Modelling seawater intrusion in the Pingtung coastal aquifer in Taiwan, under the influence of sea-level rise and changing abstraction regime

Developing industrial, agricultural and tourism activities together with human settlements in coastal areas has led to overexploitation of aquifers, increasing seawater intrusion (SWI) and decreasing groundwater quality. The sea-level rise (SLR) due to climate change associated with global warming is expected to exacerbate the problem (Stocker 2014). The coastal plain of Pingtung in Taiwan currently suffers from SWI due to overexploitation of the aquifer system by a significant number of pumping wells screened through the aquifers, at different pumping rates and depths, and using different well types. There are two different methods commonly used for analysing SWI: “sharp interface” and “variable density” methods (Sherif et al. 2012). In the sharp interface method, the saltwater and freshwater are considered as immiscible fluids with a sharp interface separating them in a homogeneous unconfined aquifer. This method is used when the width of the transition zone between saltwater and freshwater is small compared to the aquifer thickness (Bear 1979), and hence the effect of the interface (transition zone) is ignored. It can be assumed that the flow is substantially horizontal in the system (Dupuit assumption) and the vertical gradient is negligible (Vacher 1988). Based on the literature, due to the dominantly laminar pattern of groundwater flow in aquifers, where the width of the mixing zone is relatively smaller than the aquifer thickness, the boundary can be delineated as a sharp interface with a slope, where the two liquids may be considered as two immiscible fluids (Das and Datta 1999; Kacimov and Sherif 2006). Under this condition and due to pressure balance on both sides, it can generally be assumed that this sharp interface is static and there is no flow across it (Dogan and Fares 2008). The Ghyben-Herzberg method was the first analytical formulation of SWI (Bear 1979), and provides a quick estimation of the location of the freshwater–saltwater interface under steady-state conditions.

However, this method is not applicable to cases with a wider interface between freshwater–saltwater (Cooper 1959), as is the case in this study where the freshwater–saltwater interface occupies a greater depth than the thickness of the aquifer system. Therefore, the variable density approach was employed to analyze the mixing of freshwater and saline water (as miscible fluids), which is controlled by hydrodynamic dispersion, which is the combination of mechanical dispersion and physico-chemical dispersion (Bear 1979). This zone is created between freshwater and saltwater in the form of a transition zone and it is closer to the physical behaviour of the flow and solute transport in reality. The fluid density in the transition zone (freshwater–saltwater interface) varies in time and space as a function of the fluid’s temperature, concentration, and pressure (Kooi et al. 2000; Voss and Souza 1987; Pool and Carrera 2011; Diersch and Kolditz 2002; Barazzuoli et al. 2008); therefore, a variable-density flow and transport model is imperative for an appropriate representation of the saltwater intrusion mechanism (Kolditz et al. 1998; Voss and Souza 1987; Langevin and Guo 2006; Simmons et al. 2001).

While there are many natural and manmade factors affecting SWI in coastal aquifers, the main natural factors are aquifer geometry, geological parameters, depth of aquifer along the coast, hydraulic and hydrogeological parameters, and sea-level rise (Mehnert and Jennings 1985; Sherif et al. 2012; Paniconi et al. 2001a, b; Zhou et al. 2000; Tber et al. 2008; Narayan et al. 2007; Collins and Gelhar 1971). Moreover, artificial recharge, controlling water runoff from the surface and excessive pumping are some of the man-made factors affecting SWI (Sherif et al. 2012). The progress of the freshwater–seawater interface in coastal aquifers can be directly affected by changes in all of the aforementioned factors (Kim et al. 2009). Identifying the effects of these factors on SWI plays an important role in the effective management of groundwater resources.

Sea-level rise is one of these factors that reduces the land surface and affects groundwater resources (Werner and Simmons 2009; Chang et al. 2011; Ferguson and Gleeson 2012; Essink 2001). By 2100, sea levels are expected to rise globally from 0.09 to 0.88 m (Solomon et al. 2007). SLR can also increase the hydrostatic head of the sea which would, in turn, result in inland movement of the saltwater interface (Priyanka and Mahesha 2015). SLR has a negative effect on aquifers’ hydraulic equilibrium condition, in line with various natural parameters such as tides and manmade activities like overpumping (Werner et al. 2013). SLR coupled with overexploitation of groundwater has been proven to be a key factor in extending the transition zone in unconfined aquifers (Bobba 2002; Yechieli et al. 2010; Loáiciga et al. 2012; Carretero et al. 2013; Rasmussen et al. 2013; Sefelnasr and Sherif 2014; Uddameri et al. 2014). Loss of wetlands, impeded drainage, erosion, land-surface inundation and SWI are all problems caused by sea-level rise (Fitzgerald et al. 2008; Bricker 2009; Nicholls 2010, 2015). The lateral encroachment of seawater will be accelerated by increasing the hydraulic head of the saltwater in coastal boundaries due to SLR. It has been shown that a 0.5-m SLR could increase the lateral saltwater intrusion in the Nile aquifer by about 9,000 m (Sherif and Singh 1999) and change the inland penetration of the saline toe by approximately 5 km (Werner and Simmons 2009). In another study, it was shown that a 1-m SLR within the Mediterranean Sea would cause a 400-m intrusion for the Mediterranean coastal aquifer in Israel (Yechieli et al. 2010).

A number of numerical simulation codes such as SUTRA (Voss and Provost 1984), FEFLOW (Diersch 2006), and SEAWAT (Langevin et al. 2008) have been applied to simulate variable-density flow and solute transport in aquifers. Despite the availability of powerful numerical modelling tools, due to the lack of adequate and accurate data, relatively few studies have been published on the large-scale simulation of SWI (Ababou and Al-Bitar 2004; Kopsiaftis et al. 2009; Sherif et al. 2012). In general, collecting hydrogeological data for large-scale three-dimensional (3D) case studies is very expensive, time-consuming and in some cases impossible due to limited accessibility. In this study, the encroachment of saltwater at different layers of the multilayer aquifer system in the Pingtung Plain in Taiwan, is simulated in 3D using FEFLOW, considering the density-dependent flow and solute transport in the aquifer.

The Pingtung Plain, with an area of 1,210 km² (width of 20 km and length of 55 km), is located in the southwest of Taiwan. The plain extends as a 2-km deep alluvial wedge bounded by two active faults named “Feng Shan” and “Chao-Chou” from the north-west and south-east, respectively. The Taiwan Strait extends along the south border of the domain boundary. The foothills and river valleys bound the plain in the north. There is a mountain range adjacent to the alluvial deposit plain (Fig. 1), which itself can be categorised as a valley due to its position, being surrounded by hills with four specific topographic zones which are the northern mountainous region, eastern mountainous area, the western foothills and the Pingtung Valley (Ting et al. 1998b). The plain is a relatively flat area with less than 100-m elevation and is drained by three main rivers, the Kaoping, Donggang, and Linbian. The Kaoping, with a length of approximately 50 km, is the longest river within the Pingtung Plain area, originating from the mountain range and discharging to the Taiwan Strait. These rivers divide the Pingtung Plain into three zones of proximal fan, mid fan, and distal fan (Fig. 1).

The Pingtung Plain’s subsurface geology has been identified from the examination of the outcrops and interpretation of borehole tests and available wells (Ting et al. 1998b). According to Kuo and Wang (2001), sandy gravel with more than 60% gravel and 20% sand forms the proximal fan material with a transmissivity of about 9,000 m²/day and a storage factor of 6.5 × 10⁻³(Hsieh et al. 2010). However, decreasing grain size from north to south and east to west leads to having 40% gravel and 40% sand in the mid fan zone (Kuo and Wang 2001) with 2,300 m²/day transmissivity and a 9.5 × 10⁻⁴ storage factor, while the distal fan is mainly comprised of silt and clay, with a transmissivity of about 1,200 m²/day and a storage factor of about 0.00005 (Hsieh et al. 2010). The afore-mentioned storage factors are based on previous studies (Hsu 1961; Ting 1992; Ting et al. 1998b) which were defined as characteristic properties of the upper aquifer only, while the values of this parameter are not available for the other layers. Due to the lack of data, value of 0.0001 m⁻¹ (default value in FEFLOW) was assigned for specific storage (S_s) for the entire model.

Being located south of the Tropic of Cancer, the Pingtung Plain has a subtropical climate with an average temperature of 24 °C, reaching its highest value (28 °C) in July and lowest value (19 °C) in January. The plain receives, on average, 2,400 mm/year of natural precipitation; nonetheless, this amount reaches 3,000 mm/year or higher in the mountain areas, with 90% of the total annual precipitation occurring during the wet (monsoon) season (from May to October) due to typhoons (Yu et al. 2006).

The recharge pattern is highly dependent on the permeability of the soil (Hsieh et al. 2010). As the right side of the plain, the proximal fan is composed of gravel and sand, it has higher potential for recharge and percolation of precipitation and river flow (Ting et al. 1997; Hsieh et al. 2010). Also, the monitored groundwater level in the eastern part is much higher than in the western side of the plain (zone of overpumping), comprised of silt and clay with fine sediments in all aquifers (Hsieh et al. 2010; Chang et al. 2017). The precipitation follows an unsteady pattern in this area and the ratio of rainfall in the wet season to dry season is 9:1, which is much higher than the ratio in other areas in Taiwan (Huang and Chiu 2018; Fig. 2).

The rainfall distribution is a significant factor in controlling the availability of the Pingtung Plain’s groundwater resources (Hsu et al. 2007). The plain is enclosed by mountainous regions in the east and north and foothills in the west, which contain a thick series of black slate with very low porosity (Tu et al. 2011). Accordingly, the cleavages and local joints are the only options for water storage (Ting et al. 1998b), a feature that results in most of the precipitation running off the surfaces towards the rivers and streams and along the plain (Chang et al. 2004). The entire groundwater aquifer underlying the Pingtung Plain area is comprised of unconsolidated sediments. The direct infiltration of rainfall in the north and east of the plain (due to the unconsolidated deposits) through the foothills is the main mechanism of groundwater recharge. Additionally, river seepage, as local penetration in the upper regions of the plain, is another source of surface water for recharging groundwater.

Faults with impervious bedrock surround the plain from the east and west. This, together with the low permeability soil along the north boundary, creates no lateral flux boundary conditions in these parts of the domain (Hsu et al. 2007). The outflow includes abstraction via numerous pumping wells, evaporation losses and the seaward movement of brackish water. The groundwater has been extracted at a rate of 4,000 million liters per day (MLD) within the entire Pingtung Plain area, which is significantly higher than the natural recharge rate of 2,730 MLD (Hsu et al. 2007), which also strongly depends on the precipitation rate. Due to lack of data on the transient pumping rates for the simulation period (2008–2015) and depth of the wells, the pumping data of the year 1996 (the only available data) for the 25 townships in the Pingtung Plain (Figs. 1 and 3; Table 1) were used as input. The wells were considered screened through the aquifer’s whole depth (through three aquifers and two aquitards), for both steady-state and transient models. Based on Taiwan’s Water Resource Bureau (MOEA), the applied abstraction rate has not changed significantly in the last decade. Therefore, in this study, it was assumed that the rates of abstraction from the pumping wells were constant for the simulation period. However, to address the uncertainty in the data, the study included cases of a 30% increase and decrease in the abstraction rate in order to examine the importance of this parameter for the SWI.

The availability of a large amount of water, the soil texture, and the climatic conditions make the Pingtung Plain one of the highly productive areas for agriculture (Hsieh et al. 2010). Surface-water supplies only 20% of the demand which emphasises the significant dependency on groundwater resources (Lin et al. 2011) for agriculture, industry, fish breeding and drinking water in this region. Seawater intrusion has occurred in the Pingtung Plain due to overexploitation of groundwater, especially due to private well pumping (Chang et al. 2017). The Taiwan Environmental Protection Agency (TWEPA) characterises the saltwater with respect to its total dissolved solids (TDS) content and categorises the water quality according to the range of TDS into three groups: contaminated water (TDS > 10,000 mg/L), semi-brackish water (1,000 < TDS < 10,000 mg/L) and freshwater (TDS < 1,000 mg/L) (Huang and Chiu 2018). Based on an early investigation, in 1960 the groundwater in the entire Pingtung Plain aquifer was of good quality, with TDS ranging between 180 and 984 mg/L (Ellis 1999). However, overabstraction of groundwater, primarily for agricultural use, deteriorated the groundwater quality, especially in the wells located around the Kaoping River mouth, causing the inland encroachment of seawater, with TDS ranging between 1,000 and 29,000 mg/L from 1995 to 1996 (Ellis 1999). One of the main factors triggering seawater encroachment is the outcropping of the aquifers along the Kaoping River mouth, which provides a pathway between seawater and the land (Wang 1996). Overabstraction of the aquifer along the coast has led to lowering of the groundwater table below mean sea level (MSL), which created a hydraulic gradient from the sea towards the land and naturally pushed the seawater inland through this pathway. Based on the literature, nearly 30% of the Pingtung Plain area (359 km²) has been subject to seawater intrusion, with an average inland encroachment rate of 500 m/year from 1980 to 1996 (Wang 1996). Therefore, all areas in the vicinity of the shoreline are greatly susceptible to seawater intrusion.

To study the inland encroachment of saline water in the Pingtung Plain, three cross-sections were used; these are cross-sections A–A, B–B and C–C through the right, middle and left side of the plain, respectively (Table 3; Fig. 11). The contour lines of contaminated water with TDS > 10,000 mg/L were used to represent the saline intrusion (highlighted in red to yellow in Figs. 12 and 13). At the end of 2010, the coverage of the contaminated area followed an increasing pattern from the right to the left of the domain. Although the seawater intrusion maintained its forward pattern through all aquifers, an insignificant encroachment along the right-hand side region of the model (A–A section) for the three aquifers was observed (Table 3; Fig. 11), which is in good agreement with the literature (Huang and Chiu 2018). This could be due to the high potential of the eastern part of the plain (proximal fan) for natural recharge, which leads to higher groundwater levels (Fig. 1) in all aquifer layers compared with the left and middle parts of the plain, where the groundwater heads are lower due to overabstraction (Ting et al. 1998a). Increasing the freshwater level due to precipitation percolation in the eastern side of the plain reduced the occupied transition zone (freshwater–saltwater interface) by pushing it back towards the sea. This is the main reason for the relatively insignificant change of SWI in the right side of the plain.

Also, the predicted salinity concentrations at the end of 2010 (December 2010) showed a larger contaminated area in the top and bottom aquifers compared with the middle aquifer, which is in agreement with the permeability distributions in the three aquifers. The second aquifer is comprised of lower permeability soil compared with the first and third aquifers. As the SWI shows more progress in the middle region of the domain, the 2D vertical section of the model is presented through the B–B cross-section for full representation of the SWI behaviour through the entire depth in the middle of the aquifer system (Fig. 14). Considering the thicknesses of the aquitards and based on the literature (Hsu et al. 2007), no leakage was considered between aquifers and aquitards. The significant difference in conductivity of aquitards and aquifers also supports this assumption. The saline water with TDS higher than 10,000 mg/L intruded about 800 m along the first and second aquitards at the end of 2010, mostly under the influence of abstraction along Donggang Township (T03; Table 1). Although the seawater progressively intruded the aquifer during the 5 years of study (end of 2010 to the end of 2015), the transition of saltwater with TDS > 10,000 mg/L was slower than the lower TDS. The saline water with TDS higher than 10,000 mg/L intruded approximately 100 m more landward in this period (Table 3; Fig. 11). However, larger areas covered by freshwaters were superseded with brackish water in this 5-year period. The mass transfer through the aquitards was very slow (less than 15 m for TDS > 10,000 mg/L) and almost negligible during the 5 years of analysis.

The model was employed to predict the SWI for the next 25 years (2016–2040) in order to study the future trends of saltwater intrusion. In this analysis, the hydrological setting and input parameters were maintained from the previous run. In addition, the effects of sea-level rise, a 30% increase in abstraction rate, a 30% decrease in abstraction rate, and a combination of sea-level rise with a 30% increase in abstraction rate were studied (Table 4; Fig. 13).

The model predicts that in 2040, seawater may intrude mostly in the middle region of the domain by approximately 5 km along the cross-section B–B in the top aquifer (Table 4; Fig. 11).

The results show that, in the top aquifer, an SLR of 27 cm during the next 25 years would lead to more than 1 km, 900 m and 50 m additional intrusion along the cross-sections C–C, B–B, and A–A, respectively. Some townships, including Fangliao (T16) and Jiadong (T21), would be affected more by the increased abstraction, with nearly 70 m additional SWI in the bottom aquifer. By 2040, a 30% increase in the abstraction rate, without considering the effects of SLR, would cause approximately 70, 220 and 120 m of additional seawater intrusion along the A–A, B–B and C–C sections in the third aquifer, respectively.

Among the aforementioned factors, the combined effect of SLR and overabstraction would lead to the most severe intrusion of saline water due to the increased seawater head and decreased inland head, resulting in additional hydraulic gradient towards the land. The contaminated water would cover 340 km² (28% of the entire domain), which is 67.17% more than the situation without SLR or overabstraction (Table 4; Fig. 11).

One of the potential management scenarios for reducing SWI and recovering groundwater resources is to decrease the abstraction rate in order to increase the groundwater table and decrease the landward hydraulic gradient, as one of the suggested scenarios in literature for managing SWI (Hussain et al. 2019). A 30% reduction in the total abstraction rate would lead to about 50, 250 and 1,850 m backward movement (retreat) of saline water in the top aquifer along the A–A, B–B and C–C cross-sections, respectively. For the middle aquifer, this would not lead to any significant backward movement (only 44, 200 and 19 m along the A–A, B–B and C–C sections, respectively). The lower aquifer depicts a 25, 315 and 170 m retreat along the A–A, B–B and C–C cross-sections, respectively; therefore, the top aquifer would be influenced more by a reduction in abstraction. Another possible scenario to control SWI would involve relocating the coastal wells in the vicinity of the shorelines further inland (Hussain et al. 2019) in the regions of Linyuan (E31), Xinyuan (T17), Donggang (T03), Linbian (T19), Jiadong (T21), and Fangliao (T16). The right side of the plain along the A–A cross-section shows a very low response for a 1-km inland relocation of wells with less than 5-m backward movement of seawater at all depths of the aquifer. It maintained this low reaction by increasing the distance of wells from the sea by either 5 km or 10 km with less than 10 m seaward movements in all layers of the aquifer. However, the reaction of the middle part of the plain along the B–B cross-section was more significant, with a minimum of about 365 m and a maximum 375 m backward movement of seawater through the bottom and middle aquifers, respectively, as a response to the 1-km inland relocation of wells. Inland relocation of the wells further away from the sea (5 or 10 km) would not lead to a significant difference in pushing back the seawater in all layers of the aquifer. Inland relocation of the wells 10 km farther pushed the seawater a maximum of about 80 m back to the sea in the middle aquifer (Fig. 15). In the west of the plain in the distal fan, a 1-km inland relocation of pumping wells cleared a minimum of about 50 m to a maximum of 65 m seawater along the C–C cross-section in the bottom and middle aquifers, respectively. Also, this part of the plain did not show a significant difference for the relocation of the wells (either 5 or 10 km away) in all layers of the aquifer. From the predicted results, it can be concluded that the inland relocation of the abstraction wells is a more effective solution at a maximum 1 km distance from the coastal boundary, and any further movement will not change the results in terms of increasing its performance.

A 3D finite element model was developed to simulate the seawater intrusion in the Pingtung Plain coastal aquifer. The developed model can simulate density-dependent flow and allows coupling between the water flow and salt transport by tracing the interface of saltwater and freshwater. The model can address the challenges and complexity of aquifer conditions and parameters (e.g., geology, atmospheric forces, groundwater abstraction, etc.). After rigorous validation, the model was used to study SWI in all layers of the Pingtung Plain aquifer system and to predict the future behaviour of the aquifer. The model showed the relationships between seawater intrusion and aquifer parameters and examined the effectiveness of different management scenarios. Water abstraction near the shoreline is proven to be the main reason for the inland encroachment of seawater, especially in the lower left and middle parts of the plain. Although sea-level rise, the variation of pumping rates and inland relocation of wells were among the other influential parameters, a combination of SLR and increased abstraction exacerbate the situation.

The analysis depicts that the seawater intruded into the entire width of the aquifer in all investigated scenarios, however, less encroachment was observed in the second aquifer. Likewise, seawater intruded less into the lower east of the plain, due to its high potential for natural groundwater recharge and possessing higher groundwater heads.

However, the middle and left side of the plain are more vulnerable to SWI due to overabstraction around the Kaoping River mouth (lower middle) and distal fan (left). Decreasing the abstraction rate and farther inland relocating the wells could be effective solutions with a significant reversal of the seawater intrusion.

This 3D finite element model can potentially be used as an asset to investigate the influential parameters on the SWI to help the decision-makers for establishing policies especially in the era of climate change. Several limitations existed when developing this model. Firstly, the data for the abstraction wells were captured for only one specific year and the same rates were assumed for the consecutive years. Secondly, due to the lack of data, the storativity parameter was assigned as the default value in the software for the entire model.