From: "Saved by Windows Internet Explorer 7" Subject: Nonsustainable groundwater sustaining irrigation: A global assessment Date: Wed, 28 Mar 2012 18:19:46 +0200 MIME-Version: 1.0 Content-Type: multipart/related; type="text/html"; boundary="----=_NextPart_000_0000_01CD0D0F.5ADC4CF0" X-MimeOLE: Produced By Microsoft MimeOLE V6.0.6002.18463 This is a multi-part message in MIME format. ------=_NextPart_000_0000_01CD0D0F.5ADC4CF0 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Location: http://www.agu.org/journals/wr/wr1201/2011WR010562/ =EF=BB=BF
AGU: Water Resources = Research
WATER RESOURCES RESEARCH, VOL. 48, W00L06, 18 PP.,=20
doi:10.1029/2011WR010562
Nonsustainable groundwater sustaining irrigation: A global = assessment
Department of Physical Geography, Utrecht =
University,=20
Utrecht, Netherlands
Department of Physical Geography, Utrecht =
University,=20
Utrecht, Netherlands
Soil and Groundwater Systems Unit,
Deltares, =
Utrecht,=20
Netherlands
[1] Water used by=20 irrigated crops is obtained from three sources: local precipitation = contributing=20 to soil moisture available for root water uptake (i.e., green water), = irrigation=20 water taken from rivers, lakes, reservoirs, wetlands, and renewable = groundwater=20 (i.e., blue water), and irrigation water abstracted from nonrenewable=20 groundwater and nonlocal water resources. Here we quantify globally the = amount=20 of nonrenewable or nonsustainable groundwater abstraction to sustain = current=20 irrigation practice. We use the global hydrological model PCR-GLOBWB to = simulate=20 gross crop water demand for irrigated crops and available blue and green = water=20 to meet this demand. We downscale country statistics of groundwater = abstraction=20 by considering the part of net total water demand that cannot be met by = surface=20 freshwater. We subsequently confront these with simulated groundwater = recharge,=20 including return flow from irrigation to estimate nonrenewable = groundwater=20 abstraction. Results show that nonrenewable groundwater abstraction = contributes=20 approximately 20% to the global gross irrigation water demand for the = year 2000.=20 The contribution of nonrenewable groundwater abstraction to irrigation = is=20 largest in India (68 km3 yr=E2=88=921) followed by = Pakistan (35=20 km3 yr=E2=88=921), the United States (30 = km3=20 yr=E2=88=921), Iran (20 km3 = yr=E2=88=921), China (20=20 km3 yr=E2=88=921), Mexico (10 km3 = yr=E2=88=921), and=20 Saudi Arabia (10 km3 yr=E2=88=921). Results also = show that=20 globally, this contribution more than tripled from 75 to 234 = km3=20 yr=E2=88=921 over the period 1960=E2=80=932000.
Received 17 February 2011; = revised 21 November 2011; accepted 24 = November=20 2011; published 25 January 2012.
Keywords: blue water, green = water,=20 irrigated crops, irrigation water demand, non-local water resources,=20 non-renewable groundwater abstraction.
Index Terms: 1818 = Hydrology:=20 Evapotranspiration; 1834 Hydrology: Human impacts (4323); 1842 = Hydrology:=20 Irrigation; 1884 Hydrology: Water supply.
[2] Irrigated crops = play a vital=20 role in the securing global food production. It is estimated that 17% of = agricultural lands are irrigated, yet they account for 40% of the global = food=20 production, sustaining the livelihood of billions of people [Abdullah,=20 2006]. At the same time, water used by irrigated crops (i.e., crop = water=20 demand) and irrigation water demand (including evaporative and = percolation=20 losses during transport and application) are responsible for about 70% = of the=20 global water withdrawal [Shiklomanov, 2000] and account for about 90% of = the global=20 water consumption, i.e., water withdrawal minus return flow [D=C3=B6ll = et al.,=20 2009; Siebert=20 et al., 2010].
[3] Water demand for = irrigated=20 crops can be met by three different sources: (1) green water, being = water from=20 local precipitation that is temporarily stored in the soil, (2) blue = water,=20 being surface freshwater available in rivers, lakes, reservoirs and = wetlands,=20 and renewable groundwater, and (3) nonrenewable groundwater and nonlocal = water=20 resources [V=C3=B6r=C3=B6smarty et al., 2005]. The latter = comprises water=20 transported by cross-basin diversions, water from desalinization plants, = and=20 nonrenewable groundwater abstracted from aquifers. We explicitly use the = term=20 nonrenewable groundwater abstraction, because it consists of additional = water=20 gained by groundwater abstraction in surplus of groundwater recharge.=20 Groundwater can serve as a temporary source of irrigation water if = during the=20 dry season or during dry years surface water is insufficient to satisfy = demand.=20 Also, groundwater may be the main source of irrigation water in areas = overlying=20 productive aquifers wherever access to surface water is limited. = Importantly, as=20 long as abstraction is smaller than recharge, it will only reduce the=20 groundwater discharge to surface water (base flow) and as such can be = counted as=20 available blue water. However, if groundwater abstraction exceeds the=20 groundwater recharge over extensive areas for prolonged periods, = persistent=20 groundwater depletion occurs [Gleeson = et=20 al., 2010] where groundwater reserves still exist, leading to = falling=20 groundwater levels [Konikow and Kendy, 2005; Karami = and=20 Hayati, 2005; Reilly et al., 2008; Rodell = et=20 al., 2009; Tiwari et al., 2009; McGuire,=20 2009; Scanlon=20 et al., 2010; Famiglietti et al., 2011]. In that case fossil=20 groundwater, not being an active part of the current hydrological cycle, = is used=20 as an additional, albeit nonrenewable, source of irrigation water.
[4] Previous studies by = V=C3=B6r=C3=B6smarty et=20 al. [2005], Rost et al. [2008], Wisser = et al.=20 [2010], and Hanasaki et al. [2010] implicitly quantified the = amount of=20 nonrenewable and nonlocal water resources on the basis of the amount of = water=20 demand exceeding locally accessible supplies of blue water. However,=20 uncertainties of these estimates inherently remain large = (389=E2=80=931199=20 km3 yr=E2=88=921) since they are sensitive to both = estimated water=20 demand (1206=E2=80=933557 km3 yr=E2=88=921) and = simulated surface=20 freshwater availability (i.e., water in rivers, lakes, and reservoirs;=20 36,921=E2=80=9341,820 km3 yr=E2=88=921).
[5] At the same time, = estimated=20 groundwater abstraction ranges globally between 600 and 1100 = km3=20 yr=E2=88=921 [Shah et al., 2000; Zektser = and=20 Everett, 2004; D=C3=B6ll, 2009; Wada et = al.,=20 2010]. Recently, using a forward modeling approach, Siebert = et=20 al. [2010] quantified the amount of groundwater consumed = through=20 current irrigation practice to be 545 km3 = yr=E2=88=921. Yet, up to=20 now, all global studies dealing with sources of irrigation water demand = have not=20 explicitly identified which part of the irrigation water demand is = currently met=20 from nonrenewable groundwater abstraction. = Regional=20 studies by Rodell=20 et al. [2009], Tiwari et al. [2009], and Famiglietti et=20 al. [2011] using the Gravity Recovery and Climate Experiment = (GRACE)=20 [Tapley et=20 al., 2004] satellite observation revealed that considerable = amounts=20 of nonrenewable groundwater resources are being abstracted in northeast = India=20 and northwest Pakistan and California's Central Valley in the United = States,=20 most of which is used for irrigation. McGuire=20 [2009] and Scanlon et al. [2010] also reported depleting = groundwater=20 resources because of irrigation in the High Plains (Ogallala) aquifer, = United=20 States. It can thus be expected that large amounts of nonrenewable = groundwater=20 are indeed abstracted for irrigation purposes, particularly during the = growing=20 season when irrigation water demand exceeds available surface freshwater = resources. Assessing globally this contribution is important because it=20 pinpoints areas where irrigation and thus food production is sustained = by a=20 nonsustainable water resource.
[6] In this paper, = expanding on=20 existing studies, we explicitly quantify the amount of nonrenewable or=20 nonsustainable groundwater abstraction used for current irrigation = practice at=20 the global scale. In order to make use of the best available data and to = make=20 our assessment as relevant for the present-day situation as possible, we = opted=20 for the year 2000. The trend of past groundwater abstraction is = reconstructed=20 over the period 1960=E2=80=932000 to highlight the increasing importance = of nonrenewable=20 groundwater abstraction in irrigation practice over the recent past when = irrigated areas expanded rapidly. We provide a global total estimate = only for=20 the past period because of larger uncertainties caused by assumptions to = overcome a lack of historical data set.
[7] This study follows = an=20 improved method to compute nonrenewable groundwater abstraction compared = to that=20 of Wada et=20 al. [2010, 2011a].=20 Compared to these previous works we develop a new approach when = downscaling=20 country-based data on groundwater abstraction to grid-based estimates, = while we=20 additionally account for additional recharge that occurs from = irrigation. We=20 then focus on irrigated areas since the contribution of nonrenewable = groundwater=20 abstraction to irrigation has not been estimated by any of the previous = work.=20 Also, we add a reconstruction of the increased irrigation water demand = from=20 1960=E2=80=932000, as well as the contribution of other water resources = available to=20 meet crop water demand such as soil moisture (green water) and surface=20 freshwater (blue water).
[8] Green water and = blue water=20 availability are obtained from simulation with the global hydrological = model=20 PCR-Global Water Balance (PCR-GLOBWB) [Van Beek = et=20 al., 2011] at a spatial resolution of 0.5=C2=B0 =C3=97 = 0.5=C2=B0 (i.e., 50 km =C3=97 50=20 km at the equator). Nonrenewable groundwater abstraction for irrigation = is=20 calculated by taking abstraction in excess of recharge, while = considering the=20 fraction of irrigation water demand over net total water demand. We = subsequently=20 compare the amount of nonrenewable groundwater abstraction used for = irrigated=20 crops to the amount of green water and blue water used for irrigation. = It can be=20 expected that the total amount of available green water and blue water = and=20 nonrenewable groundwater is insufficient to satisfy crop water demand = since we=20 simulate optimal crop growth. This shortage can be partly supplied from = nonlocal=20 water resources [cf. Rost et al., 2008; Hanasaki = et=20 al., 2010] such as desalination and diverting water ways = (i.e.,=20 aqueducts). Again, assuming that optimal crop growth is aimed for, we = equate=20 this shortage with the additional supply from =E2=80=9Cnonlocal water = resources=E2=80=9D to be=20 consistent with the previous studies. By explicitly defining the = contribution of=20 nonrenewable groundwater we can better confine previous assessments of=20 =E2=80=9Cnonrenewable and nonlocal water resources=E2=80=9D estimated by = V=C3=B6r=C3=B6smarty et=20 al. [2005], Rost et al. [2008], Wisser = et al.=20 [2010], and Hanasaki et al. [2010] and identify regions = where=20 irrigation is dependent on nonrenewable groundwater rather than surface=20 freshwater or blue water.
[9] We start section=20 2 by defining the various terms used to calculate nonrenewable = groundwater=20 abstraction, water demand and water availability. Next, we describe the=20 calculation of water availability, irrigation water demand and the = estimation of=20 nonrenewable groundwater abstraction.
[10] In this study, the = term=20 =E2=80=9Cnonrenewable groundwater=E2=80=9D denotes groundwater gained by = abstraction in excess=20 of recharge for any region of the world, while the term = =E2=80=9Cgroundwater depletion=E2=80=9D=20 is used to denote persistent removal of groundwater from storage, which = we=20 estimate as overdraft restricted to subhumid to arid areas as done by Wada et = al.=20 [2010].
[11] Throughout the = paper we will=20 consistently use the term water =E2=80=9Cdemand=E2=80=9D to denote the = need or requirement for=20 water. The term demand is used to indicate that we can only estimate = potential=20 use, i.e., the water that would be used by a given activity if = sufficient water=20 were available. Water demand generally comes from three sectors: = domestic water=20 demand, industrial water demand and agricultural water demand. The = latter can be=20 further subdivided into irrigation water demand (by far the largest = part) and=20 livestock water demand. Environmental flow requirements are not = accounted for in=20 this study. In many analyses [e.g., D=C3=B6ll = and=20 Siebert, 2002; Wisser et al., 2008; Wada et = al.,=20 2011a] one distinguishes gross demand (including losses and return = flows)=20 from net demand (without losses and return flows). The latter is = sometimes=20 equated with consumptive water use [e.g., D=C3=B6ll = and=20 Siebert, 2002].
[12] Limiting ourselves = to=20 irrigation water demand we will use the following definitions.
[13] 1. Net crop water = demand is=20 the amount of water that is required to ensure maximum crop growth. Net = crop=20 water demand, (m d=E2=88=921), is taken equal to crop-specific = potential=20 evapotranspiration, which in turn can be related to reference=20 evapotranspiration:
where ETc is the crop-specific=20 evapotranspiration and ET0 is the reference = evapotranspiration (m=20 d=E2=88=921). kc is a = dimensionless crop=20 factor. Tc is the crop-specific = potential=20 transpiration, and ESc is the = potential bare=20 soil evaporation over the irrigated areas (m d=E2=88=921). We = chose to count=20 soil evaporation as part of the net crop water demand as it plays a = relatively=20 large role in the early stages of crop development and equally uses soil = moisture as does transpiration.
[14] 2. Net irrigation = water=20 demand is the amount of water, without counting transport and = application=20 losses, that needs to be supplied by irrigation to ensure maximum crop = growth.=20 If local precipitation is not sufficient to satisfy crop water demand, = actual=20 evapotranspiration falls below the potential rate. Thus, net irrigation = water=20 demand, (m d=E2=88=921), is the amount of water that = needs to be=20 additionally supplied by irrigation to ensure maximum = evapotranspiration:
where Ta is the crop-specific = actual=20 transpiration and ESa is the actual = bare soil=20 evaporation (m d=E2=88=921) that would occur over the = irrigated areas in case=20 no irrigation were present.
[15] 3. Green water is = the actual=20 evapotranspiration Ta + = ESa that would have occurred without irrigation. = This is=20 plant-available water stored in the soil that originates from local=20 precipitation. We will use GreenW to denote green water.
[16] 4. Gross = irrigation water=20 demand is the amount of water, including transport and application = losses, that=20 needs to be supplied by irrigation to ensure maximum crop growth. Such = losses=20 include evaporative and percolation losses during transport from source = to field=20 and during application. Note that application-related evaporative losses = are=20 those that occur before irrigation water is able to infiltrate into the = soil,=20 i.e., interception evaporation and open water evaporation when flooding = occurs,=20 while percolation losses during irrigation are often induced to avoid = topsoil=20 salinity. Here gross irrigation water demand is calculated by = multiplying the=20 net irrigation water demand with a dimensionless country-specific = efficiency=20 factor, eIrr:
The efficiency factor increases the amount of irrigation water by = about 10%=20 to 40%. The country-specific efficiency factors were taken from Rohwer = et al.=20 [2007].
[17] 5. Gross crop = water demand=20 is the amount of water that is required to ensure maximum crop growth in = irrigated areas, including applied irrigation water and losses through = transport=20 and application. We will denote this quantity by .
[18] Gross irrigation = water=20 demand is then satisfied from available blue water, BlueW,=20 nonrenewable groundwater, NRGroundW, and potential nonlocal = water=20 resources, NonLW:
while gross crop water demand can be supplied from four possible = sources:
2.2. Simulating Available Blue=20 Water
[19] To estimate the = amount of=20 blue water available to satisfy gross irrigation water demand, we use = the global=20 hydrological model PCR-GLOBWB to simulate surface freshwater or water in = rivers,=20 lakes, reservoirs and wetlands [Van Beek = and=20 Bierkens, 2009]. We refer to Van Beek = et=20 al. [2011] for an extensive description. PCR-GLOBWB is a = conceptual,=20 process-based water balance model of the terrestrial part of the = hydrological=20 cycle except Antarctica. It simulates for each grid cell (0.5=C2=B0 = =C3=97 0.5=C2=B0 globally)=20 and for each time step (daily) the water storage in two vertically = stacked soil=20 layers and an underlying groundwater layer, as well as the water = exchange=20 between the layers and between the top layer and the atmosphere = (rainfall,=20 evaporation and snow melt). The model also calculates canopy = interception and=20 snow storage. Subgrid variability is taken into account by considering=20 separately tall and short vegetation, open water (i.e., lakes, = reservoirs,=20 floodplains and wetlands), different soil types (Food and Agricultural=20 Organization (FAO) Digital Soil Map of the World; h= ttp://www.fao.org/nr/land/soils/digital-soil-map-of-the-world),=20 and the area fraction of saturated soil (improved ARNO scheme from Hagemann = and=20 Gates [2003]) as well as the frequency distribution of = groundwater=20 depth based on the surface elevations of the 1 =C3=97 1 km Hydro1k data = set. Fluxes=20 between the lower soil reservoir and the groundwater reservoir are = mostly=20 downward, except for areas with shallow groundwater tables, where fluxes = from=20 the groundwater reservoir to the soil reservoirs are possible (i.e., = capillary=20 rise) during periods of low soil moisture content. The total specific = runoff of=20 a cell consists of saturation excess surface runoff, melt water that = does not=20 infiltrate, runoff from the second soil reservoir (interflow) and = groundwater=20 runoff (base flow) from the lowest reservoir.
[20] Simulated specific = runoff=20 from the two soil layers (i.e., direct runoff and interflow) and the = underlying=20 groundwater layer (i.e., base flow) is routed along the drainage network = on the=20 basis of DDM30 [D=C3=B6ll and Lehner, 2002] by using the = kinematic wave=20 approximation of the Saint-Venant equation [Chow et = al.,=20 1988]. The effect of open water evaporation, storage changes by = lakes,=20 attenuation by floodplains and wetlands, and reservoir operations (i.e., = water=20 supply, flood control, hydropower and navigation) are taken into account = as well=20 [Van = Beek et=20 al., 2011].
[21] In this study, = PCR-GLOBWB=20 was forced with daily fields of precipitation, reference = evapotranspiration and=20 temperature over the period 1958 to 2001. Precipitation and air = temperature were=20 prescribed by the Climate Research Unit (CRU) TS 2.1 monthly data set = [Mitchell = and=20 Jones, 2005; New et al., 2000] which was subsequently = downscaled to the=20 daily fields [Van=20 Beek et al., 2011] by using the ERA40 reanalysis data [Uppala = et=20 al., 2005]. Although the Climate Research Unit (CRU) TS 2.1=20 underestimates precipitation because of snow undercatch [Fiedler = and=20 D=C3=B6ll, 2007] over the Arctic regions, this weakness is of = little=20 consequence for this study as no major irrigated areas are located = there.=20 Prescribed reference evapotranspiration was calculated on the basis of = the=20 Penman-Monteith equation according to the FAO guidelines [Allen et = al.,=20 1998] by using time series data of CRU TS 2.1 with additional inputs = of=20 radiation and wind speed from the CRU CLIM 1.0 climatology data set [New et = al.,=20 2002]. Crop-specific potential evapotranspiration was obtained from = a crop=20 factor climatology per grid cell derived from combining land cover data = (GLCC=20 version 2; U.S. Geological Survey's (USGS) Earth Resources Observation = and=20 Science Center, Global land cover characteristics data base version 2.0, = http://edc2.usgs.gov/glcc/glc= c.php),=20 data on leaf area index (LAI) values at dormancy and peak of the growing = season=20 [Hagemann et=20 al., 1999] and the monthly CRU climatology of temperature,=20 precipitation, and potential evapotranspiration over 1961=E2=80=931990 = (see Van Beek = et=20 al. [2011] for details).
[22] PCR-GLOBWB is run = with a=20 daily time step but blue water availability is subsequently evaluated at = monthly=20 time steps. Local blue water availability at a given time and for a = given cell=20 i is finally obtained by taking the cumulative = discharge=20 along the river network (including lakes, wetlands and reservoirs that = are part=20 of the drainage network) after subtracting upstream net total water = demand, , which is the sum of blue water demand of all sectors = (domestic,=20 industrial and agricultural sector) after subtracting return flows:
where WA is blue water availability, QLoc,=20 is the specific discharge or local runoff in cell i,=20 is the net total water demand of upstream cell j (all in m3 d=E2=88=921), = taken to be the local=20 water consumption [D=C3=B6ll and Siebert, 2002], and j =3D i=20 + 1=E2=80=A6n, all upstream the cells = draining to cell=20 i.
2.3. Estimating Green Water and = Water Demand=20 per Sector
[23] Although we focus = on=20 irrigation water demand, we need water demand for all sectors (domestic, = industrial and agricultural including livestock sector), in order to = calculate=20 blue water availability (equation=20 (6)) as well as to estimate groundwater abstraction at the scale of = grid=20 cells (see section=20 2.4). We will describe in brief how the various terms are = calculated. For an=20 extensive description of methods we refer to Wada et = al.=20 [2011a, 2011b].
[24] Agricultural water = demand=20 can be subdivided into livestock and irrigation water demand. As in = previous=20 studies [e.g., D=C3=B6ll and Siebert, 2002; Rost et = al.,=20 2008; Wisser=20 et al., 2008; Hanasaki et al., 2010], we combine gridded = irrigated areas=20 and crop-related data to estimate irrigation water demand. First, we = took the=20 map of irrigated areas based on the MIRCA2000 data set [Portmann = et=20 al., 2010] and combined it with crop factors and growing = season=20 lengths from GCWM [Siebert and D=C3=B6ll, 2010]. Both data sets are = representative=20 for the year 2000. Using these as input to PCR-GLOBWB and forcing the = model with=20 precipitation and reference evapotranspiration data as described in section=20 2.2, this yielded daily time series of actual evapotranspiration, = which can=20 be seen as the evapotranspiration of the crops in the irrigated areas in = case no=20 irrigation was applied. This was used as an estimate of green water,=20 GreenW. Subtracting this amount from calculated time series = of=20 crop-specific potential evapotranspiration for the irrigated areas (net = crop=20 water demand, , equation=20 (1)) then resulted in time series of net irrigation water demand, = , (equation=20 (2)). Multiplication with country-specific efficiency factors (equation=20 (3)) from Rohwer et al. [2007] finally resulted in daily = time series=20 of gross irrigation water demand, . Daily time series were aggregated to monthly values before = further=20 analysis. As stated above, the data obtained are representative for the = year=20 2000. To obtain monthly time series of GreenW, and for the period of interest 1960=E2=80=932000 we repeated = this procedure for=20 each year, while estimating the growth of irrigated areas by combining=20 country-specific statistics on irrigated areas (FAOSTAT; http://faostat.org/) with the MIRCA = 2000 data set=20 [Portmann et=20 al., 2010] (see Wisser et al. [2010] and Wada et = al.=20 [2011b] for details).
[25] Livestock water = demand from=20 1960=E2=80=932000 was reconstructed on the basis of statistics of = livestock densities=20 [Wint = and=20 Robinson, 2007; FAOSTAT, http://faostat.org/], while industrial = and=20 domestic water demand over the same period could be estimated using = statistics=20 on population and socioeconomic drivers (e.g., GDP and electricity = production)=20 taken from the FAOSTAT, the UNEP (http://www.unep.org/) and the World = Bank (http://www.worldbank.org/). To = calculate=20 return flows needed to estimate net water demand (i.e., potential = consumptive=20 water use), we used country-specific recycling ratios calculated by Wada et = al.=20 [2011a] on the basis of economic development stages. We refer to Wada et = al.=20 [2011a] for details on these calculations. Adding industrial and = domestic=20 water demand (after subtraction of return flows), livestock water demand = and=20 gross irrigation water demand yields net total water demand, .
2.4. Estimating Nonrenewable = Groundwater=20 Abstraction for Irrigation
[26] Nonrenewable = groundwater=20 abstraction in irrigated areas is obtained in three steps: (1) = calculation of=20 grid-based (0.5=C2=B0 =C3=97 0.5=C2=B0) natural groundwater recharge and = additional recharge=20 from irrigation, (2) downscaling of country-based estimates on = groundwater=20 abstraction to grid-based groundwater abstraction, and (3) for the = irrigated=20 areas, subtracting grid-based groundwater abstraction from groundwater = recharge=20 to estimate nonrenewable groundwater abstraction for irrigation. = Although=20 groundwater depletion leads to increased capture of exogenous surface = water and=20 groundwater through a reduction of groundwater discharge to streams and=20 increased recharge from streams [Bredehoeft,=20 2002], in many (semiarid) areas with extensive and long-time = groundwater=20 exploitation, with thousands of small agricultural wells dispersed over = the=20 entire groundwater region, the effects of increased capture are rather = small or=20 the time of increased capture has long passed, and removal from storage = can be=20 estimated by the difference between abstraction and recharge rates [Wada et = al.,=20 2010]. In sections=20 2.4.1=E2=80=932.4.3=20 these three steps are subsequently described in more detail.
2.4.1. Natural Groundwater = Recharge and=20 Additional Recharge From Irrigation
[27] Natural = groundwater recharge=20 can be readily obtained from the simulation of PCR-GLOBWB for the period = 1960=E2=80=932000, where the natural groundwater recharge is estimated = as the net flux=20 from the lowest soil layer to the groundwater layer, i.e., deep = percolation=20 minus capillary rise. Note that in PCR-GLOBWB, the long-term average of=20 groundwater recharge equals long-term average groundwater discharge from = the=20 groundwater layer to the surface water network. Our estimate of the = average=20 natural groundwater recharge globally amounts to 15.2 =C3=97 = 103=20 km3 yr=E2=88=921 [Wada et = al.,=20 2010]. It should be noted that the simulated recharge does not = explicitly=20 include recharge from streams and lakes, although such effects may be = implicitly=20 included when calibrating soil characteristics to reproduce observed = low-flow=20 properties.
[28] To account for = additional=20 groundwater recharge from irrigation, RIrr,=20 we use the following approximation:
where LIrr is the amount of = irrigation=20 losses estimated on the basis of the country-specific efficiency factor = eIrr (as used in equation=20 (3)) (m3 d=E2=88=921), k(=CE=B8E_FC) is the unsaturated hydraulic = conductivity at=20 field capacity (m d=E2=88=921) and AIrr is=20 the corresponding irrigated areas within the cell (m2). This=20 formulation is based on the fact that in irrigation practice water is = supplied=20 to wet the soil to field capacity during the application and the amount = of=20 irrigation water in excess of the field capacity can percolate to the=20 groundwater system. The additional recharge rate thus equals the = unsaturated=20 hydraulic conductivity of the bottom soil layer at field capacity, = assuming=20 gravitational drainage. However, the total percolation losses are = further=20 constrained by the reported country-specific loss factor on the basis of = the=20 work by Rohwer et=20 al. [2007].
2.4.2. Grid-Based Groundwater = Abstraction=20 for Irrigation
[29] To estimate = grid-based=20 groundwater abstraction, we start with groundwater abstraction rates per = country=20 and groundwater regions where major aquifers are present as stored in = the IGRAC=20 GGIS database (International Groundwater Resources Assessment Centre, http://www.un-igrac.org/). We = indexed=20 country-based groundwater abstraction to the year 2000 on the basis of=20 population statistics as most abstraction data are collected before the = year=20 2000. Data on abstraction rates were lacking for Afghanistan, North = Korea, Sri=20 Lanka, Colombia, and several countries in Africa and South America.
[30] Since the = locations where=20 groundwater is abstracted by wells are not known for most of the = countries, Wada et = al.=20 [2010] downscaled groundwater abstraction per country to a 0.5=C2=B0 = grid=20 resolution by using water demand as a proxy, i.e., on the basis of the=20 assumption that groundwater is abstracted close to where it is most = needed. In=20 this study, we improve on their approach by taking into account the = available=20 surface freshwater.
[31] First, for each = month, m, in the year 2000 and for each grid cell, i, we calculate deficits, Defsm,i, between the surface water availability,=20 WAm,i, simulated by PCR-GLOBWB = (after=20 correcting for upstream water consumption through equation=20 (6)) and the estimated net total water demand, (see section=20 2.3). Because we are interested in groundwater as an alternative = source we=20 limit this analysis to regions where the aquifers are present (major = groundwater=20 regions of the world according to the IGRAC GGIS). We subsequently = estimate=20 annual deficits, Defsa,i, for the = year=20 2000.
We assume that grid cells with deficits (i.e., water demand in excess = of blue=20 water availability) are the main locations where groundwater is = abstracted as an=20 alternative resource to satisfy the demand.
[32] Second, the annual = deficits,=20 Defsa,i, are filled by the amount = of=20 available country total groundwater abstraction until total water demand = is=20 satisfied by groundwater abstraction per grid cell. Total annual = deficits per=20 country, Defsa, are given by
where n is the number of grid cells with = deficits per=20 country. If the total annual deficits are larger than the available = annual=20 groundwater abstraction in a country, Defsa=20 > GroundWa, (e.g., Egypt, Sudan, = Mali,=20 Niger, Sudan, Turkmenistan and Uzbekistan), we distribute the country=20 abstraction according to the intensities rather than the volume of the = deficits.=20 In most cases the available abstraction is larger than the total = deficits in a=20 country and the remaining country-based abstraction (GroundWa =E2=88=92 Defsa) is further=20 allocated relative to the intensity of total water demand over its = country total=20 (again limited to cells in major groundwater regions):
[33] Third, we use the = fraction=20 of irrigation water demand over net total water demand in each grid cell = in the=20 irrigated areas to arrive at groundwater abstraction for irrigation, = :
Over the irrigated areas (MIRCA2000 [Portmann = et=20 al., 2010]) irrigation water demand is dominant, so that = groundwater=20 abstraction for irrigation is close to the total groundwater abstraction = in=20 irrigated areas.
2.4.3. Nonrenewable Groundwater = Abstraction=20 for Irrigation
[34] Nonrenewable = groundwater=20 abstraction is subsequently calculated by subtracting the sum of the = simulated=20 natural groundwater recharge and additional recharge from irrigation = from the=20 gridded groundwater abstraction in a vertical slice per grid cell over = all=20 regions of the world. Negative values indicate grid cells with overlaps = between=20 available blue water and renewable groundwater abstraction and the = remaining=20 positive values denote grid cells where nonrenewable groundwater is = abstracted.=20 Nonrenewable groundwater abstraction for irrigation is estimated again = by using=20 the fraction of irrigation water demand over net total water demand in = each grid=20 cell in the irrigated areas:
[35] We also assessed = the past=20 trend of groundwater abstraction for the period 1960=E2=80=932000 as a = first-order=20 estimate assuming that country-based groundwater abstraction increases = linearly=20 with water demand. So for a given year, k, an = estimate=20 of country-based groundwater abstraction is obtained by multiplying the=20 groundwater abstraction of the year 2000 by the ratio of country-based = water=20 demand of year, k, over that of the year 2000 = water=20 demand. Water demand is calculated according to Wada et = al.=20 [2011a, 2011b] = (see section=20 2.3). Next, by repeating for each year the methodology previously = described,=20 we thus estimate the amount of nonrenewable groundwater abstraction for=20 irrigation during the period 1960=E2=80=932000.
3.1. Model Evaluation and=20 Validation
[36] Before we present = the result=20 of nonrenewable groundwater for irrigation, we first provide the = evaluation of=20 model performance and validation result for downscaling country total=20 groundwater abstraction rates to 0.5=C2=B0 grids, and compare estimated = nonrenewable=20 groundwater abstraction to independent estimates of groundwater = depletion.
[37] To assess model = performance,=20 we compared simulated terrestrial water storage (TWS) to the GRACE = satellite=20 observations. Monthly GRACE TWS anomalies were obtained from the DEOS = Mass=20 Transport release 1/1b (DMT-1) model of Liu et = al.=20 [2010] for the period 2003=E2=80=932008. Since this period extends = beyond that of=20 the available climate forcing for PCR-GLOBWB used in this study (section=20 2.2), we forced the model by a comparable climate data set of daily = rainfall=20 and temperature fields taken from the ECMWF Operational Archive (http://www.ecmwf.int/products/data/archive/descriptions/od/ope= r/index.html).=20 For compatibility with our overall analysis, we bias-corrected this data = set by=20 scaling the long-term monthly means of these fields to those of the CRU = TS 2.1=20 data set. Monthly reference (potential) evapotranspiration was computed=20 according to the FAO guidelines [Allen et = al.,=20 1998] using the relevant long-term monthly values from the CRU TS = 2.1 and=20 CLIM 1.0 data sets (section=20 2.2) where temperature was replaced by that of the bias-corrected = ECMWF=20 Operational Archive, and subsequently downscaled on the basis of the = daily=20 temperature. Figure 1 = compares our=20 simulated TWS anomalies with those of the GRACE observations for major = basins of=20 the world. PCR-GLOBWB reproduces the seasonal and interannual variations = in TWS=20 well particularly in (semi)arid basins such as Niger and Zambezi. TWS is = also=20 reproduced reasonably well for basins where major irrigated areas are = present=20 such as Mississippi, Nile, Indus, Genges and Mekong. However, PCR-GLOBWB = underestimates TWS for most of the period for Yangtze. This might be = caused by=20 overestimation of our evapotranspiration, resulting in underestimation = of soil=20 water storage and surface runoff. For Danube, TWS is underestimated for = 2006 in=20 which significant flooding, i.e., 2006 European floods, occurred because = of=20 heavy rain and melting snow, while it is well reproduced for the other = years.=20 Overall, PCR-GLOBWB reproduces TWS adequately for most of the basins, = which=20 increases our confidence on model performance.
[38] Figure 2 compares our = estimated=20 groundwater abstraction rates for the United States with reported values = per=20 county taken from the USGS (http://www.usgs.gov/). Both estimated = and=20 reported values sum up to a total of around 115 km3 = yr=E2=88=921=20 (year 2000). Although the spatial resolution does not exactly correspond = with=20 each other, our estimated groundwater abstraction rates show a good = agreement=20 with those of the USGS throughout the United States (excluding Alaska = and=20 Hawaii). Large groundwater abstraction rates over the High Plains = (Ogallala)=20 Aquifer, the Central Valley, California, the Mississippi River Valley = alluvial=20 aquifer, the Basin and Range basin fill aquifers, Florida, the Pacific = Northwest=20 basaltic rock aquifers and the Snake River Plain basaltic rock aquifers = are well=20 reproduced by our downscaling method (see section=20 2.4.2). For the Southwest, we also have a good agreement if we add = up our=20 downscaled values per county.
[39] Table 1 shows, per country, = our=20 estimate of nonrenewable groundwater abstraction rates. An uncertainty = analysis=20 for the estimates was performed according to Wada et = al.=20 [2010]. Note that here we use estimates of total nonrenewable = groundwater=20 abstraction for all purposes including irrigation as an estimate of = groundwater=20 depletion, i.e., groundwater removal from storage. Moreover, we limit = our=20 estimates to subhumid to arid regions to prevent excessive = overestimation of=20 groundwater depletion due to enhanced recharge as would occur in areas = with=20 abundant surface water. Comparing to the previous works by Wada et = al.=20 [2010, 2011a], = we=20 rectified the groundwater recharge estimate by including additional = recharge=20 from irrigation to find that global recharge is enhanced by 420 = km3=20 yr=E2=88=921 which reduces depletion from 283 (=C2=B140) to = 256 (=C2=B138)=20 km3 yr=E2=88=921. Figure=20 3 compares for several regions our estimates of groundwater = depletion rates=20 with independent estimates of groundwater depletion [Sahagian = et=20 al., 1994; McGuire, 2003; Foster = and=20 Loucks, 2006; Rodell et al., 2009; Tiwari = et=20 al., 2009; Famiglietti et al., 2011; Konikow,=20 2011]. As shown in Figure 3,=20 our estimates compare well with the other independent assessments of = groundwater=20 depletion reported around the year 2000, showing that our approach does = not lead=20 to large structural errors. The exception here is the estimate for north = India=20 (i.e., Rajasthan, Punjab, Haryana) and northern India and adjacent areas = (NIAAs)=20 where compared to the GRACE estimates from Rodell = et al.=20 [2009] and Tiwari et al. [2009], groundwater depletion is=20 overestimated (38.3 km3 yr=E2=88=921 versus 17.7 = km3=20 yr=E2=88=921 and 71.7 km3 yr=E2=88=921 = versus 54 km3=20 yr=E2=88=921). This most likely results from the fact that = surface water=20 availability necessary to meet the large irrigation water demand is=20 underestimated as it is known that in the Indo-Gangetic plains extensive = diversion works are present [Sharma and = Kansal, 2009]=20 that are not included in our modeled surface water system. Also, it may = well be=20 that additional recharge occurs from these diversions.
[40] A comparison of = our=20 estimates with those of a recent study by Konikow=20 [2011] also shows a reasonable agreement (see Figure 3). We compare our = estimates=20 with Konikow's [2011] results for the last period = 2000=E2=80=932008=20 since these estimates are also closest to the other independent = estimates.=20 Differences between these estimates are relatively small, except for the = estimate for northern India and adjacent areas, which Konikow=20 [2011] equally takes from Tiwari = et al.=20 [2009] and one region in the United States: western U.S. alluvial = basins.=20 This region used to have depletion rates similar in magnitude to our = estimate=20 during the period 1950=E2=80=931980, after which surface water = diversions and artificial=20 recharge programs reduced depletion to small values [Konikow,=20 2011]. If we add up the depletion rates from the United States and = the other=20 5 regions evaluated by Konikow [2011], we end up with a depletion of = 132.8=20 km3 yr=E2=88=921 which is not very different from = the 101.6=20 km3 yr=E2=88=921, given the uncertainty estimate = provided for the=20 global estimate (27% for Konikow [2011] and 38 km3 = yr=E2=88=921 for=20 this study). The global estimate of groundwater depletion by Konikow=20 [2011] (145 km3 yr=E2=88=921) is however very = different from=20 that of ours (256 km3 yr=E2=88=921; see Table 1). This is largely = due to the=20 extrapolation performed by Konikow [2011] assuming that the ratio of = depletion to=20 groundwater abstraction of the rest of the world is the same as that of = the=20 United States (i.e., 15.4%). Table=20 1 shows that this is clearly not the case as the ratios considerably = vary=20 among countries (between 7% and 87%). For instance, taking the estimated = depletion from the GRACE [Tiwari et al., 2009] for northern India and = adjacent areas=20 (54 km3 yr=E2=88=921) and comparing this to = reported abstraction=20 rates (see http://www.un-igrac.org/) for=20 India (190 km3 yr=E2=88=921) and Pakistan (55 = km3=20 yr=E2=88=921) already yields a ratio of 22% and could easily = ad up to 25=E2=80=9330%=20 if we correct the total abstraction of 245 km3 = yr=E2=88=921 with=20 the part that is abstracted in southern India. These fractions are much = higher=20 than the 15.4% assumed by Konikow [2011] and largely explain the = differences between=20 his and our estimates. Major other hotspots such as Iran, Mexico, the = remaining=20 parts of India and China and a number of countries in central Asia and = the=20 Middle East are thus not adequately accounted for by Konikow=20 [2011].
3.2. Total and Nonrenewable = Groundwater=20 Abstraction for Irrigation
[41] Resulting = groundwater=20 abstraction for irrigation downscaled to 0.5=C2=B0 for the year 2000 is = shown in Figure 4c, while Figure 4a shows the = groundwater=20 abstraction per country obtained from the IGRAC GGIS database and Figure 4b shows the = estimated=20 irrigation water demand. Figure 5=20 shows the estimated nonrenewable groundwater abstraction for irrigation = for the=20 year 2000. Large amounts of groundwater are being abstracted over major=20 irrigated regions such as India, northern China, United States, = Pakistan,=20 southern Mexico, northern Iran, central Saudi Arabia, and southern = Europe.=20 Summing total and nonrenewable groundwater abstraction for irrigation = for these=20 areas amounts to 80% and 90% of the global total for the year 2000,=20 respectively.
3.3. Contribution of Water = (Re)sources to=20 Irrigated Crops
[42] On the basis of = the=20 simulations with PCR-GLOBWB and the calculations of water demand and=20 nonrenewable groundwater abstraction we estimated the contribution of = different=20 sources of water to irrigated crops, i.e., to gross crop water demand: = green=20 water, blue water, nonrenewable groundwater or nonlocal water resources. = For the=20 year 2000, water used for irrigated crops globally amounts to 2510=20 km3 yr=E2=88=921, of which 47% (1172 = km3=20 yr=E2=88=921) and 53% (1338 km3 = yr=E2=88=921) are composed of=20 green water use and gross irrigation water demand, respectively (see Table 2). Blue water = contributes 63%=20 or 844 km3 yr=E2=88=921 to the gross irrigation = water demand while=20 nonrenewable groundwater contributes 18% or 234 km3 = yr=E2=88=921.=20 Potential nonlocal water resources contribute the remaining 19% or 260=20 km3 yr=E2=88=921 to the gross irrigation water = demand. We estimate=20 that about 85% of the global nonrenewable groundwater abstraction (275=20 km3 yr=E2=88=921) is used for irrigation. Our = estimate of=20 nonrenewable groundwater used for irrigated crops is comparable to the = lower=20 range of that of V=C3=B6r=C3=B6smarty et al. [2005], who suggest = that 16% to 33% of=20 agricultural water demand is nonlocal and nonrenewable (391 to 830=20 km3 yr=E2=88=921). If we add our estimate of = nonrenewable=20 groundwater abstraction to that of nonlocal water resources, being the = remaining=20 shortage to bring irrigation water to the optimum, our total falls = within their=20 range.
[43] Focusing on = country=20 estimates in Table 2, it = can be=20 seen that India needs the amount of 600 km3 = yr=E2=88=921 of water=20 in order to satisfy its gross crop water demand in irrigated areas, = which is=20 nearly a quarter of the global total. In India, the gross irrigation = water=20 demand (353 km3 yr=E2=88=921) constitutes nearly = 60% of its gross=20 crop water demand, of which 19%, or 68 km3 = yr=E2=88=921, is=20 supplied from nonrenewable groundwater. India uses the largest amount of = nonrenewable groundwater for irrigation among the countries. Because of = the=20 scarce rainfall under its semiarid climate, in Pakistan most (80%, or = 146=20 km3 yr=E2=88=921) of the gross crop water demand = is satisfied by=20 irrigation. While a major part of the irrigation water is taken from the = Indus=20 river, nonrenewable groundwater contributes 24% to the gross irrigation = water=20 demand and amounts to 35 km3 yr=E2=88=921, the = second largest=20 volume after India. Similar to India, gross irrigation water demand = constitutes=20 60% of the gross crop water demand of the United States and Mexico. In = these=20 countries, around 20% of gross irrigation water demand comes from = nonrenewable=20 groundwater while around 60% is supplied from blue water. In Iran and = Saudi=20 Arabia, where rainfall and surface freshwater are extremely scarce, = nonrenewable=20 groundwater provides the largest contribution to gross irrigation water = demand,=20 40% and 77%, respectively. Our estimate in Saudi Arabia suggests that = its=20 irrigation practice is near-optimal in terms of productivity and = sustained by=20 the large abstraction of nonrenewable groundwater resources. In China, = on the=20 other hand, nonrenewable groundwater contributes merely around 15% to = the gross=20 irrigation water demand. This can be explained by the large share of = green water=20 used by irrigated crops which enjoy substantial but variable rainfall,=20 contributing 66%, or 267 km3 yr=E2=88=921, to the = gross crop water=20 demand, the largest volume for any of the major irrigated countries.
[44] Figure 6 shows the current=20 contribution of each water resource to irrigated crops (gross crop water = demand=20 in irrigated areas) for major groundwater users such as India, China, = United=20 States, Pakistan, Iran, Mexico, Saudi Arabia, Egypt, Spain and Libya. = Large=20 fractions of nonrenewable groundwater abstraction over gross irrigation = water=20 demand are observed predominantly in arid regions such as the Middle = East.=20 Nonrenewable groundwater supplies more than half of the gross irrigation = water=20 demand in Saudi Arabia, Qatar, Libya and UAE. However, it should be = noted that=20 gross crop water demand in most of countries is not fully covered by the = sum of=20 available green water, blue water and nonrenewable groundwater, = particularly in=20 water scarce regions such as India, United States, Pakistan, Iran, = Mexico,=20 Egypt, Kazakhstan, Spain, Italy, Turkey, South Africa, Morocco and = Algeria. In=20 some regions such as west United States and the Indo-Gangetic plains = extensive=20 larger and smaller diversion works are probably able to meet this = outstanding=20 demand by nonlocal water. In other less developed regions, it is common = that=20 farmers irrigate less than optimally because of persistent water = scarcity or to=20 minimize costs. Here, potential yields over current irrigated areas can = still be=20 improved if additional water resources were available for irrigation.=20 Conversely, in China, Russia, Saudi Arabia, Argentina, Yemen and UAE, = current=20 irrigation practice is near optimal in terms of productivity. In these=20 countries, current irrigated areas are almost fully exploited and their = yields=20 from irrigation can only be increased by improved water use efficiency = (i.e.,=20 higher crop water productivity per unit irrigated area) or by expanding = their=20 irrigated areas. However in some of these countries where available blue = water=20 is almost fully used for irrigation, additional abstraction of = nonrenewable=20 groundwater will result in further depletion of groundwater = resources.
[45] We reconstructed = past trends=20 of different water resources contributing to irrigated crops. We only = provide=20 global estimates here because of the rather strong assumption of a = linear=20 relationship between country-based groundwater abstraction and country = net total=20 water demand. Figure 7 = shows that=20 in irrigated areas the gross crop water demand more than doubled from = 1217 to=20 2510 km3 yr=E2=88=921 over the period = 1960=E2=80=932000. For the year=20 1960, green water contributed globally 48% or 589 km3 = yr=E2=88=921=20 to the gross crop water demand resulting in a gross irrigation water = demand of=20 628 km3 yr=E2=88=921. Blue water and nonrenewable = groundwater=20 supplied 73%, or 457 km3 yr=E2=88=921, and 12%, or = 75=20 km3 yr=E2=88=921, of the gross irrigation water, = respectively,=20 leaving 15%, or 96 km3 yr=E2=88=921, for nonlocal = water resources.=20 During the 1960=E2=80=932000 period the global gross irrigation water = demand more than=20 doubled to 1338 km3 yr=E2=88=921 as a result of = expansion of=20 irrigated areas to support growing food demands. The amount of blue = water=20 contributing to the global gross irrigation water demand also increased = to 844=20 km3 yr=E2=88=921 but its share decreased to 63% = for the year 2000.=20 However, the amount and share of nonrenewable groundwater rose to 234=20 km3 yr=E2=88=921 and close to 20%, respectively. = These results=20 suggest that available blue water resources have become extensively = exploited=20 for irrigation. Even though large numbers of reservoirs were constructed = to=20 supply water to irrigation, the increase in their storage capacities has = been=20 tapering off since the 1990s [Chao et = al.,=20 2008]. Consequently, the contribution of nonrenewable groundwater=20 abstraction to meet the gross irrigation water demand has been = increasing=20 rapidly, resulting in an increasing dependency on nonrenewable = groundwater for=20 irrigation in recent years.
[46] Here, we first = compare per=20 major irrigated country our results with available statistics and = previous=20 estimates to explore the differences in crop and irrigation water = demand. Next,=20 we discuss limitations and uncertainties inherent to this study.
4.1. Comparisons With Previous=20 Estimates
[47] We compared the = estimated=20 crop and irrigation water demand with reported and estimated values = taken from=20 the FAO and previous studies in Table=20 3. Siebert=20 and D=C3=B6ll [2008] and Liu and = Yang=20 [2010] assumed that the contribution of green and blue water is = sufficient=20 to meet gross crop water demand and estimated blue water demand from = this. Rost et = al.=20 [2008] quantified the amount of gross and net irrigation water = demand (see=20 equations=20 (4) and (2),=20 respectively) with and without the potential contributions of = nonrenewable and=20 nonlocal blue water resources to meet these demands (IPOT/ILIM; see Table 3). Wisser = et al.=20 [2008], on the other hand, assessed uncertainties of computing gross = irrigation water demand by using two different data sets of irrigated = areas=20 based on the FAO (GMIA [Siebert et al., 2005, 2007] and the IWMI = (GIAM) [Thenkabail et=20 al., 2006]) with two different climate inputs of the National = Centers=20 for Environmental Prediction (NCEP) (NCEP/NCAR [Kalnay = et=20 al., 1996] and the CRU (CRU TS 2.1) [Mitchell = and=20 Jones, 2005]).
[48] Our estimates of = gross crop=20 water demand for irrigated areas per country agree reasonably well with = those of=20 the other studies. For major irrigated countries such as India, China, = and=20 United States, we estimated gross crop water demand to be 600, 403, and = 204=20 km3 yr=E2=88=921, respectively, while the other = studies report=20 313=E2=80=93462, 404=E2=80=93492, and 218=E2=80=93265 km3 = yr=E2=88=921. The variation is=20 likely caused by differences in irrigated areas since we used FAOSTAT = statistics=20 to correct irrigated areas per country from the MIRCA2000 data set, = while Siebert = and=20 D=C3=B6ll [2008] and Liu and Yang [2010] used the MIRCA2000 data set = and that=20 of Siebert et=20 al. [2007], respectively. Differences in prescribed reference = evapotranspiration which is used to compute crop water demand between = this study=20 and the other studies should also explain the variation. Our estimates = of=20 irrigation water demand are close to the reported values of FAO and = those of Siebert = and=20 D=C3=B6ll [2008] for most of the countries. Range is large = for irrigation=20 water demand estimates from the other studies. This is because Wisser = et al.=20 [2008] account for uncertainties caused by different irrigated area=20 estimates combined with different climate inputs. For India, the use of = the IWMI=20 irrigated areas doubles the estimated gross irrigation water demand = compared to=20 the estimate based on the FAO irrigated areas, while the climate input = of the=20 NCEP results in lower values compared to that of the CRU. The same trend = applies=20 for China, but not for the United States and Egypt; here both FAO and = IWMI=20 report similar areas yet the gross irrigation water demand decreases = when the=20 NCEP climate data are used. Globally, gross irrigation water demand is = larger by=20 30% when using IWMI irrigated areas instead of the FAO data, while it = decreases=20 by the same magnitude when NCEP instead of CRU climate data are used [Wisser = et=20 al., 2008].
[49] Our blue water = contribution=20 to irrigation is generally lower than reported by other studies, because = we=20 partitioned blue water contribution into renewable and nonrenewable = resources,=20 while blue water contribution of the previous studies include both = terms. If we=20 combine both terms, our blue water contribution becomes close to = existing=20 estimates. Rost=20 et al. [2008] implicitly quantified amounts of nonlocal and=20 nonrenewable water resources used for irrigation (i.e., IPOT and ILIM). = For=20 India and China, their values are larger than our estimates of = nonrenewable=20 groundwater and nonlocal water resources whereas they are comparable to = our=20 values for the United States, Pakistan and Egypt. Our green water = contribution=20 to irrigated crops per country is close to those of Siebert = and=20 D=C3=B6ll [2008] and Liu and Yang [2010] since we used the same CRU = climate=20 data set as forcing.
4.2. Limitations and=20 Uncertainties
[50] Limitations of = this study=20 largely result from uncertainties caused by input data and modeling = assumptions.=20 For example, as in previous studies [e.g., D=C3=B6ll = and=20 Siebert, 2002; Rost et al., 2008; Wisser = et=20 al., 2008; Siebert and D=C3=B6ll, 2010; Hanasaki = et=20 al., 2010], growth of irrigated crops to compute crop water = demand=20 with crop calendar is simulated in a rather simple manner which only=20 approximates actual water use conditions. Simulating optimal irrigation = likely=20 causes overestimation of crop water demand in regions where persistent = water=20 scarcity leads farmers to irrigate less than the optimal conditions [D=C3=B6ll = and=20 Siebert, 2002] such as India, Pakistan, Iran and Egypt.
[51] The amount of soil = moisture=20 (i.e., green water) and surface freshwater (i.e., blue water) available = to=20 irrigated crops were simulated by PCR-GLOBWB. Extensive validations of=20 PCR-GLOBWB were performed by Van Beek et al. [2011] by comparing the = simulated river=20 discharge to observations [Global Runoff Data Centre (GRDC),=20 2008] and the estimated actual evapotranspiration to that of the = ERA-40=20 reanalysis as proxy for observed rates [K=C3=A5llberg et=20 al., 2005]. Comparisons with over 3600 GRDC stations show = that the=20 coefficient of determination (R2) is high (=E2=89=880.9) for = most of the=20 stations but the coefficient of determination decreases when the mean = minimum=20 and maximum monthly discharge are considered instead of the mean = discharge.=20 Interannual variability is mostly well reproduced in major rivers except = the=20 Niger (R2 =3D 0.54), Orange (R2 =3D 0.54), Murray=20 (R2 =3D 0.60), Indus (R2 =3D 0.62), Zambezi = (R2 =3D=20 0.75) and Nile (R2 =3D 0.87) where the simulated river = discharge is=20 often also overestimated.
[52] Although strictly = speaking=20 not observational data, the ERA-40 reanalysis set can be seen as a proxy = to=20 actual evapotranspiration measurements because of the assimilation of=20 atmospheric observations [K=C3=A5llberg et al., 2005]. Comparisons between = our simulated=20 actual evapotranspiration and that of the ERA-40 show that for the = nonirrigated=20 areas both data sets are quite similar and that differences can be = largely=20 explained by differences in rainfall estimates between ERA40 and CRU [Van Beek = et=20 al., 2011]. However, for most of the major irrigated areas = the=20 simulated actual evapotranspiration of PCR-GLOBWB is generally smaller = than that=20 of the ERA-40 reanalysis. While seasonal courses are well reproduced = during the=20 wet season, the deviation widens during the dry season in heavily = irrigated=20 regions such as the Great Plains, Spain and Pakistan. The latter comes = from the=20 fact that, unlike evaporation from PCR-GLOBWB, ERA-40 evaporation is a=20 reanalysis product. Because irrigation is not explicitly modeled in the = ECMWF=20 land surface model, screen temperatures will generally be overestimated = over=20 these areas because of an underestimation of the latent heat flux. = Consequently,=20 during the analysis steps, additional moisture is added to the soil = moisture=20 reservoirs in order to increase the subsequent latent heat fluxes and = keep=20 temperature as calculated by the numerical weather prediction model = close to the=20 observations. This =E2=80=9Cimplicit irrigation scheme=E2=80=9D = resulting from data assimilation=20 thus explains the higher evaporation rates of ERA40 during the dry = irrigation=20 season (see the work by Van Beek et al. [2011] and results therein). Van Beek = et=20 al. [2011] also showed that the difference in dry season = evaporation=20 between ERA-40 and PCR-GLOBWB could largely be explained by the = calculated=20 irrigation water demand as also used in this paper.
[53] Uncertainties in = the=20 estimated groundwater abstraction also affect our results. Because of a = lack of=20 observations, the country data does not include nonreported groundwater=20 abstractions, which might be prevalent over major irrigated regions such = as=20 northwest India and northeast Pakistan. For example, Foster = and=20 Loucks [2006] suggests the amount of groundwater abstraction = in India=20 to be around 240 km3 yr=E2=88=921 while we used = 190 km3=20 yr=E2=88=921. We identified main locations (i.e., grid cells) = where=20 groundwater is abstracted by using surface freshwater deficits over = total water=20 demand as a proxy. Validation result shows that groundwater abstraction = rates=20 were well reproduced by our downscaling approach as compared to reported = values=20 per county taken from the USGS (see section=20 3.1). Our approach thus improves upon the earlier estimate of = nonrenewable=20 groundwater abstraction of Wada et al. [2010] who used simply total water = demand as a=20 proxy.
[54] We quantified the = residual=20 between the simulated gross irrigation water demand and the available = blue water=20 resources and nonrenewable groundwater abstraction for irrigation as the = estimate of the contribution of nonlocal water resources. This residue = was=20 particularly large in India, Pakistan, the United States, Iran, and = Mexico.=20 Although actual nonlocal sources such as water diversions, e.g., the = Central=20 Valley Project in the United States, irrigation canals in the Yamuna = River, a=20 major tributary of Indus, [Sharma and Kansal, 2010] and desalinated water = use=20 (globally 5 km3 yr=E2=88=921 in the year 2000), = account for part=20 of these nonlocal water resources, a considerable part of the residual = water may=20 be attributed to the uncertainties described above. It is unlikely that = nonlocal=20 water resources are sufficient to fill all the shortage. For example, if = we use=20 the amount of groundwater abstraction suggested by Foster = and=20 Loucks [2006] for India, our estimate of nonrenewable = groundwater=20 abstraction for irrigation increases by around 50% while the amount of = nonlocal=20 water resources decreases by the same magnitude. And if we overestimated = the=20 crop water demand based on optimal crop growth in India, a discrepancy = to actual=20 crop water use further reduces the amount of nonlocal water resources. = The same=20 conditions also apply to Pakistan and Iran where persistent water = scarcity is=20 prevalent.
[55] Since our global = model does=20 not include additional capture and surface water diversions (e.g., = aqueducts),=20 overestimation of nonrenewable groundwater abstraction occurs in some = regions=20 notably in north India, western U.S. alluvial basins, and Southern = California=20 (Los Angeles and San Diego area). However, it does not demonstrate the=20 inadequacy of estimating depletion from the water budget but shows that = the=20 attribution of country total abstraction rates to grid-based rates using = local=20 surface freshwater deficit as a proxy has its limitations. This is a = general=20 limitation of all global modeling efforts: when viewed at individual = cell=20 scales, disparities are likely to occur, while the regional variation is = adequately captured. Even though our method for computing large-scale=20 nonrenewable groundwater abstraction has its limitations and = uncertainties to be=20 able to provide estimates across the entire globe, it yields adequate = results=20 when compared to the independent estimates (see Figure 3).
[56] This study = provides a global=20 overview of the amount of nonrenewable groundwater abstraction that = contributes=20 to gross irrigation water demand. Apart from estimates of green and blue = water=20 contribution to water supplied to irrigated crops by previous studies = [e.g., Falkenmark et=20 al., 1997; Jackson et al., 2001; D=C3=B6ll = and=20 Siebert, 2002; Kundzewicz et al., 2007; Rost et = al.,=20 2008; Wisser=20 et al., 2008; Siebert and D=C3=B6ll, 2010; Hanasaki = et=20 al., 2010; Liu and Yang, 2010], we explicitly quantified = the amount=20 of nonrenewable groundwater abstraction for irrigation by confronting = the sum of=20 the simulated natural groundwater recharge and additional recharge from=20 irrigation with the gridded groundwater abstraction for irrigation. = Thus, our=20 blue water denotes exclusively renewable surface freshwater and = groundwater=20 which is closer to its definition. Optimal growth of irrigated crops and = available green water and blue water to meet gross crop water demand = were=20 simulated by applying a state-of-the-art global hydrological model = PCR-GLOBWB.=20 The resulting shortage between gross irrigation water demand and = available blue=20 water and nonrenewable groundwater abstraction was calculated as an = estimate of=20 nonlocal water resources.
[57] The results of = this study=20 show that nonrenewable groundwater abstraction globally contributes = nearly 20%,=20 or 234 km3 yr=E2=88=921, to the gross irrigation = water demand for=20 the year 2000 and has more than tripled in size since the year 1960. = Country=20 assessments reveal that nonrenewable or nonsustainable groundwater = supplies=20 large shares of current irrigation water particularly for semiarid = regions where=20 surface freshwater and rainfall are very scarce: Pakistan, Iran, Saudi = Arabia,=20 Libya, UAE and Qatar. Much of current irrigation in these regions is = sustained=20 by nonsustainable groundwater.
[58] The reconstructed=20 development from 1960 to 2000 shows an increased dependency of = irrigation on=20 nonsustainable groundwater with time. Thus, irrigation is more and more=20 sustained by an unsustainable water source. Severe competition for = scarce=20 surface freshwater resources for irrigation also worsens the condition = of=20 depleting groundwater resources. We argue that the unsustainability of=20 groundwater use for irrigation is an important issue not only for the = countries=20 with intensive groundwater use, but also for the world at large since=20 international trade directly links food production in one country to = consumption=20 in another. Rising population and their food demands are likely to = increase the=20 amount of nonrenewable groundwater abstraction for irrigation, = particularly in=20 emerging countries such as India, Pakistan, China, Iran and Mexico. This = will=20 result in falling groundwater levels which may eventually become = unreachable for=20 local farmers with limited technology. Groundwater resources have = supported=20 their livelihoods to generate their food and income for decades. Limits = to=20 global and regional groundwater consumption cast large uncertainties on = their=20 livelihoods, threatening regional and global food security. Groundwater=20 depletion is a long outstanding issue, and various efforts have been = made to=20 explore solutions [Moench et al., 2003], yet it is far from = resolved. This=20 study gives further evidence to scale of the issue and its growing = trend. It is=20 urging to invest further political, institutional and economic efforts = to limit=20 the overdraft, yet important to find adaptive responses that do not = reduce=20 current food productivity.
[59] We are grateful to = three=20 anonymous reviewers, the Associate Editor, and the Editor for their = constructive=20 comments and thoughtful suggestions, which significantly helped to = improve the=20 quality of this manuscript. We are also thankful to Hassan Hashemi = Farahani and=20 Pavel Ditmar for sharing the DMT-1 GRACE data with us. This research = benefited=20 greatly from the availability of invaluable data sets as acknowledged in = the=20 references. This study was financially supported by Research Focus Earth = and=20 Sustainability of Utrecht University (project FM0906: Global Assessment = of Water=20 Resources).
Citation: =20 (2012), Nonsustainable = groundwater=20 sustaining irrigation: A global assessment, Water Resour. Res., 48, W00L06, = doi:10.1029/2011WR010562.