Arima-type brine is non-volcanic hot spring water with high salinity found in a fore-arc region with no Quaternary volcanism [2]. This water has been geochemically characterized by a high Cl content at ~40000 ppm and specific O–H isotopic ratios that are similar to magmatic/metamorphic thermal waters. Despite its occurrence in a non-volcanic region, this water is distinct from meteoric water or buried sea water [2-4]. It also shows high 3He/4He ratios comparable or close to the mantle value and characteristic ratios of both stable and radiogenic isotopes (i.e., H, C, O, Sr, Nd, and Pb) that indicate slab-fluid derived from the subducted Philippine Sea Plate (PHS), suggesting a deep origin of the brine [1,5-7].

Arima springs in southwest Japan (Figures 1, 2A and 2B), the type locality of Arima-type brine, consist of hot springs (?25°C) and associated cold springs (<25°C) including highly carbonated water. These springs generally exhibit a wide O–H isotopic range, which can be explained by mixing of meteoric waters and the deep brine component [3,4,7]. δ¹⁸O is clearly correlated with δD, forming a straight mixing line between the meteoric water and the predicted slab-fluid generated by dehydration of PHS at ~50 km depth beneath the Arima area [7] (Figure 3A). The isotopic ratios also constitute linear variations with major solute elements such as Cl and Na (Figures 3B-3D), upon which the composition of the deep brine component has been estimated [4,7].

Figure 1: Tectonic setting in Japan and location of the study area. The map shows the distribution of Quaternary volcanoes (red circles) and the geometry of the subducting Pacific and Philippine Sea slabs relative to the Itoigawa– Shizuoka Tectonic Line (ISTL) and the Median Tectonic Line (MTL). The pinkish contour lines indicate the depth of the upper surface of the Pacific slab (50 to 300 km depth with 50 km intervals), whereas the purplish contour lines indicate that of the Philippine Sea slab (10 to 200 km depth with 10 km intervals). The aseismic parts are shown by the dotted line.

Figure 2: (A) Geologic map of the Arima area after Maruyama and Lin [12], shown as the square region in Figure 1.

Figure 2: (B): Locations of the spring waters in the Arima area.

The deep-seated brine is relatively well characterized, as previously mentioned. It is thought to ascend along faults such as Atago-yama and Tenjin-yama faults associated with large tectonic lines such as the Arima–Takatsuki Tectonic Line (ATTL), as shown in Figure 2A, without undergoing significant geochemical modification [7,8]. However, the mechanism of encounter between the deep brine and the shallow meteoric waters to be mixed remains poorly understood. In this study, we aim to constraining the geochemical processes at such shallow parts in which the deep component is mixed with the meteoric waters. For this objective, we analyze and utilize rare earth elements (REEs), particularly lanthanoids, of Arima spring waters. The geochemical behavior and partitioning of REEs between solid and fluid are sensitive to temperature, fO₂, fCO₂, and pH and may provide key information on the mixing processes at the shallow level where temperature, volatile fugacity, and pH are potentially variable [9,10]. First, we provide new data on REEs for Arima spring waters and compare the results with major solute elements and O–H isotopes, resulting in the identification of several types of spring water [9,10]. On the basis of these variations, we discuss the geochemical processes and mechanism related to the mixing of deep brine with meteoric waters.

In the Arima region, spring waters occur in a narrow area of approximately 1 km² (Figure 2B) and show compositional variation in major solute elements and gases [7]. These springs appear to upwell through the ATTL and subsidiary faults striking NW–SW or NW–SE with dextral slip associated with the Median Tectonic Line formed in the late mid-Miocene in the southwest Japan arc (Figure 2A) [11,12]. The average slip rate for the eastern range-front segment of the tectonic line during the late Quaternary period is estimated to be 0.5–1.5 mm/ year dextrally and 0.1–0.8 mm/year vertically [13]. The southwest Japan arc is associated with two oceanic plates, the Pacific plate and the PHS, which subduct beneath the area from the east at 9 cm/year and from the southeast at 4 cm/year, respectively. The slab surface depth beneath the Arima area is ~400 km for the Pacific slab but 50–80 km for the PHS, showing a large uncertainty [14,15]. Despite this active subduction, a Quaternary volcano has not formed in this region because the Pacific slab is too deep and the PHS is too shallow to fulfill the physiochemical conditions for arc magma generation (Figure 1) [16].

As shown in the geological map of this area (Figure 2A), the basement around the Arima area is composed of late Cretaceous felsic volcanic rocks such as rhyolite of the Arima Group, granitic rocks such as Rokko granite, and late Eocene to early Oligocene non-marine sedimentary rocks with rhyolitic tuff layers of the Kobe Group [17]. The Arima Group directly covers the Rokko granite south of the ATTL and the sedimentary rhyolitic rocks north of the ATTL in the Arima area [18].

Sample description of Arima spring waters

The hot spring sources in the Arima area have been characterized by NaCl and hydrocarbon contents in addition to O–H–He isotopic composition [4,7], in most of which (Ginsen, Gokuraku, Gosho, Inari-Kinsen, Kinsen, Tansansen, Tenjin, and Uwanari waters) were investigated to understand their generation processes in this study (Figure 2B). During the dry season in December 2010, we collected 10 L samples of each of these waters directly from the well pipe before the water is pooled in tanks to be oxidized for commercial use. The Japanese names of these spring waters carry deep meanings. For example, “Kinsen” means gold-colored hot spring source, and “Tansansen” means high abundance of CO₂. Kusuda et al. [7] analyzed 12 solute elements/components and the isotopic ratios of H, He, C, and O in these spring waters and gases. Their results indicate a large range of elements, particularly in Cl (14–40,000 ppm) and hydrocarbon (12–1300 ppm) contents, adding to the wide range of O–H isotopic compositions. The He isotopic ratio has a smaller gap among these samples at 2–10 (×10^-6), which is similar to the mantle value, indicating deep origin [7]. The Li/Cl ratios for the spring waters are almost stable with a maximum of approximately 0.001, indicating a signature of Arima-type brine as suggested by Kazahaya et al. [19]. An exception is ‘Tansansen’ water, which has twice that value. These chemical signatures of spring waters can be explained by the mixing of meteoric water with slabderived fluid of the PHS thought to be dehydrated at 50–80 km depths. The spring water with the lowest O–H isotopic composition close to a meteoric signature is identified with the lowest content of Cl and other cations such as Na, Ca, Br, Li, and Sr and the highest He³/He⁴ ratio, named as “Tansansen” which means saturated water with isolated carbonaceous gas. That with an opposite signature close to the endmember of ‘Deep Brine’, with the highest Cl and a higher He³/He⁴ratio is identified as “Kinsen” (Figures 3A-3D). Other spring waters with moderate Cl and cation contents plotted between ‘Tansansen’ and ‘Kinsen’.

Figure 3: Correlations among O isotopic ratio versus (A) H isotopic ratio, compared with the meteoric water line and the compositional trend depending on the depth of subducting slab (tick mark) based on Rayleigh fractionation [7]. (B) Cl (C) Ca (D) Li abundances (ppm) in Arima spring waters with the spring names in blue. The estimated dense end-member deep brine is marked in red [7].

Analytical method for REE composition in high-salinity brine

The high-salinity and solute elements in the brine may interfere with quantitative analysis of the REEs due to the matrix effect. In case of several water samples contained visible particles, we dissolved the visible particles in water by using acid. In order to consider the matrix effect with high-salinity conditions, we applied the standard addition method for solutions with strong matrix effects with 0.4 ml addition of XSTC-1 (SPEX CertiPrep Co. Ltd., USA) by 0.001 ppb, 0.01 ppb, and 0.1 ppb in each 2.0 ml of each sample [20]. The samples were carefully diluted with ultra-pure water (>18.3 MΩ milli-Q water, Millipore Corporation, Japan) by using a weight scale in a class 100 clean room to minimize possible environmental contamination. The ICP-MS (iCAP-Qc, ThermoFisher Scientific) analysis was conducted at the Japan Agency for Marine–Earth Science and Technology (JAMSTEC). The analytical conditions were constantly monitored by using a standard tuning solution and diluted 5% HNO₃ acid solution. The reproducibility standard deviation (RSD) was approximately 2–4% for five repeated data acquisitions, indicating analysis stability [21]. The detection limit was 0.1 ppt. It should be noted that Eu, Nd, Sm, and Gd were quantitatively re-calculated after being combined with 135Ba for correction due to significant interference from Ba oxides, which enabled us to discuss the Eu-anomaly. The most significant interference on REEs signals come from Ba polyatomic ions, because Ba is one of the highest abundant elements in the brine samples. We carefully investigated the formation of these polyatomic ions in ICP-MS measurement by analyzing pure Ba solution, and estimated for typical brine waters that Ba interfering ions contributed to La, Ce, Nd, Sm, Eu and Gd raw signals for 92%, 24%, 48%, 78%, 96% and 75%, respectively. Pr is free from any of Ba oxide and hydroxide interference which exist farthest to Gd. Intra REEs interfering, e.g., PrO, NdO and SmO overlapping on MREE and HREE signals, were found less than 2.7% (mostly <1%), thus ignorable for this study.

REE abundances and patterns of Arima spring waters

The analytical results of REEs are listed in Table 1. The REE compositions are plotted in Figure 4 as spidergram patterns normalized by depleted mid-ocean ridge basalt (MORB) mantle (DMM). The REE abundances of Arima spring waters are approximately three to five times lower than those of DMM. We have detected four distinct patterns in these spring waters: (i) a flat pattern with a negative Euanomaly, represented in the figure by the light blue line; (ii) a slightly convex-down pattern of increasing concentration toward both the left and right, indicating light REEs (LREEs) and heavy REEs (HREEs), respectively, associated with a moderately positive Eu-anomaly (blue line); (iii) a strongly convex-down pattern with a strong positive Euanomaly (black line); and (iv) a slightly convex-down pattern with a moderately positive Eu-anomaly and a negative Ce-anomaly (dark blue line). Except for ‘Tansansen’ (light blue line), which is a highly carbonated cold spring, all spring waters have positive Eu-anomalies; ‘Kinsen’ (blue line) shows the highest REE content. Among ‘Ariake’, ‘Gosho’, ‘Gokuraku’, ‘Tenjin’, and ‘Uwanari’, the five samples classified as (iii), the degree of positive Eu-anomaly increased for samples with a stronger convex-down pattern. ‘Kinsen’, classified as (ii), maintained overall high concentrations for both LREEs and HREEs with a slightly convex-down pattern (Figure 4). ‘Ginsen’ (dark blue line), classified as (iv), is a cold spring exhibiting a relatively flat pattern and is in this sense similar ‘Tansansen’ (i). However, the former shows overall low concentrations (Figure 4).

Table 1: Rare earth element (REE) abundances (ppb) of Arima spring waters. Abbreviation: n.a. = not analyzed.

Figure 4: Depleted mid-ocean ridge basalt (MORB) mantle (DMM)-normalized rare earth element (REE) compositions of Arima spring waters analyzed in this study. The data are listed in Table 1.

Correlation of REEs and major solute elements

The solute major elements, Cl and Li abundances, and δ¹⁸O–δD– He isotopic ratios in the spring waters, the REEs of which are presented in Table 1, are summarized in Table 2 [7,22]. As previously mentioned, all spring waters are aligned along a single mixing line between the meteoric water and a dense end-member as the deep brine component (Figures 3A-3D), which approximately corresponds to a slab-derived fluid dehydrated at an approximately 50 km depth [7]. ‘Kinsen’ has high O and H isotopic ratios close to the dense end-member, indicating less involvement of meteoric water, whereas the two cold springs, ‘Tansansen’ and ‘Ginsen’, have nearly the same δ¹⁸O and δD as the meteoric water (Figure 3A). However, as shown in Table 2, ‘Tansansen’, the highly carbonated cold spring water, has the highest ³He/⁴He ratio, indicating deep origin; ‘Kinsen’ also has a high ratio indicative of its mantle origin. Other springs are plotted along a mixing line between ‘Tansansen’ and air-saturated water [7]. It is also noted that no clear relationship is observed between dissolved HCO₃ and the major solute concentrations, particularly Cl, that represent the proportion of dense end-members involved in the samples (Figure 2) (Table 2).

Table 2: Major solute elements and isotopic composition discussed in this study. The data are from [7]. Abbreviation: n.k. = not known.

Concerning REE abundance in the spring waters, no clear correlation was noted with the Cl content and thus the proportion of deep brine component (Figures 5A-5C), implying that in addition to the mixing of the dense end-member with meteoric water, additional, independent processes controlling the REE content are required. The variability of the REE pattern shown in Figure 4 likely reflects such multiple processes (Figure 4). Because the dense, deep brine is thought to ascend through the fault system from depth without significantly reacting with the surrounding rocks [7], shallow crustal processes likely affect the REE behavior, where the dense brine encounters different environments with temperature and redox states distinct from those of the deep part. Such possibilities are discussed subsequently. As shown in Figure 5A-5C, ‘Tansansen’, being similar to the He systematics, again shows distinctly high REE concentrations despite a low Cl content. If ‘Tansansen’ is excluded, a broad and non-linear positive correlation exists between REE and Cl contents, with ‘Kinsen’ as an end-member of the highest concentration for most of the elements (Figure 5A-5C). This broad correlation is reflected in the REE pattern: ‘Kinsen’ plots at the highest concentrations, and the other five hot springs plot at lower concentrations with stronger convex-down patterns (Figure 4).

Figure 5: Correlation diagrams between Cl abundance (ppm) versus (A) La, (B) Gd, (C) Yb in ppb.

The convex-down pattern is related to the Eu-anomaly (Figure 4). In this section, we examine the relationship between the major solute elements and Eu abundance as shown in Figure 6A-6C. The behavior of Eu appears to be coupled with that of Cl, Ca, and Sr, although two broadly linear trends consisting of relatively cold and hot spring waters, respectively, are noted (Figure 6A-6C). This correlation implies that the Eu abundance is roughly related to the proportion of deep brine (Figure 6A) and is different from other REEs (Figure 5A-5C). As is graphically expressed in Figure 4, the five hot spring waters are depleted in REEs compared to ‘Kinsen’ and have decreased Cl contents, although the decrease in Eu abundance is less than that in other REEs. Figure 6 also shows that Eu behaves in correlation with Ca and Sr, suggesting the possible involvement of plagioclase, which may selectively buffer Eu among the REEs [23]. This possibility is examined subsequently.

Figure 6: Correlation diagrams between Eu (ppb) abundance versus (A) Cl, (B) Ca, (C) Sr in ppm

In previous studies of the geochemical characterization of slabderived fluid beneath volcanic arcs, no Eu-anomaly has been reported for slab-fluids as a possible source of the deep brine [24]. The estimated composition of slab-fluid from the Philippine Sea slab generally shows a smooth pattern with downward sloping toward HREEs [25]. The slab-derived fluid exhibits variable REE abundances depending on the temperature of dehydration. With slab dehydration at lower temperatures beneath the fore-arc region, including the Arima area, REEs in the slab-fluid exhibit significantly lower abundances due to the large temperature dependence of the partition coefficient between fluid and residual solids [26-28]. As a result of dehydration, the REE pattern in a slab-derived fluid shows a smooth and slight convex-down pattern. Therefore, the positive Eu-anomaly observed for ‘Kinsen’ and other hot spring waters (Figure 4) is attributed to processes that occurred during their ascent from depth, likely at a shallow level when the deep brine encountered near-surface water under oxidizing conditions.

In this case with oxidizing conditions within a relatively shallow aquifer, if the temperature is decreased from the original high temperature state up to 500°C, REEs can coprecipitate as a form of Ln(OH)3 with Fe oxyhydroxide [9,10]. The chemical reaction of Ln, involving slab-derived fluid and near-surface (e.g., meteoric) water as reactants with solid precipitates as reaction products, is expressed as Ln³⁺ _(aq) + 3H₂O_(aq) ? Ln(OH)₃(ss) + 3H⁺ _(aq)’, where (aq) and (ss) denote aqueous solution and solid solution, respectively. It is also suggested that during this precipitation process, Eu in the spring waters has to be retained relative to the other REEs, resulting in a positive Eu-anomaly. If plagioclase abundant in Eu with a positive anomaly exists in the matrix of an aquifer, it may supply Eu for buffering, as is observed. Therefore, the low REE abundance with a strong positive Eu-anomaly in Figure 4, as is observed for ‘Gokuraku’, ‘Gosho’, ‘Inari-Kinsen’, ‘Tenjin’ and ‘Uwanari’ hot springs, is interpreted to be a result of overlapping effects associated with elution of Eu and precipitation of other REEs under relatively oxidized conditions. The difference in these five hot springs and ‘Kinsen’ could be attributed to the degree of the reaction progress due to retention time within the aquifer, although these springs may share a common deep brine.

Conversely, the negative Eu-anomaly observed in ‘Tansansen’ may have developed under relatively reductive conditions in presence of feldspar such as by reaction with surrounding rocks having negative Eu-anomalies such as Rokko granite or that by (organic) C in the sedimentary rocks exposed in this area [4,23]. The relationship between Eu behavior and other REEs concerning ‘Tansansen’ is discussed subsequently.

Behavior of REEs

Despite the negative Eu-anomaly and the lowest Cl content of the spring waters analyzed in this study, ‘Tansansen’ shows high REE concentrations and a flat pattern, except for Eu. The flat REE pattern is notably distinct from those of other Arima spring waters, basement igneous rocks, and river water [1,25] in particular, ‘Tansansen’ has higher HREE contents. This feature can be explained by the existence of CO₂ gas and carbonate species in aqueous fluid [26-28]. The gassaturated fluid exhibits solubility of Ln³⁺(CO₃)₂ - approximately four times higher at 1 atm [29,30]. In addition, REEs are more preferentially partitioned into fluid than Fe oxyhydroxide with increasing NaHCO₃ concentration, and the dominant REE(III) species in aqueous solutions changes from REE³⁺_(aq) to REECO³⁺_(aq) and then to REE(CO₃)^2-`_(aq). This transition decreases Racah (E1 and E3) parameters for 4f electron repulsion of REEs and rapidly modifies the REE abundances in aqueous solution from a relatively low level with a significant tetrad effect pattern to higher concentrations with a smooth pattern, depending on the carbonate abundance [9]. Therefore, highly carbonated water such as ‘Tansansen’ may have a high and flat REE pattern caused by reaction with solids. The ‘Ginsen’ cold spring, which has δ¹⁸O and δD close to the meteoric water line (Figure 3A), also exhibits a relatively flat pattern with a weaker positive Eu-anomaly compared than that in hot spring waters (type (iv) in Figure 4), suggesting a carbonate effect similar to ‘Tansansen’ but to a lesser extent. It is also noted that the cold springs are derived from a shallow aquifer seated at ~40 m in depth, yet their REE concentrations are significantly higher (up to 1000 times) than those in nearby river waters (Figure 5) [7]. This result indicates that a significant uptake of REEs has occurred even during a shallow cycling of near-surface water.

Comprehensive model for Arima spring waters

Based on the observations and arguments discussed thus far, we propose a model for shallow-level chemical evolution of Arima spring waters, as shown in Figure 7. At least two types of meteoric aquifers occur beneath the study area. One correspond to lower aquifers (aql) seated deeper than 160 m, and the other represents upper aquifers (aq-u) shallower than ~40 m [7,8]. The slab-derived fluid dehydrated at a minimum temperature of 500°C from subducting PHS migrates upward and enters the upper aquifer, where it is mixed with meteoric waters to variable extents and is cooled to ~130°C [8]. Within the lower aquifer, the precipitation reaction, discussed above, occurs to induce degassing of CO₂ and He. The separated gas phase migrates upward and enters the upper aquifer to produce highly carbonated meteoric water with high 3He/4He ratios [2,7]. The precipitates remove REEs from the aqueous solution to variable extents, whereas the solution reacts with feldspar with a strong positive Eu-anomaly to produce the geochemical characteristics of ‘Gokuraku’, ‘Gosho’, ‘Inari-Kinsen’, ‘Tenjin’, and ‘Uwanari’ hot springs (Figures 4 and 7). The differences among these five hot springs may represent the degrees of such reactions reflecting retention time in aq-l.

Figure 7: Geochemical evolution model for spring waters in the studied area. Meteoric aquifer-l is the same as lower meteoric aquifer (aq-l) in the main text. Meteoric aquifer-u is the same as upper meteoric aquifer (aq-u) in the text. The abbreviations used in the equations, (aq) and (ss), denote aqueous solution and solid solution, respectively.

Within aq-u, the host rock is thought to be composed of silicic igneous rocks with high REE contents with negative Eu-anomalies such as the Rokko granite or rhyolite exposed in this area [23]. When the CO₂-bearing gas phase that separates from aq-l dissolves into aq-u, dissolution of the REE elements from the rock matrix to the aqueous solution is enhanced, fingerprinting the Eu-negative anomaly of the host rocks. These processes within aq-u produced ‘Tansansen’ and ‘Ginsen’, the latter showing lesser dissolution. The ‘Kinsen’ hot spring water is nearly upwelled directly from the subducted slab through both meteoric aquifers, even though a small degree of interaction is inferred from the weak positive Eu-anomaly, as shown in Figure 7.

In order to test this scenario, knowledge of the composition and reaction ratio of the meteoric waters and the host rocks is required. On the basis of these data, the model calculates qualitative and quantitative revisions. In any case, the abundances and patterns of REEs in the spring waters appear to provide unique and invaluable information on the chemical evolution of the deep brine at relatively shallow levels, which has been obtained by neither major solute elements nor isotopic compositions. The REE compositions will be useful for investigating such shallow chemical processes of other springs, especially cold–hot spring association, because the nature of the deep brine is clarified through removal of the shallow effects.

In addition to examining the major solute elements and isotopic compositions including δ¹⁸O–δD, He, and Sr–Nd–Pb systematics reported in previous studies, we have performed REE analyses for both hot and cold spring waters in the Arima area. We have shown that the simple binary mixing between the deep brine and meteoric water inferred from the previous studies cannot explain the REE variations. We have identified the following four distinct REE patterns according to isotope composition and major solute element and gas compositions: (i) high REE content with a flat pattern and negative Eu-anomaly, having the lowest δ¹⁸O and δD isotopic ratios with abundant CO₂ gas and the highest He isotopic ratio; (ii) high REE content with a slightly convex-down pattern and positive Eu-anomaly, having the highest δ¹⁸O and δD isotopic ratios with a high He isotopic ratio; (iii) lower REE content with a strong convex-down pattern and a positive Euanomaly, having lower δ¹⁸O– δD isotopic ratio with lower He isotopic ratios; and (iv) a slightly convex-down pattern with a moderately positive Eu-anomaly and a negative Ce-anomaly, having similar δ¹⁸O, δD, and He isotopic ratios as that in type (i).

We have presented the following model to explain the shallowlevel chemical evolution of the deep brine resulting in distinct features of Arima spring waters. Two sources are present for REEs. The first is deep brine likely originating from the subducted Philippine Sea slab, and the other is basement crustal rocks serving as aquifers. The deep brine encounters meteoric water at the lower aquifer under relatively oxidizing conditions, in which the formed precipitates remove REEs from the solution and a CO₂-bearing gas phase is generated. In the upper meteoric aquifer, the gas from the lower aquifer dissolves in the meteoric water and enhances the solubility of REEs from the host rocks into the solution. This process produces cold spring waters with a flat REE pattern of relatively high concentrations. This information obtained by REE analysis will be useful for understanding the shallow geochemical processes for other spring waters that have not been identified by major solute or isotopic studies.

We would like to thank T. Yokoyama, K. Fujinaga, T. Ishikawa, M. Tanimizu, H. Sakuma, and H. Masuda for their help at various stages of this study, and M. Totani, the president of Arima Hot Springs Tourism Association, for permission and help with our field work and sampling. This work was supported by JSPS KAKENHI Grant Number 25400524 and the Cooperative Research Program of Earthquake Research Institute, The University of Tokyo.