Groundwater is a major resource in Hawaii and is the principal source of water for municipal, agricultural, and industrial use. With a growing population, a long-term downward trend in rainfall, and the need for proper groundwater management, a better understanding of the hydroclimatological system is essential. Proxy records from corals can supplement long-term observational networks, offering an accessible source of hydrologic and climate information. To develop a qualitative proxy for historic groundwater discharge to coastal waters, a suite of rare earth elements and yttrium (REYs) were analyzed from coral cores collected along the south shore of Moloka'i, Hawaii. The coral REY to calcium (Ca) ratios were evaluated against hydrological parameters, yielding the strongest relationship to base flow. Dissolution of REYs from labradorite and olivine in the basaltic rock aquifers is likely the primary source of coastal ocean REYs. There was a statistically significant downward trend (−40%) in subannually resolved REY/Ca ratios over the last century. This is consistent with long-term records of stream discharge from Moloka'i, which imply a downward trend in base flow since 1913. A decrease in base flow is observed statewide, consistent with the long-term downward trend in annual rainfall over much of the state. With greater demands on freshwater resources, it is appropriate for withdrawal scenarios to consider long-term trends and short-term climate variability. It is possible that coral paleohydrological records can be used to conduct model-data comparisons in groundwater flow models used to simulate changes in groundwater level and coastal discharge.
 Approximately 99% of Hawaii's domestic drinking water and about 50% of all freshwater used in the state is from groundwater [Oki et al., 1999]. Groundwater also plays a vital role in structuring coastal habitats and ecosystems, such as fishponds, anchialine pools, and coral reefs. For example, fresh groundwater is thought to be the main source of nutrients (i.e., nitrate, phosphate, and silica) to the coastal waters off leeward Hawai'i, Maui and Moloka'i [Knee et al., 2008; Street et al., 2008]. Thus, groundwater supports biological and chemical processes and is also an essential source of freshwater.
 Groundwater recharge can be defined as direct infiltration of rainfall, fog drip and irrigation water that is not lost to runoff, evapotranspiration, or stored in soil. Long-term downward trends in annual rainfall over much of the state may eventually impact future groundwater resources. Increasing expected groundwater withdrawals from urban development would exacerbate these impacts. Such trends are particularly relevant on the island of Moloka'i, where there is a heightened demand for groundwater from a population that has increased by more than 40% from 1970 to 2000 [State of Hawaii, 2000].
 In order to effectively manage groundwater resources under increased demands from rising population and declining rainfalls, a better understanding of the hydroclimatological system is essential. This includes identifying natural and anthropogenic variability. The U.S. Geological Survey (USGS) maintains a network of stream-gaging stations in Hawaii, operating a total of about 500 stations since 1909, although only about 50 are active currently. Given certain criteria (i.e., drainage basins upstream of the stations should be free of artificial changes), Fontaine  identified 16 stream-gaging stations useful for analysis of long-term trends. Based on these records, Oki  found a statewide downward trend in base flow from 1913 to 2000.
 To extend the stream gage data both spatially and temporally, coral geochemistry records have the potential to offer an accessible source of hydrologic and climate information in locations where instrumental records are sparse or not available. Scleractinian corals may preserve useful records of historical climate data for several reasons: the coral skeletons are annually banded, yielding rapid growth rates (typically 10–20 mm yr−1 for massive Porites spp.), they can grow continuously for decades to centuries, and are sensitive to changes in environmental conditions, such as sea surface temperature, salinity, sedimentation, and pollution. The utility of coral proxies is largely dependent on the fact that as corals grow they secrete a calcareous skeleton into which trace elements are partitioned from the ambient seawater, thereby reflecting changes in the environment in which the coral was growing. As such, high-resolution coral records have been shown to accurately record changes in environmental parameters such as precipitation, sea surface temperatures (SST), sediment flux, salinity and pollution [Beck et al., 1992; Dunbar et al., 1994; Le Bec et al., 2000; Fallon et al., 2002; McCulloch et al., 2003]. The most commonly used climate proxies in coral skeletons are oxygen and carbon stable isotope ratios, and trace and minor elemental composition in the coral skeleton. In particular, oxygen isotopic (δ18O) and Sr/Ca variations have proven to be robust tracers of SST and hydrological process (i.e., precipitation, evaporation, freshwater input) [Dunbar et al., 1996; Gagan et al., 1998; Quinn et al., 1998; Hendy et al., 2002].
 Dissolved concentrations of rare earths and yttrium (REYs) have been used as tracers of groundwater flow paths in the hydrogeological environment [Banner et al., 1989; Johannesson et al., 1997, 1999]. In basaltic rock aquifers, REE concentrations have been shown to be relatively constant along groundwater paths, and are believed to be in steady state with host rocks [Tweed et al., 2006]. For example, work from a basaltic rock aquifer in southeast Australia showed that consistent REE patterns and concentrations at different stages along flow paths suggest that REE steady state is achieved relatively quickly, perhaps during infiltration of slightly acidic rainfall [Tweed et al., 2006]. Within a basalt aquifer, the REYs are mainly derived from dissolution of labradorite and olivine in basalts [Tweed et al., 2005].
 In general, it is believed that the REY “dissolved load” is in the colloidal fraction (<0.22 μm), as documented by significantly lower REE concentrations with progressively smaller filtration methods [Elderfield et al., 1990; Sholkovitz, 1990; Akagi et al., 2004]. Upon contact with seawater, the REYs are desorbed (released from the colloidal form) under low salinity values (typically <10 ppt) as demonstrated with a seawater leaching experiment [Sholkovitz, 1990]. In a similar manner to desorption of particulate barium (Ba) [Coffey et al., 1997], seawater-induced desorption of REYs is believed to be the primary mechanism by which seawater concentrations are enriched. As such, submarine groundwater discharge has been shown to be an important source of REEs to coastal waters [Duncan and Shaw, 2003].
 Incorporation of trace elements into the coral aragonite is commonly described in terms of distribution coefficients (D) between coral aragonite and seawater where D is the ratio of rare earth elements (REE) to Ca in coral to seawater. The DREE has been investigated previously with various coral species with DREE values ranging from 1 to 2 in Porites corals [Wyndham et al., 2004; Akagi et al., 2004], suggesting that trivalent REEs are incorporated into the coral lattice in close proportion to their seawater concentration with minimal fractionation of the series during incorporation into coral carbonate [Wyndham et al., 2004]. Given the similar chemical behavior of Y to the lanthanide group [Stevenson and Nervik, 1961], the distribution coefficient for Y is assumed to be within the same range. Substitution in the coral lattice is augmented by the fact that the ionic radii of Y3+, Ce3+ and La3+ in ninefold coordination are similar in size to Ca2+ [Shannon, 1976]. Due to these characteristics, coral REY patterns have been used as tracers of coastal river discharge [Fallon et al., 2002; Wyndham et al., 2004; Lewis et al., 2007; Jupiter, 2008]. While accurate DREY values for Porites corals collected off the south shore of Moloka'i remain to be determined such that absolute concentrations can be reported, previously reported distribution coefficients in the literature [e.g., Wyndham et al., 2004] are sufficient to evaluate relative patterns in coral elemental ratios.
 Coral paleorecords may capture hydrological conditions that are not evident in the instrumental record as they have the potential to integrate the total freshwater discharge to the inner fringing reef over several decades. In a novel approach to develop a proxy for historic groundwater discharge to coastal waters, a suite of trace elements in coral cores, including the REYs were analyzed from sites along the south shore of Moloka'i (Figure 1). We hypothesize that high-resolution (subannual) REY records from corals are responsive to variations in groundwater input to the south shore of Moloka'i during the 20th century, which may in turn reflect larger-scale warming and drying in the region.
2. Study Area
2.1. Regional Setting
 Moloka'i is the fifth largest of the Hawaiian Islands with an area of 676 km2 [Juvik and Juvik, 1998]. The island is mainly composed of two shield volcanoes [Stearns and Macdonald, 1947], the older West Moloka'i Volcano rises to an altitude of 421 m and the younger East Moloka'i Volcano rises to an altitude of 1512 m. Moloka'i is formed primarily by the shield-stage (tholeiitic basalt and olivine tholeiitic) and postshield-stage alkali (plagioclase, labrodite-rich) basalt lavas of the older West Moloka'i Volcano and the younger East Moloka'i Volcano [Langenheim and Clague, 1987]. The fringing coral reef extends for a distance of about 40 km from Hale O Lono on the west to Kamalō on the east, as described in detail by Brown et al. .
 Geographically the coral study sites can be divided into those east of the Kaunakaki Wharf (One Ali'i and Kamalō) and those to the west ('Umpipa'a and Pālā'au) (Figure 1). According to a sediment trap study, sites in the east near Kamiloloa, are dominated by terrestrial material (91%–93%), whereas to the west near Pālā'au reworking of carbonate material dominated the sediment trap [Bothner et al., 2006]. In the east, the Kawela Gulch is a major pathway of water and sediment to the south coast of Moloka'i [Field et al., 2008]. Once deposited onto the reef, the primary transport of land-derived material is to the west, in the direction of the prevailing trade winds, with a secondary direction of slightly offshore, toward a zone of low coral abundance seaward of One Ali'i [Ogston et al., 2004; Presto et al., 2006]. Historically, the region near Pālā'au was subject to heavy flooding and runoff in the 1900s. As a result, an extensive mangrove forest was established in 1903 to curb the heavy sediment runoff [D'Iorio, 2008].
2.2. Oceanographic Setting
 The reef flat along the south coast of Moloka'i forms a shallow (<1.5 m) surface between the shoreline and the reef crest 0.1–1.9 km offshore. The reef flat experiences mixed semidiurnal tides with a mean range of 0.6 m and a maximum spring range of 0.9 m [Ogston et al., 2004; Storlazzi et al., 2004]. Strong trade winds (>10 m s−1) occur throughout most of the year to produce longshore current to the west and slightly offshore [Storlazzi et al., 2004; Presto et al., 2006]. During the trade wind period (April–November), sediment is resuspended on the reef flat on a daily basis [Ogston et al., 2004], reducing light availability for photosynthesis during afternoon turbidity events [Piniak and Storlazzi, 2008]. Salinity and SSTs values range between 26 and 34 psu and 20–29°C [Presto et al., 2006], with variations attributed to freshwater input, season, currents and storms. Offshore (1 m) nutrient values for select sites along the south shore of Moloka'i range between 37 and 129 mmol L−1 for soluble reactive silica, 0.60–7.00 mmol L−1 for dissolved inorganic nitrate, and 0.55–1.29 mmol L−1 for soluble reactive phosphorous [Street et al., 2008].
 Mild temperatures and persistent, cool trade winds characterize the climate in Hawaii, which has a rainy season from October through April, and a dry summer season from May through September [Blumenstock and Price, 1967; Sanderson, 1993]. Interannual and interdecadal rainfall variation is also influenced by well-known ocean-atmosphere systems such as the El Niño–Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO) [Lyons, 1982; Mantua et al., 1997; Chu and Chen, 2005]. For example, analysis of seasonal rainfall anomalies in Hawaii showed that a rainfall deficit occurs in winter and spring of the year following an El Niño, when there is a reduction in easterly trade winds (as a result of warmer waters and a decrease in atmospheric pressure) [Chu, 1989]. Likewise, Mantua et al.  found a negative correlation between the PDO and Hawaii winter rainfall, with the PDO index defined as the leading principal component of North Pacific monthly SST anomalies poleward of 20°N. The mechanism linking rainfall to large-scale ocean-atmosphere systems is as follows: because Hawaii is located at the centers of the ENSO and PDO, the sinking portion of the Hadley cell inhibits wintertime rainfall during an El Niño event. Furthermore, Chu and Chen  note that during these phases, there is an anomalous zonal circulation cell that is well established over the subtropical North Pacific with a pronounced descending branch over Hawaii, creating unfavorable conditions for convection and rainfall.
 Rainfall distribution in Hawaii is also strongly influenced by an orographic effect when moisture-laden air masses rise and cool, and the moisture condenses on the hillslopes. On Moloka'i, the East Moloka'i Volcano causes orographic precipitation along the crest of the East Moloka'i Volcano and on the windward (north) side of east Moloka'i. Annual rainfall in the upper slopes of the East Moloka'i Volcano can exceed 4 m yr−1. In contrast, rainfall toward the southern coast of West Moloka'i decreases drastically to less than 0.4 m yr−1 [Giambelluca et al., 2008; Tribble and Oki, 2008]. As a result, the greatest groundwater recharge is focused near the topographic peak of East Moloka'i Volcano, estimated at more than 1 m yr−1 [Shade, 1997] (Figure 1).
2.4. Hydrogeology and Groundwater
 Assuming a daily average rainfall on Moloka'i of approximately 2 million m3, Tribble and Oki  estimated that 50% evaporates or is transpired by plants, 16% runs off to streams and the remaining 34% infiltrates into the ground. Streams on the north side of the island are perennial and are sustained by base flow, while the remaining streams are ephemeral with exceptions being the Kamalō and Kawela streams on East Moloka'i. Fresh groundwater on Moloka'i exists as either a freshwater lens floating on denser, underlying saltwater within permeable, dike-free lava flows or as dike-impounded water ten to hundreds of feet above sea level [Oki, 2006]. Groundwater recharge rates were estimated at 8.26 m−3 s−1 assuming a natural vegetative cover [Shade, 1997]. Under natural conditions, average recharge rates to an aquifer must balance the amount of discharge to streams, springs or the ocean. However, human demand for groundwater has led to increased withdrawal rates along the coast, which can lead to decreases in groundwater discharge to streams and the ocean. For example, from 2000 to 2002, 16,000 m3 were pumped per day for domestic use on the island [Tribble and Oki, 2008]. In a numerical simulation of the hydrologic effects of additional groundwater withdrawals on Moloka'i, Oki  showed that groundwater levels rise and coastal discharge increases near sites of reduced withdrawal, whereas groundwater levels decline and coastal discharge decreases near sites of increased withdrawal.
 The island of Moloka'i is divided into West, Central and Northeast Aquifer Sectors with relative differences in percentage distribution of the rainwater into runoff, evaporation and transpiration, and groundwater recharge [Tribble and Oki, 2008]. As mentioned earlier, the distribution of groundwater recharge is similar to the distribution of annual rainfall, with the greatest recharge near the summit of the East Moloka'i Volcano. Along the south shore of Moloka'i, groundwater discharge is also greater in the east, with freshwater discharge estimated to be about 9400 m3 d−1 km−1 between Kawela and Pūko'o. Groundwater discharge between Kawela Gulch and Kaunakakai however may be lower than expected because of the confining effect of the fringe of sediments, as suggested by elevated hydraulic heads from test wells [Stearns and Macdonald, 1947]. In contrast to the Northeast Aquifer Sector, drier areas in the west tend to have a lower percent of runoff and recharge, and a higher percentage of rainfall that evaporates and transpires [Shade, 1997]. As such, groundwater discharge in the west, between Hale O Lono and Kūkukū is estimated at 900 m3 d−1 km−1, only one tenth as much as in the east [Tribble and Oki, 2008].
2.5. Submarine Groundwater Discharge
 Groundwater that enters the ocean below sea level has been termed submarine groundwater discharge (SGD) and is a mixture of fresh groundwater and recirculated seawater [Moore, 1996]. Coastal groundwater discharge in Hawaii has been investigated using airborne thermal infrared (thermal IR) imaging systems [Johnson et al., 2008; Peterson et al., 2009], CTDs (submersible instruments to measure conductivity, temperature and depth), and multiple geochemical tracers such as dissolved silica, and naturally occurring U/Th isotope series [Moore, 1996; Knee et al., 2008; Street et al., 2008].
 Using uncalibrated thermal IR images, Grossman et al.  observed several plumes of cool waters across the reef flat offshore from Kapuaiwa and along the Kamiloloa coast (see Figure 1 for locations). In the absence of surface runoff from streams, the authors suggest that these cool waters are best explained by groundwater discharge along the shore. Likewise, low salinities and high dissolved silica concentrations provide unambiguous evidence of meteoric groundwater input to coastal sites along the leeward side of Moloka'i [Street et al., 2008]. The role of SGD as an additional nutrient source to the nearshore and fringing reef has also been observed on Kona, Hawaii [Knee et al., 2008]. Submarine groundwater discharge may also form distinctive dissolution pits in fringing coral reefs, referred to as “blue holes” seen in combined aerial photography and LIDAR images from Moloka'i [Brown et al., 2008].
3. Material and Methods
 Coral cores were collected in November 2006 and 2007 from 4 sites along the fringing reef of the south shore of Moloka'i from living scleractinian Porites lobata (Figure 1); P. lobata is the predominant coral genus in Hawaii and forms massive colonies [Grigg and Dollar, 1980; Jokiel et al., 2008]. All cores were collected using SCUBA and a pneumatic, handheld underwater drill attached to a stainless steel coring barrel. The cores were drilled parallel to the growth axis (perpendicular to the coral surface). This is confirmed by X-radiographs that reveal annual growth bands perpendicular to the core axis. The stainless steel core barrel yielded cores with a diameter of slightly less than 5 cm and ranging in length from 17 cm to 40 cm with a total length of up to 92 cm. Cores were collected from corals in water depths of less than 1 m to 6.4 m (Table 1). Drill holes in coral colonies were sealed with concrete plugs. Within 2–3 h after collection, cores were measured, and rinsed with freshwater (using a Waterpik) to remove living tissue from the outer skeleton. Cores were slabbed (∼7 mm thick) and cleaned untrasonically in 18 MΩ·cm water.
Table 1. Coral Core Site Information for the Four Coral Records Discussed in Section 3
Distance to Shore (m)
14 Nov 2006
15 Nov 2006
15 Nov 2006
30 Nov 2007
3.1. Coral Trace Metal Analysis
 Trace element concentrations (11B, 43Ca, 84Sr, 89Y, 138Ba, 139La, 140Ce, and 238U, plus 146Nd and 174Yb for Pālā'au and Kamalō corals, and 57Fe and 27Al for 'Umpipa'a and One Ali'i corals) were measured by laser ablation inductively coupled mass spectrometry (LA-ICP-MS) with an argon-fluoride (ArF) excimer laser (193 nm wavelength) attached to a Varian 820 ICP-MS at the Australian National University. Details for the laser system, data collection, and protocol for drift correction are further discussed by Sinclair et al. , Fallon et al. , and Wyndham et al. [2004. This approach allows for highly resolved microsampling and precise concentration measurements of ∼40 trace and rare earth elements [Sinclair et al., 1998]. To account for variations in coral surface architecture and density, all concentrations were normalized to 43Ca [Fallon et al., 1999]. The coral cores were subsectioned into 95 × 25 mm pieces using a diamond blade saw to be mounted on an X-Y sample stage and placed in a sealed Perspex chamber (under helium atmosphere) for analysis of tracks selected along the major growth axis. The surface of the analytical track was subject to a cleaning step by masking the beam (3 cm × 1 cm) with a rectangular aperture (40 μm × 500 μm) and ablating approximately 5–10 μm of the sample surface, exposing clean, fresh aragonite for LA-ICP-MS analysis [Wyndham et al., 2004]. During sample ablation, the laser spot size was decreased to 40 μm × 400 μm and scanned along the coral growth axis at 40 μm s−1. Data were smoothed using a 10-point running median to remove outliers followed by a 10-point running mean to reduce data volume [Jupiter et al., 2008] (see auxiliary material). Adjacent laser scans, approximately 750 μm apart were also performed to evaluate signal replication. Based on multiple laser scans, the standard errors for La/Ca, Y/Ca, and Ce/Ca are ±8.14 × 10−6, ±2.83 × 10−5, and ±1.64 × 10−5 mmol mol−1, respectively. The adjacent scans reproduce the low-frequency signal (Figure 2).
 While the fractionation history, particularly between the heavy and light REEs (LREEs), is widely used to infer fundamental geochemical process in natural systems [Hanson, 1980; Henderson, 1984; Banfield and Eggleton, 1989; Elderfield et al., 1990], it is outside the scope of this study since we only analyzed Ce, La and Y, plus ytterbium (Yb) and neodymium (Nd) in 2 of the 4 coral cores in order to reduce total mass scan time. A U.S. Geological Survey (USGS) program to understand the dynamics of sediment transport and its impact on coral reef health [Brown et al., 2008] initiated the study presented here. Therefore, the geochemical analysis was focused on the LREEs since previous work has shown LREE enrichment during flood events [Wyndham et al., 2004; Jupiter, 2008].
3.2. Coral Chronology
 Positive X-radiographs were taken to reveal density bands and aid in chronology development. A pair of high- and low-density layers characterizes the annual growth band couplets (Figure 3) with presumably minimum density layers forming during minimum winter SSTs and maximum density layers forming during summer SSTs [Highsmith, 1979]. This assumption was tested by analyzing stable oxygen (δ18O) isotopes at 1 mm increments for the upper 218 mm of the 'Umpipa'a core as a proxy of SST variability [Weber and Woodhead, 1972; Dunbar and Wellington, 1981] (Figure 3a). Minimum and maximum coral δ18O values within a density couplet were assigned a corresponding minimum and maximum instrumental SST and calendar date based on 2x2 gridded weekly SST records [Reynolds et al., 2002]. Monthly coral δ18O values correlate to monthly SSTs (r = −0.67; p = 0.05) during the calibration period between 1991 and 2006 (Figure 3c). Based upon the chronology, an average extension (growth) rate of 8.37 to 11.87 mm yr−1 was determined (Table 2). The initial chronology based on the X-radiographs was further refined by matching maximum strontium (Sr) and uranium (U) to Ca ratios to instrumental SSTs, according to paleothermometer studies that indicate an inverse relationship between SST and U/Ca [Shen and Dunbar, 1995; Min et al., 1995], and Sr/Ca [e.g., Smith et al., 1979; Beck et al., 1992; Gagan et al., 1998]. According to the calibration period, maximum Sr/Ca and U/Ca ratios were assigned to the coldest (February) month for the remaining length of each coral record. The chronology was linearly interpolated between the SST markers. Uneven annual and seasonal growth rates result in unequal numbers of sample points. Therefore, using Arand TIMER software [Howell et al., 2006], data were linearly interpolated to 12 points per year (approximately yielding an effective monthly sample resolution) for statistical analyses. Given dating uncertainties derived from the coral X-radiographs and geochemical seasonal signals, replicate age dating trials yields an absolution chronology error of approximately ±12 months.
Table 2. Calculated Growth Rates of Coral Skeletal Record Based on the Length of Coral Core and Number of Annual Density Band Couplets Based on Coral Skeletal X-Radiographsa
Number of Density Band Couplets
Growth Rate (mm yr−1)
Band couplets are composed of a high- and low-density band. Coral cores were collected alive in 2006 and 2007.
3.3. Statistical Analyses
 Empirical orthogonal function (EOF) analysis in MATLAB was used to identify common patterns of variability in the coral REYs ratios (Ce/Ca, La/Ca, Y/Ca), SST proxies (Sr/Ca and U/Ca), and soil proxies (Fe/Ca and Al/Ca) where Fe and Al are major elements in Moloka'i's basaltic red soil (i.e., weathering of kaolinite and smectite clay minerals). The nonparametric Mann-Kendall test [Helsel and Hirsch, 2002], was used to evaluate long-term trends in both coral REY and base flow data from the longest, continuous USGS stream-gaging station on Moloka'i (Hālawa Stream, 164000000). Daily discharge was measured from 1913 to 2002 (with a hiatus between 1932 and 1938) and used to determine monthly mean and annual mean flow values for both total streamflow and estimated base flow [Oki, 2004]. Slopes of statistically significant long-term trends were estimated using Sen's method at p = 0.05 [Gilbert, 1987]. In order to identify periodicities in the coral REY data that could be related to a specific climate forcing, spectral analysis was applied using the Singular Spectrum Analysis–Multi Taper Method (SSA-MTM) Toolkit [Vautard et al., 1992; Dettinger et al., 1995]. Analysis of long-term trends from the coral records was confined to the 'Umpipa'a core, which was the only record (85 years) long enough to avoid biases of unrepresentative catchment behavior, such as those during storms, hurricanes or droughts [Feng et al., 2004].
 Covarying behavior between La/Ca, Y/Ca, and Ce/Ca, can be seen in Figure 4, and defines the first EOF from the four coral sites (average loading of 70%) (Table 3). The only other elements that showed joint behavior were Fe/Ca and Al/Ca, and Sr/Ca and U/Ca, with an average loading on the first EOF of 64% and 77%, respectively. There was no coherence between REY/Ca and Ba/Ca ratios. Neither Ba/Ca nor any of the ratios of REY/Ca were significantly related to Moloka'i instrumental records of streamflow or precipitation. The strongest relation between calculated base flow and 'Umpipa'a coral Y/Ca record was during the fall and winter rainy seasons (r = 0.23; p = 0.05).
Table 3. Percent of Common Variance Defined as the Loading in the 1st EOFa
EOF analysis was performed on the REYs ratios (Ce/Ca, La/Ca, and Y/C), sea surface temperature (SST), proxies (Sr/Ca and U/Ca), and soil proxies (Fe/Ca and Al/Ca). Fe and Al were not measured from the Kamalō and Pālā'au cores, as indicated by na.
 Given the commonality observed in the REYs and the fact that the Y/Ca ratios are almost an order of magnitude higher than La/Ca and Ce/Ca, the Y/Ca ratios are herein identified as representative of REY behavior and the following discussion is focused on coral Y/Ca variation. The most striking feature of the Y/Ca records in Figure 1 is the long-term downward trend observed at all the sites except One Ali'i (Figure 1). Measured Y/Ca records from Kamalō, 'Umpipa'a and Pālā'au have statistically significant negative trends at the 95% confidence level, while the Y/Ca trend at One Ali'i had a statistically significant positive trend (Figure 1). According to slope estimations, the decrease in Y/Ca over the various lengths of coral records was between −39% and −63%, with an increase of 17% from the One Ali'i coral (Table 4). Over the length of each coral record, the ΔY/Ca was an order of magnitude greater than the limit of detection. The 'Umpipa'a Y/Ca record also displays an accelerated decrease beginning in 1980 (Figure 1). The slope decreases by an order of magnitude from 1980 to 2006, and the % change is more than 10% (from −39% to −53%) (Table 4). These results are consistent with those calculated from the shortest coral record from Pālā'au, where the magnitude of change (−53%) is similar to 'Umpipa'a record the over the same time period (1980–2006). This change may be a function of elevated Y/Ca values centered in the early 1980s as well as relatively low Y/Ca values in the most recent decade.
Table 4. Percent Change and Difference of Y/Ca of the Length of Each Coral Record Calculated Using the Linear Function of Sen's Method of Slope Estimationa
% Change (Length of Record)
% Change From 1980 to 2006
Negative values are given in bold. Values were also calculated for the 'Umpipa'a core from 1980 to 2006 for analysis during the most recent decades. ΔY/Ca, difference of Y/Ca.
 The MTM analyses yielded statistically significant (99% confidence interval) frequencies of 0.023, 0.10, and 0.19 per year in the 85 year 'Umpipa'a Y/Ca record, characteristic of interdecadal and interannual variation, accounting for 37% of the variation (Figure 5a). No annual periodicity was detected at the 95% confidence level. The MTM spectrum of the PDO index yields statistically significant (99% confidence interval) peaks at similar frequencies (0.018, 0.10, and 0.16 per year), accounting for 45% of the PDO variation (Figure 5b). The reconstructed time series are inversely correlated with a correlation coefficient of −0.37 (p = 0.05) for the period 1920–2006 and slightly stronger after 1938 (r = −0.48; p = 0.05), which excludes the distinct drop in coral Y/Ca centered at 1935. The calculated base flow time series (not shown) is dominated by higher frequencies (around 3 per year) associated with seasonal variation.
 The most salient feature of the Y/Ca time series is the statistically significant downward trend at 3 of the sites, with the longest record suggesting an almost −40% reduction since 1920 (Figures 1 and 6and Table 4). The positive trend detected at One Ali'i is believed to be in response to surface water runoff from the Kawela Gulch, a major pathway of water and sediment to the south coast of Moloka'i [Fletcher et al., 2002; Brown et al., 2008]. While surface runoff from the Kamalō Gulch is also a potential source, it is likely that the fine-grained sediment plumes formed during floods are either discharged directly into deep water through one of two paleochannels, deposited on the well developed alluvial fan, and/or driven west along the shoreline by prevailing winds [Field et al., 2008]. The latter is supported by elevated coral Ba/Ca values from One Ali'i relative to Kamalō, with desorption from terrigenous muds being the primary source of barium to the nearshore coral reefs [McCulloch et al., 2003; Fleitmann et al., 2007]. In fact, sections of the reef around Kamalō contain some of the most densely packed coral communities in Hawaii [Jokiel and Brown, 2004; Brown et al., 2008]. Therefore stream runoff, rather than groundwater input, from the Kawela Gulch is believed to be the dominant source of terrestrially derived material to the adjacent and downstream coral reefs. Kawela stream gage data suggests an annual sediment yield of approximately 2530 t [Tribble and Oki, 2008]. Observed elevated coral Fe/Ca ratios at One Ali'i relative to 'Umpipa'a also supports the notion that the former is dominated by sediment laden surface water rather than groundwater input.
 Previous investigations also link REY variability to terrestrial runoff [e.g., Fallon et al., 2002; Lewis et al., 2007; Jupiter, 2008], as well as biological activity [Wyndham et al., 2004]. In these studies, there was a strong relationship, frequently at the seasonal scale, between Ba/Ca and Mn/Ca to the REY/Ca ratios. These alternative scenarios were investigated by the examining the seasonal signal and covariance between Ba, Mn and the REYs in the Moloka'i records. In contrast to previous work, there is neither a seasonal signal in the coral REYs nor is there covariance with Mn/Ca variability, as would be expected if REY variability was primarily responding to biological activity. Given the lack of covariance between Ba/Ca and the REYs, as well as an absence of surface runoff from streams in the west (near Pala'au and 'Umpipa'a), river discharge is not considered a primary control on REY variability in the Moloka'i records. Instead, the utility of REEs as tracers of groundwater flow paths [e.g., Tweed et al., 2005], suggests that dissolution from labradorite and olivine in the basaltic rock aquifers of Moloka'i is the primary source of REYs, and that variability in the coral record is driven by changes in submarine groundwater discharge.
 The coral REYs records from Moloka'i are therefore interpreted as being responsive to changes in groundwater discharge associated with a decrease in base flow since 1913. The similarity over the long term between the base flow and Umpipa’a coral records is illustrated in Figure 6, with both records revealing a reduction between −39% and −46%. Of additional note is the fact that the 'Umpipa'a coral was collected near a “blue hole,” signifying the proximity of potential groundwater discharge [Field et al., 2008]. On a seasonal scale, the strongest relationship between calculated base flow and coral REY/Ca is during the fall and winter rainy seasons, presumably when base flow and submarine groundwater discharge are most likely to be the greatest. As mentioned earlier, there was no statically significant relationship between coral REYs variability to Moloka'i instrumental records of stream discharge or precipitation. Oki  suggests that changes in base flow occur at much longer time scales relative to rainfall, whereas changes in direct runoff are controlled by rainfall. Thus, the time scales over which these changes occur reflect their controlling factors.
 The statistically significant downward trends at the remaining sites are consistent with data from Oki , who found a statistically significant downward trend in all measured Hālawa base flow percentiles, ranging from annual, to 10th and 90th quartile base flows, representing a −46% reduction in average base flow over the length of record (Figure 6b). Since the Hālawa gaging station data is free from artificial change (i.e., groundwater withdrawal or upstream surface water diversion), the long-term trend in base flow is believed to represent a response to a reduction in annual rainfall over much of the state since 1913 rather than an increase in withdrawal [Oki, 2004]. Presumably the coral REYs trends also reflect the long-term downward trend in annual rainfall by means of changes in groundwater storage and recharge. A primary driver of interannual and interdecadal rainfall variation in Hawaii is the PDO [Mantua et al., 1997], with the dominant PDO modes at 15–25 and 50–70 years [Minobe, 1999]. As discussed above, spectral analysis results reveal interannual and interdecadal variability in the 'Umpipa'a Y/Ca coral record that may be sensitive to SSTs changes centered in the extratropical North Pacific.
 A comparison between the PDO index [Mantua et al., 1997] and the coral Y/Ca record reveals a divergence beginning in the 1980s (Figure 7a). Beginning in the late 1980s the PDO index trends toward negative values (colder/wetter), but as seen in Figure 7a, the coral record continues its trend toward lower Y/Ca ratios (interpreted to represent warmer/drier conditions). The calculated base flow also diverges from the PDO in the late 1980s, but the divergence is not as striking (Figure 7b). The divergence between the 'Umpipa'a coral Y/Ca record and the PDO index could also reflect the impact of increased withdrawal in the Kualapuu area. Since the majority of groundwater withdrawal is from the Kualapuu area, it is reasonable to assume that coastal discharge near 'Umpipa'a is particularly sensitive to increased withdrawals [Oki, 2006]. Using a two-dimensional, steady state groundwater flow model to determine the effects of proposed groundwater withdrawals on water levels and coastal discharge, Oki  found that coastal discharge is reduced by an amount equal to the additional withdrawal. While changes in sea level may mask a lowering of groundwater in well data, chloride concentrations in some wells have increased in recent years and may be related to increased withdrawal rates (D. Oki, personal communication, 2009).
 In concert with increased withdrawal rates, a reduction in groundwater discharge may also reflect a trend in Hawaii toward drier and warmer conditions. There is empirical evidence in Hawaii that both air temperatures and SSTs have increased over the past several decades [Jokiel and Brown, 2004]. Hawaiian rainfall has also been low since the mid-1970s [Chu and Chen, 2005], despite a recent trend toward a positive PDO phase (colder/wetter). Divergence in Hawaiian air temperatures from the PDO signal in recent decades also suggests a potentially greater influence of global warming compared to large-scale modes of climate variability [Giambelluca et al., 2008]. Cao et al.  suggest that warmer and drier conditions may also be the result of increasing Trade Wind Inversion (TWI) over Hawaii since the late 1970s, which is conducive to increased atmospheric stability and reduction in cloud development [Cao et al., 2007]. A climate shift centered in the mid-1970s in not unique to Hawaii. The “1976 Pacific climate shift” has been characterized as a warming in SSTs through much of the eastern tropics [Graham, 1994; Trenberth and Hurrell, 1994], with anthropogenic forcing partly responsible for the observed warming [Knutson and Manabe, 1998]. At this point it remains difficult to determine the primary driver for a reduction in base flow, given uncertainties associated with groundwater flow paths and residence times on Moloka'i, as well as changes associated with groundwater withdrawal. Nevertheless, there is potential for the coral paleohydrological proxy to integrate both local factors (i.e., groundwater pumping) and potentially climatically driven changes to the hydrologic setting.
 Long-term downward trends observed statewide in base flow of streams correspond to downward trends in rainfall and may reflect a decline in groundwater storage and recharge in Hawaii [Oki, 2004]. According to these observations, Hawaii's climate is likely to become drier as warming continues, leading to further reductions in groundwater recharge and groundwater discharge to streams. These shifting patterns in the hydroclimatological system will undoubtedly cause changes in land surface interactions and ecological systems [Benning et al., 2002; Atkinson and LaPointe, 2009]. In a novel approach to develop a proxy for historic groundwater discharge to coastal waters, results from this study show that coral REY records from the south shore of Moloka'i are a promising paleohydrological indicator. There is a statistically significant downward trend (−40%) in subannually resolved REY/Ca ratios over the last century that is consistent with a long-term downward trend in base flow and a statewide downward trend in annual rainfall over much of the state. Most recently, increased groundwater withdrawals rates, in conjunction with evidence of a warming trend, may be driving an accelerated decrease in coral Y/Ca ratios. With greater demands on freshwater resources, it is appropriate that proposed withdrawal scenarios take into account long-term trends and short-term climate variability. Therefore, it is possible that coral paleohydrological records can be used to conduct model-data comparisons in groundwater flow models used to simulate changes in groundwater level and coastal discharge under different pumping scenarios.
 A U.S. Geological Survey Mendenhall Postdoctoral Research Fellowship to N.G.P. and USGS internal funds to M.E.F. funded this study. The authors thank Josh Logan, Tom Reiss, and Curt Storlazzi for field support and Joe Reich, the captain of the R/V Alyce C., as well as Delwyn Oki and Gordon Tribble for helpful discussion and providing supporting data. This manuscript was improved by comments by Renee Takesue, Peter Swarzenski, and two anonymous reviewers.