Assessment of groundwater quality and remediation in karst aquifers: A review

Abstract

Karst aquifers, capable of storing and transmitting large amount of water, are the main source of drinking water in many regions worldwide. Their excessive permeability leads to an enhanced vulnerability to retain and spread the contamination accordingly. From sustainability perspective, the environmental, economic and social impacts of karst contamination on water resources management are gaining more attention. In this study, an overview of hydrogeological processes and concepts regarding groundwater flow and contaminant transport in karstic systems is presented, followed by a short discussion on surface water and groundwater interaction. Due to the complexity of karstic systems, different approaches have been developed by researchers for investigating and understanding hydrogeological processes and groundwater behavior in karst which are reviewed herein. Additionally, groundwater contamination issues and the most common and effective remediation techniques in karstic terrains are discussed. Lastly, modeling techniques and remote sensing methods, as beneficial and powerful tools for assessing groundwater flow and contaminant transport in karst terrains, are reviewed and evaluated. In each section, relevant research works conducted for Puerto Rico are discussed and some recommendations are presented to complement the ongoing hydrogeological investigations on this island.

1. Introduction

Sustainable water resources management is a crucial concern in most countries across the globe. Only 3% of total water on the Earth is considered as fresh water resources and approximately 30% of that is accessible as groundwater, which is vital for human health, ecosystem, energy industry and other water-dependent topics (Shiklomanov, 1993). Karst aquifers are responsible for providing potable water for 40% and 25% of the US and world's population, respectively (Ghasemizadeh et al., 2012). The increasing demand by residential, industrial and agricultural uses have caused groundwater depletion and decreased quality in many regions. From sustainable development perspective, environmental, economic and social impacts are consequences of water pollution in any region of the world. Hence, careful attention should be paid to preserve water resources.

With particular reference to karst aquifers, this study provides a comprehensive review of hydrological concepts and novel investigation and modeling techniques followed by a discussion of groundwater contamination and remediation techniques. This paper aims to present and review the work of other researchers in the recent years (especially after 2010) and to discuss the latest improvements that have been made regarding groundwater quality and quantity assessment. In each section, the associated research work that has been done for Puerto Rico is presented to better understand what research works have been conducted and what are the research gaps on karst aquifers of this island.

Puerto Rico, as the case study location, is considered a territory of the United States (US). The island is located in northeastern side of Caribbean Sea and have an estimated population of 3.6 million (Castro-Prieto et al., 2017). Several surface water and groundwater resources across the island provide residents with fresh water and are used for agricultural, industrial and energy-based purposes. Fig. 1 exhibits the geographical location of Puerto Rico, generalized geological setting of the island, surficial geological formation of north coast limestone aquifer and cross-sectional view of geological layers along the north coast.

Fig. 1.

Geographical location of Puerto Rico (a), generalized geological setting of the island (b), surficial geological setting of north coast limestone aquifer (c) and cross section view of geological layers in Florida and Barceloneta municipalities (Vertical scale is exaggerated, d) – modified from Gómez-Gómez et al. (2014).

Due to the presence of karstic aquifers with high level of heterogeneity and anisotropy in northern coast of the island, water from precipitation can rapidly percolate into underground. This rapid movement, makes karst aquifer highly vulnerable to emerging contamination from several sources of contamination induced by agricultural and industrial activities in addition to proximity to urban areas (Cherry, 2001). Moreover, highly heterogeneous karstic aquifers with a lot of conduits cause high rate of water level fluctuation even in small temporal scaling. Yu et al. assessed the patterns of temporal scaling of groundwater level fluctuation for the karstic aquifer of Puerto Rico and pointed out that according to multifractal or singularity spectrums, there are smoother fluctuations in the alluvial aquifer and rougher ones in the northern karst aquifer (Yu et al., 2016).

2. Karst aquifers

2.1. Characterization

Comprising of chemically soluble rocks with large passages or network of conduits and caves inside, karst aquifers are very permeable and capable of storing and transmitting large amount of water. Therefore, karst regions have great potential of providing habitants with fresh water (Quinn et al., 2006). In fact, 20–25% of the world's population depends on water supplies from karst aquifer directly or indirectly. Approximately 10% of the world's land surface areas have karst aquifer beneath them. This percentage is higher in some areas such as in Europe where it is roughly 35% (Ford and Williams, 2007). Until recent years, the boundaries of karst aquifers around the world were not recognized accurately. Hence, by taking advantage of GIS tools, recent exploration of karst aquifers and Global Lithological Map that was developed before, Chen et al. completed the first World Karst Aquifer Map (WOKAM). Their map distinguishes continuous carbonate rocks and discontinuous carbonate rocks and include major karst springs, wells and caves (Chen et al., 2017). Many other resources show karst map of the US or the world. Distribution of karst aquifers within the United States and its territories based on USGS data is presented herein (Fig. 2). Detailed geological investigations in Puerto Rico show that limestone, dolomite, gypsum and anhydrite are the most common materials forming Puerto Rican karst aquifers (Fig. 1).

Fig. 2.

Distribution of karst aquifers in the United States and its territories – compiled from open files associated with the USGS report of Weary and Doctor (2014).

Karst aquifers are individually different with unique work-frame and characteristics and they should be studied case by case (Stevanović,2015) Two important characteristics of karst aquifers are heterogeneity and anisotropy which make it hard for hydrogeologists and researchers to develop models using simplifying assumptions. Karsts have the most complex system amongst terrains, causing lot of uncertainties and errors in developed models for studying and predicting their behavior (Bakalowicz, 2005). Also, the recharge and discharge rate of karst springs can vary a lot due to several reasons such as fluctuations in water table level caused by hydrological events or seasonal variations (Gárfias-Soliz et al., 2009). Table 1 elaborates hydrogeological characteristics of three main aquifer types, porous media, fractured rock and karst system based on ASTM D 5717–95 Standard (ASTM, 1995).

Table 1.

Hydrogeological characteristics of three main aquifer types, porous media, fractured rock and karst system based on ASTM D 5717–95 Standard (ASTM, 1995; Rosenberry and LaBaugh, 2008).

Aquifer characteristics	Aquifer type
	Porous (Granular)	Fractured rock	Karst
Effective porosity	Mostly primary, through intergranular pores	Mostly secondary, through joints, fractures, and bedding plane partings	Mostly tertiary (secondary porosity modified by dissolution); through pores, bedding planes, fractures, conduits, and caves
Isotropy	More isotropic	Probably anisotropic	Highly anisotropic
Homogeneity	More homogeneous	Less homogeneous	Heterogeneous
Flow	Slow, laminar	Possibly rapid and possibly turbulent	Likely rapid and turbulent
Flow predictions	Darcy's law usually applies	Darcy's law may not apply	Darcy's law rarely applies
Storage	Within saturated zone	Within saturated zone	Within both saturated zone and epikarst
Recharge	Dispersed	Primarily dispersed, with some point recharge	Ranges from almost completely dispersed- to almost completely point-recharge
Temporal head variation	Low variation	Moderate variation	Moderate to high variation
Temporal water chemistry variation	Low variation	Low to moderate variation	Moderate to high variation

A karst aquifer system comprises several elements such as caves, conduits, sinkholes and springs. Limestone karst aquifers are common in many areas around the world including Puerto Rico (Cherry, 2001; Rafael et al., 2016), Florida (Xu et al., 2016) Mexico (Bauer-Gottwein et al., 2011) and China (Luo et al., 2016). Karsts usually are evolved from fractured or fractured-porous rock networks after several years through carbonate dissolution creating large passages and caves. Several modeling methods have been employed to simulate the evolution of karst aquifers from fractured or porous-fractured rock systems (Kaufmann, 2003, 2016; Kiraly, 2003).

2.2. Hydraulic conductivity

In groundwater hydrology, hydraulic conductivity quantifies the ability of soil in transferring water. Based upon various aquifer materials, hydraulic conductivity can range from 10 cm/s for gravel to 10⁻¹⁰ cm s⁻¹ for shale. Hydraulic conductivity of karst limestone is the greatest comparing to many other aquifer types (Fig. 3).

Fig. 3.

Hydraulic conductivity (K) range for different types of rock – modified from Freeze and Cherry (1979).

Laboratory, field and numerical methods are 3 main means of measuring hydraulic conductivity. Numerical and finite element-based methods are used for determining vertical and horizontal hydraulic conductivities (Kalbus et al., 2006; Smith et al., 2016). Usually, in karst aquifers where subsurface heterogeneity exists, determining hydraulic parameters such as K requires a complicated analysis because this parameter is spatially and temporally variable throughout the aquifer. Hydraulic tomography is a novel method that can be used for imaging the heterogeneity in karstic terrains (Illman et al., 2007). Moreover, in karst aquifers, the average range of K can vary depending on several factors such as geology, slope, level of heterogeneity and karstification. Angulo et al. (2011) studied hydraulic conductivity in karst areas by applying water injection tests and electrical resistivity logging. Similarly, different researchers reported experimental values for hydraulic conductivity based on their research approach and case study area (Chen et al., 2011; Fu et al., 2015; Sudicky et al., 2010). As it is shown in Fig. 3, K of limestone karst aquifer which is dominant in northern coast of Puerto Rico, can be assumed in the range of 10⁻⁴−5cm/s⁻¹ which demonstrates high level of permeability in karst aquifers. The estimated values of K in different locations of north coast karst aquifer of Puerto Rico can be found in the water resources investigation reports (Rodriguez-Martinez, 1995) or similar sources for fine-scale studies or modeling purposes.

2.3. What methods can be used to study karst aquifers?

Based on the aforementioned complex characteristics of karst, several techniques and tools associated with modified and reformed conventional methods (such as hydrologic and hydraulic methods, geophysical and geological methods, modeling techniques and tracer tests) have been employed by researchers for understanding the behavior of karst aquifers (Goldscheider and Drew, 2007; Stevanović, 2015). Giudici et al. (2012) and Hu et al. (2009) studied karst aquifers by taking advantage of modeling methods. In addition, taking advantage of geological methods, which help understanding the aquifer geometry and hydraulic properties such as permeability in addition to orientation and characteristics of potential flow paths, can boost the accuracy of modeling results (Goldscheider and Drew, 2007). Geophysical techniques can also be employed in conjunction with geological methods to understand geologic structures and overburden thickness of the aquifer (Chalikakis et al., 2011; Ford and Williams, 2007; Goldscheider and Drew, 2007).

2.4. Surface water and groundwater interaction (SWGWI) in karst

Interaction of surface water (SW) and groundwater (GW) plays a critical role in understanding hydrological behavior of a basin. This interconnectivity incorporates the topographical, geological and morphological characteristics of terrains. Generally, water recharge from inflow of GW into the riverbed, water discharge from river bed to aquifer and also losing and gaining water for both SW and GW in some river segments are three main processes that can occur in SWGWI (Winter et al., 1998). High permeability and low attenuation capacity of karst aquifers make mixture of SW and GW problematic for fresh water use in karst terrains. Rainfall infiltration and streamflow response to base-flow discharge are separated by a much shorter delay in karst compared to other aquifer types (Bayless et al., 2014). Due to the complex patterns of SW and GW flow in karst, several studies have denied coincide of SW and GW drainage divides. A river that disappears in a SW basin and reappears in another basin can be mentioned as an example which depicts uncertainties in understanding the source of water and its associated dissolved contaminants in karst (Winter et al., 1998).

Usually, carbonate rocks are present at the land surface in karstic terrains; however, in some cases, the bedrock is covered by other deposits (e.g. surficial deposits) which is called “mantled” karst (e.g., Edwards Aquifer in south-central Texas). In mantled karst, shallow GW has interaction with lakes and wetlands in a way similar to that in sandy glacial and dune terrains. The small difference in interaction of these two can be tied with the buried carbonate rocks. Dissolution of buried carbonate rocks can lead to slumpage of an overlying confining bed, allowing water to flow readily through the confining bed. Consequently, changes in hydraulic heads within the aquifers underlying the confining bed can affect the lakes and wetlands (Winter et al., 1998). In addition to south-central Texas, mantled karst can be found in north-central Florida. This area has several sinkhole lakes and unconsolidated deposits, overlie the highly soluble limestone of the Upper Floridan aquifer. When unconsolidated surficial deposits slump into sinkholes which form in the underlying limestone, most land surface depressions containing lakes are made. Hence, in most of the times, sinkholes in the bedrock underlying lakes affect the hydrology of the lakes, although the lakes are not located directly within limestone (Winter et al., 1998).

The general term “Surface water” can be referred to wastewater on the ground, brackish water of the sea/ocean, and water in lakes and rivers. Attempts have been made to minimize the interaction between polluted SW and GW in karstic areas by placing an impermeable seal along canal bottoms or riverbeds, constructing small dams/weirs, building grouting curtains, changing the flow direction of surface waters, plunging ponors and creating reactive barriers by pumping freshwater into aquifer (Milanović et al., 2015).

Modeling, hydrographic analysis and hydrogeological mapping are among common “desktop” techniques which are used to assess SWGWI in karst (Fleckenstein et al., 2010). Hydrographic analysis, which monitors stream flow time series and define base flow component, can be used easier compared to the other two methods. However, it is only applicable for gaining stream conditions. Hence, it may not give accurate result if used for a karst aquifer. Hydrogeological mapping method, which provides comprehensive understanding of GW systems around streams and their related hydrogeological systems, can describe hydrogeological components of a GW system at a course scale (Gleeson et al., 2014). Thus, it can be used for regional assessment of karst aquifers. In addition, modeling methods can simulate water flow and contaminant transport around streams and within aquifer using mathematical equations. Although they can predict the behavior of a hydrogeological system, their application in karst is limited. Usually, simplifying assumptions are made during modeling process of SWGWI in karst which result in uncertainties. However, by taking advantage of new techniques/technologies and/or by combing this method with other in-situ measurements, researchers have shown the application and promising future of modeling methods (Guay et al., 2013).

Also, field observations, seepage measurement, hydrochemical methods such as environmental isotopes, hydrochemistry, tracers and also hydrometric and geophysical analysis are beneficial “field” tools which can be used for describing SWGWI in karst (González-Pinzón et al., 2015; Martinez et al., 2015). In the recent years, use of methods associated with geophysics (e.g. resistivity, EM, radiometrics) or remote sensing (e.g. Landsat) has been reported for mapping landscape features that indicate or control connectivity (Meyerhoff et al., 2013). However, because these methods require specific equipment, technical expertise and logistical support, researchers have had tendency to use more practical and “easy to use” methods such as Hydrometrics and Water Budgets analysis (Brodie et al., 2007). Sometimes, because of unknown and complex catchment boundaries, analyzing water budgets can also be problematic (Goldscheider and Drew, 2007; Hartmann et al., 2014; Kovács et al., 2005). Among field tools, tracers have been widely used due to their capability of providing independent ways of validating or refuting conventional-traditional methods of analyzing data and describing SWGWI (Baskaran et al., 2009; Jankowski, 2007). By using tracers such as florescent dyes, aquifer parameters such as recharge and discharge and fluid transport properties can be determined (Martinez et al., 2015). However, tracer studies require careful planning (e.g. meeting environmental regulatory controls) and dosage control. In many cases, isotropic techniques and artificial tracers are used for determining residence time and water age and for understanding the water movement through conduits. The main advantages of these techniques are determining linear flow velocities and information on contaminant transport and delineating catchment areas. Although obtained information and data from tracers are often reliable and unequivocal, limited applicability in large areas with long transit times and also change of color and toxicity concerns are some of the disadvantages of using isotropic techniques and artificial tracers (Goldscheider et al., 2008; Jones and Banner, 2003; Morales et al., 2017). Reviewing the literature, it seems like the use of tracers has been more frequent and beneficial for understanding the interconnectivity between SW and GW in karst terrains compared to other field methods (Katz et al., 1997), for example evaluating the connection between sinking streams and down-gradient springs. It should also be noted that temperature change analysis and water budget assessment can be coupled with other methods to achieve more accurate and validate results (Brodie et al., 2007).

Due to complexity of karst's behavior, different types of “integrated” techniques and/or less common approaches have been employed by researchers to evaluate their performance in describing the interconnectivity between SW and karst GW (Brian Katz et al., 1997; Chu et al., 2016; Loáiciga, 2001; Rugel et al., 2016). Sutton and Screaton described SWGWI of a karst aquifer basin in Florida by analyzing river discharge and a transient numerical GW flow modeling. Their modeling results highlight the prominence of spatiotemporal variations in head gradients that can affect streams and karst aquifers connections and aquifer martial dissolution (Sutton et al., 2014). Bailly-Comte et al. studied the hydrodynamic interactions between GW and surface water in a karst watershed in southern France. Their focus was on the effect of GW on the genesis and propagation of surface floods. They showed the role of initial water level in a karst aquifer in predicting the type of hydraulic connection between SW and GW during flood events (Bailly-Comte et al., 2009). Moreover, the analysis of SW and karst GW interaction conducted by Bayless et al. (2014) showed that analytical methods such as hydrograph separation and hysteretic loops can be used for identifying bounding conditions within the watershed. Chu et al. studied SWGWI in a karst aquifer in China using GW levels data and hydrochemistry analyses, together with isotope data based on hydrogeological field investigations. Their cross-correlation analysis showed that precipitation results in increasing GW level with a 2-month lag time. Moreover, their spectral analysis depicted the possible interconnectivity between GW level, GW exploitation and precipitation (Chu et al., 2016). Rugel et al. conducted longitudinal sampling on a creek in Georgia, USA. They realized there are stepwise GW inputs from the Upper Floridan Aquifer, (in addition to specific conductance increase) at 5 out of 50 sampled reaches. The results of sampling along the 50-km study river section revealed that 42% of total GW inputs entered through discrete preferential flow pathways were located within these 5 sampling locations (Rugel et al., 2016).

In north coast limestone aquifer of Puerto Rico, with presence of rivers and lagoons and occurring intense precipitation and seawater intrusion, SWGWI can have a major impact in the quality of fresh water within karst aquifers of the region. However, we failed in our attempts to find any detailed research works associated with SWGWI in karst aquifers of northern PR. Hence, collecting field data, doing research and developing models to fully understand the mechanism of SWGWI in northern part of the island is strongly recommended.

Despite the fact that numerous methods have been developed for describing SW-GW interrelationship, there are still uncertainties and lack of sufficient knowledge for fully understanding 1- the time lag between GW pumping and its influence on SW, 2- relationship between GW pumping and river losses and 3- exact recharge and discharge points in streams (Jankowski, 2007). Wu et al. assessed uncertainties in SWGWI modeling. Employing a probabilistic collection method, they evaluated the applicability of the frame-work through an integrated SW-GW model for a basin in China and asserted that in describing complex SWGWIs, modeling uncertainties depend on the output and have significant spatiotemporal variability. Hence, employing a systematic uncertainty analysis can be extremely helpful in understanding SWGWI (Wu et al., 2014) and also in groundwater model development process of karst aquifers (Engelhardt et al., 2014).

3. Groundwater contamination and remediation techniques

Karst aquifers which are portrayed as high permeable soil/rock systems with caves and fractures inside and also recharged by sinkholes and rivers, have shown high vulnerability to contamination (Kačaroğlu, 1999). The formation of solution channels and sinkholes facilitates the intrusion of seawater and contaminated storm water and wastewater into the aquifer. This is the main reason of accelerating the spread of contaminants in karst more than other aquifer types (i.e. porous media and fractured rock). Basically, fate and transport processes are not only different for each contaminant (depending on the physical and chemical properties of the contaminant), but also are particularly different in karst compared to other aquifers. A few studies assessed historical GW contamination in north coast limestone karst aquifer of Puerto Rico and suggested beneficial remedial actions and water resources management approaches (Biaggi, 1995; Padilla et al., 2011).

3.1. Contaminant types and sources of contamination

Agricultural, industrial, residential, commercial and municipal development are considered as the main sources of groundwater pollution in recent decades (Fetter, 2001; Wakida and Lerner, 2005). In particular, leakage of storage tanks, chemical spills, landfills, fertilizers and pesticides, sanitation systems, untreated waste discharge and sewage are some of the main sources of contamination due to anthropogenic activities (El Alfy and Faraj, 2017). Generally, regardless of the source of contamination, pathogens, organic compounds (Lapworth et al., 2012), metals (Yao et al., 2012), and other inorganic compounds (e.g., nitrates, chlorides) are often found in groundwater (Panagiotakis and Dermatas, 2017; Vidal Montes et al., 2016). The most commonly found contaminants in karst GW are presented below.

3.1.1. Nitrate

Nitrate is one of the most common GW contaminants according to several studies (Almasri and Kaluarachchi, 2004; Babiker, 2004). Point sources of Nitrogen/Nitrate leachate such as old septic systems, landfills, wastewater holding ponds and leaks from cracks in sewer pipelines can cause GW Nitrate contamination (Almasri, 2007; Kendall and Aravena, 2000; Wakida and Lerner, 2005). Additionally, non-point sources of N leaching, such as fertilizer use in agricultural areas, play a very significant role in increasing GW Nitrate contamination. In a USGS report with focus on Manati and Vega Baja municipalities in Puerto Rico, Nitrate occurrence and contamination were assessed. It was identified that the major sources of Nitrate contamination in the karst aquifer of the region are use of fertilizers for cultivation of pineapples and also septic tank effluent in rural and un-sewered (no sewer system) areas (Conde-Costas and Gómez-Gómez, 1999).

3.1.2. PPCP

Nowadays, treating water contaminated by new chemical compounds mainly originated from pharmaceutical and personal care products (PPCP) are a big concern (Metcalfe et al., 2011). Concentration of PPCPs such as Antibiotics, Anti-inflammatories, Lipid regulators, Psychiatric drugs, Stimulants, Insect Repellants and Sunscreen agents is higher than regulatory criterion in many areas. Usually, wastewater and contaminated surface water, landfills, septic systems and sewer leakages are considered as common sources of PPCP contamination especially in karstic areas (Dodgen et al., 2017; Sui et al., 2015). High concentration of PPCPs in karst aquifers of Switzerland and Jordan was reported by Morasch (2013) and Zemann et al. (2015), respectively. Moreover, Dodgen et al. assessed PPCP concentration in karst aquifers of southwestern Illinois in the USA and found out that no single water quality or meteorological parameter was significantly associated with PPCP contamination across all site profiles. They mentioned septic tanks effluent as a probable source of PPCPs (Dodgen et al., 2017). In another study, it was realized that due to high transmissivity of the shallow alluvium, the GW below alluvial deposits is subjected to pollution caused by PPCP residues from wastewater effluents through artificial aquifer recharge (Sui et al., 2015). No studies related to PPCP contamination of GW in PR were found. However, it is expected that due to large number of septic tanks and poor effectiveness of wastewater treatment plants to deal with PPCPs in addition to use of mosquito repellent creams (because of Zika virus), groundwater quality will be affected by PPCP contamination in the island (Hennessey et al.,2016).

3.1.3. Metals

In carbonated water of the karst aquifer (at natural pH), heavy metals such as Cr, Ni, Pb and Cd often precipitate as hydroxides and carbonates. Heavy metal transport, as an episodic phenomenon, mostly occurs when metals are adsorbed onto small particles of soil. When hydraulic condition of GW flow facilitates suspension of small particles, heavy metals can migrate down the flow path (Vesper et al., 2003). Also, elevated concentration of metalloids such as Arsenic has raised concerns in many regions. Heavy metals and metalloids in groundwater can be present in areas where there are mining operations or mine waste dumps. Industrial activities and urban waste are known as possible sources of metal(loid)s leachate in karstic GW (Alloway, 2013). Additionally, urban surface runoff containing high concentration of metals goes through karst aquifers via sinkholes and conduit network (Bonneau et al., 2017). Heavy metal ions are also frequently leached naturally from rocks and soils within karst media and can be introduced with acidic deposition (Drew and Hötzl, 1999). In karst GW of Puerto Rico, high concentration of metals (mainly arsenic, lead, and total chromium) have also been detected mainly at superfund sites (Fig. 4) (USEPA, 2003).

Fig. 4.

Case study location (upper picture), RCRA, NPL Superfund sites, aquifers, and the sampling wells (lower picture) based on the study of Yu et al. (2015).

3.1.4. LNAPL and DNAPL

Light Non-Aqueous Phase Liquids (LNAPLs) have tendency to float on the surface of groundwater table. They are usually present behind obstructions in cave streams within karst aquifers. Unlike LNAPLs, Dense Non-Aqueous Phase Liquids (DNAPLs), sink to the bottom of the aquifer and have tendency to be stored in lower points of the bottom of conduits. Moreover, it was reported that they tend to concentrate in open conduits (Ghasemizadeh et al., 2012). LNAPL and DNAPL Transport in karst aquifers is contingent upon runoff and GW flow hydraulics. LNAPLs can be forced into the karst system as plug flow and DNAPLs can be mobilized by transporting sediments at the bottom layer. In contrast to dissolved contaminants, the transport of NAPLs in aquifers is strongly influenced by buoyancy effects and surface tension properties. This makes the transport behavior of NAPLs less directly related to groundwater flow and more difficult to predict (Vesper et al., 2003).

By conducting transport experiments using non-aqueous phase TCE in a laboratory-scale karst limestone rock, Carmona De Jesús and Padilla showed that although DNAPLs prefer to move downward, they can also move along non-vertical conduits when subjected to GW flow fields. Furthermore, it was demonstrated that once DNAPLs penetrate preferential flow pathways, subsequent injections often follow the same route. When they reach limited flow areas, DNAPLs will be accumulated and transported through diffusive processes after dissolution (Carmona De Jesús and Padilla, 2015).

3.1.5. Volatile Organic Compounds (VOC)

Volatile and Semi-Volatile Organic Compounds (VOCs and SVOCs, respectively) flow along with the subsurface waters in karst; but their volatility, results in migrating the vapor phases of these contaminants through the vadose zone (Field, 2017). A few studies assessed the contamination of Chlorinated Volatile Organic Compounds (CVOC) in PR (Padilla et al., 2011; Rivera et al., 2017). Historical studies since 1980 show that mainly, contaminants with chlorinated solvents including TCE, dichloroethene, chloroform, carbon tetrachloride, tetrachloroethene, tetrachloroethane, dichloroethane and methylene chloride, have high concentrations causing public health concerns (Padilla et al., 2011). Several wells and sites (Fig. 4) were considered as the National Priority List (NPL) Superfund Sites and remediation actions for treating water in these sites begun years ago (Yu et al., 2015).

Regarding studying the groundwater pollution and understanding the potential exposure pathways of contaminants, some methods such as using tracers and GIS has been employed by researchers (Steele-Valentín and Padilla, 2009). The abundancy of superfund sites and high concentration of contaminants in GW recourses of Puerto Rico are suspected to be responsible for increasing rate of pre-term birth (highest amongst US states and territories) in the island (Mathews and MacDorman, 2011). However, since 2006, this rate has been declined from 20% to 11.4% due to remediation actions and enhanced awareness of habitant regarding water-borne diseases (March of Dimes Website, 2016; Rutigliano, 2016).

As it appears in Fig. 4, there is abundance of superfund sites in NPR mainly due to industrial activities, improper management of landfills, accidental spills, unidentified waste disposals, or residential septic systems. Most of these sites are located in upper aquifer of North Coast Limestone aquifer system. On the border of Arecibo and Barceloneta, there are 3 superfund sites which indicated high level of contamination in groundwater of that area.

In karst aquifers of PR, analysis of contaminant transport requires more sophisticated approaches coupled with continuous field assessments and data collection (Hu et al., 2009). Yu et al. assessed the concentration of CVOCs such as Trichloroethylene (TCE), Perchloroethene (PCE), Carbon Tetrachloride (CT), Chloroform or Trichloromethane (TCM), and Methylene Chloride (DCM), in northern Puerto Rico based on historical data. They stated that the hydrogeological conditions of the karst aquifer (intrinsic properties and the biological environment) in addition to the source origin were greatly associated with the spatiotemporal distribution patterns of the CVOCs. Since water resources pollution in northern Puerto Rico has caused a negative social, economic and environmental impact, long-term, consistent monitoring of water quality in addition to implementation of remedial actions were suggested (Yu et al., 2015).

3.1.6. Pathogens

Pathogens such as viruses and bacteria are easily transported in karstic systems. This is because of high drainage potential of karst and lack of soil layers, which can act as filters. Hence, pathogens retain their activity in karst conduits for a long time (Vesper et al., 2003). Fecal coliforms, fecal streptococci, and often a particular emphasis on Escherichia coli have been studied more compared to viruses in karst aquifers. These organisms often indicate GW contamination caused by sewage or animal waste. Studies regarding bacterial contamination in West Virginia (Mathes, 2000), Ontario (Conboy and Goss, 2000), and western Ireland (Thorn and Coxon, 1992), show the presence of fecal bacteria in carbonate wells (White, 2016). Muldoon et al. (2016) assessed sources of fecal contamination in the Silurian dolomite aquifer of Wisconsin using enteric pathogens. Moreover, it was found out that high percentage of bacteria travels within karst media by adsorbing on small particulates (mostly clay) (Mahler et al., 2000; White, 2016).

Resources for assessment of pathogens in karst aquifers of Puerto Rico are scarce. By assessing the North Karst Belt Zone of Puerto Rico, Acosta-Colón suggested that mesofauna diversity can be a possible indicator of pathogenic and opportunistic species. It was also recommended that in addition to characterizing mesofauna and species, more knowledge about the diet of mesofauna organisms is needed to clearly identify species that can be used as indicators of cave microbial pathogens (Acosta-Colón, 2016).

3.1.7. Chloride

Coastal aquifers, such as north coast limestone aquifer of Puerto Rico, are susceptible to seawater intrusion which can increase the salinity (Chloride concentration) of groundwater (Arfib et al., 2007; Mongelli et al., 2013). Having a hydraulic connection to the sea, karstic-coastal aquifers can be characterized by having groundwater flow in conduits, sub-marine freshwater springs and intrusion of seawater through the aquifer via conduit networks (Fleury et al., 2007). Due to susceptibility of karstic-coastal aquifers, developing a sustainable plan by using integrated models for managing and monitoring water resources within these aquifers is essential (Sreekanth and Datta, 2015).

Same as pathogens, not much research has been conducted regarding seawater intrusion in coastal karst aquifer of Puerto Rico. Although Zack et al. stated that the freshwater-saltwater interface of the North Coast limestone aquifer of PR is probably seaward and freshwater discharges to the seabed, by extracting water from public supply wells which results in decreasing GW level, seawater intrusion can be a potential problem in NPR (Zack et al., 1987). Not much chloride concentration data is available within USGS database and other resources. Hence, conducting a field sampling for a few years is recommended to assess the potential risk of chloride-based GW contamination.

3.2. Remediation strategies in karst aquifers

Based on increased understanding of contaminant transport behaviors and efficacies of various remediation technologies, USEPA emphasizes that combination of source and dissolved phase (plume) remediation is preferable strategy for remediation of many sites (USEPA, 2013). However, in addition to the challenges in costs and the extend of the treatment efficiencies arising at different sites, remediation in karst is even more challenging than in other hydrogological environments due to the presence of highly permeable conduits of unknown extent (Parise et al., 2015) (Fig. 5).

Fig. 5.

Preferred remediation techniques for a) source zone treatment and b) dissolved phase (plume) treatment implemented at superfund sites (MNA and VEB stand for Monitored Natural Attenuation and Vertical Engineered Barrier, respectively. Other GW Remedies include Institutional control) – modified from USEPA (2013).

For example, Sinreich (2014) stated that at the field scale, selecting the specific remediation technique for each contaminant requires comparative tracing experiments using contaminant surrogates. In spite of the dominant preferential flow pathways, solute and colloid tracers interact with aquifer material. Hence, if reactive and/or non-persistent contaminants are present in conduits, hydro-chemical properties of contaminants are more important than aquifer intrinsic vulnerability in the transport process and arrival of contaminants at karst springs. Most remediation processes in karst have lower efficiency compared to other aquifer types because of their time-dependent characteristics. Sinreich (2014) also suggested that the assumption of high vulnerability of karst may not be true for all conditions and should be identified based on fate and transport of specific contaminants. Thus, no preference should be considered for karst GW treatment methods because they can have different efficiency and cost based on contaminant properties and site condition.

Another example related to pump and treat application (Field, 2017), confirmed varying levels of success due to extreme heterogeneity and anisotropy of karst, limited diameter/length of the conduits and the behavior of flow regime within the solution conduits. It was realized that pump and treat was barely used by agencies for remediating contaminated karst GW (USEPA, 2013). As concluded by Parise et al. (2015), main difficulties with achieving feasible plume containment in karst with pump and treat systems are finding all preferential flow paths of the dissolved contaminant to the points of compliance and hydraulically stopping this transport by pumping both contaminated and clean water.

Due to all these challenges, it was pointed out that contaminated karst aquifers tend to fall under the special category of Technical Impracticability (TI) waivers. This means regardless of how much effort and financial resources are expended, actual remediation cannot be accomplished because of technical implausibility especially for aquifers contaminated by DNAPLs (Field, 2017; USEPA, 2012). It was found that most of the case studies associated with TI waivers were those in karstic hydrogeological setting (Deeb et al., 2011).

However, in situ thermal treatment, in situ chemical oxidation, and in situ bioremediation are increasingly being applied at fractured rock and karst site for source and plume remediation while point-of-use and point-of-entry has been evaluated and proposed strategy for karst groundwater treatment (Randrianarivelo et al., 2017).

3.2.1. Source and plume remediation

In-situ thermal treatment, In Situ Chemical Oxidation (ISCO), and in situ bioremediation are the main in situ remediation methods which have commonly been used in karst aquifers with certain limitations (Beyke et al., 2014; Parise et al., 2015). Studies have shown that identifying the location of conduits and major fractures is necessary for an efficient remedial action. It was reported that Electrical Resistance Heating technique was successfully implemented for removing DNAPLs and TCE in a karst aquifer in Alabama but preferential pathways within karst aquifers leading to high seepage velocity can cause heat loss (Hodges et al., 2014). The preferential flow pathways within karst system which can disperse injected materials, usually are problematic for other treatment methods such as ISCO and bioremediation as well.

Regarding in situ bioremediation, Hashim et al. pointed out that biological or biochemical methods which use microbes and nutrients for bio-precipitation, enzymatic oxidations, bio-surfactants and sulfate reductions for removing heavy metals, seems to be applicable and efficient in karst because injecting nutrients and electron donors is relatively inexpensive and non-toxic (Hashim et al., 2011). Byl and Williams stated that in areas isolated from the major groundwater flow paths, low migration of contaminant may possible to allow biodegradation of chlorinated ethenes and fuel if favorable microorganisms, food sources, and geochemical conditions are present (Byl et al., 2001; Byl and Williams, 2000). The potential for biodegradation of chlorinated organic solvents was in a karst aquifer was confirmed at the TCE contaminated site at Lewisburg, Tennessee indicating that natural attenuation should not be disregarded (Byl et al., 2002). Also, Randrianarivelo et al. conducted remedial investigations of a karst aquifer in Pennsylvania. They selected chemical oxidation in conjunction with monitored natural attenuation or land use control as preferred remediation method alternative based on the hydrogeological conditions of the site and the extent of the delineated contamination plumes (Randrianarivelo et al., 2017).

Regarding ISCO remediation method, by assuming that the contaminated area is well identified and the injection fluid has the right dosage and residence time, the possibility of delivering injection fluid to the contaminated area with minimal error is the major challenge, similar to other fluid-based remediation methods in karst aquifers. Moreover, for achieving the highest efficiency in treating contaminants diffused into the rock matrix or moving with slow advective transport, oxidizing agents should remain in the contaminated area. However, rapid movement of water through preferential flow pathways dilutes the oxidizing agents. Thus, it can be asserted that application of ISCO in karst aquifers is limited. Regardless of limitations in applying treatment methods for karst aquifers, Randrianarivelo et al. (2017) pointed out chemical oxidation as a beneficial techniques for GW remediation.

The main challenge associated with these techniques is to identify the zone that requires treatment. Based on site conditions and the type of contaminants, the most effective technology should be employed. Assessment of remediation technique requires an appropriate monitoring approach (for locations consist of springs, streams, extraction systems, and previously tested wells) and comprehensive hydrogeological and water quality sampling data (Randrianarivelo et al., 2017).

3.2.2. Remediation by mitigating exposure pathways

Exposure pathways often play a crucial role in spreading the contamination through karst aquifers. Steele-Valentin and Padilla delineated the potential exposure pathway of contaminants in Vega Alta, Puerto Rico. The pathway starts from superfund sites and is directed toward Atlantic Ocean. Moreover, it covers a noticeable lateral extension which impacts wells, wetlands and streams in the area (Steele-Valentín and Padilla, 2009). Because of interconnectivity between conduits and fractures within karst aquifer, controlling exposure pathways is more difficult compared to other aquifer types. However, mitigating exposure pathways can be done by treating at the tap, replacing drinking water supplies, treating spring flow using active and passive methods, land cover control using fences, signage, deed restriction and local law enforcement. In spite of their high treatment capabilities, these techniques often require long-term operation and maintenance costs (Randrianarivelo et al., 2017). One of the promising, cost-effective methods for minimizing the exposure pathways is electrochemical treatment: environmentally friendly (option of using solar power), independent (no other remedial actions needed) and controllable with regard to the rates of redox reactions. The high removal percentage for various contaminants in karst groundwater has been achieved in laboratory and pilot scale studies (Fallahpour, 2016; Gregor et al., 2017; Mao et al., 2012).

4. Application of modeling in karst

In order to predict the behavior aquifers based on hydrological variations, groundwater models have been developed by hydrogeologists and water resources scientists. In addition, some models are developed to chemically analyze the water quality and to simulate fate and transport of contaminants. A groundwater flow model is able to exhibit precise representation of hydrological and geological systems and also it can give a real insight into relationship and interactions between system elements. Modeling using computer programs can be truly beneficial when there are karst aquifers in the case study location. This is mainly due to the fact that karst aquifers are very heterogeneous and anisotropic and have a complex structure. Ergo, developing a relatively sophisticated model is the best option for simulating these types of aquifers.

4.1. Model parameters and development

Depending on the soil type and water table level, the percolation rate regarding the movement of water from saturated zone to ground-water differs (Ritchey and Rumbaugh, 1996). Additionally, the impact of human interferences with natural water cycle which can be caused mostly by irrigation and pumping water from wells, should be taken into account. Actually, a groundwater model can determine how much it is perilous for an aquifer and also for the ecosystem if certain level of human interference exists. This can help developing water resources management plans that can not only help optimizing water extraction, but also can preserve the environment and natural resources (Drew and Hötzl, 1999). Other physical parameters such as topographical and geological information of the region that is going to be modeled should be given to the modeling platform (Peterson and Wicks, 2006).

Anisotropy of aquifers regarding hydraulic conductivity, which is a parameter that can have a dissimilar value in each direction, can only be considered in two or three-dimensional models. Nowadays, by developing computer programs and also due to the need of acquiring more valid results as the output of modeling, three-dimensional models are more acceptable despite the possible complex procedure of setting them up (Anderson et al., 2015). In fact, while one-dimensional models can be applied for vertical flow in multiple horizontal layers and two-dimensional models considers water flow in a vertical plain and this is repeated in multiple parallel vertical plains as well, a three-dimensional model subdivides the flow region into smaller cells with possible different properties regarding aquifer condition, soil characteristics and water flow (Ebrahim, 2014).

Mostly, numerical analysis and tools should be used to solve complex differential equations of groundwater flow. In fact, a groundwater flow model is able to represent conceptual model of an aquifer mathematically and this mathematical representation enables researchers to solve the governing equations numerically by computers (Ebrahim, 2014; World Meteorological Organization, 2009). Using numerical solutions for solving groundwater flow equations in a three-dimensional scale is beneficial for models which follow the flow domain discretization approach. Usually, in a groundwater flow model, hydraulic head at each cell center and groundwater flow rate between cells can be considered as outcomes. Moreover, impacts on streamflow because of pumping or long-term impacts of current pumping can be assessed. Checking the consistency of datasets and parameters and also defining e framework for future studies related to groundwater and aquifer condition are two prominent products of a groundwater model (Hartmann et al., 2014; Ritchey and Rumbaugh, 1996).

Researchers and hydrogeologists have tried to develop groundwater models which can predict and simulate the groundwater flow in the karst aquifers in a regional or local scale. Regional groundwater models are capable of optimizing groundwater resources development plans, analyzing water budget of aquifers and assessing regional flow systems. Zhou and Li (2011) published a review paper regarding regional groundwater modeling and discussed their characteristics and associates drawbacks. Sauter assessed the quantification and prediction of regional groundwater flow and transport in a karst aquifer in Germany. He discussed how the most appropriate modeling tool can be selected and how it can be used for simulation for a specific case study location. By analyzing spring flow, spatiotemporal variations of groundwater levels and hydraulic parameters, the model was able to successfully describe the karst aquifer system in the studied location (Sauter, 1992). Fig. 6 depicts the schematic diagram of the process of developing a groundwater model.

Fig. 6.

Schematic diagram representing the process of developing a groundwater model – modified from World Meteorological Organization (2009).

4.2. Spatially lumped models and distributed parameter models

Ghasemizadeh et al. categorized groundwater models into different groups based on their capabilities and characteristics. Due to high level of heterogeneity and anisotropy in karst aquifers, accurately understanding of their behavior and distribution has always been challenging. This has forced modelers to employ approximate-based approaches and consequently consider the impact of the uncertainties caused by these approaches in their models. Hence, Spatially Lumped Models (SLM) and Distributed Models (DM) or Spatiotemporal Distributed Models (SDM) were introduced as two general approaches in modeling karst aquifers (Ghasemizadeh et al., 2012; World Meteorological Organization, 2009).

Spatially lumped models comprises of concentrated elements at spatially singular points; whereas, the elements are spatially distributed in distributed models. Hence, in distributed systems, physical quantities are spatially and temporally dependent. Spatially lumped (or global) models do not consider spatial alternation of flow patterns and are supposed to simulate a global chemical-hydrological response at the aquifer output point (for example spring discharge point) with regard to inputs of the aquifer (e.g. rivers, groundwater recharge points, net runoff etc.) (Ghasemizadeh et al., 2012; Singh, 2014). Assessing temporal alternations is an approach that spatially lumped models take to describe the global water balance and hydrological behavior of an aquifer. Moreover, in spatially lumped models, some factors that cause complexity in calculations and simulating are neglected due to simplifying assumptions and hence, using only the global parameters in simple ordinary linear differential equations and also low data requirements, are some of their properties that can be considered when trying to select the best modeling approach for groundwater flow and transport simulation. Although these models cannot produce accurate results, especially in karstic areas, they have been widely used by researchers in the areas where less data is available or only the prediction of groundwater flow, spring discharge and groundwater levels is necessary (Long, 2015; Panagopoulos, 2012). Hydrograph-Chemograph Analysis (Dewandel et al., 2003), Linear Storage Models (or Rainfall-Discharge Models) (Butscher and Huggenberger, 2008) and Soft Computing Techniques such as Fuzzy Logic (Mohd Adnan et al., 2013; Rezaei et al., 2013), Genetic Algorithm (McKinney and Lin, 1994; Nicklow et al., 2010) and Artificial Neural Network (ANN) (Hu et al., 2008), are three main approaches with regard to spatially lumped models that have been adopted by hydrological modelers.

In contrast, distributed models take complex parameters involved in groundwater flow and transport into account. In these models, dependent hydrological parameters and boundary conditions can be spatiotemporally variable and this will require the equations to be solved numerically and based on partial differential equations (Asher et al., 2015; Kuniansky, 2016). Also, due to the fact that all variables should be defied to the system, collecting more data and paying careful attention to details in this type of modeling is demanded which can make it more challenging (Dong et al., 2012; Long and Gilcrease, 2009).

For karst aquifer modeling, different type of distributed models based on the level of simplified assumptions have been used. Each of the developed models and methods treats complexity of karst aquifer differently and simulates groundwater flow based on its own logic and assumptions. Equivalent Porous Medium (EPM), Double Porosity (or Continuum) Method (DPM), Discrete Fracture Network (DFN), Discrete Channel (or Conduit) Network (DCN) and Hybrid Models (HM) are five common modeling approaches in distributed systems that have their own characteristics which will be discussed shortly in the following (Ghasemizadeh et al., 2012). DFN can be subcategorized into Discrete singular fracture set approach (DSFS) and Discrete multiple fracture set (DMFS) approaches. Sometimes, EPM and DPM are also mentioned as Single continuum porous equivalent approach (SCPE) and Double Continuum porous equivalent approach (DCPE) respectively in the literature. It is worth mentioning that employing Hybrid Models, which are the result of integrating discrete models and EPM approach and are also called coupled continuum pipe flow models, can be beneficial in many cases regarding modeling complex hydrological systems such as karst aquifers (Kiraly, 1998; Liedl et al., 2003). Fig. 7 demonstrated schematic configuration of the aforementioned distributed modeling approaches.

Fig. 7.

Distributed parameter modeling methods for karst aquifers – modified from Kuniansky (2016).

Bauer et al. have developed a numerical model to describe the influence of exchange flow between conduits and fissured system. They found out that under conditions of early karst evolution, conduit development is faster. Hence, exchange flow plays an important role in developing early karst evolution in limestone aquifers (Bauer et al., 2003). Also, some researchers have employed numerical modeling approaches to describe groundwater flow and transport in rough fractures (Briggs et al., 2014) and karst aquifers (Faulkner et al., 2009). However, modeling karst aquifers cannot only be carried out by numerical approaches (Barrett and Charbeneau, 1998). Furthermore, for simulating the genesis of karst aquifer systems, a numerical couple reactive network model, comprising of a 2D porous continuum flow module, a discrete pipe network for modeling flow and transport in the conduits and a carbonate dissolution module was developed by Clemens T et al. (1997).

4.3. Computer models and programs

MODFLOW is the most common groundwater modeling code that has been used due to its capability of simulating complex groundwater flows in a three-dimensional scale. Working based on finite difference method and block-centered approach, MODFLOW simulates the groundwater within the aquifer by considering different type of layers underground (i.e. confined, unconfined or both) and also different recharge or discharge sources such as areal recharge, groundwater flow to wells, runoff caused by rainfall, flow to riverbeds, spring flow etc. (Harbaugh, 2005). The initial version of MODFLOW (MODFLOW-2000) was released in the year 2000 and five years later, the updated version (MODFLOW-2005) started gaining attentions from groundwater modelers and hydrogeologists. To enhance the application of MODFLOW-2000, two models have been introduced by USGS which are VSF and MF2K-GWT. Basically, VSF is a version of MODFLOW-2000 that in addition to the ability of MODFLOW-2000 to model groundwater flow using a finite-difference method in a 3-D scale, can be applicable for variably saturated flow (VSF) (Thoms et al., 2006).

MF2K-GWT is an integrated model with MODFLOW-2000 that have the ability to simulate groundwater flow and solute transport (U.S. Geological Survey Website, 2012). Nevertheless, some programs that were independent to MODFLOW but developed by USGS were released as well such as HST3D (3-D Heat and Solute Transport Model) that is able to simulate ground-water flow and associated heat and solute transport in a 3D scale. Its capabilities can be used in analyzing problems associates with landfill leaching, seawater intrusion, hot-water geothermal systems etc. (Kipp, 1997).

The most updated version of MODFLOW program (MODFLOW 6) was released recently. In this program, any number of models can be used for concurrent simulation. These models can have inter-connection with each other and this can help solving complex hydrogeological problems in many cases such as the conditions in karst aquifers. Also, within this framework, multiple local GW models can be coupled with regional scale models (Langevin et al., 2017). Moreover, Conduit Flow Package (CFP), which can be coupled with MODFLOW-2005, can facilitate simulation of karstic geometry and GW movement and consequently, increase the accuracy of GW flow modeling in conduits (Shoemaker et al., 2007).

After releasing MODFLOW-2005, several associated models and packages were introduced and released based on numerous approaches and techniques. As an example, MT3D model, which is a modular, comprehensive, numerical three-dimensional solute transport model, was developed by USGS. This model has been designed to work very well regarding simulation of solute transport and reactive solute transport in complex hydrological systems. Being connected to MODFLOW, which is the USGS groundwater flow simulator, MT3D is able to simulate and analyze advection-dominated transport, especially solute transport, without refining new models (Bedekar et al., 2016). Lautz and Siegel used MT3D and MODFLOW to simulate groundwater and surface water mixing in the hyporheic zone. They took advantage of this model due to its ability to simulate advective transport and source and sink mixing of solutes (Lautz and Siegel, 2006).

Taking advantage of the features in MODFLOW and MT3D, a new computer program, SEAWAT, was released to assist hydrogeologists in simulating three-dimensional, variable-density and transient ground-water flow that can be coupled with solute transport. In the last version of SEAWAT (version 4), the effect of fluid viscosity and density fluctuations can be considered in simulation of groundwater flow and solute transport. This will allow the users to recognize this model as a tool that can be used in a wide range of simulation practices including seawater intrusion in coastal aquifers (Langevin, 2009; Langevin et al., 2008). Xu employed SEAWAT in his dissertation to study seawater intrusion into a coastal karstic aquifer in Florida and achieved accurate results. It is worth mentioning that seawater intrusion can be considered as a substantial source of brackish water in coastal aquifers such as karst aquifer in north coast of Puerto Rico (Xu, 2016).

In addition, FEFLOW (Finite Element Subsurface Flow System) is a finite-element package for simulating 3D and 2D fluid density-coupled flow, contaminant mass (salinity) and heat transport in the subsurface. It has several applications including regional groundwater management, saltwater intrusion, seepage through dams and levees, land use and climate change scenarios, groundwater remediation and natural attenuation and also groundwater-surface water interaction. As an example, a study has been conducted to simulate groundwater dynamics in an irrigation and drainage network in Uzbekistan using FEFLOW. After model calibration and validation, the results show high level of accuracy and can be used for hydrogeological management plans (Diersch, 2014; Khalid Awan et al., 2015).

SUTRA is another model that has been released for simulating 2-D saturated-unsaturated, fluid-density-dependent flow with energy transport or chemically-reactive single-species solute transport capable of analyzing saltwater intrusion and energy transport. It uses a 2D hybrid finite-element and integrated finite-difference approach to approximate the governing flow and transport equations that explain the two interdependent processes. It should be noted that the 3D version of this model has also been released recently. In SUTRA's Version 2.2 specification of time-dependent boundary conditions can be identified without programming FORTRAN code. SUTRA, can also describe chemical species transport including absorption, production and decay processes and assess well performance and pumping test data (Voss and Provost, 2002). For instance, Hussain et al. (2015) used SUTRA in their paper to study coastal aquifer systems that are subjected to seawater intrusion.

Visual MODFLOW Flex platform, an integrated modeling environment that connects MODFLOW and MT3D, is able to simulate complex 3D groundwater flow and contaminant transport. Its graphical user interface and 3D visualization capabilities in addition to its ability to simulate groundwater flow and contaminant transport can gain attention of hydrological and groundwater modelers. As an example of work, Varghese et al. successfully developed a Visual MODFLOW Flex model for simulation of groundwater flow in a region in India (Kumar and Singh, 2015; Varghese et al., 2015).

CHEMFLO-2000, which is interactive software for simulating water and chemical movement in unsaturated soils, enables users to simulate groundwater flow and chemical fate and transport in vadose zones. The model can be used as a tool that can enhance the understanding of unsaturated flow and transport processes. In this model, water movement and chemical transport are modeled using the Richards and the convection-dispersion equations, respectively. The equations are solved numerically using the finite differences approach (Nofziger and Wu, 2003).

Another 3D finite-element based model for simulating flow and transport is 3DFEMFAT. This model works for saturated/unsaturated heterogeneous and anisotropic media. Its typical applications include infiltration, agriculture pesticides, sanitary landfill, hazardous waste disposal sites, density-induced flow and transport, saltwater intrusion, etc. Its flexibility and feasibility in simulating a wide range of practical problems especially by employing its transport module, has made it valuable software for researchers and transport modelers. Also its application in studying seawater intrusion in coastal aquifer has been verified by some scientists (Lathashri and Mahesha, 2016; Park et al., 2012).

Regarding surface water and groundwater interaction which was discussed in the previous sections in detail, GSFLOW (Groundwater and Surface-water FLOW) was released by USGS in 2008 as an integrated tool that is able to couple groundwater and surface water flow models by taking advantage of the approaches used in USGS Precipitation- Runoff Modeling System (PRMS) and the USGS Modular Groundwater Flow Model (MODFLOW and MODFLOW-NWT). Meteorological and hydrological data such as rainfall, sunny hours and temperature in addition to groundwater stresses and initial/boundary conditions are involved as inputs for the process of simulation in this model. GSFLOW can also take into account the impact of land cover change, climate change and groundwater extraction on surface water and groundwater flow for spatiotemporally variable situations (Markstrom et al., 2008). However, regarding its limitations, it was asserted by researchers that its ability to simulate surface water and groundwater in karst aquifers with high level of heterogeneity is not guaranteed (Fulton et al., 2015).

By taking advantage of a control volume finite-difference method, MODFLOW-USG (Un-Saturated Grid version of MODFLOW) is able to simulate groundwater flow and its related processes. This version of MODFLOW supports different types of structured and unstructured grids. This capability is extremely useful when high resolution along rivers and around wells is needed. In addition, MODFLOW-USG couples Connected Linear Network (CLN) process to Groundwater Flow (GWF) process, which was introduced in MODFLOW-2005, to analyze and simulate the influence of karst conduits and multi-node wells. Hence, this version can help modelers to gain a deeper understanding about karst systems and conduit networks (Panday et al., 2013). Moreover, for the purpose of generating layered quadtree grids that can be used in MODFLOW-USG or other similar numerical models, a new computer program, GRIDGEN, was developed by Lien et al. in 2015. After reading a 3-D base grid, GRIDGEN continues dividing into refinement features, which has already been provided by user, until reaching the desired refinement level. After finishing the process of gridding, a tree structure file will be created and can be used in numerical models such as MODFLOW-USG. This model was used for assessing the Biscayne aquifer in southern Florida in which karst aquifers are abundant (Lien et al., 2014).

Developing a Newton-Raphson Formulation for MODFLOW-2005 for offering an enhanced solution for problems related to groundwater flow in unconfined aquifers, MODFLOW-NWT has been introduced and developed by Niswonger et al. (2011). Its main application in addition to Surface-Water Routing (Hughes et al., 2012) and Seawater Intrusion (Bakker et al., 2013) can be described as its ability to solve problems that are coupled with drying and rewetting nonlinearities in equations that govern groundwater flow in unconfined aquifers.

A recently developed model similar to MODFLOW but with a wider range of applicability in describing hydrological systems is Rainfall-Response Aquifer and Watershed Flow Model (RRAWFLOW). This lumped-parameter model receives hydrological inputs such as rainfall, recharge and discharge etc. and is able to simulate groundwater level, streamflow and spring flow. It also can be used for modeling solute transport in aquifers and assessing system response to hydrological events (Long, 2015). For classification of karst aquifers and characterizing time-variant systems, Long and Mahler developed and used this model in 2013. This model was used to predict and classify hydraulic responses to recharge in two karst aquifers in Texas and South Dakota, USA (Long and Mahler, 2013).

Usually, groundwater flow and contaminant transport models are used simultaneously using software platforms such as GMS. Several researchers have done flow and transport analysis (e.g. using MODFLOW and MT3D) and have achieved accurate and valid results (Abdalla and Khalaf, 2015; Bora and Borah, 2016). Also, few scientists have studied the groundwater flow and contaminant transport in karstic aquifer of northern Puerto Rico using GMS and their modeling results show its capability in analyzing and describing hydrological systems with complex properties such as high level of heterogeneity and anisotropy (Ghasemizadeh, 2015; Ghasemizadeh et al., 2016; Maihemuti et al., 2015). Table 2 elaborates the characteristics and application of aforementioned most commonly used groundwater modeling codes.

Table 2.

Prevalent groundwater models that have been used for simulating groundwater flow and contaminant transport. FE and FD represent Finite Element and Finite Difference respectively.

Model/Software	Modeling technique	Focus			Application and advantages/Ref.	Ease of use	Accuracy for karst modeling
		GW flow	Solute transport	Heat transport
3DFEMFAT	FE	*	*		GW modeling in saturated/unsaturated heterogeneous and anisotropic media, simulation of infiltration, agriculture pesticides, sanitary landfill, hazardous waste disposal sites, density-induced flow and transport, seawater intrusion etc. (Lathashri and Mahesha, 2016; Park et al., 2012)	High	Medium
AQUA3D	FE	*	*	*	3D groundwater flow and transport simulation for homogeneous and anisotropic flow conditions, simulation of heat and contaminant transport by taking into account the effect of dispersion	Medium to High	Low
CHEMFLO	FD	*	*		Simulation of water movement and chemical fate and transport in vadose zones and layered soil by employing improved numerical methods (Nofziger and Wu, 2003)	Low	Medium
FEFLOW	FE	*	*	*	regional groundwater management, saltwater intrusion, seepage through dams and levees, land use and climate change scenarios, groundwater remediation and natural attenuation, groundwater-surface water interaction (Diersch, 2014; Khalid Awan et al., 2015)	High	Medium
GSFLOW	FD	*			Coupled Groundwater and surface water model which can assess the hydrological behavior based on land use change, climate variability and groundwater withdrawals (Markstrom et al., 2008)	Medium	Low
HST3D	FD	*	*	*	sub-surface-waste injection, landfill leaching, saltwater intrusion, freshwater recharge and recovery, radioactive-waste disposal, hot water geothermal systems, and subsurface-energy storage (Kipp, 1997)	Medium	Medium
MODFLOW	FD	*			Simulation of steady or unsteady flow in complex flow system with irregular geometry, Simulation of flow from external stresses in a confined or unconfined aquifer (Harbaugh, 2005), High applicability for karst aquifers if it couples with CFP package	High	Medium to High
MODFLOW-NWT	FD	*			Surface water and groundwater interactions, seawater intrusion and solving problems related to drying and rewetting nonlinearities of the unconfined GW flow equation (Niswonger et al., 2011)	Medium	Low
MODFLOW-OWHM	FD	*			Simulation, analysis, and management of human and natural water movement within a physically-based supply-and-demand framework, seawater intrusion, conjunctive use of groundwater and surface water (Hanson et al., 2014)	Low to Medium	Low
MODFLOW-USG	FD	*			Unstructured grid version of MODFLOW for simulating GW flow and other related processes, simulation of the effects of multi-node wells, karst conduits and tile drains (Panday et al., 2013)	High	Medium to High
MT3D	FD		*		simulation of solute transport and reactive solute transport in complex hydrological systems and analyzing advection-dominated solute transport (Bedekar et al., 2016)	High	Medium to High
SEAWAT	FD	*	*	*	3D simulation of variable density, transient groundwater flow in porous media coupled with multi-species solute and heat transport, seawater intrusion in coastal aquifers (Langevin, 2009; Langevin et al., 2008; Post, 2011)	High	Medium
SURTA	FE	*	*	*	Simulation of saturated-unsaturated, fluid-density-dependent groundwater flow with energy transport or chemically-reactive single-species solute transport (Voss and Provost, 2002)	Medium	Low to Medium

4.4. Equivalent Porous Media (EPM) method

Several approaches have been followed to achieve accurate and valid results with acceptable efficiency at the same time. For some cases, simulation of groundwater hydraulic and contaminant transport in karst aquifers has been carried out by employing Equivalent Porous Media (EPM) method (Scanlon et al., 2003). Basically, using EPM approach for modeling a karst aquifer means considering simplifying assumptions in order to make the model more practical and applicable. Ghasemizadeh et al. developed their model based in EPM approach and found out that its result is acceptable for predicting water table fluctuations. Although their EPM-based model was not expected to be accurate enough for contaminant transport, they found good agreement between their model output and actual data regarding spreading TCE contaminant (Ghasemizadeh et al., 2015). Furthermore, in another study, by employing drainage features in regional groundwater flow modeling in karstic aquifer of northern Puerto Rico, Ghasemizadeh et al. asserted that they were able to improve their simulation by assigning arrays of adjacent model cells with drains to simulate conduits. They suggested that using this feature can be truly helpful especially when there is not sufficient data for conduit characteristics. Similarly, Maihemuti et al. developed a MODFLOW model using EPM approach for assessing karst aquifer system and groundwater resources for a case study location in northern Puerto Rico. Their model was supposed to predict the karst system response to rainfall events and high pumping demands and also to describe the hydrological behavior of the existing aquifer. They concluded that although there is high potential of conduit dominated flow, the result of their EPM-based approach is reliable in representing the hydrodynamics of the karst aquifer in their case study location (Maihemuti et al., 2015).

4.5. How remote sensing can improve karst GW modeling?

Remote sensing tools, either coupled with modeling methods or be used separate, can be very beneficial because of their strong data analysis and management capability which allows assessment of several datasets and layers simultaneously. However, lack of high resolution data for local studies can be problematic in some cases (Manda and Gross, 2006; Theilen-Willige et al., 2014). Using Geographic Information System (GIS) as a tool in groundwater modeling procedure, can be truly helpful; because all parameters such as distribution of rainfall, groundwater recharge and discharge and also land cover are defined within a spatial context (Singh and Fiorentino, 1996).

Several researchers have took advantage of this powerful tool directly or as a parallel method in integrated approaches (Dar et al., 2010; Nampak et al., 2014). Alonso-Contes (2011) used remote sensing and advanced digital image processing techniques to delineate karst features which can enhance the understanding with regard to hydrogeology of the Tanamá River and Rio Grande de Arecibo catchments located in the north coast tertiary basin of Puerto Rico. Basically, remote sensing tools have assisted the author in lineament mapping for GW exploration.

Also, Manda and Gross have employed GIS analysis to characterize solution conduits in karstic areas. Based on their study, they have shown that GIS-based methods can be used for determining depths, dimensions, shapes, apertures and connectivity of potential conduits and also for describing physical characteristics that have an effect on the groundwater flow in karst aquifers (Manda and Gross, 2006). In addition, Theilen-Willige et al. employed GIS and remote sensing methods by analyzing satellite data in order to detect of near-surface faults and fracture zones that can lead to dissolution processes in conduits of karst aquifers (Theilen-Willige et al., 2014).

Numerous researchers have taken advantage of remote sensing techniques to improve their models and solve some un-answered and complex problems by increasing the accuracy of prediction and also by taking into account other hydrological phenomena (Ashraf and Ahmad, 2012; Machiwal et al., 2012; Thakur et al., 2016; Xu et al., 2011). Table 3 elaborates the potential application of GIS and remote sensing in different phases of integrated groundwater modeling.

Table 3.

The potential role of GIS and remote sensing in different steps of groundwater modeling procedure – modified from Ashraf and Ahmad (2012).

Phase	GIS functions	Modeling steps
Data Collection and Analysis	Data input, Digitization, Data conversion (import/export, Coordinate transformation, Map retrieval	Groundwater and hydrological data collection
Developing Conceptual Model	Conversion of vector and raster layers, Data integration, Image processing, buffering, Surface generation, Linking of spatial and attribute data	Developing conceptual model
Model Design	Map calculations, Neighborhood operations, Interpolation, Theissen polygons, buffering, Surface generation	Delineating boundary conditions, Mesh generation, 3D layering of the aquifer
Model Calibration	Data layers integration	Parameter zonation, Recharge estimation, Water balance
	Overlay analysis	Steady-state and Transient-state simulations
	Statistical analysis	Parameters estimation
Model Generalization, Predictions and Result Presentation	Data retrieval	Prediction, Assessing different scenarios
	Data visualization, Presentation of simulated results	Map composition

5. Conclusion

It is vital to employ a comprehensive and efficient approach for managing water resources in regions where groundwater is the main source of water supply and vulnerable to contamination (e.g. karst aquifers). Limestone karstic aquifer systems of northern Puerto Rico has been experiencing high level of contamination resulting in accelerating rate of preterm births in the island in addition to other human health related disorders. Several wells and sites were listed as the National Priority List (NPL) Superfund Sites and remediation action for treating water in these sites are ongoing. By considering this island as a case study location and also by focusing on karst aquifers as the most complicated and hard-to-analyze forms of aquifers, a review study was presented to qualitatively and quantitatively assess groundwater resources. After an explanation of karst systems and their associated study methods in addition to surface water and GW interaction (SWGWI), a brief review discussion on groundwater contamination and risk assessment was presented. Potential contamination threats in karst aquifers were discussed and different remediation techniques were evaluated.

Furthermore, a comprehensive discussion on existing groundwater modeling methods with regard to their application, advantages and disadvantages was presented. Mostly, numerical modeling approaches are taken by modelers to simulate complex groundwater systems. Numerical modeling tools often take advantage of finite-difference and finite-element techniques to solve complicated equations that governs hydrological system dynamics in an aquifer. For karst aquifers, lumped models and spatially distributed models are considered as two general modeling approaches that can be used in certain conditions. In addition, various computer-based models such as MODFLOW have been explained and evaluated. A combined set of models can be used to simulate groundwater flow and contaminant transport either in steady state or transient form. Additionally, application of remote sensing and GIS in assessing water resources in karst and also for modeling purposes was explained.

Overall, it can be asserted that karst aquifers have the most complex behavior compared to other aquifer types. Because of their high level of permeability, they are always prone to contamination. Hence, employing different methods for studying karst aquifer and their interrelationship with surface water and ground surface may lead to different level of accuracy. Using integrated techniques by taking advantage of different methods for a case study and collecting sufficient data is costly, but will greatly enhance the understanding and investigation of a certain karst aquifer. Such integrated techniques and data collection processes are recommended for north coast karst aquifer of Puerto Rico where the human health and ecosystem are at high risk. From sustainable development perspective, environmental, economic and social impacts are consequences of water pollution in any region of the world. Hence, careful attention should be paid to preserve water resources.