Groundwater dynamics and arsenic contamination in Bangladesh
Charles F. Harvey a,⁎, Khandaker N. Ashfaque a, Winston Yu a, A.B.M. Badruzzaman b,M. Ashraf Ali b, Peter M. Oates a, Holly A. Michael a, Rebecca B. Neumann a,
Roger Beckie c, Shafiqul Islam d, M. Feroze Ahmed b
a Parsons Laboratory, CEE, MIT, Cambridge, MA, United Statesb Bangladesh University of Engineering and Technology, Dhaka, Bangladesh
c University of British Columbia, BC, Canadad Tufts University, MA, United States
Over the last 45 years most of the population ofBangladesh and West Bengal switched their watersupply from ponds and rivers to well water. As manyas 10 million new domestic wells have been installed,
providing drinking water for over 130 million people.Tragically, much of the region's groundwater isdangerously contaminated by arsenic and approximately57% of these people now drink water with arsenicconcentrations above 10 ppb, the standard of the WorldHealth Organization (Yu et al., 2003). Irrigation wells,mostly extracting from the shallow aquifer, wereinstalled across the country concurrent with thistransition of the domestic water supply. According toBADC (2003), a total of 924,023 shallow tubewells and
2 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
23,434 deep tubewells were used for irrigation inBangladesh during the 2003 dry season. Groundwaterirrigation greatly increased agricultural productionenabling Bangladesh to become self-sufficient in foodeven though the population nearly tripled over the lastfour decades. Irrigation now sustains production of dry-season rice called Boro, which provides greater yieldsthan the traditional rice grown during the wet season(Fig. 1). During the 2003 dry season, about 87% of thetotal irrigated area of about 4 million hectares (about28% of the total area of the country) was under Borocultivation and Boro accounted for about 49% of thetotal rice production (MoA, 2004). Thus, issues ofgroundwater quality and quantity have become vital forboth the supply of drinking water and the production offood in Bangladesh.
A wide range of evidence indicates the importanceof groundwater flow to the subsurface biogeochemistry
Fig. 1. (A) Cultivation of high-yielding boro rice has greatly expanded since 1of the cultivatable area. Most boro is irrigated by groundwater so extractiomaximum depth to groundwater in wells between 1988 and 1997 (data sourceyearly maximum depths in six geographic regions of Bangladesh: the northrepresents the country average of yearly maximum depths. Error bars represeroot of n−1). The country average, as well as all the hydrologic regions, ha
in Bangladesh, however little work has been directedtowards understanding the physical groundwatersystem. The hydrogeologic characterization that hasbeen conducted across Bangladesh is small relative tothat conducted at groundwater contamination sites inthe US where groundwater is not used for drinking.This paper focuses on how groundwater flow andsolute transport through Bangladeshi aquifers affectsarsenic concentrations. We first describe the biogeo-chemical processes that control the aqueous/solidphase partitioning of arsenic in Bangladesh, andconsider possible explanations for observed verticalprofiles of chemical parameters within the aquifers. Wethen discuss the processes that drive groundwater flowin Bangladesh by analyzing hydraulic data we havecollected from our field site in Munshiganj, whichincludes detailed diurnal and seasonal hydraulic headcycles, as well as seasonal water levels in ponds, rice
970 to cover approximately 20% of Bangladesh, or approximately 45%n has also risen. Data taken from Hossain et al. (2003). (B) Yearly: WARPO, 2000). The filled and unfilled symbols represent the averageand south, divided into east, central and west sections. The solid linent the standard error of the mean (standard deviation divided by squares a pattern of increasing maximum depths.
3C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
fields and rivers, and budgets for irrigation pumping.Finally, we use a lumped-parameter model of thegroundwater system to show how irrigation pumpingchanges the source of recharge to the aquifers andreduces the residence time of groundwater byredirecting natural discharge from the rivers intoirrigated rice fields and increasing recharge fromponds.
2. Biogeochemistry and arsenic mobility
Researchers agree that dissolved arsenic in thegroundwater of Bangladesh originates from the sedi-ments. However, there is no evidence of widespread,unusually high, levels of solid phase arsenic in theaquifer material-concentrations are typically less than 10ppm in sandy sediment and less than 100 ppm in claysand peats (Nickson et al., 1998; McArthur et al., 2004;Swartz et al., 2004). Thus, it appears that high dissolvedarsenic concentrations in groundwater are the result ofparticular hydrologic and biogeochemical conditionsthat partition arsenic from the solid to the aqueous phaseand perhaps transport arsenic into contaminated aqui-fers, but have not yet flushed dissolved arsenic fromthese aquifers.
The original source of the arsenic was most likelyoxidation of sulfide minerals, principally pyrite, derivedfrom the granitic and metamorphic source regions of theHimalayas. In an accompanying paper in this issue,Polizzotto et al. (2006-this issue) show that arsenic-bearing pyrite grains have reached the Ganges delta andare incorporated in the aquifers. This work supports thehypothesis thatminerals are cyclicallyweathered near theland surface, where the water table rises and falls eachyear. When oxygen is introduced into the near surface,sulfide minerals are oxidized, iron oxides form, andarsenic is transferred from pyrite to iron oxides. Duringanoxic conditions, which may coincide with periods ofrecharge as return flow from rice fields, iron-oxidesdissolve and arsenic is released into the water columnwhere it is transported to depth with the recharge water.Previous researchers have suggested that pyrite oxidationoccurredduringweatheringat the source in theHimalayasand that arsenic was transported and deposited in theGanges delta in association with the resulting iron oxides(McArthur et al., 2004). Some arsenic was likelytransported into the Ganges Delta in both states, but theimportant difference between these two explanations isthat the redox cycle scenario provides an explanation for asource of dissolved arsenic near the land surface, whereasthe distributed iron oxide explanation places the source ofdissolved arsenic deeper within the aquifers.
The reducing conditions of almost all groundwater inBangladesh (demonstrated by high levels of dissolvedferrous iron and methane, and low measurements of Eh),and the weak but statistically significant positivecorrelation of dissolved arsenic to iron and bicarbonate,suggest that most arsenic is liberated by dissolution ofiron (oxy)hydroxides, or perhaps desorption of arsenicafter reduction from arsenate to arsenite (Nickson et al.,1998; BGS and DPHE, 2001; Harvey et al., 2002). Thelow concentrations of sulfate (and in some areas thenegative correlation between arsenic and sulfate) as wellas the generally reducing conditions indicate that arsenichas not been directly mobilized into groundwater fromsulfide minerals (e.g., Harvey et al., 2002).
Microbiological processes drive many geochemicaltransformations in Bangladeshi soils and groundwater(see, for example, Oremland and Stolz, 2005), andmicrobial activity likely ensures that the timescales ofbiochemical processes are much less (∼weeks) thanthe residence times of groundwater (∼decades).Laboratory batch experiments demonstrate that biotic(van Geen et al., 2003; Islam et al., 2004) and abiotic(Polizzotto et al., 2006-this issue) transformationsoccur within weeks or even hours, and these resultsare confirmed by field perturbation experiments(Harvey et al., 2002). Consequently, the rate ofchemical transformations within aquifers is likelycontrolled by the rate that chemical loads aretransported into the subsurface, not by the rates ofmicrobiological activity. Microbial activity appears todrive the aquatic geochemical system quickly towardslocal equilibrium, therefore groundwater mixing maybe the limiting process for chemical transformations,as is often the case in groundwater systems. Becausemicrobes are ubiquitous, understanding specific detailsof microbial enzymatic and metabolic pathways mayprove important if: (1) Geochemical transformationsnear the land surface, where rapid water tablemovement may change chemical conditions on thetime-scale of microbial response, prove to be impor-tant; (2) Evidence is found that significant transforma-tions occur against a thermodynamic gradient, such asmay result from detoxification; (3) Important microbialprocesses are found that occur only slowly, over thedecades long time-scale of solute transport andmixing.
2.1. Geochemistry of brown/orange and grey sediment
Several research teams (BGS and DPHE, 2001;Harvey et al., 2002; van Geen et al., 2003; McArthur etal., 2004) describe two distinct types of aquifer
4 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
sediment: brown (or orange to yellow) sedimentpresumably containing iron (oxy)hydroxides wheredissolved arsenic concentrations in porewater are low,and grey sediments where dissolved arsenic concentra-tions may be high. The brown sediments are found atdepth in the older Pleistocene aquifers such as the DupiTila formation, where low-arsenic water is obtained, aswell as near the surface. Dissolved arsenic is presum-ably low in these sediments because of the capacity ofiron (oxy)hydroxides to adsorb arsenic. Islam et al.(2004) showed that arsenic is liberated from sedimentscollected in West Bengal by the addition of organiccarbon. They do not report the in-situ arsenic concen-trations in the pore-water, but the sample contains iron(oxy)hydroxides and is described as coming from atransition zone between a region of oxidizing conditionsand a region with reducing conditions.
High concentrations (N200 ppm) of solid-phasearsenic have been found in orange, (oxy)hydroxide-rich,bands ∼1.5 m deep in soils (Breit et al., 2001) at severallocations in the country. Such horizontal bands of ironoxides have previously been described in the soil scienceliterature (Brammer and Brinkman, 1977) for Bangla-desh, and indeed have provided the evidence for a soildevelopment process termed ferolysis (Brammer, 1996;van Ranst and de Coninck, 2002) by which oxide layersdevelop. If such arsenic-rich iron bands prove to beubiquitous, then their biochemical transformations couldpotentially provide an important source of dissolvedarsenic into water that recharges the aquifers, consistentwith the analysis of Polizzotto et al. (2006-this issue).
The role played by iron (oxy)hydroxides within thecontaminated grey sediments of the Holocene aquifer,where most wells withdraw water, is still enigmatic. Iron(oxy)hydroxides must exist, or have existed veryrecently, according to the theory that arsenic is releasedfrom iron (oxy)hydroxides in local sediments by organiccarbon oxidation. However these iron (oxy)hydroxideshave not been definitively documented in the greysediment and high concentrations of methane andhydrogen (Harvey et al., 2002) in strongly reducingwater indicate that geochemical conditions are notconducive to stability of iron (oxy)hydroxides. Giventhat dissimilatory iron reduction, the primary means ofiron reduction, precedes methane generation in sedimentdiagenesis, active iron reduction would most likely haveoccurred at an earlier stage in diagenesis as opposed tothe present time. Furthermore, iron (oxy)hydroxideshave not been detected in grey sediments fromMunshiganj using either bulk- or micro-X-ray absorp-tion spectroscopy (XAS), indicating that they compriseless than 5% of the total Fe (Polizzotto et al., 2006-this
issue) and, in sequential extractions, all of the ferric ironcan be accounted for as magnetite (Swartz et al., 2004)within experimental error. Thus, the role that iron (oxy)hydroxides may have played in controlling the currentconcentrations of dissolved arsenic is difficult todetermine.
Further complicating the puzzle over the role ofiron (oxy)hydroxides, Swartz et al. (2004) show thatonly very small quantities would be required toexplain the current ratio of sorbed to dissolved arsenic,and McArthur et al. (2004) provides a geologicexplanation for why the Ganges Delta sedimentwould have been deposited with relatively little iron(oxy)hydroxides. Thus, it is conceivable that slowreductive dissolution within aquifer sediments couldbe responsible for high dissolved arsenic concentra-tions, but only if the geochemical system happens tobe in a state where iron (oxy)hydroxides have releasedalmost all of their sorbed arsenic. In other words, theaquifer sediments must be poised in a geochemicalstate where the inventory of iron (oxy)hydroxides isnearly (or recently) exhausted, yet arsenic has not beenflushed away by flowing groundwater. Other explana-tions, that we explore below, are that both the physicalflow system and the biogeochemical system haverecently been perturbed, and that dissolved arsenicoriginates from near-surface sediments above theaquifer which may have a much larger compositionof (oxy)hydroxides.
Dissolved arsenic concentrations are maintained inpart because geochemical factors conspire to preventarsenic that has been dissolved from sorbing back ontoaquifer sediment. First, the paucity of ferric (oxy)hydroxides implies there may be few adsorption sites.Second, high concentrations of other anions, such assilicate and phosphate, which compete with arsenic forsurface sorption sites, are prevalent in groundwaterthroughout most of the arsenic-affected areas. However,there is no convincing correlation between the concen-trations of these anions and arsenic that would indicatethat the spatial pattern of competing anions can explainthe pattern of dissolved arsenic. Equilibrium chemicalmodeling using the parameters measured at our siteindicates that silicate, phosphate and other anionscompete, preventing arsenic from sorbing (Swartz etal., 2004). These conceptual geochemical models arefurther complicated by the fact that arsenic likelyadsorbs to surfaces of many solid phases other thanoxyhydroxides, such as magnetite, green rust, andpotentially siderite and apatite. Arsenic is known tosorb readily to magnetite (Dixit and Herring, 2003) andthe results of our density and magnetic separations show
5C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
that the magnetite fraction has the highest arsenicconcentration by weight (Swartz et al., 2004).
2.2. Irrigation pumping and time trends in arsenicconcentration
Several research groups postulate that irrigationpumping may flush arsenic from aquifers (e.g. Harveyet al., 2003; McArthur et al., 2004). Harvey et al. (2003)support this contention by comparing concentrationssampled from irrigation wells to concentrations fromdrinking water wells to show that irrigation wells, whichflush much greater quantities of water, have significantlylower arsenic concentrations. At a national scale, Ali(2003) estimated that each year groundwater irrigationremoves from aquifers, and then applies to fields, aboutone million kilograms of arsenic.
On the other hand, some evidence suggests thatarsenic concentrations may rise after pumping com-mences. Kinniburgh et al. (2003), van Geen et al. (2003)and McArthur et al. (2004) all provide strong statisticalevidence that arsenic concentrations in domestic wellwater correlate to the age of the well, suggesting thatarsenic concentrations may rise after a well is installed,perhaps because irrigation wells, which have muchgreater effects on the local groundwater system, areinstalled in the region at the same time as the domesticwells where arsenic is measured. Can these apparentlycontradictory suggestions of both falling and risingarsenic concentrations be reconciled? Clearly pumpingremoves some arsenic from the aquifer. In fact,irrigation pumping can be viewed as analogous to“pump-and-treat” groundwater remediation methodsemployed in North America and Europe, but withoutthe “treat” and with extraction rates that are actuallyhigher than at many sites! However, increased flushingmay be concurrent with increased input of dissolvedarsenic, caused either by simple groundwater transportor by release driven by input of organic carbon fromsurface sources, or from sediments, including peatlayers. This concurrent enhancement of both sinks andsources of arsenic to the groundwater system by humanperturbation could potentially create very complextemporal and spatial behavior of dissolved arsenic.
At our site in Munshiganj, processes appear to becompeting to both increase and decrease arsenicconcentrations: arsenic is being extracted from thesystem and radiocarbon dating of dissolved carbonindicates that arsenic has been mobilized recently(Harvey et al., 2002). The radiocarbon data show thatdetrital organic carbon has not driven recent biogeo-chemical reactions. The byproducts of microbial activity,
both inorganic carbon and methane, have much youngerradiocarbon dates than the dissolved organic carbon orthe sediment, and the concentration of this inorganiccarbon is much larger than that of the older organiccarbon. In fact, at 20 m depth, the inorganic carbon haslevels of carbon-14 higher than 100% modern. This iscarbon from bomb testing, so it entered the aquifer in thelast 50 years. At three 30-m wells dissolved inorganiccarbon (DIC) radiocarbon ages are 462, 770, and 823years, much younger than the local sediment and theradiocarbon age of dissolved organic carbon (DOC)(3636, 1538, and 1890 years). Tree roots and burrowinganimals are unlikely to penetrate below 10 m because theaquifer remains saturated all year, and in our discussionof the hydrologic model, we will consider whether rootsin villages may penetrate through 6m of clay to reach theaquifer. Thus, dissolved carbon with a radiocarbon ageyounger than the sediment age was transported down-ward and laterally by flowing groundwater.
The presence of young DIC and old DOC in the samewater does not appear to result from the mixing ofyoung, DIC-containing water and old, DOC-containingwater. The concentrations correlate strongly (i.e. waterhigh in young DIC is also high in old DOC); they do notfollow a mixing-line that would have a negativecorrelation. Thus it appears that the older DOC wasmobilized from the sediment concurrently with theproduction or inflow of young DIC. McArthur et al.(2004) argue that buried peat deposits have provided theorganic carbon that drives reduction at our field site, butthey do not attempt to reconcile the different radiocar-bon ages of dissolved organic carbon and inorganiccarbon.
3. Geochemical profiles with depth
In this section, we consider how geochemicalcharacteristics vary with depth in aquifers, and hencehow chemical conditions relate to flow paths andgroundwater age. Fig. 2 compares depth profiles ofsolute concentrations measured at our field site inMunshiganj with averaged values from the BGS andDPHE (2001) data set (http://www.bgs.ac.uk/arsenic/Bangladesh). Our site in the Munshiganj district (Fig. 3)is located 30 km south of Dhaka and 7 km north ofthe Ganges. It contains a small intensive-study area(100 m2) with 25 sampling wells (Fig. 3B) that extractwater from depths ranging between 5 and 165 m belowthe land surface (Fig. 4B). We also monitor water levelsat 87 other locations in the surrounding 16-km2 region.We describe some similarities between results at oursingle site and the averaged national data set that suggest
Fig. 2. Concentrations of sulfate, arsenic, calcium and ammonium from the Munshiganj field site (Harvey et al., 2002) compared to the BGS andDPHE (2001) national data set. The median, mean and 90th percentile are plotted for the BGS data for a bin size of∼200 data points. Using a constantbin size causes unequal depth intervals; intervals are small where many wells are screened at ∼25 m depth.
6 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
some general characteristics of geochemical evolutionand transport across the region.
3.1. Arsenic as a function of depth
At the site in Munshiganj, dissolved arsenic has apeak at approximately 30 m depth, but we find nochemical characteristic of the solid sediment to explainthis pattern (Harvey et al., 2002; Swartz et al., 2004).The geochemistry of the local sediment and thedissolved components in the groundwater indicate anarsenic source that is hydrologically upgradient. Fur-thermore, several types of data, when taken together,suggest a relation between the arsenic peak andgroundwater flow patterns.
The pumping-test hydraulic conductivity estimates(Fig. 4A) indicate that a lower conductivity layer at 24 m
Fig. 3. (A) IKONOS satellite image of 16 km2 study area in Munshiganj wponds and bridges over the Ichhamati river channels. The rice field where irrighighlighting the Ichhamati river main channel, side channel, and intensive fielshown in (B) between March 2003 and July 2004. The flooding season extenirrigation extends from January to April with a groundwater minimum in Mtheodolite indicted that level errors were less than 1 cm. (D) Monthly rainfall,for last 30 years prior to 2002 (WARPO, 2000) and 2003 rainfall (BWDB, 200southwest of our field site. ET was calculated from data measured in Mumonitoring rice field in our study area (A) during the 2003–2004 irrigation
may partially separate horizontal flow paths at our smallintensive study site. Higher conductivity strata areevident at 18 m and 60 m, and at 24 m the well hassuffered significant silting, so that the estimatedconductivity value is uncertain, but appears to be muchlower than above or below. The head data (Fig. 4B) showthat, at least during some times of year, there isconvergent vertical flow that mixes water from aboveand below 38 m, indicating that horizontal flow mustaccelerate to conserve water mass at this depth. Evidencefor groundwater mixing at 30 m is supported by the O-18profile (Fig. 4C). The range of isotope ratios at 30 m isconsistent with mixing of lighter water from above andheavier water from below. The heavier water below 30mcould represent infiltrated pond, river or rice-field waterthat has been subject to relatively more evaporation.Measurable tritium was found to 30 m depth and at
ith hydrologic monitoring locations at irrigation wells, drinking wells,ation was monitored is also shown. (B) Center section of the study aread site near the center. (C) Water levels within the Munshiganj study areads from June to October with the flood peak in August, and dry seasonarch. Repeated surveys of well casing and benchmark elevations byevapotranspiration (ET), and irrigation data. Both the monthly average4) were measured at the meteorological station in Bhagakul, about 4 kmnshiganj district (WARPO, 2000). The irrigation water applied to aseason is also shown.
7C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
60 m, where hydraulic conductivity appears highest(although not between 30 and 60 m) (Harvey et al.,2003). These values indicate the presence of at least acomponent of water that is less than 45 years old.
Furthermore, the tritium values show a sharp decreaseto less than 1 T.U. below 24 m, so that the peak ofhigh dissolved arsenic corresponds with the depthwhere older water mixes with younger recharge. We
Fig. 4. (A) Horizontal hydraulic conductivity as estimated from pumping tests (Yu, 2003) at the intensive study site shown in Fig. 3B. Theconductivity at 22 m is uncertain because of well silting, but appears to be lower than at other depths. (B) Hydraulic heads relative to the 20-mpiezometer during the dry season, early irrigation (January) and late irrigation (May). The inset represents heads in May measured with a manometerthat obtains relative differences to within 1 mm. (C) Oxygen-18 isotope ratios relative to SMOW. The convergence of vertical flow at 38 m in May isbelow the low conductivity at 22 m, but is not consistent throughout the year, as evidenced by the January data. These data are all local to our smallintensive study site.
8 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
do not have detailed data beyond our small intensivestudy site to extrapolate these results to the region.
An obvious possibility for the recharge source of theshallow groundwater at our intensive study site is theneighboring pond (Fig. 3B, P-1). Ponds lose water at afaster rate than potential evaporation and maintain ahydraulic gradient towards our sample wells. However,the cause of the variable O-18 ratios (Fig. 4C) isunknown, and we do not currently have a sufficienttransient 3D characterization of the flow field to supportthis hypothesis.
Kinniburgh et al. (2003) and McArthur et al. (2004)both describe typical depth profiles of arsenic concen-trations as “bell shaped”. Although the general trend ofdecreasing arsenic levels with depth is obviously evidentfrom the national data set (Fig. 2) the upper part of the“bell shape”, increasing arsenic with depth, is notstatistically robust for the combined data set and isonly evident in histograms when certain bin intervals arechosen. However the “bell shaped” pattern is evident at avariety of specific study sites (Harvey et al., 2002; van
Geen et al., 2003; McArthur et al., 2004). Depth trendscan also be considered within different geologic regions,and Yu et al. (2003) tabulate the geologic regions ofBangladesh where there is a statistically significant trendof decreasing arsenic with depth (they did not considernon-monotonic trends). Their geostatistical analysisshows that the trend of decreasing arsenic concentrationswith depth explains much of the differences in arsenicconcentrations between nearby wells: neighboring wellsoften have different arsenic levels because one with-draws water from deeper in the aquifer where arsenicconcentrations are lower.
3.2. Sulfate, calcium and ammonium as a function ofdepth
At the Munshiganj site, the inverse relation ofdissolved sulfate with As (Fig. 2), and the presence ofacid volatile sulfide (AVS) in the sediments near thedissolved As peak, suggest that arsenic has not beenreleased directly by sulfide oxidation. Instead, low
9C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
dissolved sulfur levels appear to limit the precipitationof sulfides near the arsenic peak. The BGS and DPHE(2001) countrywide data set also shows a distinct,statistically robust, pattern of decreasing sulfate withdepth. This rapid decline of sulfate with depth isconsistent with the previously described scenario forinput by cyclical sulfide weathering with water-tableoscillations, followed by sulfate reduction in the aquifer.Furthermore, monthly measurements of sulfate by theBGS and DPHE (2001) in very shallow dug wells arealso consistent with this scenario, with some containingvery high concentrations with seasonal oscillations.
At our site, peaks in ammonium and calcium mirrorthe sharp peak in arsenic (Fig. 2), and these solutessuggest inflow and oxidation of organic carbon, andsubsequent mixing of solutes. Ammonium is anoxidation product of natural organic matter, and calciummay be released from solid carbonate after organiccarbon oxidation (Swartz et al., 2004). Furthermore,dissolved organic carbon, with radiocarbon ages inaccord with estimated sediment ages, also shows a bellshaped profile, indicating that recalcitrant detritalorganics may have been liberated from sedimentconcurrent with the processes that liberate arsenic. TheBGS and DPHE (2001) do not report ammonium fortheir countrywide data set, but the profile of calciumwith depth appears to show a “bell shape” similar to thatfound at our site.
3.3. “Bell shaped” depth profiles
The bell shaped pattern of solutes (arsenic, ammo-nium, calcium, and dissolved carbon) with depth istypical of vertical profiles of contaminant plumes thatoriginate from localized surface sources before movinglaterally and downward into aquifers. After solute entersan aquifer, a plume migrates laterally away from thissource location and is pushed deeper into the aquifer byrecharge from above. Thus, the basic hydrologic processof lateral transport creates vertical profiles of contam-ination with typical “bell” shapes. Such profiles havebeen characterized on multitudes of groundwatercontamination sites in North America and Europewhere release of pollutants at distinct surface locationscontaminates groundwater as migrating plumes.
Simple geometric considerations indicate thatgroundwater fluxes at our site have a large lateralcomponent, consistent with conventional understandingfor horizontal alluvial aquifer systems (Freeze andWitherspoon, 1966). After water enters the aquifer fromthe surface, it must move laterally to reach dischargeareas. At our site, the lateral velocity component must be
relatively large in much of the aquifer because thespacing between discharge areas, irrigation wells andriver channels, is as great or greater than the thickness ofthe aquifer (∼100 m). Furthermore, at the location ofour intensive study site, where the geochemical profileswere characterized, there is no irrigation well beneathour sampling piezometers; without such a sink at depth,the convergence of vertical flow indicated by our headgradients and isotope data must be accommodated byaccelerating horizontal flow.
Thus, the average vertical component of flow inrecharge areas determines the net chemical flux from thesurface down into the aquifer, and the 3-D pattern ofgroundwater flow determines the pattern of solutemigration from the sources. While one can postulatethat the bell-shaped pattern at our site results from aparticular local source of organic carbon (i.e. a nearbypond, river or rice field), it remains an open questionwhether the much less distinct bell-shaped pattern of thenational BGS and DPHE (2001) data, or similar patternsobserved at other small-scale sites, results from the samemechanisms. It is intriguing that the bell-shape patternof arsenic is mirrored by the distribution of well depths(not shown) in the BGS and DPHE (2001) data set. Mostwells withdraw water from depths near 25 m. Similarly,at our site in Munshiganj nearly all wells (drinking andirrigation) are completed at the same depth where wefind the arsenic peak, 30 m. This correspondence maysuggest a hydraulic component to the cause of arsenicmobility with depth. Because deeper wells are moreexpensive, villagers complete wells only to a depthsufficient to provide adequate yield, perhaps below thelow conductivity layer we find at 24 m (Fig. 4). Thus, arelationship may exist between the depth at which theaquifer becomes more conductive to groundwater flow,and the depth of maximum arsenic concentration.However, such speculation has not been confirmed bydetailed hydrogeologic studies.
The correspondence of the depth profiles measured atour site to the average BGS and DPHE (2001) profilesshould not be interpreted as indicating that the sameprocesses are occurring everywhere. There is spatialheterogeneity in aqueous chemical characteristics, muchof which is likely induced by the complex mosaic ofrecharge and discharge areas at the surface which havevariable water chemistry. However, the correspondencedoes suggest the existence of dominant reactive-transport processes that affect the subsurface biogeo-chemistry at many locations across the country.
One interesting observation from our site is that thestable water isotopes indicate a transition betweensurface sources of recharge that corresponds with the
10 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
arsenic peak. Thus, it is possible that the high arsenicvalues at the peak are caused by the mixing ofgroundwater from two different sources rather than theinput from one particular source. This points to the needto identify sources of recharge at our site.
4. The relation of groundwater flow and chemicaltransport to arsenic concentrations
Much of the existing literature on groundwater flowin Bangladesh focuses on isotopic inference, and not onphysical understanding of flow and solute transport.However, several simple lines of reasoning, and someisotope data, indicate that flow and transport playimportant roles in subsurface arsenic concentrations.
4.1. Irrigation pumping
Fig. 1 indicates that, at a national level, pumpingalone drives significant groundwater flow. Boro ricerequires ∼1 m of irrigation annually (Hossain et al.,2003), and roughly 20% of the country cultivatesgroundwater irrigated Boro. Therefore, assuming aporosity of 20%, this withdrawal cycles 5 m ofgroundwater flow annually below rice fields, whichamounts to 1 m of vertical groundwater circulationwhen averaged over the non-irrigated areas. Thus,pumping-induced groundwater flows reach depths of20 or 30 m within two or three decades, on average. (Ofcourse, pumping-induced flow is much greater in thevicinity of irrigation wells and will not be evenlydistributed, but somewhat channeled into higher-conductivity pathways.).
Some researchers have suggested that pumping doesnot greatly change groundwater flow because naturalflow is already rapid (Aggarwal et al., 2003). Pumpingmust change the pattern of flow because it introducesspatially distinct sinks (well screens) into the ground-water system, altering the 3-D flow paths of groundwa-ter. Because hydraulic heads are described by a 3-dimensional parabolic differential equation, inserting asink in the system will change heads, and hence flowthroughout the domain. Groundwater flow is not likerapid river flow; extracting groundwater changes thegradients everywhere, including up-gradient flow.
However, it is possible that pumping would notgreatly change the average residence time of ground-water in the aquifer. Irrigation extraction could be offsetby decreased natural discharge to rivers during the dryseason. Although pumping must change groundwaterflow paths, it would not change the average residencetime if there was no return flow (re-infiltration of
irrigation water) and if total irrigation withdrawalsequaled the reduction in discharge to the rivers. Outflowfrom the system would simply be switched fromdischarge to rivers to increased evapotranspirationfrom crops. Such a scenario assumes perfectly efficientirrigation (i.e. no return flow) and an exact tradeoffbetween discharge to rivers and pumping. However,since there is return flow, there must be an increase inflow through the system (i.e. reduction in averageresidence time) even if the pumping withdrawal is offsetby a decrease in discharge to the river. In the hydrologicmodeling sections below, we address this questiondirectly by estimating fluxes with and without irrigatedagriculture.
4.2. Groundwater tritium
Measured tritium values indicate groundwater flowthrough at least the upper 30 m is often rapid.Measurable values of tritium (≳0.1 T.U.) indicate thatat least a component of the groundwater precipitated inthe last 50 years, during which time atomic bomb testinggreatly increased atmospheric level of tritium. Thelargest set of tritium data has been gathered by the IAEA(Aggarwal et al., 2002) over the last 30 years and theirfinal report concludes that groundwater ages in theupper 100 m are generally less than 100 years. Theirrecent plot (Aggarwal et al., 2003) shows a somewhatmore complex picture, with tritium values greater than 1tritium unit (TU) penetrating below 25 m in 1999, butnot in 1979. At our site, we also find tritium valuesabove 0.2 TU to a depth of 60 m indicating a componentof water less than 45 years. Clearly some areas ofstagnant water (van Geen et al., 2003; Dowling et al.,2002) exist, either because of low hydraulic conductiv-ity (e.g. clay layers) or because of local recharge anddischarge patterns. However, the common occurrence ofmeasurable tritium in groundwater indicates thatgroundwater flow is sufficient to rapidly transportarsenic, or solutes that interact with arsenic, throughlarge portions of the shallow aquifers where high arsenicconcentrations are found.
Irrigation return flow, which now comprises asignificant component of groundwater recharge, posesa challenge for the interpretation of tritium and heliummeasurements in contemporary groundwater. Becauseirrigation return flow is recycled groundwater, it hasdifferent tritium concentrations when infiltrating thanprecipitation, and may have very low tritium concentra-tions. Irrigation water is often withdrawn from below 30m, and available data indicate that tritium levels can below beneath this depth. Helium concentrations in return
11C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
flow also may differ from precipitation if they have notequilibrated with the atmosphere before the water re-enters the subsurface.
We know of no measurements of either the tritiumor helium concentrations in irrigation return flow.However, we can gain insight into how irrigationreturn flow may affect tritium-helium dating byconsidering several possible scenarios. First, thetritium concentrations in return flow may be belowdetection, and the helium may have re-equilibratedwith the atmosphere in the rice fields. When this typeof water is mixed with resident groundwater in thesubsurface, the tritium-helium ratio, from which theage is calculated, remains the same; thus the estimatedtritium-helium age remains unchanged, although thehelium concentration is reduced. A second possibilityis that helium does not equilibrate with the atmo-sphere. In this case, recently infiltrated low-tritiumreturn flow will simply appear to be old groundwater.These two examples both show that tritium and heliumconcentrations may not distinguish irrigation return-flow from resident groundwater.
4.3. Patchiness of dissolved arsenic
Arsenic concentrations are extremely patchy oversmall spatial scales. In the vertical dimension, highconcentrations can be found within tens of meters of lowconcentrations. If there is any groundwater flow acrossthese concentration gradients, arsenic will be trans-ported from areas of high concentrations to areas of lowconcentration, and vice versa. If flow does not crossthese arsenic gradients, then the patchy spatial pattern ofdissolved arsenic must correspond to the spatial patternof groundwater flow paths. In either case, understandingthe effects of flow and transport is important forunderstanding the behavior of dissolved arsenic;groundwater is either transporting arsenic, or the spatialpattern of flow paths is related to the spatial pattern ofdissolved arsenic.
The complex mosaic of recharge and discharge areasat the surface provide one simple potential explanationfor this spatial complexity in the subsurface. Becausethe topography is essentially flat, local flow systemsdominate. Discharge areas (irrigation wells and rivers)and likely recharge areas (pond, rice fields, and rivers)are all spaced within tens and hundreds of meters ofeach other as demonstrated by the satellite image in Fig.3. This spacing of sinks and sources drives groundwaterflow through a complex transient 3-dimensional systemof flow paths which also must have spatial scales of tensand hundreds of meters.
4.4. Why is dissolved arsenic still in the groundwater?
The short residence time (decades) of some contam-inated groundwater, as indicated by both tritiumconcentrations and pumping rates, suggests that arsenicis being flushed from the system. Indeed, estimatedgroundwater ages, and the rates of pumping-inducedcirculation described above, combined with the lowestimated retardation coefficients for arsenic, raises thequestion of how such high concentrations of dissolvedarsenic can remain in the groundwater. Kinniburgh et al.(2003) estimates the retardation factor for arsenic to beas low as two. We also find that, where arsenic is high,the effective retardation factor is less than ten (Harveyet al., 2002; Swartz et al., 2004). These values imply aresidence time for arsenic of decades to centuries inaquifers that are thousands of years old. So why is thearsenic still there? Possible explanations include: (1)Groundwater flow has not been rapidly flushingaquifers during past centuries, but rather is the resultof the recent advent of massive irrigation; (2)Geochemical conditions have recently shifted tomobilize arsenic in greater concentrations, or; (3)Dissolved arsenic is provided hydrologically upgradientof the sampling wells by the near-surface processdescribed in Polizzotto et al. (2006-this issue). Wesuspect that aspects of all three explanations apply tovarying degrees at different locations, and that theirrelative importance will only be elucidated when thegroundwater flow system is better understood.
5. The annual cycle of groundwater flow in centralBangladesh
Discharge and recharge of groundwater follows adramatic annual cycle, with floodwater returning theaquifer to full conditions every year. After floodwatersrecede in the fall, groundwater is lost both by dischargeto rivers and by evapotranspiration, which is enhancedby irrigation pumping. Then in the early summer,groundwater is recharged by direct rainfall, and byrising river levels.
5.1. Hydrologic characteristics of the Munshiganj fieldsite
Here we present hydrologic data that characterizethe annual cycle of water levels at our field site anddemonstrate several features of groundwater flow thatare important for arsenic mobilization and transport.Fig. 3B shows a map of a region within our largerstudy area. The hydrostratigraphy of our site (Fig. 4)
12 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
is representative of the deltaic depositional environ-ment that is pervasive in Bangladesh and West Bengal(Khan, 2000). The surface landscape consists of: (1)organic-rich clay/silt-lined fields; (2) ponds that areseveral meters deep with low-permeability claybottoms; (3) villages and roads, rising 2–3 m abovethe level of the fields, that are constructed on clay/siltborrowed to excavate the ponds; (4) and the IchamatiRiver, which traverses the site and eventually flowsinto the Ganges approximately 7 km away. The surfaceoverbank clay/silt deposit extends to a depth ofapproximately 3 m below ground surface, and acts asa confining or semi-confining layer to the underlyingsand aquifer. The approximately 80–100 m thick sandaquifer is fairly homogeneous at our site, althoughlocalized peat and silts have been reported in otherlocations (Rahman and Ravenscroft, 2003; McArthur etal., 2004). Groundwater wells are completed in thisaquifer, most screened at around 30 m depth, wherearsenic concentrations appear to be highest, with someirrigation wells screened deeper.
According to BADC (2002), about 45% of the totalarea (203 km2) of Sreenagar Thana, where our field sitein Munshiganj is located, was irrigated during the 2002dry season. Of this total area, 52 km2 (or about 26% ofthe total area of the Thana) was irrigated by 996 shallowirrigation wells, 3 km2 was irrigated by 12 deep wells,and 37 km2 by surface water using 223 low lift pumps(LLP). Thus there are about 5 irrigation wells per squarekilometer area of the Thana and each shallow irrigationwell covers about 0.053 km2 of agricultural land. Itshould be noted that significantly higher density ofirrigation wells could be found in the northern and insome south-western districts of Bangladesh (Ali, 2003),where groundwater is easily available at shallow depthand availability of surface water for irrigation during thedry season is extremely limited. For example, in DhunatThana of the northern district of Bogra there are about34 shallow irrigation wells per square kilometer of theThana and each shallow well covers about 0.02 km2 ofagricultural land (BADC, 2002).
Annual hydrographs for irrigation wells, domesticwells, ponds, and the river are presented in Fig. 5A forthe 14 months from May 2003 to June 2004. Some basicfeatures of these water levels are the following: (1)During monsoon flooding, all water levels and hydraulicheads become much closer than the rest of the year.Because the head differences are relatively small, and theonly pumping is domestic (irrigation ceases), flow isgreatly reduced during flooding. (2) As the floodrecedes, but before irrigation begins, the river leveldrops more quickly than the groundwater level, which
drops more quickly than pond levels. (3)When irrigationbegins in January, the decline of the river level slows butgroundwater decline accelerates such that the hydraulichead in the aquifer falls below that in the river. (4) Whenmonsoon flooding begins in June, all water levels riserapidly together as the aquifer is recharged and both riverwater and rain inundate the land.
Our water-level observations allow us to develop aconceptual model of water flow in the vicinity of ourfield site. Generally, recharge enters the subsurfacethrough pond and river bottoms, and the overlyingconfining unit, into the aquifer below. Undisturbed, thewater would then flow laterally out to discharge into therivers, or be drawn up by transpiration. Indeed, the dropin water levels in the aquifer after flooding recedes inDecember, but before irrigation pumping, is largelycaused by discharge to the river.
The groundwater levels recorded every hour at ourintensive study area (Fig. 5B) show that beginning inJanuary, irrigation pumping greatly changes flow in thesystem. At the onset of irrigation pumping, the rate atwhich groundwater heads decline doubles, and dramat-ic diurnal head oscillations develop as pumps are turnedon during the day and off at night. The dramatic diurnalhead differences demonstrate that the aquifer isconfined, or semi-confined. Also in the beginning ofJanuary, the river reverses flow direction and no longerdischarges to the Ganges, but rather flows, at arelatively low rate, from the Ganges to the field area.
While spatial gradients in the aquifer are very small,temporal fluctuations, caused by irregular pumpingschedules, are more significant and spatially extensive,consistent with the high transmissivity of the aquifer.Heads recorded in different wells at approximately thesame time rarely differ by more than several centi-meters, yet the effect of the seasonal trend over a day ismore than a centimeter and the daily oscillation fromirrigation pumping can be more than 20 cm. From apractical point of view, this makes it very difficult tomap groundwater flow directions. In the time required towalk from one well to another, the head in both wellsmay change by an amount larger than the instantaneousdifference in head between the two wells. Thus,groundwater flow patterns are not readily apparentfrom the water level data in Fig. 3 and are a focus ofcurrent 3-D numerical modeling.
5.2. Dynamic lumped-parameter model of the Mun-shiganj aquifer
Here we construct a dynamic model of recharge anddischarge from the aquifer (Fig. 7). We use this model to
13C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
estimate the fluxes between the aquifer and the river,ponds, rice fields and villages, flows that cannot all beexplicitly calculated from the data. First we use themodel to estimate unknown hydrologic parameters.Then we apply the calibrated model to estimate thevarious fluxes in and out of the aquifer, with and withoutpumping, and to characterize the uncertainty in thesefluxes given current data.
We use a lumped-parameter, or box, model thatdoes not consider spatial gradients within the aquifer.This modeling choice is supported by the facts that:(1) Temporal gradients in hydraulic head overwhelmspatial gradients within the aquifer over both theannual cycle and the daily cycle, and; (2) Spatialgradients within the aquifer are too small to bemeasured by conventional methods, but gradientsbetween the aquifer and surface water bodies are
larger. The lumped-parameter model (Fig. 6) couplesthe mass-balance equation for the aquifer with mass-balance equations for irrigated fields, ponds, and non-irrigated areas (mostly villages and roads).
Fig. 6. Schematic diagram of the lumped-parameter model highlight-ing different recharge and discharge fluxes along with the parameterand variables used in for the numerical model.
14 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
Pond:dhpdt
¼ ha−hp� �
kp−apET0 þ R ð4Þ
The following time-varying model output was fit toobserved values to estimate the system parameters:
ha head in the aquifer [L]hf head (water level) in the rice fields [L]hp head (water level) in the ponds [L]hv head in the non-irrigated areas (e.g. villages)
[L]
The following time series were set to measured datavalues:
hr river stage, a function of time [L]qI pumping rate per unit area L
T
� �ET0 reference crop evapotranspiration L
T
� �R rainfall rate L
T
� �
The following system parameters were estimated inall cases:
kr hydraulic conductance between the river andthe aquifer 1
T
� �kf hydraulic conductance between the rice fields
and the aquifer 1T
� �kp hydraulic conductance between the ponds and
the aquifer 1T
� �kv hydraulic conductance between the non-irri-
gated areas and the aquifer 1T
� �
The following dimensionless parameters were fixedto independently estimated values in all cases:
αv the scaling factor for non-irrigated area tran-spiration (i.e. trees)
αf the scaling factor for rice field evapotranspirationαp the scaling factor for pond evaporation (i.e. pan
evaporation) S Storativity of AquiferSy specific yield of near-surface clay (This value
is set to 1 when the head is above the landsurface indicating standing water in the ricefields.)
ff fraction of area covered by fields (65%)fp fraction of area covered by ponds (10%)fv fraction of area covered by nonirrigated areas
(e.g. villages; 22%)fr fraction of areas covered by rivers (2%)fav aquifer-clay partition coefficient (fraction of
ETtree coming out of aquifer)
The left side of the aquifer equation describes thechange in storage in the aquifer. The first four terms onthe right side describe exchange with irrigated fields,ponds, the river, and non-irrigated areas, respectively.The final two terms represent pumping and transpirationfrom trees that may extend some roots into the aquifer.The hydraulic conductances (not conductivities) used inthe model characterize the flow rate per unit areabetween two model reservoirs. For example the flowrate between the field and aquifer isQf =kf (hf−ha), L3
T
h i.
The transient mass balance equations for the fields,ponds and non-irrigated areas are each coupled to themass balance equation for the aquifer through thehydraulic head in the aquifer, ha. Because the water levelin the river is controlled by the Ganges level (asdiscussed below) it is prescribed by the measured waterlevels, and there is no mass balance equation for theriver. In the next sections we describe how our field datawere used to construct the lumped parameter model.
5.2.1. Rates of groundwater extractionWe determined extraction rates for our study area by
combining direct measurements of instantaneous pump-ing rates from local irrigation wells with the extractionschedule estimated from daily oscillations in aquiferhydraulic head. We then apply these pumping rates tothe mass balance equations in order to model seasonalhydrological fluxes. Fig. 7 shows a histogram of themeasured instantaneous pumping rates from 41 of the 54irrigation wells we have identified in the area surround-ing our field site shown in Fig. 3A. Our averagemeasured rate of 24.1 L/s is consistent with the range
Fig. 7. Distribution of measured pumping rates of 41 irrigation wellswithin the Munshiganj study area. The mean and median of thedistribution is 24.1 L/s and 25.0 L/s, respectively. Rates were measureddirectly by recording the time to fill known volume.
15C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
recently published by Shankar et al. (2005), butsignificantly greater than the value of 15 L/s cited inolder reports for shallow irrigation pumps (Rahmanand Ravenscroft, 2003). The daily extraction was then
Fig. 8. (A) Aquifer head data for a 30-m well at the intensive field site for thefluctuations in groundwater head, calculated by subtracting a 24-h moving apumping duration for the 2003–2004 irrigation season. (D) Rate of daily aquiby the pumping duration for each day. Data is used to scale the number of w
computed as the average pumping rate multiplied byboth the number of wells in use and the duration ofpumping. As we discuss next, the last two quantitieswere estimated from observed water level fluctuationsrecorded by our automatic pressure transducers. Ourcalculations are applied specifically to a circular areawith a 1.5 km radius centered at our intense field area(Fig. 3A) that contains 32 irrigation wells. This area islarge enough to be representative of a typical area; itcontains many ponds, irrigation wells and a smallriver.
Irrigation pumping causes both daily fluctuations ingroundwater heads (pumps are typically run from about9 in the morning until 7 or 8 in the evening), and asecular decline in heads that is the accumulated effect ofextraction over the season. Both the daily and theseasonal fluctuations are clearly visible in the hydro-graph shown in Fig. 8A. The daily duration of pumpingcan be determined from the hydrograph as the timebetween the peak head at the start of daily pumping, andthe minimum head at the end. These daily oscillations
period of December 2003 to July 2004 as shown in Fig. 5A. (B) Dailyverage from the raw head data presented in panel (A). (C) Calculatedfer drawdown, calculated by dividing the amplitude of head oscillationsells pumping on each day.
16 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
are shown in Fig. 8B after a 24-h moving average issubtracted from the hydrograph to remove the effects thelong-term seasonal decline, and precipitation andrecharge. The magnitude of the oscillations show thatthe net extraction increases in January, the first month ofthe irrigation season, and then begins to decrease duringApril, with much less pumping in early May. Fig. 8Cshows that the daily duration of pumping is nearlyconstant at around 10 hours. Thus, it appears that thetotal daily extraction is controlled more by the numberof wells pumping than by the hours of operation, afinding that is consistent with our own observations oflocal practices.
We used the magnitude of the daily drawdown tocompute the number of wells extracting each day.Because drawdown is linearly proportional to the rate ofextraction in a confined aquifer, we compute theproportion of the 32 wells in the modeled area that arepumping each day as the ratio of the daily drawdown tothe maximum daily drawdown. To account for thevariable duration of pumping each day, we perform thiscalculation with normalized daily drawdowns defined asthe daily drawdown divided by the duration of pumping,i.e., a rate of drawdown. Fig. 9 confirms that the rate ofdrawdown is nearly constant over a day, a necessarycondition for this normalization. This computed rate ofdrawdown is shown in Fig. 8D. We estimate themaximum daily decline, corresponding to all 32 wellsextracting, as the mean daily rate plus one standard
Fig. 9. Head data and exponential curve fit for: (i) five days in January with a 1duration, (iii) 100 consecutive days beginning January 7, 2002, with variablemaximum, then averaged on each subsequent hour of pumping, ending at thedecline increases from January to February, and that the rate is approximatelyfrom concave to convex.
deviation, since it is unlikely that the data recordcontains only one day where all 32 wells are pumping.This assumption in effect implies that all wells wereused together 19 days out of the entire 104-day irrigationseason. By summing the product of the number of wellspumping on each day, the average pumping rate, and theduration of pumping, and dividing by the irrigated areawithin the 1.5 km radius (calculated from the IKONISimages Fig. 3A to be 65% of the total area), we estimatethat 0.72 m of water was applied to the fields during theirrigation season.
We corroborated these results by employing a localfarmer to record his schedule of pumping from oneirrigation well over the entire season. We determinedthat the local farmer irrigated his field with a total of 0.9m of water over the season, a number consistent with thevalue of 0.72 estimated for our model area, and alsoconsistent with the 1 m of irrigation water for Boro ricecultivation used earlier in our national calculations (Fig.1). However, the proportion of land (∼65%) used forirrigated rice in our research area is larger than thenational proportion that includes large parts of thecountry with less or no rice cultivation. The monthlyextraction rates are plotted in Fig. 3D. Both theextraction rate of this well (∼13 L/s) and the areairrigated (3.17 ha) (indicated on map in Fig. 3A) arerelatively small. Irrigation return flow to the aquifer canbe roughly estimated as the water applied to the ricefields minus the water stored in the rice fields and the
0-h pumping duration, (ii) five days in February with an 11-h pumpingpumping duration. The data on each day was normalized to zero at theminimum head value, or end of pumping. The fits show that the rate oflinear with time. However, the data also show an unexplained inflection
Fig. 11. Comparison between Ganges water level (as recorded inBhagakul Meteorological Station BWDB, 2004) and Ichhamati waterlevel (within the study area). The water levels follow each otherclosely. The pink squares represent the average of both the mainchannel and side channel (Fig. 3) that we used to fit the model, whereasthe green triangles represent the water levels in the main channel only.The oscillations in the Ganges water level at a 14-day frequency are
17C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
water lost to evapotranspiration. Given that rice fieldsare usually irrigated to maintain a standing water depthof ∼20 cm (Ali, personal communication), and weestimate evapotranspiration at 28 cm (see below), thisreturn flow is 0.72 m− (0.2 m+0.28 m)=0.24 m for themonths of January, February and March. This roughcalculation is consistent with previous calculations ofirrigation efficiency from the agricultural literature (Yu,2003) and suggests that irrigation return flow is animportant component of the hydrologic system.
5.2.2. Exchange with the river and pondsThe large head difference between the aquifer and the
Ichamati river (1–2 m, Fig. 10) during the dry seasondemonstrates the potential for significant exchangebetween the river and aquifer, but also shows that thehydraulic conductivity of the river sediments is muchlower than the aquifer conductivity. The sediments in
Fig. 10. River stage of the Ichamati River (also in Fig. 3C) andhydraulic head measured in nearby wells (mapped in Fig. 4B) duringDecember 2003 to July 2004. The error bars (of 20 cm) associated withthe groundwater levels (from January to April) account for the waterlevel fluctuation due to irrigation pumping (Fig. 5B). The groundwaterheads are located along a transect roughly perpendicular to the river,but show no appreciable hydraulic gradient even though the river headdiffers from the aquifer head by more than a meter.
likely the effect of aliasing with tides: measurements are take twiceevery day at the same time, so tidal oscillations in the Ganges appear ascycles with 14 day frequencies. The groundwater heads at our site (Fig.8) do not show any evidence of tides.
the river channel must severely restrict exchange withthe aquifer, however we cannot directly calculate thisflux. Because we do not have measurements of thehydraulic conductivity or thickness of these sediments,we cannot directly estimate the exchange between theaquifer and the river. Furthermore, the exchange withthe river cannot be estimated from Darcy's law cal-culations of flux in the aquifer perpendicular to the riverbecause spatial hydraulic gradients within the aquiferare too small to be measured (Fig. 10).
The level of the Ichamati river appears to becontrolled by the Ganges river even in the dry seasonwhen parts of the river become very shallow (b1 mdepth). Our measured water levels for the Ichamati riverclosely follow the water levels recorded by theBangladesh Water Development Board for the GangesRiver at the station in Bhagakul, about 7 km south to ourfield site (Fig. 11), indicating that the Ichamati ishydraulically connected to the Ganges. Thus, the riverlevel is not a function of local aquifer discharge andrecharge, and may be modeled simply as a prescribedhead.
Ponds cover ∼10% of the land and the majority ofthese ponds have likely been excavated over the last 50years as the population has greatly increased. The pondsare excavated to provide clay/silt material for construc-tion of villages above the monsoon flood levels. Theyreceive most of the human waste from the villages andare also used for aquiculture. These ponds do not have a
18 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
direct connection to the rivers during the dry season.Most are surrounded by clay banks that maintain thewater level several meters above both the river stage andthe underlying aquifer (Fig. 3C). However, waterovertops these banks during the flooding season,connecting the ponds to the floodwaters. During theperiod after flooding and before irrigation, and also theperiod following irrigation but before the next flood,pond levels are often manipulated for aquiculture byopening and closing canals, so the water levels in theponds may change dramatically.
During the irrigation season, pond levels drop due toboth evaporation and flow to the aquifer. Fig. 12 showsthis rate of decline computed from heads monitored ineight separate ponds. All eight ponds lose water at a ratefaster than can be explained by evapotranspiration,indicating loss to the aquifer. The average decline for theponds over the period from January 15, 2004 to April 24,2004 indicates that ponds contribute about 0.53 mm/dayof water to the aquifer, which is about 42% of irrigationpumping.
5.2.3. Rainfall, evapotranspiration, and storage fluxesFig. 3D shows the average total monthly rainfalls,
and one standard deviation, calculated from the last30 years of data recorded in Bhagakul MeteorologicalStation, about 4 km south–west to our field site inMunshiganj (WARPO, 2000). The monthly rainfallfor 2003 (BWDB, 2004) is also plotted. Our hy-drologic model uses the 2003 monthly hydrologicdata, but such data is not yet available for 2004, so
Fig. 12. The rate of pond water level (PWL) decline for 7 ponds (mapped inperiod of December 2003 to April 2004. Rates of PWL decline were calculatmeasurements. Positive differences indicate a decline.
we must use averages over the last 30 years for 2004.Fortunately, rainfall is a minor component of thehydrologic system during the first months of the year,the important irrigation period during which the headsdecline.
Fig. 3D also shows the total monthly referenceevapotranspiration for the Munshiganj district averagedover the last 30 years (WARPO, 2000). These valueswere calculated from the FAO Penman-Monteithequation (Allen et al., 1998), and vary little from yearto year. The reference evapotranspiration, ET0, wasconverted into pond evaporation, rice field evapotrans-piration, and evapotranspiration from non-irrigatedareas using the following factors: ETpan=αpanET0
(αpan=1.4), ETcrop=αcropET0 (αcrop=0.9), ETtree=α tree
ET0 (αtree=0.95), where, ET0, ETpan, ETcrop, and ETtree
are used for the reference crop, pond water surface, riceand trees, respectively.
Transpiration by trees in villages and along roads(non-irrigated areas) is an important component of thehydrologic system. If trees primarily extract theirwater from the near-surface clay sediments, they willhave little impact on aquifer flow. However, if theirroots extend 5 or 6 m (through both the clay moundedto form the village and the underlying overbank-deposit clay) to the much higher yielding aquifersediments, then transpiration from the trees will be asignificant flux of water from the aquifer. ETtree iscalculated from ET0 values shown in Fig. 3D. Duringthe irrigation period of mid-January to late-April, thetotal value of ETtree is ∼40 cm. Since villages cover
Fig. 3A), and the rate of pan evaporation (ET) and rainfall (RF) for theed by dividing the difference in PWLs by the duration between the two
19C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
22% of the land, this evapotranspiration flux fromtrees amounts to ∼65% of irrigation pumping. Wehave found no literature specifically describing theroot depth of these trees, however many other speciesare known to extend roots well below 5 m, and it maybe to the advantage of these trees to extract waterfrom below the very low permeability clay of thevillages. However, it is also possible that the rootsextract most of their water from the clay. During thedry season, the estimated total ETtree from thevillages), ∼40 cm, can be supplied by the approxi-mately 5 m of clay underlying the villages byreducing the water content of the clay by ∼8%.Since clay is compressible, this would imply that theelevation of villages and roads could rise and fall∼40 cm a year! Given this uncertainty, we presentmodeling results for the two extreme cases where allET in the non-irrigated areas is extracted from eitherthe near-surface clay or from the aquifer, but find thatit makes little difference for other aquifer fluxes.
Water also flows in and out of storage, both frompore-space in the near-surface clays, and in and out ofelastic storage in the aquifer. For the model calcula-tions plotted here, the aquifer storativity S wasspecified at 0.01, corresponding to a specific storageof 10−4 m−1 over a thickness of 100 m, and thespecific yield of the clay was set at 0.02. We chooseboth of these values to be at the high end of previousestimates (Rahman and Ravenscroft, 2003; Yu, 2003),because previous estimates are largely based onpumping tests conducted over durations shorter thanthe seasonal time scale of our model. Over longertime-scales, the high-storage deep clay that forms thelower confining unit to the aquifer will play a greaterrole in water storage, and in effect, increase thestorage of the aquifer. The specific yield of the surfacesediments is only significant in our hypotheticalsimulations with no pumping, and we also suspect
Table 1The estimated conductance parameter values when the storage coefficients amodeled residence times for the aquifer
Kf (1/d) [conductance for field]Kv (1/d) [conductance for village]Kp (1/d) [conductance for pond]Kr (1/d) [conductance for river]Objective function w/pumpingResidence time (years) w/pumping
w/o pumping
that a large value is appropriate for this parameterbecause over the long duration of a year more waterwill drain from a soil. To test the importance of thesestorage coefficient values, we also conducted severalmodel parameter estimation and simulation runs withS=0.001 and Sy=0, i.e. no storage in the soil.
5.3. Model fitting
We simultaneously solved the equations for thelumped-parameter model numerically with the Matlabfunction ODE45, and embed this numerical solutioninto Matlab's lsqnonlin.m least squares algorithm toestimate the four conductances by minimizing the sumof square differences between the measured andmodeled heads. The terms of the sum of square errorsare:
SSE ¼Xna
iðha;i− hma;iÞ2 þ
Xnv
iðhv;i− hmv;iÞ2
þXnf
iðhf ;i− hmf ;iÞ2 þ
Xnp
iðhp;i− hmp;iÞ2 ð5Þ
where na, nv, nf, np, are the number of headobservations in (respectively) the aquifer, the claysin the non-irrigated villages and roads, the clays in thefield and the ponds, ha,i, hv,i, hf,i, hp,i, are the observedheads in the aquifer, non-irrigated areas, fields andponds, and ha,i
m , hv,im , hf,i
m , hp,im , are the modeled heads in
the aquifer, non-irrigated areas, fields and ponds. Weassessed the quality of the parameter fits from the sumof square errors by computing the parameter covari-ance matrix as the Cramer-Rao lower bound (Miltonand Arnold, 1995) using the Matlab function n4sid.The covariance matrix provides information aboutparameter uncertainty and correlation. The diagonalproves the variance of the parameter estimates and theoff-diagonal terms indicate the covariance of theestimates. If parameters are highly correlated, then
re fixed, the respective objective functions (sum of square errors), and
Village ETtree from:
Clay (Case A) Aquifer (Case B)
8.9×10−4 8.9×10−4
6.3×10−6 9.1×10−4
9.3×10−3 8.3×10−3
7.7×10−2 8.7×l0−2
5.9×10−1 5.7×10−1
19 1342 22
Table 3The correlation coefficients and the coefficient of variation (CV) ofestimated parameters when the four conductance parameters and thetwo storage parameters are all estimated simultaneously
Kf Kp Kr Kv Sy S CV
Case A: Village ET out of clayKf 1 0.15Kp −0.17 1 0.09Kr −0.31 0.04 1 0.10Kv −0.34 0.02 −0.22 1 0.77Sy −0.26 0.02 −0.29 0.92 1 0.14S −0.24 −0.07 0.81 −0.15 −0.19 1 0.13
Case B: Village ET out of aquiferKf 1 0.18Kp −0.11 1 0.09Kr −0.39 −0.06 1 0.11Kv −0.36 0.10 −0.07 1 0.08Sy −0.36 0.11 −0.10 0.79 1 0.09S −0.26 −0.14 0.83 −0.16 −0.22 1 0.15
Both the cases where ET from the village trees is withdrawn fromvillage clay and from the underlying aquifer are given.
Table 2The correlation coefficients and the coefficient of variation (CV) ofestimated parameters when the four conductance parameters areestimated simultaneously
Kf Kp Kr Kv CV
Case A: Village ET out of clayKf 1Kp −0.18 1 0.11Kr −0.28 0.29 1 0.11Kv −0.30 0.01 0.09 1 5.45
Case B: Village ET out of aquiferKf 1 0.10Kp −0.21 1 0.10Kr −0.31 0.23 1 0.10Kv −0.11 0.05 0.00 1 0.06
Both the cases where ET from the village trees is withdrawn fromvillage clay and from the underlying aquifer are given.
20 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
they cannot be independently estimated with reliabil-ity, suggesting that more data is required or that themodel is incorrectly structured.
5.4. Model results
The fitted conductances and their estimated uncer-tainties, Tables 1–3, suggest that the simple lumpedparameter model may accurately represent the basicbehavior of the true system. First, the modeled headsclosely fit the observed heads and water levels (Fig. 13).Second, the estimated parameter uncertainties are low asevidenced by their small coefficients of variation, andthey are not strongly correlated, suggesting that themodel is not over parameterized. Third, the estimatedvalues for all four conductances (Table 1) are typical forthese types of clay or silty sediments (Freeze andCherry, 1979).
Because we do not know the depth of tree roots, wefit conductances for the two extreme cases wheretranspiration is drawn up by roots either in the near-surface clay only, or in the underlying aquifer only. Aswe describe later, both approaches lead to the sameconclusions about the role of hydrology on arsenicbehavior.
The conductance estimates are in accord withexpectations for interface sediments between the aquiferand various reservoirs. To compute the hydraulicconductivity of these interface sediments, we mustmultiply the conductance by the thickness of the unit inthe direction of flow. For the river-aquifer interface, withan assumed 2 m thick interface unit, the hydraulicconductivity is 1.8×10−6 m/s, and 2.0×10−6 m/s for thecase where transpiration from trees is drawn from theclay and aquifer, respectively. This corresponds to a
silty-clay. Similarly, assuming a 1-m thick unit at thebottom of the pond, a 2-m thick unit for the fields and a6-m thick unit for the non-irrigated villages and roadsyields hydraulic conductivities for the pond of1.0×10−7 m/s and 9.6×10−8 m/s, for the fields of2.0×10−8 m/s and 2.0×10−8 m/s and for the non-irrigated areas of 4.4×10−10 m/s and 6.3×10−8 m/s,where the two figures correspond to cases where treetranspiration is drawn from the clay and aquifer,respectively.
The coefficient of variation of the estimatedparameters is well below 20% for all the conductancesexcept for the conductance of non-irrigated areas whenthe ET is taken from the non-irrigated clay and withspecified storage parameters. In this case, the coefficientof variation is∼550% indicating that the modeled headsare insensitive to this parameter. When the ET is takenfrom the clay, there is very little flow from non-irrigatedareas to the aquifer and the heads in the non-irrigatedareas agree well with the data (Fig. 13A). Thus, for thiscase, the conductance of the clay in the non-irrigatedareas does not influence the modeled heads significantlyand so cannot be estimated accurately. These sedimentscontribute a relatively small component of the hydro-logic cycle and the model is therefore relativelyinsensitive to this parameter, as indicated by the largecoefficient of variation.
We fixed the value of the two storage parameters, thespecific yield of the clay (Sy=0.2) and the storativity ofthe aquifer (S=0.01) because neither of these parameterscan be reliably estimated from available data. Also, the
Fig. 13. (A) Model fits and predictions for the case where transpiration from the villages (trees) is extracted from the village clay. The first panel(upper left) shows the best model fit to the data, with the corresponding model fluxes in and out of the aquifer system plotted below. The pumping fluxis prescribed. The upper right panel shows the predicted heads in the absence of pumping and irrigation, with the corresponding fluxes plotted below.(B) The same set of plots as in (A), except here the transpiration of the villages is modeled as coming from the aquifer (i.e. tree roots are all modeled asextending through the village clay). The model results differ because they now show significant ET from the aquifer and a roughly correspondingincrease in recharge to the aquifer from the village clay.
21C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
22 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
values assigned to these parameters have little impact onour results. The specific yield of the clay underlying therice fields has little impact during irrigation becausewater is ponded above the clay, and accordingly theseareas are given a storage parameter value of 1 duringponding in the model. The specific yield of the villageclay also has little impact because: (1) in the case wheretranspiration is extracted from the clay, there isessentially no flow to the aquifer, and; (2) in the casewhere trees draw water from the aquifer, the specificyield cannot be estimated independently of the conduc-tance of the clay (Sy and kv are highly correlated, Table3) since increasing either parameter will increase theflux to the aquifer. Furthermore, the specific yield of theclay has little impact on predicted flows in the absenceof irrigation because the rice fields yield little water tothe aquifer (Fig. 14).
The storativity of the aquifer cannot be estimatedaccurately because the estimated value is stronglycorrelated to the estimated conductance of the river.Including aquifer storativity in the parameter estimationprocedure results in a correlation coefficient of 0.81 forS and kr (Table 3). The parameters are linked becausethey affect the modeled heads in the same way: in the
Fig. 14. The estimated (with pumping) and predicted (without pumping)transpiration from the village trees is extracted from the clay (A) and from thethe instantaneous fluxes shown in Fig. 13 over time.
first half of the dry season, as heads are falling,increasing either the model aquifer storage or the riverconductivity will decrease the rate of head decline;likewise, during the latter period as heads are rising,increasing either parameter will decrease the rate of headrise. Thus, if the aquifer storage is much lower than0.01, the flux in and out of the river will be decreased bythe same amount as the decrease of flux in and out ofstorage. This flux is relatively small (Fig. 14), so thecalculated residence times will only slightly increase forcurrent conditions, but the effect on residence times ismore significant in the absence of pumping becauseother fluxes are smaller. Without pumping, modelresults for S=0.001 and Sy=0 (no soil flux) indicatethe groundwater residence time may be increased toover 120 years (not shown).
5.5. Recharge and discharge fluxes and their implica-tions for arsenic
Themodel was used to examine water residence timesas well as fluxes between the various reservoirs, with andwithout irrigation pumping. The net yearly flux into orout of the aquifer, shown in Fig. 14, can be used to
annual water fluxes into and out of the aquifer, for the case whereunderlying aquifer (B). These yearly fluxes are calculated by integrating
23C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
compute water residence times within the aquifer bydividing the aquifer volume by the yearly flux. Theseresidence times, Table 1, show that for the pumpingscenario the residence time is 26 and 38 years,respectively, for the two ETcases, and without pumping,residence times are 44 and 84 years, respectively(assuming the presence of the modern number ofponds). These residence times are consistent with thebomb-levels of tritium and carbon-14 measured to adepth of 30 m. An important conclusion can beimmediately drawn from the analysis—the flushingtime of the aquifer, with or without pumping, is rapidcompared to the age of the basin. Accordingly, the recentintroduction of large-scale pumping itself may not solelyexplain why arsenic has not been flushed from theaquifer long ago. As we discuss next, pumping does,however, change the hydrologic balance and watersources to the aquifer, which likely has major implica-tions for arsenic dynamics. It should also be rememberedthat the residence time is not the same as the averagegroundwater age; a low residence time may imply rapidflow at only very shallow depths with stagnant waterbelow. The lumped-parameter model does not distin-guish flow rates at different depths.
Fig. 13 shows the estimated water fluxes (volumeflux per total land area in mm/day) as a function of time,for the two different ET cases: with and withoutpumping. The line labeled “storage” shows the net rateof water release from (positive) or replenishment to theaquifer (negative), which must balance the sum of all theother fluxes at each instant.
We first discuss the current situation with pumping.For the ET from clay case, the model predicts that duringpumping, water initially leaves the aquifer to the river,but then the fluxes reverse, and the river becomes asource of water to the aquifer. Water is released fromstorage in the aquifer at the onset of pumping, and isreturned to storage at the cessation of the pumpingseason. During most of the irrigation season, the aquiferis neither releasing nor taking significant water fromstorage, but is acting as a flow-through reservoir. Watermainly leaves the aquifer by pumping, and enters atroughly equal rates from the irrigated fields and pondsand, in the later half of the irrigation season, from rivers.There is very little flux to or from the non-irrigatedvillage areas. The water balance for the case where treetranspiration is taken directly from the aquifer appearsroughly the same as for the case of tree transpiration lossfrom the non-irrigated clays, except that the transpirationloss from the aquifer, which is only slightly smaller thanthe pond and field flux, is approximately balanced by aleakage flux from the non-irrigated areas. In summary,
with pumping, water principally leaves the aquifer bypumping, and to a lesser degree is initially lost to theriver. Water principally enters the aquifer from theirrigated fields and ponds, and later in the season fromthe river. Because the aquifer is semi-confined, andbecause there is little rainfall during the irrigation season,direct rainfall is a minor component of aquifer recharge.The ET and leakage from the non-irrigated village areasbalance each other out.
While the vigor of the aquifer circulation is reducedin the non-pumping scenario, the more importantobservation may be that water balance shifts. Again,transpiration and leakage from the village areas (non-irrigated) more or less balance. Water principally leavesthrough the river over most of the year, but water fluxesfrom the ponds are approximately two times greater thanfluxes from the fields. The fluxes from the fields arereduced by two compounding effects. First, the fieldsare no longer flooded in the non-pumping scenario, sothere is no ponded water to drive water down into theaquifer. Second, the downward gradient into the aquiferis less without pumping, because the hydraulic head inthe aquifer is not drawn down. The large source of waterfrom the ponds does not necessarily represent pre-irrigation conditions, because many of the ponds likelydid not exist more than 50 years ago. If the ponds areremoved from the model to represent past conditions,then the estimated flux through the system is greatlyreduced. Consequently, the groundwater residence timesmay have been much greater in the past when there weremany fewer excavated ponds.
The hydrologic balances computed above forpumping and non-pumping suggest that the pondsand flooded rice fields are likely recent sources ofanaerobic, organic-rich recharge waters. Indeed, wide-spread flooding during the dry season coupled withpumping would both promote anaerobic conditionsand provide the gradients necessary to drive water intothe aquifer. In the absence of pumping, the flux ofwater from the fields is diminished by half but moreimportantly would likely be much less reducing in theabsence of flooding and rice cultivation. In contrast,wide-spread flooding during the monsoon causes littleinfiltration of reducing bottom water because heads inthe aquifer, rivers and fields equilibrate. Because theaquifer is semi-confined, heads rise quickly afterirrigation ceases from the input of relatively smallamounts of surface water. Thus, pumping-inducedshifts in the hydrology of Bangladesh coupled withpopulation growth, rice cultivation and excavation ofponds may be creating large perturbations in thesubsurface biogeochemistry of Bangladesh.
24 C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
6. Conclusion
Arsenic concentrations in groundwater in Bangla-desh vary greatly over short distances; water withdangerously high levels is found in wells that are tensof meters from wells with safe water. Undoubtedlysome of this variation is caused by differences in bulksediment composition, such as variations in thesorption capacity of grey sediment and orange/brown sediment that contains more oxides. However,large gradients in arsenic concentrations are foundwithin grey aquifer sediment where there is littledifference in the solid characteristics and manydrinking water wells extract groundwater. At ourstudy site in Munshiganj, a wide variety of datasupport the hypothesis that the pattern of groundwaterarsenic concentrations is related to the pattern ofgroundwater flow. Where the mineral composition ofaquifer material is homogeneous, the most likelyexplanation for large differences in groundwaterarsenic concentration is that concentrations are relatedto the flow path of the groundwater. Some flow paths,such as fast paths terminating in irrigation wells, mayflush arsenic from the system, and may now have lowarsenic concentrations. Other flow paths may intro-duce oxidants that immobilize dissolved arsenic andhence contain relatively low concentrations. Yet otherflow paths may have high arsenic concentrationsbecause they originate from areas where reductiveprocesses are mobilizing arsenic. Such a flow path isevident at 30 m depth at our field site where arsenicand the products of organic carbon oxidation arestrongly correlated. Because areas of recharge (ricefields, ponds, river channels) and discharge (irrigationwells, river channels) are spaced tens and hundreds ofmeters apart, the chemical heterogeneity caused bydifferent flow paths will be similarly spaced.
Groundwater irrigation has greatly changed thelocation, timing and chemical content of recharge tothe aquifer. During the dry season, irrigation water isponded in rice fields over approximately half of ourtotal research site, thereby changing both the hydro-logic budget and the biogeochemistry of rechargewater. Irrigation enhances recharge by both loweringthe head in the aquifer beneath the near-surface clay,and by raising the head above the clay. The resultingreturn flow to the aquifer is anoxic and probably rich inorganic matter, potentially driving reduction of oxidesthat may be present in the near surface sediments, ordeeper in the aquifer. This process does not repeatduring the flood season. When the land is inundatedduring the monsoon, near surface sediment porewater
may become anoxic, but recharge ceases because thereis no driving potential gradient in the water column.
Permanent constructed surfacewaterbodies, primarilyponds excavated near villages, provide a large source ofrecharge to the aquifer. Because these ponds receive thewaste from nearby villages, they are a potential source oforganic carbon to the subsurface, and because they arelocatednear thedrinkingwaterwells in thesevillages, theymay affect the groundwater withdrawn by these wells.
The hydrologic data and modeling from our siteindicate that groundwater flow is vigorous, flushing theaquifer over time-scales of decades but also rapidlytransporting a solute load into the aquifer with rechargewater from rice fields, ponds and rivers. The net effect iscertainly a complex transient three-dimensional patternof groundwater flow paths with differing soluteconcentrations. Fully understanding the spatial andtemporal patterns of dissolved arsenic concentrationswill require detailed distributed-parameter hydrogeolo-gic models that describe how groundwater carriessolutes into the subsurface, and flushes solutes fromthe subsurface. Constructing such models will requirethe level of hydrogeologic characterization that isemployed at groundwater contamination sites in NorthAmerica and Europe. Only with such characterizationand modeling will we address broad questions such aswhether arsenic concentrations will generally rise or fallover time, or whether dissolved arsenic will penetratedeeper into aquifers in the future. The combinedhydrologic and biogeochemical results presented hereimply that the system may not be in steady-state and thatthe net effect of competing processes could eitherincrease or decrease arsenic concentrations over the nextdecades.
Acknowledgements
We thank three anonymous reviewers for their veryhelpful suggestions. This work was funded by theNational Science Foundation (Grant No. EAR-0001098). [DR]
References
Aggarwal, P.K, Basu, A.R., Poreda, R.J., 2002. Isotope hydrology ofgroundwater in Bangladesh: implications for characterization andmitigation of arsenic in groundwater. IAEA-TC Project Report:BGD/8/016, IAEA, Vienna.
Aggarwal, P.K, Basu, A.R., Kulkarni, K.M., 2003. Comment on“Arsenic mobility and groundwater extraction in Bangladesh”.Science 300 (5619), 584.
Ali, M.A., 2003. In: Ahmed, M.F. (Ed.), Fate of Arsenic in theEnvironment, Arsenic Contamination: Bangladesh Perspective.ITN, Bangladesh.
25C.F. Harvey et al. / Chemical Geology xx (2006) xxx–xxx
ARTICLE IN PRESS
Allen, R.G., Pereira, L., Raes, S., Smith, D., 1998. Crop Evapotrans-piration- Guidelines for Computing Crop Water Requirements.FAO Irrigation and Drainage Paper, vol. 56. FAO-Food andAgriculture Organization of the United Nations, Rome. ISBN: 92-5-104219-5. M-56.
BADC, 2002. Survey Report on Irrigation Equipment and IrrigatedArea in Boro/2002 Season. Bangladesh Agricultural DevelopmentCorporation.
BADC, 2003. Survey Report on Irrigation Equipment and IrrigatedArea in Boro/2003 Season. Bangladesh Agricultural DevelopmentCorporation.
Bangladesh Water Development Board (BWDB), 2004. Surface waterand rainfall data for Bhagakul Meteorological Station, Munshiganj(personal communication).
Brammer, H., 1996. The Geography of the Soils of Bangladesh.University Press Limited, Dhaka, Bangladesh. ISBN 984 05 1328.
Brammer, H., Brinkman, R., 1977. Surface-water gley soils inBangladesh: environment, landforms and soil morphology.Geoderma 17 (2), 91–109.
Breit, G., Foster, A.L., Whitney, J.W., Uddin, N.Md., Yount, J.C.,Welch, A.H., et al., 2001. Variable arsenic residence in sedimentfrom eastern Bangladesh: clues to understanding arsenic cyclingin the Bengal delta. Abstract Presented at InternationalConference on Arsenic in Drinking Water, Hosted by ColumbiaUniversity Superfund Basic Research Program, ColumbiaUniversity, November, 2001.
Dixit, S., Herring, J.G., 2003. Comparison of arsenic(V) and arsenic(III) sorption onto iron oxide minerals: implications for arsenicmobility. Environ. Sci. Technol. 37 (18), 4182–4189.
Dowling, C.B., Poreda, R., Basu, A., Peters, S., Aggarwal, P., 2002.Geochemical study of arsenic release mechanisms in the BengalBasin groundwater. Water Resour. Res. 38 (9), 12.
Freeze, R.A., Cherry, J.A., 1979. Groundwater. Prentice-Hall, UpperSaddle River, NJ.
Freeze, R.A., Witherspoon, P.A., 1966. Theoretical analysis ofregional groundwater flow: 1. Analytical and numerical solutionsto the mathematical model. Water Resour. Res. 3, 623–634.
Harvey, C.F., Swartz, C.H., Badruzzman, A.B.M., Keon-Blute, N., Yu,W., Ali, M.A., et al., 2002. Arsenic mobility and groundwaterextraction in Bangladesh. Science 298 (5598), 1602–1606.
Harvey, C.F., Swartz, C., Badruzzman, A.B.M., Keon-Blute, N., Yu,W., Ali, M.A., et al., 2003. Response to comments on “arsenicmobility and groundwater extraction in Bangladesh. Science 300(5619), 584.
Hossain, M., Lewis, D., Bose, M., Chowdhury, A., 2003. RiceResearch: Technological Progress and Impacts on the Poor: TheBangladesh Case (Summary Report). International Food PolicyResearch Institute (IFPRI).
Islam, F.S., Gault, A., Boothman, C., Polya, D., Charnock, J.,Chatterjee, D., et al., 2004. Role of metal-reducing bacteria inarsenic release from Bengal delta sediments. Nature 430, 68–71.
Khan, F.H., 2002. Geology of Bangladesh. The University PressLimited, Dhaka, Bangladesh. ISBN: 984 05 1148 3.
Kinniburgh, D.G., Smedley, P.L., Davies, J., Milne, C.J., Gaus, I.,Trafford, J.M., et al., 2003. The scale and causes of thegroundwater arsenic problem in Bangladesh. In: Welch, A.H.,Stollenwerk, K.G. (Eds.), Arsenic in Ground Water: Geochem-istry and Occurrence. Kluwer Academic Publishers, Boston,MA, pp. 211–257.
McArthur, J.M., Banerjee, D.M., Hudson-Edwards, K.A., Mishra, R.,Purohitb, R., Ravenscroft, P., et al., 2004. Natural organic matter insedimentary basins and its relation to arsenic in anoxic groundwater: the example of West Bengal and its worldwide implications.Appl. Geochem. 19 (8), 1255–1293.
Milton, J.S., Arnold, J.C., 1995. Introduction to Probability andStatistics: Principles and Applications for Engineering and theComputing Sciences. McGraw-Hill, New York.
MoA, 2004. Handbook of Agricultural Statistics, Sector MonitoringUnit. Ministry of Agriculture.
Nickson, R., McArthur, J., Burgess, W., Ahmed, K.M., Ravenscroft,P., Rahman, M., 1998. Arsenic poisoning of Bangladeshgroundwater. Nature 395, 338.
Polizzotto, M.L., Harvey, C.F., Li, G., Newville, M., Fendorf, S., 2006-thisissue. Solid-phases and desorption processes of arsenicwithinBangladeshsediments. Chem. Geol. doi:10.1016/j.chemgeo.2005.11.026.
Rahman, A.A., Ravenscroft, P. (Eds.), 2003. Groundwater Resourcesand Development in Bangladesh. The University Press Ltd, Dhaka,Bangladesh. ISBN: 984 05 1643 4.
Shankar, B., Halls, A., Barr, J., 2005. The effects of surface waterabstraction for rice irrigation on floodplain fish production inBangladesh. Int. J. Water 3 (1), 61–83.
Swartz, C.H., Keon, N.E., Badruzzman, B., Ali, A., Brabander, D.,Jay, J., Islam, S., Hemond, H.F., Harvey, C.F., 2004. Subsurfacegeochemistry and arsenic mobility in Bangladesh. Geochim.Cosmochim. Acta 68 (22), 4539–4557.
van Geen, A., Zeng, Y., Versteeg, R., Stute, M., Horneman, A., Dhar,R., et al., 2003. Spatial variability of arsenic in 6000 tube wells in a25 km2 area of Bangladesh. Water Resour. Res. 39 (5) (Art. No.1140).
van Ranst, E., de Coninck, F., 2002. Evaluation of ferrolysis in soilformation. Eur. J. Soil Sci. 53, 513–551.
Water Resources Planning Organization (WARPO), 2000. NationalWater Management Plan Project: Draft Development Strategy.Ministry of Water Resources, Government of the People'sRepublic of Bangladesh.
Yu, W., 2003. Socio-Hydrologic approaches for managing groundwa-ter contamination problems: strategies for the arsenic problem inBangladesh. Doctoral thesis, Division of Engineering and AppliedSciences, Harvard University, Cambridge, MA, USA.
Yu, W.H., Harvey, C.M., Harvey, C.F., 2003. Arsenic in thegroundwater in Bangladesh: a geostatistical and epidemiologicalframework for evaluating health effects and potential remedies.Water Resour. Res. 39 (6) (Art. No. 1146).
