Release of arsenic to deep groundwater in the Mekong Delta, Vietnam, linked to pumping-induced land subsidence

Abstract

Deep aquifers in South and Southeast Asia are increasingly exploited as presumed sources of pathogen- and arsenic-free water, although little is known of the processes that may compromise their long-term viability. We analyze a large area (>1,000 km²) of the Mekong Delta, Vietnam, in which arsenic is found pervasively in deep, Pliocene–Miocene-age aquifers, where nearly 900 wells at depths of 200–500 m are contaminated. There, intensive groundwater extraction is causing land subsidence of up to 3 cm/y as measured using satellite-based radar images from 2007 to 2010 and consistent with transient 3D aquifer simulations showing similar subsidence rates and total subsidence of up to 27 cm since 1988. We propose a previously unrecognized mechanism in which deep groundwater extraction is causing interbedded clays to compact and expel water containing dissolved arsenic or arsenic-mobilizing solutes (e.g., dissolved organic carbon and competing ions) to deep aquifers over decades. The implication for the broader Mekong Delta region, and potentially others like it across Asia, is that deep, untreated groundwater will not necessarily remain a safe source of drinking water.

Arsenic in groundwater poses a massive and growing human health threat throughout South and Southeast Asia. An estimated 100 million people (1) are chronically exposed to arsenic, a potent carcinogen also linked to a variety of other health risks in adults and children (2), through consumption of naturally contaminated groundwater. Despite widespread awareness of this crisis, groundwater exploitation continues to rise, with demand increasingly being met by deep wells (>150 m). Deep wells typically exhibit low arsenic concentrations and have been promoted as an alternative to those tapping contaminated shallow groundwater. “Dig deep to avoid arsenic” (3) has been touted as a safe answer to the provisioning of drinking water in Bangladesh, despite a lack of evidence that deep aquifers indeed remain uncontaminated under prescribed (4, 5) or unregulated pumping. In fact, recent studies indicate that arsenic occurrence may be on the rise where deep aquifers are intensively pumped in parts of Bangladesh, West Bengal, India and the Red River Delta, in northern Vietnam (6–8). In some cases, isolated deep arsenic contamination may be caused by downward leakage through well bores. However, in the Mekong Delta, in southern Vietnam, deep aquifers show pervasive arsenic contamination that may be directly linked to groundwater exploitation via a causal mechanism not previously considered and described presently.

Arsenic occurs naturally in sediments throughout the depth profile of the major river basins of South and Southeast Asia. Solid-phase arsenic is primarily released to groundwater during the microbially mediated reductive dissolution of ferric (hydr)oxides found in buried river-borne sediments. Dissolution is controlled by a suite of physicochemical conditions that vary widely within and among hydrogeologic units (9), largely as a result of variability in depositional and paleoclimatic conditions during their formation. Across basins, dissolved arsenic concentrations tend to be highest in the shallow (<100 m) subsurface (10), where the reactivity of host minerals and the organic carbon needed to dissolve them is also greatest (11, 12). As a result, considerable attention has been paid to contamination mechanisms in Holocene units (up to ∼0.011 Ma in age), where affected wells are most commonly found, and to older Pleistocene units (∼0.011–2.6 Ma), where they are usually more rare (6–8, 13). Little is known of arsenic occurrence in older Pliocene–Miocene-age (∼2.6–23 Ma) aquifers or in the thick sequences of interbedded confining clays (i.e., aquitards), which are known to mobilize high levels of dissolved arsenic in near-surface Holocene clays (12).

Here, we focus on the Mekong Delta, Vietnam, where heavily exploited Pliocene–Miocene-age aquifers are extensively contaminated at depths of 200–500 m. A recent nationwide survey of arsenic in wells conducted from 2002 to 2008 by the Department of Water Resources Management, Vietnam includes 42,921 observations in the Delta alone (Fig. 1). Whereas prior synoptic studies in the Delta have focused on near-river areas, where the highest population densities and arsenic concentrations are found (14–18), wells in this new survey were sampled in proportion to their abundance in all populated areas, providing unprecedented spatial coverage. Our analysis uses (i) the richness of these observations along with (ii) simulation of the spatio-temporally explicit flow and pumping history of the multiaquifer system and (iii) validation of pumping-induced compaction by radar remote sensing of land subsidence. This complementary suite of methods allows us to reveal a human-influenced contamination mechanism in deep aquifers.

Fig. 1.

Groundwater arsenic concentrations in the Mekong Delta, Vietnam. (A) Plan view. (B) North-looking perspective (vertical exaggeration, 150×), highlighting the focus area of this work. Topography and bedrock surface shown above and below zero elevation (mean sea level), respectively. Coastline is lightly dashed.

Results

In the Mekong Delta, groundwater is widely pumped from seven major aquifers ranging from Holocene to Miocene age. The delineation of aquifers and their ages used here is based on the work of the Division for Geological Mapping for the South of Vietnam, which used a suite of standard techniques including mud logging of drill cuttings, radiometric dating, analysis of microfossils, and geophysical surveys to describe the >1,000-m-deep, fault-blocked basin underlying the Delta and the complex stratigraphy of its fill. On the Vietnamese side of the Delta, numerous productive, sandy aquifers are separated by and embedded with similarly thick and laterally extensive sequences of significantly less-permeable clays. The confining nature of these clays is indicated by distinctly separable hydraulic heads in nested monitoring wells. Well nests, or sets of wells in which each measures the hydraulic head in one of four to six aquifers at the same location, are well distributed throughout the Delta. They allow for examination of both horizontal and vertical hydraulic gradients, and they indicate that several aquifers are experiencing widespread head declines (Supporting Information). Pumping wells, in use since the early 1900s, have become increasingly common since the early 1980s. Today these wells are widely used for a variety of domestic, agricultural, and industrial purposes. Throughout much of the Delta, deep aquifers are the most heavily exploited.

Arsenic occurrence in the Mekong Delta exhibits some classic characteristics observed in many other South and Southeast Asian river basins (Fig. 1A). Dissolved concentrations are highest (maximum 1,470 µg/L) in the shallow subsurface (<100 m), in close proximity (<5 km) to the main river and its distributaries, and drop off sharply with distance. Wells with concentrations up to 1,000 µg/L, two orders of magnitude greater than the 10 µg/L World Health Organization drinking water standard, are often less than 100 m away from others with no detectable arsenic. Pervasive arsenic is found in several extensive hot-spot regions.

The most prominent arsenic hot-spot region is located ∼50 km southwest of Ho Chi Minh City (HCMC) and is over 1,000 km² in extent. This “focus area” contains 1,059 wells with arsenic exceeding 10 µg/L (Fig. 1A). Based on stratigraphic cross-sections and the evident partitioning of these wells by depth (Fig. 1B), we divide wells in the focus area into two sets: those in shallow Holocene–Pleistocene aquifers (Fig. 2C) and those tapping the deep Pliocene–Miocene aquifers (Fig. 2D). The deep set (170–500 m) contains the majority (84%) of arsenic-contaminated wells.

Fig. 2.

(A) Locations of bores in cross-sections in the vicinity of the focus area. (B) Cross-section illustrating complex stratigraphy and distinction between shallow and deep zones of focus area. (C and D) Arsenic concentrations in wells of the shallow and deep zones. Wells with As <5 µg/L outside the focus area are shown in gray. (E) Pumping intensity of wells in the deep zone estimated by multiplying the age of each well by the 2007 unit pumping rate (pumping rate for each district and aquifer divided by the total number of wells). Wells in districts for which pumping data are unavailable are shown in gray.

The Mekong Delta focus area shows far more deep, contaminated wells than other major arsenic-affected regions of South and Southeast Asia. Prior surveys from Nepal, India, Bangladesh, and Vietnam’s Red River Delta (compiled in ref. 10) indicate that the fraction of contaminated wells diminishes with depth (Fig. 3). In those areas, wells were rarely found with arsenic in excess of 10 µg/L at depths greater than 200 m. Deep wells, however, tend to be less well represented in surveys, in terms of both total quantity and spatial coverage. As such, our understanding of the extent and mechanisms of deep arsenic contamination has remained incomplete.

Fig. 3.

Depth profile of groundwater arsenic occurrence in surveys of major affected areas in South and Southeast Asia.

Evidence suggests that deep contamination in the Mekong Delta focus area is unlikely to have been caused by the only presently acknowledged mechanism: pumping-induced vertical migration of arsenic or dissolved organic carbon (DOC), which can trigger arsenic release, from the surface or shallow subsurface (6, 7, 19–21). The number of wells above 10 µg/L in the deep zone is seven times greater than in the shallow zone. The mean concentration in the shallow zone is significantly less than in the deep zone (4 vs. 20 µg/L, respectively). There are no clusters of overlying shallow wells with concentrations exceeding those of contaminated wells at depth. Moreover, vertical velocities through the layered system are estimated, using the hydraulic head record and thicknesses of sand and clay in all 1,020 1-km² locations within the focus area, to be less than 0.2 m/y. Given that the mean distance between contaminated wells in the shallow and deep zones is 100 m, and considering a generous 28 y of travel time since the onset of increasing well installations, downward transport of dissolved arsenic between zones has not likely occurred over any significant area. Short-circuiting through leaky bores cannot account for the regional-scale contamination of deep wells in the focus area given the distribution of the concentration data just described and sound well sealing practices described in documentation from the Division for Water Resources Planning and Investigation for the South of Vietnam (DWRPIS).

Although downward transport of contaminants from the near-surface can be excluded, deep pumping since the mid 1990s has caused hydraulic heads in the deepest Pliocene–Miocene-age aquifers to decline by several meters. Such head declines induce compaction, as water that previously supported the mineral structure is removed by pumping. Pumping-induced compaction is most pronounced in clays, far more compressible than sands, which are effectively squeezed during persistent overexploitation of adjacent aquifers. Water expelled from compressible clays, including formations at depths of hundreds of meters, has met a large portion of pumping demand in confined aquifer systems around the world, as evidenced by resulting land subsidence (22). During compaction owing to pumping, dissolved arsenic and any other, potentially toxic, solutes stored in deep interbedded clays would be expelled into adjacent aquifers. Expelled solutes from clay could also include DOC (23) or competing ions that could promote arsenic dissolution or desorption within the aquifers.

According to the clay-compaction release mechanism, high densities of arsenic-contaminated wells should correspond to areas of significant cumulative pumping. Indeed, pumping rates among the 1,365 wells in the deep zone of the focus area are much greater and the median age of wells is twice that of those within the 10-km surrounding buffer. Deep wells in this buffer (1,218 total) largely have been installed since 1996 (Supporting Information) and do not show arsenic (only four wells with arsenic >10 µg/L). This implies that before significant pumping lasting a decade or more, deep aquifers are not pervasively contaminated through in situ processes or by solute diffusion out of interbedded clays. Rather, there is a likely ∼10-y time lag before contaminant arrival at pumping wells, consistent with the requisite travel time between compacting clays and these wells.

We compute compaction rates within aquifers and confining beds of the focus area and surrounding region using 3D transient aquifer simulation. We recreate the spatially explicit pumping history in each of the seven major aquifers according to 2007 pumping data from the DWRPIS. We assign typical low permeability values for interbedded clays along with high storage property values, which are known to be one to two orders of magnitude greater for clay than for sand. The model is calibrated to fit the historical trends in hydraulic head data in all aquifers measured from 1996 to 2008 in nested monitoring wells across the region. Observed head declines in these wells of up to 74 cm/y in the Delta and 230 cm/y in HCMC are well reproduced by our model (Supporting Information). Our simulations indicate land subsidence rates in the focus area of 1.1–2.4 cm/y, as pumping demand is met largely by water released from compacting clay storage. Simulated total subsidence is greater within the focus area (maximum 27 cm) compared with adjacent areas owing to the greater pumping intensity of wells within it (Fig. 2E). Pumping in HCMC, more than 50 km away, is not responsible for subsidence in the focus area. Measured hydraulic heads, interferometric synthetic aperture radar (InSAR)-based subsidence estimates, and aquifer simulation all indicate that the effect of HCMC pumping is local.

Pumping-induced clay compaction is measurable as land subsidence. We measure land subsidence rates using InSAR data collected by the Phased Array type L-band Synthetic Aperture Radar (PALSAR) instrument aboard the Advanced Land Observing Satellite (ALOS), because no ground-based measurement record is available. Two PALSAR tiles cover our study area and images of them were acquired over the period 2007–2010 every 2–12 mo. We form interferograms from all pairs of scenes that span a 1-y interval, selected to minimize seasonal effects, and average them to reduce atmospheric errors. Based on InSAR, subsidence is occurring at a rate of 1–3 cm/y (∼1.2–3.6 cm in the vertical assuming no horizontal deformation) relative to a coherent reference area near the Cambodian border and is highest in localized (1–3 km) subsidence bowls centered on many of the regions’ small cities (Fig. 4). Estimated errors in rates are spatially variable, ranging from ±0.5–1 cm/y (Supporting Information). Comparison of land subsidence rates based on InSAR and based on aquifer system simulation is shown in Fig. 4.

Fig. 4.

(A) InSAR-based line-of-sight land subsidence rate for 2007–2010 with superimposed focus area outline (white). Areas of low correlation are excluded. Data from the Alaska Satellite Facility, Japanese Aerospace Exploration Agency, Ministry of Economy, Trade and Industry (© JAXA, METI 2011). (B) Comparison of subsidence rates derived from InSAR and aquifer simulation. (C) Subsidence rates derived from aquifer simulation only. InSAR-based rates shown in B were upscaled to ∼1-km resolution using the median of values from the finer grid in A to facilitate comparison with the aquifer simulation results. Northeast and southwest corner areas fall outside of the aquifer simulation domain and are indicated in light gray.

Subsidence measured with InSAR is seen throughout the focus area and further extends to the north and south of it where either (i) we do not have arsenic measurements or (ii) wells have generally been pumping for less than 10 y. In the latter case, it seems that the onset of pumping-induced subsidence has begun, and arsenic transport to well screens may be in progress. Subsidence in HCMC is also evident. HCMC is, however, located outside of the Mekong floodplain such that provenance there differs from the Himalayan sediments responsible for most arsenic in wells of the Delta proper. Wells in the HCMC area are, not surprisingly, largely uncontaminated anywhere in the depth profile, and it seems that the city’s excessive pumping, although inducing subsidence, is therefore not causing release of arsenic from a deep source. Additional factors complicating the relationship between observed subsidence and arsenic occurrence may be related to the network of regional faults, local depositional conditions, and sulfide attenuation (24, 25).

The paleoclimatic record supports the occurrence of dissolved arsenic in deep clays. Holocene clays are known to maintain high concentrations of dissolved arsenic over millennia (12, 26). Modern climate features, notably tropical temperatures and precipitation, and high sea levels, are conducive to sustaining the biogeochemical conditions that are favorable to arsenic dissolution within Holocene clays. The Pleistocene was marked by frequent glaciations that substantially lowered global sea levels. In the Mekong Delta, dramatic changes occurred in vegetation types, flooding patterns, hydraulic gradients, and mineral weathering (27, 28). Evidence of these changes is seen in the abundance of ferric (hydr)oxides that give Pleistocene sands in Bangladesh their oft-noted brown or orange color (8, 10, 13, 29, 30). These oxidized deposits have a higher capacity for arsenic adsorption (4, 21, 31) and are associated with aquifers that are low in dissolved arsenic, which are found at greater depths in the Bengal Basin owing to higher sedimentation rates. In contrast to the Pleistocene in the Mekong Delta region, temperatures during the Pliocene–Miocene epochs were likely similar or elevated relative to the Holocene (27, 32), suggesting biogeochemical conditions may also have been like those responsible for mobilizing arsenic in shallow clay today.

Through our analysis, we put forth a conceptual model that describes the vertical distribution of arsenic seen in the Mekong Delta in terms of its historical development (Fig. 5). From Miocene times to the present, fresh clays rich in arsenic and organic carbon were broadly deposited. Upon burial, solid-phase arsenic provided a persistent source that contaminated pore waters trapped in thick clay units. Over millions of years of deposition, dissolved arsenic concentrations in the pore fluids of more permeable aquifer sands were drawn to low levels by advection, mixing, and dilution. Concurrently, clays lost some of their original solute load to diffusion through the limited connected pore network. Slow diffusion out of occluded pores and slow dissolution and desorption of the persistent solid-phase arsenic supply (12, 33), however, maintained a continuous dissolved arsenic load in deep clays, as found elsewhere in much older aquitards (>70 Ma) (34). When low-arsenic, deep aquifers were overpumped during recent decades, clay compaction began, leading to water containing arsenic or possibly other, arsenic-mobilizing solutes being squeezed out of dead-flow storage in confining clays to adjacent aquifers, a process taking a decade or more.

Fig. 5.

Conceptual model for depth distribution of arsenic in groundwater.

The implication of these findings for the Mekong Delta region, and potentially other arsenic-prone aquifer systems like it, is that deep, untreated groundwater is not a safe long-term water source. Deep wells that test clean upon installation, as do those bordering the focus area, may not remain arsenic-free over time as pumping promotes compaction and release of arsenic or arsenic-mobilizing solutes from deep clays. The potential for a deep source of arsenic, resulting from this unrecognized clay expulsion mechanism, should be considered as deep groundwater resources are further developed in these settings. In other confined aquifer systems around the world, solute release from compacting clays could also affect groundwater quality in as yet unknown ways.

Demand for deep groundwater is created by limited freshwater availability, arsenic-contaminated shallow groundwater, and nonpotable surface water. However, deep groundwater exploitation in the Mekong Delta presents new potential hazards: land subsidence, saline intrusion, and human-induced arsenic contamination. Management of water resources in the complex, deltaic aquifer systems of South and Southeast Asia should seek to minimize human exposure to all relevant threats including, but not limited to, those due to arsenic. To reduce the impacts of arsenic contamination from deep groundwater extraction, water managers should consider a suite of measures. These include first understanding the nature and extent of deep groundwater arsenic, limiting intensive extraction, treating or blending extracted groundwater to meet health standards, and possibly screening pumping wells over intervals of deep aquifers that are distant from confining clays, among other water management strategies aimed at health-risk reduction.

Materials and Methods

Here we provide a summary of the data and methods used in the paper. Greater detail is provided in Supporting Information.

Arsenic data were acquired from the Department of Water Resources Management (DWRM) in Hanoi, Vietnam. Samples were collected according to national standard TCVN 4556-88 and analyzed in certified laboratories (Vietnam Laboratory Accreditation Scheme) by hydride-generation atomic absorption spectroscopy (ISO 11969:1996). In the Mekong Delta, 50,532 wells were surveyed, of which 42,921 were considered, excluding samples with no recorded depth and very shallow (≤10 m), potentially open wells that may be exposed to surface conditions. Along with arsenic concentration measurements, depth and the Global Positioning System coordinates of each well, the year of installation was generally reported (202 missing values). Missing year of installation values were ignored in analyses involving well ages.

Ancillary hydrogeologic data were acquired from the DWRM and the DWRPIS of Vietnam, HCMC, Vietnam. These include (i) 10 stratigraphic cross-sections compiled from 81 well logs at ∼20-km spacing including mapped geologic faults, originally interpreted by the Division for Geological Mapping for the South of Vietnam; (ii) pumping rates in 22 districts, specific to the six major pumped aquifers (excluding the Holocene aquifer), from a 2007 survey of 5,282 wells, classified by purpose of water use; (iii) transient hydraulic head data from 45 nested monitoring wells, measured monthly over the period 1996–2008; and (iv) hydraulic head maps for the years 1987 (35) and 2004 (provided by DWRM).

The 3D extents of the major aquifer and confining units were determined based on the well logs and cross-sections. Clay pods embedded within aquifers seem to be discontinuous and were ignored in the interpolation of the major hydrogeologic unit boundaries and subsequent regional groundwater and subsidence simulations. Each data point in the DWRM survey was assigned to an aquifer according to its latitude, longitude, and well-screen elevation, taken as the well’s recorded depth relative to the Shuttle Radar Topography Mission (STRM) 90-m digital elevation model (DEM).

Estimates of vertical flow between zones of the focus area were made for all 1-km² cells within it. We considered the shortest reasonable distance through which shallow contamination could have been transported to depth: the middle of the Upper Pleistocene aquifer (the most shallow of three Pleistocene aquifers), where the bulk of shallow zone wells are found, to the top of the deep zone (i.e., the top of the Upper Pliocene aquifer). For each cell of the focus area, the effective hydraulic conductivity (K_effective) over this minimum transport distance was calculated using the harmonic mean of the (n) layer conductivities, scaled by their thicknesses (b) as follows:

where n is the number of layers. Hydraulic conductivity values for each layer type were initially chosen from well construction specification materials acquired from the DWRPIS (for aquifer type) and literature values (for confining unit type) and updated after calibrating the groundwater flow model (described below). Velocity, v, in each cell was calculated by Darcy’s law for flow through porous media:

where K_effective is the effective hydraulic conductivity defined above, I is the vertical hydraulic gradient component, and θ is the effective porosity, taken as the mean of literature values for the two materials, weighted by their relative thicknesses. The average gradient between the Upper Pleistocene and Upper Pliocene aquifers was calculated over two periods determined by data availability: 1987–1997 and 1997–2005. All parameters for the vertical flow calculations are provided in Supporting Information.

Groundwater flow simulation was conducted using the US Geological Survey’s vetted software MODFLOW-2005. The Interbed Storage (IBS1) package for MODFLOW was used to simulate compaction and land subsidence according to

where S_s is specific storage, b is thickness, and Δh is the temporal change in local hydraulic head values. Details of the model discretization, boundary conditions, assignment of pumping conditions, and calibration to monitoring well data in the seven major aquifers considered are provided in Supporting Information.

Radar imagery was acquired by the Alaska Satellite Facility from the Japanese Aerospace Exploration Agency for UPASS proposal ID: 589. Data were collected by the PALSAR instrument, a phased-array L-band synthetic aperture radar (23.8-cm wavelength) carried on the ALOS satellite, which has a 46-d repeat period. A total of 39 images covering the two tiles were analyzed. Interferograms were formed using a geodetically accurate, motion-compensating InSAR processor (36). Elevation correction, negligible in this nearly flat landscape, was made using the SRTM 90-m DEM. Results were resampled to a final resolution of ∼60 m. Orbital ramps were removed from interferograms by subtraction of a linear phase plane. The average phase was computed on coregistered stacks of all available 1-y interval interferograms. The stacking procedure was justified by inspection of the nested hydraulic head data, which consistently show linear average annual declines: stationary subsidence rates are expected.

InSAR-based subsidence rates were retained only for areas in the landscape with high correlation (a measure of radar signal quality), namely, developed areas that are elevated above the floodplain. Errors were estimated by taking the SD of the stacked phase in 400-pixel windows and range from ±0.5–1.0 cm, with lowest errors in highly urbanized, radar bright areas (Supporting Information). All subsidence rates and error estimates are reported in the satellite’s line-of-sight direction, which is approximately equivalent to, if slightly less than, the vertical rate.