Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 Mountains are important sources of freshwater for the adjacent lowlands. In view of increasingly scarce freshwater resources, this contribution should be clarified. While earlier studies focused on selected river systems in different climate zones, we attempt here a first spatially explicit, global typology of the so-called “water towers” at the 0.5° × 0.5° resolution in order to identify critical regions where disproportionality of mountain runoff as compared to lowlands is maximum. Then, an Earth systems perspective is considered with incorporation of lowland climates, distinguishing four different types of water towers. We show that more than 50% of mountain areas have an essential or supportive role for downstream regions. Finally, the potential significance of water resources in mountains is illustrated by including the actual population in the adjacent lowlands and its water needs: 7% of global mountain area provides essential water resources, while another 37% delivers important supportive supply, especially in arid and semiarid regions where vulnerability for seasonal and regional water shortage is high.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 During the past two decades, mountains have gained increasing attention in science and politics. Especially the mountain-specific chapter 13 in the Rio Earth Summit's Agenda 21 [UNCED, 1992] underscores the global role that mountains play in debates on environment and development issues [Funnell and Price, 2003]. What concerns water resources, the symbolic term “water towers” is widely adopted today, expressing the importance of mountains for providing freshwater for the adjacent areas downstream [e.g., Bandyopadhyay et al., 1997; Liniger et al., 1998]. Furthermore, population distribution is closely linked to occurrence of large river flows [Meybeck et al., 2001], which often originate in mountains. With the International Year of Mountains 2002 and the following International Year of Freshwater 2003, the concerns of mountain water resources have obtained further legitimacy. In a changing environment with population growth, the role that mountains play for lowland water supply has to be identified more closely.
 Although there is great need for closer knowledge about mountain runoff for the sustainable management and planning of water resources, progress in view to these questions lags behind other fields of hydrological research [Kundzewicz and Kraemer, 1998]. Sparse and topographically biased gauging networks [Briggs and Cogley, 1996] as well as large errors especially in precipitation measurements [Sevruk and Kirchhofer, 1992] contrast with the high spatial heterogeneity of hydrological, meteorological and climatological patterns in mountains. Consequently, accurate figures for the hydrological importance of mountains on a global scale are still disputed today, although the principles of the hydrological highland-lowland linkage have been recognized long since. While regional studies in densely gauged areas may attain higher accuracy [e.g., Viviroli and Weingartner, 2004b; Weingartner et al., 2007], uncertainties are large on the local and the global scale.
 A recent study including all climate zones found an estimate of 32 percent for mountain contribution to global discharge [Meybeck et al., 2001], while regionally, mountain discharge may represent up to 95 percent of total flow in a catchment [Liniger et al., 1998]. A comparative assessment of case studies [Viviroli, 2001] demonstrated that the contribution from mountains to total runoff is about twice what would be expected on basis of their share in surface area. A global assessment assembled from these case studies [Viviroli et al., 2003] identified the most marked water towers as “wet islands” within dry climate zones. Examples for inferior importance of mountains on downstream hydrology were found in humid tropical climates (e.g., Mekong, Orinoco, and Amazon rivers). It was concluded that each region has its own particularities and that it is therefore necessary to consider climate zones (e.g., humid tropical versus arid subtropical versus temperate) and scales (global, continental, regional) to reveal the hydrological importance of mountains under different conditions [Viviroli and Weingartner, 2004a].
 Moving on from the single-basin perspective, we attempt a global assessment of the hydrological importance of mountains in order to identify regions which should receive special attention in research and water resources planning. First steps toward a spatially distributed global-scale analysis were made by Dürr . With the currently available global data and the possible resolution of 0.5° × 0.5° grid cells, a certain degree of generalization has to be taken into account, while on the other hand, it becomes possible to identify critical zones on the global scale by connecting information about the available freshwater resources in the highlands with human water needs in the adjacent lowlands. Since the aim is to assess actual water resources in river systems, we explicitly focus on long-term mean values of runoff and not on precipitation. We consider natural water fluxes, disregarding reduction of river flow due to water use, now frequent in dry regions [Vörösmarty and Meybeck, 2004; Meybeck and Vörösmarty, 2005]. Statements concerning groundwater flows have to be omitted because of insufficient data.
2. Definitions and Data Sources
2.1. Water Towers
 The term “water tower” originally refers to “a tower supporting an elevated tank, whose height creates the pressure required to distribute the water through a piped system” [Soanes and Stevenson, 2004]. In the context of hydrology, it is used as a symbolic term for a mountain area that supplies disproportional runoff as compared to the adjacent lowland area. The concept is therefore a relative one, the extent of disproportionality also depending on the location from which the mountains are looked at. We will focus on the disproportionality of mountain runoff as compared to average conditions in the respective lowlands. Furthermore, a “water tower” can have multiple functions, such as for ecosystems and water resources (irrigation, industrial or domestic supply) as considered in this article, but also for hydropower, sediment loads and nutrient balance.
2.2. Mountains and Lowlands
 Universally valid definitions for mountains can only be vague and depend on the connotation in focus, e.g., geology, tectonics, climatology, soil science, ecology, or hydrology. Goudie  defines mountains as “substantial elevations of the Earth's crust above sea level which result in localized disruptions to climate, drainage, soils, plants and animals”. If altitude is considered solely, lower-lying areas with marked relief are not included, although roughness is a major criterion for defining mountains; the treatment of high altitudes, on the other hand, is disputed as well [Price, 1981; Gerrard, 1990; Meybeck et al., 2001].
 For precisely delineating a mountain range, it is most promising to incorporate local features of, e.g., climate, relief, geology, vegetation, or accessibility [Browne et al., 2004; Ives et al., 1997] (see Baumgartner et al.  for the European Alps). For the global definition necessary for our study, however, the use of locally variable criteria is neither objective nor efficient. Therefore the two uniform definitions by Kapos et al.  and Meybeck et al.  were considered. We are using here the definition by Meybeck et al. since it offers a better distinction, particularly in high-elevation areas. It distinguishes a total of 15 relief categories, derived from average cell altitude (eight classes) and an average roughness indicator (seven classes). The definition was established on a 0.5° × 0.5° resolution, while the necessary relief and roughness indicators were derived from the 0.5' × 0.5' resolution digital elevation model GTOPO 30 (United States Geological Survey (USGS), 1996, GTOPO 30 – Global 30 Arc Second Elevation Data Set, data available online at http://edcdaac.usgs.gov/gtopo30/gtopo30.asp).
 We extended the original definition of Meybeck et al.  by including all areas above 1000 m above sea level (asl) plus those areas between 200 and 1000 m asl with a relief roughness of more than 20 ‰ (relief roughness is a function of cell length and minimum and maximum elevation in a cell). This is equivalent to adding hills, midaltitude plains above 1000 m asl and plateaux of medium, high and very high altitude to the four mountain classes (low, middle, high and very high altitude) already identified (see Meybeck et al.  for a detailed global map of relief types). This extension follows our concept of a “water tower” which encompasses an elevated area that potentially delivers disproportional runoff. By including all potentially important areas, we will be able to investigate whether hills and plateaux actually should be considered mountainous from the hydrological point of view, i.e., if they produce disproportional runoff. For certain regions such as China or the river Niger basin, hills may still function as “water towers” although they do not meet the definition of “mountainous” in the original definition of Meybeck et al. . Because of the rather generous extent of our definition, mountains and higher-altitude areas (hills, midaltitude plains and platforms of medium to very high altitude) make up 39% of global continental surface excluding Antarctica and Greenland, as compared to 25% in the definition by Meybeck et al. . This ensemble will be termed “mountains” in the rest of this article, while the remaining relief classes correspond to our definition of “lowlands”.
 To achieve their data set, Fekete et al.  combined observed river discharge information with a climate-driven water balance model. Snowmelt is incorporated in the model as a simple function of temperature and elevation. Runoff from glaciers is mainly locally important, but downstream contribution is mostly low [see, e.g., Viviroli and Weingartner, 2004b]. However, the accuracy of single cells or the seasonal importance of snowmelt delayed runoff is less important here than the runoff over larger areas. The latter is assured by a correction to measured runoff values applied by Fekete et al. : The authors have selected 663 gauging stations from the Global Runoff Data Center (GRDC) database, and annual interstation runoff was then calculated for the regions between the gauging stations. These interstation values were used to adjust the outputs from the UNH Water Balance Model (UNH-WBM) [Vörösmarty et al., 1998] which is forced by long-term mean monthly climate (precipitation and air temperature) and uses land cover and soil information (see Vörösmarty et al.  for details). The result is a spatially and temporally distributed runoff simulation which reproduces at-site discharge measurements. The quality of mountain area runoff is discussed in section 5.3.
2.4. Analysis Basins
 Since the significance of mountains in water resources is assessed in relation to the adjacent lowland areas, a hydrological reference between mountain and lowland areas has to be established. Following the resolution and referencing of the runoff data (see section 2.3), the river basins from the simulated topological network STN–30p [Vörösmarty et al., 2000a, 2000b] were used. STN–30p represents potential flow pathways at 0.5° × 0.5° spatial resolution across the entire nonglaciated surface of the globe. It was derived from provisional flow directions for each land-based grid cell in a digital elevation model [Edwards, 1989] which were then adjusted with help of digital overlays of rivers and independent map sources.
 A major problem was how to handle small catchments since the resolution of a global-scale analysis is not sufficient to represent them accurately. Therefore the basins from the STN–30p were combined with the coastal catchments presented by Meybeck et al. . This typology is based upon a coastal segmentation with respective upstream basin areas and was originally developed to facilitate the budgeting of global riverine transfers from land to oceans; care was also taken to produce climatologically homogeneous coastal zones. For our combination, river basins with a surface area of more than 100,000 km2 (165 basins with a total area of 92.0 M km2) were taken from the STN–30p, while the remaining small catchments (total area 41.6 M km2) were merged using the coastal catchment typology, resulting in another 231 basins. Figure 1 illustrates this procedure for western Europe; for example, the Loire River basin (1a, 118,282 km2) is a large river basin, while the remaining small basins in the Iberian-Biscay coastal segment are aggregated to a single unit (1b, 1,109,150 km2). All basin-relative computations are based upon the resulting “analysis basins” and carried out for the continental surface excluding Antarctica and the glaciated parts of Greenland.
 For 30 basins without lowland cells (mostly upper altitude endorheic areas), the lowest-lying cell according to the topological network STN–30p was changed to lowland manually in order to establish a plausible reference for calculation of mountain-lowland-relative figures. No changes were necessary for basins with only lowland cells since they are not relevant for our analysis.
 In order to understand the potential of mountains as “water towers”, it is essential to examine their hydrological significance in different climate zones. To achieve this, the life zones scheme according to Holdridge  was employed, offering appropriate resolution regarding critical climate transitions (see also discussion in section 5.2). The typology uses temperature, growing season length, mean annual precipitation and a potential evapotranspiration ratio to discern between five and eight humidity provinces for six latitudinal regions. The originally 39 classes were aggregated to 15 classes by Leemans  which we further condensed into seven climate types that suit our hydrological point of view (Table 1).
A: Share in continental surface area (total: 133.6 M km2); POP: Share in global population (total: 6.06 billion people); Q: share in global total discharge (total: 40,606 km3 a−1); q: Average total runoff; Antarctica and glaciated parts of Greenland are excluded.
Polar and cold
tundra and polar; cold parklands
forest tundra; boreal forest
temperate forest; warm temperate forest
cool desert; hot desert
tropical semiarid; tropical dry forest
tropical seasonal forest; tropical rain forest
 For reasons of compatibility with the runoff and mountain typology data sets in use, population data prepared by Vörösmarty et al. [2000c] was employed. They used current country level population statistics, distributing them with the help of spatial extents of urban areas, populated places and with city lights from remote sensing. With a resolution of 0.5° × 0.5° and the reference year 2000, these data are sufficiently accurate and up-to-date for our global analysis. The incorporation of more recent and higher-resolution data (e.g., Oak Ridge National Laboratory (ORNL), 2004, LandScan 2004 Global Population, data available online at http://www.ornl.gov/sci/landscan/) proved difficult because they are not fully congruent with the hydrological data sets in use.
 To map and assess the hydrological significance of mountain areas, two indicators are introduced. They are later applied to all mountain cells and can either be mapped globally or tabulated according to relevant regions (e. g. by climate zone). Departing from the mere relation of highland to lowland runoff (indicator 1, relative water yield), four types of mountain cells are defined by incorporating lowland climate, representing the natural Earth system state. This is followed by adding human water needs (indicator 2, water resources contribution) which represents the demand perspective of man.
3.1. Runoff Contribution From Mountains (Relative Water Yield)
 First, the disproportionality of mountain runoff to lowland runoff is assessed in order to present an overview of the hydrological significance of mountains. This is done through the dimensionless coefficient RWY (relative water yield) which applies to all mountain cells i, each situated in a respective analysis basin j:
where rmi,j is runoff in each mountain cell i which belongs to analysis basin j (mm a−1) and RLj is average lowland runoff in analysis basin j (mm a−1). RLj is thus an aggregate value for the whole lowland area, while rmi,j as well as the resulting RWYi,j are specific for each mountain cell.
3.2. Mountain Typology in Earth System Context
 To emphasize the importance of the lowland-relative view, lowland climate is incorporated explicitly in the analysis. We distinguish mountain areas cell-wise into four classes using the following scheme:
 First, as a climatological distinction, it is determined whether climate conditions in the lowland part of an analysis basin are dry. Lowland dryness is identified here by the predominant aggregated Holdridge life zone, with steppe, arid, and subtropical considered as dry, the remaining classes considered as wet (see Table 1).
 Then, as a hydrological distinction, the contributing potential of a mountain cell to the respective analysis basin's lowland runoff is assessed. This is achieved by comparing mountain cell runoff to average lowland runoff using the following definition: With a dry lowland, a mountain cell's contributing potential is assumed high if its annual runoff exceeds 125% of mean annual lowland runoff; otherwise, its contributing potential is medium or low. With a wet lowland, a mountain cell's contributing potential is assumed low if its annual runoff is less than 75% of mean annual lowland runoff; otherwise, its contributing potential is high or medium.
 This procedure assigns to each individual mountain cell one of the four following classes (cf. Figure 2):
 1. Earth system mountain type I: High contributing potential to a dry lowland area. This matches the original “water towers” concept of mountains being elevated humid islands within a dry region.
 2. Earth system mountain type II: Medium or low contributing potential to a dry lowland area. These mountain cells are not significantly wetter than the lowland area but occasionally or seasonally still may be of essential importance, even with low discharges.
 3. Earth system mountain type III: Medium or high contributing potential to a wet lowland area. Although located in a region where water supply in the lowlands is sufficient on average, these mountain cells have an important supporting function and make important seasonal contributions to lowland runoff.
 4. Earth system mountain type IV: Low contributing potential to a wet lowland area. This represents rain shadow areas or high altitudes.
 An interpretation of these mountain types in terms of hydrological significance will be done on basis of the results in section 5.1.
3.3. Mountains in Water Resources Context (Water Resources Contribution)
 For this first spatially explicit, global assessment, the water resources aspect is introduced by considering potential minimum anthropogenic water demand in lowlands as well as in mountains. While global maps of human water use such as irrigation now become available [e.g., Siebert et al., 2005], total anthropogenic river water depletion can be determined with high resolution only at the regional scale. Therefore a per capita need of 500 m3 a−1 is assumed here, being the threshold value for severe water scarcity according to Falkenmark and Widstrand . It should be noted that this represents an absolute minimum need which we consider here to be permanently withdrawn. Actual needs may vary locally and regionally, depending on a variety of factors such as the extent of irrigation and the state of economic development [see, e.g., Sullivan, 2002]. We introduce an index for water resources contribution WRC which is the ratio of lowland water availability (surplus or deficit) to water supply (only surplus is considered, see below) from mountains, again applicable to all mountain cells:
where RLj is average runoff in lowland area of analysis basin j (mm a−1), NLj is minimum water need of population in lowland area of analysis basin j (mm a−1), rmi,j is runoff from mountain cell i which is situated in analysis basin j (mm a−1), and NMj is minimum water need of population in mountain area of analysis basin j (mm a−1).
WRC identifies the importance of each mountain cell i for water resources supply in the related lowland of analysis basin j. Mountain water need NMj is calculated as a basin average (as opposed to the cell-wise treatment of mountain runoff rmi,) to smooth out high local water demands which can be satisfied with runoff from adjacent cells. The interpretation of WRC values is given in Table 2.
Table 2. Interpretation of Water Resources Contribution Index WRC for Individual Mountain Cells
lowland deficit which cannot be satisfied through the mountain contribution; with this absolute deficit, water can only be supplied from other sources such as fossil groundwater pumping, river diversions or desalination.
−1 – 0
lowland deficit which can be transformed into a surplus thanks to the mountain contribution: the supply situation is still comfortable, but mountains are essentially important for satisfying lowland water need; appropriation of the mountain supply may require water storage in reservoirs.
0 – 1
lowland surplus, with an additional contribution from mountains which is larger than this surplus; mountains are not needed on the average, but may still contribute important seasonal supply or may be important for other uses such as hydroelectricity.
lowland surplus, which is even larger than the mountain contribution; mountains are not significant for lowland water need.
 The following example adopted from the Rhine River catchment illustrates the calculation and interpretation of WRC: Minimum water need for the lowland population of 22 million is 11 km3 a−1 or, relative to an area of 77,200 km2, 142.5 mm a−1 (NL). Average lowland runoff is 239 mm a−1 (RL), resulting in a small, but positive lowland surplus of 96.5 mm a−1 (RL-NL). For the total mountain population of 18.6 million, minimum water need is 9.3 km3 a−1 or, relative to an area of 87,900 km2, 105.9 mm a−1 (NM). Runoff from a selected mountain cell in Central Switzerland is 1350 mm a−1 (rm), the balance for this specific cell thus being a large surplus of 1244.1 mm a−1 (rm–NM). According to equation (2), the cell's WRC is 0.08, indicating that lowland water resources are only just sufficient, while mountain discharge has an important supportive role and shows a considerable surplus which, for instance, is available for hydropower use. Note that the extent of the mountainous area for the Rhine River catchment is rather high here as compared to other studies [e.g., Viviroli and Weingartner, 2004b] because an extensive mountain definition was employed in order to identify all potentially important elevated areas (see section 2.2).
 The mountain water balance term (rmi,j − NMj) may be negative because of one or both of the following reasons: Either, mountain population is very high so its demand exceeds the available supply, or, mountain discharge is very low so that the demand of the population cannot be met. In these cases, WRC is not calculated. Instead, the causing conditions (high population and/or low runoff) are identified. We assume high population for densities above 21 p km−2 (half of the average population density in coastal catchments according to Meybeck et al. ) and low runoff for values below 30 mm a−1 (just under one tenth of average runoff depth in exorheic regions, generally used as threshold for nonpermanent river flow [Vörösmarty and Meybeck, 2004]).
4.1. Runoff Contribution From Mountains (Relative Water Yield)
 The shares of mountains in global area, discharge and runoff are shown in Table 3 for different life zones, with lowland runoff values given for comparison. Compared to their share in total area (29.8%), mountains in the arid zone clearly deliver most disproportional discharge (share: 66.5% of total). This disproportionality is second highest for the temperate zone (share in total area: 43.4%, share in total discharge: 60.8%). Note that calculation of RWY from mountain and lowland runoff is not allowed here because runoff values refer to averages over different life zones; for RWY, a hydrological linkage in the sense of mountains-lowlands is necessary since it is calculated always relative to an analysis basin.
Table 3. Discharge and Runoff From Mountains for Condensed Holdridge  Life Zonesa
qM, mm a−1
qL, mm a−1
QM: mountain share in total discharge; AM: overall surface area; qM: average runoff in mountains; qL: average runoff in lowlands.
Polar and cold
Outside humid tropical
Steppe, arid, and subtropical
 To assess relative water yield correctly, Figure 3 shows the area-weighted distribution of RWY, distinguished by the predominant life zone in the lowlands. The essential importance of mountains for lowland areas with dry climates is highlighted: Although area-relative median values (50% of summed up area) are very low for the steppe, arid, and subtropical life zones, between 18% and 31% of mountain cells have very high RWY values of more than 5. Regionally, these contributions are essential.
Figure 4 maps the cell-by-cell values of relative water yield in mountains (RWY), thus identifying disproportionality of runoff formation in mountains. Details for runoff from mountain ranges with superior water yield are found in Table 4. Extended areas with disproportionally high mountain runoff can be identified especially in semiarid and arid regions (e.g., Ethiopian Highlands, Drakens Mountains, Great Dividing Range), but also in more humid regions (e.g., European Alps). Low relative water yield from mountains, in contrast, is found inside of large mountain massifs such as the Himalayas (e.g., Tarim Basin) and the Rocky Mountains (e.g., parts of Colorado Plateau) as well as in river basins which on the average show strongly humid conditions (e.g., Amazon and Mississippi rivers).
Table 4. Selected Mountain Ranges With Superior Relative Water Yield and Corresponding Major River Basins Which Benefit From the Disproportionally High Runoffa
Major Beneficiary River Basin(s)
q*, mm a−1
The q* is the maximum smoothed runoff (mean in circular-shaped neighborhood with radius of 1°) in the mountainous part of the respective beneficiary river basin(s), being representative for the highest runoff values which can be expected on the mesoscale. The rivers selected here have a basin area of more than 100,000 km2 and an average RWY of more than 2.5.
 The relation to lowland climate conditions is given explicitly in the map of Earth system mountain types (Figure 5, top), where the global mountain area is distinguished cell-wise according to contributing potential and lowland climate (cf. Figure 2). The resulting map particularly reveals the location of mountain areas contributing disproportionally to lowland discharge, with distinction of associated lowland climate being dry (I, blue) or wet (III, green). This distinction can even concern one single mountain range which is orientated toward different lowland climates, as is the Highland of Bihé: north to the wet lowlands of the Zaire (Congo) River basin, south to the drier Zambezi River basin. It is further revealed where disproportionally low discharge from mountain areas occurs, again distinguished by associated lowland climate being dry (II, yellow) or wet (IV, brown). Theoretically, type IV should not be directly adjacent to lowland areas since it represents rain shadow area or high altitudes. Occurrences are found, however, as an actual precipitation increase at the rise of a mountain massif may not be captured by the typology when other parts of the basin are excessively wet (e.g., Mississippi and Amazon River basins). Regional examples for all four mountain types are listed in section 5.1 where the interpretation of this typology for water resources will be discussed.
 As Table 5 reveals, mountain type I clearly shows the highest relative water yield (average RWY = 6.0) with a runoff of 389 mm a−1; with an extent of 12 M km2, it covers 23% of global mountain area. The highest absolute discharge (11,512 km3 a−1) and runoff (734 mm a−1) is found in mountain type III, which is explained by the high precipitation amounts that mountains can extract in humid climates; this type covers 30% of global mountain area. The respective average RWY of 1.5 admits a supportive role and points at a large surplus. The average RWY for type II is relatively low (0.5), but may still have high relevance in terms of seasonal and local contributions (cf. Figure 3); it has a share of 28% in global mountain area.
Table 5. Characteristic Hydrological Values for the Relief Types [Meybeck et al., 2001] Considered as Mountainous for this Article, Distinguished Into Earth System Mountain Types I to IVa,b
Mountain Type I: Dry Lowland Climate, High Contributing Potential
Mountain Type II: Dry Lowland Climate, Low or Medium Contributing Potential
Mountain Type III: Wet Lowland Climate, High or Medium Contributing Potential
Mountain Type IV: Wet Lowland Climate, Low Contributing Potential
 The changing importance of the Earth system mountain types is examined in Figure 6, which shows their contribution to the lowlands for different predominant lowland life zones:
 1. With all climate zones in focus, type III clearly makes highest contributions, providing 63% of mountain discharge with a share in mountain area of only 30% (RWY = 2.6). Type I is also of importance for runoff formation (discharge: 26%, area: 23%), although less markedly disproportional (RWY = 1.4), while types II and IV participate to a lesser degree in runoff formation.
 2. If the humid tropics are disregarded, RWY values are clearly higher. While type I gains importance (discharge: 39%, area: 27%, RWY = 2.8), type III has somewhat lower shares, although with an increased RWY (discharge: 51%, area: 25%, RWY = 3.9).
 3. For the temperate zone, shares in discharge are much higher for type III (90%) than for type IV (10%), although their shares in area differ far less (type III: 60%, RWY = 2.6; type IV: 40%, RWY = 0.5).
 4. In steppe, arid, and subtropical zones, types I and II show a similar pattern as types III and IV in the temperate zone, however with a markedly higher RWY values for type I (5.2). Especially in view of strong seasonal variability of lowland runoff, these contributions are essential.
 Note that the predominance of types I and III complies with our definition of Earth system mountain types, as well as the absence of types I and II in the temperate zone and of types III and IV in the dry zones.
4.3. Mountains in Water Resources Context (Water Resources Contribution)
 Mapping the water resources contribution index WRC (Figure 5, bottom) finally illustrates the importance of mountains for human water needs. Critically important mountain regions are found in the Middle East, South Africa, parts of the Rocky Mountains and the Andes. Particularly marked is the importance of mountain water resources in the western and eastern Himalayas, which partly cannot compensate the large lowland deficits any more (WRC < −1). Striking is the frequent occurrence of overuse of mountain water resources which is found primarily in the dry mountain areas already mentioned.
 With the help of Figure 7 (left), it is revealed that over one third (37%) of global mountain area has a supportive role for lowland water resources (WRC between 0 and 1), which may also include important seasonal contributions, while about 7% are of essential significance (WRC < 0). Following our classification, we expect mountain type I to show highest significance for lowland water supply; this is corroborated through Figure 7 (right), where the distribution of WRC noticeably shifts toward lower values, indicating a greater importance for the lowlands. Around the critical value of WRC = 0, a stronger concentration is found, with 27% of type I mountains being very important (WRC < 0) and 48% providing additional resources (WRC between 0 and 1). This indicates a vital importance of type I mountains for water resources supply, particularly in the densely populated subtropical regions. Note that the assumed per capita need of 500 m3 a−1 represents a minimum value and that with assumption of higher values, the resulting WRC index would become smaller (i. e. even more in favor of mountain water supply).
4.4. Importance of Different Relief Types
 With the rather extensive definition chosen for the delineation of mountains, it is possible to identify those relief types which actually function as a “water tower”. Low and midaltitude mountains produce the highest discharge and make up the largest area in all four Earth system mountain types (Table 5). Highest runoff values are found for hills (770 mm a−1 in mountain type III and 550 mm a−1 in mountain type I) which is explained by the marked precipitation increase at the foot of mountain ranges, especially in humid areas. Values are only little lower for low- and midaltitude mountains (756 mm a−1 in type III, 408 mm a−1 in type I) and for high and very high altitude mountains (611 mm a−1 in type III, 409 mm a−1 in type I). Elevated Plateaux, on the other hand, produce significantly less runoff. However, it should be kept in mind that these values refer to an annual average. Seasonal and regional importance is therefore always subject to specific investigations.
5.1. Mountain Types in Earth System Context
 Considering above results, the typology of mountains in Earth system context (cf. sections 3.2 and 4.2) is rephrased, now with focus on the “water tower function” instead of the mere contributing potential to lowland discharge. Table 6 summarizes our conclusions with regards to water resources and hydropower potential. This results in a revised type description:
Table 6. Relative Importance of Mountains as Water Towers With Regards to Different Functions, Distinguished by Earth System Mountain Typea
Compare with section 3.2; negative sign: very low; 0: low; plus sign: medium; double plus sign: high.
Lowland water resources
0 to +
− to +
0 to +
+ to ++
0 to +
 1. Earth system mountain type I: Essential water tower. Most of the discharge in the respective lowland basins originates from the mountain cells, which are therefore of superior importance. Examples: Ethiopian Highlands (Nile River), Taurus and Zagros Mountains (Euphrates and Tigris rivers), Pamir and Tien Shan (Amu-Darya and Syr-Darya rivers), southwest Himalayas (Indus River), Highlands of Lesotho (Orange River), and parts of the southern Rocky Mountains (Colorado River and Rio Grande).
 2. Earth system mountain type II: Occasional water tower. Mountain cells may be essential for downstream discharge on the regional scale, either seasonally or in single years. Examples: Hoggar and Tibesti Mountains, central and northwest Himalayas.
 3. Earth system mountain type III: Supportive water tower. Mountain cells contribute considerably to downstream discharge and bear an essential hydropower potential. Examples: European Alps (Danube, Rhine, Rhone, and Po rivers), northern Rocky Mountains (Mackenzie), eastern Himalayas (Chang Jiang/Yangtse River).
 4. Earth system mountain type IV: Limited water tower. Mountain cells are dryer than the lowland average and therefore of only limited or local value; potential for hydropower generation is available. Examples: northeast Andes (Amazon River), eastern Rocky Mountains (Mississippi River).
 The importance for geofluxes (fluxes of water and associated solid, gaseous, and dissolved material as well as thermal and mechanical energy) is summarized as well in Table 6. Geofluxes are essentially derived from wet mountain areas [Milliman and Syvitski, 1992], and if the potential for river transport is also high in the associated lowlands (such as in Earth system mountain type III downstream areas), geofluxes are generally high in the respective catchments. In Earth system mountain type I downstream areas, contrarily, transport may be limited because of lowland storage (except for large allochthonous rivers such as the Nile river capable of crossing large dry areas).
 We are aware that the choice of thresholds and definitions in our methods influences the results to a certain degree. Assessing climate parameters such as precipitation-evaporation relationships, used for example for the Holdridge  climate definition, is inherently difficult in mountain areas. While all methods have their particular benefits and shortcomings, the choice for the Holdridge life zone scheme was made because it produces the most plausible overall picture for the combined distributions of climate zones in general and of lowland dryness in particular.
 Capturing lowland dryness with a globally uniform definition is difficult, especially in lowland regions which are climatologically very heterogeneous. One of the most difficult cases is the Niger River. It originates from a hilly area (the Fouta Djalon) in a tropical monsoon climate, flows northeasterly to reach a hot desert area and then returns southeasterly to a tropical monsoon climate [cf. Viviroli and Weingartner, 2004a]. Additionally, it is joined in its lowest course by the Benue River which is located in a humid climate and therefore further masks the marked dryness which occurs in the middle course of the Niger River. In our analysis basin typology, this particular problem was eased by separating the Benue River from the Niger River and treating it as an individual basin.
 The typology established here holds valid for today's climate only. This is particularly relevant for the mountain areas that have a low contributing potential to a dry lowland area. The Hoggar and Tibesti Mountains, for instance, were important “water towers” for the Sahara Area in the humid Holocene, as can be deduced from Elephant bones and rock paintings discovered [cf. Mauny, 1956]. In a more humid climate, these mountain areas might be reactivated.
5.3. Global Runoff Model
 An important source of uncertainty is the underlying runoff model. In order to assess the accuracy and plausibility of runoff values, the continental water balances of the employed UNH model [Fekete et al., 2002] were compared with outputs from the WaterGAP [Döll et al., 2003] model, and the world water balance according to Baumgartner and Reichel  (Table 7). The comparison generally shows a high degree of correspondence, with more notable differences for Africa and Australia. Altogether, uncertainties of modeled runoff are significantly larger in areas where the model could not be adjusted properly because of missing long-term gauge values [see Fekete et al., 2002].
Table 7. Comparison of Mean Annual Runoff for Six Continents From Various Sourcesa
 In mountain areas, however, higher uncertainties must be expected because of the high heterogeneity of hydroclimatological patterns and the sparser gauging network. The comparison of runoff from mountain areas for the UNH-GRDC and WaterGAP models in Table 8 shows that, with the exception of South Africa, the UNH-GRDC model produces higher runoff values from mountain areas, most clearly for Asia and Africa. On a smaller scale, comparisons for the densely gauged region of the European Alps (model: 822 mm a−1 [Fekete et al., 2002], reference: 910 mm a−1 [Baumgartner et al., 1983]) and Switzerland (model: 1061 mm a−1 [Fekete et al., 2002], reference: 991 mm a−1 [Schädler and Weingartner, 2002]) suggest that at least in areas with extensive measurements, water balance in mountains is represented with reasonable accuracy. While the underestimation for the European Alps was expected because of the smoothing effects in models, the overestimation of strongly mountain influenced Switzerland is rather surprising.
Table 8. Comparison of Modeled Mean Annual Runoff From Mountain Areas For Six Continentsa,b
 Although global runoff models now under development will be improved using full calibration instead of adjustment methods, mountain runoff will presumably remain hard to assess because of the inherent difficulties concerning representativity and accuracy of the gauging networks.
 As consequence of the current uncertainties, the maps presented should not be interpreted on basis of single pixels, but rather in the regional and global patterns they produce. We assume larger uncertainties for low runoff values and the figures derived from them, respectively, and suggest cautious interpretation for areas with a mean annual runoff of less than 30 mm (shown in hatch marks in the relevant world maps). This applies especially to the “occasional water towers” (type II) as defined above where precipitation and runoff are low, but highly variable from year to year. Consequently, the validity of mean annual values is limited, while interannual variability causes a significant vulnerability for crises in water resources [cf. Viviroli et al., 2003].
6. Summary and Outlook
 An exhaustive assessment of mountain areas with regards to water resources supply for the adjacent lowlands was presented, providing the basis for identification of critically important regions on a global scale. The significance of mountain areas was assessed first in a hydrological and ecological context, ranging between essential (23% of global mountain area), supportive (30%), occasional (28%) and limited (19%) importance. When lowland water users are considered, 7% of global mountain area has an essential role in water resources, while another 37% provide important supportive supply. The distinction of climate zones in typology and analysis was indispensable since significant findings may be masked in global average values. With the resulting maps, the disproportionality of mountain runoff is identified on a cell-by-cell basis, and statements for individual basins, climate zones and relief types are summarized from this data. The methods that have been established could serve as a starting point for more detailed investigations, for instance with increased spatial (regional) and temporal (seasonal) resolution.
 Studies rooted in regional analysis (e.g., Barnett et al.  on possible changes in snowmelt-dominated regions) are able to provide higher spatial accuracy, and specific regional analyses (such as Bales et al.  for the western United States) may point more clearly to research issues such as the combination of remote sensing and ground-based data. Similar analyses will only be feasible at global scale with highly improved global hydrological modeling efforts, but they are inherently hampered by the lack of accurate monitoring data in many regions of the world.
 Within the next decades, effects of climate change and population growth are expected to worsen water resources supply significantly, particularly through altered discharge patterns from mountains [see Messerli et al., 2004] and increasing demand for food production. This will affect most heavily regions prone to scarcity already today, such as arid regions or also the monsoon belts, especially when seasonal deficits occur which nowadays are attenuated by mountain supply. Since at least 40% of the world's population lives in a river basin shared by two or more countries [Wolf, 2002], a considerable conflict potential exists about sharing water between upstream and downstream neighbors. To date, more than twice as much positive interactions are reported than negative ones [Wolf et al., 2003], but with increasing pressure, basin-wide agreements are likely to meet harder challenges. The expansion of human water resources appropriation (today 56% of global accessible discharge [Postel et al., 1996]) will foster the construction of dams and river transfers, which will also have serious ecological, economical and social costs [Gleick, 2003]. The most severe problems are to be expected in developing regions of the tropics and subtropics, where water resources management is hindered by the lack of reliable and long-term data sets, missing technical, scientific and financial resources as well as political instability. At the same time, dependency on mountain water resources is increased by demographic pressure as well as by strong urbanization and industrialization processes.
 An integrated approach is necessary for sustainable water management, with an adequate monitoring network and the best possible assessment of the resource as basis for decision making [WWAP, 2006; Oki and Kanae, 2006]. This applies especially to water resources originating from mountain areas which are essential for mitigation of seasonal and regional deficits, and there particularly to the developing world. In these regions, the data available today are not yet sufficient for a long-term successful water resource management which is necessary for the benefit of the rapidly growing population and the urgently needed increase of food production. Furthermore, climate and hydrology modelers should focus on a better representation of the even more uncertain mountain areas, because they represent highly relevant areas for today's and tomorrow's water resources.
 The cooperation between the participating research groups was supported by the Association of Scientific Staff, University of Berne (MVUB), through a project grant for Daniel Viviroli. Hans Dürr has been funded by a Ph.D. grant from the French Ministère de la recherche, by the EU programme Si WEBS (contract number HPRN-CT-2002-000218), and by Utrecht University (high potential project G-NUX). We thank Thomas Ninck for his help with several GIS analysis steps for an earlier version of this article and Kerstin Verzano for providing WaterGAP [Döll et al., 2003] model data. Charles Vörösmarty is acknowledged for communication of various data sets. Roger C. Bales, Jeff Dozier, and an anonymous reviewer provided valuable thoughts that helped to improve the original manuscript.