Release of arsenic to deep groundwater in the MekongDelta, Vietnam, linked to pumping-inducedland subsidenceLaura E. Erbana, Steven M. Gorelicka,1, Howard A. Zebkerb, and Scott Fendorfa

Departments of aEnvironmental Earth System Science and bGeophysics, Stanford University, Stanford, CA 94305

Edited by Jerome Nriagu, University of Michigan, Ann Arbor, MI, and accepted by the Editorial Board July 1, 2013 (received for review January 11, 2013)

Deep aquifers in South and Southeast Asia are increasingly ex-ploited as presumed sources of pathogen- and arsenic-free water,although little is known of the processes that may compromisetheir long-term viability. We analyze a large area (>1,000 km2) ofthe Mekong Delta, Vietnam, in which arsenic is found pervasivelyin deep, Pliocene–Miocene-age aquifers, where nearly 900 wells atdepths of 200–500 m are contaminated. There, intensive ground-water extraction is causing land subsidence of up to 3 cm/y asmeasured using satellite-based radar images from 2007 to 2010and consistent with transient 3D aquifer simulations showing sim-ilar subsidence rates and total subsidence of up to 27 cm since1988. We propose a previously unrecognized mechanism in whichdeep groundwater extraction is causing interbedded clays to com-pact and expel water containing dissolved arsenic or arsenic-mo-bilizing solutes (e.g., dissolved organic carbon and competing ions)to deep aquifers over decades. The implication for the broaderMekong Delta region, and potentially others like it across Asia, isthat deep, untreated groundwater will not necessarily remain asafe source of drinking water.

groundwater contamination | clay compaction | InSAR | aquifer system

Arsenic in groundwater poses a massive and growing humanhealth threat throughout South and Southeast Asia. An

estimated 100 million people (1) are chronically exposed to ar-senic, a potent carcinogen also linked to a variety of other healthrisks in adults and children (2), through consumption of naturallycontaminated groundwater. Despite widespread awareness ofthis crisis, groundwater exploitation continues to rise, with de-mand increasingly being met by deep wells (>150 m). Deep wellstypically exhibit low arsenic concentrations and have been pro-moted as an alternative to those tapping contaminated shallowgroundwater. “Dig deep to avoid arsenic” (3) has been touted asa safe answer to the provisioning of drinking water in Bangla-desh, despite a lack of evidence that deep aquifers indeed remainuncontaminated under prescribed (4, 5) or unregulated pump-ing. In fact, recent studies indicate that arsenic occurrencemay be on the rise where deep aquifers are intensively pumpedin parts of Bangladesh, West Bengal, India and the Red RiverDelta, in northern Vietnam (6–8). In some cases, isolateddeep arsenic contamination may be caused by downwardleakage through well bores. However, in the Mekong Delta,in southern Vietnam, deep aquifers show pervasive arseniccontamination that may be directly linked to groundwaterexploitation via a causal mechanism not previously consideredand 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 duringthe microbially mediated reductive dissolution of ferric (hydr)oxides found in buried river-borne sediments. Dissolution iscontrolled by a suite of physicochemical conditions that varywidely within and among hydrogeologic units (9), largely asa result of variability in depositional and paleoclimatic con-ditions during their formation. Across basins, dissolved arsenic

concentrations tend to be highest in the shallow (<100 m) sub-surface (10), where the reactivity of host minerals and the or-ganic carbon needed to dissolve them is also greatest (11, 12). Asa result, considerable attention has been paid to contaminationmechanisms in Holocene units (up to ∼0.011 Ma in age), whereaffected wells are most commonly found, and to older Pleisto-cene units (∼0.011–2.6 Ma), where they are usually more rare(6–8, 13). Little is known of arsenic occurrence in older Plio-cene–Miocene-age (∼2.6–23 Ma) aquifers or in the thick se-quences of interbedded confining clays (i.e., aquitards), whichare 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 con-taminated at depths of 200–500 m. A recent nationwide survey ofarsenic in wells conducted from 2002 to 2008 by the Departmentof Water Resources Management, Vietnam includes 42,921 ob-servations in the Delta alone (Fig. 1). Whereas prior synopticstudies in the Delta have focused on near-river areas, where thehighest population densities and arsenic concentrations arefound (14–18), wells in this new survey were sampled in pro-portion to their abundance in all populated areas, providingunprecedented spatial coverage. Our analysis uses (i) the rich-ness of these observations along with (ii) simulation of the spatio-temporally explicit flow and pumping history of the multiaquifersystem and (iii) validation of pumping-induced compaction byradar remote sensing of land subsidence. This complementarysuite of methods allows us to reveal a human-influenced con-tamination mechanism in deep aquifers.

ResultsIn the Mekong Delta, groundwater is widely pumped from sevenmajor aquifers ranging from Holocene to Miocene age. The de-lineation of aquifers and their ages used here is based on thework of the Division for Geological Mapping for the South ofVietnam, which used a suite of standard techniques includingmud logging of drill cuttings, radiometric dating, analysis of mi-crofossils, and geophysical surveys to describe the >1,000-m-deep, fault-blocked basin underlying the Delta and the complexstratigraphy of its fill. On the Vietnamese side of the Delta,numerous productive, sandy aquifers are separated by and em-bedded with similarly thick and laterally extensive sequencesof significantly less-permeable clays. The confining nature ofthese clays is indicated by distinctly separable hydraulic heads in

Author contributions: L.E.E., S.M.G., H.A.Z., and S.F. designed research, performed re-search, analyzed data, and wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. J.N. is a guest editor invited by the EditorialBoard.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. E-mail: gorelick@stanford.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1300503110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1300503110 PNAS | August 20, 2013 | vol. 110 | no. 34 | 13751–13756

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

July

7, 2

020

nested monitoring wells. Well nests, or sets of wells in which eachmeasures the hydraulic head in one of four to six aquifers at thesame location, are well distributed throughout the Delta. Theyallow for examination of both horizontal and vertical hydraulicgradients, and they indicate that several aquifers are experienc-ing widespread head declines (Supporting Information). Pumpingwells, in use since the early 1900s, have become increasinglycommon since the early 1980s. Today these wells are widely usedfor a variety of domestic, agricultural, and industrial purposes.Throughout much of the Delta, deep aquifers are the most heavilyexploited.Arsenic occurrence in the Mekong Delta exhibits some classic

characteristics observed in many other South and SoutheastAsian river basins (Fig. 1A). Dissolved concentrations are highest(maximum 1,470 μg/L) in the shallow subsurface (<100 m), inclose proximity (<5 km) to the main river and its distributaries,and drop off sharply with distance. Wells with concentrations upto 1,000 μg/L, two orders of magnitude greater than the 10 μg/LWorld Health Organization drinking water standard, are oftenless 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 arsenicexceeding 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) andthose tapping the deep Pliocene–Miocene aquifers (Fig. 2D).The deep set (170–500 m) contains the majority (84%) ofarsenic-contaminated wells.The Mekong Delta focus area shows far more deep, contam-

inated wells than other major arsenic-affected regions of Southand Southeast Asia. Prior surveys from Nepal, India, Bangla-desh, and Vietnam’s Red River Delta (compiled in ref. 10) in-dicate that the fraction of contaminated wells diminishes withdepth (Fig. 3). In those areas, wells were rarely found with ar-senic in excess of 10 μg/L at depths greater than 200 m. Deepwells, however, tend to be less well represented in surveys, in

terms of both total quantity and spatial coverage. As such, ourunderstanding of the extent and mechanisms of deep arseniccontamination has remained incomplete.

Fig. 1. Groundwater arsenic concentrations in the Mekong Delta, Vietnam. (A) Plan view. (B) North-looking perspective (vertical exaggeration, 150×), highlightingthe focus area of this work. Topography and bedrock surface shown above and below zero elevation (mean sea level), respectively. Coastline is lightly dashed.

Fig. 2. (A) Locations of bores in cross-sections in the vicinity of the focusarea. (B) Cross-section illustrating complex stratigraphy and distinction be-tween shallow and deep zones of focus area. (C and D) Arsenic concen-trations in wells of the shallow and deep zones. Wells with As <5 μg/Loutside the focus area are shown in gray. (E) Pumping intensity of wells inthe deep zone estimated by multiplying the age of each well by the 2007unit pumping rate (pumping rate for each district and aquifer divided by thetotal number of wells). Wells in districts for which pumping data are un-available are shown in gray.

13752 | www.pnas.org/cgi/doi/10.1073/pnas.1300503110 Erban et al.

Dow

nloa

ded

by g

uest

on

July

7, 2

020

Evidence suggests that deep contamination in the MekongDelta focus area is unlikely to have been caused by the onlypresently acknowledged mechanism: pumping-induced verticalmigration of arsenic or dissolved organic carbon (DOC), whichcan trigger arsenic release, from the surface or shallow sub-surface (6, 7, 19–21). The number of wells above 10 μg/L in thedeep zone is seven times greater than in the shallow zone. Themean concentration in the shallow zone is significantly lessthan in the deep zone (4 vs. 20 μg/L, respectively). There are noclusters of overlying shallow wells with concentrations exceedingthose of contaminated wells at depth. Moreover, vertical veloc-ities through the layered system are estimated, using the hy-draulic head record and thicknesses of sand and clay in all 1,0201-km2 locations within the focus area, to be less than 0.2 m/y.Given that the mean distance between contaminated wells inthe shallow and deep zones is 100 m, and considering a gener-ous 28 y of travel time since the onset of increasing well in-stallations, downward transport of dissolved arsenic betweenzones 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 thedistribution of the concentration data just described and soundwell sealing practices described in documentation from theDivision for Water Resources Planning and Investigation forthe South of Vietnam (DWRPIS).Although downward transport of contaminants from the near-

surface can be excluded, deep pumping since the mid 1990s hascaused hydraulic heads in the deepest Pliocene–Miocene-ageaquifers to decline by several meters. Such head declines inducecompaction, as water that previously supported the mineralstructure is removed by pumping. Pumping-induced compactionis most pronounced in clays, far more compressible than sands,which are effectively squeezed during persistent overexploitationof adjacent aquifers. Water expelled from compressible clays,including formations at depths of hundreds of meters, has meta large portion of pumping demand in confined aquifer systemsaround the world, as evidenced by resulting land subsidence (22).During compaction owing to pumping, dissolved arsenic and anyother, potentially toxic, solutes stored in deep interbedded clayswould be expelled into adjacent aquifers. Expelled solutes from

clay could also include DOC (23) or competing ions that couldpromote arsenic dissolution or desorption within the aquifers.According to the clay-compaction release mechanism, high

densities of arsenic-contaminated wells should correspond toareas of significant cumulative pumping. Indeed, pumping ratesamong the 1,365 wells in the deep zone of the focus area aremuch greater and the median age of wells is twice that of thosewithin the 10-km surrounding buffer. Deep wells in this buffer(1,218 total) largely have been installed since 1996 (SupportingInformation) and do not show arsenic (only four wells with ar-senic >10 μg/L). This implies that before significant pumpinglasting a decade or more, deep aquifers are not pervasivelycontaminated through in situ processes or by solute diffusion outof interbedded clays. Rather, there is a likely ∼10-y time lagbefore contaminant arrival at pumping wells, consistent with therequisite 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 transientaquifer simulation. We recreate the spatially explicit pumpinghistory in each of the seven major aquifers according to 2007pumping data from the DWRPIS. We assign typical low per-meability values for interbedded clays along with high storageproperty values, which are known to be one to two orders ofmagnitude greater for clay than for sand. The model is calibratedto fit the historical trends in hydraulic head data in all aquifersmeasured from 1996 to 2008 in nested monitoring wells acrossthe region. Observed head declines in these wells of up to 74 cm/yin the Delta and 230 cm/y in HCMC are well reproduced by ourmodel (Supporting Information). Our simulations indicate landsubsidence rates in the focus area of 1.1–2.4 cm/y, as pumpingdemand is met largely by water released from compacting claystorage. Simulated total subsidence is greater within the focusarea (maximum 27 cm) compared with adjacent areas owing tothe greater pumping intensity of wells within it (Fig. 2E). Pumpingin HCMC, more than 50 km away, is not responsible for sub-sidence in the focus area. Measured hydraulic heads, interfer-ometric synthetic aperture radar (InSAR)-based subsidenceestimates, and aquifer simulation all indicate that the effect ofHCMC pumping is local.Pumping-induced clay compaction is measurable as land sub-

sidence. We measure land subsidence rates using InSAR data

Fig. 3. Depth profile of groundwater arsenic occurrence in surveys of major affected areas in South and Southeast Asia.

Erban et al. PNAS | August 20, 2013 | vol. 110 | no. 34 | 13753

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

July

7, 2

020

collected by the Phased Array type L-band Synthetic ApertureRadar (PALSAR) instrument aboard the Advanced Land Ob-serving Satellite (ALOS), because no ground-based measure-ment record is available. Two PALSAR tiles cover our study areaand images of them were acquired over the period 2007–2010every 2–12 mo. We form interferograms from all pairs of scenesthat span a 1-y interval, selected to minimize seasonal effects,and average them to reduce atmospheric errors. Based onInSAR, subsidence is occurring at a rate of 1–3 cm/y (∼1.2–3.6 cm in the vertical assuming no horizontal deformation) rel-ative to a coherent reference area near the Cambodian borderand is highest in localized (1–3 km) subsidence bowls centeredon many of the regions’ small cities (Fig. 4). Estimated errors inrates are spatially variable, ranging from ±0.5–1 cm/y (SupportingInformation). Comparison of land subsidence rates based onInSAR and based on aquifer system simulation is shown inFig. 4.Subsidence measured with InSAR is seen throughout the fo-

cus area and further extends to the north and south of it whereeither (i) we do not have arsenic measurements or (ii) wells havegenerally 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. Sub-sidence in HCMC is also evident. HCMC is, however, locatedoutside of the Mekong floodplain such that provenance therediffers from the Himalayan sediments responsible for mostarsenic in wells of the Delta proper. Wells in the HCMC areaare, not surprisingly, largely uncontaminated anywhere in thedepth profile, and it seems that the city’s excessive pumping,although inducing subsidence, is therefore not causing release ofarsenic from a deep source. Additional factors complicating therelationship between observed subsidence and arsenic occur-rence may be related to the network of regional faults, localdepositional conditions, and sulfide attenuation (24, 25).The paleoclimatic record supports the occurrence of dissolved

arsenic in deep clays. Holocene clays are known to maintain highconcentrations of dissolved arsenic over millennia (12, 26). Modernclimate features, notably tropical temperatures and precipita-tion, and high sea levels, are conducive to sustaining the bio-geochemical conditions that are favorable to arsenic dissolutionwithin Holocene clays. The Pleistocene was marked by frequentglaciations that substantially lowered global sea levels. In theMekong 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 offerric (hydr)oxides that give Pleistocene sands in Bangladeshtheir oft-noted brown or orange color (8, 10, 13, 29, 30). Theseoxidized deposits have a higher capacity for arsenic adsorption(4, 21, 31) and are associated with aquifers that are low in dis-solved arsenic, which are found at greater depths in the BengalBasin owing to higher sedimentation rates. In contrast to thePleistocene in the Mekong Delta region, temperatures duringthe Pliocene–Miocene epochs were likely similar or elevatedrelative to the Holocene (27, 32), suggesting biogeochemicalconditions may also have been like those responsible for mobi-lizing arsenic in shallow clay today.Through our analysis, we put forth a conceptual model that

describes the vertical distribution of arsenic seen in the MekongDelta in terms of its historical development (Fig. 5). FromMiocene times to the present, fresh clays rich in arsenic andorganic carbon were broadly deposited. Upon burial, solid-phasearsenic provided a persistent source that contaminated pore

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) Com-parison of subsidence rates derived from InSAR and aquifer simulation. (C) Subsidence rates derived from aquifer simulation only. InSAR-based rates shown inB 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.

Fig. 5. Conceptual model for depth distribution of arsenic in groundwater.

13754 | www.pnas.org/cgi/doi/10.1073/pnas.1300503110 Erban et al.

Dow

nloa

ded

by g

uest

on

July

7, 2

020

waters trapped in thick clay units. Over millions of years of de-position, dissolved arsenic concentrations in the pore fluids ofmore permeable aquifer sands were drawn to low levels by ad-vection, mixing, and dilution. Concurrently, clays lost some oftheir original solute load to diffusion through the limited con-nected pore network. Slow diffusion out of occluded pores andslow dissolution and desorption of the persistent solid-phasearsenic supply (12, 33), however, maintained a continuous dis-solved arsenic load in deep clays, as found elsewhere in mucholder aquitards (>70 Ma) (34). When low-arsenic, deep aquiferswere 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 inconfining clays to adjacent aquifers, a process taking a decadeor more.The implication of these findings for the Mekong Delta region,

and potentially other arsenic-prone aquifer systems like it, is thatdeep, untreated groundwater is not a safe long-term watersource. Deep wells that test clean upon installation, as do thosebordering the focus area, may not remain arsenic-free over timeas pumping promotes compaction and release of arsenic or ar-senic-mobilizing solutes from deep clays. The potential for adeep source of arsenic, resulting from this unrecognized clayexpulsion mechanism, should be considered as deep groundwa-ter resources are further developed in these settings. In otherconfined aquifer systems around the world, solute release fromcompacting clays could also affect groundwater quality in as yetunknown ways.Demand for deep groundwater is created by limited freshwa-

ter availability, arsenic-contaminated shallow groundwater, andnonpotable surface water. However, deep groundwater exploi-tation in the Mekong Delta presents new potential hazards: landsubsidence, saline intrusion, and human-induced arsenic con-tamination. Management of water resources in the complex,deltaic aquifer systems of South and Southeast Asia should seekto minimize human exposure to all relevant threats including, butnot limited to, those due to arsenic. To reduce the impacts ofarsenic contamination from deep groundwater extraction, watermanagers should consider a suite of measures. These include firstunderstanding the nature and extent of deep groundwater arse-nic, limiting intensive extraction, treating or blending extractedgroundwater to meet health standards, and possibly screeningpumping wells over intervals of deep aquifers that are distantfrom confining clays, among other water management strategiesaimed at health-risk reduction.

Materials and MethodsHere 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 ResourcesManagement (DWRM) in Hanoi, Vietnam. Samples were collected accordingto national standard TCVN 4556-88 and analyzed in certified laboratories(Vietnam Laboratory Accreditation Scheme) by hydride-generation atomicabsorption spectroscopy (ISO 11969:1996). In the Mekong Delta, 50,532 wellswere surveyed, of which 42,921 were considered, excluding samples with norecorded depth and very shallow (≤10 m), potentially open wells that may beexposed to surface conditions. Along with arsenic concentration measure-ments, depth and the Global Positioning System coordinates of each well,the year of installation was generally reported (202 missing values). Missingyear of installation values were ignored in analyses involving well ages.

Ancillary hydrogeologic data were acquired from the DWRM and theDWRPIS of Vietnam, HCMC, Vietnam. These include (i) 10 stratigraphic cross-sections compiled from 81 well logs at ∼20-km spacing including mappedgeologic faults, originally interpreted by the Division for Geological Map-ping for the South of Vietnam; (ii) pumping rates in 22 districts, specific tothe six major pumped aquifers (excluding the Holocene aquifer), from a2007 survey of 5,282 wells, classified by purpose of water use; (iii) transienthydraulic head data from 45 nested monitoring wells, measured monthlyover 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 determinedbased on the well logs and cross-sections. Clay pods embedded withinaquifers seem to be discontinuous and were ignored in the interpolationof the major hydrogeologic unit boundaries and subsequent regionalgroundwater and subsidence simulations. Each data point in the DWRMsurvey was assigned to an aquifer according to its latitude, longitude,and well-screen elevation, taken as the well’s recorded depth relative tothe Shuttle Radar Topography Mission (STRM) 90-m digital elevationmodel (DEM).

Estimates of vertical flow between zones of the focus area were made forall 1-km2 cells within it. We considered the shortest reasonable distancethrough which shallow contamination could have been transported todepth: the middle of the Upper Pleistocene aquifer (the most shallow ofthree 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). Foreach cell of the focus area, the effective hydraulic conductivity (Keffective)over this minimum transport distance was calculated using the harmonicmean of the (n) layer conductivities, scaled by their thicknesses (b) as follows:

Keffective =

Xn

1bn

Xn

1

bn

Kn

; [1]

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

v = −KeffectiveI

θ; [2]

where Keffective is the effective hydraulic conductivity defined above, I is thevertical hydraulic gradient component, and θ is the effective porosity, takenas the mean of literature values for the two materials, weighted by theirrelative thicknesses. The average gradient between the Upper Pleistoceneand Upper Pliocene aquifers was calculated over two periods determined bydata availability: 1987–1997 and 1997–2005. All parameters for the verticalflow calculations are provided in Supporting Information.

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

Δb= SSbΔh; [3]

where Ss is specific storage, b is thickness, and Δh is the temporal change inlocal hydraulic head values. Details of the model discretization, boundaryconditions, assignment of pumping conditions, and calibration to moni-toring well data in the seven major aquifers considered are provided inSupporting Information.

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

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

Erban et al. PNAS | August 20, 2013 | vol. 110 | no. 34 | 13755

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

July

7, 2

020

ACKNOWLEDGMENTS. We thank Y. Chen and C. Wortham for interfero-metric synthetic aperture radar processing advice and Prof. C. Harvey atMassachusetts Institute of Technology for his review. We thank the De-partment of Water Resources Management, Hanoi, for the arsenic data, theDepartment of Water Resources Planning and Investigation for the Southof Vietnam for hydrogeologic datasets, and Alaska Satellite Facility,Japanese Aerospace Exploration Agency and National Aeronautics and

Space Administration for synthetic aperture radar data. We gratefullyacknowledge the United Parcel Service Endowment Fund and the GlobalFreshwater Initiative of the Woods Institute for the Environment at Stanford.This work is being supported by National Science Foundation Grant EAR-1313518 to Stanford University. Any opinions, findings, and conclusions orrecommendations expressed in this material are those of the authors anddo not necessarily reflect the views of the National Science Foundation.

1. Ravenscroft P, Brammer H, Richards K (2009) Arsenic Pollution: A Global Synthesis(Wiley-Blackwell, Chichester, UK).

2. Smith AH, Lingas EO, RahmanM (2000) Contamination of drinking-water by arsenic inBangladesh: A public health emergency. Bull World Health Organ 78(9):1093–1103.

3. Dig deep to avoid arsenic, 8483, Phillip Downey.4. Radloff KA, et al. (2011) Arsenic migration to deep groundwater in Bangladesh

influenced by adsorption and water demand. Nat Geosci 4(11):793–798.5. Michael HA, Voss CI (2008) Evaluation of the sustainability of deep groundwater as an

arsenic-safe resource in the Bengal Basin. Proc Natl Acad Sci USA 105(25):8531–8536.6. Mukherjee A, Fryar AE, Scanlon BR, Bhattacharya P, Bhattacharya A (2011) Elevated

arsenic in deeper groundwater of the western Bengal basin, India: Extent and con-trols from regional to local scale. Appl Geochem 26(4):600–613.

7. Winkel LH, et al. (2011) Arsenic pollution of groundwater in Vietnam exacerbated by deepaquifer exploitation for more than a century. Proc Natl Acad Sci USA 108(4):1246–1251.

8. Burgess WG, et al. (2010) Vulnerability of deep groundwater in the Bengal AquiferSystem to contamination by arsenic. Nat Geosci 3(2):83–87.

9. Smedley PL, Kinniburgh DG (2002) A review of the source, behavior and distributionof arsenic in natural waters. Appl Geochem 17(5):517–568.

10. Fendorf S, Michael HA, van Geen A (2010) Spatial and temporal variations of ground-water arsenic in South and Southeast Asia. Science 328(5982):1123–1127.

11. Postma D, et al. (2012) Groundwater arsenic concentrations in Vietnam controlled bysediment age. Nat Geosci 5(9):656–661.

12. Kocar BD, et al. (2008) Integrated biogeochemical and hydrologic processes drivingarsenic release from shallow sediments to groundwaters of the Mekong delta. ApplGeochem 23(11):3059–3071.

13. Zheng Y, et al. (2005) Geochemical and hydrogeological contrasts between shallowand deeper aquifers in two villages of Araihazar, Bangladesh: Implications for deeperaquifers as drinking water sources. Geochim Cosmochim Acta 69(22):5203–5218.

14. Hoang TH, Bang S, Kim KW, Nguyen MH, Dang DM (2010) Arsenic in groundwaterand sediment in the Mekong River delta, Vietnam. Environ Pollut 158(8):2648–2658.

15. Nguyen KP, Itoi R (2009) Source and release mechanism of arsenic in aquifers of theMekong Delta, Vietnam. J Contam Hydrol 103(1-2):58–69.

16. Buschmann J, Berg M, Stengel C, Sampson ML (2007) Arsenic and manganese con-tamination of drinking water resources in Cambodia: Coincidence of risk areas withlow relief topography. Environ Sci Technol 41(7):2146–2152.

17. Berg M, et al. (2007) Magnitude of arsenic pollution in the Mekong and Red RiverDeltas—Cambodia and Vietnam. Sci Total Environ 372(2-3):413–425.

18. Stanger G, Truong TV, Ngoc KSLTM, Luyen TV, Thanh TT (2005) Arsenic in ground-waters of the Lower Mekong. Environ Geochem Health 27(4):341–357.

19. Neumann RB, et al. (2009) Anthropogenic influences on groundwater arsenic con-centrations in Bangladesh. Nat Geosci 3(1):46–52.

20. Polizzotto ML, Harvey CF, Sutton SR, Fendorf S (2005) Processes conducive to therelease and transport of arsenic into aquifers of Bangladesh. Proc Natl Acad Sci USA102(52):18819–18823.

21. Harvey CF, et al. (2002) Arsenic mobility and groundwater extraction in Bangladesh.Science 298(5598):1602–1606.

22. Poland JF, ed (1984) Guidebook to Studies of Land Subsidence Due to GroundwaterWithdrawal (UNESCO, New York).

23. McMahon PB, Chapelle FH (1991) Microbial production of organic acids in aquitardsediments and its role in aquifer geochemistry. Nature 349(6306):233–235.

24. Buschmann J, Berg M (2009) Impact of sulfate reduction on the scale of arseniccontamination in groundwater of the Mekong, Bengal and Red River deltas. ApplGeochem 24(7):1278–1286.

25. Lowers HA, et al. (2007) Arsenic incorporation into authigenic pyrite, Bengal Basinsediment, Bangladesh. Geochim Cosmochim Acta 71(11):2699–2717.

26. Benner SG, et al. (2008) Groundwater flow in an arsenic-contaminated aquifer, Me-kong Delta, Cambodia. Appl Geochem 23(11):3072–3087.

27. Heaney LR (1991) A synopsis of climatic and vegetational change in Southeast Asia.Clim Change 19(1-2):53–61.

28. Bird MI, Taylor D, Hunt C (2005) Palaeoenvironments of insular Southeast Asia duringthe Last Glacial Period: A savanna corridor in Sundaland? Quat Sci Rev 24(20-21):2228–2242.

29. Swartz CH, et al. (2004) Mobility of arsenic in a Bangladesh aquifer: Inferences fromgeochemical profiles, leaching data, and mineralogical characterization. GeochimCosmochim Acta 68(22):4539–4557.

30. Van Geen A, et al. (2003) Spatial variability of arsenic in 6000 tube wells in a 25 km2

area of Bangladesh. Water Resour Res 39(5):1140–1151.31. Stollenwerk KG, et al. (2007) Arsenic attenuation by oxidized aquifer sediments in

Bangladesh. Sci Total Environ 379(2-3):133–150.32. Robinson MM, Chandler MA, Dowsett HJ (2008) Pliocene role in assessing future cli-

mate impacts. Eos Trans AGU 89(49):501–502.33. Tufano KJ, Reyes C, Saltikov CW, Fendorf S (2008) Reductive processes controlling

arsenic retention: Revealing the relative importance of iron and arsenic reduction.Environ Sci Technol 42(22):8283–8289.

34. Yan XP, Kerrich R, Hendry MJ (2000) Distribution of arsenic (III), arsenic (V) and totalinorganic arsenic in porewaters from a thick till and clay-rich aquitard sequence,Saskatchewan, Canada. Geochim Cosmochim Acta 64(15):2637–2648.

35. Nguyen VC (1987) Hydrogeological Map of Vietnam (General Department of Geol-ogy, Hanoi, Vietnam).

36. Zebker HA, Hensley S, Shanker P, Wortham C (2010) Geodetically accurate InSAR dataprocessor. IEEE Trans Geosci Rem Sens 48(12):4309–4321.

13756 | www.pnas.org/cgi/doi/10.1073/pnas.1300503110 Erban et al.

Dow

nloa

ded

by g

uest

on

July

7, 2

020