While past strategies for agricultural water management have focused on irrigation (use of blue water), this paper demonstrates the dominance of green water in food production. A global, yet spatially disaggregated, green-blue analysis of water availability and requirement, using the LPJmL dynamic vegetation and water balance model, indicates that many countries currently assessed as severely water short are able to produce enough food for their populations if green water is considered and is managed well. The need to integrate green and blue water management is highlighted in a future scenario of water availability under climate change and population growth (HadCM2 A2). For 2050, the scenario indicates that 59% of the world population will face blue water shortage, and 36% will face green and blue water shortage. Even under climate change, good options to build water resilience exist without further expansion of cropland, particularly through management of local green water resources that reduces risks for dry spells and agricultural droughts.
 Most forms of land use are closely linked to water resources in two different ways [Falkenmark and Mikulski, 1994]. On the one hand by vegetation being genuinely water-dependent, on the other by interaction with fundamental hydrological processes taking place in the same domain near the ground surface where the rain input is being partitioned between vapor flow and liquid water flow. This paper has its focus on one particular form of land use, agriculture, and its links to the challenge of feeding a growing humanity. This is a form of land use where management of soil, crops and water are fundamental tools to make crop production more effective, increasing the output of crops per unit of water evapotranspired.
 The rainwater partitioning takes place at two different partitioning points: the upper partitioning point at the land surface between surface runoff and infiltration, and the lower, between root water uptake forming transpiration, evaporation from wet soil, and percolation for groundwater recharge. One fundamental question is to what degree intensification of crop production on the land already under agriculture will be enough to produce the amount of food required and what that will imply in terms of soil, crop and water management, and consequently redirection of water flows, impacting on water resources for other uses. A second question is to what degree horizontal expansion of agriculture will be required, thereby “grabbing water” from other ecosystems, i.e., green water now used by other terrestrial ecosystems, or blue water, functioning as habitats for aquatic ecosystems.
 The core of this paper involves an analysis of water availability and requirements for food production today and some 40 years from now. Water availability is estimated in terms of both the naturally infiltrated green water in the soil (green water resource) and the liquid blue water in rivers and aquifers (blue water resource) available for complementary irrigation in cases where green water is too scarce to meet crop water requirements (for a definition of green and blue water, see Figure 1). Future water requirements and availability are tightly linked to land use. Food security alternatives in regions where water deficiencies make agriculture on current croplands insufficient necessitate strategic choices between horizontal expansion of croplands, involving altered land use; irrigation even in regions where water is already overappropriated, leaving aquatic ecosystems under severe pressure; and food import influencing crop production in more water abundant regions.
 While population growth increases water requirements by about 1,300 m3 per capita per year for each additional person [Falkenmark and Rockström, 2004], climate change is projected to have regionally different effects on water availability. Increasing temperatures will increase agricultural water demand everywhere, whereas changes in precipitation vary strongly from region to region and also among climate models and emissions scenarios [Intergovernmental Panel on Climate Change (IPCC), 2007]. In addition, rising atmospheric CO2 concentration has a direct effect by increasing plant water use efficiency [Leipprand and Gerten, 2006; Gerten et al., 2007]. In terms of climate change, it is of particular concern what can be foreseen for the already water scarce tropical and subtropical regions, which represent today's poverty and hunger hot spots [Rockström et al., 2007a] and for which much of the past debate has had its focus on drought and desertification, i.e., absence of water rather than the opportunities linked to presence of water [Falkenmark and Rockström, 2008]. Hence, the water scarcity situation in the future will be determined by the detailed pattern of change in climatic, demographic, and other processes such as land use change and technical development. (For a definition of water scarcity concepts, see Figure 2.)
 The ongoing population growth and the expected rising incomes will have implications in terms of rising food demands, not only to feed the growing population but also to meet changing food preferences [Lundqvist et al., 2007]. Since food production on either rain-fed land (supplied mainly by green water as provided by rainfall, and sometimes augmented by small additions of supplementary blue water) or irrigated land (supplied mainly by blue water withdrawn from rivers, reservoirs and aquifers in addition to the green water) is inevitably coupled to consumptive water use, these changes will also have implications for the local water resources (cf. Figure 1).
 Already in 1989, an effort was made to analyze green-blue water scarcity for the African countries. In that study the agricultural yield increase needed for food self-sufficiency was related to difficulty of access to irrigation water for the agricultural intensification [Falkenmark, 1989]. An index-based approach was used, in which the required yield increase, expressed as degree of technological sophistication (basically additional green water requirement) was related to the level of water crowding (number of persons sharing every flow unit of blue water). The result was a two-digit index for green water requirement/degree of blue water shortage for individual countries. In this paper we continue along a similar track.
 In two recent studies, the water requirements for food self-sufficiency on a country by country basis were analyzed, using slightly different approaches. One of them [Rockström et al., 2007a] had its focus on population growth and how to meet the longer-term millennium development goal (MDG) of hunger alleviation in 92 developing countries. It was based on future calorie levels as foreseen by FAO for the developing world by 2030 (3,000 kcal per capita per day), assuming 20% of animal protein. The other study had its focus on the global implications of globalization-related income rising in terms of altered food preferences, especially the animal protein component [Lundqvist et al., 2007]. Both studies went on to analyze the available water resources that could be relied upon to meet those water requirements, assuming that both naturally infiltrated soil moisture/green water and blue water (surface and groundwater) would be made use of. For both of these types of water resources, there are huge losses that could be put to productive use: In the former study this was expressed by the potential to increase water productivity and crop yields, in the latter study by trying to cut losses in the food chain from field to plate. Both studies also analyzed the additional land requirements needed for self-sufficient food production. The basic conclusions drawn in these two studies were that, given realistic improvements of water productivity, horizontal expansion of cropland will be unavoidable, remaining at the same order of magnitude as currently ongoing.
 Some 15 years after the development of the water crowding index [Falkenmark, 1989; Stockholm International Water Institute (SIWI), 2007], the green-blue thinking and modeling advances have made it possible to make a process-based analysis of the green and blue water presently used for food production, as well as the total available water resources that potentially could be used within the agricultural sector [Rost et al., 2008]. An advantage with using spatially explicit physically based models that estimate water flows, is that upstream consumption of water is accounted for and is thus not available downstream.
 This study goes beyond previous global analyses of availability, demand and scarcity of water resources, by jointly addressing green and blue water, with the help of a physically based, spatially distributed water model. The aim is to analyze the possibility of different countries for meeting current and future water requirements for food production by increasing green and blue water productivity in agriculture, after assessing availability of these two resources. The potential of green water for increasing resilience to global change is finally addressed. More specifically, the study was guided by the following objectives: (1) quantifying spatially explicit green and blue water use and availability around the year 2000, and green and blue water availability around 2050 accounting for both climate change impacts (including CO2 effects) and growing population; (2) comparing total green and blue water availability and requirements for food production; and (3) analyzing the options for resilience building by comanagement of blue and green water and improving its productivities, to meet a future with growing social and environmental vulnerability.
2.1. General Approach
 The analysis in this paper is made from data estimated by the global vegetation and hydrology model LPJmL (described in detail below), which computed green and blue water availability and consumption at a spatial scale of 0.5°. Simulations were run to calculate present (i.e., 1996–2005 average) and future (2046–2055 average) water availability. In the latter case, two scenarios were distinguished, one including climate change only and the other one including both climate and population change.
 The assessment of green and blue water flows has been made at two spatial scales. (1) At the pixel level (0.5°), it provides a basis for evaluating the potential for rain-fed farming, e.g., for poor subsistence farmers to increase their yields, producing food also for local market and income raising. This analysis made it possible to identify water scarcity-driven global poverty and undernutrition hot spots where risk for conflict and/or urban, transcountry or international migration may culminate. (2) At the country level, it allows evaluation of potential for food self-sufficiency, which if not met will indicate either risk of out migration or, alternatively, food trade needs.
2.2. Definitions of Green and Blue Water Availability, Use, and Requirement
 In this paper, we have defined blue water availability from the perspective of the proportion of runoff water that can be sustainably withdrawn for food production, given the sum of the blue water available in the rivers BR, stored in lakes and reservoirs BL, and in groundwater BG [Rost et al., 2008], multiplied by a factor 0.7 to account for environmental flow requirements (30% of mean annual river discharge were found to be a good approximation for the latter, at least for semiarid regions [Smakhtin et al., 2004]). Blue water use is the sum of all evaporative flows of blue water, which is almost exclusively due to irrigation, as municipal and industrial water consumption is comparatively low. Throughout this paper the term use is equal to consumptive use, i.e., the proportion of water appropriation that does not return as blue water flow (in rivers, groundwater or lakes). While blue water flows have a consumptive (evaporation and transpiration) and a nonconsumptive (return flow) portion, green water flow is per definition always a consumptive use of freshwater.
 Green water availability is defined as the soil moisture available for productive vapor flows from agricultural land (evapotranspiration ET from cropland and permanent pasture), i.e., total rainfall infiltration over agricultural land minus runoff from this area multiplied by a reduction factor 0.85 that describes minimum evaporation losses in agricultural systems that cannot be avoided (generally evaporation losses at pregermination and early germination stages of a crop) [de Wit, 1958]. The productive green water flow (i.e., transpiration, T) was used to represent green water use. That means that the difference between green water use and green water availability represents the potential for improving green water productivity by shifting vapor flows from unproductive evaporation/interception to productive transpiration and, hence, increasing plant growth and crop yield without compromising downstream blue water resources. Moreover, the ratio between green water use and green water availability is thus the transpiration efficiency for each country. To estimate the green water availability per capita, the population that does not live in an agricultural area (i.e., in a pixel that does not contain agricultural areas) is assumed to be evenly distributed over pixels that contain agricultural areas, within the respective country.
 Finally, 1,300 m3 per capita per year is defined as the green and blue water required per capita for sustaining a “standard diet” [Falkenmark and Rockström, 2004]. This figure is a starting point for developing a global combined green-blue water shortage index. It builds on assumptions of diet composition (20% meat-based products and 80% vegetarian products) at total calorific intake of 3,000 kcal per capita per day, and assumed water productivities, and conforms to earlier estimates of human water requirements [e.g., de Fraiture et al., 2007]. As such, it does not take into account any future dietary changes, such as shifts from beef to poultry, or to more vegetarian diets, or any future improvements of water productivity. It is also a global average, and does not account for regional differences. Moreover, it is used for both croplands and pasturelands, since these two categories are not separated in the present study.
2.3. General Overview of the LPJmL Model
 For the above analyses, we applied the well-established LPJmL dynamic global vegetation and water balance model, which has been extensively validated against biogeochemical and hydrological observations including leaf phenology, crop yields, river discharge, soil moisture, and green and blue water use [e.g., Sitch et al., 2003; Gerten et al., 2004; Bondeau et al., 2007; Rost et al., 2008]. LPJmL simulates the establishment, growth, seasonal phenology, productivity and water use of natural and agricultural vegetation at daily time steps on the basis of explicitly coupled physiological, biogeochemical and hydrological process representations [Sitch et al., 2003; Gerten et al., 2004]. Nine natural and twelve agricultural plant functional types are considered. The latter include the world's most important field crops (temperate and tropical cereals and roots, rice, maize, pulses, sunflower, soybean, groundnuts, rapeseed) as well as pasture (grazing land). The plant functional types can principally coexist at any location, if the prevailing climate permits their establishment and survival. While the distribution of natural vegetation as a function of climate was simulated by LPJmL, the distribution and fractional coverage of pasture and cropland was prescribed for each year using (1) a data set of the yearly fractional coverage with cropland per grid cell [Ramankutty and Foley, 1999], (2) a data set of the present distribution of crop types within every grid cell [Leff et al., 2004], and (3) data on the contemporary distribution of grazing land [Klein Goldewijk and Battjes, 1997]. Croplands and pastures were merged into one category (managed lands) when calculating green water availability over agricultural land, since there are large uncertainties in the global data sets mapping pasture and cropland, which has been explained by difficulties in separating the two in mosaic landscapes where field sizes are small and commonly located in between pasturelands [Hannerz and Lotsch, 2008]. The growing season of the crop types was initiated in the model by sowing (computed as dependent on long-term average temperature or precipitation), and harvest occurred when maturity or the fixed end of the growing period was reached [Bondeau et al., 2007].
 Irrigation was assumed to occur in the equipped areas indicated by Siebert et al. , but only to the extent that the green water available in the soil was insufficient to guarantee optimal growth of the present agricultural vegetation. These irrigation water requirements were assumed to be withdrawn from local (within a pixel or one neighboring pixel) surface water bodies, aquifers and river discharge accumulated along the river network (i.e., BR, BL and BG), after accounting for domestic and industrial water needs and regional differences in irrigation efficiency. In the case that irrigation water requirements exceed these resources we assume that the irrigation water requirements can be met by withdrawal from an outside source not accounted for in the model (e.g., fossil groundwater, desalinization plants and river diversions). Note that while in rain-fed areas all crop and grassland production relies on green water (i.e., local precipitation), on irrigated land both green and blue water are involved. In LPJmL we traced the flows of green (precipitation) and blue (irrigation) water in the soil in order to distinguish their individual contributions to productive (transpiration) and unproductive (soil evaporation and interception loss) water consumption. In general, the partitioning of precipitation into runoff, transpiration, evaporation and interception was calculated for each grid cell as a function of climate (precipitation, net radiation and temperature) and vegetation (crop/plant functional type, leaf area index, seasonal phenology, and management practices such as irrigation) (details in the work by Rost et al. ).
2.4. Simulation Protocol
 For this study, the LPJmL model was forced by the CRU TS2.1 gridded data set of monthly air temperature, precipitation and cloud cover [Mitchell and Jones, 2005] for the period 1901–2002, preceded by a 900-year spin-up period during which the climate of 1901–1930 was repeated. Daily weather variability was emulated by linear interpolation between the midmonth values, and by using a stochastic procedure in the case of precipitation [Gerten et al., 2004]. To derive a future scenario of climate, CO2 and population change, the model was run until the year 2055 forced by the climate from the HadCM2 scenario normalized with the CRU TS2.1 1961–1990 climatology [Mitchell et al., 1995] under the economy-oriented SRES A2 emissions trajectory [Nakicenovic and Swart, 2000], as described by Müller et al. . To compute water availability per person, we used gridded population data for the present and the future, which are consistent with the emission and climate change scenario [Bengtsson et al., 2006]. Future changes in land use and in the extent of irrigated areas were not considered here (as our focus was on effects of climate and demographic change, and because our aim was to explore water shortage and the potential for its alleviation on present agricultural land); instead, the values for the year 2002 were held constant until the end of the simulation period (2055). Present blue and green water availability and use were analyzed as annual averages for the period 1996–2005, whereas potential future conditions were analyzed for the period 2046–2055. The individual influences of population change and climate change by this period were derived as the difference from a simulation in which population was held constant at the 2000 level. All analyses were performed for every 0.5° grid cell separately (while accounting for the fact that blue water consumed upstream is not available downstream). Results shown for countries or continents always represent values aggregated from the grid cell scale.
3. Hunger Alleviation Challenge: The Role of Green Water
3.1. Dominance of Green Water
 Currently, the consumptive use of water in agriculture is dominated by green water (Figure 3). Food production in Europe, Africa and South America depends almost exclusively on green water. It is only in a few regions, such as in parts of south Asia and North America, where consumptive use of blue water on cropland and pasture exceeds that of green water (for more details, see Rost et al. ).
 Globally, consumptive use of green water on cropland is four times that of blue water, indicating the enormous potential for making green water more productive in agriculture (in particular by reducing the difference between total green water flows and productive green water flows, i.e., transpiration) (Table 1). Basically, the huge difference (5,300 km3 a−1) between total and productive green water consumption on rain-fed cropland constitutes a sizable potential water resource.
Table 1. Approximate Global Totals of Contemporary Green and Blue Water Use in Agriculturea
Both total evapotranspiration, ET, and productive transpiration, T, are shown. Units are km3 a−1.
Growing period of crops only (excluding transpiration from intermittent grassland in fallow periods).
Green, rain-fed cropland
Blue, irrigated cropland
3.2. Conceptualization of Water Predicament
 Given this dominance of green water in agriculture (and hence in human appropriation of water resources), the conventional measures of blue water scarcity do not reflect the real water situation. An illustration of the water predicament that some countries find themselves in now and around 2050 provides an integrated green and blue water analysis for future food water requirements (Figure 4). Total water requirement for food self-sufficiency is, as already indicated, taken as 1,300 m3 per capita per year [Falkenmark and Rockström, 2004]. A location within the green square illustrates the potential to meet food water requirements by green water only, seeing 100% as the resource that can potentially be exploited. When food water requirements are beyond 100% of the green water resource on current agricultural land, irrigation will be needed and can be added to the field, provided that blue water shortage does not prevent irrigation. At locations in the red square, green water requirements exceed availability on agricultural land and blue water cannot be used to augment the green water resource because of chronic blue water shortage; these countries will have to rely on cropland expansion and/or imports of food. In the top left corner, requirements exceed green water availability but it is still possible to use more blue water for irrigation, while in the bottom right corner, it is not possible to exploit more blue water for irrigation, while there is still room for more efficient green water use.
 The only country of those depicted in Figure 4 that is currently facing chronic blue water shortage (<1,000 m3 per capita per year [SIWI, 2007]) is Iran, while both Ethiopia and India are heading toward chronic blue water shortage when having to meet the food water requirements of their growing populations in around 2050. Nigeria has ample rainfall, but not that much over current cropland or pasture, which in combination with rapid population growth explains the high ratio of agricultural water requirements in 2050 in relation to available green water (ET over agricultural land). In Kenya, the climate is expected to get wetter in 2050 in HadCM2, explaining why blue water crowding decreases despite population growth.
 To conclude, a blue water focus excludes the green water dimension which is, as shown in Table 1, the most important water source for food production globally. Figure 4 indicates the need for comanagement of green and blue water resources to meet future needs.
3.3. Local Blue and Green Water Shortage
 A comparison between blue water shortage (Figure 5a) and combined blue and green water shortage (Figure 5b) shows that the majority of the extensive areas that are presently characterized by chronic blue water shortage (i.e., below 1,000 m3 per capita per year), actually have an adequate overall supply of water (i.e., green and blue water availability greater than 1,300 m3 per capita per year) to meet the water required for producing the standard diet. This includes most of those areas in sub-Saharan Africa that are blue water scarce.
3.4. Potential Green Water Reserve at the Country Scale
 When aggregating water availability from grid cell to country level, many countries which are presently classified as chronically blue water short actually have enough blue plus green water to produce a standard diet (see Figure 6). Such countries are located in the top left quadrant (green area). In some of these countries (e.g., Eritrea, Uganda and Lesotho) the total water availability is beyond 2,600 m3 per capita per year; twice the amount needed to produce a standard diet, making them potentially able to be self-sufficient in food production. Other countries within the same group (e.g., Algeria, Morocco and Afghanistan) have slightly less total water available (1,300–2,600 m3 per capita per year); here total water availability makes self-sufficiency potentially possible, provided water losses are kept to a minimum. Countries that are too water short even when the green water availability is included, and which therefore cannot be self-sufficient in food production (total water availability below 1,300 m3 per capita per year) in the bottom left quadrant (red area) include Jordan, Israel, West Bank, Pakistan, Lebanon, Iran, Iraq and also Rwanda and Burundi. In yet another group of countries (blue area) including for example Egypt, Japan and Bangladesh, green water availability is too low to meet food water requirements even if nonproductive green water losses are minimized (green water availability <600 m3 per capita per year), while blue water could be utilized for irrigation (total water availability above 1,300 m3 per capita per year); however within this group there are some countries that are approaching chronic water shortage (i.e., already below 1,700 m3 per capita per year) and thus irrigation expansion might be less feasible. Finally, there are some countries that are approaching chronic blue water shortage (i.e., already below 1,700 m3 per capita per year) but where green water resources keep them potentially self-sufficient if green water is carefully managed (yellow area). This group includes India, China and Malawi.
4. Analysis of Future Green and Blue Water Scarcity
 When comparing green-blue water shortage projected for the 2050s with the present situation, a large number of additional countries will fall below the critical threshold of 1,300 m3 per capita per year, indicating insufficient water for food self-sufficiency (Figure 7). Water shortage will thus become a serious problem throughout the MENA countries and in a number of countries in South Asia (Table 2). In sub-Saharan Africa, Burkina Faso, Malawi, Rwanda, Burundi and Uganda will also enter the water scarce cluster. Thus, what the HadCM2 scenario under SRES A2 suggests is that these countries will have to expand cropland area and/or rely to a considerable degree on food imports in the future, no matter how well blue and green water are managed.
Table 2. Population in Different Regions Facing Water Shortage in 2000 and 2050a
Blue Shortage 2000
Green-Blue Shortage 2000
Blue Shortage 2050
Green-Blue Shortage 2050
Values account for only blue water and for blue and green water together (numbers in billion persons). Threshold value chronic blue water shortage is 1,000 m3 per capita per year; blue-green shortage is 1,300 m3 per capita per year.
World water shortage
 The dramatic scale of the green-blue water shortage developing in the next few decades is illustrated in Figure 8 as follows.
 1. While more than 3 billion people suffer from chronic blue water shortage at present, only less than 300 million or 4.5% remain after taking the green water resource into account.
 2. This situation will however change radically in the next 4–5 decades, so that by mid century, some 36% will be water short even when green water on current agricultural land is taken into account.
 3. This change suggests the need for cropland expansion and/or dramatic changes in world food trade.
 Note that the increasing water shortage computed here, encompassing mainly countries in the MENA and South Asia regions (Figure 7), is due primarily to increasing population and, thus, increasing freshwater requirements. In more recent scenarios [IPCC, 2007] than the one chosen here, the MENA region is projected to be subject also to severe reduction in precipitation and hence water availability [IPCC, 2007].
 The LPJmL simulation results confirm earlier assessments [Vörösmarty et al., 2000], that at least over the next few decades population change seems to be a stronger driver of change in water availability than climate change (data not shown). In individual countries, blue and green water shortage may develop differently. While we generally expect a decrease in per capita availability of both resources green and blue, there may be exceptions, e.g., in cases where precipitation (and hence total water availability) is projected to increase as in the present scenario in eastern Africa.
5. Options for Water Resilience
5.1. Importance of Resilience Building
 The green-blue conceptualization of available water resources or overall (“effective”) water shortage has now made possible a more thorough, future-oriented analysis of the complementarity between rain-fed and irrigated agriculture aiming at improved comanagement of land and green-blue water. The earlier studies have clarified the special need to look closer on the large dryland region in the (sub) tropics, better denoted as the savannah region [Falkenmark and Rockström, 2004]. Our analysis has shown that the blue water approach to water scarcity for food production is too narrow and in many cases misleading. Figure 5 demonstrated the enormous local level difference when incorporating also green water in the conceptualization of water availability and shortage. With the new approach, a country level outlook shows that in the year 2000, practically no country was chronically water short (Figure 7a). During the following 50 years, however, climate change and population growth probably contribute in pushing several countries into such water-short situations (below 1,300 m3 per capita per year): the MENA region countries, Eritrea, Uganda, Burkina Faso, Iran, Iraq, Pakistan, Nepal and Uzbekistan (Figure 7b).
 The envisaged growing pressure on water resources, from human demand and climate change, will require adaptation to increasing water shortage in many countries. The situation will be exacerbated by the fact that rainfall variability will be increasing even further in savannah regions, adding another pressure on top of the overall trend in water scarcity. These combined pressures will increase the frequency of water-related shocks: floods, droughts, and dry spells, which will require a strong emphasis on building water resilience in agriculture as a key adaptation strategy to deal with human and climate related water scarcity.
5.2. Avoiding Water-Related Shocks
 In the critical savannah regions, resilience (i.e., the capacity of a society or system, such as a farming system, to withstand a disturbance, such as a drought or flood, and continue to develop [Folke et al., 2002]) will be of crucial importance to avoid the risk for conflicts and large-scale out migration if poverty and hunger alleviation would not be possible to achieve. Such out migration would have severe implications for the rest of the world in view of emerging risks for political instability.
 As shown in Table 3, our green-blue analysis of country-level water predicaments reveals a differentiated spectrum of adaptation and resilience strategies to cope with rising water shortage challenges. Only one extreme category of countries (category A in Table 3) emerges in terms of limitations to water development, namely, the countries subject to extreme water shortage (category A) because of lack of both green and blue water resources, where social resilience through food imports and insurance systems will be the dominating strategy. In the other end of the spectrum, we find the countries (category D) with apparently adequate availability of both green and blue water resources, where water governance and management is less constrained by water limitations. However, for blue water availability, category D includes two different sets of hydrological situations as a result of the way our blue water availability analysis is carried out. We have estimated the weighted average blue water availability per capita at a country level. The implication is that a country with apparent high blue water availability can still have a large portion of its population suffering from blue water shortage, as all the blue water may not be accessible (people live far from the rivers). The result (as shown in Table 3) is that many of the blue water scarcity prone countries in the savannah regions appear in this category (Niger, Senegal, Kenya, Zimbabwe, South Africa, Tanzania and Mali). These countries have a larger river passing through a limited portion of their countries, giving a large but not easily accessible blue water resource.
A, chronic blue water shortage, green plus blue water shortage (chronic blue water shortage (<1,000 m3 per capita per year), total water shortage (blue plus green) (<1,300 m3 per capita per year)); B1, green water freedom under chronic blue water shortage (chronic blue water shortage (<1,000 m3 per capita per year), total (blue plus green) water availability (>1,300 m3 per capita per year)); B2, green water freedom under blue water shortage (blue water shortage (<1,700 m3 per capita per year), total (blue plus green) water availability (>1,300 m3 per capita per year)); C, blue water freedom under green water shortage (blue water availability (>1,700 m3 per capita per year), green water shortage (<600 m3 per capita per year)); D, blue and green water freedom (blue water availability (>1,700 m3 per capita per year), green water availability (>600 m3 per capita per year)).
A: chronic blue water shortage, blue + green water shortage
Rw, Bu, Iran, Leb, Pak, WBank, Isr, Jo, UAR, Quat
Food imports, social and financial insurance systems, investments in unconventional water sources
B1: green water freedom under chronic blue water shortage
Rainwater management and soil moisture conservation, runoff water harvesting systems, spatial catchment planning, adaptive water governance at microcatchment and mesocatchment scales
B2: green water freedom under blue water shortage
Iq, India, China, Mwi, Tajik, Turk, Azer, BuFa, Ethiop
Same as above plus microcatchment and supplemental irrigation
C: blue water freedom under green water shortage
SKor, Egypt, NKor, Bang,
Integrated water governance at river basin scale, infrastructure development for irrigation
D: blue and green water freedom
Sri L, El S, Togo, Niger, Senegal, Kenya, Zim, South Africa, Tanz, Mali
Simultaneous strategies in irrigation development and water, management in rain-fed agriculture
 Countries facing blue water freedom, while green water is inadequate to meet requirements, include South Korea, Egypt, North Korea and Bangladesh (category C). In these countries there is an option to developing infrastructure for irrigation.
 Both categories B and D (in Table 3) include countries with significant degrees of freedom to develop green water use, which sets new priorities for water resilience. In these countries sources of resilience to rising water shocks due to growing competition for scarce water resources and increased frequency of floods, droughts and dry spells, will shift governance from the conventional focus on blue water resources, to a wider green-blue approach, offering new opportunities to build water resilience. This is particularly the case for the regions subject to blue water shortage but with present degrees of green water freedom (B1 and B2 with blue and green water availability >1,300 m3 per capita per year), and also to regions in countries of category D where the green water is available while blue water is inaccessible.
 The most poverty and hunger stricken countries in the world fall into these categories (B and D), which also correspond to the savannah (or so-called dryland) regions of the world, which normally are portrayed as having very limited opportunities for building water resilience in agriculture. This analysis has presented evidence that the green water source is significant in these countries and possible to exploit as a strategy for water resilience to climate change. In this case, water resilience is focused on improved management of local green water resources in rain-fed systems. A key emphasis is to minimize risks for droughts and dry spells. Water harvesting and conservation agriculture form two key management strategies in this regard. Water harvesting for upgrading rain-fed agriculture includes a wide spectrum of in situ green water conservation strategies (e.g., terracing, microbasin, bunding) aimed at maximizing rainfall infiltration, and external water harvesting systems where local runoff water is concentrated to cropland (runoff diversion systems) or stored in ponds and microdams for supplemental irrigation [Hatibu et al., 2006]. Small-scale supplemental irrigation systems has been shown to have a particular potential in building water resilience in rain-fed agriculture systems by bridging dry spells [Molden, 2007], which in tropical savannah farming systems are seasonal occurrences [Barron et al., 2003]. Conservation agriculture includes management practices that avoid soil inversion (generally through plowing) which enables the buildup of green water availability in the root zone. These reduced and minimum tillage systems already play a large-scale role in agricultural development in, e.g., Latin America [Derpsch, 2001], and increasingly in sub-Saharan Africa [Rockström et al., 2008].
 However, there is a limited ability of any management system to cope with meteorological droughts, where the minimum crop water needs exceed cumulative volumes of available water, and evidence shows that dry spells dominate as the critical source of water scarcity in savannah farming systems [Barron et al., 2003]. The focus of green water resilience building is thus to maximize the capacity to bridge intra-annual dry spells [Falkenmark and Rockström, 2008]. For any agricultural system, avoiding agricultural droughts (i.e., droughts caused not primarily from lack of rainfall but from poor soil management) and dry spells requires the adoption of soil, crop and water management strategies that (1) secures adequate soil water/green water availability and (2) maximizes plant water uptake capacity.
 The same amount of rainfall may result in very different productivity levels in the savannah regions [Rockström and Falkenmark, 2000]. Management related failures often “lock” the production system in an extremely low agroecological productivity state, with yield levels for basic staple grains generally <1 t ha−1, which are commonly prevailing in e.g., savannah regions in sub-Saharan Africa [Rockström et al., 2007b]. Such systems consume less than 10% of the seasonal rainfall available, and present “desertification-like” conditions, despite having a potential to often triple or even quadruple sustainable productivity levels. Here, apparent droughts are common but are not due to real rainfall deficits. What is scarce is instead plant-available soil moisture/green water in the root zone. This presents a key win-win opportunity for agricultural productivity improvements with limited downstream reductions in blue water availability. Shifting nonproductive evaporation losses in favor of productive transpiration flows is feasible through improved crop and soil management practices, which improves both yields and water productivity [Rockström et al., 2007b]. This can be achieved both through in situ reductions of evaporation flows, e.g., through improved tillage and mulching practices (E to T direct conversion), but also by increasing overall water productivity (ET per unit yield) through supplemental irrigation (which increases the ratio of T/ET) [Oweis and Hachum, 2001].
5.3. Green Water-Based Potential for Food Self-Sufficiency by Vapor Shift
 When looking at the current green water use at country level in relation to requirement, we find that most countries in theory have a green water-based self-sufficiency potential and are in other words in the position to produce all required food locally (Figure 9). The concave line corresponds to 600 m3 per capita per year of transpiration required for production of a standard diet (i.e., the fraction of productive water flow of the estimated total water need of 1,300 m3 per capita per year). Thus, countries that fall above (or to the right of) this line have an availability >600 m3 per capita per year of productive green water under current water productivity levels, and are thus theoretically able to be self-sufficient in terms or food production from a water perspective. Countries, which normally are portrayed as subject to water shortages, but which through this analysis present significant water potentials to meet their food needs, include Kenya, Ethiopia, Mali and Mozambique.
 Many countries on the left side of the curve show a productive fraction of green water use (i.e., green water use over availability, or transpiration efficiency) of between 0.2 and 0.8. This clearly demonstrates a potential for increasing food production through vapor shifts from unproductive evaporation to productive transpiration without requiring addition of blue water resources. Largest potential is available for countries such as Bangladesh, Pakistan, and India (now around 30% productive use of green water flow), and somewhat less so China, Iran, Iraq, and Jordan. Countries where this potential is very low (beyond 60%) include Israel, West Bank, Lebanon, Togo, Cameroon, Benin, which thus appear to already have a high fraction of productive to total vapor flow.
 The present LPJmL-based analysis demonstrates that green water flows dominate agriculture globally and in most regions of the world. Even basins that according to conventional assessments [World Water Assessment Programme, 2006] are dominated by irrigation (i.e., blue water use), such as the Jordan River basin, have in fact green water contributions to food production of similar magnitude as blue water contributions. This is not only a reflection of existing rain-fed agriculture in these basins, but is also due to the fact that most irrigated areas do receive and consume a considerable amount of rainfall during the cultivation season.
6.1. Efforts to Better Characterize Water Shortage
 Accordingly, water shortage of basins and countries changes when taking green water into account. This suggests that the original blue water crowding index [Falkenmark, 1989] to express water shortage on the basis of blue water availability only is inadequate when addressing the new challenges of meeting rapidly growing water needs for future food production. A blue water approach may be appropriate for extremely dry environments, where transboundary rivers (e.g., Egypt) or groundwater resources (e.g., the Middle East) play a dominant role in meeting human water requirements. For the rest of the world, a broadened green-blue approach to water shortage is critical to identify options for water resilience building. This has set us in the quest for the new green-blue index of water shortage, which we tried to develop from the analyses presented in this paper.
 However, research remains before we are able to define a threshold of green water availability (corresponding to the blue water threshold of 1,000 m3 per capita per year denoting chronic water shortage according to the original water crowding index). The reason is the difficulty in defining the green water resource; how to deal with the productivity of green water use (T/ET ratio) and to identify at what intensity of green water use a farming system, region or country faces water shortage. For this paper, therefore, we chose to assess our water availability analyses against an average cumulative green and blue water need of 1,300 m3 per capita per year, instead of the conventional blue water index of 1,000 m3 per capita per year, in order to translate availability analyses to water scarcity predicaments.
 We believe the analysis presented in the paper to be relatively robust and conservative. By defining the green water resource to only include actual ET flow on cropland and permanent grazing land, we restrict the “resource” to water that is already today under agriculture (i.e., not allowing for “grabbing” of green water flows currently sustaining other terrestrial ecosystems such as forests, grasslands, wetlands, or overestimating the green water potential by assuming that green water flows from marginal lands are accessible for other use). This excluded green water resource is very significant, accounting for approximately 75% of the global green water flow from land [Rockström et al., 1999; Rost et al., 2008]. The choice of an average water requirement threshold of 1,300 m3 per capita per year to meet a standard dietary demand, is a relatively high figure, based on current water productivity estimates, a diet requirement of 3,000 kcal per capita per day and a mix of 20% animal protein-based foods (at a water productivity of 4 m3 1,000 kcal−1) and 80% vegetarian foods (at a water productivity of 0.5 m3 1,000 kcal−1).
 Despite this conservative approach (considering only a limited portion of accessible green water and applying a relatively high water requirement to meet human needs), a new green-blue approach moves (in the analysis of the current situation) most people in today's world out of physical water shortage. When projecting into the future, according to the HadCM2 A2 scenario chosen, LPJmL simulations indicate that the MENA region and western and southern Asia will experience the most serious increases in green and blue water shortage. This is mostly due to population growth, and to a lesser extent due to climate change (the latter is expected to significantly reduce water availability in the Mediterranean region in particular).
 We recognize the inconsistency of taking all available blue water into account when calculating blue water availability, but only the green water on agricultural land when calculating green water availability. This inconsistency seems to be acceptable, since in most countries only a small fraction of blue water withdrawals is consumed outside of agriculture, i.e., by households and industry. We intend to derive a more realistic measure for blue water availability in forthcoming studies, e.g., subtracting per country the blue water in very remote areas and blue water during flood events that is actually not accessible to humans [Postel, 1998]. Moreover, as pointed out by Vörösmarty et al. , blue water shortage is more appropriately expressed at pixel level rather at country level, in order to give a more realistic relationship between the water resource and the actual accessible water for people. In this paper blue water availability, when expressed as country level averages, is still calculated as a weighted average; that is, we do not take into account the spatial inaccessibility of blue water concentrated in a certain part of a country. This means that our analysis exaggerates blue water availability in two respects; that all blue water is available from the source, and that all is spatially accessible. The dilemma with this approach is illustrated, e.g., by Niger, which is a country where the Niger river passes through only the southern tip of the country, generating a high average blue water availability at a country level, which however is not accessible for the vast majority of the population.
6.2. Some Interesting Country Cases
 Bringing together the results from the different sections, some interesting country profiles appear. Kenya is particularly interesting as the outlook for the country in terms of actual water shortage is transformed when green water is included. When looked at from a purely blue water perspective, Falkenmark  found Kenya to be one of the most exposed countries in terms of rapidly growing water requirement for food self-sufficiency, impossible to be met by irrigation because of chronic water shortage. In this analysis, we note however that there is plenty of unused and available green water that Kenya may benefit from. There is also a potential for water productivity increase through vapor shift from nonproductive evaporation to productive transpiration. Not even by 2050 will the country become water short when both blue and green water availability and climate change are taken into account.
 Pakistan and Iran are interesting in the sense that they are chronically water short even when the green water is taken into account. This poses severe challenges for the future, particularly for Pakistan by even having passed the “water barrier” of 500 m3 per capita per year [Falkenmark, 1989]. Both countries fortunately have some potential for water productivity increase through vapor shift.
 China and India both find themselves in rather equivalent situations as far as average green-blue water availability is concerned. In terms of blue water they have already passed below the 1,700 m3 per capita per year threshold for blue water shortage, and social driving forces at work are pushing them toward chronic blue water shortage. Their predicament however improves when green water is included. By 2050, including green water, India will have passed below 1,300 m3 per capita per year, whereas China is somewhat better off. Even here, both countries present good opportunities for water productivity improvements.
 A whole cluster of sub-Saharan countries, finally, are in fact water rich with more than 4,000 m3 per capita per year, altogether. Many of these countries have plenty of blue water, and have earlier been denoted “economically water scarce countries” [de Fraiture et al., 2001] in the sense that they do not have the economic resources and infrastructure for exploiting their blue water resources more intensively (dams, pipelines, boreholes, pumps). However, as discussed above, much of this blue water may in fact not be easily accessible, even with significant investments, because of the distance from the resource. For the sub-Saharan countries, a green water strategy, which aims at raising green water productivity with low-cost practices to manage the local on-farm water balance in rain-fed agriculture, appears more appropriate to meet increasing food water requirements.
 The green water resource, as defined in this paper, includes green water flows from both cropland and permanent pasturelands. Exploiting more green water for food production, may occur at the expense of pasturelands, i.e., putting more pastureland under cropland. As pointed out by the UN Millennium Ecosystem Assessment , grazing lands provide ecosystem functions apart from only fodder for livestock (e.g., biological diversity), and spatial diversity in landscapes contributes to ecological resilience (e.g., water regulation). This means that increasing green water use may in fact include critical trade-offs between different ecosystem services (food versus other ecological functions).
 The analysis of grid cell-based calculations of green water availability in relation to water requirement, assumes self-sufficiency within a grid cell, which is unrealistic as food and other agricultural commodities are traded both within and between countries. Thus, the degree of actual water scarcity (i.e., the actual social water availability through virtual water movement of food) may be slightly different from our analysis suggests.
 The analysis in this paper has broadened the freshwater resource to include green vapor flows, and has focused on a spatially distributed analysis of water availability. However, a key future challenge is the projected increase in temporal variability of rainfall due to climate change. Already today, the most water scarcity prone regions of the world are also those subject to the highest variability in rainfall distribution, causing recurrent droughts and dry spells. With climate change, the frequency and amplitude of extreme rainfall events are expected to increase, raising water related vulnerability. In future analyses the risk for deviations in green and blue water availability across temporal and spatial scales will have to be taken into account, in order to truly understand the predicament, and thereby identify solutions, for the majority of the inhabitants on Earth deemed to face various degrees of green and blue water shortage in the future.
6.3. Land Use Change Implications
 This study suggests that if the assumed slow fertility decline scenario (A2) would in fact materialize, additional food water requirements by 2050 may be of the order of 5,700 km3 a−1. This raises the question of from where all this additional water may be drawn. There are basically four different sources to turn to (1) water productivity improvement (turning E into T), (2) irrigation expansion as deemed acceptable in relation to environmental flow requirements, (3) horizontal cropland expansion, involving land use change, and (4) virtual water import by food trade, involving land use change elsewhere.
 The scale of the additional green water requirements is by no way neglectable. Figure 10 illustrates the challenges in four different transnational river basins, showing both the increase in green water requirement beyond cropland green water availability on agricultural land, and the green water availability over remaining land. It should be noted that the possibility for horizontal expansion of cropland is quite limited in some regions of the world where almost all arable land is already under agriculture.
 Global estimates of future additional water requirements, assuming food self sufficiency at country level as the politically preferred food policy, varies with assumption on rate of fertility decline and water productivity improvements (Table 4). Starting with the additional requirements arrived at in our calculations (some 5,700 km3 a−1) a more rapid fertility decline toward the UN medium population projection would reduce this amount by almost 30% down to around 4,100 km3 a−1. Water productivity improvements have an even larger effect, bringing down the per capita food water requirements from 1,300 m3 per capita per year to 1,000 m3 per capita per year [Rockström et al., 2007a]. This would lower the additional water requirement by almost 40%. Combining both these approaches would reduce the additional requirements by some 60% to around 2,300 km3 a−1.
Table 4. Additional Water Requirements for Food Self-Sufficiency and the Influence of Fertility Decline and Water Productivity Improvementsa
Additional Requirements (km3 a−1)
Fertility decline population scenarios are for 2050. Low WP (water productivity), 1,300 m3 per capita per year; high WP, 1,000 m3 per capita per year; A2, low fertility decline; UN medium, high fertility decline.
 Three sources remain for covering additional water requirements: irrigation, horizontal expansion and virtual water import/trade. Even if assuming that as much as an additional 15% of blue water accessibility (70% of blue water availability) could be designated for irrigation, except for situations with chronic blue water shortage, the possible contribution would however be very limited (less than 10% of what remains after the earlier much more effective reductions) (Figure 11). Large water requirements in fact remain to be met by horizontal expansion or virtual water trade. (It is to be noted that only water availability to meet requirements in each country is accounted for in the analysis. The result is that in countries with water surplus, green water availability is equal to food water requirement, and is therefore dependent on assumptions on population size and water productivity; hence the slight difference in green water availability estimates between population and water productivity scenarios.)
 What this analysis of the options for meeting the additional food water requirements by 2050 has shown, is that the best way to avoid the massive food water requirements by 2050 is through water productivity improvement which is almost 10% better than speeding up fertility decline. In combination these two approaches would save some 60% of additional water requirements. We have also seen that irrigation will only contribute a marginal amount of water in spite of the generous assumptions to increase the water withdrawal by as much as 15% of the resource. The analysis also indicates, however, that it will not be possible to escape a horizontal cropland expansion by some 1,020 Mha, unless that can be avoided by virtual water import/trade. This conclusion fits very well with the findings of Rockström et al. [2007a] for 92 developing countries, and implies a cropland expansion continuing at about the same rate as today.
 To conclude, a blue-only focus on water for food production excludes the green water dimension. As shown in Table 1, green water is the most important water resource for food production globally, indicating the enormous potential for making green water more productive in agriculture. Many countries which are presently classified as chronically blue water short actually have enough blue plus green water to produce a standard diet. This includes for example Algeria, Tunisia, Syria, Morocco and Uganda. There is however a cluster of countries that remain chronically water short even when green water is included (e.g., Rwanda, Burundi, Pakistan, Jordan, West Bank, Iran). In these countries the buildup of social resilience will be essential to sustain food security and/or to rely on virtual water imports. Some countries that are rich in both blue and green water, still suffer from limited access to blue water due to economic (so-called economically water scarcity), or to spatial constraints (water passing through part of the country only). These countries thus rely heavily on their green water resource.
 In the scope of future resilience building to feed humanity in a world characterized by increasing population and climate change, a green-blue approach to water management will be necessary. This involves productivity improvements across the full range of agricultural water management options, from purely rain-fed agriculture, to supplementary irrigation with water harvesting, and finally to full irrigation. While the conventional blue water approach expects 6.5 billion people to be living in countries with chronic water shortage by 2050 (Table 2), the green-blue conceptualization has made possible a more thorough future-oriented analysis of where countries are heading. Quite a number of countries across the MENA region all the way to India will be developing severe water shortage problems even when looking at the overall blue and green water availability. We have suggested that by mid century some 36% of the world population will be living in countries without ability to be self-sufficient in food production. Major implications will have to be foreseen in terms of cropland expansion and world food trade. In view of the ongoing climate change which will intensify climate variability, and therefore increase challenges of coping with droughts and desertification, it will be important to build up a green water potential or “improved green and blue water productivity,” especially important in blue water-short countries with green water potential many of which belong to the MENA group.
 As indicated in the discussion of this paper, much research remains in order to advance our understanding of the linkages between green and blue water availability, water shortage and strategies to govern and manage water resources sustainably across scales. Global models like LPJmL need to be developed further to simulate the cumulative effects of changes in agricultural water management in the global water system for different scenarios.
 The quest for a new integrated green-blue water shortage index to support integrated water resource management is still on, and will particularly require interdisciplinary research linking social indicators of water shortage with hydrological data of green and blue water availability. Also, the future changes in the shortage index will have to be evaluated by use of more climate and socioeconomic scenarios than the one studied here, since it is known that especially the precipitation patterns vary among projections from different climate models [IPCC, 2007], and that, e.g., population projections differ depending on the SRES scenario chosen. Furthermore, in this paper, future land use change is not explicitly considered (only population growth and climate change). Major implications will have to be foreseen in terms of cropland expansion and world food trade. The analysis has clarified that even at low fertility increase, cropland area would have to be expanded by some 1,000 Mha, i.e., continue at around the same rate as today. Such land use change, both in terms of expansion of agricultural land, but also shifts in farming systems, e.g., expansion of bioenergy crops, and changes in agricultural productivity, need to be included in future analyses of green and blue water resources in a world subject to rapid social and ecological change.