Hydrologic Impacts of Engineering Projects on the Tigris–Euphrates System and Its Marshlands

Journal of Hydrology (2008) 353,59– 75

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/jhydrol

Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands

C. Jones a, M. Sultan a,*, E. Yan b, A. Milewski a, M. Hussein c, A. Al-Dousari d, S. Al-Kaisy e, R. Becker a a Department of Geosciences, Western Michigan University, 1903 West Michigan Avenue, Kalamazoo, MI 49008, USA b Environmental Science Division, Argonne National Laboratory, Chicago, IL, USA c Iraq Reconstruction Management Ofﬁce, US Embassy, Baghdad, Iraq d Kuwait Institute for Scientiﬁc Research, Kuwait City, Kuwait e Department of Applied Geology, University of Tikrit, Tikrit, Iraq

Received 25 June 2007; received in revised form 16 January 2008; accepted 22 January 2008

KEYWORDS Summary Rising demands for fresh water supplies are leading to water management Tigris–Euphrates practices that are altering natural flow systems world-wide. One of the most devas- watershed; tated of these natural systems is the Tigris–Euphrates watershed that over the past Continuous rainfall– three decades has witnessed the construction of over 60 engineering projects that elim- runoff model; inated seasonal flooding, reduced natural flow and dramatically reduced the areal SWAT; extent (1966: 8000 km2; 2002: 750 km2) of the Mesopotamian Marshes downstream. Engineering projects We constructed a catchment-based continuous (1964–1998) rainfall runoff model for the watershed (area: 106 km2) using the Soil Water Assessment Tool (SWAT) model to understand the dynamics of the natural flow system, and to investigate the impacts of reduced overall flow and the related land cover and landuse change downstream in the marshes. The model was calibrated (1964–1970) and validated (1971–1998) against stream flow gauge data. Using the calibrated model we calculated the temporal variations in the average monthly flow rate (AMFR), the average monthly peak flow rate (AMPFR), and annual flow volume (AFV) of the Tigris and Euphrates into the marshes at a location near Al-Basrah city (31N, 47.5E) throughout the modeled period. Model results indicate that the AMPFR (6301 m3/s) and average annual flow volume (AAFV: 80 · 109 m3/yr) for period A (10/1/1965–09/30/1973), preceding the construction of the major dams is progressively diminished in periods B1 (10/1/1973–09/30/1989; AMP- FR: 3073 m3/s; AAFV: 55 · 109 m3/yr) and B2 (10/1/1989–09/30/1998; AMPFR, 2319 m3/s; AAFV: 50 · 109 m3/yr) that witnessed the construction of the major dams (B1: Keban, Tabqa, Hamrin, Haditha, Mosul, Karakaya; B2: Ataturk) due to the combined effects

* Corresponding author. Tel.: +1 269 387 5487. E-mail address: [email protected] (M. Sultan).

of filling artificial lakes, evaporation and infiltration of impounded water and its utilization for irrigation purposes. To investigate the impacts of reduced flow on the areal extent of the marshes, we examined the variation in marsh size extracted from temporal satellite data (1966, 1975, 1976, 1977, 1984, 1985, 1986, 1987) acquired around the same approximate time period (July to September) of the year versus simulated AFV for the period preceding the onset (1987) of major local engineering projects (e.g., Crown of Battles River, Loyalty to the Leader Canal, Mother of Battles River) in and around the investigated marshes. Results indicate that the areal extent of the Central and Al-Hammar marshes (e.g., 1966: 7970 km2, 1977: 6680 km2, 1984: 5270 km2) decreases with a decrease in AFV (e.g., 1966: 60.8 · 109 m3, 1977: 56.9 · 109 m3, 1984: 37.6 · 109 m3). Using a relationship that describes the impact of reduced AFV on the areal extent of the marshes, we evaluated the impact of additional reductions in flow that will result from the implementation of the planned engineering projects on the Tigris–Euphrates system over the next few years. Upon completion of the ongoing South Eastern Anatolia project, with projected reductions in AFV exceeding 5 · 109 m3/yr, the sustainable marshes in the Central and Al-Hammar area will be reduced by at least an additional 550 km2. ª 2008 Elsevier B.V. All rights reserved.

Introduction Recently (2002–2007), there has been a reversal in this devastating trend, with the marshland’s area increasing 2 2 Water shortages in the arid parts of the world are affecting from 750 km in 2002–3980 km in 2005 (Jones et al., the human welfare, economic activity, and political stabil- 2005) largely due to piecemeal interventions by locals and ity of these areas. Faced with overpopulation problems the coalition forces. Drainage canals (e.g., COB, LTL, Pros- and demand for development of new agricultural lands to perity River, MOB) were dammed at the inflow point to al- support increasing population, many countries of the arid low the water to continue flowing as it naturally would world are adopting aggressive policies to develop new agri- into the marshes (Basgall, 2003; IMET, 2004). Fig. 2b illus- cultural communities without careful analysis of the envi- trates how a local engineering project, the MOB in this case, ronmental and hydrologic impacts of these projects. can devastate the marshes in a short time and illustrates These problems are exemplified in the countries of the how the revival of the marshes is attainable if these pro- Middle East. Aggressive water management programs by jects were reversed. Similarly, it could be demonstrated Iraq and its neighbors have brought drastic modification that engineering projects, in and around the marsh area, to the Tigris and Euphrates watershed (106 km2) and the had dramatic, but reversible effects on the spatial extent Mesopotamian Marshes downstream in southern Iraq of the marshes. (Fig. 1). The marshlands that extended over an area of In this manuscript, we show that although the major 8000 km2 in the early sixties declined to about 750 km2 in destruction of the marshes could be related to the local the summer of 2002 (Jones et al., 2005). These observed engineering projects in Iraq, it is the reduced flow due to land cover and landuse changes (LCLUC) over the Mesopo- damming in neighboring countries that presents the largest tamian Marshes reflect the impacts of large engineering and more permanent threat to the marshes. The local pro- projects in upstream countries (Turkey, Iran, and Syria) jects were damaging but their devastating impacts could and local engineering projects in Iraq. Examples of up- be readily reverted if the named projects could be elimi- stream projects include the Tabaqa dam (1975; storage nated. The reduced flow due to damming on the other hand capacity (SC): 11.7 · 109 m3) and Tishrine dam (1999; SC: cannot be as easily reverted. 1.9 · 109 m3) in Syria, the Ataturk dam (1992; SC: We constructed a catchment-based continuous (1964– 48.7 · 109 m3) and Keban dam (1975; SC: 31 · 109 m3)in 1998) rainfall–runoff model for the entire watershed and Turkey and the Karkheh dam (2001; SC: 7.8 · 109 m3)in calibrated the model against stream flow data in Turkey Iran (Table 1). Examples of local engineering projects that and in Iraq using the Soil and Water Assessment Tool devastated the marshes include the diversion canals that (SWAT). The calibrated model was used to characterize were largely constructed in the late 1980s and early the natural flow of the system for the time period (1964– 1990s (Jones et al., 2005) to drain the marshes (e.g., Pros- 1970) preceding the construction of the major engineering perity River; Aybas Canal, and Dike; Fig. 2a), or to divert projects and to investigate the hydrologic impacts that re- surface water in and around the marshlands (e.g., Crown sulted from the implementation of the engineering projects of Battles River (COB); Loyalty to the Leader Canal inside and outside of Iraq in the following decades (from (LTL); Mother of Battles River (MOB); Fig. 2a), water that 1970 to 1998). Moreover, the calibrated model was used would have flowed under natural conditions into the as a predictive tool to investigate the impacts of projected marshlands. Fig. 2a also shows the reduction in marshland reductions in river flow with the construction of additional area associated with the implementation of local projects dams in neighboring countries. The SWAT model was se- and large upstream engineering projects between 1969 and lected because (1) the model is an open source code giving 2002. the user the flexibility for making modifications as needed, Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands 61

Figure 1 Location and extent (dark blue outline) of the Tigris–Euphrates watershed plotted on a color coded digital elevation map derived from Shuttle Radar Topography Mission data (SRTM). Also shown in black contours are the annual precipitation amounts. The lowest annual precipitation (<300 mm) occurs in the Mesopotamian plain and the highest precipitation (>600–1200 mm) falls over the mountainous belt to the north (e.g., Taurus Mountains) and east (Zagros Mountains) of the plains.

(2) it is a continuous model, allowing rainfall–runoff and Mountains, across the Mesopotamian plain, into the Central groundwater-recharge estimates to be made over extended and Al-Haweizah marshes, and finally drains in the Arabian periods of time, and (3) Geographic Information System Gulf (Figs. 1 and 2a). The highlands of Turkey provide 88– (GIS) data sets which were generated for the watershed 98% of the Euphrates total flow, modest contributions come could be readily imported into the model. from the Syrian highlands and only minimal additions occur inside the Iraqi borders (UNEP, 2001). The Tigris River, on Site description the other hand, receives only 32–50% of its flow from these highlands, whereas the remaining flow comes from numer- The Euphrates and Tigris Rivers originate from the highlands ous tributaries that derive their water from the Zagros of Turkey, Iran, and Syria, flow down the Mesopotamian Mountains that straddle the Iraq–Iran borders (UNEP, plain, merge near the city of Al-Basrah and distribute 2001). Precipitation feeding both rivers is temporally dis- greatly near this confluence feeding the Mesopotamian tributed from November to April, most of which falls as Marshlands (Fig. 1). The Euphrates River originates in Turkey snow, and remains in the snow pack until the spring snow at the foothills of the Taurus Mountains, heads south melting season. Because snow melt is the dominant source through Syria across Iraq, into the Mesopotamian Marshlands for the flow in the Tigris and Euphrates rivers, annual peak (Al-Hammar Marshes) before it reaches its final destination, flows are observed to coincide with seasonal melting of the Arabian Gulf, via the Shatt-al-Arab (Fig. 1). The Tigris snow in May on both rivers. By the month of October most River originates in Turkey, near Lake Van (elevation: of the snow on the highlands would have melted and it is 3000 m a.s.l.), flows south along the base of the Zagros in this month that the lowest flow on both rivers is observed. 62 C. Jones et al.

Table 1 Dams in the Tigris–Euphrates watershed Name Completion aUse Country Basin bMax. vol cPrin. vol 109 m3 dMax. SA km2 ePrin. SA fRes. K 109 m3 Dokan 1961 IRR Iraq Tigris 6.8 4.6 270 200 1 Derbendikhan 1962 IRR Iraq Tigris 3 2 121 78 1 Keban 1975 HP Turkey Euphrates 31 21.6 675 458 2.5 Tabaqa 1975 HP Syria Euphrates 11.7 9.6 610 447 2.5 Hamrin 1980 IRR Iraq Tigris 3.95 2.15 440 195 2.5 Mosul 1985 HP, IRR Iraq Tigris 11.1 8 371 285 2.5 Haditha 1984 HP, IRR Iraq Euphrates 8.2 7 500 415 4 Karakaya 1987 HP Turkey Euphrates 9.6 7.1 268 198 1 Ataturk 1992 HP, IRR Turkey Euphrates 48.7 38.7 817 707 2 Kralkizi 1997 HP Turkey Tigris 1.9 1.1 57.5 39.2 2 Batman 1998 HP, IRR Turkey Tigris 1.175 0.745 49.3 37.3 2 Tishreen 1999 HP Syria Euphrates 1.9 1.8 166 142 2 a IRR: irrigation, HP: hydroelectric power. b Maximum volume in billion cubic meters. c Principal volume in billion cubic meters. d Maximum surface area in km2. e Principal surface area in km2. f Reservoir hydraulic conductivity in mm/h.

Model construction and data collection parameter (S) for the moisture content of the soil profile on any given day (Neitsch et al., 2005). It varies non-linearly Model framework from AMC I (dry) at wilting point to AMC III at field capacity, and approaches 100 at saturation (Arnold et al., 1993). In The model for the Tigris–Euphrates watershed was con- our case, an average soil moisture condition was found to structed within the SWAT framework to simulate the hydro- be the overwhelming condition throughout the calibration logic processes using its physically based formulations. period across the watershed. The Penman–Monteith meth- SWAT is a semi-distributed continuous watershed simulator od (Monteith, 1981) was used to estimate evaporation on that computes long-term water flow over large basins using bare soils and evapotranspiration (ET) on vegetated areas. daily time steps. Major model components include: flow The variable storage routing method (a variation of the generation, stream routing, pond/reservoir routing, ero- kinematic wave model) developed by Williams (1969) was sion/sedimentation, plant growth, nutrients, pesticides, used for channel routing. The Arcview GIS interface for and land management. In SWAT, a large-scale watershed SWAT2005, developed by DiLuzio et al. (2001), was utilized can be divided into a number of subbasins, which are further to provide the inputs for model construction. subdivided into small groups called hydrologic response units (HRUs) that possess unique land cover, soil, and man- Data collection agement attributes. The water balance of each HRU is calculated through four water storage bodies: snow, soil A major component in model development was the con- profile, shallow aquifer, and deep aquifer. Flows generated struction of co-registered digital datasets for the entire wa- from each HRU in a subbasin are then summed and routed tershed that serve as inputs to our hydrologic model. A large through channels, ponds and/or reservoirs to the outlets number of digital data sets were generated for modeling of the watershed. The detailed descriptions of formulation and calibration purposes: (1) A digital terrain mosaic used in modeling hydrologic processes in HRUs/subbasins (approximately 85 m resolution) constructed from 324 Digi- and routing can be found in Arnold et al. (1998), DiLuzio tal Terrain Elevation Data (DTED) (spatial resolution: et al. (2004), Neitsch et al. (2005), and Srinivasan et al. 85 m; vertical accuracy: +/À15 m) data cells (NIMA, (1998). The model has been successfully validated over sev- 1990, 1991a–c), each covering 1 latitude by 1 longitude eral watersheds (e.g. Arnold et al., 1998). was constructed for stream delineation, and for computing In this study, a non-linear areal depletion curve (Ander- reservoir storage related parameters. (2) Temporal and spa- son et al., 1976) was used to estimate seasonal growth tial distribution of precipitation, temperature, relative and recession of snow pack and to determine snowmelt. A humidity, wind speed, and solar radiation collected from Soil Conservation Service (SCS) curve number method, 36 stations over time period of 1964–1998 to extract (SCS, 1972) adjusted according to soil moisture conditions, monthly precipitation and/or snow accumulation through- was used to calculate direct surface runoff and infiltration out the modeling period, and to constrain evapotranspira- in and through a soil profile (Arnold et al., 1993). The mod- tion, surface runoff, and ground water contributions. (3) A ified SCS method (SCS, 1985) introduces the antecedent mosaic for landuse types generated from various landuse moisture condition (AMC) to account for the impacts of soil maps for Turkey, Iraq, Iran, and Syria (CIA, 1993a–d) and moisture content on surface runoff. In SWAT model, the a mosaic for reclassified (in SCS soil hydrologic groups) soil curve number (CN) was estimated based on the retention types extracted from the United States Department of Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands 63

Figure 2 (a) Diversion and drainage projects in and around the marshes for the area covered by red box in Fig. 1 plotted over a Landsat ETM+ band 2, 4, 7 false color image acquired in 2002. The areas outlined in red and aqua represent the spatial extent of the marshes in 1969 and 2002, respectively. Diversion projects (Glory River, COB River, MOB River, LTL Canal) and drainage projects (Aybas Canal, Prosperity River, Dike) are highlighted in yellow and green, respectively. White arrows indicate the direction of water drainage to the named projects. (b) Temporal time series (1973 MSS, 2002 ETM+, and 2006 ETM+) for the area covered by the black box in Fig. 2a showing the desiccation of a section in the Al-Hammar marsh in 2002 due to the implementation of the MOB River project (outlined by yellow line with black hachuring) and the restoration of this area in 2006 after the project was reversed.

Agriculture (USDA) global soil data set (FAO-UNESCO, 2005); the impacts of reduced flow on the areal extent of the the generated landuse and soil type mosaics were used to marshes for the period preceding the onset (1987) of major construct digital CN maps for soil horizons (typically 3–7 local engineering projects in and around the investigated horizons for each soil type) across the entire watershed. marshes. Five of these mosaics are false color mosaics (4) Stream flow and relevant dam storage parameters (bands 2,4,7) generated from Landsat Thematic Mapper (e.g., principle and maximum storage and holding capaci- (Landsat TM) images (spatial resolution: 28.5 m) acquired ties) on the Euphrates (4 gauges) and Tigris (7 gauges) in Iraq in 1984, 1985, 1986, and 1987, one false color (bands and Turkey for running and calibrating the model. (5) Two 1,3,2) mosaic was generated from Landsat Multi Spectral mosaics of the Landsat Thematic Mapper (Landsat TM) Scanner (Landsat MSS) images (spatial resolution: 70 m) ac- scenes, one generated from scenes acquired in the late quired in 1977, and three black and white mosaics were gen- 1980s (Landsat GeoCover Dataset 1990; spatial resolution: erated from ARGON KH-9 (spatial resolution: 10 m) and 30 m; Tucker et al., 2004) and the other from scenes ac- CORONA images (spatial resolution: 10 m) acquired in quired in the late 1990s (Landsat GeoCover Orthorectified 1976, 1975, and 1966). To examine the overall disintegra- Thematic Mapper Dataset 2000; spatial resolution: 15 m; tion and restoration of the marshlands from 1969 to 2005, Tucker et al., 2004) were used to refine the DTED derived the nine mosaics together with additional 29 false color stream network and the landuse maps, derive dam related Landsat TM and MSS mosaics and Moderate Resolution Imag- storage parameters, and as a base map (Landsat GeoCover ing Spectrometer (MODIS) false color (bands 1,2,3 resam- Dataset 2000). (6) Nine mosaics were generated from satel- pled to 250 m) images acquired from 1998 to 2005 were lite imagery acquired around the same approximate time generated. Tens of individual temporal Landsat (MSS and period (July to September) over the marshes to investigate TM) scenes were examined over specific areas of interest 64 C. Jones et al.

(e.g., dams) to enable the extraction of reservoir storage found in higher altitudes. In contrast, adjacent subbasins parameters (refer to Section ‘‘Temporal satellite data’’). in the lowlands with similar physiographic characteristics All digital products were co-registered using the TM global were re-grouped into relatively larger subbasins. Improve- GeoCover dataset as a reference mosaic and brought to ments in the definition of stream networks and subbasin the same projection (UTM WGS84). delineation in areas experiencing gentle slopes (e.g., <1 cm/km in Mesopotamian plain) (Kolars and Mitchell, Digital elevation – watershed and stream delineation, 1991) were attained using a reference stream network ex- and reservoir storage tracted from the Landsat TM mosaics with high spatial res- The DTED mosaic covering the entire Tigris–Euphrates wa- olution (15–30 m) and applying a DEM ‘‘burning’’ method tershed was used to delineate the watershed boundaries (DiLuzio et al., 2004). The DTED data was also used to ex- and stream network using the Topographic Parameteriza- tract dam storage parameters (e.g., reservoir surface area tion (TOPAZ) program (Garbrecht and Martz, 1995). The and volume). Applying standard GIS techniques to DTED spatial resolution (3-arc-second; approximately 85 m) of data, we computed the emergency surface area, the area the DTED data had to be reduced to 540 m due to computa- enclosed by the contour having the surface elevation of tional restrictions. The TOPAZ method compares the eleva- the emergency spillway. We then extracted the volume of tion of each cell to that of the neighboring cells and water subtended between the delineated surface and the identifies the flow direction as being towards the cell with surface of the Earth portrayed in the DTED data, by sum- lowest elevation. The watershed was subdivided into 149 ming all intervals between these two surfaces again using subbasins, with sizes ranging from 200 to 10,000 km2. To ac- standard GIS applications (ESRI, 2001). count for the large disparities in physiogeographic characteristics of the examined watershed, a low Critical Source Climate datasets – temporal and spatial precipitation Area (CSA) value of 970 km2 was selected to define the sub- over watersheds basins across the watershed. Manual adjustments were then The climate datasets compiled for model construction in- applied to subdivide these into smaller subbasins in areas of clude the following data: (1) precipitation, temperature, varying physiogeographic characteristics, especially those relative humidity, and wind speed from 36 stations over

Figure 3 Distribution of landuse types and locations of climate data stations across the Tigris–Euphrates watershed. Also shown here are soil types and data (e.g., precipitation, relative humidity, temperature, solar radiation, and wind speed) from climate stations that were used as inputs to the SWAT model. Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands 65 the time period of 1964–1998, and (2) solar radiation from 4 done for the month of December. Inspection of available gauging stations for the time period 1964–1998. Inspection daily precipitation data for months other than March and of Fig. 3 shows that climatic gauges are unevenly distributed December did not reveal any obvious gradients and hence across the watershed. Fortunately, the stations are concen- approximations similar to ones reported above or applica- trated in areas where the majority of precipitation occurs in tions of alternative spline functions for extracting daily val- the upper reaches of the watershed on the highlands sur- ues from monthly averages were not pursued. Similar rounding the Mesopotamian plain. For modeling purposes, observations were noted for the remaining available daily the largest concentration of subbasins (119 subbasins from climatic datasets. Climatic data sets were imported to con- a total of 149 subbasins) was generated over the highlands strain spatial distribution of precipitation, to determine on- to increase the accuracy of the SWAT-derived spatial distri- set of snow pack growth and melting, and to compute butions of precipitation in this area. SWAT assigns a subba- evapotranspiration using the Penman–Monteith equation sin the precipitation value of the gauge closest to the (Monteith, 1981) and to constrain surface runoff and ground subbasin center. For small subbasins with multiple temper- water contributions. ature gauges, the gauge that was most representative of the overall elevations of the subbasin was kept and others were Landuse and soil type discarded. Continuous daily precipitation data (1969–1998) A digital landuse mosaic was constructed from individual was extracted from two sources: (1) global gauge data ex- landuse maps (CIA, 1993a–d) for Iraq, Turkey, Syria, and tracted from the Global Historical Climatology Network Iran. Six major landuse units were mapped throughout the (GHCN) (EarthInfo, 1996–2003) dataset, and (2) the histor- watershed and surrounding domains: ‘‘Arable lands’’, ‘‘Irri- ical global gridded precipitation (latitude cell: 2.5·longi- gated farm land’’, ‘‘Pasture’’, ‘‘Wetland’’, ‘‘Mixed For- tude cell: 3.75) data set (URL: gu23wld0098 dataset version est’’, and ‘‘Desert’’ (Fig. 3). The highlands surrounding 1, March 1999; Hulme, 1992, 1994; Hulme et al., 1998). The the Mesopotamian plain are largely composed of ‘‘Mixed latter data set was only used when rain gauge data was ab- Forest’’ and ‘‘Pasture’’ landuse types, whereas the remain- sent for a particular gauge. For the time periods where such ing lowlands are dominated by the ‘‘Desert’’ landuse type. data were absent, the precipitation values of the enclosing There were few noted exceptions including small pockets of cell were assigned to the enclosed gauge. ‘‘Arable lands’’, or ‘‘Irrigated farm land’’ in the River flood With very few exceptions (four stations, Fig. 3), available plains. Also the soil type ‘‘Irrigated farm land’’ is wide- climatic datasets are average monthly datasets (e.g., spread in the area bound by the Tigris and Euphrates rivers. monthly total precipitation, average monthly minimum Using the SWAT landuse library, we selected the landuse temperature, average monthly maximum temperature, types that closely resemble the mapped units. For example, average monthly solar radiation, and average monthly wind the mapped ‘‘Desert’’ landuse type was modeled as SWAT’s speeds) that were extracted from the GHCN global climatic ‘‘Desert Southwest Range’’. Similarly, the following data (EarthInfo, 1996–2003). We ran the model on daily mapped units: ‘‘Mixed Forest’’, ‘‘Arable lands’’, ‘‘Wet- steps to take advantage of the daily data where available lands’’, ‘‘Irrigated farm land’’, and ‘‘Pasture’’ were mod- and applied reasonable approximations to extract daily data eled as ‘‘Mixed forest’’, ‘‘Agricultural lands’’, ‘‘Mixed from monthly averages. Because no apparent trends were wetlands’’, ‘‘Agricultural row crops’’, and ‘‘Pasture’’, identified in the available daily data for the climatic param- respectively, in SWAT. The description for the selected eters, simple conversions were applied. Daily precipitation landuse types in the SWAT landuse library closely resembles was computed throughout the entire modeling period by the landuse classification produced for Iraq (CIA, 1993a–d). dividing total monthly precipitation by the number of days For example, the CIA landuse classifies ‘‘Arable Lands’’ as in each month. This is a reasonable approximation, given land cultivated for crops that are replanted after each har- the fact that the Tigris–Euphrates system is primarily driven vest (e.g. wheat, corn, and rice), whereas the description by contributions from melting snow (77% from melting for the equivalent unit, ‘‘Agricultural Lands’’ in SWAT, is snow) and the majority of the precipitation is stored as snow land that is annually harvested by plants such as wheat pack in the winter months and released during the snow and corn. Landuse type distributions are used within the melting season in the spring. Monthly average values for SWAT domain to estimate the spatial variations in evapora- wind speed, relative humidity, and solar radiation were tive demand and surface roughness coefficients using the treated as daily estimates. With two exceptions, the months Erosion Productivity Impact Calculator (EPIC) model incor- of December and March, the average monthly minimum and porated in SWAT (Williams and Sharply, 1989). maximum values were treated as daily minimum and maxi- A multi-step process was adopted to define soil types mum temperature values. across the watershed. The USDA global soil data set (FAO- Inspection of the few available daily temperature data- UNESCO, 2005) was downloaded from the Natural Resources sets (larger black cross, Fig. 3) indicated that the first two Conservation Service (NRCS) web site (www.nrcs.usda.gov). weeks in March are colder than the last two weeks. The Nine main soil types were identified in the watershed and opposite trend was observed for the month of December. surroundings: ‘‘rock’’, ‘‘sand’’, ‘‘vertisols’’, ‘‘aridisols’’, A first order approximation was applied to account for these ‘‘ultisols’’, ‘‘mollisols’’, ‘‘alfisols’’, ‘‘inceptisols’’, and temperature gradients. In computing the daily temperature ‘‘entisols’’ (Fig. 4). The alfisol, entisols, and vertisols soil minima and maxima from average monthly values for the types in the area are located along the river flood plain month of March, temperatures (minimums and maximums) and in the Mesopotamian plain; they are reworked river were lowered for the first two weeks and increased for channel deposits. Aridisols comprise much of the modeled the third and fourth weeks in ways that preserve the re- area, mollisols and ultisols occur in the highlands, and ported average monthly temperatures. The opposite was sands are located in the southern hyper-arid section of the 66 C. Jones et al.

Figure 4 Distribution of soil types derived from the NRCS database and water management projects (dams: black squares; flow barrages: pink squares) throughout the Tigris–Euphrates watershed. Also shown are locations of stream flow gauges (yellow circles) used for calibration (Keban, Hit, Mosul, and Baghdad) and verification (Hit and Baghdad). Green circles denote gauge locations that were used to define dam outflows upstream from the gauge locations. watershed (Fig. 4). For each of the nine soil types, and for tained in the soil and landuse coverages allowed for the every horizon within a particular soil type, particle size extraction of a CN distribution map for soil types across properties (percent clay; percent sand) were obtained using the entire watershed. USDA Soil Taxonomy classifications (NRCS and USDA, 1999). Particle size properties (USDA soil texture triangles; Rawls Stream flow data et al., 1992) were used to estimate the saturated hydraulic We compiled daily (5 stations) and monthly stream flow conductivities for each of the investigated horizons (Rawls data (6 stations) across the watershed (Fig. 4). With one and Brakensiek, 1983) and the estimated conductivities exception (Keban station in Turkey), all our stream flow sta- were then used to classify each of the soil horizons into tions are located in Iraq. Stream flow data served two pur- SCS soil hydrologic groups: Group A: >0.76 cm/h; Group B: poses: (1) to calibrate the model and (2) to constrain 0.38–0.76 cm/h; Group C: 0.13–0.38 cm/h, and Group D outflow for dams upstream whenever this data was unavail- soils: 0.0–0.13 cm/h (Rawls et al., 1992). The shapefiles able. Four stations were selected for calibration purposes; depicting distributions of soil hydrologic groups, together two stations (Keban and Hit) are located on the Euphrates with the generated landuse digital maps (also shapefiles), and the other two (Mosul and Baghdad) on the Tigris (Fig. 4). were imported into the model and used to define multiple Daily flow data sets were converted to monthly flow HRUs for each of the 149 subbasins. With the threshold lev- averages and used for model calibration and validation els of 20% and 10% for landuse and soil, respectively, recom- throughout the simulation period. Daily outflows are avail- mended by SWAT (DiLuzio et al., 2001), 1–7 HRU’s were able for each of the dams being modeled in Iraq such as defined for each subbasin, resulting in a total of 401 HRUs the Saddam, Mosul, and Dokan reservoirs but not for dams for the watershed. The integration of the information con- outside of Iraq such as the Keban reservoir in Turkey and Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands 67 the Tabqa reservoir in Syria (Fig. 4). Stream flow data from delineate the emergency spillway surface area and volume the nearest gauges downstream were treated in our model using DTED data and the known surface elevation of the as outflow values for the latter types of dams. For example, principal emergency spillway. stream flow data from the Husybah gauge was modeled as outflow for all dams located up gradient on the Euphrates. Similarly, the values from the Mosul gauge were used as in- Model calibration and validation puts for the upstream dams on the Tigris (Fig. 4). The model was calibrated (1964–1970) by adjusting model Temporal satellite data parameters of snowpack/snowmelt, soil, groundwater, and We generated temporal satellite mosaics over the marshes channel characteristics to achieve the best possible match to monitor the disintegration (1969–2002) and restoration to the stream flow data observed at two upstream (Mosul (2003–2006) of the marshes. False color images were gen- and Keban) and two downstream (Hit and Baghdad) erated from Landsat TM bands 2 (blue), 4 (green), 7 (red); (Fig. 4) stations on the Tigris and Euphrates Rivers, respec- Landsat MSS bands 1 (blue), 3 (green), 2 (red); and MODIS tively. The calibration was conducted for the time period bands 1 (blue), 2 (green), 3 (red). In generating the mosa- preceding the onset of the construction of the major dams ics, we minimized inter-scene spectral variations related across the watershed in the early seventies. An equilibration to differences in acquisition time and atmospheric contri- period of one year was adopted to get the hydrologic cycle butions. Digital numbers (DN) were converted to space fully operational (Neitsch et al., 2002). reflectance values to reduce spectral variations related Our examination of the temporal and spatial distribution to differences in sun angle elevation and scenes were nor- of precipitation, temperature and stream flow data indi- malized using areas of overlap between adjacent scenes cated that snowmelt is the dominant source for stream to the one with the least atmospheric effects (Sultan flows in the study area. Snowmelt-runoff is a relatively slow et al., 1992, 1993). In generating these false color images, and gradual hydrologic process, compared to the rainfall– we selected the band combinations that discriminate veg- runoff process. Model calibration, thus, was conducted via etation from surrounding units taking advantage of one or a stepwise, iterative process by adjusting the following: more of the following: chlorophyll absorption feature in snow pack/melt parameters, followed by key soil/ground- the 0.6 lm wavelength region (MSS, TM, and MODIS bands water parameters, and finally channel routing parameters. 2, 2, and 1, respectively), the relatively high reflectance The calibration was conducted against the observed of vegetation in the 0.69–0.83 lm wavelength region as monthly flow hydrographs for the upstream gauges followed compared to soils (TM band 4 and MODIS band 2), and by the downstream gauging stations. Following the calibra- the relatively low reflectance of vegetation in the 0.4– tion period, dams were constructed on the Euphrates and 0.6 lm wavelength region compared to the surrounding Tigris rivers. The calibrated model was validated (1971– soil. Given the selected band and color schemes, vegeta- 1998) by comparing the simulated stream flow rates to the tion appears in shades of green on all mosaics, water as observed flow rates at the downstream stations (Hit and dark bodies, and soil in shades of other colors depending Baghdad). on their composition and texture (e.g., grain size) (Sultan et al., 1987). On these mosaics, the marshlands were Snow pack/melt parameters mapped as contiguous areas of shades of green and black. Several main snow pack/melt related parameters were ad- The earliest multi-spectral satellite data utilized in this justed to calibrate the simulated flow rates against ob- study is a Landsat MSS 1973 scene. For time periods pre- served flow rates at the Mosul and Keban gauging stations. ceding 1973, only panchromatic satellite images are avail- These two stations are proximal to the mountainous area able (CORONA satellite data; McDonald, 1995). Because and receive the majority of the snow melt from the high- these images are acquired in the visible (0.4–0.7 lm) lands further north. Keban is located in the Taurus Moun- wavelength region, the vegetation and the water bodies tains, whereas Mosul is located near the base of the that define the areal extent of the marshes appear as rel- Zagros Mountains. The calibration of these two stations is atively dark areas compared to their surroundings. essential for the calibration of the whole system. The SWAT The digital mosaics were used to delineate the temporal model classifies precipitation as snow or rain depending on variations in marshland extent, and to extract reservoir the snow fall temperature (SFTMP). For example, if the storage parameters. These parameters are required for SFTMP was set to 2 C and the maximum daily and minimum modeling reservoir storage and routing in SWAT. The princi- daily temperatures were 10 C and À10 C, respectively, the pal surface area for each dam was determined by delineat- calculated average daily temperature (0 C) will be less than ing the average areal extent of the lake that is being the SFTMP and precipitation will be modeled by SWAT as impounded by the dam using as many satellite images avail- snow. The seasonal growth and recession of the snow pack able following the filling of the lake. We applied standard is defined using an areal depletion curve (Anderson, 1976). GIS techniques to DTED data to compute the average areal The latter is determined by the threshold snow water con- extent of the lake, the area enclosed by the contour having tent (SNOCOVMX) that corresponds to 100% snow cover the average surface elevation of the examined satellite and a specified fraction of SNOCOVMX at 50% snow coverage images. We then extracted the volume of water subtended (SNO50COV). The areal depletion curve affects snowmelt if between the delineated surface and the surface of the Earth the snow water content is below SNOCOVMX. In addition to portrayed in the DTED data, by summing all intervals the areal coverage of snow, snowmelt is also controlled between these two surfaces again using standard GIS by the snow pack temperature, which is influenced by a lag- applications (ESRI, 2001). Similar procedures were used to ging factor (TIMP), threshold temperature for snow melt 68 C. Jones et al.

(SMTMP), and melting rate estimated by maximum and min- default value, default range, and the source for SWAT’s imum snowmelt factors (SMFMX and SMFMN). The adjust- defaults. An iterative process was applied over the calibra- ments for the seven parameters (SFTMP, SNOCOVMX, tion period to provide a coarse fitting for the main peaks SNO50COV, TIMP, SMTMP, SMFMX, and SMFMN) were made that develop in May and June. The characteristics of the heuristically. Seven parameters related to snow pack/melt simulated hydrographs were found to be more sensitive to were adjusted one at a time following a general order from the SNOCOVMX, SNO50COV, SFTMP, SMTMP parameters. a relatively sensitive to less sensitive parameters. The adjustment were performed through trial-and-error within Soil and ground water parameters a reasonable range for each parameter and the entire pro- The second step in the calibration process aimed at cess was manually repeated a few times until a general achieving a finer fitting for the higher and lower flows agreement between simulated and observed monthly hydro- by adjusting soil and groundwater parameters. The im- graphs was reached. pacts of these adjustments were more pronounced on SWAT’s default values were adopted as our initial param- the lower gauges since they are largely controlled by eter values and the implemented adjustments were con- the overall groundwater system, whereas the upper strained by the ranges of parameter variation provided by gauges are more impacted by the runoff system. The SWAT. The reported SWAT default and range values were groundwater parameters include soil available water based on results from previous studies (Anderson, 1976; Hu- capacity (SOL_AWC), soil evaporation compensation coeffi- ber and Dickinson, 1988; Sangrey et al., 1984; Smedema and cient (ESCO), threshold water levels in shallow aquifer for Roycroft, 1983; Lane, 1983; and Chow, 1959). Table 2 lists base flow (GWQMN) and for re-evaporation/deep aquifer for each of the seven snow parameters (this section) and percolation (REVAPMN), re-evaporation coefficient other parameters listed in Sections ‘‘Soil and ground water (GW_REVAP), deep percolation fraction (RCHRG_DP), de- parameters’’, ‘‘Channel characteristics’’, and ‘‘Calibration lay time for groundwater recharge (GW_DELAY), and base- criteria and model evaluation’’ the following: parameter flow recession constant (ALPHA_BF). These parameters definition, adopted parameter value, SWAT’s parameter dictate the amount as well as the timing of water flow

Table 2 Calibration parameters aParameter SWAT default range (value) Final value Deﬁnition SFTMP (À)5–5 (1.0)a 2.0 Snowfall temperature (C) SNOCOVMX 0–500 (1.0)b 150.0 Minimum snow water content that corresponds to 100% snow

cover, SNO100 (mm H2O) SNO50COV 0–1 (0.5)a 0.58 Fraction of snow volume represented by SNOCOVMX that corresponds to 50% snow cover TIMP 0–1 (1.0)a 0.01 Snow pack temperature lag factor SMTMP (À)5–5 (0.5)a 5.0 Snow melt base temperature (C) c SMFMX 0–10 (4.5) 4.5 Melt factor for snow on June 21 (mm H2O/ C day) c SMFMN 0–10 (4.5) 0.0 Melt factor for snow on December 21 (mm H2O/ C day) SOL_AWC Variesa Varies (0.01–0.50) Available water capacity of the soil layer (mm/mm soil) ESCO 0–1 (0.95)a 0.1 Soil evaporation compensation factor GWQMN 0–5000 (0)a 2000 (Aridisol) Threshold depth of water in the shallow aquifer required for

return ﬂow to occur (mm H2O) REVAPMN 0–500 (1.0)a 400.0 Threshold depth of water in the shallow aquifer for ‘‘revap’’

to occur (mm H2O) GW_REVAP 0.2–1.0 (0.2)a 0.2 Groundwater ‘‘revap’’ coefficient GW_DELAY 0–500 (31)d Varies (20–65) Groundwater delay time (days) ALPHA_BF 0–1 (.048)e Varies (0.1–0.9) Baseflow alpha factor (days) RCHRG_DP 0–1 (0.05)a 0.05 Deep aquifer percolation fraction CH_K1 0–150 (0.50)f 50.0 Effective hydraulic conductivity in tributary channel alluvium (mm/h) CH_K2 0–150 (0.0)f 0.0 Effective hydraulic conductivity in main channel alluvium (mm/h) CH_N1 0–0.3 (0.014)g Varies (0.08–0.12) Manning’s ‘‘n’’ value for the tributary channels CH_N2 0–0.3 (0.014)g Varies (0.035–0.05) Manning’s ‘‘n’’ value for the main channel a Neitsch et al. (2002) b Anderson (1976). c Huber and Dickinson (1988). d Sangrey et al. (1984). e Smedema and Roycroft (1983). f Lane (1983). g Chow (1959). Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands 69 through the soil zone and underlying aquifer to the stream Calibration criteria and model evaluation channel. Fine-tuning calibration was performed manually by A classification scheme for GW_DELAY response was tied repeating the 3-step adjustment described above until a to the topographic expressions within the watershed. Soils best fit was achieved between modeled and observed flow within subbasins in the highlands were assigned lower delays values. The values for the parameters that achieved the factors (20–35) compared to areas in the plains downstream optimum calibration are given in Table 2; all of the se- (40–65). The baseflow recession constant (ALPHA_BF) and lected parameter values are within the reported SWAT de- the groundwater delay (GW_DELAY) were found to be the fault ranges. Two statistical measures, coefficient of major parameters affecting the shape of the high flows; determination (r2) and coefficient of efficiency (E)(Nash these parameter and others listed above (e.g., GWQMN, and Sutcliffe, 1970) were used to quantify the achieved REVAPMN) were optimized manually through trial-and-er- levels of calibration and to evaluate the overall perfor- ror, whereas default values were adopted for a number of mance of the model. the less sensitive parameters in our model (e.g., RCHRG_DP, Fig. 5a and b compares the simulated flow rates (blue GW_REVAP) (Table 2). curve) to the observed rates (pink curve) for the Keban, Manual calibration methods were performed instead of Mosul, Hit, and Baghdad gauges over the calibration period automated methods due to the vast differences in physical (1964–1970). Visual inspection of Fig. 5a shows a general characteristics experienced throughout the modeled region. agreement between simulated and observed monthly flow Heuristic calibration scenarios for rainfall–runoff models rates throughout the calibration period. The observed min- for such complex systems are difficult to implement using or deviations between simulated and observed flow rates automated methods. For example, and as described earlier, on the Euphrates could be in part attributed to uncertain- for each of the major physiographic areas (e.g., mountains, ties introduced by the adopted approximations converting lowlands), each parameter had to be calibrated separately average monthly climatic data into daily values. The more to achieve the best results. Some 400 runs were completed significant deviations observed on the Tigris River are prob- to optimize the model fit. As is the case with the previous ably largely related to uncertainties introduced by the pau- calibration step, this step focused on calibrations of simu- city of climatic stations in the highlands compared to the lated hydrographs against observations from the upper Euphrates system. Only two climatic stations are located gauging stations. The adjustment for all key parameters in the northeast highlands of the Tigris subbasin. The sta- was repeated until the fitting of the monthly hydrographs tions are located at extreme elevations (>4000 m), was significantly improved compared to fittings achieved whereas the remaining highlands that are of relatively low- in the previous step. er elevations (1200 m) lack any climatic stations. This sit- uation could factiously lower the modeled temperatures Channel characteristics and overestimate the modeled snow accumulation at the The third step of the calibration process extended the fit- expense of runoff. This could explain why our simulated ting process to simulate the observed hydrographs at the flow on the Tigris Mosul station is lower than the observed two downstream gauging stations with main consideration flow during the winter months (Fig. 5a). The model given to parameters related to channel routing (CH_N). In characterizes some of the precipitation as snow and under- the model, a variable storage routing method (a variation estimates observed runoff. Adjustment of SFTMP was of the kinematic wave model) developed by Williams found to improve the simulated hydrograph on the Tigris (1969) was adopted for channel routing calculations. The River but the simulation for the Euphrates River was key parameter, the manning roughness coefficient (n), was compromised. initially assigned for all HRUs using the SWAT default value The final calibration was achieved on the basis of the (0.014). Adjustments were then made guided by classifica- overall model performance. The coefficient of determina- tions described in Chow (1959) to improve the time to peaks tion (r2) for simulated flows ranges from 0.6 to 0.8 and and peak flows for the two lower gauging stations. The fol- the coefficient of efficiency (E) exceeds 0.6. Compared to lowing coefficients were adopted: mountainous area 0.05; the other three stations, the relatively larger differences foothills 0.04; and vegetated channel plain 0.035. between modeled and observed values at the Baghdad Transmission losses through channels were calculated in gauge are largely related to the absence of quantitative SWAT for two different channel types: tributary and main estimates for the amount of water diverted during peak flow channels. The key parameter for losses is the effective seasons from the Tigris River into surrounding natural hydraulic conductivity (CH_K). The main channels in the wa- depressions. For example, the Sammarra Barrage System tershed receive continuous groundwater, and thus the was constructed (1954) to divert water from the Tigris River transmission losses for these channels are minimal. A zero into the Thartar Depression to protect Baghdad against transmission loss for the main channels was simulated by flooding (Manley and Robson, 1994; UNEP, 2001). In absence setting effective hydraulic conductivity (CH_K2) to zero. of such data, we could not simulate the diversion of these Most tributary channels are ephemeral or intermittent waters resulting in higher modeled simulated peak flows receiving little groundwater and are thus likely to lose water compared to observed flows. to bank storage or to the underlying aquifer. These channels Fig. 6 provides a comparison between the simulated were assigned the initial SWAT default hydraulic conductiv- monthly flow rates and the observed rates at the Hit and ity (CH_K1) values (0.5 mm/h) and adjustments (adopted Baghdad gauging stations for the time period 1971 through value: 50.0 mm/h) were conducted within reported ranges 1998, the validation period. The model simulations corre- (Lane, 1983) to improve hydrograph fitting for all four gaug- late reasonably well with the observed flow rates; the ing stations. coefficient of determination (r2) and coefficient of 70 C. Jones et al.

a 4000 0 5000 0

2000 4000 2000 3000 Keban Gauge Mosul Gauge 4000 3000 4000

2000 6000 2000 6000

8000 /s) 3 1000 1000 8000 10000 0 10000 Cumulative Snow Melt (mm) Cumulative Snow Melt (mm) 0 12000 5/3/1964 5/3/1965 5/3/1966 5/3/1967 5/3/1968 5/3/1969 5/3/1964 5/3/1965 5/3/1966 5/3/1967 5/3/1968 5/3/1969 Simulated Flow Observed Flow Snow Melt 7000 6000 Monthly Flow Rate (m 6000 5000 Hit Gauge 5000 Baghdad Gauge 4000 4000 3000 3000

2000 2000

1000 1000

0 0 5/3/1964 5/3/1965 5/3/1966 5/3/1967 5/3/1968 5/3/1969 5/3/1964 5/3/1965 5/3/1966 5/3/1967 5/3/1968 5/3/1969 Time b Euphrates River Tigris River 4000 4000 KebanKeban Gauge Gauge MosulMosul Gauge Gauge 3000 3000

2000 2000 /s) 3

1000 2 1000 R = 0.8053 R2 = 0.7161 0 0 0 1000 2000 3000 4000 0 1000 2000 3000 4000

5000 6000 Baghdad Gauge 4000 HitHit Guage Gauge 5000 Baghdad Gauge 4000 3000 3000 2000 2000 Simulated Monthly Flow Rate (m 2 1000 R = 0.7636 1000 R2 = 0.6041 0 0 0 2000 4000 6000 0 1000 2000 3000

Observed Monthly Flow Rate (m3/s)

Figure 5 (a) Calibration results comparing the simulated monthly flow rates (blue curve) to observed flow rates (pink curve) for the upstream gauges (Keban and Mosul) and downstream gauges (Hit and Baghdad) throughout the calibration period (1964–1970). Also shown is the simulated cumulative snow melt (green curve) from all subbasins contributing to flow at Keban and Mosul gauges. (b) Observed versus simulated flow rates at the Keban, Hit, Mosul, and Baghdad gauges throughout the calibration period. Red lines show the 1:1 correlation.

efficiency (E) throughout the validation period are 0.62 years indicating that climatic variations are not likely to and 0.6, respectively. The figure shows a general trend have caused the observed general trend of decreasing of decreasing flow rate with time. Inspection of precipita- flow rate with time. Next we show that the reduced flow tion data over the watershed shows that this trend is not is primarily caused by the major engineering projects on related to any variations in precipitation over the past 40 the Tigris–Euphrates System. Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands 71

70007000 800800 60006000 Hit Gauge 600 50005000 40004000 400 3000

3000 Precipitation (mm) /s) 3 2000 2000 200200 10001000 00 0 5/3/1964 5/3/1968 5/3/1972 5/3/19725/3/1976 5/3/19765/3/1980 5/3/19805/3/1984 5/3/19845/3/1988 5/3/19885/3/1992 5/3/19925/3/1996 5/3/1996 60006000 800800 50005000 Baghdad Gauge 600 40004000 Monthly Flow Rate (m Monthly Flow Rate 30003000 400 20002000 200 10001000

0 0 5/3/19645/3/1964 5/3/1968 5/3/1968 5/3/1972 5/3/19725/3/1976 5/3/19765/3/1980 5/3/19805/3/1984 5/3/19845/3/1988 5/3/19885/3/1992 5/3/19925/3/1996 5/3/1996 Time Time Observed Flow Simulated Flow No Data Precipitation Observed Flow Simulated Flow No Data Precipitation

Figure 6 Calibration and validation results for the downstream gauges (Hit and Baghdad). Simulated monthly ﬂow rates (blue curve) were calibrated (1964–1970) and validated (1971–1998) against observed ﬂow rates (pink curve). Also shown are the variations in total annual precipitation over the entire watershed (red curve).

Discussion reduction in annual flow (Fig. 7; blue line) throughout the calibration and validation periods. Inspection of Fig. 7 dem- The calibrated model was used to investigate the impacts onstrates two ways by which the dams reduce the flow to of the construction of dams inside and outside Iraq on the the marshlands. The first is a one-time reduction that is size and vitality of the marshes. We addressed this issue by associated with the impoundment of water behind the new- examining the changes in the general patterns and magni- ly constructed dams forming large artificial lakes with stor- tude of simulated inflow of the Tigris and Euphrates into age capacities ranging from a few billion cubic meters to the marshlands, changes that are associated with the con- tens of billions of cubic meters. The second is a permanent struction of the major dams. The simulated inflows were reduction in flow due to loss of impounded water through computed at a location near the city of Al-Basrah. It is evaporation, infiltration and irrigation. noteworthy that the nearest available stream flow data is The construction of each of these large dams led to the upstream from Al-Basrah, some 560 km from the Hit sta- development of large water impoundments behind the dams tion on Euphrates and 350 km from the Baghdad station and the abstraction of large volumes of water that would on Tigris. The simulated flow for the time period (10/1/ have otherwise reached the marshlands. For the period pre- 1965 to 09/30/1973) that preceded the construction of ceding the construction of large dams, a sharp decrease in the major dams (SC: >3 · 109 m3), hereafter referred to flow is observed on the percent reduction curve mimicking as period A, and the flow for the time periods, B1: 10/1/ the filling of the reservoir. For example, for the period 1973 to 09/30/1989, and B2: 10/1/1989 to 09/30/1998 1970–1973 preceding the completion of the Keban and Tab- that witnessed the construction of the major dams are qa dams in 1974, the simulated AAFV was reduced by up to shown in Fig. 7. The Keban, Tabqa, Haditha, and Karakaya 30%. Similarly, during the filling of the Ataturk reservoir, the dams on the Euphrates, and the Mosul and Hamrin dams AAFV was reduced by yet another 31% from its 1988 values (total SC: 76 · 109 m3) on the Tigris were constructed in (Fig. 7) over the course of three years (1989–1992). The period B1, whereas the Ataturk (SC: 50 · 109 m3) was con- construction of the major dams represented in Fig. 7 led structed in period B2. The AMPFR (6301 m3/s) in period A is to an abstraction of over 75 · 109 m3 over the course of progressively diminished in periods B1 (AMPFR: 3073 m3/s) the calibration and validation periods. If we include the and B2 (AMPFR: 2319 m3/s) due to the combined effects of smaller dams (<3 · 109 m3), the one time extraction would evaporation and infiltration of the water impounded be- have reached over 100 · 109 m3. hind dams and the utilization of stored water for irrigation Inspection of the percent reduction in AFV shows that purposes. following the lake-filling period, (1–3 years on the average), The impacts of dam related reduced flow on the vitality flow volumes increase, but generally to levels less than pre- of the marshes was further investigated by comparing the dam levels. For example, following the filling of the Keban simulated inflow from Tigris and Euphrates (Fig. 7; black and Tabqa reservoirs, the simulated AFV bounced back to line) to the inflow that would have occurred if no dams were approximately 85% of its original volume in 1969 and follow- constructed (Fig. 7; pink line). The simulated flow under ing the filling period (1989–1993) of the Ataturk reservoir, these two scenarios was then used to compute percent the AFV reached 70% of the 1969 values. It is these 72 C. Jones et al.

A B1 B2 14000 -10 Haditha (8.2) Keban (31) ? C 0 Tabqa (11.7) Hamrin (3.95) Karakaya (9.58) /s) 12000 3 Mosul (11.1) ? ? 10 Ataturk (48.7) 10000 20 30 8000 40 50 6000 60 (Dams /NoDams) 4000 70 80

Average Monthly Flow Rate (m 2000 90 Percent Reduction Annual Flow Volume 0 100 1/1/65 1/1/69 1/1/73 1/1/77 1/1/81 1/1/85 1/1/89 1/1/93 1/1/97 Time Flow With Dams Flow Without Dams Percent Reduction in Flow

Figure 7 Simulated monthly flow rates from 1965 to 1998 (black curve) are compared to flow rates that would have existed if no dams were constructed (pink curve). Also shown are the percent reductions in AFV that resulted from the construction of dams (blue curve). Three major time periods are represented on the figure: time period A (1965–1973) refers to the time where no major dams were yet constructed; time period B1 (1973–1989) refers to the time span where the Keban, Tabqa, Hamrin, Haditha, Mosul, and Karakaya dams were constructed, and time B2 (1989–1998) refers to the period where the Ataturk dam was completed. The pink and black curves overlap in period A, the time period where no major dams were constructed, but progressively diverge in periods B1 and B2 as more dams are being constructed. Reported AFV values were calculated for each year starting in October 1st and ending in September 30th in the following year. permanent reductions in flow volumes that pose, we be- MSS and Landsat 5 scenes acquired in either July, August, lieve, the largest threat to the livelihood of the marshes. or September of 1975, 1976, 1977, 1984, 1985, 1986, and The percent reduction in AFV for period B1 ranged from 1987. Analysis of these images showed that the areal ex- 12% to 43% and averaged 21%, for period B2 it ranged from tent of the marshes was reduced (1966: 7970 km2; 1977: 26% to 48% and averaged 35%. 6680 km2; and 1984: 5270 km2) in response to the progres- Next we demonstrate that the areal extent of the marsh- sive construction of dams upstream (Fig. 8). Unfortunately, lands is directly related to the reduction of the inflow from satellite imagery over the examined area, season, and the Tigris and Euphrates. The higher the flow, the larger the time period are limited; to the best of our knowledge marsh area becomes and vice versa. We restrict our analysis there are only nine coverages between 1966 and 1987, to the time period 1966–1987 (before period C, Fig. 7), the the ones reported here. time period preceding the construction of the large engi- Spatial variations were primarily observed along the wes- neering projects (Fig. 7) within and around the marshes to tern and northern margins and to a much lesser extent along separate the spatial variations in the marshlands related the eastern and southern boundaries, where the marshes to changes in inflow from those related to implementation are generally bound by steeper topographic boundaries of engineering projects. We also make our comparisons in (e.g., cliffs). Because most of the flow in the Tigris and marshland extent using scenes acquired around the same Euphrates is generated in the snow melting season, it could approximate time period of the year, the months of July be demonstrated to a first order that the areal extent of the through September, to avoid complications arising from investigated marshes decreases with the decline in AFV marshland spatial variations related to seasonal flow (Fig. 9). Naturally, one should not expect a 1:1 relationship variations. because other factors such as local topography, average We investigated the effects of reduced inflow arising annual temperature, amount of evaporation and infiltration from the construction of the Keban and Tabqa dams on can influence the size of the marshland as well. Neverthe- the Central and Al-Hammar marshes. These marshes are less, we could use this relationship and apply the same fed by the Euphrates and Tigris Rivers and we analyzed procedures to extract similar relationships for marshes variations in the marshes’ size by comparing the areal ex- other than the Central and Al-Hammar marshes within the tent of theses marshes before and after the construction Tigris–Euphrates system. Such relationships could be used of these dams. The areal extent of the marshes for the to achieve a first order understanding of what would be time period preceding the dams was extracted from a mo- the impact of additional reductions in flow that will result saic of CORONA KH4-B images acquired in September, from the implementation of the planned engineering pro- 1966, and the areal extent of the marshes for the period jects on the Tigris–Euphrates system over the next few following the construction of the dams was extracted from years. Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands 73

needed to irrigate a hectare of land in this area (Gollehon and Quinby, 2006), we estimate an additional water allocation of approximately 5.3 · 109 m3/yr will be required to irrigate the new farmlands. This increase in irrigated farm- land will further reduce the inﬂow of the Tigris–Euphrates system into the marshes. At a minimum, the AFV will be reduced by more than 5 · 109 m3/yr not to mention losses to evaporation and inﬁltration in reservoirs behind newly constructed dams. Fig. 9 shows that such a reduction in the average AFV will result in an additional decline in the area of the Central and Al-Hammar marshes amounting to approximately 550 km2.

Summary

We constructed a catchment-based continuous (1964–1998) rainfall–runoff model for the entire Tigris–Euphrates watershed (area: 1,000,000 km2) using the SWAT model to understand the natural flow system, and to investigate the impacts of reduced overall flow and the related land cover Figure 8 Spatial variations in the areal extent of the Central and landuse change downstream in the marshes. Precipita- and Al-Hammar marshlands extracted from temporal (1966– tion was derived from daily and monthly rain gauge data 1984) satellite data. The variations correspond to progressive (36 stations) from 1964 to 1990 and from the historical glo- reduction in AFV caused by engineering projects (e.g., Keban, bal gridded precipitation dataset (1990–1998). Six landuse Tabqa, Hamrin, and Haditha dams) outside of the marshes. The types were derived from landuse map sheets and nine soil maximum extent (September, 1966; black line) of the marshes types were derived from global soil map coverages. The spa- under natural flow conditions was reduced (August, 1977; blue tial extent of drainage basins, reservoir volume, and marsh- line) following the post filling of the Keban and Tabqa dams and land volume were extracted from DTED data. Evaporation was further decreased (July, 1984; yellow line) following the on bare soils and ET on vegetated canopy was calculated construction of additional dams (Hamrin and Haditha). using the Penman–Monteith method. A SCS curve number method was adopted to estimate direct runoff using rainfall

70 excess and channel routing was conducted using the variable storage method. The model was successfully calibrated

) 60 3 against stream ﬂow data in Iraq (3 gauges) and Turkey m

9 (1 gauge). 50 The calibrated model was used to simulate the AMFR, AMPFR, and AFV of the Tigris and Euphrates into the 40 marshes at a location near Al-Basrah city for the following 30 time periods: time period A (10/1/1965 to 09/30/1973), the time period that preceded the construction of the major 20 dams; and time periods B1 (10/1/1973 to 09/30/1989) and B2 (10/1/1989 to 09/30/1998), the time periods that wit-

Annual Flow Volume (10 10 2 R = 0.730 nessed the construction of the major dams. The model 3 9 0 shows that the AMPFR (6301 m /s) and AAFV (80 · 10 3 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 m /yr) in period A is progressively diminished in periods 3 9 3 Area: Al-Hammar and Central Marshes (km2) B1 (AMPFR: 3073 m /s; AAFV: 55 · 10 m /yr) and B2 (AMP- FR: 2319; AAFV: 50 · 109 m3/yr) due to the combined ef- Figure 9 Plot of spatial extent of the marshlands extracted fects of evaporation and infiltration of the water from temporal satellite images acquired in September 1966, impounded behind dams and the utilization of stored water July 1975, August 1976, August 1977, July 1984, August 1985, for irrigation purposes. July 1986, September 1986, and August 1987 against the We demonstrated that for the time period preceding simulated AFV for these nine time periods. the construction of the major engineering projects in and around the investigated marshes, the areal extent of the Central and Al-Hammar marshes extracted from temporal The ongoing South Eastern Anatolia project calls for an satellite images acquired around the same approximate increase in irrigated area of 1,700,000 hectares by 2010 time period of the year (months of July to September) de- (SPO, 1990). This will be achieved by the completion of a creased (e.g., marsh area 1966: 7970 km2; 1977: 6680 km2; series of dams and canal systems to channel the reservoir 1984: 5270 km2) with reductions in AFV (e.g., AFV 1966: water to new irrigated land. At present, nine of the 22 dams 60.8 · 109 m3; 1977: 56.9 · 109 m3; 1984: 37.6 · 109 m3). A in the project have been completed. The new dams under relationship that describes the impact of reduced flow on construction or planned equate to a combined additional the areal extent of the examined marshes was then used storage of 15.5 · 109 m3. Assuming 3100 m3 of water are to provide first order simulations of what would be the 74 C. Jones et al. impact of additional reductions in flow that will result Huber, W.C., Dickinson, R.E., 1988. Storm Water Management from the implementation of the planned engineering pro- Model, version 4: User’s Manual, Athens, GA. jects on the Tigris–Euphrates system over the next few Hulme, M., 1992. A 1951–80 global land precipitation climatology years. The ongoing South Eastern Anatolia project calls for the evaluation of general circulation models. Climate for an increase in irrigated area of 1,700,000 hectares Dynamics 7, 57–72. Hulme, M., 1994. Validation of large-scale precipitation fields in by 2010. Upon completion of the project, we estimate that 9 3 general circulation models. In: Desbois, M., Desalmand, F. an additional water allocation of 5.3 · 10 m /yr will be (Eds.), Global Precipitations and Climate Change. Springer- needed. This together with additional losses to evapora- Verlag, Berlin, pp. 387–406. tion and infiltration behind dams will reduce the inflow Hulme, M., Osborn, T.J., Johns, T.C., 1998. Precipitation sensi- of the Tigris–Euphrates system into the marshes by at tivity to global warming: comparison of observations with least 5 · 109 m3/yr. This reduction in the AFV will result HadCM2 simulations. Geophysical Research Letters 25, 3379– in an additional decline in the area of the Central and 3382. Al-Hammar marshes amounting to a minimum of an addi- IMET, 2004. The New Eden Project: Final Report, Italian Ministry for tional 550 km2. the Environment and Territory. United Nations CSD-12 New York, April 14–30, 2004. Jones, C.K., Sultan, M., Al-Dousari, A., Salih, S.A., Becker, R., Milewski, A., 2005. Who did What to the Mesopotamian Marsh- Acknowledgements lands? Inferences from Temporal Satellite Data. Geological Society of America Annual Meeting Abstracts with Programs, 37(7), paper 231-6. Funding was provided by the National Science Foundation Kolars, J.F., Mitchell, W.A., 1991. The Euphrates River and the through OISE Grant 0455896. We thank the reviewers of Southeast Anatolia Development Project. Southern Illinois Uni- the Journal of Hydrology for their constructive comments. versity Press. Lane, L.J., 1983. Chapter 19: Transmission Losses, SCS – National Engineering Handbook, Section 4: Hydrology. US Government References Printing Office, Washington, DC, pp. 19.1–19.21. Manley, R., Robson, J., 1994. Hydrology of the Mesopotamian Anderson, E.A., 1976. A point energy and mass balance model of Marshlands, Unpublished Report. Wetland Ecosystems Research snow cover. NOAA Technical Report NWS 19, US Department of Group, University of Exeter. Commerce. McDonald, R.A., 1995. Corona – success for space reconnaissance, Anderson, J.R., Hardy, E.E., Roach, J.T., Witmer, R.E., 1976. A a look into the cold-war, and a revolution for intelligence. land use and land cover classification system for use with remote Photogrammetric Engineering and Remote Sensing 61 (6), 689– sensor data. US Geological Survey Professional Paper 964, 28pp. 720. Arnold, J.G., Allen, P., Bernhardt, G., 1993. A comprehensive Monteith, J.L., 1981. Evaporation and surface temperature. Quar- surface-groundwater flow model. Journal of Hydrology 142, 47– terly Journal of Royal Meteorological Society 10 (451), 1–27. 69. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through Arnold, J.G., Srinisvan, R., Muttiah, R.S., Williams, J.R., 1998. conceptual models, Parts I-a discussion of principles. Journal of Large area hydrologic modeling and assessment, part I: model Hydrology 10, 282–290. development. Journal of American Water Resource Association Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Srinivasan, R., Williams, 34 (1), 73–89. J.R., 2002. Soil and Water Assessment Tool: User Manual, Basgall, M., 2003. DUWC Joins Iraqi Marsh Restoration Project. Version 2000. Grassland, Soil and Water Research Laboratory, Wetland Wire 6 (2), 1–3. Temple, TX. Chow, V.T., 1959. Open Channel Hydraulics. McGraw-Hill Inc., New Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2005. Soil York. and Water Assessment Tool Theoretical Documentation. Grass- CIA, 1993a. Iran Land Use. World Factbook. land, Soil and Water Research Laboratory, Temple, TX. CIA, 1993b. Iraq Land Use. World Factbook. NIMA, 1990. CD: DTED 134. National Imagery Mapping Agency. CIA, 1993c. Syria Land Use. World Factbook. NIMA, 1991a. CD: DTED 133. National Imagery Mapping Agency. CIA, 1993d. Turkey Land Use. World Factbook. NIMA, 1991b. CD: DTED 147. National Imagery Mapping Agency. DiLuzio, M., Srinivasan, R., Arnold, J.G., Neitsch, S.L., 2001. NIMA, 1991c. CD: DTED 148. National Imagery Mapping Agency. Arcview Interface for SWAT2000 User’s Guide. US Department of Rawls, W.J., Brakensiek, D.L., 1983. A procedure to Predict Green Agriculture – Agricultural Research Service, Temple, TX. Ampt infiltration parameters. Advances in Infiltration, American DiLuzio, M., Srinivasan, R., Arnold, J.G., 2004. A GIS-coupled Society Agriculture Engineering, 102–112. hydrological model system for the watershed assessment of Rawls, W.J., Ahuja, L.R., Brakensiek, D.L., Shirmohammadi, A., agricultural nonpoint and point sources of pollution. Transac- 1992. Infiltration and soil water movement. In: Maidment, D.R. tions in GIS 8 (1), 113–136. (Ed.), Handbook of Hydrology. McGraw-Hill Inc., pp. 5.1–5.5. EarthInfo, 1996–2003. GHCN Global Climate. EarthInfo Inc. Sangrey, D.A., Harrop-Williams, K.O., Klaiber, J.A., 1984. Predict- ESRI, 2001. Using ArcGIS Spatial Analyst, 232 pp. ing groundwater response to precipitation. ASCE Journal of FAO-UNESCO, 2005. Soil map of the world. In: ESRI (Ed.), Soil Geotechnical Engineering 110 (7), 957–975. Climate Map. USDA-NRCS, Soil Survey Division, World Soil SCS, 1972. Section 4: Hydrology, National Engineering Handbook. US Resources, Washington, DC. Department of Agriculture, Soil Conservation Service, Engineer- Garbrecht, J., Martz, L., 1995. TOPAZ: an automated digital ing Division, Washington, DC, USA. landscape analysis tool for topographic evaluation, drainage SCS, 1985. Section 4: Hydrology, National Engineering Handbook. US identification, watershed segmentation, and sub-catchment Department of Agriculture, Soil Conservation Service, Engineer- parameterization: overview. Agricultural Research Service ing Division, Washington, DC, USA. (NAWQL) 95 (1). Smedema, L.K., Roycroft, D.W., 1983. Land Drainage – Planning Gollehon, N., Quinby, W., 2006. Irrigation Resources and Water and Design of Agricultural Drainage Systems. Cornell University Costs, AREI. USDA. Press, Ithaca, New York. Hydrologic impacts of engineering projects on the Tigris–Euphrates system and its marshlands 75

SPO, 1990. The Southeastern Anatolia Project Master Plan from correlations of arabian and nubian neoproterozoic out- Study, Final Master Plan Report, vol. 2, second ed., June crops. Tectonics 12 (6), 1303–1319. 1990. Tucker, C.J., Grant, D.M., Dykstra, J.D., 2004. NASA’s global Srinivasan, R., Ramanarayanan, T.S., Arnold, J.G., Bednarz, S.T., orthorectified landsat data set. Photogrammetric Engineering 1998. Large area hydrologic modelling and assessment. Part II: and Remote Sensing 70 (3), 313–322. model application. Journal of American Water Resource Associ- UNEP, 2001. The Mesopotamian Marshlands: Demise of an Ecosys- ation 34 (1), 91–101. tem, Early Warning and Assessment Report, UNEP/DEWA/TR.01- Sultan, M., Arvidson, R.E., Sturchio, N.C., Guinness, E.A., 1987. 3 Rev.1, Division of Early Warning and Assessment, United Lithologic mapping in arid regions with Landsat thematic mapper Nations Environmental Programme, Nairobi, Kenya. data – Meatiq Dome, Egypt. Geological Society of America USDA, 1999. Soil Taxonomy: A Basic System of Soil Classification for Bulletin 99 (6), 748–762. Making and Interpreting Soil Surveys Report, Washington, DC. Sultan, M., Becker, R., Arvidson, R.E., Shore, P., Stern, R.J., El Williams, J.R., 1969. Flood routing with variable travel time or Alfy, Z., Guinness, E.A., 1992. Nature of the Red-Sea crust – a variable storage coefficients. Transactions 12 (1), 100–103. controversy revisited. Geology 20 (7), 593–596. Williams, J.R., Sharply, A.N., 1989. EPIC – Erosion/Productivity Sultan, M., Becker, R., Arvidson, R.E., Shore, P., Stern, R.J., El Impact Calculator: 1. Model Documentation. USDA Technical Alfy, Z., Attia, R.I., 1993. New constraints on Red-Sea rifting Bulletin No. 1768.