (Caviglia and Sadras, 2001) and Syria (Oweis et al., 2000), where both
parameters were measured and reported. Water productivity was low in plots without
fertilizer (usually control plots in experimental trials). In all 3 experiments, water
productivity increased sharply if small amounts of nitrogen (0-80 kg ha-1) were
applied. Thereafter the curve continued to increase, but at slower rates (see Figure
2-5). This indicates that the largest improvements in water productivity can be made
in agricultural areas with low inputs (including fertilizers), rather than in areas where
the use of fertilizers is already high. However, water availability may constrain the
use of fertilizers by farmers. If water diversions for irrigation are unreliable, or rainfall
is highly unpredictable, investments in fertilizer are not profitable or too risky to
invest in. This should be recognized when defining policy for improvement and
investment. In such areas, a combined approach should be adopted where fertilizers
and supplemental irrigation (or irrigation improvement) are promoted to increase
water productivity (Oweis et al., 2000).
– 84 –
water productivity, WPET (kg m )
zone of
high impact
seasonal applied N (kg ha )
Figure 6-1: The impact of applied nitrogen (N) on water productivity levels and the effect of the use of
fertilizers to increase water productivity. Hypothetical data are used.
Water availability and management
Water availability, whether under rainfed or irrigated conditions, strongly impacts the
water productivity values that are found in the field. In this research, only seasonal
totals of rainfall and applied irrigation water were considered. The timing of rainfall
or irrigation events was not investigated, although it is realized that these also impact
crop production and evapotranspiration. At field level, four experiments, two on
maize (Turkey, Oktem et al., 2002; India, Mishra et al., 2001) and the other two on
wheat (USA, Al Kaisi, 1997; China, Zhang et al., 1999), were conducted to measure
the amount of applied irrigation water as well as the water productivity. Although, the
total amounts of water applied in a season varied strongly from no-irrigation (rainfed
conditions) to 900 mm, all four experiments depicted a parabolic function. With no
irrigation water applied, water productivity was low, but it increased strongly when
limited water was added to the root zone. Under highest levels of irrigation, the water
productivity was again lower (see Figure 2-4). Similar finding were reported by
Oweis et al. (2000) who found that maximum wheat yields were obtained at full
irrigation, though maximum water productivity was reached at 2/3 of the seasonal
irrigation water requirement. Wheat yields under stressed conditions were, however,
only slightly lower than under highest irrigation level. Similar conclusions were
drawn by Hargreaves (1975) who showed for different climates that the highest crop
water productivity for cereals is obtained at a moisture adequacy (i.e. ratio of the
actual moisture available to the amount for which the yield is maximum) of 0.394,
which proves that significant water deficits enhance water productivity.
This field scale analysis proves that under rainfed conditions water productivity can
be increased significantly if supplemental irrigation is applied, or if rainwater is
harvested and retained in the root zone. Harvesting of rain water focuses on reducing
– 85 –
losses due to deep drainage, runoff and soil evaporation (Passioura, 2006). Crops with
quick leaf expansion shade the soil earlier, and soil evaporation is therefore lower.
Variation in row spacing, seeding rate, soil cover and soil tillage also affect soil
evaporation and infiltration of water to the root zone (Van Herwaarden and Passioura,
Similarly, the field scale studies showed that water productivity increases when crops
receive less water than the maximum that is required to meet the full crop water
demand (Figure 6-2). Deliberately stressing crops may improve the water productivity
and may have only a limited effect on final crop yields if crops are exposed to water
stress during certain phenological stages (Fereres and Soriano, 2004). Major reasons
for an increased water productivity as a result of deficit irrigation are a reduction of
soil surface evaporation, a better partitioning of reproductive and vegetative biomass
(a higher harvest index), a better synergy between fertilizers and irrigation, and a
reduction of negative growing conditions (waterlogging, anaerobic conditions in the
root zone, pests and diseases). Deficit irrigation diminishes irrigation water
applications and improves water productivity, though it requires high level
management skills and a reliable supply of water to irrigate crops during drought
sensitive phenological stages (flowering, grain filling). It is also believed that droughttolerant and drought-sensitive varieties of crops react differently to deficit irrigation,
and caution should be exercised when exposing crops to deficit irrigation practices
(Geerts and Raes, 2009).
water productivity, WPET (kg m )
zone of
high impact
seasonal applied irrigation water, I (mm)
Figure 6-2: The impact of applied irrigation water (I) on water productivity levels, and the effect of two
irrigation practices (supplemental and deficit irrigation) on increasing water productivity.
Hypothetical data are used.
The analysis above focussed on the effects of irrigation water at field scale. It is
important, however, to recognize that approximately 80% of the crops are cultivated
– 86 –
under rainfed conditions. The global water productivity map was therefore analysed
using global precipitation data sets. It was found that total seasonal precipitation
strongly affects the level of water productivity attained. With approximately 600 mm
precipitation or more, water productivity is around 1.2 kg m-3. With half of this
precipitation, it decreases linearly to approximately 1.0 kg m-3 (Figure 6-3).
Thereafter, water productivity levels decrease faster until levels are reached where
precipitation is insufficient and irrigation water becomes dominant. Unlike the
example in Figure 6-2, the data presented below are based on a global data set that
includes all climate zones where seasonal reference evapotranspiration varies between
300 and 1000 mm. The data in the figure below are not standardized for atmospheric
demand and represent global average conditions. When interpreting this figure, it
should be realized that due to climatic constraints, the scope for improvement is not
the same everywhere, but dependent on climatic conditions during the growing
season. However, the zone where precipitation ranges from roughly 40 to 280 mm is
believed to depict the highest scope for improvement in water productivity.
supplemental irrigation
rain water harvesting
zone of
high impact
water productivity, WPET (kg m )
seasonal precipitation (mm)
Figure 6-3: The impact of water availability from rainfall on water productivity at a global level (see
also Figure 5-7).
6.5 Final considerations
Water productivity in agriculture must be improved and it is generally believed that
this is feasible. But where are the major potentials and what are the constraints? Two
systems may show the largest potentials: rainfed agriculture in region with low and
erratic rainfall, and intensively irrigated areas.
– 87 –
In areas where limited and unreliable rainfall prevails, the use of fertilizers is inhibited
as the costs of fertilizers as well as the risk of failure are high. It was shown, though,
that the largest scope for improvement in water productivity is possible where limited
quantities of fertilizers are used, and where crops are cultivated under rainfed
conditions. Investing in rain water harvesting techniques and/or systems for
supplemental irrigation, in combination with improved agronomic management and
the use of fertilizers, may give a significant boost to the productive use of water
resources within a basin. The impact of a planned intervention on the hydrological
cycle must be considered since increased yields are associated with higher evaporative
losses. This may impact on water availability to downstream users (Seckler, 2006;
Comprehensive Assessment of Water in Agriculture, 2007; Perry, 2007). The
importance of rainfed agriculture to global food production is, however, still
underestimated and it requires a change in conceptual thinking to stress the
importance of investments aiming at upgrading rainfed agriculture (Falkenmark and
Rockstrom, 2006). Sub-Saharan Africa is often mentioned as a region where
significant gains can be made (Rockstrom and Barron, 2007; Rockstrom et al., 2009),
but also Central Asia, Iran, Turkey and the Middle-East were shown from the global
analysis to depict low water productivity (Molden et al., 2010).
Improvements in irrigated agricultural systems can be made using deficit irrigation
strategies. However, the increase in water productivity is thought to be more moderate
than in rainfed areas (Molden et al., 2010). This study revealed that the scope for
improvement is lower in systems with a high average water productivity. Also, the
large scale introduction of deficit irrigation management requires a change in mind set
by farmers to move from maximizing irrigation for maximized yields, to deliberately
stressing crops to maximize water productivity. Moreover, deficit irrigation requires a
higher level of management skills, as yields may go down and the risk of crop failure
increases. This is contradictory to investments in rainfed systems, which aim at
increasing yields and reducing the risks of crop failure, which finds more support
from farmers.
– 88 –
Abbate, P.E., J.L. Dardanelli, M.G. Cantanero, M. Maturano, R.J.M. Melchiori, E.E. Suero,
2004. Climate and water availability effects on water-use efficiency in wheat. Crop
Sci., 44, 474-483.
Abou Kheira, A.A., 2009. Macromanagement of deficit-irrigated peanut with sprinkler
irrigation. Agric. Water Manage. 96, 1409-1420.
Aggarwal, P.K., A.K. Singh, G.S. Chaturvedi, S.K. Sinha, 1986. Performance of wheat and
triticale cultivars in a variable soil-water environment. II. Evapotranspiration, water
use efficiency, harvest index and grain yield. Field Crops Res., 13, 301-315.
Ali, M.H., M.R. Hoque, A.A. Hassan, K. Khair, 2007. Effects of deficit irrigation on yield
water productivity, and economic returns of wheat. Agric. Water Manage., 92, pp
Al-Kaisi, M.M., A. Berrada, M. Stack, 1997. Evaluation of irrigation scheduling program and
spring wheat yield response in southwestern Colorado. Agric. Water Manage. 34,
Allen, R. G., L. S. Pereira, D. Raes, M. Smith. 1998. Crop evapotranspiration: guidelines for
computing crop water requirements. FAO Irrigation and Drainage Paper 56. Food and
Agriculture Organization of the United Nations, Rome.
Amir, J, J. Krikun, D. Orion, J. Putter, S. Klitman, 1991. Wheat production in an arid
environment. 1. Water-use efficiency, as affected by management practices. Field
Crops Res., 27, 351-364.
Anac, M.S., M. Ali Ul, I.H. Tuzel, D. Anac, B. Okur, H. Hakerlerler, 1999. Optimum
irrigation schedules for cotton under deficit irrigation conditions. In: Kirda, C., P.
Moutonnet, C. Hera, D.R. Nielsen (Eds.), Crop yield response to deficit irrigation,
Developments in Plant and Soil Sciences, vol. 84. Kluwer Academic Publishers,
Dordrecht, pp. 196-212.
Anderson, M.C., J.M. Norman, W.P. Kustas, R. Houborg, P.J. Starks, N. Agam, 2008. A
thermal-based remote sensing technique for routine mapping of land-surface carbon,
water and energy fluxes from field to regional scales. Rem. Sens. Environ., 112,
Asrar G., R.B. Myneni, B.J. Choudhury, 1992. Spatial heterogeneity in vegetation canopies
and remote sensing of absorbed photosynthetically radiation: a modelling study. Rem.
Sens. Environ., 41, 85-103.
Asseng, S., T.C. Hsiao, 2000. Canopy CO2 assimilation, energy balance, and water use
efficiency of alfalfa crop before and after cutting. Field Crops Res., 67, 191-209.
Bandyopadhyay, P. K., S. Mallick, 2003. Actual evapotranspiration and crop coefficients of
wheat (Triticum aestivum) under varying moisture levels of humid tropical canal
command area. Agric. Water Manage. 59, 33-47.
Bastiaanssen, W.G.M., 1998. Remote Sensing in water resources management: the state of the
art. International Water Management Institute, Colombo, Sri Lanka,
– 89 –
Bastiaanssen, W.G.M., M.D. Ahmed, Y. Chemin, 2002. Satellite surveillance of evaporative
depletion across the Indus Basin. Water Resour. Res. 38, 1273-1282.
Bastiaanssen, W.G.M., E.J.M. Noordman, H. Pelgrum, G. Davids, B.P. Thoreson, R.G. Allen,
2005. SEBAL model with remotely sensed data to improve water-resources
management under actual field conditions. J. Irr. Drain. Eng., 131, 85-93.
Bastiaanssen, W.G.M., M. Menenti, R. Feddes, A.A.M. Holtslag, 1998. A remote sensing
surface energy balance algorithm for land (SEBAL): 1. Formulation. J. Hydrol., 212213, 198-212.
Bastiaanssen, W.G.M., S. Ali, 2003. A new crop forecasting model based on satellite
measurements applied across the Indus Basin, Pakistan. Agric. Ecosyst. Environ., 94,
Bastiaanssen, W.G.M., S. Thiruvengadachari, R. Sakthivadivel, D.J. Molden, 1999. Satellite
remote sensing for estimating productivities of land and water. Int. J. Water Res.
Dev., 15, 181-196.
Baumhardt R.L., O.R. Jones, 2002. Residue management and tillage effects on soil-water
storage and grain yield of dryland wheat and sorghum for a clay loam in Texas. Soil
Till. Res., 68, 71-82.
Baumhardt, R.L., R.J. Lascano, 1999. Water budget and yield of dryland cotton intercropped
with terminated winter wheat. Agron. J. 91, 922-927.
Bethune, M., N. Austin, S. Maher, 2001. Quantifying the water budget of irrigated rice in the
Shepparton Irrigation Region, Australia. Irr. Sci. 20, 99-105.
Bhuiyan, S.I., M.A. Sattar, M.A.K. Khan, 1995. Improving water use efficiency in rice
irrigation through wet-seeding. Irr. Sci. 16, 1-8.
Bierhuizen, J.F., O.R. Slatyer, 1965. Effect of atmospheric concentration of water vapor and
CO2 in determining transpiration-photosynthesis relationships of cotton leaves.
Agric. Meteorol., 2, 259-270.
Bordovsky, J.P., W.M. Lyle, 1996. Protocol for planned soil water depletion of irrigated
cotton. In: Cramp, C.R., E.J. Sadler, R.E. Yoder (Eds.), Evapotranspiration and
irrigation scheduling: proceedings of the international conference, November 3 - 6,
1996, San Antonio Convention Center, San Antonio, Texas, pp. 201-206.
Bos, M.G., M.A. Burton, D.J. Molden, 2005. Irrigation and drainage performance assessment:
practical guidelines. CABI Publishing, Wallingford, United Kingdom.
Bouaziz A., H. Chekli., 1999. Wheat yield and water use as affected by micro-basins and
weed control on the sloping lands of Morocco. J. Crop Prod., 2, 335 - 351.
Bouman, B.A.M., E. Humphreys, T.P. Tuong, R. Barker, 2007. Rice and Water. Adv. Agron.,
92, 187-237.
Bouthiba, A., P. Debaeke, S.A. Hamoudi, 2008. Varietal differences in the response of durum
wheat (Triticum turgidum L. var. durum) to irrigation strategies in a semi-arid region
of Algeria. Irr. Sci., 26, 239-251.
Briggs, L.J., H.L. Shantz, 1913. The water requirements of plants, l: Investigation in the Great
Plains. 2: A review of literature. U.S. Dept. Agr. Bur. Plant. Ind. Bull., 284: l -48: 285
: 1-96.
Cabangon, R.J., T.P. Tuong, N.B. Abdullah, 2002. Comparing water input and water
productivity of transplanted and direct-seeded rice production systems. Agric. Water
Manage. 57, 11-31.
Caviglia, O.P., V.O. Sadras, 2001. Effect of nitrogen supply on crop conductance, water- and
radiation-use efficiency of wheat. Field Crops Res., 69, 259-266.
Cayci, G., L.K. Heng, H.S. Öztürk, D. Sürek, C. Kütük, M. Salama, 2009. Crop yield and
water use efficiency in semi-arid region of Turkey. Agric Water Manage., 103, 65-72.
– 90 –
Chapagain, A.K., A.Y. Hoekstra, 2004. Water footprints of nations. Value of water: research
report series no. 16. UNESCO-IHE, Delft, the Netherlands.
Chapagain, A.K., A.Y. Hoekstra, H.H.G. Savenije, R. Gautam, 2006. Water saving through
international trade of agricultural products. Hydrol. Earth Syst. Sci., 10, pp.455-468.
Corbeels, M., G. Hofman, O. van Cleemput, 1998. Analysis of water use by wheat grown on a
cracking clay soil in a semi-arid Mediterranean environment: weather and nitrogen
effects. Agric. Water Manage., 38, 147-167.
Cracium, I., M. Cracium, 1999. Water and nitrogen use efficiency under limited water supply
for maize to increase land productivity. In: Kirda, C., P. Moutonnet, C. Hera, D.R.
Nielsen (Eds.), Crop yield response to deficit irrigation, Developments in Plant and
Soil Sciences, vol. 84. Kluwer Academic Publishers, Dordrecht, pp. 87-94.
Davidson, B.R. Crop yields in experiments and on farms. Nature 194, 458-459.
De Bruin, H.A.R., J.N.M. Stricker, 2000. Evaporation of grass under non-restricted soil
moisture conditions. Hydrol. Sci. J., 45, 391-406.
De Fraiture, C., D. Wichelns, 2010. Satisfying future water demands for agriculture. Agric.
Water Manage. (in press)
De Fraiture, C., M. Giordano, Y. Liao, 2008. Biofuels and implications for agricultural water
use: blue impacts of green energy. Water Pol. 10 Suppl. 10, 67-81.
De Fraiture, C., X. Cai., U. Amarasinghe, M. Rosegrant, D. Molden, 2004. Does cereal trade
save water? The impact of virtual water trade on global water use. Comprehensive
Assessment Research Report 4. Colombo, Sri Lanka.
De Wit, C.T., 1958. Transpiration and crop yields. Verslag. Landbouwk. Onderz. 64.6. Inst.
of Biol. and Chem. Res. on Field Crops and Herbage, Wageningen, the Netherlands,
pp. 1-88.
Deju, Z., L. Jingwen, 1993. The water-use efficiency of winter wheat and maize on a saltaffected soil in the Huang Huai Hai river plain of China. Agric. Water Manage. 23,
Del Blanco, I.A., S. Rajaram, W.E. Kronstad, M.P. Reynold, 2000. Physiological
performance of synthetic hexaploid wheat-derived populations. Crop Sci. 40, 12571263.
Dong, B., R. Loeve, Y.H. Li, C.D. Chen, L. Deng, D. Molden, 2001. Water productivity in
the Zhanghe Irrigation System: issues of scale. In: Barker, R., R. Loeve, Y.H. Li, T.P.
Tuong (Eds.), Water-saving irrigation for rice, Proceedings of an international
workshop, 23-25 March 2001, Wuhan, China, pp. 97-115
Doorenbos, J., A.H. Kassam, 1979. Yield response to water. FAO Irrigation and drainage
paper no. 33. Food and Agriculture Organization of the United Nations, Rome, Italy.
Ertek, A., R. Kanber, 2001. Water-use efficiency (WUE) and change in the yield-response
factor (Ky) of cotton irrigated by an irrigation drip system. Turk. J. Agric. For. 25,
111-118 (in Turkish, with English abstract).
Evett, S.R., T.A. Howell, A.D. Schneider, D.R. Upchurch, D.F. Wanjura, 1996. Canopy
temperature based automatic irrigation control. In: Cramp, C.R., E.J. Sadler, R.E.
Yoder (Eds.), Evapotranspiration and irrigation scheduling: proceedings of the
international conference, November 3 - 6, 1996, San Antonio Convention Center, San
Antonio, Texas, pp. 207-213.
Falkenmark, M, J. Rockstrom, 2004. Balancing water for man and nature: the new approach
to eco-hydrology. Earthscan, London.
Falkenmark, M, J. Rockstrom, 2006. The new blue and green water paradigm: breaking new
ground for water resources planning and management. J. Water Resour. Plan.
Manage. 132, 129-132.
– 91 –
FAO, 1978. Crop calendars. FAO Plant and Protection Papers No. 12. Food and Agriculture
Organization of the United Nations. Rome, Italy.
FAO, 2009. How to feed the world in 2050. Issue brief from the High-Level Expert Forum
held in Rome, 12-13 October. FAO, Rome, Italy.
Fengrui, L., Z. Songling, G.T. Geballe, 2000. Water use patterns and agronomic performance
for some cropping systems with and without fallow crops in a semi-arid environment
of northwest China. Agric. Ecosyst. Env. 79, 129-142.
Fereres, E., M.A. Soriano, 2007. Deficit irrigation for reducing agricultural water use. J. Exp.
Bot., 58, 147-159.
Fernández, J. E., F. Moreno, J. M. Murillo, J. A. Cayuela, E. Fernández-Boy, F. Cabrera,
1996. Water use and yield of maize with two levels of nitrogen fertilization in SW
Spain. Agric. Water Manage. 29, 215-233.
Field, C.B., J.T. Randerson, C.M. Malmstrom., 1995. Global net primary production:
combining ecology and remote sensing. Rem. Sens. Environ., 51, 74-88.
Fischer, A., 1994. A model for the seasonal variations of vegetation indices in coarse
resolution data and its inversion to extract crop parameters. Rem. Sens. Environ., 48,
Fletcher, A.L., P.D. Jamieson, 2009. Causes of variation in the rate of increase of wheat
harvest index. Field Crops Res., 113, 268-273.
Garatuza-Payan, J., C.J. Watts, 2005. The use of remote sensing for estimating ET of irrigated
wheat and cotton in Northwest Mexico. Irr. Drain. Syst. 19, 301-320.
Geerts, S, D. Raes, 2009. Deficit irrigation as an on-farm strategy to maximize crop water
productivity in dry areas. Agric. Water Manage., 96, 1275-1284.
Gencoglan, C., A. Yazar, 1999. The effects of deficit irrigation on corn yield and water use
efficiency. Turk. J. Agric. For. 23, 233-241 (in Turkish, with English abstract).
Giunta, F., R. Motzo, 2004. Sowing rate and cultivar affect total biomass and grain yield of
spring triticale (×Triticosecale Wittmack) grown in a Mediterranean-type
environment. Field Crops Res., 87, 179-193.
Grismer, M.E., 2002. Regional cotton lint yield, ETc and water value in Arizona and
California. Agric. Water Manage. 54, 227-242.
Hanks, R.J., 1974. Model for predicting plant yield as influenced by water use. Agron. J. 66,
Hargreaves, G.H., 1975. Moisture availability and crop production. Trans. ASAE, 1, 980984.
Hargreaves, G.H., Z.A. Samani, 1985. Reference crop evapotranspiration from temperature.
Appl. Eng. Agric., 1, 96-99.
Hatfield, J.L., G. Asrar, E.T. Kanemasu, 1984. Intercepted photosynthetically active radiation
estimated by spectral reflectance. Rem. Sens. Environ., 14, 65-75.
Hatfield, J.L., T.J. Sauer, J.H. Prueger, 2001. Managing soils to achieve greater water use
efficiency: a review. Agron. J. 93, 271-280.
Hay, R.K.M., 1995. Harvest index: a review of its use in plant breeding and crop physiology.
Ann. Appl. Biol., 126, 197-216.
Hochman, Z., D. Holzworth, J.R. Hunt, 2009. Potential to improve on-farm wheat yield and
WUE in Australia. Crop Past. Sci., 60, 708-716.
Hoedjes, J.C.B., R.M. Zuurbier, C.J. Watts, 2002. Large aperture scintillometer used over a
homogeneous irrigated area, partly affected by regional advection. Boundary-Layer
Meteorol. 105, 99-117.
Howell, T.A., A. Yazar, A.D. Schneider, D.A. Dusek, K.S. Copeland, 1995. Yield and water
use efficiency of corn in response to LEPA irrigation. Trans. ASAE 38, 1737-1747.
– 92 –
Howell, T.A., K.R. Davis, R.L. McCormick, H. Yamada, V.T. Walhood, D.W. Meek, 1984.
Water use efficiency of narrow row cotton. Irr. Sci. 5, 195-214.
Howell, T.A., M. Meron, K.R. Davis, C.J. Phene, H. Yamada, 1987. Water management of
trickle and furrow irrigated narrow row cotton in the San Joaquin Valley. Appl. Eng.
Agric. 3, 222-227.
Howell, T.A., S.R. Evett, J.A. Tolk, A.D. Schneider, J.L. Steiner, 1996. Evapotranspiration of
corn - Southern High Plains. In: Cramp, C.R., E.J. Sadler, R.E. Yoder (Eds.),
Evapotranspiration and irrigation scheduling: proceedings of the international
conference, November 3 - 6, 1996, San Antonio Convention Center, San Antonio,
Texas, pp. 158-165.
Hsiao, T.C., K.J. Bradford, 1983. Physiological consequences of cellular water deficits. In
H.M. Taylor, W.R. Jordan, T.R. Sinclair (eds.). Limitations to efficient water use in
crop production. ASA, CSSA, and SSSA, Madison, WI.
Huang, Y., L. Chen, B. Fu, Z. Huang, J. Gong, 2005. The wheat yields and water-use
efficiency in the Loess Plateau: straw mulch and irrigation effects. Agric. Water
Manage., 72, 209-222.
Hunsaker, D.J., A.J. Clemmens, D.D. Fangmeier, 1998. Cotton response to high frequency
surface irrigation. Agric. Water Manage. 37, 55-74.
Immerzeel, W.W., A. Gaur, S.J. Zwart, 2008. Integrating remote sensing and a process-based
hydrological model to evaluate water use and productivity in a south Indian
catchment. Agric. Water Manage., 95, 11-24.
Istanbulluoglu, A., 2009. Effects of irrigation regimes on yield and water productivity of
safflower (Carthamus tinctorius L.) under Mediterranean climatic conditions. Agric.
Water Manage. 96, 1792-1798.
Jin, M., R. Zhang, L. Sun, Y. Gao, 1999. Temporal and spatial soil water management: a case
study in the Heilonggang region, PR China. Agric. Water Manage. 42, 173-187.
Jongschaap, R.E.E., 2006. Run-time calibration of simulation models by integrating remote
sensing estimates of leaf area index and canopy nitrogen. Europ. J. Agron. 24, 316324.
Kamilov, B., N. Ibragimov, S. Evett, L. Heng., 2002. Use of neutron probe for investigations
of winter wheat irrigation scheduling in automorphic and semi-hydromorphic soils of
Uzbekistan. In: Proc. International Workshop on Conservation Agriculture for
Sustainable Wheat Production in Rotation with Cotton in Limited Water Resource
Areas, October 13-18, 2002, Tashkent, Uzbekistan.
Kamilov, B., N. Ibragimov, Y. Esanbekov, S. Evett, L. Heng, 2003. Irrigation scheduling
study of drip irrigated cotton by use of soil moisture neutron probe. In: Proceedings
of the UNCGRI/IAEA National workshop "Optimization of water and fertilizer use
for major crops of cotton rotation", December 24-25, 2002, Tashkent, Uzbekistan.
Kang, S., W. Shi, J. Zhang, 2000a. An improved water-use efficiency for maize grown under
regulated deficit irrigation. Field Crops Res. 67, 207-214.
Kang, S.Z., P. Shi, Y.H. Pan, Z.S. Liang, X.T. Hu, J. Zhang, 2000b. Soil water distribution,
uniformity and water-use efficiency under alternate furrow irrigation in arid areas. Irr.
Sci. 19, 181-190.
Karam, F., J. Breidy, C. Stephan, J. Rouphael, 2003. Evapotranspiration, yield and water use
efficiency of drip irrigated corn in the Bekaa Valley of Lebanon. Agric. Water
Manage. 63, 125-137.
Keller, A., J. Keller, 1995. Effective efficiency: A water use efficiency concept for allocating
freshwater resources. Water Resources and Irrigation Division (Discussion Paper 22).
Arlington, Virginia, USA: Winrock International.
– 93 –
Kijne, J.W., R. Barker, D. Molden, 2003. Water Productivity in Agriculture: Limits and
Opportunities for Improvement. CAB International, Wallingford UK.
Kijne, J.W., T.P. Tuong, J. Bennett, B. Bouman, T. Oweis, 2003b. Ensuring food security via
improvement in crop water productivity. In: Challenge program on water and food:
background papers to the full proposal. The Challenge Program on Water and Food
Consortium, Colombo, Sri Lanka.
Kruse, E.G., D.F. Champion, R.E. Yoder, 1991. High saline water-table effect on wheat
irrigation. In: Allen, R.A., T.A. Howell, W.O. Pruitt, I.A. Walter, M.E. Jensen (Eds.),
Lysimeters for evapotranspiration and environmental measurements, Proceedings of
the international symposium on lysimetry, Honolulu, Hawaii, July 23-25, 1991, pp.
Latta J., G.J. O'Leary, 2003. Long-term comparison of rotation and fallow tillage systems of
wheat in Australia. Field Crops Res., 83, 173-190.
Leff, B., N. Ramankutty, J.A. Foley, 2004. Geographic distribution of major crops across the
world. Global Biogeochem. Cycles, 18, GB1009.
Li, F-M., Q-H. Song, H-S. Liu, F-R. Li, X-L. Liu, 2001. Effects of pre-sowing irrigation and
phosphorus application on water use and yield of spring wheat under semi-arid
conditions. Agric. Water Manage. 49, 173-183.
Libardi, P.L., S.O. Moraes, A.M. Saad, Q. de Jong van Lier, O. Vieira, R.L. Tuon, 1999.
Nuclear techniques to evaluate water use of field crops irrigated in different growth
stages. In: Kirda, C., P. Moutonnet, C. Hera, D.R. Nielsen (Eds.), Crop yield response
to deficit irrigation, Developments in Plant and Soil Sciences, vol. 84. Kluwer
Academic Publishers, Dordrecht, pp. 109-120.
Liu, J., J.R. Williams, A.J.B. Zehnder, H. Yang, 2007. GEPIC - modelling wheat yield and
crop water productivity with high resolution on a global scale. Agric. Syst., 94, 478493.
Liu, W.Z., Y.S. Li, 1995. Crop yield response to water and fertilizer in Loess tableland of
China: a field research. Pedosphere 5, 259-266.
Lopez-Bellido, R.J., L. Lopez-Bellido, J. Benitez-Vega, F.J. Lopez-Bellido, 2007. Tillage
system, preceding crop, and nitrogen fertilizer in wheat crop: II. water utilization.
Agron. J., 99, 66-72.
Lobell , D.B., J.I. Ortiz-Monasterio, G.P. Asner, R.L. Naylor, W.P. Falcon, 2005. Combining
field surveys, remote sensing, and regression trees to understand yield variations in an
irrigated wheat landscape. Agron. J. 94, 241-249.
Lobell, D.B., G.P. Asner, J.I. Ortiz-Monasterio, T.L. Benning, 2003. Remote sensing of
regional crop production in the Yaqui Valley, Mexico: estimates and uncertainties.
Agric. Ecosys. Environ. 94, 205-220.
Lobell, D.B., J.I. Ortiz-Monasterio, C.L. Addams, G.P. Asner, 2002. Soil, climate, and
management impacts on regional wheat productivity in Mexico from remote sensing.
Agric. For. Meteor. 114, 31-41.
Mandal, K.G., K.M. Hati, A.K. Mishra, K.K. Bandyopadhyay, M. Mohanty, 2005. Irrigation
and nutrient effects on growth and water-yield relationship of wheat (Triticum
aestivum L.) in central India. J. Agron. Crop Sci., 191, 416-425.
Mishra, H.S., T.R. Rathore, R.C. Pant, 1990. Effect of intermittent irrigation on groundwater
table contribution, irrigation requirement and yield of rice in Mollisols of the Tarai
Region. Agric. Water Manage. 18, 231-241.
Mishra, H.S., T.R. Rathore, U.S. Savita, 2001. Water-use efficiency of irrigated winter maize
under cool water conditions of India. Irr. Sci. 21, 27-33.
– 94 –
Mishra, H.S., T.R. Rathore, V.S. Tomar, 1995. Water use efficiency of irrigated wheat in the
Tarai Region of India. Irr. Sci. 16, 75-80.
Mo, X., S. Liu, Z. Lin, R. Guo, 2009. Regional crop yield, water consumption and water use
efficiency and their responses to climate change in the North China Plain. Agric.
Ecosyst. Environ., 134, 67-78.
Molden, D., H. Murray-Rust, R. Sakhtivadivel, I. Makin. A water-productivity framework for
understanding and action. In: J.W. Kijne, R. Barker, D. Molden (eds), Water
productivity in agriculture: limits and opportunities for improvement. CABI
Publishing, Wallingford, United Kingdom.
Molden, D., R. Sakthivadivel, 1999. Water accounting to assess use and productivity of water.
Int. J. Water Resour. Dev., 15, 55-71.
Molden, D., T. Oweis, P. Steduto, P. Bindraban, M.A. Hanjra, J. Kijne, 2010. Improving
agricultural water productivity: between optimism and caution. Agric. Water Manage.
(in press).
Molden, D., T.Y. Oweis, P. Steduto, J.W. Kijne, M.A. Hanjra, P.S. Bindraban, 2007.
Pathways for increasing agricultural water productivity. In: D.J. Molden (ed.). Water
for food, water for life: a comprehensive assessment of water management in
agriculture. London: Earthscan, and Colombo: International Water Management
Molle, F., P. Wester, P. Hirsch, 2010. River basin closure: processes, implications and
responses. Agric. Water Manage. (in press).
Monteith, J.L., 1972. Solar radiation and productivity in tropical ecosystems. J. Appl. Ecol., 9,
Moody, E. G., M. D. King, S. Platnick, C. B. Schaaf, F. Gao, 2005. Spatially complete global
spectral surface albedos: Value-added datasets derived from Terra MODIS land
products. IEEE Trans. Geosci. Rem. Sens., 43, 144-158.
Moran, M.S., S.J. Maas, P.J. Pinter, 1995. Combining remote sensing and modelling for
estimating surface evaporation and biomass production. Rem. Sens. Rev. 12, 335353.
Mrabet, R., 2002. Wheat yield and water use efficiency under contrasting residue and tillage
management systems in a semiarid area of Morocco. Exp. Agric. 38, 237-248.
Myneni, R.B., D.L. Williams, 1994. On the relationship between FAPAR and NDVI. Rem.
Sens. Environ., 49, 200-211.
Nangia, V., C. de Fraiture, H. Turral, 2008. Water quality implications of raising crop water
productivity. Agric. Water Manage., 95, 825-835.
Navarro Dujmovich, M, C. Bongiorno, A. Confalone, 1996. Evapotranspiration efficiency of
corn grown under rainfed conditions. In: Cramp, C.R., E.J. Sadler, R.E. Yoder (Eds.),
Evapotranspiration and irrigation scheduling: proceedings of the international
conference, November 3 - 6, 1996, San Antonio Convention Center, San Antonio,
Texas, pp. 1056-1057.
New, M., D. Lister, M. Hulme, I. Makin, 2000. A high-resolution data set of surface climate
over global land areas. Clim. Res. 21, 1-25.
New, M., M. Hulme, P.D. Jones, 1999. Representing twentieth century space-time climate
variability. Part 1: development of a 1961-90 mean monthly terrestrial climatology. J.
Clim., 12, 829-856.
Norwood, C.A., 2000. Water use and yield of limited-irrigated and dryland corn. Soil Sci.
Soc. Am. J. 64, 365-370.
– 95 –
Nwadukwe, P.O., V.O. Chude, 1998. Manipulation of the irrigation schedule of rice (Oryza
sativa L.) as a means of maximizing water use efficiency and irrigation efficiency in
the semi-arid tropics. J. Arid Environ. 40, 331-339.
Oktem, A., M. Simsek, A.G. Oktem, 2003. Deficit irrigation effects on sweet corn (Zea mays
saccharata Sturt) with drip irrigation in a semi-arid region. I. Water yield relationship.
Agric. Water Manage. 61, 63-74.
O'Leary G.J., D.J. Connor, 1997. Stubble retention and tillage in a semiarid environment: 3.
Response of wheat. Field Crops Res., 54, 39-50.
Orgaz, F., L. Mateos, E. Fereres, 1992. Season length and cultivar determine the optimum
evapotranspiration deficit in cotton. Agron. J. 84, 700-706.
Oweis, T., H. Zhang, M. Pala, 2000. Water use efficiency of rainfed and irrigated bread wheat
in a Mediterranean environment. Agron. J. 92, 231-238.
Oweis, T., J. M. Pala, Ryan, 1998. Stabilizing rainfed wheat yields with supplemental
irrigation and nitrogen in a Mediterranean-type climate. Agron. J. 90, 672-681.
Pandey, R.K., J.W. Maranville, A. Admou, 2001. Tropical wheat response to irrigation and
nitrogen in a Sahelian environment. I. Grain yield, yield components and water use
efficiency. Europ. J. Agron. 15, 93-105.
Passioura, J, 2006. Increasing crop productivity when water is scarce - from breeding to field
management. Agric. Water Manage. 80, 176-196.
Payero, J.O., D.D. Tarkalson, S. Irmak, D. Davison, J.L. Petersen, 2009. Effect of timing of a
deficit-irrigation allocation on corn evapotranspiration, yield, water use efficiency
and dry mass. Agric. Water Manage. 96, 1387-1397.
Peng , S., R.C. Laza, G.S. Khush, A.L. Sanico, R.M. Visperas, F.V. Garcias, 1998.
Transpiration efficiencies of indica and improved tropical japonica rice grown under
irrigated conditions. Euphytica 103, 103-108.
Perry, C., 2007. Efficient irrigation; Inefficient communication; flawed recommendations. Irr.
Drain. 56, 367-378.
Postel, S.L., 1998. Water for food production: will there be enough in 2025? Biosci. 48, pp
Prieto, D., C. Angueira, 1999. Water stress effect on different growing stages for cotton and
its influence on yield reduction. In: Kirda, C., P. Moutonnet, C. Hera, D.R. Nielsen
(Eds.), Crop yield response to deficit irrigation, Developments in Plant and Soil
Sciences, vol. 84. Kluwer Academic Publishers, Dordrecht, pp. 161-179.
Raes, D., B. Sy, L. van Passel, 1992. The water balance of rice irrigation schemes in the
Senegal river delta. In: Feyen, J., E. Mwendera, M. Badji (Eds.), Advances in
planning design and management of irrigation systems as related to sustainable land
use, Proceedings of an international conference, September 14-17, 1992, Leuven,
Belgium: Volume 2, pp. 835-844.
Raes, D., P. Steduto, T.C. Hsiao, E. Fereres, 2009. AQUACROP Reference manual. FAO,
Land and Water Division, Rome, Italy.
Rahman, S.M., M.I. Khalil, M.F. Ahmed, 1995. Yield-water relations and nitrogen utilization
by wheat in salt-affected soils in Bangladesh. Agric. Water Manage. 28, 49-65.
Regan, K.L., K.H.M. Siddique, D. Tennant, D.G. Abrecht, 1997. Grain yield and water use
efficiency of early maturing wheat in low rainfall Mediterranean environments. Aus.
J. Agric. Res., 48, 595-603.
Rijsberman, F.R., 2006. Water scarcity: fact or fiction? Agric. Water Manage. 80, 5-22.
Ritchie, J.T., 1972. A model for prediciting evaporation from row crop with incomplete
cover. Water Resour. Res., 8, 1204-1213.
– 96 –
Rockstrom, J., M. Lannerstad, M. Falkenmark, 2007. Assessing the water challenge of a new
green revolution in developing countries. Proc. Nat. Acad. Sci., 104, 6253-6260.
Rockström, J., P. Kaumbutho, J. Mwalley, A.W. Nzabi, M. Temesgen, L. Mawenya, J.
Barron, J. Mutua, S. Damgaard-Larsen, 2009. Conservation farming strategies in East
and Southern Africa: Yields and rain water productivity from on-farm action
research. Soil Till. 103, 23-32.
Rodriguez, D., V.O. Sadras, 2007. The limit to wheat water-use efficiency in eastern
Australia. I. Gradients in the radiation environment and atmospheric demand. Aus. J.
Agric. Res. 58, 287-302.
Roel, A., J.H. Heilman, G.N. McCauley, 1999. Water use and plant response in two rice
irrigation methods. Agric. Water Manage. 39, 35-46.
Roygard, J.K.F., M.M. Alley, R. Khosla, 2002. No-till corn yields and water balance in the
Mid-Atlantic Coastal Plain. Agron. J. 94, 612-623.
Sadler, P.J. Bauer, W.J. Busscher, J.A. Millen, 2000. Site-Specific Analysis of a droughted
corn crop: II. Water use and stress. Agron. J. 92, 403-410.
Sadras V.O., J.F. Angus, 2006. Benchmarking water-use efficiency of rainfed wheat in dry
environments. Aus. J. Agric. Res., 57, 847-856.
Sadras, V.O., D. Rodriguez, 2007. The limit to wheat water-use efficiency in eastern
Australia. II. Influence of rainfall patterns. Aus. J. Agric. Res. 58, 657-669.
Saranga, Y., I. Flash, D. Yakir, 1998. Variation in water-use efficiency and its relation to
carbon isotope ratio in cotton. Crop. Sci. 38, 782-787.
Sastri, A.S., B.L. Chandrakar, B.R. Chandrawanshi, 1985. Evapotranspiration and water use
of rice in central India. Int. Rice Res. Newsletter 10, 30-31.
Savenije, H.H.G., 2004. The importance of interception and why we should delete the term
evapotranspiration from our vocabulary. Hydr. Earth Syst. Sci. 18, 1507-1511.
Schneider, A.D., T.A. Howell, 2001. Scheduling deficit wheat irrigation with data from an
evapotranspiration network. Trans. Am. Soc. Agric. Eng., 44, 1617-1623
Seckler, D., 1996. The new era of water resources management: from "dry" to "wet" water
savings. Research report 1. International Irrigation Management Institute, Colombo,
Sri Lanka.
Sezen, S.M., A. Yazar, 1996. Determination of water-yield relationship of wheat under
Cukurova conditions. Turk. J. Agric. For. 20, 41-48 (in Turkish, with English
Sezen, S.M., A. Yazar, 2006. Wheat yield response to line-source sprinkler irrigation in the
arid Southeast Anatolia region of Turkey. Agric. Water Manage., 81, 59-76.
Sharma, K.D., A. Kumar, K.N. Singh, 1990. Effect of irrigation scheduling on growth, yield
and evapotranspiration of wheat in sodic soils. Agric. Water Manage. 18, 267-276.
Sharma, K.S., J.S. Samra, H.P. Singh, 2001. Influence of boundary plantation of poplar
(Populus deltoides M.) on soil-water use and water use efficiency of wheat. Agric.
Water Manage. 51, 173-185.
Shi, Q., X. Zheng, A. Zhu, M. Li, S. Fu, F. Liu, H. Hengsdijk, P. Henstra, P. Bindraban, 2003.
Effects of intermittent irrigation on growth, yield and water use efficiency of rice in a
field lysimeter experiment. In: Proceedings of the international symposium on
transitions in agriculture for enhancing water productivity. September 23-25, 2003.
Killikulam, India.
Shih, S.F., G.S. Rahi, G.H. Snyder, D.S. Harrison, A.G. Smajstrla, 1983. Rice yield, biomass,
and leaf area related to evapotranspiration. Trans. ASAE 82, 1458-1464.
Shunqing, A., L. Gengshan, G. Anhong, 2003. Consumption of available soil water stored at
planting by winter wheat. Agric. Water Manage., 63, 99-107.
– 97 –
Shuttleworth, W.J., 1989, Micrometeorology of temperate and tropical forest. Proc. Roy. Soc.
Lond. B324, 299-334.
Siddique, K.H.M., D. Tennant, M.W. Perry, R.K. Belford, 1990. Water use and water use
efficiency of old and modern cultivars in a Mediterranean-type environment. Aus. J.
Agric. Res., 41, 431-447.
Sinclair, T.R., 1998. Historical changes in harvest index and crop nitrogen accumulation.
Crop Sci., 38, 638-643.
Singh, A.K., B.U. Choudhury, B.A.M. Bouman, 2002. Effects of rice establishments methods
on crop performance, water use, and mineral nitrogen. In: Bouman, B.A.M., H.
Hengsdijk, B. Hardy, P.S. Bindraban, T.P. Tuong, J.K. Ladha (Eds.), Proceedings of
the international workshop on water-wise rice production, 8-11 April 2002, Los
Banos, Philippines. Los Banos (Philippines): International Rice Research Institute
Singh, K.B., P.R. Gajri, V.K. Arora, 2001. Modelling the effects of soil and water
management practices on the water balance and performance of rice. Agric. Water
Manage. 49, 77-95.
Singh, R., J.C. van Dam, R.A. Feddes, 2006. Water productivity analysis of irrigated crops in
Sirsa District, India. Agric. Water Manage., 82, 253-278.
Singh, R.V., H.S. Chauhan, 1996. Irrigation scheduling in wheat under shallow groundwater
table conditions. In: Cramp, C.R., E.J. Sadler, R.E. Yoder (Eds.), Evapotranspiration
and irrigation scheduling: proceedings of the international conference, November 3 6, 1996, San Antonio Convention Center, San Antonio, Texas, pp. 103-108.
Stanhill, G., 1960. The relationship between climate and the transpiration and growth of
pastures. In: Boyle, P. J., L. W. Raymond (Eds.), Proceedings of the Eighth
International Grasslands Congress, University of Reading, Berkshire, England, pp.
Steduto, P, T.C. Hsiao, E. Fereres, 2007. On the conservative behaviour of biomass water
productivity. Irr. Sci., 25, 189-207.
Steduto, P., R. Albrizio, 2005. Resource use efficiency of field-grown sunflower, sorghum,
wheat and chickpea: II. Water use efficiency and comparison with radiation use
efficiency. Agric. Forest Meteorol., 130, 269-281.
Steduto, P., T.C. Hsiao, D. Raes, E. Fereres, 2009. AquaCrop - the FAO crop model to
simulate yield response to water: I. concepts and underlying principles. Agron. J. 101,
pp 426-437.
Steele, D.D., E.C. Stegman, B.L. Gregor, 1994. Field comparison of irrigation scheduling
methods for corn. Trans. ASAE 37, 1197-1203.
Stewart, J.I., R.E. Danielson, R.T. Hanks, E.B. Jackson, R.M. Hagan, W.O. Pruitt, W.T.
Franklin, J.P. Riley, 1977. Optimizing crop production through control of water and
salinity levels in the soil. Utah Water Research Lab. PR. 151-1, Logan, Utah, 191 pp.
Tanner, C.B., T.R. Sinclair. 1983. Efficient water use in crop production: Research or reresearch? In H.M. Taylor et al. (ed.) Limitations to efficient water use in crop
production. ASSA, CSSA, SSSA, Madison, WI.
Tennakoon, S.B., S.P. Milroy, 2003. Crop water use and water use efficiency on irrigated
cotton farms in Australia. Agric. Water Manage. 61, 179-194.
Thiruvengadachari, S., Murthy, C. S., Raju, P.V., 1997. Remote sensing study, Bhakra canal
command area, Haryana State, India: Final report to the International Irrigation
Management Institute, Colombo, Sri Lanka. Water Resources Group, National
Remote Sensing Agency, Hyderabad, India.
Tolk, J.A., T.A. Howell, S.R. Evett, 1999. Effect of mulch, irrigation and soil type on water
use and yield of maize. Soil Till. Res. 50, 137-147.
– 98 –
Tuong, T.P., B.A.M. Bouman, 2003. Rice production in water scarce environments. In: Kijne,
J.W., R. Barker, D. Molden (Eds.), Water productivity in agriculture: limits and
opportunities for improvement. CAB International, Wallingford, UK.
Tuong, T.P., 1999. Productive water use in rice production: opportunities and limitations. J.
Crop Prod. 2, 241-264.
United Nations Population Division, 2008. World population prospects: the 2008 revision Population data base. (accessed October, 2009).
Van Dam, J.C., R. Singh, J.J.E. Bessembinder, P.A. Leffelaar, W.G.M. Bastiaanssen, R.K.
Jhorar, J.G. Kroes, P. Droogers, 2006. Assessing options to increase water use
productivity in irrigated river basins using remote sensing and modeling tools. Int. J.
Water Resour. Dev., 22, 115-133.
Van Dam, J.C., R.S. Malik (Eds), 2003. Water productivity of irrigated crops in Sirsa district,
India. Integration of remote sensing, crop and soil models and geographical
information systems. WATPRO final report, including CD-ROM, pp. 173.
Van Herwaarden, A.F., J. Passioura, 2001. Improving estimates of water-use efficiencies in
wheat. Aust. Grain 11, 3-5.
Waheed, R.A., M.H. Naqvi, G.R. Tahir, S.H.M. Naqvi, 1999. Some studies on pre-planned
controlled soil moisture irrigation scheduling of field crops. In: Kirda, C., P.
Moutonnet, C. Hera, D.R. Nielsen (Eds.), Crop yield response to deficit irrigation,
Developments in Plant and Soil Sciences, vol. 84. Kluwer Academic Publishers,
Dordrecht, pp. 180-195.
Wang, H., L. Zhang, W.R. Dawes, C. Lu, 2001. Improving water use efficiency of irrigated
crops in the North China Plain - Measurements and modelling. Agric. Water
Manage., 48, 151-167.
Wilson, K.B., A.H. Goldstein, E. Falge, M. Aubinet, D. Baldocchi, P. Berbigier, C.
Bernhofer, R. Ceulemans, H. Dolman, C. Field, A. Grelle, B. Law, T. Meyers, J.
Moncrieff, R. Monson, W. Oechel, J. Tenhunen, R. Valentini, S. Verma, 2002.
Energy balance closure at FLUXNET sites. Agric. For. Meteor. 113, 223 - 243.
Xianqun, X., 1996. The combined field experiment for determining evapotranspiration in
north China Plain. In: Cramp, C.R., E.J. Sadler, R.E. Yoder (Eds.),
Evapotranspiration and irrigation scheduling: proceedings of the international
conference, November 3 - 6, 1996, San Antonio Convention Center, San Antonio,
Texas, pp. 69-74.
Xie Z-k, Y-j. Wang, F-m. Li, 2005. Effect of plastic mulching on soil water use and spring
wheat yield in arid region of northwest China. Agric. Water Manage., 75, 71-83.
Xinyuan, W., L. Mengyu, L. Xionan, M. Xuesen, 1993. Water use and water use efficiency of
winter wheat in an?? water-deficient and salt-affected area in Hebei Province. In:
Shalhevet, J., L. Changming, X. Yuexian, V.R. Priel (Eds.). Water use efficiency in
agriculture: proceedings of the binational China-Israel workshop, April 22-26,
Beijing, China.
Yazar, A., S.M. Sezen, B. Gencel, 2002a. Drip irrigation in the Southeast Anatolia Project
(GAP) area in Turkey. Irr. Drain. 51, 293-300.
Yazar, A., S.M. Sezen, S. Sesveren, 2002b. LEPA and trickle irrigation of cotton in the
Southeast Anatolia Project (GAP) area in Turkey. Agric. Water Manage. 54, 189-203.
Yazar, A., T.A. Howell, D.A. Dusek, K.S. Copeland, 1999. Evaluation of water stress index
for LEPA irrigated corn. Irr. Sci. 18, 171-180.
Zachariasse, L.C., 1974. Farmer and farm returns: an analysis of causes of differences in
profitability of similar arable farms in the North-East-Polder. Ph.D. thesis
Wageningen University, Wageningen, the Netherlands, p.p. 113.
– 99 –
Zhang, B., F-M Li, G. Huang, Z-Y Cheng, Y. Zhang, 2006. Yield Performance of spring
wheat improved by regulated deficit irrigation in an arid area. Agric. Water Manage.,
79, 28-42.
Zhang, H., T.Y. Oweis, S. Garabet, M. Pala, 1998. Water-use efficiency and transpiration
efficiency of wheat under rainfed conditions and supplemental irrigation in a
Mediterranean-type environment. Plant Soil, 20, 295-305.
Zhang, H., X. Wang, M. You, C. Liu, 1999. Water-yield relations and water-use efficiency of
winter wheat in the North China Plain. Irr. Sci. 19, 37-45.
Zhang, J., X. Sui, B. Li, B. Su, J. Li, D. Zhou, 1998. An improved water-use efficiency for
winter wheat grown under reduced irrigation. Field Crops Res., 59, 91-98.
Zhang, X., D. Pei, C. Hu, 2003. Conserving groundwater for the North China Plain. Irr. Sci.
21, 159-166.
Zima Szalokine, I., S. Szaloki, 2002. Relationships of water- and nutrient supply, yield and
evapotranspiration of maize. Idojaras 106, 197-213.
Zwart, S.J., W.G.M. Bastiaanssen, 2004. Review of measured crop water productivity values
for irrigated wheat, rice, cotton and maize. Agric. Water Manage. 69, 115-133.
Zwart, S.J., W.G.M. Bastiaanssen, 2007. SEBAL for detecting spatial variation of water
productivity and scope for improvement in eight irrigated wheat systems. Agric.
Water Manage. 89, 287-296.
Zwart, S.J., W.G.M. Bastiaanssen, C. de Fraiture, D.J. Molden, 2010a. A global benchmark
map of water productivity for rainfed and irrigated wheat. Agric. Water Manage. (in
Zwart, S.J., W.G.M. Bastiaanssen, C. de Fraiture, D.J. Molden, 2010b. WATPRO: a remote
sensing based model for mapping water productivity of wheat. Agric. Water Manage.
(in press).
Zwart, S.J., L.M.C. Leclert, 2010. A remote sensing based irrigation performance assessment:
a case study of the Office du Niger in Mali. Irr. Sci. (in press).
Zwart, S.J., H. Pelgrum, W.G.M. Bastiaanssen, 2010c. A global map of the growing season of
wheat using high resolution NDVI time-series of SPOT-Vegetation. (forthcoming).
– 100 –
Curriculum Vitae
Sander Zwart is a mixed pixel from the three northern provinces of the Netherlands.
He was born in Drachten, Friesland on March 30,1976 from a Frisian father and a
Stellingerwerver mother. He lived for 18 years in Roden, Drente and went to
Nienoordcollege in Leek, Groningen. His dream was to become a tropical doctor to
help people in need. But after being refused admission by lot to the study of medicine,
he decided to study Tropical Land Use at the Agricultural University in Wageningen.
He planned to go for one year to Wageningen, and then again join in the lottery for
admission to medicine. This never happened. Feeling at home in Wageningen and
enjoying the new life with new friends and an intriguing study programme, he decided
to finish Tropical Land Use. During his studies, Sander stayed six months in Lahore
Pakistan thanks to a traineeship with Euroconsult and he spent three months in
Palermo, Sicily, Italy, and in Simferopol, Crimea, Ukraine to conduct field work for
the three M.Sc. theses he wrote. Later, he decided to enter the M.Sc. programme Geoinformation Science. In January 2002, after seven and a half years of studying, Sander
graduated with two M.Sc. degrees; one in Irrigation and Water Engineering, and one
in Remote Sensing. That same year Sander joined WaterWatch in Wageningen, where
he could combine and develop his multi-disciplinary expertise in many projects
related to water management, agriculture and remote sensing. In 2003 Sander started
his Ph.D. research on water productivity with Wim Bastiaanssen as his supervisor.
This work was entirely conducted in his spare time. Sander wrote six peer-reviewed
articles, of which four were used for this thesis, and he contributed to four other
scientific articles. His first article, a review of crop water productivity measurements,
has been cited more than 80 times since the end of 2009 and it is quoted in the top5
most cited articles of the Elsevier journal Agricultural Water Management.
Peer-reviewed publications
Zwart, SJ, H Pelgrum, WGM Bastiaanssen, 2010. A global map of the growing season of
wheat using high resolution NDVI time-series of SPOT-Vegetation. (forthcoming).
Klok, L, SJ Zwart, H Verhagen, E Mauri, 2010. The surface heat island of Rotterdam
derived from satellite imagery. (forthcoming).
Immerzeel, WW, P Droogers, WGM Bastiaanssen, SJ Zwart, 2010. Calibrating hydrological
models using remotely sensed evapotranspiration: the importance of scale. Advances
in Water Resources (submitted).
Zwart, SJ, WGM Bastiaanssen, C de Fraiture, DJ Molden, 2010. A global benchmark map of
water productivity for rainfed and irrigated wheat. Agricultural Water Management
(in press).
Zwart, SJ, WGM Bastiaanssen, C de Fraiture, DJ Molden, 2010. WATPRO: a remote
sensing based model for mapping water productivity for wheat. Agricultural Water
Management (in press).
– 101 –
Zwart, SJ, LMC Leclert, 2010. A remote sensing based irrigation performance assessment: a
case study of the Office du Niger in Mali. Irrigation Science (in press).
Wipfler, EL, K Metselaar, JC van Dam, RA Feddes, E van Meijgaard, B van Ulft, B van den
Hurk, SJ Zwart, WGM Bastiaanssen, 2009. Seasonal evaluation of the ECMWF land
surface scheme against remote sensing derived energy fluxes of the Transdanubian
region in Hungary. Hydrology and Earth System Sciences Discussions 6, pp. 62936334.
Immerzeel, WW, A Gaur, SJ Zwart, 2008. Integrating remote sensing and a process-based
hydrological model to evaluate water use and productivity in a south Indian
catchment. Agricultural Water Management 95(1), pp. 11-24.
Zwart, SJ, WGM Bastiaanssen, 2007. SEBAL for detecting spatial variation of water
productivity and scope for improvement in eight irrigated wheat systems. Agricultural
Water Management 89(3), pp. 287-296.
Zwart, SJ, WGM Bastiaanssen, 2004. Review of measured crop water productivity values for
irrigated wheat, rice, cotton and maize. Agricultural Water Management 69(2), pp.
– 102 –
– 103 –
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF