The circulation in the tropical South Pacific generally follows a SE to NW direction at all depths (Figure 1B) and represents the northern limb of the wind-driven geostrophic anticyclonic gyre that affects the entire water column of the South Pacific (Reid, 1997).

The surface hydrography of the westernmost equatorial Pacific is dominated by the West Pacific Warm Pool (WPWP) (Figure 1B), the largest oceanic warm-pool on Earth that plays a key role in the heat transfer between ocean and atmosphere in the tropics and in particular for the dynamics of the El Niño Southern Oscillation (ENSO) (Hu et al., 2015). The WPWP is connected to the central Pacific circulation system via the Equatorial Undercurrent (EUC), which crosses the entire equatorial Pacific from West to East and acts as the counterpart of the equatorial westward surface currents that push warm waters toward the WPWP (Tsuchiya et al., 1989; Tomczak and Godfrey, 2003). Another important feature of the surface waters of the tropical central Pacific is the High Surface Evaporation Area (HSEA) centered at ~15°S/130°W (see Figure 1B and Figure S1), where South Pacific Equatorial Water (SPEW) forms as a consequence of strong evaporation. This water mass is subsequently subducted and carries its high salinity and high oxygen signal to the NW into the WPWP (Tomczak and Hao, 1989; Tsuchiya et al., 1989; Qu and Lindstrom, 2002; Tomczak and Godfrey, 2003). Antarctic Intermediate Water (AAIW) is the major intermediate water mass prevailing between 600 and 1,300 m in the entire South Pacific (Hartin et al., 2011) and is characterized by low salinities (34.3–34.5), low temperatures (3.5–10°C), and high oxygen content (4.5–5.5 ml/l) (Bostock et al., 2010). AAIW forms mainly in the southeastern Pacific as a consequence of cooling of Subantarctic Mode Water (SAMW) (Sloyan and Rintoul, 2001) and flows in a NW direction following the subtropical gyre circulation (Tomczak and Godfrey, 1994) (Figure 1B). Equatorial Pacific Intermediate Water (EqPIW) (Bostock et al., 2010) dominates in the tropical region (~15°N to ~15°S) and is a mixture of AAIW and Pacific Deep Water (PDW). It carries distinct non-conservative tracer signatures such as very low oxygen concentrations (below 2 ml/l) originating from high biological activity and subsequent demineralization of organic material at the equator as a consequence of the upwelling of nutrient rich waters. EqPIW can be further subdivided into South EqPIW (SEqPIW) and North EqPIW (NEqPIW), separated by the equator and characterized by higher biological activity to the north of the equator (NEqPIW) (Bostock et al., 2010) (see Figure S1). The main mixing zone between pure AAIW and SEqPIW at the longitude of our seawater profiles (~165°W) is located between ~13°S and ~17°S. The locations of the stations of our study at 13°S (53-4), 9°S (2-2), and 3°S (21-1) allow investigation of the consequences of the mixing of these water masses for REE concentrations and Nd isotope compositions (see Figure S1). Below the intermediate water layer Upper Circumpolar Deep Water (UCDW) prevails, which detaches from the Antarctic Circumpolar Current (ACC) in the South Pacific near 120°W and first flows northward and then northwestward (similar to AAIW) toward the equatorial western Pacific (Figure 1B; Reid, 1997; Kawabe and Fujio, 2010). Along its pathway through the Pacific Ocean, this initially cold, salty and oxygen rich water mass loses its properties through mixing with PDW that carries a significantly lower oxygen content and dominates the mid-depth water column of the South Pacific (e.g., Molina-Kescher et al., 2014). As a result, the dissolved oxygen maximum of UCDW essentially vanishes north of 15°S in the Samoa/New Guinea region (Reid, 1997; Kawabe and Fujio, 2010). Lower Circumpolar Deep Water (LCDW) is the bottom water mass responsible for the ventilation of the entire deep Pacific given that no deep water is formed in the North Pacific. This water mass enters the Pacific basin southeast of New Zealand as a Deep Western Boundary Current and flows northward embracing the Manihiki Plateau (Kawabe and Fujio, 2010; Figure 1B).

A total of 27 water samples collected at different locations in Tahiti during two sampling campaigns in 2007 and 2009 were measured for REE concentrations. These include coastal and lagoon waters, stream/river waters, and the mixing of these with seawater as well as some samples from an active SGD spring “La source” (Table 1). These samples were filtered directly after collection through 0.2 μm 25 mm Sartorius Minisart syringe filters with a CA membrane into acid cleaned PE bottles. The samples were acidified within 24 h of collection with trace metal grade HNO₃ to a pH of ≤2.

The open ocean seawater samples of this study were obtained during RV Sonne cruise SO-225 from three profiles above the Manihiki Plateau in December 2012 (Figure 1, Tables 2, 3) using a CTD rosette equipped with 24 × 12 L standard Niskin bottles (Ocean Test Equipment Inc.). All samples were filtered directly from the Niskin bottles using AcroPak500 filter cartridges with 0.8/0.2 μm pore size into acid-cleaned PE bottles and cubitainers, 125 ml for REE analysis and 10 L for Nd isotope compositions, respectively. All samples were acidified to pH 2 onboard using distilled 6 N HCl and were stored at 4°C. REE concentrations for profile 21-1 (except for 5,172 m) and sample 53-4-3088 m were previously published in Osborne et al. (2017).

Yttrium and REE concentrations were measured at GEOMAR using a seaFAST online pre-concentration system (Elemental Scientific Inc., Nebraska, USA) directly connected to an Agilent 7500ce ICP-MS (Hathorne et al., 2012, 2015; Osborne et al., 2015). This method was part of the GEOTRACES intercalibration for REE concentrations (van de Flierdt et al., 2012). Tahitian samples were measured with a 4 mL loop as in Hathorne et al. (2012), while Manihikian samples were measured with an 8 mL sample loop. Calibration standards were prepared with a mixed YREE solution of a seawater-like composition in a natural seawater matrix, and the Manihikian samples were measured by time resolved analysis (Osborne et al., 2015). As the seaFAST column quantitatively removes the matrix (e.g., Hathorne et al., 2012) the same calibration standards were used for all samples regardless of their salinity. Total procedural blanks for the Tahitian samples, including filtering pure water in the field and storage in sample bottles, before being run through the preconcentration system just like samples, were < 0.1 ppt for all REE and < 0.2 ppt for Y. These were below the detection limits for most YREE and negligible except for the fresh water samples with very low conductivity. Although no procedural blanks were taken during the Manihiki campaign, comparable shipboard blanks from the GEOMAR lab obtained during other cruises using essentially the same methods were < 0.1 ppt for La and Ce and below the detection limit for the other REEs. Repeated analyses of GEOTRACES BATS reference seawater samples (van de Flierdt et al., 2012) were used to assess the accuracy and precision of the analyses (Table 3).

For details on the laboratory procedure see Behrens et al. (2018). Briefly, Nd was preconcentrated from 10 L of seawater using C₁₈ SepPak® cartridges (modified after Shabani et al., 1992; Jeandel et al., 1998). Neodymium isotope ratios were measured on a Thermo Neptune Plus multicollector-inductively coupled plasma-mass spectrometer (MC-ICP-MS) at the ICBM, University of Oldenburg. The measured ¹⁴³Nd/¹⁴⁴Nd ratios were corrected for instrumental mass fractionation to a ¹⁴⁶Nd/¹⁴⁴Nd ratio of 0.7219 (O'Nions et al., 1977) using an exponential law. The mass bias-corrected ¹⁴³Nd/¹⁴⁴Nd ratios were normalized to 0.512115, the accepted value for the JNdi-1 standard (Tanaka et al., 2000), which was measured at the beginning and end of a session and after every third sample together with an in-house Nd standard. External reproducibility, checked with multiple JNdi-1 analyses during each session, was typically better than ± 0.3 ε_Nd units (2σ, n = 11–19 per session). For individual samples, the 2σ propagated error derived from the internal error and external error (mean values of JNdi measurements of the respective session) is provided (Table 2). Standards and samples were run at the same concentrations. The accuracy of our measurements was checked using total procedural duplicates of seawater samples from two water depths of the GEOTRACES intercalibration station BATS (Bermuda Atlantic Time-Series Station) and one sample from the SAFe intercalibration (Sampling and Analysis of Iron) (Pahnke et al., 2012; van de Flierdt et al., 2012). Our measured average ε_Nd values of seawater at BATS (15 and 2,000 m) and SAFe (3,000 m) are identical within analytical uncertainties to the measurements of the average ε_Nd values reported by van de Flierdt et al. (2012) (see Behrens et al., 2018). Column chemistry blanks were <1.5 pg Nd (n = 6).

The Tahitian water samples show large differences in their absolute REE concentrations, with the largest range found for the freshwaters, ranging from 1.7 pmol Nd/kg (t141; a well water with very low conductivity) and 875 pmol Nd/kg (t197 river water), but at the same time exhibiting similarities in their PAAS normalized patterns, mostly resembling Tahitian basaltic rocks (Figures 2A–C). The REE concentrations and pattern indices (e.g., HREE/LREE, MREE enrichment, Ce and Eu anomalies), of the Tahitian water samples do not correlate with properties such as salinity, alkalinity or dissolved silicate (Figure S2). However, the classification of the samples according to salinity (Table 1) reveals systematic variations based on different REE pattern indices (Figure S3) as discussed in section “Basaltic influence on Tahitian coastal waters.”

Two seawater samples (t354 and t351) (Table 1) have clearly anomalous enrichments of La (520 and 192 pmol/kg) and Ce (652 and 323 pmol/kg) indicative of contamination by petroleum products (Olmez et al., 1991) and are not discussed further.

Despite the warm and wet conditions favorable for the chemical weathering of the fresh basaltic minerals, the Tahitian waters have relatively low total dissolved solids due to extremely high precipitation rates (up to 8,000 mm/a), which dilutes the solute concentrations, and short groundwater residence times in the sediments (<50 a) (Hildenbrand et al., 2005) that limit the exposure time of solids to the solutions. For comparison, relative to the size of the river catchment area, the discharge of Tahitian rivers from all parts of the island is similar to that of the biggest rivers on Earth (Table S1).

The large range of REE concentrations found for the Tahitian waters and the fact that there are no significant correlations between REE concentrations or pattern indices with salinity, alkalinity or dissolved silicate (section REE compositions of Tahitian waters, Figure S2) suggests a reversible absorption-desorption based exchange process, likely involving precipitation-dissolution of secondary minerals, where recirculating seawater via SGD (e.g., Duncan and Shaw, 2003) through the basaltic and coral reef sedimentary rocks plays an important role. This is supported by both well-developed negative Ce anomalies (≤0.5) (seawater derived signal) and high Eu anomalies (≥1.5) (basaltic weathering signal) found in many of the samples regardless of their origin (Figures 2A–C). A multitude of systematic variations in the REE patterns and concentrations (Figure 5 and Figure S3) reveals the consistent offset of SGD influenced waters from the other seawater, freshwater and mixture samples. A clear example of this can be seen in the plot of Nd concentrations against Y/Ho where the SGD samples exhibit the most fractionated values (Figure 5). Y and Ho have very similar speciation in seawater so such a strong fractionation is likely to be the result of the preferential removal of Ho by secondary precipitates such as Fe-Mn oxides (Bau et al., 1997). Y/Ho fractionation during carbonate mineral precipitation is also well documented (e.g., Tanaka et al., 2008) and the high Y/Ho ratios of the SGD may also reflect secondary precipitation within the reef carbonate rocks. Given the affinity of REEs for carbonate complexes in seawater (e.g., Cantrell and Byrne, 1987; Schijf et al., 2015) it should be expected that reef carbonate rocks surrounding the tropical basaltic islands play an important role in SGD recirculation process. This is the case for the Faaromai river mouth, which exhibits the highest Nd concentrations that decrease exponentially with increasing salinity (t197 < t196 < t195). This river is located on the northern coast of Tahiti where an edifice collapse caused two giant landslides (Hildenbrand et al., 2006), which most likely occurred around 870 ka, and coral reefs have not yet grown back. Extrapolation to seawater salinity of 36.2 using the exponential fit (Figure 6) gives a Nd concentration of 89 pmol/kg, which is the largest flux of REEs from the studied rivers. In contrast, the two other salinity gradients sampled at river mouths exhibit a more typical estuarine REE behavior with removal during initial mixing and subsequent release as salinity increases (e.g., Elderfield, 1988; Sholkovitz, 1995) producing concentrations around 40–50 pmol/kg in the saltiest samples.

The highly variable REE concentrations of freshwater and SGD observed on Tahiti make the selection of an endmember value for traditional flux calculations (e.g., Kim and Kim, 2011) beyond the feasibility of such spatially and temporally coarse sampling. What is remarkably consistent in the Tahitian data are the REE concentrations of the high salinity (~36) water samples (Table 1), from both lagoon and beyond the reef break “open ocean” settings (Figure 1). With an average Nd concentration of 21.8 ± 3.1 pmol/kg (1 SD), Tahitian coastal waters have Nd concentrations at least 4 times that of South Pacific surface waters (<5 pmol/kg, Amakawa et al. (2013); Jeandel et al. (2013); this study). The MREE enrichment and Eu/Eu^* found in these Tahitian coastal seawaters, both 1.4 ± 0.1 (1 SD), is clearly derived from the basaltic rocks present on the island (Figures 2C, 7), with a very similar REE pattern like those found on other central Pacific islands [see compilation by Grenier et al. (2013) in their Figure 12] and in groundwaters from the basaltic islands of Jeju (Korea) (Kim and Kim, 2011) and Hawai'i (Johannesson et al., 2017) (Figure 7), thus suggesting a typical REE signal for land-sourced waters supplied from Pacific volcanic islands. Estuarine REE patterns from Papua-New Guinea (Sepik and Fly rivers), although showing common characteristics (such as the Eu peak) to the basaltic island PAAS normalized distributions described above, also present some differences resembling the brackish Amazon estuarine water samples (Figure 7; Rousseau et al., 2015). This suggests that the presence of suspended particulate matter carried by relatively larger rivers (such as these found in Papua-New Guinea) exerts a considerable influence on the coastal seawater REE patterns which differ from those of small islands that lack high volumes of channeled water, like Tahiti, Jeju, or Hawaii. Johannesson et al. (2017) suggested the aridity of the Kona area (in the rain shadow of both Mauna Loa and Mauna Kea) may play a role in the lack of modification of SGD REEs upon mixing with seawater.

The similar PAAS normalized pattern observed between Tahiti, Jeju, and Hawaii (Figure 7) clearly indicates a common signature of basaltic rock REE sources in the Pacific Ocean.

Similar to other locations such as offshore Hawaii (Fröllje et al., 2016), the markedly radiogenic ε_Nd signatures of shallow Pacific waters strongly suggest the influence of basaltic island weathering (Figure 8). This observation is consistent with previous studies (e.g., Sholkovitz et al., 1999; Lacan and Jeandel, 2001; Amakawa et al., 2004; Hongo et al., 2006; Allègre et al., 2010; Fröllje et al., 2016), which proposed island weathering as a prominent source of REE to the surface ocean, in particular taking into account the high solubility of basaltic rocks in weathering solutions and seawater (Pearce et al., 2013). However, about 2,000 km downstream from Tahiti, Manihiki surface waters have evolved different REE patterns compared to Tahitian coastal waters (Figure 9). Despite transiting the South Pacific gyre, presumably one of the most particle poor regions of the oceans, REE concentrations decrease (e.g., Nd 4-fold), Ce anomalies strengthen, and Eu anomalies disappear. Interestingly, Y/Ho ratios remain elevated and perhaps even strengthen slightly between Tahiti and Manihiki, producing values similar to those found in surface waters of the SW Pacific (Zhang and Nozaki, 1996). Unfortunately, most studies, including recent works (e.g., Grenier et al., 2013; Jeandel et al., 2013; Fröllje et al., 2016; Johannesson et al., 2017), don't include Y data, therefore limiting the extent of this comparison. However, given the results here and the fact that Y/Ho fractionation most likely requires interaction with particles (e.g., Bau et al., 1997), future studies should include Y in their measurement routines. A volcanic island imprint on the open ocean surface waters cannot be unambiguously attributed but remains the most likely explanation for the radiogenic Nd isotope signatures observed. Hopefully, Nd isotope compositions of Tahitian waters will be obtained in the frame of future investigations but these will likely be very radiogenic, similar to the groundwater and surface water data from Oahu Island (Fröllje et al., 2016), clarifying the impact of basaltic island weathering on the tropical South Pacific Nd budget. To make a first order calculation of the potential Tahitian REE flux to seawater, we have chosen to assume that the consistently measured coastal seawater Nd concentrations of ~20 pmol/kg are found only in the upper layers and close to the coastline. Future work should test this assumption with transects perpendicular to the shoreline sampling the entire water column and measurements of current velocities near the coast. With such conservative assumptions in mind, water transport in the South Equatorial Current south of 9°S (between 150° and 160°W) around 13.4^*10⁶ m³/s and a mean speed (ú) of 4.9 cm/s (4234 m/d) near Tahiti (Wyrtki and Kilonsky, 1984) and a general 15 pmol/kg Nd enrichment around the entire 193 km of the Tahitian coast with respect to superficial open ocean values (<5 pmol/kg, Amakawa et al. (2013); Jeandel et al. (2013); this study), requires a Nd flux of between 126 and 0.016 mol/d depending on the area used for the transport calculation (Table 4). The largest flux estimate is remarkably similar to the estimated SGD Nd flux of 120 ± 60 mol/d from the similar sized Jeju island (Kim and Kim, 2011) but this is unlikely as it would require all waters transiting the Tahitian coastline to have open ocean Nd concentrations, a scenario which is unrealistic as waters moving along the coast would likely already have elevated Nd concentrations. The smaller estimates are similar in size to estimates of the SGD Nd flux from the Kona coast (Tahiti is approximately 1/10 the size of Hawai'i). In order to tentatively assess the effect of volcanic islands such as Tahiti in the global REE budget, we take our conservative estimates from an island of 1,045 km² and extrapolate that using the global “island volcanic rock surface area” of 2.02 x 10⁶ km² estimated by Allègre et al. (2010) to obtain estimates ranging from 1.1 × 10⁴ mol Nd year⁻¹ to 8.87 × 10⁷ mol Nd year⁻¹ (Table 4). A second way to estimate the flux is to ignore the large variability in freshwater and SGD Nd concentrations measured on Tahiti, and follow the approach of Allègre et al. (2010) for Sr, use the averaged rainfall minus evapotranspiration on the total cumulative surface area of volcanic islands globally with an average Tahitian freshwater Nd concentration (249 pmol/kg) to obtain a flux of 3.17 × 10⁵ mol Nd year⁻¹ (assuming 70% removal on mixing with seawater; e.g., Rousseau et al., 2015). These first order extrapolations produce a global Nd flux from volcanic island weathering that is probably at least two orders of magnitude lower than the global sediment pore water flux estimated to be >18 × 10⁶ mol Nd year⁻¹ (Abbott et al., 2015). However, that estimate is also extrapolated from a single location where a reactive radiogenic phase (possibly basalt minerals) enhance fluxes (Abbott et al., 2016). If the basaltic island SGD fluxes from Jeju island are more representative, basaltic island weathering could easily fulfill the model budget requirements for the missing Nd source to the ocean (Tachikawa et al., 2003; Arsouze et al., 2009). Knowing which basaltic island fluxes are most representative will require many more studies from other basaltic islands but we have identified factors such as groundwater residence time, the presence of coral reefs or carbonate sediments and landslides spreading more of the volcanic material into contact with seawater, that should all be studied further. Although surface water REE distributions can be clearly altered by volcanic islands there may also potentially be an influence on intermediate depth water masses.

On AAIW's pathway from its source region in the SE Pacific sector of the Southern Ocean until it loses its geochemical identity in the volcanic WPWP region (Lacan and Jeandel, 2001; Grenier et al., 2013, 2014), AAIW characteristics are modified by mixing with other water masses, and by potential input/exchange from basaltic islands of the South Pacific. At the Manihiki Plateau, AAIW corresponds to a potential density of around 27 σ_θ between ~400 and ~1,300 m (see section “General Hydrography”), and is considerably diluted as observed in Figure 3 and Figure S1, displaying ε_Nd signatures ranging between −3.1 and −4.0, similar to deeper PDW (Figure 4F). This indicates that the original Nd isotope composition of AAIW (−8 to −9 ε_Nd, Molina-Kescher et al., 2014) vanishes in this area, which is evident when compared to other stations south of our studied stations (Figure 8). Particularly remarkable is the comparison of stations 53-4 and SX-22 (Figure 8), which are located close to each other (see locations on Figure 1 and Figure S1) but present a marked offset of about 3 epsilon units in the depth range of AAIW. This change corresponds to the progressive mixing of AAIW and SEqPIW in this area (Figure 3 and Figure S1; Bostock et al., 2010), which is most evident in [O₂] (Figure 10). Purer SEqPIW sampled at equatorial station 21-1 (649 m water depth) is even more radiogenic at ε_Nd value of −1.5 (Figure 8), which is accompanied by an oxygen minimum of 1.5 ml/l, with both, ε_Nd and [O₂] changing toward more negative ε_Nd and higher [O₂] values above and below that depth. Plotting the Manihiki intermediate waters on the AAIW – SEqPIW mixing array (Figure 10) clearly reveals an offset toward more radiogenic signatures suggesting an additional Nd input/exchange along route which may derive from volcanic Society, Tuamotu and Tubuai Islands (including Tahiti). This observation supports the suggestions of Amakawa et al. (2013), who estimated a large Nd supply from these islands to AAIW of 2.08 ± 1.53 × 10⁵ mol Nd year⁻¹, within error of the estimate provided in section “Imprint of volcanic islands on surface and subsurface waters (0–300 m water depth, <26σ_θ) of the open central Pacific.”

In order to determine changes in the REE composition of AAIW we normalized the available REE data along its pathway to averaged REE concentrations in the formation region and depth of this water mass. This is shown in Figure 11, which includes averaged data in the depth range (370–~1,000 m) of AAIW from the EGY and GYR stations (unfiltered samples) in the SE Pacific near the formation region of AAIW (Jeandel et al., 2013); stations 53-4 and 2-2 (this study) located downstream of the first encounter of AAIW with volcanic islands; and FLUSEC 22 in the WPWP (Grenier et al., 2013). Station 21-1 (this study) is also shown in order to evaluate differences with respect to SEqPIW, which dominates the intermediate water at this location. In general, all samples have lower LREE concentrations compared with original AAIW, which is expected as the more readily scavenged LREEs are preferentially removed from the water column as the water masses age (e.g., Osborne et al., 2017). The fact that the source AAIW normalized REEs are < 1, especially for the LREEs, clearly indicates the importance of removal by particles during transit through the gyre, despite relatively low productivity. Station GYR shows a LREE pattern more similar to Manihiki stations than the much more proximally located EGY station (Figure 1). However, the GYR station is located in ocean current downstream (only ~500 km to the East) of another volcanic island, Easter Island (Rapa Nui) including three inactive volcanoes. Given the similarity between REE patterns of basaltic islands separated by such large distances in the Pacific (section “Basaltic influence on Tahitian coastal waters”), the similar REE signals of the GYR and Manihki stations posisibly originate from volcanic REE inputs into intermediate waters. Nevertheless, the interpretation of the GYR and EGY data needs to be considered with caution given that these samples were not filtered prior to analysis, which potentially explains the apparent HREE enrichment with respect to original AAIW observed at these stations. SEqPIW (station 21-1) reflects characteristics similar to neighboring stations 2-2 and 53-4, indicating similar geochemical characteristics of SEqPIW and AAIW in terms of REE concentrations.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.