2.1. Study Area
Istanbul is located in the north-west of Turkey within the Marmara region, intersecting two continents, namely Asia and Europe. The surface area of Istanbul is 5540 km
2. Forest territories cover 44% of the city, mainly in the northern part. A large majority of the settlements (with a 2700 capita/km
2 population density) are concentrated on the south coast of the city (
Figure 1). The city has a transitional climate, impacted by the Black Sea to the north, and the Marmara Sea and the Aegean Sea to the south. The northern parts of the city, where forested areas mostly lie, is affected by northerly colder air masses of maritime and continental origins, whereas the southern part shows the general characteristics of the Mediterranean climate [
31]. The average temperature in winter months is between 2 °C and 9 °C, and in summer months between 18 °C and 28 °C. The city receives about 815 mm of precipitation per year as a long-term average, according to the recorded stations in the city [
32].
Istanbul has a population of nearly 15 million people, and it is expected that this number will grow to 21 million by 2050 [
16,
18]. In parallel with the increase in population, daily water consumption will grow due to changes in lifestyle, income level and eating habits. Today, gross water demand in the city is estimated to be 175 L/capita-day, and this figure is expected to reach 225 L/capita-day by 2050 including industrial usage and NRW.
Figure 2 shows the historical population changes and water demand of Istanbul.
As of today, 15 drinking-water reservoirs operate to meet the demand for potable water in Istanbul. To meet the water demand of the Istanbul metropolitan area, ISKI had to expand its service area beyond the city border (
Figure 3). Currently, ISKI is responsible for the management and the protection of the water resources located in different administrative regions to supply drinking water to the Istanbul metropolitan area. There are several ongoing projects to increase the potential water capacity and protect catchments for the future. Six of the main drinking-water reservoirs and their watersheds are within the city border, namely, Terkos, Buyukcekmece, Alibey, Sazlıdere, Omerli, Elmalı and Darlık; and the rest, the Kazandere and Papucdere Reservoirs, Istranca Creeks and the Melen System, are in neighboring cities (
Figure 3). Istanbul has an unbalanced distribution in terms of its water resources and population between the Asian and European sides. In numbers, the Asian side has 77% of the water resources (including the Melen System) while it hosts 35% of the population (
Table 1).
There are plans for the Melen watershed, located in the western part of the Black Sea Region and 180 km to the east of Istanbul (
Figure 3), to provide water to Istanbul in the medium and long term [
33]. In order to convey water from the Asian side to the European side, a 6-m diameter and 5551-m long Bosphorous tunnel was constructed. The tunnel goes 135 m below sea level, crossing the two continents, with a capacity to transfer 3 million m
3 of water daily [
23].
In order to investigate the water-resources availability of Istanbul, we studied the current watersheds of the city as well as the surrounding potential catchments of Istanbul. This area is located between 40.3 to 42.1 north latitude and 27.1 to 31.7 east longitude, which includes the area next to Istanbul, the Black Sea coast of Trakya Region (Istranca Sub-region), Kapıdag Peninsula, Izmit Bay, the Sapanca and Iznik Lake watersheds, downstream of Sakarya River, and the Melen watershed in the Western Black Sea Basin in Turkey (
Figure 4). The total study area is around 20,790 km
2. Although there are more flow stations maintained by the State Hydraulic Works (DSI) in the region, 25 stations were found to be suitable over the area for this study. Among these, 12 gauge stations are within the current watersheds of Istanbul, and 13 are located in the remaining parts of the study area (
Figure 4). Thus, we used these hydrometric stations for regionalization of the hydrological model parameters in the ungauged catchments.
2.2. SWAT Model
SWAT is a hydrological model developed by the US Department of Agriculture (USDA) Agricultural Research Service [
34,
35]. It is a continuous-time, semi-distributed, process-based model, developed to evaluate alternative management strategies on water resources and non-point source pollution in large river basins [
36].
Water balance is the driving force behind all the processes in SWAT because it impacts plant growth and the movement of sediments, nutrients, pesticides and pathogens [
36]. In SWAT, a watershed is divided into multiple sub-basins, which are then further subdivided into hydrologic response units (HRUs) based on unique combinations of land use, soil, management and topographical features. The model simulates hydrology of a watershed in two phases. In the first phase, called the land phase, the hydrological processes of a watershed are simulated at the HRU level and water balance calculated for each sub-basin. The pathways of water movement in the land phase simulated by SWAT are given as canopy storage, surface runoff, evapotranspiration, infiltration, lateral sub-surface flow, return flow, revap from shallow aquifers, and percolation to the deep aquifer. In the second phase (the routing phase), after the loadings of water, sediment, nutrients and pesticides are determined, and loadings are routed through streams and reservoirs within the watershed [
37]. A schematic representation of hydrological cycle elements simulated by SWAT is given in
Figure 5.
More details and model equations can be found in the SWAT technical documentation (
http://swatmodel.tamu.edu) and in Arnold et al. [
34]. A general overview of SWAT model use, calibration and validation is discussed by Arnold et al. [
36], and historical development, applications, and future research directions are summarized in Gasmann et al. [
38] and Douglas-Mankin et al. [
35].
2.3. Model Inputs and Setup
The SWAT model requires a land-use map, climate data, soil map, and topography. Due to the lack of local data to build a model, data required for this study were compiled from global datasets. River discharges and water consumption rates were obtained from local administrations (
Table 2).
The soil map was produced by the Food and Agriculture Organization/United Nations Educational, Scientific and Cultural Organization (FAO–UNESCO) global soil map [
39], which provides data for 5000 soil types (65 for Turkey) comprising two layers (0–30 cm and 30–100 cm depth) at a spatial resolution of 10 km. The land-use map was obtained from the CORINE 2000 Land Cover datasets (
http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-2000-raster-3) at a resolution of 100. A digital elevation model (DEM) was constructed from the Shuttle Radar Topography Mission (SRTM) database at a 90-m spatial resolution (
http://srtm.csi.cgiar.org/). Three different climate database sources were available for the region: (1) measured data collected from the State Meteorological Service (MGM) in 17 temperature and rainfall climate stations with <15% missing data for the period 1960–2013; (2) gridded data constructed from Climate Research Units (CRU) with a 0.5° resolution for the period 1970–2007, and 0.25° gridded data from Climate Forecast System Reanalysis (CFSR) for the period 1979–2014, amounting to 48 and 103 grid points, respectively.
We used the ArcSWAT 2012 interface to set up the model. Despite using a high-resolution DEM in the model, in order to avoid the discrepancies, particularly during the stream network delineation, we used the burn-in feature of ArcSWAT with river data obtained from the Google Earth software. Also, to delineate coastal catchment areas more accurately, a threshold drainage area of 100 ha was chosen. Inland sub-basin outlets were manually added to represent reservoirs, gauge stations, main river channels, and other topographical features in the watershed; while coastal outlets were created automatically by the software based on the given threshold. As a result, the study area was configured with 1335 sub-basins, which were further discretized into 3315 HRUs. The model could not represent 4% of the real area as a result of the precision of basin delineation near the coastal zones. However, this missing area does not affect principal objectives of the study. The total simulated area in the current model is 19,960 km2.
Except for two stream gauges, none of the gauges are affected by the reservoir operations. Therefore, we did not use the reservoir operation rule in the current model. The available river discharge data in the region varies between 10 years and 32 years. Five elevation bands in each sub-basin were established to adjust for orographic change in temperature (−6.5 °C km
−1) and rainfall (100 mm km
−1). Potential evapotranspiration (PET) is simulated using Hargreaves method [
40], actual evapotranspiration (ET) is estimated based on the methodology of Ritchie [
41], and surface runoff is calculated by the Soil Conservation Service (SCS) curve number procedure [
42].
According to the acquired data (
Table 2), the model was simulated from 1977 to 2013 (37 years), and the first 3 years were used as a warm-up period to allow the processes simulated to reach a dynamic equilibrium and decrease the uncertainty of the initial conditions of the model. The simulation includes both dry and wet years occurring in the historical period.
Figure 6 depicts the yearly cumulative precipitation in Istanbul between the years 1977 and 2014, including both drought and wet periods.
Due to the availability of more than one climate database, we evaluated three different datasets as a preliminary analysis of the model. According to the preliminary model results during the model set-up, CFSR outperformed the local dataset and CRU captured better the streamflow dynamics and also the total rainfall distribution over the study area. Most of the watershed area evaluated in this study are protected zones without any settlements [
24], thus land-use changes are negligible.
For model calibration, validation and uncertainty analysis we used the SUFI-2 algorithm [
43,
44] in the SWAT-CUP software. All uncertainties in the model, such as a parameter, measured data (e.g., stream flow), driving variables (e.g., rainfall), and conceptual framework, are expressed as a set of parameter ranges by SUFI-2. The algorithm tries to capture most of the measured data within the 95% band of prediction uncertainty (95PPU). The uncertainty (95PPU) is quantified at the 2.5% and 97.5% levels of the cumulative frequency distribution of an output variable obtained using the Latin hypercube sampling technique. Two indices are used to measure the goodness of calibration/uncertainty performance, the P-factor (ranges 0 to 1), which is the percentage of data captured by 95PPU band, and the R-factor (ranges 0 to ∞), which is the average thickness of the uncertainty band divided by the standard deviation of the related measured variable [
43,
44]. These two indices are used to judge the strength of the calibration procedure where a value of >0.7 for the P-factor and a value of around 1 for the R-factor would be satisfactory, depending on the study [
11,
12]. More information about the algorithm is given by Abbaspour et al. [
11,
44]. SUFI-2 allows the use of different objective functions such as R
2, RSR or Nash–Sutcliffe efficiency (NSE) [
45]. Despite the fact that we used NSE as an objective function, percent bias (PBIAS) [
46] and R
2 of the calibration/validation results were also evaluated as well as the P-factor and the R-factor in order to assess the model performance and model output uncertainty.