Introduction

Coastal plains in arid and semi-arid climate zones constitute favorable sites for greenhouse farming that can be sustained during the entire year. Such plains have a mild winter climate and receive more precipitation compared to inland plains. In general, the groundwater is the main source of irrigation water in such plains because of the uneven distribution of annual precipitation and streamflow. On the other hand, global population growth increases the demand for additional potable water and food, and human activities escalate the pollution risk of available water resources. Hence, groundwater depletion (e.g. Konikow and Kendy 2005) and pollution are two major problems of water and food security in the future. Furthermore, climate change poses another threat that is growing steadily. Different climate models anticipate different spatiotemporal changes for temperature, precipitation, and evapotranspiration. But, all models agree with the fact that increasing temperatures will cause appreciable variations in precipitation, evaporation, and evapotranspiration regimes. For instance, Thomas and Famiglietti (2019) suggest that climate-induced decrease in recharge would have a greater influence on the global groundwater depletion as compared to groundwater pumping. Sea-level rise due to the melting of the glaciers by increasing temperature poses another threat for the coastal aquifers. Projecting future sea-level change is a challenging issue because of the complexity of the earth’s climate. However, a conservative estimate is that a temperature rise of 1 °C could cause a sea-level rise of 1.15 mm/year over the next 2,000 years (Sweet et al. 2017). Under these circumstances, there is a need for more efficient groundwater use policies that account for the effects of climate change.

Numerical groundwater flow models allow simulation of the response of natural systems under different stress conditions. Once based on a reliable conceptual model of the hydrogeologic system, these models can be used for testing various hypotheses regarding future climate change and water use scenarios. Sustainable groundwater management studies utilizing numerical groundwater flow models have been used for many purposes. For instance, Wang et al (2008); Sanz et al. (2011); Gilfedder et al. (2012) investigated the effects of land use and anthropogenic activities on groundwater management. El Idrysy and DeSmedt (2006); Alwathaf and El Mansouri (2012); Mc Callum et al. (2013); Hogeboom et al. (2015); Tian et al. (2015) developed numerical groundwater flow models for semi-arid and arid climate conditions where intensive irrigation water supplied from groundwater. Pusatli et al. (2009) presented a groundwater-dependent irrigation water management plan for the Küçük Menderes Basin in Turkey. In recent years, studies emphasized also on climate change (e.g. Green et al. 2011; Jackson et al. 2011; Taylor et al. 2013; Arnell and Llyod-Hughes 2014; Garrote et al. 2015; Meixner et al. 2016; Joseph et al. 2018; Salem et al. 2018) and the effects on aquifer recharge (Crosbie et al. 2013; Moeck et al. 2016). Studies like Green and MacQuarrie (2014) investigated the potential effects of seawater intrusion in coastal aquifers under changing climate conditions.

Demre Plain, which is the site of this study, is renowned for the vast greenhouse and citriculture fields where yearlong agriculture is possible due to mild winter conditions. Groundwater, supplied from drilling wells and caisson wells, is the sole source of potable and irrigation water in the plain. In the following, first, we present the data used in the hydrogeological conceptual model and then develop a numerical groundwater flow model. The model is used to assess the spatiotemporal hydraulic head variations in response to inputs from the RCP 4.5 to RCP 8.5 climate change scenarios for the period between 2015 and 2050. The methodology used here can be applied in similar sites elsewhere in the world.

Materials and methods

Location climate, morphology, and geology

Demre is a typical plain, located along the coast of the Mediterranean Sea in SW Turkey. The plain is nested in the Taurus Mountains Range and bounded by the Mediterranean Sea from the southeast. Demre town (Myra of Roman Empire period) is famous also as being the place where St Nicholas (or Santa Claus), a 4th-century Bishop, lived. The 24 km2-large plain is located at the terminus of a 960 km2-large drainage basin. Average elevations of Demre Plain and its drainage area are 10 m a.s.l. (above sea level) and around 1000 m a.s.l., respectively. Mediterranean climate with hot and dry summers and warm and wet winters dominates in the plain. According to the Köppen-Geiger climate classification, this climate is considered to be Csa class (e.g. Geiger 1954). The average annual temperature is 18.4 °C in the 1981–2014 period and the average annual precipitation in the 1969–2014 period is 790.5 mm. The basin is drained by Demre Stream into the Mediterranean Sea. The stream flows only after intensive winter–spring precipitations and/or during the spring snow-melt period at the upper reaches of the basin. There is no flowrate observation along the stream. Floods up to 650 m3/s occur with a frequency of 5–10 years (Avcı 2015). Demre plain has been formed after the Last Glacial Maximum (LGM) which terminated about 20,000 years before the present. During the peak of Quaternary glaciations at around the LGM, the global sea level dropped 130 m below the current sea level (e.g. Spratt and Lisiecki 2016). Since LGM, melting ice waters fill in the oceans, and the global sea-level rise to the current level about 8,000 years ago (e.g. Fleming et al. 1998; Atalay 2005). This process resulted in the accumulation of alluvium brought by Demre Stream into a gulf surrounded by the Mesozoic carbonate rocks which are karstified extensively. Today, the alluvial fill constitutes a gently sloping plain extending from the sea coast to about 25 m a.s.l. where the plain meets the limestone cliff (see Fig. 1). The elevation of the carbonate rock mountains surrounding the plain ranges from sea level nearby the plain to more than 2000 m at the northern end of the basin.

Fig. 1
figure 1

Geology and land use map of the study area and its vicinity

Full size image

The alluvium comprises sediments ranging in size from clay to cobble. Coarse material is observed along the streambed whereas sand, silt, and clay-sized materials are more common in the plain. The thicknesses of alluvium range from about 100 m at the north to about 200 m along the coastline. The carbonate rock units exposed around the plain are fractured and karstified extensively (Şenel et al. 1981).

Land use

The agricultural production pattern in the plain determines the amount of groundwater used for irrigation. A historical analysis of the land use has been realized using a chronologic series of air photographs and satellite images taken from the 1980s to 2014. The plants grown in the plain comprise vegetables and citrus (i.e. lemon, orange, and grapefruit). Vegetables and citrus are grown in greenhouses and orchards, respectively. The greenhouses are visible in the images as dirty white rectangular fields whereas the orchards are recognized as dotted lines of trees. Historical images have been digitized to determine the magnitude of greenhouse and orchard areas. Data show that the area of the greenhouses increased from 6.9% to 80.2% from the 1980s to 2010s due to the increasing market price of the greenhouse products. Accordingly, the citriculture farming areas decreased from 93.1% to 19.8% in the same period (Avcı et.al. 2014). Analyses of the images taken since the early 2000s revealed that the ratio of the greenhouse to citriculture fields unchanged during the last two decades (see Fig. 1).

Hydrogeology

The alluvium in the plain is a prolific aquifer in which the groundwater can be accessed everywhere (see Fig. 2). The carbonate rocks (i.e. the karst aquifer) surrounding the plain underlie the alluvium and the wells drilled in this unit, provide groundwater only if they cut fracture zones and dissolution conduits. Massive blocks of carbonate rocks do not yield a satisfactory amount of groundwater. All wells in the plain drilled in the alluvium. Transmissivity (T) of alluvium aquifer varies from 100 m2/day to 4000 m2/day depending on the grain size distribution of the sediments around the well (DSİ 1978). No information exists for the transmissivity of the karst aquifer. However, studies nearby the plain (e.g. Bayari et al. 2011) revealed substantial groundwater discharge from the karst aquifer toward the sea and coastal plains. Recharge of aquifers is supplied by precipitation while the alluvium aquifer receives also a lateral groundwater recharge from the karst aquifer. Discharge of alluvium aquifer occurs via evapotranspiration, abstraction from wells, and groundwater outflow to the Mediterranean Sea.

Fig. 2
figure 2

Hydrogeology map and observed hydraulic head distribution in 2014

Full size image

Two irrigation cooperatives operate 24 drilling wells to provide irrigation and potable water. The wells have been drilled to depths from 20 m to 100 m from the surface of the alluvium aquifer. These wells have been drilled in the northern part of the plain to minimize the risk of seawater intrusion due to over-pumping. Their discharge rates vary between 40 L/s and 90 L/s. In addition to the drilling wells, there are about 200 private caisson wells with depths ranging from 10 to 30 m, dug into the alluvium. They are located mostly in the southern part of the plain. There are four brackish karst springs located on the alluvium-limestone contact at the western and eastern parts of the plain. The location of these springs marks roughly the sites of coastal discharges of karst aquifer before the sea-level rise after LGM.

Water chemistry and environmental isotope data reveal that (a) these springs discharge deep-circulating karst groundwater and groundwater contacts with seawater at depths of karst aquifer before reaching the surface, (b) the shallow groundwater along the coast chemically enriched due to evaporation, and (c) none of the groundwater samples are affected by seawater intrusion (Avcı 2015). According to major ions composition, 3 different water groups were identified in the plain; Group 1 represents springs of deep-circulating karst groundwater in contact with seawater, Group 2 represents fresh groundwater in alluvium belonging to calcium–magnesium carbonate (Ca/MgCO3) facies and, Group 3 is the partly evaporated shallow groundwater along the coast, belonging to calcium–magnesium and chloride–bicarbonate (Ca/Mg-Cl−/HCO3−) facies (Avcı 2015).

The general direction of groundwater flow in the alluvium aquifer is from northwest to southeast where the Mediterranean Sea is located. Groundwater head decreases from 5 m a.s.l. in parts of the plain unaffected by pumping to about 0.5 m a.s.l. near the coast. Lower head values occur in parts of the plain where the pumping wells are clustered. The average hydraulic gradient in the alluvium aquifer is 1.3 * 10–3.

Temporal variation of hydraulic head

Temporal groundwater level measurements in the plain have been conducted in the pumping wells numbered as 19001, 19002, 19004, 19005, 19006, 26722, 26723, 26724, and 26725 (see Fig. 2 for locations) between 1983 and 1999. Measurements performed manually except for the well 19004 which has been equipped with a mechanical level recorder. Groundwater level measurements have been conducted monthly between 1984 and 1992. Later on, measurements have been taken in April and October until 1999. Thereafter, irregular monthly manual measurements were taken in 2014 and 2015. The groundwater level in the wells 19004 and 19006 was monitored also by electronic data loggers in 2014 (e.g. Avcı et al. 2019). Characteristics of the wells used for groundwater level monitoring are given in Table 1. According to monthly groundwater level observations in the 1983–1999 period, the highest and lowest groundwater levels in the plain were 6.8 m a.s.l. (well 19001, February 1985) and − 2.6 m a.s.l. (well 19002, March 1984), respectively. In the 2014–2015 observation period, the highest and lowest groundwater levels were measured as 5.2 m a.s.l. (well 26722, February 2015) and − 1.4 m a.s.l. (well 19004, May 2014), respectively (see Fig. 3). Two additional groundwater level monitoring wells (numbered 63068 and 63070) have been drilled in 2014 to observe the groundwater level variation outside the pumping fields. Comparison of long-term groundwater levels and precipitation showed that the rainfall rises rapidly the groundwater level whereas the pumping is the main reason for local groundwater level decline. The groundwater levels measured in the pumping wells have been affected by the abstractions either in the same well or in the nearby wells. However, the groundwater level in the plain seems to be in a steady state during the 30 years-long observation period (see Fig. 3). The long-term stability of groundwater level in the plain is attributed to the lateral recharge from the karst aquifer which surrounds the alluvium aquifer from east, north, and west. The karst aquifer is about 1,000 km2 large and drains the mountains that extend above 2000 m a.s.l. at the north forms an effective flux boundary.

Table 1 Some properties of the wells used for groundwater level observations (after Avcı 2015)
Full size table
Fig. 3
figure 3

Temporal variation of precipitation and groundwater level in the alluvium aquifer

Full size image

Groundwater budget

Data used in groundwater budget calculations have been compiled from literature and the database of local organizations. Precipitation and temperature data collected in Demre meteorological station in the 1969–2014 period have been used to calculate the recharge from precipitation and discharge by evapotranspiration. Recharge components of the alluvium aquifer include the recharge from precipitation and the lateral recharge from the karst aquifer. Discharge components include evapotranspiration, surface run-off, abstraction for irrigation and domestic water use, and the groundwater outflow to the Mediterranean Sea. It is assumed that there is no long-term change in groundwater storage based on the fact that the annual hydraulic head distribution in the aquifer has not changed for over three decades. The following equations have been used to calculate the groundwater budget of the alluvium aquifer.

$$\Delta S\left( {\text{Change in groundwater storage}} \right) = {\text{Recharge}} - {\text{Discharge}}$$
(1)

An explicit form of Eq. 1 is as follows:

$$\Delta S = \left( {P + R_{k} } \right) - \left( {\text{ET + SR + IW + PW + GOF}} \right)$$
(2)

where ΔS is the change in groundwater storage; P is recharge from precipitation; Rk is recharge from karst aquifer; ET is discharge by evapotranspiration; SR is discharge by surface run-off; IW is irrigation water use; PW is domestic (potable and household) water use and GOF is the groundwater outflow into the Mediterranean Sea.

According to observations of the Demre meteorological station in the period 1969–2014, the average annual evapotranspiration was calculated as 561.4 mm/year. Evapotranspiration was calculated with the Turc method (Turc 1954). There is no stream gauging station in the basin of Demre Stream, Therefore, surface run-off was calculated using the Soil Conservation Service Curve Number method (SCS-CN) (Soulis and Valiantzas 2012). Groundwater is the sole source of irrigation water but no record has been kept on how much water is obtained from each drilling and caisson well. For that reason, the annual volume of abstracted groundwater was estimated from the irrigation and domestic water use. Total plant water demand was determined as 13.7 Mcm (million cubic meters) by the Blaney–Criddle method (Blaney and Criddle 1962) based on the plant pattern. Potable and household water need in the plain was calculated as 1.3 Mcm/year using a typical water consumption of 137 L/day per person for the total population of 25,700 people (Antalya Province Environmental Status Report 2010). The population in the plan has not changed in the long term. The amount of groundwater outflow into the sea was calculated as 4.6 Mcm/year based on average transmissibility, hydraulic gradient, and coastline length values of 2000 m2/day (DSİ 2000), 1.3 * 10–3 and 4950 m, respectively (Avcı 2015). Lateral recharge from the karst aquifer was estimated as 16.2 Mcm based on the assumption of no long-term change in the groundwater storage. The average groundwater budget components of the alluvium aquifer in the period of 1969–2014 are given in Table 2. Groundwater budget components were calculated also for the years 1980, 1990, 2004, and 2009 to determine the effect of land-use changes on the groundwater budget during the last 35 years (see Table 2).

Table 2 Groundwater budget components of Demre coastal aquifer for different years
Full size table

Results show that the temporal variation of the amount of lateral recharge from the karst aquifer is driven by the change in local hydraulic gradient. Almost all pumping wells are located nearby the karst aquifer and more lateral groundwater flow (i.e. recharge) is supplied from the karst aquifer when the hydraulic gradient between alluvium and karst aquifers is increased due to pumping. In years when recharge from other budget components are low and the irrigation water demand is high, more water is supplied to the alluvium aquifer from the karst aquifer. The groundwater budgets calculated for the years of 1980, 1990, 2004, and 2009 show that the amount of irrigation water use is unchanged after 2004 because of the fixed plant pattern (or the agricultural land use).

Numerical flow model

The MODFLOW-2000 (Harbaugh et al. 2000) numerical groundwater flow model coupled to GMS 10.0 pre/post processor interface software (http://www.aquaveo.com/ software/gms-groundwater) has been used in modeling studies. GMS is capable of creating a conceptual model in a GIS (Geographic Information System) environment so that the generated data are transferred easily to the numerical model domain (e.g. Gogu et al. 2001). Therefore, the “conceptual model method” of GMS has been used for modeling the alluvium aquifer in Demre plain. The model run in a transient mode for annual time steps in which the recharge and discharge components varied depending on the changing climate parameters. The numeric models include 35 time steps (stress periods) for 1980–2014 and 2015 to 2050 periods.

Boundary conditions and grid mesh

The conceptual model was defined to a numerical model as a single layer and the flow domain was divided into 150 m by 150 m by 100 m grid cells. The total model area is about 26 km2 and the alluvium aquifer was represented by 1155 active cells. All cells within the model boundaries are active. Bottom elevations of all active cells were defined as 100 m below sea level. Surface topographic elevation values were added to the model as 2D scatter points (see Fig. 4).

Fig. 4
figure 4

Numerical groundwater flow domain showing the boundary conditions, locations of drilling wells, caisson wells, and the ground surface elevation contours

Full size image

Environmental isotope and hydrochemistry data do not indicate a seawater intrusion anywhere in the flow domain (Avcı 2015). Therefore, the aquifer boundary with the Mediterranean Sea and the Beymelek Lagoon at the east (see Fig. 1) was defined as a specified head boundary (CHD = 0). Sea and lagoon levels were assumed not to change during the modeling period because current studies (e.g. Group WGSLB 2018) infer a negligible sea-level rise during the modeling period which ends in 2050. The boundary between the karst aquifer and alluvium aquifer was defined as a general head boundary (GHB) to allow for lateral groundwater flow from the karst aquifer.

Recharge and discharge

Model area, well and river coverages were created to define recharge and discharge parameters of the numerical flow model. Recharge and maximum evapotranspiration rate parameters were defined in the model. To define the evaporation to MODFLOW simulation; maximum evaporation rate, evaporation surface elevation, and the evaporation extinction depth definitions are required. The evaporation extinction depth was defined as 2 m for all cells and the evaporation surface was set equal to the topographical elevation of each cell. The evapotranspiration was calculated with the Turc method (Turc 1954) and defined to the model for each year (i.e. stress period) of the modeling period of 1980–2014. Run-off was calculated by SCS (Soulis and Valiantzas 2012) method and the net recharge was defined to the model.

To describe the abstraction from deep and shallow wells for irrigation water, each well was described with its bottom and top elevation values. The amount of abstraction from each well was described by the pumping rate. Since the exact construction dates of the caisson wells are not known, the year 1990, when greenhouse farming started to become widespread, was accepted as the first year of their operation period. Annual groundwater abstraction rates were determined from the annual irrigation water demands in the period of 1980–2014.

River package was used to simulate the recharge from Demre Stream. Conductance of riverbed (m2/y), river base elevation (thalweg, m), and stage of river data were defined in the model. Conductance (m2/y) which is defined as hydraulic conductivity (k, m/y) divided by river length (L, m) and multiplied by seepage area (A, m2) was calculated as 3394 m2/year. The highest water elevation of the stream was defined as 0.5 m above the streambed elevation.

Geohydrological parameters

The geo-hydrological parameters required by the model include porosity (n, %) and hydraulic conductivity (k, m/year). The hydraulic conductivity of alluvium was determined as 52 m/day based on the pumping tests (DSİ 2000). Thus, a hydraulic conductivity value of 19,000 m/year was assigned to the alluvium. The porosity of alluvium was accepted as 30% based on the values for silty gravel-coarse sand in the literature (e.g. Freeze and Cherry 1979). Similarly, the specific yield and specific retention were defined as 25% and 5%, respectively.

Calibration and validation

The groundwater flow model is calibrated with the observed hydraulic heads in pumping wells between 1983 and 1999 (see Fig. 3). Measurements have been taken at times when the pump stopped but, the observed data have likely been affected by the drawdown created in the same well and/or in the nearby pumping wells. All pumping wells have been operated for varying time lengths irregularly during the irrigation period between February and October. Frequently, groundwater has been supplied from another well if a pump nearby the irrigated plot fails. Therefore, the exact amount of water abstracted from a well is not known precisely. Under these conditions, the calibration strategy is based on the difference between the calculated and observed minimum groundwater level in each year of the calibration period. Figure 5 shows the average, maximum, and minimum difference between the calculated and observed minimum groundwater levels for each well during the calibration period. The values in boxes along the solid line shows the average difference between the calculated and observed groundwater levels in each well during the calibration period. The difference varies from − 0.2 m a.s.l. to + 1.4 m a.s.l. The vertical bars show the 1 sigma uncertainty and the maximum and minimum difference values are marked along the dashed lines that extend above and below to the average line, respectively.

Fig. 5
figure 5

Statistics of the difference between the calculated and observed minimum groundwater levels in each pumping well between 1983 and 1999

Full size image

The mean of the average, minimum and maximum difference between the calculated and observed groundwater levels among all wells during the calibration period are 0.6 m a.s.l., − 0.4 m a.s.l. and 1.5 m (a.s.l.), respectively. The 1 sigma standard deviation of the mean is ± 0.6 m a.s.l. A better match between the observed and calculated groundwater levels could be obtained by modifying the geohydrologic parameter values (i.e. hydraulic conductivity, porosity) as in the case of the inverse modeling approach. However, this was not attempted because the tuned values would be representative only for a small number of model cells where most of the groundwater pumping wells are clustered (see Fig. 2). Results of the model based on modified K and n values indicated that the groundwater head distribution is controlled mainly by the GHB condition along the boundary with the karst aquifer and the specified head boundary along the coastline (CHD = 0 m a.s.l.). Consequently, it was decided to keep the model as simple as possible so that a match between observed and calculated values with ± 2 m tolerance is assumed acceptable. This tolerance limit is based on the mean short-term level fluctuation observed in the pumping wells when the pump is turned off before the groundwater level measurement.

Groundwater level observations of 17 wells in 2014 were used for the validation of the numerical model (see Fig. 6). Continuous level monitoring performed in wells numbered 63068, 63070, 19004, and 19006 using electronic groundwater level loggers in 2014. The wells 63068, 63070, and 19004 were used only for groundwater monitoring whereas 19006 was used also for groundwater abstraction. The rest of the wells have been used for irrigation. Eight of them, indicated by pumping well numbers > 30000 have not been used in the calibration stage. The observed and calculated levels in the monitoring wells fall on the 1:1 line. The lowest and highest difference between observed and calculated groundwater levels are 0.0 m and 0.76 m, respectively. The average difference between observed and calculated groundwater levels for all of the 17 wells is 0.25 m (1 sigma standard deviation is 0.38 m). It has been concluded that the numerical flow model simulates the aquifer’s behavior reasonably well because the mean difference of 0.28 m between the observed and calculated levels remains below the ± 2.0 m assumed error margin of the observations.

Fig. 6
figure 6

Observed and calculated groundwater levels of 2014 used for model validation

Full size image

A simple sensitivity test was performed to assess the effect of the uncertainties in boundary conditions using modified values of general head boundary (GHB) between the alluvium and karst aquifers, rainfall (P) and evapotranspiration (ET). The values of GHB (3 m a.s.l.) P (0.795 m/year) and ET (0.561 m/year) representing the year 2014 (i.e. Scenario #1) are arbitrarily increased and decreased 15% (Scenario #2: GHB = 3.45 m a.s.l., P = 0.990 m/year, ET = 0.610 m/year and Scenario #3: GHB = 2.55 m a.s.l., P = 0.671 m/year, ET = 0.527 m/year). Air temperature is assumed unchanged.

Results of sensitivity calculations for Scenario #1, Scenario #2 and Scenario #3 revealed average calculated groundwater levels (and 1 sigma standard deviation) of 2.69 m a.s.l. (± 0.80), 2.44 m a.s.l. (± 0.70), 2.99 m a.s.l. (± 0.80) and 2.06 m a.s.l. (± 0.64), respectively. The average calculated groundwater levels are similar to the respective averages of the observed values within the range of corresponding standard deviations.

Results

Water budget of the numerical flow model

The numerical model ran for every year during the 35 years between 1980 and 2014. The spatial distributions of hydraulic head calculated for the years 1981 and 2014 are shown in Fig. 7. These years represent the time steps when the minimum and maximum needs for the irrigation water occurred because the annual precipitations in 1981 and 2014 were 1205 mm and 650 mm, respectively. Besides that, the change in agricultural land use from the orchard-dominant type in the 1980s to the greenhouse-dominant type in the 2010s increased the irrigation water demand. Eventually, more groundwater was abstracted in 2014 compared to 1981 and the plain-wide groundwater levels in 2014 were lower than those in 1981. In both cases, groundwater levels along the coastline approximate the sea level. Along the shoreline, the ground surface is very close to mean sea level. Recharge from precipitation evaporates rapidly so that the groundwater level becomes equal to sea level. This process is dominant particularly around Beymelek Lagoon at the southeast and around the sand groin at the southwest ends of the shoreline. According to the model results, the groundwater level decreased about 1 m throughout the plain during the last 35 years. The flow budget calculated by the model reveals that the aquifer volume reduced from 35.9 Mcm in 1980 to 28.1 Mcm in 2014.

Fig. 7
figure 7

Calculated hydraulic head distribution of Demre coastal aquifer for the years 1981 and 2014

Full size image

The numerical flow model produced results that agree with the hypothesis of the strong lateral recharge from the karst aquifer to the alluvium aquifer as envisaged by the conceptual model. The annual amount of recharge supplied by the karst aquifer is inversely proportional to the recharge from precipitation. In other words, more groundwater is abstracted from the alluvium aquifer if the recharge from precipitation is relatively low and the amount of abstraction is replenished by the recharge from the karst aquifer. For example, about 11 Mcm groundwater was recharged from the karst aquifer in 1981 (annual precipitation 1205 mm), whereas in 2014 (annual precipitation 640 mm), this recharge was reached about 15 Mcm. According to the groundwater budget of the numerical flow model, the recharge from precipitation was 20.9 Mcm, and the recharge from the karst aquifer was 15 Mcm in 2014. In the same year, the infiltration from Demre Stream, groundwater abstraction, evapotranspiration, and discharge into the sea (both from the aquifer and Demre Stream) were 0.2 Mcm, 15 Mcm, 14.5 Mcm, and 6.5 Mcm, respectively. A conceptual model of the groundwater system in the Demre coastal plain is presented with the water budget components in Fig. 8.

Fig. 8
figure 8

Conceptual model and groundwater budget components of the coastal aquifer

Full size image

Recharge and discharge components of the numerical flow model agree also with hydrological groundwater budget calculations. For example, recharge from precipitation was calculated as 18.9 Mcm in groundwater budget and 20.9 Mcm in model calculations. Similar agreements are observed in hydrological budget/model budget values of the recharge from karst aquifer as 16.2 Mcm/15.0 Mcm, evapotranspiration as 13.4 Mcm/14.5 Mcm surface run-off, and groundwater outflow to sea as 6.7 Mcm/6.5 Mcm. The difference between the values was partly due to slight differences in the magnitudes of areas used in hydrological budget and model budget calculations.

Models based on future climate scenarios

The responses of Demre coastal aquifer to future climate change have been assessed based on two scenarios called RCP 4.5 and RCP 8.5. An RCP (Representative Concentration Pathway) is a temporal greenhouse gas concentration trajectory adopted by the IPCC (Intergovernmental Panel on Climate Change, http://www.ipcc.ch/) for its fifth Assessment Report (AR5) in 2014. Currently, Earth has almost reached the conditions represented by the scenario RCP 4.5. Considering that there is no evidence for an international agreement of a hard-brake that would decrease greenhouse gas emissions in foreseeable future, the scenarios RCP 4.5 and RCP 8.5 seem to bracket the more likely cases. Even, an abrupt termination of greenhouse gasses occurs, the present moment of radiative forcing would probably continue to increase for a considerable amount of time because of the century-long lifetimes of some of the greenhouse gases. Therefore, scenarios of RCP 1.9, RCP 2.6 and RCP 6.0 are not considered in this study.

According to AR5, Turkey is one of the Mediterranean Basin countries where the effects of global climate change will be felt strongly (Türkeş et al. 2013). The reason for global climate change is the increasing concentration of greenhouse gases in the atmosphere. The RCP 4.5 scenario assumes the greenhouse gas emissions will reach a peak around 2040, then decline whereas, in the RCP 8.5 scenario, emissions will continue to rise throughout the 21st century.

State Meteorological Service of Turkey (MGM) produced annual temperature and precipitation projections for the hydrological basins of Turkey (MGM 2014, see Supplementary Material for the data). These projections are based on RCP 4.5 and RCP 8.5 scenarios and, have a spatial resolution of 20 km, and cover the period between 2013 and 2050 (Demircan et al. 2014). According to RCP 4.5 and RCP 8.5 scenarios, mean annual air temperatures over the study area are expected to increase almost linearly from about 19.5 °C in 2015 to about 21 °C in 2050 (see Fig. 9). Both scenarios infer the same trend of temperature increase although annual temperature estimates may deviate as much as 2 °C between the scenarios for a given year. Such a temperature rise is expected to increase irrigation water demand because of the increasing evapotranspiration. Another adverse effect of temperature rise is the reduction of the amount of snow-type precipitation on the mountainous part of the Demre basin where the karst aquifer, recharging laterally the alluvium aquifer, outcrops. Contrary to rainfall, part of which tends to leave rapidly a basin as surface run-off, slowly melting snow recharges the aquifers more efficiently. Only the episodic warm air masses or extreme rain events falling on snow result in fast snow-melt that can generate surface run-off and reduce the aquifer recharge.

Fig. 9
figure 9

Temporal variation of annual air temperature and annual precipitation in the study area according to RCP 4.5 and RCP 8.5 climate scenarios

Full size image

Figure 9 shows also the temporal variation of annual precipitation estimated by RCP 4.5 and RCP 8.5 scenarios between 2015 and 2050. Unlike temperature, the annual precipitations do not reveal an apparent linear increase trend. For Demre coastal plain, average values of the mean annual air temperature and the mean annual precipitation of the RCP 4.5 and RCP 8.5 scenarios in the period of 2015–2050 are 20.2 °C, 828.4 mm/year and 20.6 °C, 737.2 mm/year, respectively (see Supplementary Data). The anticipated mean annual temperature values are remarkably higher than 18.4 °C which is the average of the 1981–2014 period. However, the mean annual precipitation of 790.5 mm, observed in the 1969–2014 period is between the mean annual precipitation values anticipated by the RCP 4.5 and RCP 8.5 scenarios.

The temperature and precipitation data of the climate scenarios were used as input parameters of the transient numerical groundwater flow model which was run for the period 2015–2050. The precipitation data were used directly. Total of annual irrigation and potable water demands were used as the annual groundwater abstraction rate in the numerical model. Annual estimated evapotranspiration values were calculated from temperature and precipitation data using Turc’s method (Turc 1954). The potable water demand was kept unchanged during the modeling period because an accountable population increase in the plain is not expected. The annual amounts of irrigation water demands for greenhouse and citriculture irrigation were calculated separately by Blaney–Criddle (1962) method in a manner that the precipitation input was not considered for the greenhouses which comprise more than 80% of the plain area.

Annual groundwater abstraction amounts calculated for various years are compared in Table 3. The year 1981 represents the situation when most of the plain (about 93%) was dominated by orchard fields that required an annual groundwater abstraction of 10 Mcm. In 2014, when greenhouses dominated the plain, groundwater abstraction increased to 15 Mcm, mainly because of the increasing plant water use. Greenhouses prevent the recharge from precipitation. Annual groundwater abstractions for the year 2050 were calculated as 19.1 Mcm and 19.6 Mcm based on the climate scenarios of RCP 4.5 and RCP 8.5, respectively. These values indicate about a 25% increase in groundwater abstraction. Annual variation of groundwater abstraction during the 2015–2050 period seems to be associated both with the amount of precipitation and the temperature for both of the climate scenarios (see Supplementary Material). The amount of groundwater abstraction increases in years with low precipitation and high temperature because of the reduced soil moisture in orchard fields and increased evapotranspiration in the orchard fields and greenhouses (see Fig. 10). In general, there is a more obvious relationship between the temporal trends of groundwater abstraction amount and the temperature.

Table 3 Groundwater abstraction amounts and the corresponding properties of seawater invasion zone (SIZ) for different groundwater abstraction conditions
Full size table
Fig. 10
figure 10

Temporal variation of annual air temperature, precipitation, and groundwater abstraction amount according to RCP 4.5 climate scenario

Full size image

Results of the transient numerical groundwater flow modeling suggest that climate change will remarkably affect the groundwater’s head distribution for both of the climate scenarios. This is mainly because the demand for groundwater abstraction in each scenario is similar (see Supplementary Material). The groundwater head around the pumping well field in the north of the plain will be lowered to 1.8 m in 2050 (see Fig. 11) from 2.4 m in 2014 (see Fig. 7). Moreover, the 2.4 m headline which extended along the coastline in 2014 will move to the north of the pumping well field.

Fig. 11
figure 11

Calculated hydraulic head distribution in Demre coastal aquifer for 2050 with RCP 4.5 and RCP 8.5 scenarios

Full size image

According to the numerical flow model describing the conditions in 1981, the 0-m groundwater headline coincided with the sea-level except for the western and eastern ends of the coastline where recharge from precipitation is negligible (on the west and east) and lateral recharge from brackish springs and lagoon is effective (on the east) (see Fig. 7). The maximum distance of 0-m headline from the coastline in 1981 was 70 m and the seawater invasion covered an area of 0.8 km2 (see Table 3). An increase in groundwater abstraction due to the agricultural land-use change increased the area invaded by seawater to 1.2 km2 and the maximum distance of the 0-m groundwater headline to the coastline reached 166 m. Beyond that, the growing demand for groundwater due to climate change seems to increase the area invaded by seawater to 1.2 km2 and the distance of intrusion to the coastline will increase to 206 m (see Table 3). These changes show that the groundwater system in the plain will reach its limits in terms of long-term sustainability. If the groundwater abstraction rates anticipated from climate change scenarios are exceeded, seawater intrusion by invasion from the coastline and rise from the bottom of the aquifer seems to be inevitable.

An important problem regarding the projections of future behavior of the groundwater system is the veracity of the RCP scenarios used in the groundwater flow model predictions (see Table 4). To evaluate this issue, the RCP-anticipated precipitation and temperature values in the period 2015 and 2019 are compared in Table 4 to the observations of the Demre meteorological station, which is located in the study area. In this period, the means (and 1 sigma standard deviation) of the difference between observed and anticipated precipitation values of RCP 4.5 and RCP 8.5 scenarios are -235 mm/year (± 568.3) and − 8.9 mm/year (± 396.1), respectively. Similarly, the values of the difference between observed and anticipated temperatures of the RCP 4.5 and RCP 8.5 scenarios are 0.1 °C (± 10.5) and − 0.4 °C (± 10.5), respectively. Among all data, reasonable deviations from the observed data occur only in RCP 4.5 precipitations of the years 2015 and 2019 and RCP 8.5 temperature of the years 2016 and 2017. These values indicate that the predictions of an RCP scenario for individual years can be highly unrealistic although the general trend of precipitation and temperature agrees with the trend of global climate change.

Table 4 Comparison of the observed and RCP-anticipated precipitation and temperature values for the study area
Full size table

Discussion

Results of the numerical groundwater flow model run based on future climate change scenarios of RCP4.5 and RCP8.5 reveal a major risk of the sustainable use of the groundwater in Demre coastal plain. However, this conclusion is based on several assumptions, such as continuity of the present agricultural land use, hydrogeologic boundary conditions, and the validity of precipitation and temperature data series anticipated by the RCP scenarios.

The agricultural land use pattern is determined usually by the magnitude of income obtained from the type of plants grown. If the present plant pattern is changed to plants that require more irrigation water, then the groundwater system which seems to reach its safe limit by 2050 can move to a negative water balance condition which in turn would result in seawater invasion. Similarly, if less water-demanding plants are preferred, then the groundwater system would be in much safer conditions in terms of sustainable use.

The reliability of the results obtained from modeling depends also on the continuity of hydrogeologic boundary conditions in time. The coastal aquifer is fed strongly by the mountainous karst aquifer. Models assumed that this lateral recharge will continue in the same manner during the modeling period. This could only be possible if the recharge conditions of the karst aquifer are unchanged until 2050. However, the magnitude of the lateral recharge from the karst aquifer may be affected also by climate change. In an assessment of the climate, impacts on eight representative aquifers in the western United States, Meixner et al. (2016) point out an expected decline of the mountain system recharge (MSR) due to decreased snowpack. Increasing temperatures due to climate change both decrease the winter precipitation and increase the rain-to-snow ratio which may lead to decreased groundwater recharge. However, the actual effects of these changes on MSR are uncertain due to the poor understanding of mountain aquifers (e.g. Viviroli et al. 2006). Effect of sea-level rise on the boundary condition of numerical groundwater flow model along the coastline seems to be negligible because the maximum amount of rising would be about 6 cm according to a recent estimate of 1.15 mm/year_ rise per 1 °C increase in temperature (Sweet et al. 2017).

As shown in this study, the RCP-anticipated temperature and precipitation data used in numerical groundwater flow models may be highly uncertain at least in some part of the modeling period. Some researchers (e.g. Crosbie et al. 2013) use the probability distributions of the outputs of a large number of GCMs (Global Climate Models) that utilize a large number of RCPs, to make more robust estimates of future groundwater recharge. However, such an approach may not be useful as stated by Jackson et al. (2011) who use an ensemble average of thirteen GCMs to infer the future groundwater recharge but found that the result is not statistically significant at the 95% confidence level. Similarly, Joseph et al. (2018) point out the need for a convergence of climate models before attempting hydrological impacts assessment studies to be used for subsequent groundwater adaptation and management strategies.

Conclusion and Outlook

Results of this study anticipate a sustained use of groundwater resources in the plain until 2050 if the present land (or irrigation water) pattern use is not changed, the model inputs obtained from RCPs are generally valid and the lateral recharge from mountainous karst aquifer continues as it does currently. However, a comparison of the observed and RCP-anticipated model input parameters (i.e. temperature and precipitation) shows substantial deviations from the observed data in some years of the modeling period. Therefore, uncertainties associated with GCMs, RCPs, and with the GCM downscaling algorithms limit the reliability of the forecasts made by numerical groundwater flow models. Future work should concentrate on using more accurate input data and assessing the effect of uncertainties in the input data on the results of the numerical groundwater flow model. Sufficient data should be collected for future model calibrations that should include inverse modeling to assess the multiple permutations of all likely parameters affecting the model results.