Global change and global scenarios of water use and availability:

An Application of WaterGAP1.0
Joseph Alcamo, Petra Dll, Frank Kaspar and Stefan Siebert
Center for Environmental Systems Research (CESR),
University of Kassel, Germany
June, 1997
Table of contents
SUMMARY
1. INTRODUCTION......................................................................................................................................................5
2. OVERVIEW OF WATERGAP MODEL...............................................................................................................5
2.1 WATER USE............................................................................................................................................................6
2.2 WATER AVAILABILITY............................................................................................................................................6
3. SCENARIOS...............................................................................................................................................................7
3.1 GENERAL FEATURES ..............................................................................................................................................7
3.2 BASIC DRIVING FORCES OF SCENARIOS ..................................................................................................................8
3.3 WATER USE ASSUMPTIONS.....................................................................................................................................9
3.3.1 Domestic water use .......................................................................................................................................9
3.3.2 Industrial water use.....................................................................................................................................12
3.3.3 Agricultural water use.................................................................................................................................14
3.4 WATER AVAILABILITY ASSUMPTIONS CLIMATE CHANGE SCENARIOS...............................................................16
4. RESULTS..................................................................................................................................................................17
4.1 PRESENTATION OF RESULTS.................................................................................................................................17
4.2 INDICATORS OF CRITICALITY................................................................................................................................17
4.3 THE CURRENT OCCURRENCE OF WATER SCARCITY..............................................................................................18
4.4 THE FUTURE OCCURRENCE OF WATER SCARCITY ................................................................................................18
4.5 THE EFFECT OF CLIMATE CHANGE ON THE WATER SITUATION.............................................................................20
4.6 WATER USES AND WATER SCARCITY....................................................................................................................20
5. DETAILED DESCRIPTION OF THE WATERGAP 1.0 MODEL..................................................................21
5.1 SPATIAL RESOLUTION...........................................................................................................................................21
5.2 TEMPORAL RESOLUTION......................................................................................................................................21
5.3 WATER AVAILABILITY - THE HYDROLOGICAL MODEL.........................................................................................23
5.3.1 Water balance at the grid cell scale............................................................................................................23
5.3.2 Precipitation................................................................................................................................................24
5.3.3 Temperature ................................................................................................................................................24
5.3.4 Evapotranspiration and sublimation..........................................................................................................24
5.3.5 Runoff ..........................................................................................................................................................28
5.3.6 Urban areas.................................................................................................................................................30
5.3.7 Routing between grid cells..........................................................................................................................31
5.3.8 Preliminary calibration to annual discharge of large rivers and regionalization ....................................31
5.3.9 Comparison of WaterGAP estimates with other estimates.........................................................................36
5.4 WATER USE MODEL..............................................................................................................................................36
5.4.1 Domestic water use .....................................................................................................................................37
5.4.2 Industrial water use.....................................................................................................................................38
5.4.3 Agricultural water use.................................................................................................................................39
6. FRESHWATER CRITICALITY RATIO............................................................................................................44
APPENDIX
MAPS OF GLOBAL WATER SITUATION....................................................................................................................... A
TABLES OF WATER SITUATION IN DIFFERENT COUNTRIES........................................................................................ B
3
Summary
Although the availability of water far exceeds its use on the global scale, water resources are not
evenly distributed in space or time. This report addresses the questions: What is the current
distribution of global water resources? How will growth in population and economy, and
changing climate affect the distribution of water in the future, and will these changes worsen or
improve existing water scarcities? In this report we present some preliminary answers to these
questions by using the techniques of global modeling and scenario analysis.
The WaterGAP global water model. The global scenario analysis is performed with a new
global water model, WaterGAP (Water Global Assessment and Prognosis) Version 1.0.
WaterGAP computes water use and water availability in each of 1,162 watersheds and 160
countries covering nearly the entire terrestrial surface of the world. The model takes into account
basic socio-economic factors that lead to domestic, industrial and agricultural water use, and
physical and climate factors that lead to river runoff and groundwater recharge. Water uses in
this version of the model are, specifically, water withdrawals. Water availability is the sum of
annual river runoff and groundwater recharge. Although the model has been calibrated and
tested against existing data, it nevertheless contains many limitations in its description of water
use and water availability. As such, it is especially suited for drawing insights about global-scale
trends, but unsuitable for detailed conclusions about individual watersheds.
Assumptions of the scenario analysis. The purpose of the scenario analysis is to estimate the
influence of growth in population, economy and changing climate on future water use and
availability. To do so, we investigate three scenarios of water use and two scenarios of water
availability. The water use scenarios include assumptions about the trend in driving forces of
water use in domestic, industrial, and agricultural sectors. They are called Scenario L(low), M
(medium) or H (high) according to their level of water use. Scenario M is the best guess water
use scenario. The water availability scenarios assume that surface temperature and precipitation
will change in the future because of climate change. Results of the latest general circulation
model of the Max-Planck-Institut are used as the best guess water availability scenario, while
computational results of the Geophysical Fluid Dynamics Laboratory model are applied for
comparison. The climate scenarios are based on the trend of greenhouse gas emissions given by
the IS92a scenario of the Intergovernmental Panel on Climate Change (IPCC). Other
assumptions about future developments in global population and the global economy come from
the same IS92a scenario. Although WaterGAP computes changes in water use and availability
for the period 1990 to 2100, in this paper we only report calculations for two time slices 2025
and 2075.
Indicators of criticality. Our scenario analysis focuses on the relative abundance or scarcity of
water in watersheds and countries. For this purpose we use two simple indicators, the criticality
ratio and the criticality index. The criticality ratio (CR) is the ratio of water use to water
availability in a watershed or country. The criticality index (CI) combines two factors the
criticality ratio and the water availability per capita into a single indicator of water
vulnerability in a watershed and country. The index ranges from 1 for water surplus to 4 for
water scarcity.
The current occurrence of water scarcity. If we add up computations from all watersheds, we
estimate that 30% of the world's population of 5.6 billion currently live in watersheds with water
surplus (criticality index CI=1), while 20% live in marginally vulnerable watersheds (CI = 2),
4
6% under water stress conditions (CI = 3) and 44% in watersheds that suffer from water scarcity
(CI = 4). These percentages refer to conditions in 10-percentile dry years (precipitation lower
than this occurs only once every ten years). In years with average precipitation, the percentage
living in watersheds with water scarcity drops to 29%.
The future occurrence of water scarcity. Under the best guess scenarios for water use (Scenario
M) and water availability, we estimate that the percentage of the worlds population facing water
scarcity (CI = 4) increases from 44% (2.5 billion people) in 1995 to 57% (4.7 billion people) in
2025 and to 69% (7.2 billion people) in 2075. Thus, the affected population affected by water
stress is predicted to double by 2025 and to triple by 2075. Under average precipitation
conditions, these percentages are 29% in 1995, 37% in 2025 and 58% in 2075.
For the year 2075, the percentage of the worlds population living in water scarce
watersheds ranges from 61% under Scenario L to 78% under Scenario H. These data refer to
dry-year conditions (the 10-percentile dry years). Water conservation measures as represented in
Scenarios M and L result in a drastically improved water situation for some countries. On the
other hand, most countries that have critical water scarcities (CI = 4) under Scenarios M and H
remain in this category under Scenario L. In these countries it is very possible that absolute
water scarcity will hinder further economic and population development. Alternatively, these
countries will need to substantially increase their water availability (through desalination or
inter-basin water transfers, as examples), or reduce their water demand even more radically than
assumed in Scenario L.
To sum up to this point, the medium water use scenario provides a best guess of future
trends, and the low and high scenarios form an uncertainty range around this guess. A robust
conclusion from these three scenarios is that the intensity of current water scarcity is not likely to
abate in the coming decades, and indeed may affect many more billions in the future.
The effect of climate change on the water situation. On the global average, overall annual runoff
increases and water scarcity is somewhat less severe under climate change. We can summarize
these results in terms of the number of people confronted with water scarcities in the future (CI =
4): In 2075, the percentage of world population living in water scarce watersheds is 69% under
Scenario M with climate change, and 74% under Scenario M without climate change. However,
some 25% of the earths land area experiences a decrease in runoff, and this occurs in some
countries that are already facing severe water scarcity. Moreover, this analysis omits many
aspects of possible climate change impacts on water resources. For example, it does not take into
account a possible change in interannual and inter-seasonal variability of precipitation and
temperature. Moreover, although increased precipitation is likely to increase water availability, it
may also increase the rate of flooding in a watershed.
Water uses and water scarcity. The areas of water scarcity identified in this report occur
throughout the world, and in both industrialized and developing countries. For current 1995
conditions, the domestic sector accounts for 7.3%, the industrial 18.0%, and the agricultural
74.7% of global water use (under the 10-percentile dry year conditions). In water scarce
industrialized countries, the largest user is typically industry, whereas in water scarce developing
countries it is almost always agriculture. For year 2075, according to Scenario M, these figures
greatly change to 11.2% for the domestic sector, 55.3% for the industrial sector, and 33.5% for
the agricultural sector. These results follow from assumptions about the trends in population,
economy, and water intensity, and point out the possibility that industry will supersede
agriculture as the worlds largest water user.
5
1. Introduction
Water plays an essential role in the existence of society. Our food supply depends on
precipitation collected in an irrigation system or falling directly on a field. In our households
water has a central place in day-to-day activities. In industry water is indispensable as a raw
material for industrial product, or tool cool combustion processes. Hence, the question, Is there
enough water? has always been important, and remains so. The problem is that water is not
evenly distributed both in time and space. Put another way, the availability of water is not ideally
matched to human water needs.
To assess this problem, this paper provides a preliminary estimate of the global
distribution of water use and availability, and how this might change over the next 80 years. We
examine water quantity, but not water quality. No single report can give a comprehensive picture
of the effect of all possible factors on water distribution. Therefore this paper concentrates on a
few major driving forces of global change, namely, the effects of population growth, economic
growth, climate change, and changes in the intensity of water use. Through the techniques of
global modeling and global scenario analysis, we present quantitative estimates of water use and
water availability for a specified set of scenarios. In addition to the estimates for present-day
conditions, we show water use and availability values for the years 2025 and 2075.
Here, water use is defined as the volume of water withdrawn from surface water or
groundwater within a certain year. Water availability refers to the renewable water resources, i.e.
the amount of precipitation that is not evaporated or transpirated by plants but becomes runoff.
This is the sum of surface runoff and groundwater recharge. Comparing water availability and
use on the watershed scale, we determine a freshwater criticality ratio as the ratio of use over
availability. In addition, we relate this ratio to the water availability per capita, and obtain a
freshwater criticality index that describes the degree of water scarcity in the watershed.
In this paper we first briefly describe the model, and then introduce the basic assumptions
of the scenarios. Following that, we present key results of our computations of present and future
water criticality. In the last section, we describe in detail the global model WaterGAP 1.0 used
for the calculations.
2. Overview of WaterGAP model
The global scenario analysis is performed with a new global water model, WaterGAP (Water
Global Assessment and Prognosis) Version 1.0, developed at the Center for Environmental
Systems Research of the University of Kassel, with assistance from the National Institute of
Public Health and the Environment of the Netherlands. The model computes water use and
water availability in each of 1,162 watersheds covering nearly the entire terrestrial surface of the
world. Fig. A0 in the appendix shows a map of the watersheds. Some aspects of the models
design and data come from an integrated model of global environmental change called IMAGE
2 (Alcamo et al., 1994a) as well as previous modeling of the global water situation carried out by
Klepper (1996). WaterGAP takes into account basic socio-economic factors that lead to
domestic, industrial and agricultural water use, and physical and climate factors that lead to river
runoff and groundwater recharge. The model is described in detail in Section 5 of this paper.
Here we present only a brief overview of its features.
The smallest spatial entity of the model is the grid cell of size 0.5 longitude by 0.5
latitude. The basic calculations are done on the cell level, but model output is aggregated to both
the watershed and the country scale. The appropriate scale for assessing freshwater criticality is
the watershed scale. Computations are done using different temporal resolutions, depending on
the computed variable. Computational results are shown for the years 1995, 2025 and 2075.
6
"1995" broadly stands for "present-day conditions", as some data from other years were also
used. In particular, the climatic conditions of "1995" are the long-term average conditions of the
period of 1931-1960.
Water availability and use is not only computed for average precipitation of the respective
years, but also for 10-percentile dry years (i. e. precipitation lower than this occurs statistically
only once every ten years). The availability in such a year is assumed to be the sum of the
surface water runoff produced in the dry year and the groundwater recharge produced in a
average precipitation year, as groundwater is stored between years. The water situation in those
periodically returning dry years gives a clearer indication of criticality than the situation in
average years.
2.1 Water use
Water use calculations for domestic and industrial sectors are based on existing data at the
country-scale that are downscaled to the cell scale, and a procedure for scaling these data to the
years between 1995 and 2100 according to assumptions about population and economic growth.
Water use calculations for agriculture are divided into: (i) livestock water use which is computed
from the grid-scale distribution of livestock and their water consumption, and (ii) irrigation
water use which is computed from the climate-related consumptive use of crops, and other
factors. Both livestock and irrigation water use are computed on the grid-scale.
Water uses in this version of the model are, specifically, water withdrawals. Part of the
withdrawn water becomes return flow and may be reused by other water users. However, for the
estimation of water criticality, it is more appropriate to take into account withdrawal water use
instead of consumptive use (difference between withdrawal use and return flow) because
1. instream uses are not considered here
2. the quality of return flow is unknown
3. the location of the water users within the watershed is not known
2.2 Water availability
Version 1.0 of WaterGAP uses a simple approach to compute water availability on a grid-scale
that is consistent with the data available on the global scale. WaterGAP 1.0 calculates the daily
soil water balance of each grid cell, taking into account physical characteristics of watersheds
such as soil, vegetation, slope and aquifer type. Calculations are detailed enough to be tested and
calibrated to observed runoff data.
On the watershed scale, water availability is defined as the sum of annual river runoff and
groundwater recharge of all the grid cells in the watershed. This is, of course, just a rough
approximation of available water supply, since it does not take into account the spatial
variability within the watershed nor the month-to-month or week-to-week hydrologic variations.
On the other hand, many rivers have reservoirs which store river water from month-to-month, so
computing the water availability over an entire year can be a reasonable first indicator of the
water situation in a watershed. On the country scale, water availability is computed in the same
way as on the watershed scale, and the computed value is therefore equal to the internal
renewable resources of the country. With this method, the influence on water availability of
watersheds which cross country boundaries is not correctly taken into account.
7
3. Scenarios
3.1 General features
In this paper we investigate three scenarios of water use and one scenario of water availability.
The water use scenarios include assumptions about the trend in driving forces of water use
in domestic, industrial, and agricultural sectors (Table 3.1):
Scenario M (medium) This is a best guess scenario in which intermediate estimates
are made for changes in water use.
Scenario L (low): This contains a set of assumptions that lead to a lower estimate of
water use than Scenario M.
Scenario H (high): This contains a set of assumptions that lead to a higher estimate of
water use than Scenario M.
The water availability scenario assumes that average surface temperature and precipitation
will change in the future because of climate change. For this paper we use a climate change
scenario based on the trend of greenhouse gas emissions given by the IS92a scenario of the
Intergovernmental Panel on Climate Change (IPCC). As input to our model, we use
temperature and precipitation values computed with the general circulation models of the
Max-Planck-Institut and the Geophysical Fluid Dynamics Laboratory.
Although WaterGAP computes changes in water use and availability for the period 1990 to
2100, in this paper we only report calculations for two time slices 2025 and 2075.
Table 3.1: Distinguishing assumptions of water use scenarios.
Scenario
*
Domestic Industry Agriculture
L Water intensity increases with
income up to $15,000/cap-yr.
Then rapidly declines by 50%
and then further to a stringent
water conservation target
value for domestic water use.
Water intensity is constant with
income up to either $5,000 or
$15,000/cap-yr. Then rapidly
declines by 50% and then
further to a stringent water
conservation target value for
industrial water use.
Irrigated area constant. Water
use efficiency improves.
M Water intensity increases with
income up to $15,000/cap-yr.
Then declines by 50% and
remains constant afterwards.
Water intensity is constant with
income up to either $5,000 or
$15,000/cap-yr. Then declines
by 50% and remains constant
afterwards.
New irrigation areas in most
developing countries. Water use
efficiency improves.
H Water intensity increases with
income up to $15,000/cap-yr.
Afterwards, remains constant
Water intensity increases with
income up to either $5,000 or
$15,000/cap-yr. Afterwards,
remains constant
New irrigation areas in most
developing countries. Water use
efficiency does not improve.
*
Scenario L = Low, Scenario M = Medium, Scenario H = High
8
3.2 Basic driving forces of scenarios
Estimating future changes in water use and availability requires a basic set of assumptions about
future developments in global population and the global economy. These assumptions are taken
from the IS92a scenario of the IPCC. This scenario is one of six scenarios developed by the
IPCC to provide a set of feasible global scenarios of greenhouse gas emissions for use in the
analysis of climate change (Leggett et al., 1992). In so doing, an international committee of
experts agreed upon a set of assumptions about future developments in the growth in population
and the economy. In this paper we use population and economic data from IS92a, one of the
intermediate scenarios, as the basis for our scenarios. This and other information from the IS92a
scenario is only available for 13 world regions. In this paper, we scale all country data within a
region by the regional trends in IS92a. Unfortunately, this smoothes out differences in trends
between countries.
The assumed trend in population is based on the medium scenario of the United Nations
(1991). Assumptions vary substantially from region to region, ranging from a leveling off trend
in the coming few decades (or sooner) in industrialized regions, to an increase by a factor of 3.4
in Africa between 1995 and 2075. According to this scenario, global population increases to
about 10.4 billion people by 2075 (Table 3.2).
Table 3.2: Regional population assumptions
Region Population in millions
1995 2025 2075
Canada 28 32 31
United States 259 301 296
Latin America 463 672 800
Africa 683 1412 2310
OECD-Europe 375 395 380
Eastern Europe 121 137 142
Former SU 291 328 341
Middle East 235 501 816
India+South Asia 1233 1878 2389
China+C.P.Countries 1404 1838 2020
East Asia 364 554 706
Oceania 22 23 23
Japan 124 132 127
World 5602 8205 10380
Assumptions about economic growth in the IS92a scenario are based partly on earlier IPCC
work and partly on short-term estimates of the World Bank (1992). The scenario assumes
continuing growth in all regions, ranging from a factor of 3.5 increase of per cap GDP in the
United States to a factor of 17 in China over the period from 1995 to 2075 (Table 3.3). Growth
of industry as expressed by industrial GDP ranges from a factor of 3.5 for the United States to a
factor of 20 for China. Despite the assumed growth, a large income gap remains between
industrialized and developing regions.
9
Table 3.3: Assumptions of per capita GDP and per capita industrial GDP
Region GDP per capita in 1990 US-$ IGDP per capita in 1990 US-$
1995 2025 2075 1995 2025 2075
Canada 23822 43658 90878 8297 15206 31652
United States 25421 45332 89800 7639 13559 26782
Latin America 2719 4958 15210 966 1536 4273
Africa 724 1259 4321 246 545 1736
OECD-Europe 21021 38528 80197 7405 13571 28226
Eastern Europe 2501 8934 16850 1078 3304 6108
Former SU 3408 12231 23068 1757 5821 10592
Middle East 5091 7860 21372 1485 2064 4989
India+South Asia 400 991 4384 111 422 1721
China+C.P.Countries 388 1396 6651 166 824 3335
East Asia 1500 3719 16451 597 1372 5401
Oceania 19021 41274 83872 5576 12009 24323
Japan 27901 60544 123028 11716 25382 51460
World 4431 7314 15685 1562 2661 5528
3.3 Water use assumptions
3.3.1 Domestic water use
Future domestic water use in a country is estimated by multiplying the number of people in a
country times its water intensity (water use per capita). Population assumptions are the same for
all scenarios, and come from the IS92a scenario described above. Three different scenarios are
assumed for changes in water intensity, and all assume that water intensity will change along
with changes in income. As an indicator of income we use the Gross Domestic Product per
person per year (GDP/cap-yr). Assumptions for changes in GDP/cap-yr come from the IS92a
scenario, and the same income assumptions are used for Scenarios L, M and H.
An examination of current domestic water use in different countries shows an upward
trend of per cap domestic water use (water intensity) as per cap GDP increases, at least in the
range of lower incomes (Fig. 3.1). However, time series data from a few industrialized countries
indicate a clear downward trend in the last few years in their total water use. Unfortunately, we
do not have independent time series of the sectoral distribution. Therefore, we assume that the
temporal changes were the same in each sector. Fig. 3.2 shows the development of per capita
domestic water use. The downward trend seems to have occurred only in countries where the
annual GDP per capita is above $10,000 to $20,000/cap-yr. Based on these considerations we
make the following assumptions about different scenarios, which describe how the country-
specific water use intensity changes with change in per capita GDP:
For Scenario M we assume that the country-specific water intensity of the base year
increases with income up to $15,000/cap-yr, and afterwards rapidly declines to 50% of its
value at $15,000/cap-year; afterwards water intensity remains constant (Fig. 3.3). The 50%
value is reached at a GDP of $18,750/cap-year ($15,000/cap-year + 25%). The rate of
increase is given by the best fit line in Fig. 3.1, which was drawn by excluding countries
with GDP above $15,000/cap-yr. The development of water use in countries, which were
already above a GDP above $15,000/cap-yr in 1990 is adjusted by taking into account the
year of maximum water use, while the 25% increase of GDP relates to the actual GDP of
10
this country in 1990. The GDP turning point of $15,000/cap-year, and the 50% decline
should be viewed as only very rough approximations.
For Scenario H, water intensity increases in the same way as Scenario M up to $15,000/cap-
year, but above this point it remains constant (Fig. 3.3).
For Scenario L, water intensity behaves like in scenario M until it has declined to 50%.
Then, it further declines to a stringent water conservation target of 15 m
3
/cap-year (Fig. 3.3).
This is 38% of the current lowest domestic water intensity of any industrialized country
(Switzerland and Great Britain).
Some wealthy countries have decreased water use already to such an extent that scenarios M and
H are rather similar.
y = 0.3x
0.6
1
10
100
1000
10 100 1000 10000 100000
per cap GDP in US-$
p
e
r

c
a
p

d
o
m
e
s
t
i
c

w
a
t
e
r

u
s
e

(
m
3
/
y
r
)
AFRICA ASIA EUROPE AMERICA
Fig. 3.1: Per cap domestic water use as a function of the per cap GDP. Shown are data from 112
countries (WRI, 1990, 1992, 1994, 1996). The value from the year closest to 1990 was selected
for the calculation of the best fit line. Depending on the country, the depicted water use data are
from the years 1980 to 1994. The country's per capita GDP is the GDP of the respective year in
1990-US$ (at constant prices) and was derived from data in UN (1995).
11
0
20
40
60
80
100
120
140
160
1970 1975 1980 1985 1990 1995
p
e
r

c
a
p

d
o
m
e
s
t
i
c

w
a
t
e
r

u
s
e

(
m
3
/
y
r
)
FINLAND ($23000-$27000) U.K. ($11000-$17000)
SWITZERLAND ($30000-$33400) JAPAN ($18600-$23734)
TURKEY ($2253-$2650) KOREA REP. ($3483-$6280)
BOTSWANA ($1911-$2900)
Fig. 3.2: Time series of per capita domestic water use in various countries. The GDP/cap-yr
range during the respective time period is given in parentheses.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5000 10000 15000 20000 25000 30000 35000
per cap GDP in 1990 US-$
s
c
a
l
i
n
g

f
a
c
t
o
r

f
o
r

d
o
m
e
s
t
i
c

w
a
t
e
r

u
s
e

p
e
r

c
a
p
u
t
scenario H
scenario M
scenario L
Fig. 3.3: Domestic water use scenarios.
12
3.3.2 Industrial water use
Future industrial water use in a country is estimated by multiplying the industrial GDP (IGDP) in
a country times its water intensity (water use per unit IGDP). The same basic approach is used
for industrial water use scenarios as was used for the domestic sector. Three different scenarios
are assumed for changes in water intensity, and all assume that water intensity will change along
with changes in industrial GDP per person per year (IGDP/cap-yr). The reasoning is that as a
country becomes more industrialized, the type of industry changes and the approach to using
water in industry also changes. The scenario for the change of industrial GDP is taken from the
Baseline A scenario of IMAGE 2.1 (Alcamo et al., 1996) for thirteen world regions and then
applied to each country within the region. In this scenario, the ratio of industrial to total GDP is
assumed to be constant for industrialized countries, but increases in developing countries.
An examination of current industrial water use in different countries shows a slight, but
unconvincing trend downward with increasing GDP/cap-yr (Fig. 3.4). On the other hand, we
noted above that several industrialized countries have had a clear downward trend in the last few
years in their total water use. Moreover, a statistical analysis of available country data provided
some evidence that countries with a lower water availability (arbitrarily defined as having less
than 1000 m
3
/cap-yr) began reducing their industrial water intensity at a GDP/cap-yr roughly
around $5000, whereas countries with higher water availability began reducing it at around
$15,000 (Kulshreshtha, 1997). Based on these considerations we make the following
assumptions for different scenarios, which describe how the country-specific water use intensity
as given by the data changes with change in per capita GDP:
For Scenario M, for countries with higher water availability (arbitrarily defined as equal or
above 1000 m
3
/cap-yr; values taken from WRI, 1996) we assume that water intensity
remains constant at its current level up to an income of $15,000/cap-yr, and afterwards
rapidly declines to 50% of its value at $15,000/cap-year (Fig. 3.5). Furthermore we assume
that countries with a lower water availability (less than 1000 m
3
/cap-yr) have a stronger
motivation for conserving water and therefore begin to reduce water intensity at $5,000/cap-
yr. After the $5,000/cap-yr point, water intensity in these countries declines as in other
countries by 50%.
For Scenario H, water intensity is simply assumed to remain constant at its current value
(Fig. 3.5).
For Scenario L our approach for the industrial sector is similar to that of the domestic sector.
Scenario L follows Scenario M up to 50% decline, then decreases more sharply down to a
stringent water conservation target value for industry (Fig.3.5). We assume, that these
minimum needs are 25% of the lowest current industrial water intensity in any industrialized
country. This amounts to 0.22 m
3
/ $100-yr, which is 25% of Denmarks present industrial
water intensity.
13
0.001
0.01
0.1
1
10
10 100 1000 10000 100000
per cap GDP in US-$
i
n
d
u
s
t
r
i
a
l

w
a
t
e
r

u
s
e

p
e
r

i
n
d
u
s
t
r
i
a
l

G
D
P

(
m
3
/
U
S
-
$
)
AFRICA ASIA EUROPE AMERICA
Fig. 3.4: Industrial water use per industrial GDP as a function of per capita GDP in 112
countries.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5000 10000 15000 20000 25000 30000 35000
per cap GDP in 1990 US-$
s
c
a
l
i
n
g

f
a
c
t
o
r

f
o
r

i
n
d
u
s
t
r
i
a
l

w
a
t
e
r

u
s
e

p
e
r

I
G
D
P
scenario H
scenario M for high
availability countries
scenario L for high
availability countries
scenario M for low
availability countries
scenario L for low
availability countries
Fig. 3.5: Industrial water use scenarios.
14
3.3.3 Agricultural water use
3.3.3.1 Livestock water use
As noted above, agricultural water use in WaterGAP 1.0 is divided into two parts: water
requirements for livestock and water requirements for irrigation. Data on the present density of
livestock is provided by NCAR on the GlobalARC GIS Database (1996). Future use of water by
livestock is estimated by multiplying the number of livestock animals times the water use per
head of livestock (see Section 5.7). To estimate future water use by livestock, we set water use
per head constant at its 1995 level (Table 5.1), and change the number of livestock in each
country. For the number of livestock we use the estimates for thirteen world regions from the
Baseline A scenario of the IMAGE 2.1 model (Alcamo et al., 1996). The Baseline A scenario
uses the same assumptions for population and economic growth (from the IS92a scenario) as
used in this paper, and is therefore somewhat consistent with other assumptions in this paper.
Regional figures from Baseline A are used to scale the 1995 grid-scale density of livestock
(described in Section 5.7). The same scenario for future number of livestock is used for
Scenarios L, M, and H.
3.3.3.2 Irrigation water use
Future use of water for irrigation is computed by multiplying the number of hectares used for
irrigation times the water used per hectare. Both of these variables must be specified for each
scenario. We begin with the future extent of irrigated area. Since a detailed calculation of this is
very complicated, we make the following simplifying assumptions:
For developing countries:
For Scenarios M and H: We assume that the amount of irrigated land increases up to
year 2010 according to an FAO study by Alexandratos (1995). (Alexandratos only
gives estimates up to 2010). We found that the cereal production estimates of
Alexandratos up to 2010 are rather similar to those estimated in the Baseline A
scenario of IMAGE 2.1 (described above). Therefore, for the period 2010 to 2075,
we scale the amount of irrigated area in each country by the increase in cereal
production given in the Baseline A scenario of IMAGE 2.1. We noted above that
Baseline A is consistent with the population and economic assumptions made
elsewhere in this paper. The resulting increase in irrigated land for years 2025 and
2075 are given in Table 3.4. These assumptions are used for Scenarios M and H. The
new irrigated land is allocated to a particular cell based on the algorithm of Klepper
(1996) as described in Section 5.7.2.1.
For Scenario L: we assume no increase in irrigated land.
For industrialized countries, the amount of agricultural land is already stable or declining.
Hence, for Scenarios M, H, and L we assume that the location and extent of irrigated land
remains constant at its current situation.
15
Table 3.4: Assumed change of irrigated area.
world region irrigated area 2025 / irrigated area
1995
irrigated area 2075 / irrigated area
1995
Canada 1 1
USA 1 1
Latin America 1.60 2.33
Africa* 2.14 3.98
OECD Europe 1 1
Eastern Europe 1 1
former USSR 1 1
Middle East 1.61 2.57
India and South Asia 1.61 2.44
China and centrally planned countries 1.15 1.57
East Asia 1.15 1.57
Oceania 1 1
Japan 1 1
* except Northern African states Egypt, Libya, Tunisia, Algeria, Morocco with 1.59 and 2.95.
The second step in calculating future irrigation water use is to compute the water used per
hectare. This is calculated in WaterGAP 1.0 as a function of climate, cropping intensity, and
water use efficiency (described in Section 5.7). Hence, these variables must be specified for the
scenarios.
For climate data we use results for monthly precipitation and surface temperature, as
computed with the climate models of MPI and GFDL, based on the IS92a emissions scenario
(see Section 3.4). The same climate scenario is used for Scenarios L, M, and H.
For cropping intensities in developing countries , we use estimates given in Alexandratos
(1995) for the year 2010. These estimates are used for Scenarios L, M, and H for the years 2025
and 2075 (Table 5.8). In the developed countries, cropping intensity (like irrigated area) is
assumed to remain constant.
Water use efficiency is the remaining variable that is needed to compute irrigation water
per hectare. For simplicity, in Scenarios L and M we assume that water use efficiency improves
at the same rate in all countries, namely by 0.5%/year. This gives a 16% improvement in water
use efficiency for the period between 1995 and 2025, and an 49% improvement from 1995 to
2075. This assumed rate of improvement is in line with estimated future rates of technological
improvement of crop yield (see, for example, Alexandratos, 1995, and Alcamo et al., 1996). For
Scenario H, no improvement in water use efficiency is assumed.
To sum up the irrigation water use scenario assumptions (see Table 3.1 for an overview):
For Scenario L, we assume that the amount of irrigated land remains constant at its 1995
level. In addition, we assume that water efficiency improves on all irrigated land.
For Scenario M, we assume that food production increases in developing countries such that
the amount of irrigated land also increases. Water use efficiency improves on all irrigated
land.
For Scenario H, we assume that the amount of irrigated land increases in developing
countries. Water efficiency is assumed not to improve.
The result of these assumptions is that Scenarios L, M, and H have an increasing level of
irrigation water use.
16
3.4 Water availability assumptions Climate change scenarios
The precipitation estimates of general circulation models (GCMs) are generally less reliable than
their temperature estimates. Therefore, in order to determine the sensitivity of water availability
and thus criticality to GCM predictions we use the predictions of two GCMs: The ECHAM4-
OPYC model of the Max-Planck-Institut (Rckner et al., 1996), hereafter referred to as the MPI
model) and the model of the Geophysical Fluid Dynamics Laboratory (Manabe et al., 1991;
Manabe et al., 1992), the GFDL model. The MPI model results are from a transient run based on
the IS92a emissions scenario and include the effect of sulfate aerosols; they are more recent than
the GDFL model results and therefore probably represent more up-to-date knowledge about
climate processes.
In the case of the MPI model, we use the following monthly total precipitation and surface
temperature values:
1. the mean of the values of 1950 to 1980 (the climatic normal)
2. the mean of the values of 2020 to 2030 (for the year 2025)
3. the mean of the values of 2070 to 2080 (for the year 2075), when CO
2
-equivalent
concentrations will have doubled with respect to mean concentrations between 1980 and
1990.
Precipitation in 2025 and 2075 in each 0.5 by 0.5 cell is computed by scaling the
precipitation values from the Cramer and Leemans database (P(CL), comp. section 5.3.2) with
the total precipitation values of the GCMs (interpolated from the GCM resolution down to
WaterGAP resolution):
P or P CL
P mean of or
P mean of
WaterGAP WaterGAP
GCM
GCM
( ) ( )
( )
( )
2025 2075
2020 2030 2070 2080
1950 1980

(1)
Temperature is computed using the GCM surface temperature values as follows:
[ ]
T or T CL T
T T mean of or T mean of
WaterGAP WaterGAP GCM
GCM GCM GCM
( ) ( )
( ) ( )
2025 2075
2020 2030 2070 2080 1950 1980
+

(2)
In the case of the GFDL model, we do not use the GCM results directly to scale our
current precipitation and temperature values. Instead, the GFDL results are only used to
downscale climate modeling results of Baseline A of IMAGE 2.1, which has similar
assumptions to the IS92a scenario, and gives similar emissions (See Section 3.1.3). IMAGE 2.1
contains a two-dimensional climate model which computes temperature and precipitation in 18
latitudinal bands of 10 each, taking into account the different emissions within each zonal band.
The cell values of precipitation and temperature at a given time are then obtained within IMAGE
2.1 by scaling the cell values of current precipitation and temperature with both the zonal results
of IMAGE 2.1 and the gridded GCM results. For WaterGAP 1.0 simulations of the years 2025
and 2075, we apply the precipitation and temperature values computed by IMAGE 2.1 for 2025
and 2075.
17
4. Results
4.1 Presentation of results
Scenario results are presented for the years 1995, 2025 and 2075 in two forms:
As maps showing water availability, water use and the freshwater criticality index on the
watershed scale (Appendix A).
As tables showing water availability, water use, the freshwater criticality ratio and the
freshwater criticality index of each country
The tables present information for the various scenarios presented in Section 3. The maps (with
the exception of A0 and A1) show results only for the best guess scenario of future trends, i.e.
changes in water use according to Scenario M, and changes in water availability according to the
climate scenario computed by the MPI climate model. In addition, all map information pertains
to the 10-percentile dry year (precipitation lower than this occurs only once every ten years). One
should also keep in mind that we consider the watershed scale to be the appropriate scale for
assessing water criticality, as water is transferred within a watershed but generally not between
watersheds. The country values given in the tables in Appendix B, however, do not consider the
existence of watersheds, but are a mere summation of all cells within a country.
4.2 Indicators of criticality
Our scenario analysis focuses on the relative abundance or scarcity of water in watersheds and
countries. For this purpose we use two simple indicators, the criticality ratio and the
criticality index.
The criticality ratio (CR) is the ratio of water use to water availability in a watershed or
country. In both cases it is computed as the sum of grid cell values. Recall that water
availability refers to the renewable water resources (the runoff) generated inside the entity of
interest. Values for this ratio range from near zero in sparsely-inhabited watersheds where
water use is small compared to water availability, to greater than one in arid watersheds
where water use is computed to exceed its availability. In the latter case, inhabitants of these
watersheds obtain some of their water supply from deep groundwater, from recycled sources,
from seawater desalination, from upstream sources, or from elsewhere.
The criticality index (CI) is based on the work of Kulshreshtha (1993) and combines two
factors the criticality ratio and the water availability per capita into a single indicator of
water vulnerability in a watershed and country. The reasoning is that vulnerability increases
as two conditions become more critical: (i) total water resources are used up (critical ratio
becomes larger), and (ii) the pressure on existing resources increases (water availability per
capita declines). The index ranges from 1 for water surplus to 4 for water scarcity, as defined
in Table 4.1. To illustrate the use of this index, if one-half of a watersheds annual water
availability is used (criticality ratio = 0.5), and if pressure on water resources is relatively
low (water availability per capita is 12,000 m
3
/cap yr), then CI = 1, i.e. there is a water
surplus. On the other hand, if the criticality ratio is the same, but the pressure on water
resources is much greater (for example, if water availability per capita drops to 1,500 m
3
/cap
yr), then the CR = 3, a much more critical state.
Further details about the criticality indicators are given in section 6.
18
Table 4.1: Definition of fresh water criticality index (CI)
criticality ratio (use/availability)
per cap water
availability
[m
3
/(cap yr)]
< 0.4 0.4 - 0.6 0.6 - 0.8 > 0.8
< 2,000 2 3 4 4
2,000-10,000 1 2 3 4
> 10,000 1 1 2 4
Source: Kulshreshtha, S.N. (1993): World Water Resources and Regional Vulnerability: Impact of Future Changes.
RR-93-10, IIASA , Laxenburg, Austria.
1: water surplus
2: marginally vulnerable
3: water stress
4: water scarcity
4.3 The current occurrence of water scarcity
Critical water areas occur on each continent (Figure A12 and Tables B1 and B2). The criticality
index is high in arid regions, as expected, but also in densely populated areas of India, China,
Europe and the United States, where high water use occurs (Figure A7). If we add up
computations from all watersheds, we estimate that 30% of the world's population of 5.6 billion
currently live in watersheds with water surplus (criticality index CI=1), while 20% live in
marginally vulnerable watersheds (CI = 2), 6% under water stress conditions (CI = 3) and 44%
in watersheds that suffer from water scarcity (CI = 4). These percentages refer to conditions in
10-percentile dry years (as explained previously). In years with average precipitation, the
percentage living in watersheds with water scarcity drops to 29%.
4.4 The future occurrence of water scarcity
The best guess scenario
While the WaterGAP model Version 1.0 cannot yet give very reliable results for individual
watersheds, it can present an overview of possible trends on the global-scale. We compute a
best guess" of future trends by assuming that factors related to water use will change according
to Scenario M, and that factors related to water availability will change according to the climate
scenario computed with the MPI climate model. Under these scenario assumptions, and under
10-percentile dry year conditions (as explained above), we estimate that the criticality index will
tend to increase in more areas of the world than decrease (Figures A13 to A16, and Tables B1
and B2). The percentage of watershed area with critical water scarcity (CI=4) will increase from
31.7% in 1995 to 33.2% in 2025 and 37.3% in 2075. This is despite the increase in runoff in
most areas (Figures A5 and A6). Apparently, the growth in water use due to the scenarios of
growth in population and the economy is the decisive factor. This can be seen in the maps of
annual water use (Figures A7 to A11). The greatest increases in water uses are computed for
China, India, Taiwan and the Philippines, where water use is in some areas more than 1000
mm/a higher than in the year 1995. High water use growth rates occur whenever industry is an
important water user and industrial GDP grows fast, while the effect of population increase on
water use growth (via domestic and irrigation water use) is relatively small. This is illustrated by
the water use development in Africa (Figs. A10 and A11). Here, population strongly increases
19
but not GDP or industrial GDP; hence, the increase in water use in most watersheds is less than
50 mm/a.
One way to summarize the situation is to compute the number of people living in these
areas. We estimate that the percentage of the worlds population facing water scarcity (CI = 4)
increases from 44% (2.5 billion people) in 1995 to 57% (4.7 billion people) in 2025 and to 69%
(7.2 billion people) in 2075. Thus, the affected population affected by water stress is predicted to
double by 2025 and to triple by 2075. Note that the percentage of population living in water
scarce regions increases much faster than area of these regions. Under average precipitation
conditions, these percentages are 29% in 1995, 36.6% in 2025 and 57.8% in 2075.
A higher use estimate
The best guess scenario for water use (Scenario M) assumes that water intensity will
substantially decrease in the domestic and industrial sectors, and also that the efficiency of water
use in irrigation will significantly increase. By contrast, Scenario H is more pessimistic because
it assumes that these improvements do not occur. As a consequence, for almost all watersheds
and countries the criticality ratio (use/availability) is higher than under scenario M (Table B3).
The differences are largest in countries with a GDP of more than $15,000/cap-yr, as only for
these countries, domestic and industrial water use differ for the two scenarios M and H. Model
calculations show, however, that water saving under scenario M does not make a big difference
on the estimated number of people facing water scarcity in 2025 (Table B3). However, over the
longer run in 2075, the two scenarios are quite different. Under Scenario H 78% of the worlds
population is faced with water scarcity under dry year conditions (the 10-percentile dry years),
as compared to 69% in Scenario M.
A lower use estimate
As compared to the best guess scenario, Scenario L is more optimistic in its assumptions about
future water use. For domestic and industrial water use, it assumes a decline to a stringent water
conservation target. For agricultural water use, Scenario L assumes that the area of irrigated
land remains constant, and that the use of water on this land becomes more efficient.
As can be expected, under scenario L, the criticality ratio of almost all watersheds and
countries is lower than under scenario M (Table B3). With respect to the criticality index,
stringent water saving shows very positive results for some countries, e.g. Austria, Estonia and
Ethiopia (Table B3). On the other hand, most countries that have critical water scarcities (CI = 4)
under Scenarios M and H, remain in this category under Scenario L (Table B3). Specifically, the
percentage of world population facing water scarcity (CI = 4) in 2025 is 53% under Scenario L,
as compared to 57% under Scenario M. In year 2075, the percentage is 61% as compared to
69%. These numbers refer to dry-year conditions (the 10-percentile dry years). In general, while
the water conservation of scenario L leads to some improvement in all countries, it does neither
radically change the overall global situation nor is it stringent enough to strongly improve the
water situation in countries that already today experience water scarcity.
To sum up to this point, the medium water use scenario provides a best guess of future trends,
and the low and high scenarios form an uncertainty range around this guess. A robust
conclusion from these three scenarios is that the intensity of current water scarcity is not likely to
abate in the coming decades, and indeed may affect many more billions in the future.
20
4.5 The effect of climate change on the water situation
On the global average, the climate change scenario assumed in this analysis leads to increases in
both temperature and precipitation. These changes have an opposite effect on water vulnerability
because increased temperature leads to greater evapotranspiration, higher agricultural water use,
and lower runoff, while increased precipitation leads to lower agricultural water use and higher
runoff. From the global perspective, we compute that the annual precipitation effect is more
important than the temperature effect, and that overall annual runoff increases and water scarcity
is somewhat less severe under climate change. Mean global runoff in a 10-percentile dry year
increases from 178 mm/yr in 1995 to 191 mm/yr in 2025 and 211 mm/yr in 2075. We can
summarize these results in terms of the number of people confronted with water scarcities in the
future (CI = 4): In 2075, the percentage of world population living in water scarce watersheds is
69% under Scenario M with climate change, and 74% under Scenario M without climate change
A somewhat different picture emerges when the spatial variability of these changes are
examined (Figure A2 to A6, and Table B5). Depending on the watershed, annual runoff in 2025
is up to 489 mm/yr higher than in 1995, but is also as much as 252 mm/yr lower than in 1995.
Increases of more than 50 mm/yr occur in Iceland, on the Norwegian coast, in the Ganges basin
in North India, in Bangladesh, Burma, Southern Thailand, Malaysia and most of Indonesia, in
Central Africa and parts of Mozambique and Madagascar, in Southeastern Alaska, in the
Northwest of South America (including Panama) and in Uruguay and Southern Brazil. Summed
up, annual runoff increases in 2025 and 2075 in approximately 75% of the land area with respect
to 1995.
The other 25% of the land experiences less runoff, with decreases of more than 50 mm/yr
computed to occur in Northern Brazil, Chile, Taiwan and at the Indian west coast. Some
countries that already experience severe water scarcity (CI=4) have even less runoff under
climate change, such as Cyprus, Israel, Jordan, and Morocco (Table B5). Here, climate change
will further intensify the competition for water.
It is also very important to note that this analysis omits many aspects of possible climate
change impacts on water resources. For example, it does not take into account a possible change
in interannual and inter-seasonal variability of precipitation and temperature. Moreover,
although increased precipitation is likely to increase water availability, it may also increase the
rate of flooding in a watershed.
An important source of uncertainty in estimating the impact of climate change is the
uncertainty of grid-scale precipitation calculations of GCMs. To investigate this uncertainty, we
compared WaterGAP calculations using the Max Planck climate model and the GFDL model
(see Section 3.4, and Table B5). Indeed, the two models compute considerably different water
availabilities for many countries (Table B5). On the other hand, the two models lead to very
similar estimates of the number of countries experiencing severe water scarcity (CI=4).
4.6 Water uses and water scarcity
The areas of water scarcity identified in this report occur throughout the world, and in both
industrialized and developing countries. Given this situation, which sectors of water use
contribute most to water scarcity? For current 1995 conditions, the domestic sector accounts
for 7.3%, the industrial 18.0%, and the agricultural 74.7% of global water use (under the 10-
percentile dry year conditions). However, these global averages mask differences in water use
between industrialized and developing countries: in water scarce industrialized countries, the
largest user is typically industry, whereas in water scarce developing countries it is almost
always agriculture (Table B4).
21
For year 2075, according to Scenario M, these figures will greatly change to 11.2% for
the domestic sector, 55.3% for the industrial sector, and 33.5% for the agricultural sector. These
results follow from assumptions about the trends in population, economy, and water intensity,
and point out the possibility that industry will supersede agriculture as the worlds largest water
user. This is because of the assumed industrial growth in developing countries. In industrialized
countries, the industrial sectors continue to have the highest use, and in developing countries the
industrial and domestic sectors sometimes exceed agricultural water use (Table B4).
From this global perspective, special attention should be given to the agricultural sector
in the coming decades, but more effort in the future should be devoted to other sectors.
However, we must stress that this order of priorities for water conservation could very well
change once cultural and economic factors are included in this picture.
5. Detailed description of the WaterGAP 1.0 model
WaterGAP 1.0 consists of
a module to compute water availability, the hydrological model that computes runoff (comp.
section 5.3), and
a module to compute withdrawal water use, the water use model (comp. section 5.4).
Fig. 5.1 provides an overview of the model system and its inputs and outputs.
5.1 Spatial resolution
The land mass of the earth is represented by a 0.5 by 0.5 grid, which results in 59831 cells.
Each cell is part of a watershed, of a country and of a world region. For most freshwater related
problems, the watershed is the appropriate entity. However, some data are only available on the
country or world region scale. The cell-based calculations allows any necessary or desirable
translation between spatial entities like countries or watersheds. 1162 watersheds of widely
varying size are distinguished. The watershed boundaries were obtained from Klepper (1996)
who corrected those identified by Lozar (1992, in GlobalARC GIS Database, 1996), in particular
the European ones. In addition, Klepper subdivided larger basins into smaller ones. Fig. A0 in
the appendix shows a map of the watersheds.
One problem with the current land mask is that large freshwater lakes with a size of grid
cell or larger are not included in the grid and thus in the calculations. For watersheds which
include such lakes, water availability might be overestimated as evaporation from large water
bodies tends to be larger than precipitation onto them.
5.2 Temporal resolution
Simulations are performed for 1995, 2025 and 2075. Here, the year 1995 stands for "present-day
conditions", not exactly for the actual year 1995, as data from a variety of years were used. For
each year of interest, the hydrological model is run with daily time steps; however, climatic input
(temperature, precipitation and cloudiness from the CLIMATE data base of Cramer and
Leemans (van Woerden, 1995) is only available as monthly mean values which are interpolated
to daily values. The climatic input data represent long-term climatic averages. For model output
representative of present-day conditions (year 1995), the model is run with these input data over
two years when an equilibrium is reached. For the years 2025 and 2075, the CLIMATE data
base value are modified according to predictions of General Circulation Models (GCM).
Consumptive use of irrigation is equally computed on a daily basis, while all the other water use
components are computed just once per year of interest.
22
effective
precipitation
snow
max. available
soil water content
in the effective
root zone
potential
evapotrans-
piration
actual
evapotrans-
piration
soil water
balance
domestic water
use (DWU)
industrial water
use (IWU)
livestock water
use
consumptive
irrigation
water use
natural growing
season
urbanization
GDP
population
land cover
distance
to river
temperature
precipitation
cloudiness
land cover
soil
access to safe
drinking water
industrial
GDP
irrigated area
food demand
runoff
(groundwater
recharge +
surface water
runoff)
hydro-
geology
net
radiation
water
availability
water use
slope
livestock
density
current national
agricultural
water use
current national
IWU
current
national DWU

Fig. 5.1: Overview of inputs and processes in WaterGAP 1.0.
23
5.3 Water availability - The hydrological model
5.3.1 Water balance at the grid cell scale

S
t
P E R
eff a
(3)
S = actual water content of the soil within the effective root zone [mm]
P
eff
= effective precipitation [mm/d]
E
a
= actual evapotranspiration [mm/d]
R = runoff [mm/d]
t = time [d]
Total runoff R is partitioned into surface and fast subsurface runoff R
s
and slow groundwater
runoff (or base flow) R
g
.
R R R
s g
+ (4)
The effective precipitation P
eff
is the sum of rain and meltwater. From the available total
precipitation P and temperature T, P
eff
is computed applying the following snow model:
P
S
t
P E
eff
s
p

if T T
t
P P
S
t
S
t
K T T
eff
s
s
s t

( )
if T > T
t
and S
s
> 0
P P
eff
if T > T
t
and S
s
= 0 (5)
T

= temperature [C]
S
s
= snow-water equivalent [mm]
T
t
= threshold temperature [C] (0C)
E
s
= sublimation [mm/d]
K
s
= degree-day factor [mm/(d C)]
Based on a large number of model applications in Sweden by Bergstrm (1990), a threshold
temperature of 0C is chosen, while 2 mm/(d C) is used for the degree-day factor in the case of
forests and 4 mm/(d C) in the case of other landcover (comp. Table 2.1)
24
5.3.2 Precipitation
Cramer and Leemans derived a data set of long-term mean monthly terrestrial precipitation
values on a 0.5 by 0.5 grid using measured monthly precipitation from 1931 to 1960 at about
16000-20000 stations (van Woerden, 1995), using at least data from five years within this
period. They applied the Hutchinson interpolation algorithm (Hutchinson, 1995), which takes
into account topography. From the monthly values, we calculated daily ones by linear
interpolation.
An important consideration in water resources assessment is variability. Water availability
should meet water demand not only in average years, but also in dry years. In general, variability
of runoff is higher than variability in precipitation. Dry years are particularly critical in areas
with irrigated agriculture where irrigation water demand is higher in these years. For the
computation of a water criticality index, one should therefore not only look at average monthly
conditions but more critical ones with lower than average precipitation values. However, to our
knowledge, there do not exist long-term estimates of interannual variability of monthly
precipitation. There is, however, a rather old map of long-term annual precipitation variability.
This global map of Biehl (1943) (given in Riehl, 1979), which shows the coefficient of variance
C
v
of annual precipitation in seven classes ("below 10%" to "over 40%"), was digitized by
Klepper (1996) and was kindly provided to us. Assuming that annual precipitation is normally
distributed, we computed the precipitation of a 10-percentile dry year P
10d
(i.e. only 1 in 10 years
has less precipitation) as follows from the mean precipitation values P
average
of Cramer and
Leemans:
( ) P P C
d average v 10
1 128 . (6)
5.3.3 Temperature
Similar to precipitation, Cramer and Leemans determined a dataset of long-term mean monthly
terrestrial air temperature (van Woerden, 1995). Daily values are obtained by using cubic
splines. Temperature variability is not taken into account.
5.3.4 Evapotranspiration and sublimation
Evapotranspiration is the combined water loss from the soil due to evaporation directly from the
soil and transpiration of soil water via plants (root water uptake). Sublimation refers to the phase
change of water from snow to atmospheric vapor.
Of the many approaches to compute evapotranspiration, the Penman-Monteith equation
(Shuttleworth, 1993) appears to be suited best as it includes the main driving forces of
evapotranspiration: net radiation, temperature, vapor saturation deficit of the air and wind
velocity. The diffusional resistances of the plant surface and the atmospheric boundary layer are
parameterized by a stomatal and an aerodynamic resistance. Following Jarvis (1976), stomatal
resistance is a function of leaf area index, net radiation, temperature, saturation deficit of
ambient air, leaf water potential (taken to be expressed as soil moisture potential or soil water
content) and external concentration of CO
2
. Aerodynamic resistance is computed as a function
of vegetation height and wind velocity. According to Penman-Monteith, actual
evapotranspiration [mm/d] is
25
E
L
s R G c D r
s r r
a
w
n a p a
s a

+
+ +

1
]
1
1000
1

( ) /
( / )
(7)
L = latent heat of evaporation [J/kg]
= (2.50110
6
- 2361* T) J/kg

w
= density of liquid water [kg/m
3
]
s = increase of saturated vapor pressure with temperature [Pa/K]
= 2.50310
6
exp[17.269 T
c
/(237.3 + T
c
)]/(237.3 + T
c
)
2
Pa/K
= psychrometric constant [Pa/K]
R
n
= instantaneous net radiation [J/(d m
2
)]
G = heat flux from/to soil [J/(d m
2
)]

a
= density of air [J/(kg K)]
c
p
= specific heat of moist air [kg/m
3
]
D = vapor pressure (or saturation) deficit (saturated vapor pressure minus actual vapor
pressure) [Pa]
r
s
= surface (or stomatal) resistance [d/m]
r
a
= aerodynamic resistance [d/m]
With the Penman-Monteith equation, the negative feedback processes between plant
behavior and atmospheric variables (e.g. vapor saturation) as described by McNaughton and
Jarvis (1991) can be taken into account in a coupled model of the atmosphere, vegetation and
soil. However, if evapotranspiration is computed with fixed atmospheric boundary conditions, as
is done in WaterGAP, the feedbacks are neglected, and thus the influence of vegetation on
evapotranspiration will be overestimated. The main reason for not using the Penman-Monteith
equation in WaterGAP 1.0 is that there are not enough measurement values of vapor saturation
and wind velocity available to produce datasets of climatic averages of the variables at the
required spatial resolution. The Priestley-Taylor equation is instead chosen here to compute
potential evapotranspiration E
p
, which is defined as the evapotranspiration that occurs if
evapotranspiration is not restricted by moisture availability. The original Priestley-Taylor
equation has been shown to successfully predict evapotranspiration from well-watered
lysimeters in humid regions, and has been modified for use in arid regions (Jensen et al., 1990).
Actual evapotranspiration under restricted water availability is then computed from potential
evapotranspiration as a function of soil water content.
Evapotranspiration from the soil is assumed to occur only at temperatures above 0C,
while sublimation from snow only happens at 0C and below.
5.3.4.1 Potential evapotranspiration
Priestley and Taylor (1972) evaluated measured evaporation over well-watered vegetation and
over open water bodies when advection of air was negligible. They reason that at a larger scale
the effect of advection over land will be generally minor, and that thus the evaluation of the plot
scale events without advection is applicable to the grid-scale under most circumstances. Their
measurements were all done in humid regions. Following Priestley and Taylor, potential
evapotranspiration E
p
is computed as
26
E
L
s R G
s
p
w
n

1
]
1

1000 ( )
(8)
where = 1.26. Obviously, Eq. 8 can be derived from Eq. 7 if stomatal resistance r
s
goes to
infinity and the vapor saturation deficit related term is 0.26 times as large as the radiation related
term. Jensen et al. (1990) tested a large variety of evaporation formula against evaporation from
well-watered lysimeters. For the five lysimeters in humid regions, Priestley-Taylor equation gave
very good agreement, while for the six lysimeters in semi-arid and arid areas, a mean of 1.74
gave a better fit. This was explained by heat advection to the well-watered (irrigated) lysimeters.
Thus, it is not clear, whether a value of 1.74 is adequate at the grid scale. Shuttleworth (1993)
recommends to set = 1.26 for humid areas with relative humidity of 60% or more (error +-
15%, Shuttleworth, 1993, p. 4.17 and 4.21), and = 1.74 for other areas. Shuttleworth (1993)
states that now more substantial evidence supports this empirical relationship on a regional
average for regions with uniform vegetation cover or with land cover which is heterogeneous at
the scale of a few kilometers, if E
p
is considered to be reference crop evaporation E
rc
. Then, the
crop/forest specific evapotranspiration without soil moisture stress could be computed using Eq.
8 and potential crop coefficients K
co
. In WaterGAP 1.0, potential evapotranspiration is assumed
to be independent of vegetation. However, arid and humid areas of the globe are distinguished
based on the potential vegetation as computed by IMAGE 2.1 (Table 5.1), such that their extent
and that of the areas with relative humidities above and below 60% correspond approximately.
For the temporal scale of interest in WaterGAP 1.0, soil heat flux G can be neglected
(Shuttleworth, 1993, p. 4.10). The net radiation R
n
is the sum of the net short-wave radiation
(difference between incoming and reflected short-wave solar radiation) and the net long-wave
radiation (difference between the incoming and the outgoing long-wave radiation). Daily net
radiation R
n
is computed as a function of the day of the year, latitude, cloudiness and short-wave
albedo following Shuttleworth (1993), except for the computation of the sunset hour angle
which is better approximated by the CBM model of Forsythe et al. (1995). Cloudiness is
provided by the CLIMATE database (Cramer and Leemans, cited in van Woerden, 1995).
Albedo is taken to be a function of landcover as assigned by IMAGE 2.1 (Table 5.1). Whenever
a cell is snow-covered, an albedo of snow of 0.40 is used; this value is set to a rather low value
for snow because in reality, a cell might only be partly snow-covered.
27
Table 5.1: Potential evapotranspiration coefficient and rooting depth.
Potential vegetation (landcover) class
of IMAGE 2.1
in Priestley-Taylor
equation as a function
of potential vegetation
short-wave albedo
(values similar to
those used in
IMAGE 2.1)
rooting depth [m]
as a function of landcover
tundra 1.26 0.20 0.5
wooded tundra 1.26 0.15 1.0
boreal forest 1.26 0.13 2.0
cool coniferous forest 1.26 0.11 2.0
temperate mixed forest 1.26 0.11 2.0
temperate deciduous forest 1.26 0.13 2.0
warm mixed forest 1.26 0.11 2.0
steppe 1.74 0.25 1.0
hot desert 1.74 0.35 0.1
shrubland 1.74 0.22 1.5
savanna 1.26 0.16 1.5
tropical woodland 1.26 0.11 2.0
tropical forest 1.26 0.07 2.0
(agricultural land) n.a. 0.23 1.0
(regrowth forest) n.a. 0.15 2.0
(urban areas, major cities) n.a. 0.15 1.0
* n.a.: not applicable
5.3.4.2 Actual evapotranspiration
Actual evapotranspiration E
a
is computed as a function of potential evapotranspiration E
p
, the
actual water content of the soil in the effective root zone S and the total available soil water
capacity in the effective root zone S
max
. The smaller E
p
, the smaller can the ratio of water content
in the soil and total available water capacity become before E
a
drops below E
p
.
E E E
S
S
a p p

_
,
min ,
max
max
(9)
E
pmax
= maximum potential evapotranspiration [mm/d] (9.5 mm/d)
S
max
= total available soil water capacity within the effective root zone [mm]
S
max
is computed as the product of the total available water capacity in 1 m soil TAWC from
Batjes (1996) and the vegetation-specific rooting depth. The arithmetic mean TAWC of each
cell is used; it ranges between 30 and 300 mm. Glaciated areas are not given any TAWC value
and are thus not included in the calculation of water availability. Rooting depth values are
assigned to landcover classes as used in IMAGE 2.1 (Table 5.1).
5.3.4.3 Sublimation
If temperature is at or below the threshold temperature, evapotranspiration is assumed to be zero,
while sublimation occurs if snow is present. Sublimation is computed like potential
evapotranspiration using the Priestley-Taylor approach (with = 1.26) but using the latent heat
of sublimation (2.8310
6
J/kg) instead of the latent heat of evaporation. That sublimation can be
an important part of the hydrological cycle is shown by Marks and Dozier (1992) who measured
500 mm sublimation out of a total of 2500 mm precipitation as snow in the Southern Sierra
Nevada, California.
28
5.3.5 Runoff
Total runoff R is computed as
R P
S
S
eff

_
,

max

(HBV approach; Bergstrm, 1994) (10)

= runoff factor that is varied in the course of calibration
Total runoff R is partitioned into surface and fast subsurface runoff R
s
and slow groundwater
runoff (or base flow) R
g
such that
R R R
f R f R
s g
g g
+
+ ( ) 1
(11)
f
g
= fraction of total runoff which recharges the groundwater [-]
5.3.5.1 Slow groundwater runoff
On the cell scale, groundwater runoff is conceptually equal to groundwater recharge.
Groundwater recharge is computed following Klepper (1996); all the input data files were
provided by him.
The fraction f
g
of total runoff which recharges the groundwater and contributes to
baseflow is assumed to be a function of soil texture, slope and aquifer type. Thus, for a given
cell, groundwater recharge is a constant fraction of runoff; however, total runoff is limited by a
maximum groundwater recharge value R
gmax
that depends on soil texture.
R R f R
with
f f f f
g g g
g t s a

min( , )
max
(12)
f
t
= texture-related factor [-]
f
s
= slope-related factor [-]
f
a
= aquifer-related factor [-]
Soil texture and slope are derived from the UNESCO Soil Map of the World (1974) at a
resolution of 1/12 by 1/12, in which both parameters are divided into three classes (Table 5.2).
The soil texture refers only to the uppermost 30 cm of the soil. Cell-specific texture and slope
values (on a 0.5 by 0.5 resolution) were computed as the arithmetic mean of the 1/12 values,
such that values between 10 and 30 result for both texture and slope. For each of the three
texture classes, typical values for R
gmax
and f
t
are assigned, while for each slope class, a typical
value of f
s
is given (Table 5.2). These values are expert guesses. Cell-specific values of R
gmax
, f
t
and f
s
were computed by interpolating the values according to the cell-specific texture and slope
values.
29
Table 5.2: FAO texture and slope classes with the values assigned to them in WaterGAP (from
Klepper, 1996).
FAO soil texture class texture value R
gmax
[mm/d]
f
t
coarse:
sands, loamy sands and sandy loams with less than 15%
clay and more than 70% sand
10 3 0.95
medium:
sandy loams, loams, sandy clay loams, silt loams, silt,
silty clay loams and clay loams with less than 35% clay
and less than 70 % sand; the sand fraction may be as high
as 85% if a minimum of 15% clay is present
20 1.5 0.75
fine:
clays, silty clays, sandy clays, clay loams and silty clay
loams with more than 35% clay
30 0.75 0.25
FAO slope class slope value f
s
level to gently undulating:
dominant slopes 0-8%
10 0.75
rolling to hilly:
dominant slopes 8-30%
20 0.20
steeply dissected to mountainous:
dominant slopes > 30%
30 0.05
Aquifer types were derived from a hydrogeological map of Europe (IAH, 1974-1990) and
of Africa (UN, 1988), and a geological map of the world based on Bederke and Wunderlich
(1968) and Dierke (1975). The hydrological maps were digitized at RIVM, the Netherlands,
while the last map was provided in electronic form by Wolters-Noordhoff B.V. Six aquifer types
are distinguished. Geologic maps show the age of the rocks and furthermore differentiate
intrusive and effusive igneous rocks, but do not directly show the rock type (e.g. consolidated
sedimentary rock, fissured metamorphic rock), which would be indicative of the hydrogeological
behavior. However, there exists some correlation between the units in geological maps and the
hydrogeologically relevant rock type, which was derived here from comparisons of the
geological map to the hydrogeological maps of Africa and Europe. There was a fairly good
correspondence between the hydrogeological map of Africa and the geological map of Africa,
while the correspondence of the hydrogeological and geological maps of Europe was less
accurate. This is probably the result of the fact that the African hydrogeological map is actually a
re-interpretation of (another) geological map, while the European map provides primary
material. Thus global aquifer types were based on the correspondence between the geological
units of the geological map of the world and the aquifer types of the African hydrogeological
map. The resulting global hydrogeological map is rather coarse (with a resolution of ca. 2 by
2).
Like the texture- and slope-related factors, aquifer-specific values of f
a
are "just" expert
guesses. Klepper and colleagues derived values of f
a
for Africa, while other researchers
(Meinardi et al.) from the RIVM derived such values for Europe (
Table 5.3). Differences between the two parameter sets are explained by the deeper
weathering under warm and humid conditions. Klepper (1996) then suggests to obtain cell-
specific f
a
-values by interpolating between the two parameter sets based on annual cell
temperature. The values of Europe are valid for an average of 10C, while the values for Africa
represent an average of 25C. In order to avoid unrealistic values, the limits for f
a
are set to 0.05
30
and 1. In the future, the soil-, texture- and aquifer-related factors will be calibrated by analysing
base flow measurements.
Table 5.3: Aquifer types and corresponding values of f
a
(table modified from Klepper, 1996).
aquifer unit description corresponding
European aquifer
unit
f
a
(Africa)
f
a
(Europe)
1 Tertiary:
unconsolidated
sedimentary aquifers
of good permeability
Tertiary:
unconsolidated
sedimentary aquifers
of good permeability
plus brackish
aquifers
1.00 0.95
2 Quaternary:
unconsolidated
sedimentary aquifers
of medium to poor
permeability
Quaternary:
unconsolidated
sedimentary aquifers
of medium to poor
permeability
0.80 0.80
3 Paleozoic:
consolidated
sedimentary aquifers
of good permeability
Paleozoic:
consolidated
sedimentary aquifers
and fissured igneous
and metamorphic
rocks
0.95 0.40
4 Mesozoic:
consolidated
sedimentary aquifers
of medium to poor
permeability
Mesozoic:
consolidated
sedimentary aquifers
of medium to poor
permeability
0.80 0.70
5 Cenozoic and
Mesozoic volcanic
rocks
Cenozoic: volcanic
rocks
0.90 0.50
6 Precambrian:
igneous and
metamorphic rocks of
low to negligible
permeability
Precambrian:
igneous and
metamorphic rocks
of low to negligible
permeability
0.70 0.10
5.3.5.2 Fast surface and subsurface runoff
Fast surface and subsurface runoff is computed as the difference between total runoff and
groundwater runoff.
5.3.6 Urban areas
In urban areas, the soil is partially sealed to infiltration, such that surface runoff increases while
evapotranspiration and groundwater recharge decrease. The extent of this effect does not only
depend on the degree of sealing, but on the type and location of the sealing material and the
sewer system. While in the city centers typically 85-100% of the areas are sealed, in residential
areas with single family units, it may be 10% to 50%. In WaterGAP 1.0, 380 out of 59831 cells
are urban. As a first guess, it is assumed that 10% of the urban cell area is sealed in such a way
31
that the precipitation turns into surface runoff immediately. Only the remaining precipitation is
partitioned into evapotranspiration, surface and groundwater runoff.
5.3.7 Routing between grid cells
Routing between grid cells is not implemented in WaterGAP 1.0. The water availability within a
watershed is computed by summing up the cell values of surface and groundwater runoff. This
sum can be considered to be equivalent to the discharge at the mouth of the watershed.
5.3.8 Preliminary calibration to annual discharge of large rivers and regionalization
5.3.8.1 Calibration
Hydrologic models require calibration to measured river discharge even though they are
physically based models in which a variety of independently determined data is used (e.g. soils
and vegetation data). Consistent with our goal to compute annual water availability, consistent
with the long-term mean climate data and consistent with our model that does not take into
account routing, we calibrate the runoff coefficient in Eq. 10 by comparing computed total
annual runoff against long-term annual means of measured river discharge. The computational
value is the sum of all cell values of total runoff (Eq. 10) within all the watersheds that drain to
the respective measurement point of river discharge.
We cannot, however, use measurement data of river basins of which the digital
representation appears to be inconsistent with topographic maps. A further prerequisite is that
the measurement station is close to the mouth of one of the digitized subbasins of the river basin
(compared to the total size of the basin). We have chosen 43 large river basins on the five
continents for which long-term discharge measurements are available. They cover 33% of the
total land surface of the earth. For 39 rivers, discharge data from the GRDC (Grabs et al., 1996)
are used, while for four, data from Probst and Tardy (1987, as cited in Gleick, 1993, p. 149) are
taken. Table 5.4 lists the rivers and measurement stations and periods as well as the given river
basin areas and mean annual discharges. The WaterGap watersheds are combined to larger units
corresponding to the drainage area of the measurement point where necessary. The thus obtained
watershed areas can differ significantly from the areas given in the literature. While for 19
basins, the difference is less than 10 %, the Senegal and Nile river basin areas differ by more
than 65%; for such dry basins, it is difficult to determine the exact drainage area. For all basins
except the Guadalquivir, the WaterGAP basins appear to be reasonably similar to the basin area
that can be deduced from maps.
By trial-and-error, a good fit of the annual discharge of 38 of the 43 basins was obtained
by using the -values listed in Table 5.4, which range between 0.05 and 9.5. The higher the , the
lower the computed runoff. The arithmetic mean of is 1.8. For only five of the 38 basins, a
larger than 3 was fitted. These are relatively dry basins with a ratio of annual precipitation over
annual potential evapotranspiration E
p
/P > 2 (except the Loire).
However, for five river basins (Murray, Nile, Niger, Senegal, Don), the WaterGAP 1.0
hydrological model considerable overestimates runoff, no matter how large is. All of these
basins except the Don basin are very dry basins with E
p
/P ca. 3 (Only the Indus basin has an
equally high E
p
/P but a low ). In the case of the Murray, this is probably due to an
underestimation of the available water capacity in the root zone S
max
; in the model, too much
runoff occurs in winter, when E
p
is close to zero, which could be precluded by a value of S
max
which is approx. twice the assigned value of ca. 80 mm. The situation is different in the three
African basins of the Nile, the Niger and the Senegal. The overprediction of runoff of at least the
32
first two rivers is due to the fact that our hydrological model cannot simulate correctly the huge
swamps which evaporate a large percentage of the runoff produced in upstream regions;
averaged over the (long-term mean) year, in our model a cell can only evaporate what it receives
as precipitation. These swamps, however, receive water horizontally from the river (i.e. from
neighboring cells), which then evaporates at a high rate. For each of the five river basins, Table
5.4 lists the factor F
c
by which the discharge computed with = 10 has to be divided in order to
match the measured discharge.
33
Table 5.4: Calibration of runoff coefficient of Eq. 10 to measured annual discharge. Measurement data from GRDC (Grabs et al., 1996) except
where noted.
river station latitude
(+ N, - S)
longitude
(+ E,- W)
basin
area
[1000
km
2
]
measurement
period
mean annual
discharge
[m
3
/s]
mean annual
runoff [mm/a]
river basin area
in WaterGAP
[1000 km
2
]
runoff
coefficient to
fit discharge
E
p
/ P
Amazon Obidos -1.9 -55.5 4640 1928-1983 155432 1056 5155 0.8 1.14
Parana Corrientes -29.97 -58.85 1950 1904-1983 16358 265 2093 1.9 1.56
Orinoco Puente
Angostura
8.15 -63.6 836 1923-1987 31061 1172 967 0.8 1.03
Magdalena Calamar 10.27 -74.92 257 1971-1979 6974 854 275 0.9 1.05
Parnaiba Porto Formosa -3.47 -42.5 290 1963-1975 785 85 320 9.5 2.11
Uruguay Concordia -31.40 -58.02 249 1968-1979 5218 660 350 1.5 1.07
Brazos Richmond 29.58 -95.53 117 1965-1984 200 54 141 1.9 2.65
Mississippi Tarbert Landing 31.02 -91.62 3924 1965-1984 14703 118 3466 1.7 1.75
Fraser Hope 49.38 -121.45 217 1912-1990 2709 394 183 0.05 1.41
Columbia* The Dalles 45.6 -121.7 614 1879-1982 5452 280 648 0.1 2.86
Santiago El Capomal 21.83 -105.12 129 1965-1981 291 71 141 3.7 2.56
Murray Lock 9 Upper -34.18 141.6 991 1965-1984 257 8 1038 no fit; with =
10, F
c
= 3.23
3.90
Burdekin Clare -19.77 147.23 130 1965-1984 360 88 138 2.1 3.06
Zaire Kinshasa -4.3 15.3 3475 1903-1983 40250 3653 2797 1.4 1.39
Nile* Aswan 24.1 32.9 1500 1800-1976 2830 59 2497 no fit; with =
10, F
c
= 1.27
2.93
Niger Gaya 11.88 3.4 1000 1952-1990 1153 36 733 no fit; with =
10, F
c
= 2.25
3.21
Senegal Dagana 16.5 -15.5 268 1903-1974 691 81 476 no fit; with =
10, F
c
= 1.80
3.62
Zambezi* Matundo-Cais -15.85 33.58 940 1925-1983 2429 82 984 5.2 2.20
Rufiji Stigler -7.80 37.9 158 1954-1978 790 157 134 1.7 2.32
* Probst and Tardy (1987), cited in Gleick (1993).
34
Table 5.4 continued.
river station latitude
(+ N,
- S)
longitude
(+ E,
- W)
basin
area
[1000
km
2
]
measurement
period
mean annual
discharge
[m
3
/s]
mean annual
runoff [mm/a]
river basin
area in
WaterGAP
[1000 km
2
]
runoff
coefficient
to fit
discharge
E
p
/ P
Amur Komsolmolsk 50.63 137.12 1730 1933-1984 9739 178 1995 0.9 1.51
Yenisei Igarka 67.48 86.5 2440 1936-1984 17847 231 2318 0.3 1.35
Olenek Mouth of Pur 72.12 123.22 198 1965-1984 1000 159 157 0.05 1.61
Ob Salekhard 66.57 66.53 2950 1930-1984 12504 134 2947 0.9 1.52
Lena Kusur 70.7 127.65 2430 1935-1984 16622 216 2243 0.2 1.48
Godavari Polavaram 16.92 81.78 299 1901-1979 3061 323 337 1.7 1.92
Ganges Farakka 25.0 87.92 935 1949-1973 12037 406 976 1.8 1.60
Indus Kotri 25.37 68.37 832 1973-1979 2396 92 830 0.8 4.52
Brahmaputra Bahadurabad 25.18 89.67 636 1969-1975 19674 975 514 0.05 0.95
Chao Phraya +
Mekong
656 1976-1990 10094 485 950 2.3 1.36
Xijiang Wuzhou 3 23.48 111.3 330 1976-1983 7085 678 351 0.5 1.19
Changjiang Datong 30.77 117.6 1705 1923-1986 25032 463 1649 0.5 1.20
Huanghe Huayuankou 34.92 113.65 730 1946-1979 1465 63 789 1.4 2.61
Al Furat Hindiya 32.72 44.27 274 1964-1972 678 78 355 4.4 5.07
Guadalquivir Alcala del Rio 34.52 -5.98 47 1913-1984 434 291 69 0.6 (fit to
runoff)
2.14
Loire Montjean 47.38 -0.83 110 1863-1979 838 240 116 7.5 1.12
Odra Gozdowice 52.77 14.32 110 1900-1987 536 154 129 1.7 1.36
N. Dvina Ust-Pinega 64.10 42.17 348 1881-1985 3315 300 445 0.6 1.17
Volga* Volgograd 48.40 44.5 1350 1879-1975 8137 190 1412 1.7 1.23
Don Razdorskaya 47.50 40.7 378 1881-1984 787 66 481 no fit; with =
10, F
c
= 1.30
1.91
Kura Surra 40.12 48.67 178 1932-1984 547 97 95 2.4 1.33
Rhein Rees 51.77 6.4 160 1936-1984 2291 453 154 0.5 1.03
Danube Ceatal Izmail 45.18 28.8 807 1921-1984 6499 254 792 1.8 1.13
Wisla Tczew 54.1 18.8 194 1900-1987 1055 171 227 1.9 1.24
* Probst and Tardy (1987), cited in Gleick (1993).
5.3.8.2 Regionalization
For the computation of water availability in the watersheds belonging to the above 43 river
basins, the calibrated -values (and F
c
values where necessary) are used. Each cell of a
watershed gets the same parameter value. To the other watersheds (covering two thirds of the
land surface of the earth), we assign parameter values using a simple regionalization algorithm
that we have deduced from the calibration results. The regionalization parameters are the total
available water capacity in the effective root zone S
max
, the number of days below 0C and the
ratio of potential evapotranspiration over precipitation E
p
/P averaged over the watershed. The
regionalization proceeds as follows:
1. If S
max
> 150 mm or the number of days per year below 0C > 90, then is set to 0.7.
2. Otherwise, if E
p
/P > 2.8, = 10 and the computed runoff R of each cell is divided by a
correction factor F
c
of 1.4
3. Otherwise, = 2.0
This scheme can only be expected to give a very rough indication of the correct value of . The
scatter of the calibration data is obvious in Fig. 5.2, where the 's of all calibration basins are
drawn as a function of E
p
/P, and those which fulfil condition 1 or condition 2 are designated.
The parameter values assigned in the regionalization scheme correspond approximately to the
mean parameter values of the three different groups of calibration basins.
0
1
2
3
4
5
6
7
8
9
10
0 1 2 3 4 5 6
E
p
/P

Smax > 150 mm or

number of days below
0C > 90
Ep/P > 2.8
rest
Fig. 5.2: Derivation of regionalization scheme from calibration to annual discharge in 43 river
basins.
36
5.3.9 Comparison of WaterGAP estimates with other estimates
In Table 5.5, continental estimates of long-term average annual runoff computed by WaterGAP
are compared with literature estimates. Generally, the WaterGAP estimates are rather low. For
Asia, Europe & former USSR and North America, the WaterGAP estimate is the lowest of all
five estimates.
A map of long-term average annual runoff computed by WaterGAP is given as Fig. A1
in the Appendix; it shows the runoff generated in each cell. The map is representative for the
climatic conditions of the years 1931 to 1960. The overall spatial distribution of total long-term
average runoff was compared to maps by the Institute of Geography, USSR Academy of
Sciences, as depicted in WRI (1990, p. 169). The is a close correspondence between the two
maps. However, WaterGAP 1.0 estimates lower total runoff values in Northern Algeria and
Morocco as well as in Southeast China and Indochina.
Table 5.5: Comparison of WaterGAP 1.0 estimates of water availability in km
3
/yr with literature
estimates (based on summary table in Kulshreshtha, 1993, except for data from Baumgartner
and Reichel, 1975).
Continent WaterGAP 1.0
This paper
a
Lvovich (1979) Shiklomanov
(1990)
World
Resources
Insitute (1990)
Baumgartner
and Reichel
(1975)
Asia (excl. Asian
parts of former
USSR)
8,974 9,865 9670 10,485 NA
Africa 4,057 4,225 4,570 4,184 3,397
Australia & New
Zealand
1,142 731 348 2,011 2,394
Europe & former
USSR
4,996 6,475 7,950 6,705 NA
Central America 603 545 NA NA NA
South America 10,471 10,380 11,760 10,377 11,062
North America 3,730 5,415 NA NA NA
North America &
Central America
4,333 5,960 8,200 6,945 5,832
NA = Not available
a
long-term average 1931-1960


5.4 Water use model
The various uses of water can be divided into uses for which water is withdrawn from its
original location (withdrawal use) and uses of water in situ (instream use, such as shipping,
hydroelectric power plants, habitat for aquatic life). Most water use data refer to withdrawal
water use, which is divided into domestic, industrial and agricultural water use.
Part of the withdrawn water is returned to its source or to another location (return flow).
The difference between withdrawal use and return flow is referred to as consumptive use, i.e.
the amount of water which evaporates due to its withdrawal. Thus, a user might reuse the water
that has already been withdrawn by a first user who returns (part of) it to the stream. If return
37
flow is high, like, for example, in the case of thermoelectric power plants (included in industrial
water use), the amount of water used by two plants along a river in not much larger than that
used by one. Then withdrawal use, which would be computed as the sum of the withdrawals for
both plants, would overestimate the actual amount of water necessary to run the plants. On the
other hand, consumptive use would underestimate it because for the first plant along the river,
the amount equivalent to the withdrawal use of the first power plant must be available in the
first place. Thus, actual water use lies between withdrawal and consumptive use. On a global
scale, however, it is not possible to obtain this estimate. One would need to have information on
1. the quantity of return flows
2. quality of return flows and quality requirements of users
3. location of users within a watershed
4. instream uses
Available water use data mostly refer to withdrawal water use, except sometimes in the
case of irrigation water use. The World Resources Institute (WRI) (1990, 1992, 1994, 1996)
provides national withdrawal water use data of most countries of the globe, distinguishing
domestic, industrial and agricultural water use. Altogether, national water use data for 160
countries are required. We directly use the national data of withdrawal use in the domestic and
the industrial sector, and compute consumptive water use of these sector by multiplying
withdrawal water use with an efficiency factor. We translate the national water use data, which
are given as water use per person, to water use within a cell by taking into account population
density, urbanization and access to safe drinking water (using different algorithms for domestic
and industrial water use). In the case of agricultural water use, we use a different methodology.
We subdivide agricultural water use into water use for livestock and water use for irrigation.
Livestock water use is computed by multiplying livestock numbers with typical water use of one
animal; it can be considered to be withdrawal water use. In the case of irrigation water use,
however, we compute consumptive use from independent data on soil, climate and irrigated
area using our hydrological model. We then relate the consumptive use to the national data of
withdrawal use for agricultural purposes and determine an "irrigation correction factor" that
reflects not only the irrigation efficiency but also the uncertainty of model and data. This
procedure allows us to assess the influence of change of climate and irrigated area on irrigation
water use.
5.4.1 Domestic water use
5.4.1.1 National data of withdrawal use
National data of domestic water use are taken from WRI (1990, 1992, 1994, 1996), except for 8
countries, for which we use values given by Niederlnder and Dogterom (1996). The domestic
water use is expressed as the domestic water use per person (per capita). From the available
water use data, the value from the year closest to 1990 was selected. Depending on the country,
the water use data are from the years 1965 to 1994 (most values being from the 80s and 90s).
However, one must keep in mind that only the total water use is from the indicated years, while
the division of total water use into the sectoral water uses (domestic, industrial and agricultural)
was for most countries done for another year. For most countries outside Africa, this year was
1987 (WRI, 1996, p. 307). In order to reflect present-day water use, each national value is
extrapolated to 1990 by using the same assumptions as made in Scenario M (see Section 3.3.1),
and 1990 values of GDP (only if given year of maximum water use, in case of countries above a
per cap GDP of US$ 15,0000, was before 1978, extrapolation to 1990 was done using scenario
L if computations were to be made for Scenario L).
38
5.4.1.2 Allocation to grid cells
The domestic water use of people having easy access to drinking water is known to be higher
than that of people without access. The World Bank (1996) and WRI (1996) provide national
data on access to safe drinking water in urban and rural areas as well as the percentages of urban
and rural population in each country. Access to safe drinking water in rural areas implies that a
family member need not spend a "disproportionate" part of the day fetching water. In urban
areas, access to safe drinking water is defined as access to piped water or a public standpipe
within 200 m of a dwelling or housing unit.
We distribute the domestic water use of a country to the cells pertaining to it based on
population density (data set "Global Population Density Map at 5 x 5 Minutes Resolution" of
Tobler et al., 1995, as cited in van Woerden et al., 1995)
percentage of rural and urban population in each country (WRI, 1996)
percentages of inhabitants with access to safe drinking water in rural and urban areas (World
Bank, 1996; WRI, 1996).
The steps for allocating national domestic water use values to cells are as follows:
1. Rural population density of a country is assumed to be the same in each of the country's cells.
If the total population of a cell is less than the rural one, the difference is distributed equally
within the country.
2. The difference between the total population of each cell and the rural population is assumed
to be urban population.
3. The number of inhabitants in each cell that have access to safe drinking water is computed.
4. Those inhabitants without access to safe drinking water are assumed to have a domestic
water use per capita of 3.65 m
3
/yr (10 l/d). The rest of the national water use is equally
distributed to each inhabitant with access to safe drinking water. If the national data for a
country list a per capita domestic water use of less than 3.65 m
3
/yr, the listed value is
assumed to be valid for each inhabitant of this country.
5. The total domestic water use in a cell is computed as the sum of the water use by all the
inhabitants without access to safe drinking water and those with access.
5.4.2 Industrial water use
5.4.2.1 National data of withdrawal use
National industrial water use data are taken from the same sources as the domestic water use
data. Like the domestic water use, industrial water use is generally expressed as industrial water
use per capita (WRI, 1996). Functionally, however, industrial water use rather depends on
industrial production. Thus, for WaterGAP, per capita industrial water use is translated to
industrial water use per industrial GDP. Industrial GDP is determined as a fraction of GDP, the
fraction being given by the World Bank (1996) and WRI (1990, 1992, 1994, 1996), while GDP
values of the years, for which water use data are available, are derived from UN (1995). As in
the domestic sector, each national value for industrial water use is extrapolated to 1990 using
the assumptions of scenario M (see Section 3.3.2) and 1990 values of GDP and Industrial GDP
(only if given year of maximum water use, in case of countries above a per cap GDP of US$
15,0000, was before 1978, extrapolation to 1990 was done using scenario L if computations
were to be made for Scenario L).
39
5.4.2.2 Allocation to grid cells
Industry is assumed to exist only in urban areas. Thus, national industrial water use is allocated
to grid cells according to their urban population. The total industrial water use of a country is
equally distributed to all urban inhabitants (comp. section 3.1.2), and the industrial water use in
a cell is the product of industrial water use per urban inhabitant and the number of urban
inhabitants.
5.4.3 Agricultural water use
Agricultural water use can be roughly defined as the sum of water use for livestock and the
water use for irrigation.
5.4.3.1 Livestock
Livestock water use is computed by multiplying the numbers of livestock in each cell by a
livestock-specific water use. Table 5.6 lists the types of domestic animals and their specific
water use. Data on the present density of livestock on a 1 by 1 resolution is provided by
NCAR on the GlobalARC GIS Database (1996) except for chicken. Chicken population density
is derived from country-specific data (WRI, 1996); the chicken were then equally distributed
over the cells of each country. Total livestock water use in a cell is computed by multiplying
livestock-specific water use per animal (UN, 1986; cited in Kulshreshtha, 1993) with the
numbers of the specific livestock, and then summing over all the livestock types. The UN
livestock-specific water use values are low compared to water requirements given in many
books on animal production (e.g. 5 l/d might be typical for sheep, and 50 l/d for cattle according
to Jeroch, 1986).
Table 5.6: Livestock-specific water use.
livestock water use [l/d per animal]
cattle 25
pigs 4
sheep 2.25
goats 2.25
chicken 0.028
horses 15
camels 25
buffalo 25
5.4.3.2 Irrigation
Only part of the water withdrawn for irrigation is actually used up by the crop. This part is
called consumptive use. The rest is either "lost" by evaporation or seepage between the
withdrawal point and the application point or is necessary to leach the salts in the soil. The ratio
of consumptive use and withdrawal use is called irrigation efficiency. Once the location and
extent of irrigated areas is known, consumptive use can be determined as a function of climate,
crop, cropping intensity and soil type. In WaterGAP, the dependence on crop and soil type is
neglected mainly because it is not known which crop grows where.
5.4.3.2.1 Irrigated area
The FAO supplies data of total irrigated area for each country; there do not, however, exist
continental-scale maps of irrigated areas except for Africa. Therefore, we use an algorithm of
Klepper (1996) that distributes the total irrigated area (1990 values from FAO) within each
40
country. 1/12 by 1/12 grid cells are assigned to be either irrigated or not irrigated based on the
following scheme.
1. Irrigation only occurs if grid cell is less than 50 km away from a river.
2. The grid cells of each country which fulfill this requirement are ranked according to a weight
computed as
W(1/12 cell) = W
suitability
* [soil suitability] +
W
landcover
* [land cover irrigation likelyhood class] +
W
population
* [population density]
3. Within each country, the cells with the highest weights are consecutively assigned to be
irrigated until the total irrigated area of the country is allocated.
4. The resulting global binary map at 1/12 resolution is then transformed to a map at 0.5
resolution, where the fraction of land that is irrigated is represented.
Soil suitability is computed based on the FAO Soil Map of the World and a slightly
modified table of soil specific reduction factors listed by Leemans and van den Born (1994)
which takes into account fertility, salinity, acidity, drainage and rooting depth. Soil suitability
ranges from 0 (not suitable) to 100 (very suitable). Land cover irrigation likelihood class is
based on the landcover map of Olson et al. (1985), the resolution of which is too coarse (1/6,
but mostly 1/2) to be directly applicable. Klepper thought that seven of the 58 land cover types
of Olson et al. may indicate that irrigation occurs, and assigned them to three land cover
irrigation likelihood class (Table 5.7). Population density is expressed in inhabitants per km
2
;
the data set "Global Population Density Map at 5 x 5 Minutes Resolution" of Tobler et al.
(1995, as cited in van Woerden et al. (1995) is used. Klepper found that the irrigation map of
Africa could best be simulated by setting W
suitability
to 0.1, W
landcover
to 25 and W
population
to 1.
Obviously, population density is a strong "predictor" of irrigation, while soil suitability is only a
weak one.
Table 5.7: Conversion table between the land cover types of Olson et al. (1985) and land cover
irrigation likelihood class (from Klepper, 1996).
land cover irrigation likelihood class Olson et al. land cover
1
(may contain some irrigated land if other factors are
highly favorable)
CFS (cool farmland and settlements, more or less snowy)
MFS (mild/hot farmland and settlements)
FWG (field/woods with grass and/or cropland)
2
(may contain irrigated land)
CCI (cool cropland with irrigation of variable extent)
CCP (cold cropland and pasture, irrigated locally)
3
(may include considerable portion of irrigated land)
PRA (paddy rice and associated Land Mosaics)
WCI (warm/hot cropland, irrigated extensively)
5.4.3.2.2 Consumptive use
The consumptive use of irrigation is equal to the additional amount of evapotranspiration that
occurs due to irrigation. It is the fraction of actually applied irrigation amount which
evapotranspirates. The consumptive use depends on climate as well as on crop, cropping
41
intensity and soil type. In this assessment, the dependence on crop and soil type is neglected
mainly because it is not known which crop grows where. Consumptive use in each irrigated cell
is computed by
1. determining the days on which irrigation is necessary based on conditions for "natural"
growing and the required length of the growing period
2. computing the (net) irrigation amount on the irrigation days such that the sum of
precipitation and irrigation (consumptive use) is equal to potential evapotranspiration
3. multiplying the thus computed consumptive use with the fraction of irrigated area within the
cell.
According to FAO (1981), a normal natural growing season is characterized by precipitation P
eff
being larger than potential evapotranspiration E
p
and temperature T being higher than 5C. The
thus computed natural growing season is increased by the time it takes for the crop to empty the
soil moisture supply. Based on this concept, we devised a simple algorithm to compute the
irrigation period.
The period required for crop growth is set to a standard value of 150 d. This growing
period corresponds to a cropping intensity of 1. If cropping intensity is smaller than 1, not the
whole irrigated area is cropped every year. If cropping intensity is larger than 1, more than a
single crop is grown. Cropping intensities on irrigated areas for 13 world regions are given in
Table 5.8. The 13 world regions are those used in IMAGE (Alcamo et al., 1994a). The values
for developing countries are specific for irrigated areas in the period of 1988-1990 and are taken
from Alexandratos (1995), while the other values are from an estimate of Klepper (1996). In
order to account for cropping intensities smaller than 1, the consumptive use of each irrigated
cell is multiplied with the cropping intensity. In order to account for cropping intensities larger
than 1, a fraction of the cells of a world region is assumed to require a growing period of 300 d
instead of 150 d. With a cropping intensity of, for example, 1.2, the consumptive use of 20
percent of a region's irrigated area is computed with a growing period of 300 d. Those cells
within a world region are favored which due to their temperature regime allow a 300 d growing
period and require the least irrigation days (see next two paragraphs). A day is defined to be a
"natural" growing day if
1. T > 5C
2. P
eff
> E
p
(P
eff
> 0.5 E
p
in first 30 days of the growing season)
3. S > 0 mm
4. P
eff
> 0.01 mm/d
As expressed in condition 2, during the first month of the growing season it is sufficient for
precipitation to be half the potential precipitation, while later precipitation should be larger than
potential evapotranspiration. This requirement roughly takes account of the fact that during the
early growing stage the plants are still small and thus transpire much less water than later in the
growing season. It is consistent with the mean behavior of a variety of crops if we assume that
reference crop evapotranspiration is equal to potential evapotranspiration (comp. the crop
factors and duration of growth stages by Brouwer and Heibloem, 1986).
If, on a given day, condition 1 is fulfilled, but conditions 2 to 4 are not, irrigation is
necessary for crop growth. For each irrigation day, consumptive use is computed as E
p
- P
eff
(for the first 30 days: 0.5 E
p
- P
eff
). The cell-specific consumptive use is obtained by
multiplication with the fraction of irrigated area within the cell. The cell-specific growing period
is computed by finding the period of 150 days length during which the number of irrigation days
is smallest. The procedure to find the 300 d growing period is the same. In the case of a few
irrigated cells, it happens that no period of 150 d can be found with T > 5C. Then, the growing
period continues into the time period with T 5C. If not enough cells with 300 days greater or
equal 5C can be found within a region, consumptive use is first computed with the possible
42
number of cells with 300 d growing period; the consumptive use is then corrected by
multiplication with the ratio of the regional cropping intensity of Table 5.8 and the (lower)
cropping intensity implemented in the model. If in very arid regions P
eff
is never larger than 0.5
E
p
, January 1
st
is assumed to be the start of the irrigation period.
The above described method to compute consumptive water use can result in an
underestimation of actual consumptive use because
actual use is prone to be larger than the modelled optimal use
interannual variability leads to irrigation use in low precipitation years, while the surplus
water of high precipitation years cannot be transferred to low precipitation years
daily precipitation variability may lead to irrigation use (we only use mean monthly
precipitation values)
Table 5.8: Regional cropping intensities and water use efficiencies.
IMAGE region cropping
intensity
1995
a
cropping intensity
2025 and 2075
b
irrigation water use
efficiency
(estimated from project
efficiencies)
c
computed
irrigation
correction
factor
Canada 1.0 1.0 0.45 0.42
USA 1.0 1.0 0.30 0.17
Latin America 0.9 1.0 0.26 0.30
Africa 0.9 1.0 0.41 0.29
OECD Europe 1.0 1.0 0.35 0.48
Eastern Europe 1.0 1.0 n.a. 0.52
former USSR 0.8 0.8 n.a. 0.36
Middle East 1.0 1.1 0.31 0.96
India and South Asia 1.2 1.4 0.20 0.39
China and centrally planned
countries
1.5 1.5 n.a. 0.51
East Asia 1.2 1.3 0.33 0.05
Oceania 1.5 1.5 n.a. 0.89
Japan 1.5 1.5 0.22 0.03
a
from 1988/90 data in Alexandratos (1996) for developing countries excl. China and Klepper (1996) for others
b
from 2010 data in Alexandratos (1996) for developing countries excl. China; no change with respect to 1995 for
others
c
Bos and Nugteren (1982)
n.a.: data not available
5.4.3.2.3 Withdrawal use
Estimates of irrigation efficiency of 50 irrigation projects in various countries were obtained by
Bos and Nugteren (1982) by sending out questionnaires to irrigation project officer (Table 5.8).
On the other hand, there exists data on agricultural withdrawal water use for (almost) each
country (WRI, 1996; Niederlnder and Dogterom, 1996). We subtract livestock water use as
computed by WaterGAP from these values and obtain "literature values of irrigation withdrawal
use". Then, for each country, we determine the so-called irrigation correction factor as the ratio
of consumptive use (computed by WaterGAP with the 1995 average precipitation) and the
"literature value". If model and data were correct, the computed irrigation correction factor
should be similar to the irrigation efficiencies of Bos and Nugteren. However, the correction
factor also includes, on the model side, the errors due to the allocation of irrigated area within a
country, due to the algorithm to compute consumptive use per irrigated area and due to the
43
selection of the cropping intensity. On the data side, the errors in the total irrigated area per
country and in the agricultural water use per country are reflected. All the errors might be
considerable. Therefore, the correction factor varies widely, ranging from 110
-6
to 9, the
majority ranging between 0.001 and 1. The mean irrigation correction factors of each region are
listed in Table 5.8. Note that the coefficient of variance is typically around 1.
For the computation of irrigation withdrawal use, we have decided to trust more the
country data on agricultural water use than our model. Therefore, we divide the consumptive
use computed by WaterGAP by the (national) irrigation corrections factors. Keeping these
correction factors constant for each year and also when applying the 1-in-10-years precipitation
instead of the average precipitation, our model can compute the influence on irrigation water
use of the following factors:
climate change due to greenhouse gases
increase of irrigated agricultural area and cropping intensity
increased water use in dry years
Any possible increase in water use efficiency can be expressed by an increase of the correction
factor.
However, if the irrigation correction factor is less than, let us say, 0.1, it might well be that
the model underestimates consumptive use for average precipitation in 1995 considerable. If
such a low correction factor is used for low flow conditions, the withdrawal use for a 10-
percentile dry year might well be overestimated.
The algorithm to compute the distribution of irrigated area still uses the former USSR and
CSSR as single entities, instead of the new countries. Now that data of irrigated areas for each
of the new countries exist, it is obvious that the algorithm to allocate irrigated areas does not
function well in these areas. The best solution at the moment is to still compute the effect of
climate change, water use efficiency increase and 1-in-10-years precipitation with the
consumptive use model, then to average the effect over all the countries of the former USSR
and CSSR, respectively, and finally to scale the literature values of irrigation withdrawal use
with the computed factors.
A particular problem is that there are some countries for which our model does not
compute any consumptive use for average precipitation of 1995 although there exists an
appreciable irrigated area of more than 500 km
2
(Costa Rica, El Salvador, Guatemala, Guyana,
Suriname, Uruguay, Lebanon, The Philippines, Swaziland). Some of the reasons for this are
given in Section 5.7.2.2. For many of these countries, a consumptive use is computed for 1-in-
10 dry years (or increased irrigated area in the future). For these countries, we cannot, however,
determine an irrigation correction factor; Therefore, in all countries, where a consumptive use of
zero is computed in the case of 1995 average precipitation, we just take the "literature values of
irrigation withdrawal use" as the irrigation withdrawal use under average precipitation. In case
of the three countries with irrigation correction factors of less than 0.001, we do the same. For
these countries, we cannot model the increased irrigation water use in a 10-percentile dry year,
nor the impact of climate change on irrigation water use. The effect of increased water use
efficiency and increase in irrigated area, however, can be taken into account by simple scaling.
44
6. Freshwater criticality ratio
A freshwater criticality index can be based on the ratio of water need over available water
resources. Here, however, only water use instead of water need could be computed. Thus, areas
were water use is already restricted by (natural) water availability (but also lack of infrastructure
etc.) might not appear to be critical. To be on the safe side, one should look at withdrawal water
use because
1. instream uses are not considered here
2. the quality of return flow is unknown
3. the location of the water users within the watershed is not known
The available water resources are defined as the amount of water that is sustainably
available. It is the runoff (as surface runoff and groundwater recharge) produced within the
spatial entity of interest during a given time period under the indicated climatic conditions.
Very important for the estimation of water availability but also of water use is the spatial and
temporal scale. For this report, we compute the internally renewable water resources of whole
watersheds throughout a year. We do not consider the water flowing into or out of the
watershed, which might over- or underestimate the amount of water that is actually available
within a watershed. However, it is appropriate to just look at the internally renewable resources,
because imported water should be considered to be a resource that does not "belong" to the
watershed but that is just a (temporary?) "loan" from the upstream watershed. Looking at whole
watersheds neglects the differences in water available upstream and downstream. Looking at
annual values results in underestimating the criticality during months of high irrigation water
demand and low precipitation if water supply is from surface water. At this stage of our
research, we cannot, however, look at monthly values because we have not calibrated our
hydrological model to measured monthly discharge values.
Here, we call the ratio of annual water use within a watershed over the annual runoff
produced within a watershed the freshwater criticality ratio. This definition implies the
assumption that withdrawn water can be transported within the watershed without restrictions; it
is consistent with
the fact that generally water cannot be transported across watershed boundaries but relatively
easy within a watershed, and
the idea that water supply and use should be managed on the watershed scale (i.e., for
example, not within country boundaries in case of an international watershed).
Such a criticality ratio should not be computed for average annual conditions of water
availability and use only, but also for more critical conditions representative of years that are
drier than normal. Here, we determine both a criticality index for average precipitation
conditions (C
average
) and one for 10-percentile dry conditions (C
10d
). For the latter index, we
compute the water available and used during a 10-percentile dry year (i.e. only 1 in 10 years has
less precipitation than such a 10-percentile dry year). While the amount of surface water
available is computed with the 10-percentile precipitation, we assume that the amount of
groundwater which is sustainably available in this year is that of the average year. The amount
of irrigation water required is also particular for the 10-percentile precipitation, while domestic
and industrial water use do not vary with precipitation. The freshwater criticality ratios of
watersheds are defined as
45
C
annual withdrawal water use in cell during average year
annual runoff in cell during average year
average
watershed
watershed

(13)
C
annual withdrawal water use in cell during percentile dry year
annual runoff in cell during percentile dry year
d
watershed
watershed 10
10
10

(14)
Following Kulshreshtha (1993), we combine the freshwater criticality ratio with the water
availability per capita to obtain a freshwater criticality index. This index ranges between 1 for
water surplus to 4 for water scarcity. Given the same criticality ratio, the watershed is more
vulnerable with respect to water resources if the per capita water availability is low. For
example, if water use is restricted by (low) water availability, the criticality ratio may be
relatively low, but the situation should be called critical. Table 6.1 shows the definition of the
freshwater criticality index; it is defined on the watershed scale.
The criticality ratio and the criticality index of countries computed exactly like that of
watersheds. This method does not reflect correctly the influence of international watersheds on
the water availability of a country. However, as we do not currently take into account the
topology within and between watersheds, this approach is appropriate.
Table 6.1: Definition of the freshwater criticality index (after Kulshreshtha, 1993).
criticality ratio (use/availability)
per capita water
availability
[m
3
/(cap yr)]
< 0.4 0.4 - 0.6 0.6 - 0.8 > 0.8
< 2,000 2 3 4 4
2,000-10,000 1 2 3 4
> 10,000 1 1 2 4
1: water surplus
2: marginally vulnerable
3: water stress
4. water scarcity
Acknowledgments
The authors are very grateful to O. Klepper for making available his data bases and calculations
for the purpose of developing the WaterGAP model Version 1.0. They also are indebted to R.
Leemans, E. Kreileman and their colleagues at RIVM for cooperation in the development of the
WaterGAP model. They are also very grateful to S. Kulshreshtha for providing advice and data
for the water use calculations of the WaterGAP model. The climate data from the Max-Planck-
Institut climate model was kindly provided by the DKRZ.
46
References
Alcamo, J., Kreileman, G.J.J., Bollen, J.C., van den Born, G.J., Gerlagh, R., Krol, M.S., Toet, A.M.C., de Vries,
H.J.M. (1996): Baseline scenarios of global environmental change. Global Environmental Change, 6, 261-
304.
Alcamo, J., Kreileman, G.J.J., Krol, M.S., Zuidema, G. (1994a) Modeling the global society-biosphere-climate
system: Part 1: Model description and testing. Water, Air and Soil Pollution 76, 1-35.
Alcamo, J., van den Born, G.J.J., Bouwman, A.F., de Haan, B.J., Klein Goldewijk, K., Klepper, O., Krabec, J.,
Leemans, R., Olivier, J.G.J., Toet, A.M.C., de Vries, H.J.M, van der Woerd, H.J. (1994b): Modeling the
global society-biosphere-climate system: Part 2: Computed scenarios. Water, Air and Soil Pollution 76, 37-
78.
Alexandratos, N. (ed.) (1995): World Agriculture: Towards 2010 - An FAO Study. John Wiley & Sons, Chichester.
Batjes, N.H. (1996): Development of a world data set of soil water retention properties using pedotransfer rules.
Geoderma 71, 31-52.
Baumgartner, A., Reichel, E. (1975): Die Weltwasserbilanz. R. Oldenbourg.
Bederke, E., Wunderlich, H.G. (1968): Groer physikalischer Weltatlas. II. Geologie. Meyers Verlag.
Bergstrm, S. (1994): The HBV model. In Singh, V.P. (ed.): Computer Models of Watershed Hydrology. Water
Resources Publications 443-476.
Bergstrm, S. (1990): Parametervrden fr HBV-modellen i Sverige. SMHI Hydrologi no. 28.
Bos, M.G., Nugteren, J. (1978): On Irrigation Efficiencies. International Institute for Land Reclamation and
Improvement.
Dierke (1975): Atlas.
FAO (1991): FAO Soil Map of the World. World Soil Resources Report 67 (Release 1.0). FAO, Rome.
FAO (1981): Report on the agro-ecological zones project. Vol. 3 Methodology and results for South and Central
America. FAO, Rome.
Forsythe, W.C., Rykiel Jr., E.J., Stahl, R.S., Wu, H., Schoolfield, R.M. (1975): A model comparison for daylength
as a function of latitude and day of year. Ecol. Modelling 80, 87-95.
Gleick, P.H. (1993): Water in Crisis: A Guide to the World's Fresh Water Resources. Oxford University Press, New
York.
GlobalARC GIS Database (1996) by CRSSA, Rutgers University and U.S. Army CERL.
Grabs, W., De Couet, T., Pauler, J. (1996): Freshwater fluxes from continents into the world oceans based on data
of the Global Runoff Data Base. Report no. 10, Global Runoff Data Centre, Federal Institute of Hydrology,
Koblenz, Germany.
Hutchinson, M.F. (1995): Interpolating mean rainfall using thin plate smoothing splines. Int. J. Geogr. Inf. Syst. 9,
385-403.
IAH (1974-1990): International Hydrogeological Map of Europe 1:1 500 000. Various sheets. Bundesanstalt fr
Geowissenschaften und Rohstoffe, Hannover, Germany - UNESCO Paris.
Intergovernmental Panel on Climate Change: J.T. Houghton, B.A. Callendar, S.K. Varney (eds.) (1992): Climate
Change 1992. The Supplementary Report to the IPCC Scientific Assessment. Cambridge University Press.
Jarvis, P.G. (1976): The interpretation of the variations in leaf water potential nd stomatal conductance found in
canopies in the field. Phil. Trans. R. Soc. B237, 593-610.
Jeroch, H. (1986): Vademekum der Ftterung. VEB Gustav Fischer Verlag Jena.
Klepper, O. (1996): Documentation for the calculation of world-wide water satisfaction on catchment basin level.
Unpublished report, RIVM, The Netherlands.
Kulshreshtha, S.N. (1997): Personal communication.
Kulshreshtha, S.N. (1993): World Water Resources and Regional Vulnerability: Impact of Future Changes. RR-93-
10, IIASA , Laxenburg, Austria.
Leemans, R., van den Born, G.J (1994): Determining the potential distribution of vegetation, crops and agricultural
productivity. Water, Air and Soil Pollution 76, 133-161.
Leggett, J., W.J. Pepper and R.J. Swart. 1992: Emission Scenarios for the IPCC: An Update. In: IPCC, Climate
Change 1992: The Supplementary Report to the IPCC Scientific Assessment. Cambridge University Press,
Cambridge.
Lozar, R. (1992): Global climatic change management by watershed basin units. Proceedings of the American
Society of Photogrametry and Remote Sensing, August 92, ASPRS, Bethesda MD, Volume 4, 150-.
Manabe, S., Spelman, M.J., Stouffer, R.J. (1992): Transient response of a coupled ocean-atmosphere model to
gradual changes of atmospheric CO
2
. Part 2: Seasonal response. J. Climate 5, 105-126.
47
Manabe, S., Stouffer, R.J., Spelman, M.J., Bryan, K. (1991): Transient response of a coupled ocean-atmosphere
model to gradual changes of atmospheric CO
2
. Part 1: Annual mean response. J. Climate 4, 785-818.
Marks, D., Dozier, J. (1992): Climate and energy exchange at the snow surface in the alpine region of the Sierra
Nevada, 2. Snow cover energy balance. Water Resour. Res. 28, 3043-3054.
Mays, L.W. (1996): Water Resources Handbook. McGraw-Hill.
McNaughton, K.G., Jarvis, P.G. (1991): Effects of spatial scale on stomatal control of transpiration. Agric. For.
Meteorol. 62, 279-302.
Niederlnder, H.A.G., Dogterom, J. (1996): Water demand as supply and water related emissions in the domestic,
agricultural and industrial sectors - An inventory of world-wide data coverage. ICWS Report 96.03.
Olson, J., Watts, J.A., Allison, L.J. (1985): Major world ecosystem complexes ranked by carbon in live vegetation:
A database. Report NDP-017, Oak Ridge National Laboratory, Oak Ridge, Tennessee 164 pp.
Priestley, C., Taylor, R. (1972): On the assessment of surface heat flux and evaporation using large scale
parameters. Mon. Weather Rev. 100, 81-92.
Riehl, H. (1979): Climate and Weather in the Tropics. Academic Press.
Rckner, E., Arpe, K., Bengtsson, L., Christoph, M., Claussen, M., Dmenil, L., Esch, M., Giorgetta, M., Schlese,
U., Schulzweida, U. (1996): The atmospheric general circulation model ECHAM-4: Model description and
simulation of present day climate. MPI-Report No. 218, MPI fr Meteorologie, Hamburg.
Shuttleworth, W.J. (1993): Evaporation. In Maidment, D.R. (ed.): Handbook of Hydrology. McGrawHill 4.1-4.53.
UN Department for Economic and Social Information and Policy Analysis Statistical Division
(1995): Statistical Yearbook 1993. UN, New York.
UN (1988): Africa: Major Hydrogeological Formations. After Geological Map of Africa 1:5 000
000. Map no. 2033, revision 1. UN, New York.
UN (1987): Fertility Behavior in the Context of Development. Population Studies 100.ST/ESA/SER.A/100. Dept of
International Economic and Social Affairs: New York.
van Woerden, J. W., Diederiks, J., Goldewijk, K.Klein (1995): Data Management in Support of Integrated
Environmental Assessment and Modelling at RIVM - Including the 1995 RIVM Catalogue of International
Data Sets . RIVM Report No. 402001006.
World Bank (1991): World Development Report 1991. Oxford University Press, New York.
World Bank (1996): World Development Report 1996. Oxford University Press.
World Resources Institute (1996): World Resources (1996-97) - A Guide to the Global Environment: The Urban
Environment. Oxford University Press