Alteration, mass analysis, and magmatic compositions of the Sentinel Bluffs Member, Columbia River flood basalt province

There is perhaps no more important aspect to understanding flood basalt volcanism, and to applying this knowledge to derivative geologic applications, than establishing the magmatic compositions of the lavas. The ability to correlate lavas, determine their chemostratigraphy, and understand their petrogenesis relies on the accurate characterization of their magmatic chemistry.

The Columbia River Basalt Group (CRBG; northwestern USA) is the youngest and, due to its excellent exposure, preservation, and access, the most thoroughly studied continental flood basalt province on Earth. Despite being the smallest province in terms of erupted magma volume, the CRBG includes some of the most far-traveled lavas known. Several dozen lava flows (flow fields) span ~600 km from eastern Washington and Oregon or western Idaho to the Pacific Ocean. The CRBG also includes some of the largest eruptions of basaltic magma; some eruptions correspond to supereruptions (magnitude, M >8) (Self, 2006; Bryan et al., 2010).

The CRBG stratigraphy has been developed from numerous studies in which chemical composition, magnetic polarity, petrography, physical flow characteristics, and stratigraphic position have been used to distinguish individual flows or packages of compositionally similar flows (e.g., Wright et al., 1973; Swanson et al., 1979; Mangan et al., 1986; Reidel et al., 1989, 2013; Reidel and Tolan, 2013). Although nonchemical characteristics have been useful in distinguishing some post–Grande Ronde Basalt (GRB) lavas, field identification of GRB flows is problematic in that, with few exceptions, most are aphyric or rarely to sparsely porphyritic (see Reidel and Tolan, 2013, table 2 therein). Swanson et al. (1979) could reliably subdivide and map the GRB only on the basis of four magnetic polarity intervals. Mangan et al. (1986) and Reidel et al. (1989) later subdivided GRB lavas within the four polarity intervals on the basis of chemical differences and stratigraphic position, and Reidel and Tolan (2013) recognized several additional members within the GRB. Increased resolution of the GRB chemostratigraphy has corresponded to improvements in analytical precision of several generations of analytical instruments over the past four decades. Chemical criteria have therefore emerged as the defining criteria for characterizing GRB stratigraphic units.

The use of chemical criteria to discriminate among stratigraphic units is straightforward where the chemical differences between units exceed the variations within members. This applies to many post-GRB lavas (Wanapum Basalt and Saddle Mountains Basalt). Chemical differences between GRB members, as currently defined, vary from distinct to ambiguous (see Reidel and Tolan, 2013, figure 7 therein). For example, within the GRB R₂ polarity interval, the Meyer Ridge Member is readily distinguished from other R₂ flows by its markedly higher MgO and compatible trace elements (e.g., Cr). However, a TiO₂ threshold was adopted to distinguish the Grouse Creek and Wapshilla Ridge members among a compositional continuum among low-Mg R₂ flows (Reidel and Tolan, 2013). It is evident from the current chemostratigraphic criteria that the chemical variations within GRB members, and even between some members, are not yet understood, and that the distinguishing chemical identity of individual flows within GRB members, including the Sentinel Bluffs Member (SB), cannot be reliably determined. Accordingly, the correlation of GRB flows across the nearly 600 km extent of the GRB (Fig. 1) has mainly been limited to GRB members comprising multiple flows.

The problem of distinguishing individual flows within a GRB member is illustrated in Figure 2, which shows the TiO₂-MgO distribution of SB samples from earlier data sets. These analyses are from samples collected during the Basalt Waste Isolation Project (BWIP¹) (e.g., Landon and Long, 1989), geologic mapping studies (e.g., Hooper and Gillespie, 1996; Reidel, 1988), theses, and previously unpublished analyses, and are mainly from the compilation of Reidel and Valenta (2000). Because this compilation was titled preliminary, these data were reviewed to ensure that only SB lavas would be considered here. As described in Supplemental File 1², this review showed that some analyses from the older data set are not of SB lavas and these samples were excluded from consideration. These earlier SB data define a continuum of compositions, which, at higher MgO values, form a point cloud; with decreasing MgO, two diffuse, inversely correlated arrays are apparent. A key objective of this study is to assess whether this variation truly represents a continuum in SB magmatic compositions, or whether this variation results from postemplacement modification of a limited number of chemically distinct magmatic compositions. This question has implications beyond understanding the lava chemistry and chemostratigraphy of the SB. The stratigraphy of the entire GRB is based largely on chemical differences between lavas, mainly in TiO₂ and MgO abundances, in concert with magnetic polarity and stratigraphic position (see Reidel and Tolan, 2013, table 2 and figure 7 therein).

Due to their great expanse and rapid emplacement, individual GRB lavas can serve as strain markers in documenting regional structure, identifying faults and their displacements, and defining the geometries of CRBG-hosted aquifers. Regional aquifer systems hosted by GRB lavas (Kahle et al., 2011; Burns et al., 2011; Conlon et al., 2005) support agriculture province-wide, and supply the greater Portland-Vancouver metropolitan area. Previously considered as potential repositories for nuclear waste under the BWIP and predecessor studies conducted from 1968 to 1988 (Atlantic Richfield Hanford Company, 1976; Myers and Price, 1979, 1981; Landon and Long, 1989), GRB lavas have recently been evaluated as sites for CO₂ sequestration (McGrail et al., 2011, 2014; Zakharova et al., 2012). The precision in the use of GRB lavas as strain markers is dependent on the level of specificity in the identification of chemostratigraphic units.

Although CRBG lavas are commonly regarded as well preserved, several lines of evidence suggest that they have undergone water-rock interaction, which potentially can modify their chemical composition. Secondary minerals in lavas from surface exposures and the subsurface have been amply documented (Ames, 1980; Benson and Teague, 1982; Hearn et al., 1985). R.E. Evarts (2008, personal commun.) and Wells et al. (2009) recognized chemical modifications to GRB lavas, notably iron depletion, and applied an empirically derived, minimum FeO threshold to screen samples deemed weathered. It is important that they noted that many flows with a fresh appearance (e.g., dark gray color, unaltered plagioclase) were nonetheless chemically altered. Their observations corroborated earlier observations (Benson and Teague, 1982) made on samples obtained from deep in the subsurface. Given the indications of post-magmatic water-rock interaction in CRBG lavas, it is therefore important to establish a basis for evaluating whether lava compositions have maintained fidelity to their original magmatic compositions.

The late Pleistocene Missoula floods, which traversed the CRBG, removed much of the surficially weathered rock in areas where the floods were erosive. Examination of only flood-scoured exposures of the CRBG can lead to the misimpression that CRBG lavas were somehow resistant to weathering over the past 15–16 m.y. Surficial alteration (under oxic conditions) of CRBG lavas is widespread across the province and, except in those areas deeply scoured by Missoula floods, presents an impediment to sampling unweathered rock. In the western parts of the province (Coast Range and Willamette Valley), saprolite commonly is developed in CRBG lavas to depths of 5–10 m and, in places, to more than 30 m, where lavas may form laterite. In the eastern parts of the province, in the Columbia Basin and vicinity, thick weathering rinds (to 15 cm) are common in surface exposures of lava in those areas not scoured by the Missoula floods.

CRBG lavas generally have been considered to be chemically homogeneous (e.g., Wright et al., 1973; Hooper, 1984, 1988, 2000; Reidel et al., 1989; Tolan et al., 2009), although a few exceptions have been reported (Reidel and Fecht, 1987; Reidel, 1998, 2005; Vye-Brown et al., 2013a; Reidel and Tolan, 2013). Chemical variations within a flow or package of flows are also observed. For example, among stratigraphic units of the Saddle Mountains Basalt, variations in Ti and P abundances are commonly 5%–10% (e.g., Hooper, 2000, figure 3 therein), and such variations have gone unexplained. Intraflow chemical differences have been reported in several studies (Mangan et al., 1986; Reidel, 1998, 2005; Vye-Brown et al., 2013a; Reidel and Fecht, 1987). In accounting for such variations, several studies (Reidel and Fecht, 1987; Reidel, 1998, 2005) have appealed to mixing, distal to vents, of lava flows that were simultaneously erupted but independently sourced. Vye-Brown et al. (2013a) attributed intraflow variations to inflation of lava fed by chemically zoned magma.

Along with an updated assessment of SB geochemistry and chemostratigraphy, the development of a mass-based methodology is presented to account for changes in SB compositions resulting from alteration. This methodology, referred to as mass analysis, derives from quantitative relationships among immobile elements. This report focuses on the immobile element variations, using mass analysis, to define the magmatic chemical compositions and chemostratigraphy of SB lavas, and how these enable province-wide correlations of individual flows. Mobile element variations are discussed here only to a limited extent where they bear on the conditions of alteration or on criteria used for chemostratigraphic distinctions. The variations in mobile element abundances due to alteration are sufficiently complex that this topic will be addressed subsequently elsewhere.

Lavas composing the SB, the uppermost member of the GRB, are the subject of this geochemical study because of their widespread distribution (Fig. 1), well-exposed stratigraphic sections of multiple flows throughout much of the province, a relatively large range in composition, and large numbers of analyses available from prior studies noted here. The SB composes a significant part, nearly 5% by volume, of the entire CRBG, and SB lavas span much of the extent of the GRB (Reidel et al., 2013; Reidel and Tolan, 2013) (Fig. 1). SB lavas were erupted during the C5Cn.1n polarity chron (Jarboe et al., 2010), which according to Hilgen et al. (2012) spanned the time interval 16.27–15.97 Ma. Reidel and Tolan (2013) estimated that the SB comprises 11–15 flows or flow lobes. Given the estimated SB volume and number of flows, SB eruptions averaged ~680 km³.

Despite the accumulation of >1000 chemical analyses of SB lavas, the number of compositionally distinct flows remains uncertain. In their study of SB flows within the Pasco Basin and vicinity, Landon and Long (1989) recognized three flow packages based on chemical criteria, and a total of 12 flows distributed equally among the flow packages. Individual flows within the packages, however, were identified from their thicknesses, other physical characteristics, and paleomagnetic inclinations in multiple boreholes from within a relatively limited area compared to the entire extent of the SB. Among these 12 flows (recognized from contacts), some could possibly correspond to successive lobes of one lava flow or to successive lava flows within a flow field (for definitions of a flow lobe, lava flow, and flow field, see Thordarson and Self, 1998, table 1 therein).

Based on his observation of chemical differences within some flows, Reidel (2005) suggested that the SB stratigraphy is more complex than that of Landon and Long (1989). Relying mostly on TiO₂ and P₂O₅ abundances, he assigned flows, or intervals within flows, to one of six compositional types. Reidel’s (2005) chemical stratigraphy centered on a locally recognized flow within the Pasco Basin (informally named the Cohassett flow) that he proposed was formed from the incomplete mixing of four simultaneously erupted, but compositionally distinct, lava flows, each of which was designated a SB compositional type. Multiple SB lavas stratigraphically below the Cohassett flow were assigned to one compositional type and others above this flow were assigned to another. Reconciliation of the chemical groups defined here with earlier versions of SB stratigraphic nomenclature [i.e., informal flow names from the BWIP studies (Landon and Long, 1989), and the compositional types of Reidel (2005)] is beyond the scope of this report and will be presented elsewhere.

Samples of multiple SB lava flows were collected for this study from well-exposed sections on the Columbia Plateau and in the Columbia River Gorge, and include SB reference sections (Reidel and Tolan, 2013) at Sentinel Gap (Sentinel Bluffs) and at Devils Hole (Columbia Hills Water Gap section of Reidel, 2005; Fig. 1). Samples from the Coast Range are typically from isolated exposures lacking detailed stratigraphic context, usually from quarries or other artificial cuts where only one to several lavas are exposed.

For this study I analyzed 112 samples from 75 sites. Major and trace elements were determined by X-ray fluorescence (XRF) using a ThermoARL instrument at the Peter Hooper GeoAnalytical Lab, Washington State University, Pullman (herein, the GeoAnalytical Lab). Samples were analyzed from 2009 through 2013, except for one sample analyzed in 2007. Analyses, along with supporting documentation, are provided in Supplemental File 2³. Field and laboratory sampling protocols, descriptions of the analytical methods, and the accuracy and precision of analyses are provided in Supplemental File 3⁴.

Because chemical differences among SB lavas are small, a clear understanding of analytical precision is required to distinguish differences in magmatic compositions from those resulting solely from analytical uncertainty. Thus, an assessment of the analytical precision of the GeoAnalytical Lab XRF analyses of SB samples was carried out. This yielded a relationship that quantifies analytical precision as a function of element abundance, thereby enabling the precision to be closely estimated at any measured abundance of an element. Sources of intralaboratory bias due to changes in analytical strategy (e.g., background determinations) and grinding media over time are also described in Supplemental File 3 (see ^{footnote 4}).

At all sites, a primary sample internal to alteration rinds (inter-rind sample) was collected from minimally vesicular rock within the dense flow interior, or flow core (per Self et al., 1996, 1997; Vye-Brown et al., 2013b). Sampling was carried out to minimize the amounts of macroscopic secondary minerals, if present, within the analyzed material. The vesicular parts of lavas were thus avoided because secondary minerals are typically found within vesicles and crystal-bound voids. This has been shown in samples from surface exposures (Hearn et al., 1985), as well in samples from deep in the subsurface (see Zakharova et al., 2012, figure 2 therein). Sampling of the freshest possible flow cores restricts the effects of alteration to dissolution of solid phases and transport of their soluble constituents, such that any potential change in sample mass will be in one direction, negative. Weathered rock was avoided (except in one instance where strongly weathered lava was intentionally sampled), thereby limiting alteration (if any) to conditions that were anoxic or nearly so. As will be shown, these constraints imposed by careful sampling are critical to understanding the chemical effects of alteration. At several sites, unoxidized alteration rinds were also analyzed in addition to the primary sample from which alteration rinds were removed. The type of material analyzed is identified for each analysis in Supplemental File 2 (see ^{footnote 3}).

Samples from the Bingen, Butler Canyon, Devils Hole, Sentinel Gap, and Winter Water Creek sections were additionally subsampled in the lab to minimize pervasive, texturally distinct seams, referred to here as hydration seams that are described in Supplemental File 3 (see ^{footnote 4}). Subsamples of hydration seams were analyzed at 32 sites to test whether these represented incipient alteration. No attempt was made to minimize such seams in samples from the Armstrong Canyon and Patrick Grade sections, and from sites within the Coast Range and Willamette Valley. Subsample pairs of inter-seam rock and hydration seams showed minimal chemical differences, if any, between them. In part, the lack of clear differences between these pairs might be caused by the difficulty in separating inter-seam rock from the seams. Most inter-seam samples were not pure separates, but usually contained 30% or more of the seams. In any case, small chemical differences between hydration-seam and inter-seam rock are subordinate to the observed changes associated with bulk rock alteration, and the geochemistry of hydration seams will not be addressed further in this report.

Thus, at most sites one primary sample (±samples of hydration seams and/or alteration rind) was taken from one site within the dense flow core. Five samples were collected, however, from the core of one SB flow exposed in a small quarry located in the Coast Range along Salmon Creek near the city of Saint Helens, Oregon (Fig. 1). Among these samples, three were unoxidized rock from which alteration rinds were removed, one was an alteration rind, and one was severely weathered lava from the quarry margin. As shown here, the chemical variations among these samples provide important clues to interpreting the effects of alteration on lava chemistry, notably on immobile element abundances.

In this report the term lava flow or, more simply, flow, is used to refer to lava cooling units identified from physical contacts (lower and upper chilled vesicular crusts bounding a denser lava core). In order to distinguish between SB lava flows, flow lobes, and flow fields (per Thordarson and Self, 1998), precise characterizations of magmatic chemical compositions are required, which is a key objective of this study.

On the basis of the XRF analyses presented herein, SB lavas are here categorized into five chemical series, each of which comprises three or four chemical groups (Fig. 3). In concept, the series divisions are similar to the three-level separation of SB lavas by Landon and Long (1989). Each SB chemical group spans a limited range in composition and includes one or two flows showing only minor differences in composition. SB chemical series include successive chemical groups having similar immobile element ratios and/or immobile element abundances defining distinct chemical trends.

Primary chemostratigraphic distinctions between SB lavas were made using Ti/Zr and Sc/Cr ratios (Fig. 3). Several flow sequences (chemical series), and even several individual flows, are distinguished by these ratios. The earliest SB lavas (series I) are characterized by high Sc/Cr ratios compared to all other SB lavas, due mainly to their low Cr values, and these ratios differentiate lavas of SB groups 1 and 2 from those of group 3. Series V lavas, the youngest package of SB flows, are characterized by low Ti/Zr ratios (<65.5) relative to older SB lavas.

Three of the four series V chemical groups (groups 13, 15, and 16) have Ti/Zr and Sc/Cr ratios that are indistinguishable, indicating that these lavas are closely related in their petrogenesis. The group 14 lava has even lower Ti/Zr ratios than other series V lavas due to its higher Zr. Series IV lavas have somewhat lower Sc/Cr and Ti/Zr ratios that differentiate these lavas from series II and III, but these ratios cannot distinguish series II from series III lavas. These immobile element ratios, although successful in differentiating chemical series of lavas erupted sequentially, are only partially successful in differentiating individual flow compositions within the SB series. After describing a method for determining initial magmatic compositions of both fresh and altered lavas, I will return to the topic of differentiating individual SB flow compositions.

In order to identify magmatic compositions of altered SB lavas, a suite of samples is required that includes essentially unaltered samples spanning the entire range of compositions. Note that a sufficient number of minimally altered samples were readily obtained for this study using an appropriate sampling protocol (Supplemental File 3; see ^{footnote 4}). Selection of an inversely correlated immobile element pair (e.g., Al₂O₃, TiO₂) is also necessary to differentiate magmatic from alteration-derived trends, the latter forming linear trends with positive slopes in bivariate plots (Fig. 4).

The mass of altered rock decreases from dissolution of solid phases (minerals and glass) and transport of mobile (soluble) elements, and increases by the transfer of such elements from aqueous fluids, such as by precipitation of secondary minerals. Measured abundances of immobile (insoluble) elements therefore depend on the net change in rock mass, whether by depletion and/or enrichment of mobile elements. The immobile element abundances, however, have actually neither decreased nor increased, because low-temperature aqueous fluids do not transport these elements. Abundances of immobile elements vary proportionally to one another in response to rock mass changes, thereby preserving the ratios among them.

The mass analysis methodology presented here is based on the premise that Al₂O₃ and TiO₂ are enriched in equal proportions due to alteration-generated mass change, and that Al₂O₃/TiO₂ ratios remain constant throughout the alteration. Figure 4A presents Al₂O₃ⁿ and TiO₂ⁿ data (the superscript n herein indicates values from analyses normalized to 100% volatile free) for five samples of the same lava flow exposed in the quarry located in the Coast Range near Saint Helens, Oregon (Fig. 1). The lava sampled at the quarry margin, with the highest Al₂O₃ⁿ and TiO₂ⁿ, is severely weathered and is now composed largely, if not entirely, of clay and other secondary minerals. The other four samples are unoxidized; two of these are a rind–inter-rind pair. The Al₂O₃ⁿ and TiO₂ⁿ abundances of all of these samples are strongly correlated (r = 0.99), and plot along a line of constant Al₂O₃/TiO₂. As these elements are effectively insoluble in water, the nearly 2× difference between the lowest and highest abundances is attributed to a ~50% difference in sample mass.

Among the samples having lower Al₂O₃ⁿ and TiO₂ⁿ abundances, the magmatic composition cannot be identified from these variations alone. Additional criteria are needed to identify the point along the line of constant Al₂O₃/TiO₂ that corresponds to the magmatic composition from which immobile elements have increased or decreased. As shown in the following, the sample having the lowest Al₂O₃ⁿ and TiO₂ⁿ corresponds most closely to the original magmatic composition of the lava flow.

An expanded view of the Al-Ti variations for samples from the quarry near Saint Helens, Oregon (exclusive of the severely weathered sample) and for other samples assigned to chemical group 7 (Fig. 3) is shown in Figure 4B. These samples define two groups characterized by slightly different Al₂O₃ⁿ and TiO₂ⁿ: the samples with lower Al₂O₃/TiO₂ ratios define a line of constant Al₂O₃/TiO₂ (average 6.955) and are clearly from the same lava flow despite differences in Al₂O₃ⁿ and TiO₂ⁿ abundances. Four other samples assigned to group 7 have higher Al₂O₃/TiO₂ (average 7.128) and likely represent a different, compositionally distinct lava flow within this chemical group. Among the samples with lower Al₂O₃/TiO₂, the large variations in Al₂O₃ⁿ and TiO₂ⁿ at constant Al₂O₃/TiO₂ demonstrate that Al and Ti are immobile and are conserved during alteration; their range in abundances is attributed to changes in rock mass from changes in the abundances of soluble constituents.

In this study, inversely correlated Al₂O₃ and TiO₂ abundances are used to derive a parameter quantifying remaining rock mass, here termed mass index (MI). Al and Ti are typically immobile during low-temperature alteration and weathering, and the immobility of Al and Ti are demonstrated here by the variations within one lava flow, as shown (Fig. 4). Al-Ti variations for SB lavas sampled for this study (except for the one severely weathered sample) are shown in Figure 5A. Most samples define a prominent inverse correlation baseline; other samples having higher Al₂O₃ⁿ and TiO₂ⁿ values are scattered above this baseline. As shown herein, the seemingly random scatter in sample abundances above the baseline is the result of systematic chemical changes produced by varying degrees of alteration.

Note that no samples plot significantly below the baseline of Al₂O₃ⁿ and TiO₂ⁿ abundances. The absence of samples below the baseline arrays indicates that none of the samples had gained any appreciable mass from precipitation of secondary minerals or other secondary processes. As described here and detailed in Supplemental File 3 (see ^{footnote 4}), secondary minerals that could contribute to mass gain were carefully avoided during sampling.

The Al-Ti baseline is interpreted to be magmatic in origin, as inversely correlated immobile element abundances cannot be generated by alteration. Those samples along the baseline have the lowest observed Al₂O₃ⁿ and TiO₂ⁿ abundances at a given Al₂O₃/TiO₂ ratio, and these samples are interpreted as having retained essentially all of their original mass. Altered samples having elevated Al₂O₃ⁿ and TiO₂ⁿ and plotting above the baseline arrays (Fig. 5A) have therefore undergone mass loss due to water-rock interaction.

In order to quantify parent (magmatic) compositions, regressions were performed on those samples defining baseline arrays (solid black lines in Fig. 5A), and these compositions are defined to have a mass index of 100 (MI¹⁰⁰). It became apparent that series V samples (Fig. 3) do not share a baseline trend colinear with that of older SB lavas, so separate baseline regressions were made on series I–IV and series V samples.

The regressions to determine MI¹⁰⁰ baselines were calculated iteratively, starting with a visually estimated line at the base of the arrays, and applying an interval of MI units, or MI window, to constrain samples included in each regression. An MI window of 0.5 was used, meaning that MI values for the included samples are within 0.25 MI units of MI¹⁰⁰. A regression is calculated for samples within the initial MI window, and this regression is used as the input line for a new regression as the sample population within the MI window changes. The iterative regressions proceed until the input and output lines coincide. These calculations are illustrated in the spreadsheet provided as Supplemental File 4⁵, and the function of this spreadsheet and the iterative calculation process are described in Supplemental File 3 (see ^{footnote 4}).

The multiplication by 100 converts the sample mass fraction to percent. Note that this equation has the same form as Equation 1.

Samples from 14 of the 16 SB groups recognized in this study have MI^s values within the MI window. The maximum MI^s value among group 3 samples from this study is 98.5; however, an unpublished data set of SB samples provided by R. Evarts (2013, personal commun.) includes a high-MI^s (99.73) sample. Among group 11 samples, the maximum MI^s value (99.68) is slightly lower than the MI window minimum (99.75). This difference in MI^s corresponds to a difference in TiO₂ⁿ (0.001 wt%) that is within analytical precision (Supplemental File 3; see ^{footnote 4}). One sample has slightly higher MI^s (100.36) than the MI window maximum (100.25); however, the difference of 0.11 MI units corresponds to a small difference in TiO₂ (0.002 wt%) that also is within analytical precision.

Where MI¹⁰⁰ values are lacking for a chemical group or particular flow within a stratigraphically coherent sequence of lavas, MI^s values may be determined by assuming that the parent (magmatic) compositions for the lavas in question are represented along the Al-Ti baseline. This equates, in effect, to an inference that the negatively correlated baselines of compositionally similar lavas derive from a common magmatic process. The origin of the inverse Al-Ti correlation is discussed herein.

Mass-normalized abundances of immobile elements allow more precise, distinctions than those made using immobile element ratios. For the SB samples analyzed in this study, mass-normalized TiO₂ and Cr abundances (TiO₂^mn and Cr^mn, respectively) are consistent with, but allow more precise distinctions (Fig. 6) than, those made using Ti/Zr and Sc/Cr ratios (Fig. 3). Within four of the five chemical series, TiO₂^mn and Cr^mn are inversely correlated, similar to the Al₂O₃ⁿ–TiO₂ⁿ variations, but define trends offset from each other, mainly due to differences in Cr^mn abundances. Mass-normalized abundances (colored symbols, Fig. 6) and normalized abundances (black plus symbols, Fig. 6) are also compared in Figure 6 for samples having MI^s < 98.4. For a given sample, tie lines (blue, dotted, Fig. 6) connect normalized abundances with mass-normalized abundances. These tie lines correspond to residual concentration vectors, are along constant TiO₂/Cr ratios, and have positive slopes defined by magmatic TiO₂/Cr ratios. Differences in the tie-line slopes are caused mainly by a nearly 3× difference in Cr abundances.

For immobile elements, mass normalization enables the comparison of magmatic values for samples of fresh and variably altered lava, and prevents the misidentification of samples with low MI^s values. For example, the normalized TiO₂ and Cr abundances of a group 15 sample plot within the group 9 field (series III), but this lava has a low Ti/Zr ratio characteristic of series V lavas, which erupted later in the SB sequence (Fig. 6). Its mass-normalized TiO₂ and Cr abundances are consistent with the group 15 field. Notably, the mass-normalized TiO₂ and Cr abundances of a severely weathered (MI⁵³) group 7 sample plot among those for unaltered group 7 samples.

As shown in Figure 6, mass normalization eliminates variations in TiO₂ⁿ and Crⁿ abundances within a single chemical group (e.g., group 11) that might otherwise be interpreted as unique compositions of separate lava flows or as internal variations within a single lava flow. The mass-normalized TiO₂ and Cr abundances for group 11 samples are within analytical uncertainty, and these samples are from locations ~175 km apart, demonstrating the chemical homogeneity of this flow over large distances. Similarly, groups 13 and 15 each include samples from locations more than 250 km apart.

A step-by-step summary of key procedures in mass analysis, including the applicable equations, is presented in Figure 7, and steps mentioned in the following refer to the numbered steps in this figure. Initially (step 1), flows are identified that have equivalent immobile element ratios and equivalent stratigraphic positions. These constraints are used to define MI¹⁰⁰ baseline values (step 2); unaltered samples would ideally span the range of sample Al₂O₃/TiO₂ ratios. Definition of this line is the foundation for calculations of parent TiO₂ (step 3), sample mass index (step 4), and mass-normalized abundances (step 5). Characterization of chemostratigraphic units using mass-normalized immobile element abundances can then be made (step 6).

Sampling was designed to minimize oxidation, textural indications of alteration, and secondary minerals that can potentially contribute elemental additions. By avoiding rock altered under oxic conditions, the samples collected were composed primarily of either unaltered rock or rock altered under anoxic conditions. The resultant data set of chemical analyses for such samples mainly shows inverse baseline trends in Al-Ti variations. Earlier I noted that these trends cannot be generated by alteration, and therefore must be generated by magmatic processes. The origin of the SB Al-Ti baselines is briefly discussed.

Stepwise fractional crystallization (FC) experiments on tholeiitic basalt at 0.7 GPa (Villiger et al., 2007) show that, following initial enrichment of liquids in Al₂O₃ and TiO₂ (from crystallization and removal of mafic silicates), the withdrawal of subequal amounts of clinopyroxene (Cpx) and plagioclase (Pl), along with trace amounts of spinel, generates an inverse Al-Ti trend in liquid abundances (Fig. 8). This trend is linear (r = –0.997) over a large range of temperatures (90 °C) and amount of crystallization (84%). The trend’s linearity indicates that the crystal assemblage maintained a constant Al₂O₃/TiO₂ ratio despite changes in Cpx and Pl compositions with decreasing temperature. The 0.7 GPa pressure of these experiments corresponds to lower crustal depths in the eastern CRB. It is also comparable to the maximum pressure (0.66 GPa) determined from Cpx geobarometry in GRB and other CRBG lavas by Caprarelli and Reidel (2005), from which they also inferred lower crust crystallization of CRBG magmas.

The SB Al-Ti baseline trends, also shown in Figure 8, are subparallel to the trend of experimental basaltic liquids. The similarity of the SB and experimental trends, derived from Cpx + Pl crystallization, suggests their generation by the same process. The smaller range of Al-Ti compositions in SB lavas indicates that it can be generated by a much smaller amount of fractionation. Offset of the SB trend to lower Al₂O₃ and/or TiO₂ values could potentially result from a variety of factors such as differences in the parent magma composition and pressure conditions. In the context of magma evolution from a primary, mantle-derived melt, it is clear that the erupted SB magma compositions represent only a small fraction of their overall magmatic history.

Mass balance calculations relating major element abundances of the least and most evolved lavas within four of the SB series (I–III, V) by crystal removal yield results consistent with the FC experiments discussed here. These relatively simple mass balance models used SB augite compositions from Ames (1980), and allowed plagioclase compositions to vary between anorthite and albite. The mass balance models also allowed olivine, pigeonite, and spinel or magnetite to be part of the assemblage, but these phases did not appear in the solutions. The model plagioclase compositions are mainly within the range of analyzed Pl compositions, but in some models were slightly higher (to ~10% An content) than the analyzed Pl compositions. Models relating group 16 to other series V groups, however, yielded anomalously calcic plagioclase (An₁₀₀). The anorthite contents of model plagioclase compositions do not warrant precise interpretation because these values varied with the Cpx alumina content. For example, plagioclase compositions differed from An₇₂ to An₈₀ for equally valid models of series I compositions obtained with Cpx compositions having Al₂O₃ contents that differed by a factor of 2.

Successful mass balance models included assemblages of Cpx and Pl in subequal amounts. The amounts of intraseries fractionation indicated by the models are minor, and, for example, are 6%–10% for series I–III and V. The entire range of series I–IV compositions corresponds to ~15% fractionation if these compositions were generated from one parent magma, but it appears that each series evolved separately. The correspondence of model mineral compositions with those of SB groundmass minerals (except for Pl in series V models) is consistent with fractionation at crustal pressures. Although spinel was present in small amounts in the Villiger et al. (2007) experiments, the mass balance models indicate that spinel is not required to generate the SB Al-Ti array.

The removal of crystal assemblages containing ~50% Cpx, as indicated by both the mass balance models and FC experiments, is consistent with the SB intraseries Cr^mn variations (Fig. 6). Within all SB series except series IV, Cr^mn abundances decrease by ~20%–40% from the least to most evolved compositions. These Cr^mn variations are consistent with the strong partitioning of Cr into Cpx ⁠. The compositional gaps in Cr^mn variations between series (Fig. 6) suggest that fractionation of SB magmas occurred in batches (series), and not by differentiation of a single parent magma. In addition, Al₂O₃ and TiO₂ abundances of the SB series partially overlap, rather than occupying separate segments of the Al-Ti baseline (Fig. 9). This also indicates that the magmas of each SB series evolved separately, at least in their latest phase of differentiation.

Although the main SB compositional variations can be explained by relatively simple Cpx + Pl fractionation, some exceptions are evident. For example, group 14 compositions have anomalous enrichments in incompatible trace elements (e.g., Zr, K, Ba) relative to other series V lavas; the enrichment in Zr accounts for their distinctly low Ti/Zr ratio (Fig. 3). The similar enrichments in both immobile and mobile incompatible elements indicate a magmatic origin and require a more complex petrogenesis. In addition, as shown by Ti-Cr variations (Fig. 6), the small variations among series IV lavas do not define a coherent trend. Because these issues are peripheral to the main focus of this paper, they will not be examined further.

Despite their apparently separate evolution in magma batches, compositions of SB lavas nonetheless define a linear Al-Ti trend. This is attributed to the saturation of all SB magmas in Cpx and Pl, such that SB major element compositions are constrained to a boundary curve (cotectic) along which liquid compositions vary systematically, as observed in the FC experiments discussed above. It appears, therefore, that processes other than Cpx + Pl fractionation that might have influenced SB compositions to a minor extent did not cause SB major element abundances to depart from cotectic compositions.

Inverse Al-Ti variations are not unique to the SB. The Al-Ti arrays of GRB N₂ and R₂ members sampled in multiple GRB stratigraphic sections have slopes subparallel to those of the SB lavas, but are offset to higher or lower Ti values. The compositions of these upper GRB units are more evolved than the SB lavas and, as such, would be expected to be saturated in Cpx and Pl. Thus, Cpx + Pl fractionation apparently is a common process in the latter evolutionary stages of GRB and other CRBG magmas. The protracted Cpx + Pl fractionation observed at lower crustal pressures (Villiger et al., 2007) likely explains the common occurrence of inverse Al-Ti arrays among GRB lavas.

As discussed herein, positively correlated Al-Ti trends to above-baseline Al₂O₃ and TiO₂ abundances are attributed to the concentration of immobile elements in altered, mass-depleted rock. Positive Al-Ti trends can also be generated by crystal fractionation in magmatic systems with removal of an assemblage whose bulk composition lacks or contains only minor amounts of Al and Ti. This occurs when the assemblage is composed of mafic silicates (e.g., olivine, clinopyroxene, orthopyroxene), and Al-rich phases, such as plagioclase, are absent or compose only a minor amount of the assemblage. The effect of mafic silicate fractionation on liquid Al and Ti abundances is illustrated by the highest-temperature FC experiments of Villiger et al. (2007) discussed herein. As shown in Figure 8, removal of such crystal assemblages from the melt prior to the onset of plagioclase crystallization generates a positive Al-Ti trend in their experiments from 1270 to 1210 °C.

Considered in isolation, removal of mafic silicates may appear to be a viable explanation for the positive Al-Ti trends observed among SB compositions. The fractionation of mafic silicates would affect the abundances of other elements as well. In a magmatic system, mineral-melt partitioning controls element behavior, whereas in a low-temperature water-rock environment, element solubility is the dominant control. These markedly different processes can be clearly distinguished from the changes to lava chemistry. For some elements, the chemical changes are in the opposite sense.

The mafic silicates (olivine, orthopyroxene, and clinopyroxene) preferentially incorporate Mg over Fe relative to basaltic liquid (i.e., K_D(Fe-Mg)^min-liq ≤ 0.3; see Putirka, 2008). Crystallization and removal of any of these silicates therefore would yield liquids with progressively higher FeO/MgO. Along the positive Al-Ti trends in SB lavas, FeO/MgO decreases with increasing Al and Ti. This change in FeO/MgO is in the opposite sense to that expected from mafic silicate fractionation. The decrease in FeO/MgO results from the strong depletion of FeO due to alteration. This difference in the sign of changes in FeO/MgO is sufficient to unambiguously distinguish positive Al-Ti variations resulting from crystal fractionation from those generated by alteration under anoxic conditions.

Trace element behavior also differs sharply between magmatic and groundwater-rock systems. For example, if increased Al and Ti abundances were due to olivine fractionation, Ni abundances should be severely depleted in the fractionated magmas due to the strong partitioning of Ni into olivine. For SB lavas, which have low MgO (~4.4–5.2 wt%) in comparison to primary mantle-derived melts, is likely very high (> 20) given the relationship between and magma MgO content (Hart and Davis, 1978). In SB lavas, the positive above-baseline Al-Ti trends are not accompanied by strong depletions in Ni abundances.

As discussed here, fractionation of Cpx in subequal amounts with plagioclase is inferred from intraseries variations of SB lavas. Removal of Cpx, into which Cr is partitioned relative to the melt (i.e., ⁠), should significantly reduce Cr abundances. Systematic decreases in Cr of 20%–40% (mass normalized abundances, Cr^mn) are observed within 4 SB series (Fig. 6) and are consistent with Cpx fractionation as part of a Cpx + Pl assemblage. For above-baseline samples, normalized Cr abundances (Crⁿ) instead increase due to the concentration of insoluble Cr in mass-depleted rock (Fig. 6).

The increases in Cr abundances are proportional to increases in Al and Ti abundances, and to Zr, Nb, and V abundances. The equivalent sample:parent enrichments of elements that are both compatible (e.g., Cr, V) and incompatible (e.g., Zr, Nb) in terms of mineral-melt partitioning in a magmatic system are at odds with crystal fractionation of mafic silicates. The similar behavior of compatible and incompatible elements is instead attributed to their mutual insolubility, which produces proportional changes in their abundances due to alteration-generated changes in rock mass.

The chemical homogeneity of CRBG lavas across hundreds of kilometers has been a common inference of past geochemical studies, which allowed for some chemical variation within the individual flows (e.g., Wright et al., 1973; Reidel et al., 1989; Hooper, 1984, 1988, 2000; Tolan et al., 2009). For example, many compositionally distinct Saddle Mountains Basalt flows exhibit ranges in P₂O₅ and TiO₂ abundances that are positively correlated (see Hooper, 2000, figure 3 therein), and the varying slopes of these correlations are similar to the arrays of residual concentration trends due to mass loss (Figs. 4 and 6).

Reports of chemical heterogeneity within a single CRBG lava flow have been few, and such chemical differences have been attributed to differences in magmatic compositions, in combination with lava emplacement mechanisms (Reidel and Fecht, 1987; Reidel, 1998, 2005; Reidel and Tolan, 2013; Vye-Brown et al., 2013a). These studies, however, did not evaluate the possibility that secondary processes might be responsible for the chemical differences within flows, perhaps because of a lack of a framework in which to differentiate the effects of chemical alteration from those of magmatic processes.

In this study chemical differences within a flow are observed when analyzed abundances (normalized to 100% volatile free) are considered. Mass-normalized immobile element abundances show, however, that such intraflow differences are the result of alteration, and not of magmatic processes. With the modifications to lava chemistry from chemical alteration taken into account, it appears that SB lavas have a high degree of homogeneity that is indistinguishable from analytical uncertainty, or nearly so.

The precise SB chemostratigraphy developed here allows the correlation of lavas throughout the SB. This study shows that SB flows, which were sampled in different locations and in equivalent stratigraphic position, have compositions that are within narrow ranges of immobile element ratios and mass-normalized immobile element abundances. Samples for which stratigraphic context is only broadly constrained (i.e., samples from the Coast Range and Willamette Valley) also have immobile element characteristics that coincide with the fields defined by samples from continuously sampled sections of multiple SB flows in the Columbia Plateau and Columbia River Gorge.

Secondary mineral assemblages documented in CRBG lavas indicate that alteration occurred at low temperature (<100 °C) in the presence of groundwater, under conditions equivalent to the present-day environment (Benson and Teague, 1982). As shown in Figure 10, FeOⁿ abundances of SB lavas correlate to sample mass index (MI^s), indicating that depletion of iron is a significant part of the observed mass loss. In this plot, magmatic FeOⁿ abundances correspond to the values for high-MI (~100) samples (11.1–12.7 wt%) and, with decreasing mass index, FeOⁿ is progressively depleted. This relationship indicates that alteration occurred under anoxic conditions, thereby enabling transport of reduced iron (Fe²⁺) from the altered lava.

Because iron can occur as soluble Fe²⁺ or the nearly insoluble Fe³⁺, iron mobility depends on the groundwater’s redox potential, which is largely a function of dissolved oxygen. Dissolved oxygen is consumed by oxidation reactions with redox-active elements such as iron. SB lavas, as well as other CRBG lavas, have a large reservoir of reduced iron (>10 wt% FeO) ~85% of which (molar basis) is likely ferrous iron in unaltered rock. Consumption of dissolved oxygen in water from reaction with GRB lava (Winter Water Member) has been shown in the experiments of Lane et al. (1984). For example, in their experiments performed at 100 °C over 130 days the dissolved oxygen content of initially air-saturated, synthetic GRB groundwater (see Jones, 1982) decreased nearly 80%. Comparison of dissolved oxygen consumption for a GRB lava (Winter Water Member) and several minerals (from elsewhere) shows similarly high rates of oxygen uptake by basalt and magnetite as compared to the rates for augite and plagioclase (White et al., 1985). This suggests that Fe-Ti oxides, which are abundant in the groundmass of GRB lavas, play an important role in oxygen depletion and, therefore, in influencing the redox conditions of groundwater. To account for rapid oxygen diffusion rates and a lack of Fe on exterior mineral surfaces in laboratory experiments simulating alteration, White et al. (1985) further suggested that oxygen diffuses into interior pore spaces and along the boundaries of grains dominated by Fe-Ti oxides.

Marked increases in dissolved Fe²⁺ coincident with strong decreases in dissolved oxygen and Eh have been observed in the groundwater of aquifers with distance from recharge areas (e.g., Champ et al., 1979; Edmunds et al., 1987), as well as within a relatively shallow aquifer (sampled to 60 m depth) in which the flow direction is primarily vertical (White et al., 1990). In these studies, the observed decrease in Eh between oxic and anoxic waters is on the order of 400–500 mV, with oxygen-depleted waters commonly having negative Eh values. These studies were performed on aquifers hosted by sedimentary rocks or granite-derived alluvium having relatively low amounts of ferrous iron, but that could contain other reductants such as carbon in sedimentary rocks.

Iron mobility and immobility under anoxic and oxic conditions, respectively, have been shown in batch dissolution experiments performed on lava of the Wapshilla Ridge Member of the GRB (Neaman et al., 2005). The composition of the sample used in the experiments is broadly similar to SB lavas, but has ~0.5 wt% less MgO than the most evolved SB lavas. These experiments were carried out at a pH of ~6–6.5 (end-experiment pH) near the lower range of the modern circum-neutral to moderately alkaline GRB-hosted aquifers (pH 6.7–9.4; Steinkamph and Hearn, 1996). These experiments showed significant iron release under anoxic conditions in which redox potentials measured at the end of experiments were reducing (Eh < 100 mV). Iron release under oxic conditions at atmospheric pressure and Eh ~300–450 mV was not observed.

Secondary pyrite lining vesicles and crystal-bound (diktytaxitic) voids in GRB lavas has been observed in GRB lavas in outcrops in several field sections, and has been reported in fractures and vesicles within CRBG lavas in borehole core samples (Benson and Teague, 1982; McKinley et al., 2000). The pyrite occurrences are consistent with reducing anoxic conditions.

Reduced groundwaters (Eh <–200 mV) from deep modern CRBG-hosted aquifers contain anaerobic bacteria (e.g., methanogenic, sulfate reducing; Stevens et al., 1993). Stevens and McKinley (1995) further interpreted kerogenous bacteriomorphs associated with secondary minerals in CRBG lavas as evidence that microbial populations have existed in reducing groundwater of CRBG-hosted aquifers in the geologic past.

Overall various field and laboratory studies of water-rock interaction support the interpretation that Fe²⁺ is mobile under anoxic and reducing conditions. The depletion of FeO with increasing mass loss is in agreement with the observations of Wells et al. (2009), who attributed low FeOⁿ (<11 wt%) in GRB lavas to alteration. The strong iron depletion is also in agreement with petrographic observations of borehole samples by Benson and Teague (1982), who noted that Fe-rich minerals, clinopyroxene and magnetite, are altered to a greater degree than feldspar.

The behavior of iron under anoxic conditions contrasts markedly with its behavior during near-surface weathering, under oxic conditions, in which Fe³⁺ behaves as an immobile element and is concentrated in the weathered rock residue. For example, the severely weathered and mass-depleted (MI⁵³) group 7 sample analyzed in this study (see Fig. 4A) has FeOⁿ (21.3 wt%) that is nearly twice that of unaltered samples having MI^s values of ~100. Its mass-normalized iron abundance (11.3 wt%) suggests that only minor Fe loss occurred from alteration. Although total iron is discussed here in terms of FeO (Fe²⁺), the iron in this surficially weathered rock is undoubtedly oxidized to Fe³⁺.

Low analysis totals (herein, totals) of the abundances of nonvolatile elements are commonly interpreted as indicators of the degree of alteration. The difference between 100% and the analysis total is attributed mainly to glass hydration and/or the presence of hydrous secondary minerals. Totals for SB lavas do not correlate with MI^s, and nearly all samples, including those with MI^s values as low as 91, have totals >97 wt% (Fig. 11). This indicates that alteration resulting from dissolution of minerals ± glass under anoxic conditions is not accompanied by the hydration of glass in proportion to mass loss or by the formation of significant amounts of hydrous secondary minerals. As noted herein and described in Supplemental File 3 (see ^{footnote 4}), minor amounts of secondary minerals that may line the surfaces of or fill vesicles, fractures, and other cavities were avoided with careful sampling.

Mass-depleted rock that contains minimal amounts of hydrous secondary minerals and/or hydrated glass can have high totals in part because the depleted residue is analyzed as a unit mass. In effect, the depleted abundances are inherently normalized to 100 wt% in the process of analysis. Rock in which minerals and glass have been dissolved, and in which negligible amounts of hydrous secondary minerals have formed, can therefore yield analyses with totals approaching 100 wt%. The large range in the totals of high-MI^s samples (MI^s > 99) indicates that volatile acquisition, such as from glass hydration, may occur without accompanying mineral dissolution. Among group 11 samples, which span a 5% range in mass retention, totals are higher in the low-MI^s samples. This suggests a sequence of alteration in which glass was initially hydrated and later the hydrated glass was dissolved. In any case, for rock altered under anoxic conditions, analysis totals are not a reliable indicator of the amount of mass loss. The persistence of high totals for rock altered under anoxic conditions does not apply to rock altered under the oxic conditions of surficial weathering. For example, the strongly weathered sample from the quarry near Saint Helens, Oregon, has both very low MI^s (~53) and a low analysis total (of nonvolatile constituents) of only 74 wt%.

The occurrence of only minimal amounts of hydrous secondary minerals in mass-depleted rock altered under anoxic conditions perhaps results from a lack of available aluminum for clay and zeolite formation due to the preferential dissolution of ferromagnesian minerals having low Al₂O₃ contents. This is consistent with the petrographic observations of Benson and Teague (1982), who noted that clinopyroxene and magnetite appeared to be most affected by alteration, with plagioclase less so. This is also in agreement with the observations of Wells et al. (2009), who characterized plagioclase in rocks showing evidence of chemical alteration as having a fresh appearance.

The use of MgO variation diagrams was initially advocated by Wright et al. (1973), and TiO₂-MgO variations have continued to play an important role in differentiating GRB members (Mangan et al., 1986; Reidel et al., 1989; Reidel and Tolan, 2013). Published chemical characterizations of members in terms of TiO₂ and MgO abundances have been imprecise (being only greater or less than a given value), however, and the compositions of individual flows within GRB members have not been explicitly defined. Qualitative characterizations of GRB lavas in terms of MgO abundances (as low, intermediate, high, and very high) have since become commonplace.

The TiO₂-MgO variations from previously reported SB analyses define a seemingly random distribution at higher MgO that separate into two diffuse, inversely correlated trends at lower MgO (Fig. 2). Here I examine TiO₂-MgO variations in SB lavas from analyses of this study (Fig. 12) using a threshold of sample mass index (MI^s = 99) to differentiate samples that have lost <1% of their mass (if at all) from those samples that have lost 1%–9% of their mass due to alteration. A lower MI^s threshold (98.4) was applied to group 3 samples because samples with higher MI^s are unavailable.

The high-MI^s samples define two inversely correlated trends, one for series I–IV samples and the other for series V samples, although some series IV samples are between these trends (Fig. 12). Within each of series I–III, successive chemical groups show an increase in MgO with decreasing TiO₂. This progression from more evolved to less evolved compositions is comparable to that observed for mass-normalized Cr abundances (Fig. 6), and supports the interpretation that the MgO abundances of high-MI^s samples represent magmatic abundances. When observations are limited to samples that have retained nearly all of their original mass (using an MI threshold), the compositional trends observed for mobile elements are similar to those observed for immobile elements.

Due to the mobility of MgO, the low-MI^s samples are not displaced from high-MI^s samples of the same chemical group along lines of constant TiO₂-MgO ratios, as for immobile elements (Figs. 4–6). Some low-MI^s samples have compositions corresponding to those of high-MI^s samples of other chemical groups, or have compositions not represented by any high-MI^s samples, such as the samples with <~4.4 wt% MgO. Overall, the trends of low-MI^s samples within a chemical group have slightly lower slopes than the magmatic trends, thereby broadening the combined high-MI^s and low-MI^s sample trends, extending them to lower MgO values, and causing them to merge at higher MgO values. This pattern is similar to that observed in the previously published SB data (Fig. 2).

The older analyses of SB samples collected during the BWIP and subsequent investigations include many samples from deep boreholes, and this earlier data set provides an opportunity to examine whether alteration has generated mass depletion of SB lavas in groundwater environments far from the influences of near-surface weathering. Mass analysis was applied to these data in the same manner as to the data from this study.

The Al₂O₃-TiO₂ variations of the older SB data set are shown in Figure 13, in which symbols for data points are given different colors depending on whether samples were collected from the surface (blue) or the subsurface (red). Subsurface samples, identified from sample number suffixes indicating depth (Supplemental File 1; see ^{footnote 2}), come from drill core or rotary drill chips collected from boreholes in and around the Pasco Basin. Most borehole samples are from depths between 540 and 1120 m; the depth of the shallowest borehole sample is ~120 m. The remaining samples are assumed to have been collected at the surface. The MI¹⁰⁰ lines calculated for these data were defined to represent the baselines of the main body of samples, and the relatively few samples having lower Al₂O₃ and TiO₂ that were disregarded in this process are discussed in the following. As in this study, MI¹⁰⁰ baselines were calculated separately for series I–IV and for series V. Assignments of samples to series I–IV or to series V were made mainly on the basis of Ti/Zr ratios (see Fig. 3) (described further in Supplemental File 1; see ^{footnote 2}).

The Al₂O₃-TiO₂ data from these earlier analyses exhibit a pattern similar to that shown by the data of this study (Fig. 5A), i.e., a baseline having negative slope with scattered abundances plotting above it (Fig. 13). These sample populations differ from those of this study, however, in that scattered samples also plot below the main populations. The few outlier samples that plot below the main populations were ignored because these samples apparently have gained mass from alteration, as discussed in the following.

Both surface and subsurface samples show evidence of mass loss, and several surface samples have MI^s values <92 (Fig. 13). Subsurface samples generally have MI^s values ranging from ~100 to 96, similar to the values for most surface samples. These Al₂O₃-TiO₂ relationships demonstrate that mass loss has occurred in SB lavas buried deep in the subsurface. The decreases in mass are attributed to mineral ± glass dissolution from alteration.

A small number of samples from the earlier SB data set that plot below the MI¹⁰⁰ baselines indicate minor mass gains (to ~2.6 wt%). The sample with the highest MI^s has anomalously high CaO, at least 2.5 wt% higher than the CaO values of samples with MI^s ~100, and presumably contains secondary minerals with abundant calcium, such as calcite. The other analyses with anomalously high MI^s may likewise contain significant amounts of secondary minerals.

The anomalously high MI^s values for a minority of the previously published SB analyses likely represent a net mass change whereby the mass gain from precipitation of secondary minerals exceeded the mass loss (if any) from mineral dissolution. This underscores the importance of applying (and documenting) a sampling protocol in which secondary minerals are excluded, thereby limiting mass change to negative values. If visibly altered samples are analyzed for purposes other than investigating the magmatic chemistry of CRBG lavas, this should be made clear so that these data are not later misinterpreted as magmatic compositions.

Prior characterizations of the geochemistry of SB and other GRB stratigraphic units have been developed without consideration of the effects of low-temperature alteration on lava chemistry, and analyzed abundances of altered samples undoubtedly have often been incorrectly interpreted as being magmatic in origin. As shown here, TiO₂ and other immobile elements are concentrated in the remaining rock after soluble constituents, such as Fe and Mg, have been removed by mineral dissolution. Inverse correlations of TiO₂ and MgO can also result from magmatic processes (Fig. 12), but without application of mass analysis the magmatic compositions cannot be distinguished from those modified by alteration. The use of an immobile element (e.g., TiO₂ⁿ) as a function of a mobile element (e.g., MgOⁿ) therefore has limitations in defining the chemostratigraphy of SB and other GRB lavas.

The occurrence of alteration under anoxic conditions undoubtedly has contributed to the difficulty in recognizing this alteration using physical characteristics. The freshest-appearing altered rocks, showing evidence of chemical alteration by anoxic groundwater, are typically dark gray, similar to the color of truly unaltered rock. The abundances of secondary clay and zeolite minerals tend to be minimal, and hydration is limited, so that analytical totals (or loss on ignition) are not significantly affected. In a study characterizing the secondary minerals of CRBG lavas obtained from coring to depths >1300 m, Benson and Teague (1982) described the disparity between an apparent lack of visible alteration in hand sample and the pervasive alteration, mainly to clay and zeolite minerals; they concluded that fluids were able to migrate throughout the entire basalt thickness via fractures, microfractures, and micropore systems. This view is consistent with groundwater studies documenting regional subsurface flow throughout the CRBG lava pile (Burns et al., 2011, 2015; Kahle et al., 2011; Ely et al., 2014).

Although visible alteration of CRBG lavas has been reported in thin section (Ames, 1980; Benson and Teague, 1982; White et al., 1985; Caprarelli and Reidel, 2004), such observations may not detect all chemical alteration. Even for comparatively young lavas (younger than 200 ka) on Hawai`i, lavas appearing fresh in thin section are nonetheless chemically altered (Lipman et al., 1990). Ames (1980) also noted that alteration in CRBG lavas, apparent under reflected light, was not apparent in transmitted light petrography. In addition, thin sections of GRB lavas from fine-grained, irregularly fractured parts of the lava flow core typically show nearly opaque, cryptocrystalline masses, from which little mineralogical information can be ascertained.

The narrowly defined chemical groups identified in this study greatly increase the resolution of the SB chemostratigraphy. Due to the lack of recognition of the chemical effects of alteration in prior geochemical studies of SB lavas, lava compositions of unaltered and variably altered rock were compared to one another as though all compositions were magmatic in origin. This undoubtedly resulted in stratigraphic conflicts and seemingly continuous variations, and in a generalized SB chemostratigraphy of compositional types (Reidel, 2005). For example, the oldest SB compositional type (McCoy Canyon) of Reidel (2005) corresponds to groups 1–9 of this study, and his youngest type (Museum) corresponds to groups 13–16.

The ability to correct for or avoid the effects of chemical alteration reveals that the SB comprises a greater number of flows and indicates a higher frequency of eruptions within the SB than previously estimated. The 16 chemical groups recognized in this study indicate that the SB comprises at least this number of compositionally distinct lava flows. Small differences in immobile element ratios (Fig. 4B), and/or mass-normalized immobile element abundances, indicate that some SB groups comprise two flows. On this basis, the SB comprises an estimated 21 compositionally distinct flows.

The flows recognized here on the basis of chemical differences are distinct from flow lobes that are distinguished mainly by cooling surfaces and vesicle size-abundance distributions used to identify flow contacts. In the continuous sections sampled, a single, compositionally distinct SB chemical group may be represented by a vertical succession of two or three physically defined flows, which may or may not represent flow lobes or flows of the same eruption.

A prerequisite for assessing the chemostratigraphy of SB (and other CRBG) lavas, therefore, is the application of mass analysis to reverse the influences of secondary processes on lava chemistry. To demonstrate heterogeneity along a vertical chemical profile within a single lava flow, it is necessary to show differences in mass-normalized immobile element abundances, not simply differences in analyzed (normalized to 100%) immobile element abundances, which are sensitive to changes in rock mass.

A magmatic trend defined by unaltered samples, such as the SB inverse Al-Ti baseline, is not essential to the evaluation of magmatic intraflow variations from vertical sampling through a flow at a given location. In addition, a completely unaltered sample is not required for mass normalization because the objective would be to examine only relative differences within a flow, not to identify the original magmatic composition as in this study. Analyses from a vertical profile can be mass normalized to the sample having the lowest abundance of Ti or other immobile element. Tests of intraflow chemical heterogeneity therefore can be performed on a lava flow that is not part of a related sequence of compositions such as the SB.

Alteration of CRBG lavas in anoxic groundwater has occurred across the Columbia River flood basalt province. This alteration is subtle in that it is not accompanied by distinct visual cues, such as oxidation and the abundant secondary minerals that accompany surficial weathering under oxic conditions. The groundwater alteration, however, can only be identified by chemical criteria.

The primary effect of alteration under anoxic conditions is mineral dissolution and reduction in rock mass. Due to the inherent normalization resulting from the analysis of mass-depleted rock as a unit mass, immobile element abundances change in inverse proportion to the remaining mass. The magnitude of these changes can be comparable to or exceed the subtle magmatic variations within SB lavas as well as those within other GRB members.

Mass analysis allows the quantitative assessment of alteration-generated mass transfer caused by water-rock interaction. Mass analysis employs systematic immobile element variations (i.e., inverse Al-Ti arrays) among rocks that have retained their original mass (MI¹⁰⁰) to quantify the sample mass (MI^s) remaining after alteration-generated mass loss (if any). Mass normalization, the product of MI^s and a sample’s analyzed abundances (normalized to 100%), yields pre-alteration, magmatic immobile element abundances.

With high-precision data, mass analysis permits detailed chemical characterizations of individual lava flows from mass-normalized immobile element abundances, and these abundances are consistent with broader chemical distinctions made using immobile element ratios. Immobile element abundances corrected for mass loss indicate that individual lavas are remarkably homogeneous and can even be identified from rock that has undergone as much as ~50% mass loss. The use of mass-normalized immobile element abundances for chemical characterization of SB lavas eliminates conflicts between chemical characteristics and stratigraphic position that result when only analyzed abundances (normalized to 100%) are considered, thereby simplifying the chemostratigraphy. Individual SB flows can be correlated across the entire extent of the SB, which spans much of the GRB.

Prior interpretations of intraflow differences in the chemistry of CRBG lavas, mainly between the abundantly vesicular, pervasively fractured flow crusts and the dense flow core, have called upon lava emplacement mechanisms to combine magma fed from a heterogeneous magma body or from multiple, simultaneously erupted, chemically distinct but homogeneous magmas into a single lava flow. Such explanations have spawned from the recognition that CRBG lavas are inflated pahoehoe lavas. Given that the more porous and permeable parts of CRBG lavas currently host regional aquifers, there has been ample opportunity for water-rock interaction in these parts of flows in the subsurface over much of the ~16 m.y. since these lavas erupted. Commonly, the porous parts of lavas have elevated immobile element abundances compared to the dense flow core. Such intraflow chemical differences are therefore more likely due to alteration of chemically homogeneous lava flows that were fed by well-mixed magma reservoirs. Application of mass analysis is thus a prerequisite to assessing magmatic intraflow chemical variations in the CRBG.

Major and trace element variations of SB lavas indicate that several batches (series) of magmas evolved independently, at least in their later stages of differentiation. Although SB series evolved separately, their colinear Al-Ti variations and overlapping Al and Ti abundances are attributed to the saturation of all SB magmas in clinopyroxene and plagioclase, reflecting the adherence of liquid compositions to a cotectic. Similar inverse Al-Ti arrays in other GRB members suggest that Cpx + Pl fractionation is a common process that has generated the chemical variations within GRB members.

Prior interpretations of SB chemostratigraphy and lava heterogeneity have been based on the often incorrect assumption that analyzed abundances are equivalent to magmatic abundances. As a result, earlier interpretations of SB chemostratigraphy have been generalized because magmatic compositions could not be distinguished from those modified by alteration. In addition, those magmatic processes previously thought to have been involved in the genesis and emplacement of CRBG lavas warrant reevaluation. With the application of mass analysis, future geochemical studies of CRBG lavas have an opportunity to more precisely refine our understanding of this flood basalt province.

Jon Hagstrum and Russ Evarts provided invaluable detailed constructive reviews of earlier versions of the manuscript. Thoughtful reviews by Marie-Nöelle Guilbaud, Steve Self, and John Wolff, and the editorial work of Julie Roberge and Shan de Silva, contributed to significant improvements of the manuscript and are appreciated. Rick Conrey generously provided unpublished data used in determining analytical precision of analyses performed at the Peter Hooper GeoAnalytical Lab, Washington State University (Pullman) as well as information on analytical procedures. I am grateful to Russ Evarts, who graciously served as a sounding board in numerous discussions as the concepts presented in this paper were developed. Russ Evarts and Ray Wells provided samples and/or analyses from locations mainly in the Coast Range and Willamette Valley, and Jon Hagstrum provided core from paleomagnetic sampling sites in several flows from which geochemical samples were prepared. Field introductions to the Patrick Grade and Wallace Canyon sections by Steve Reidel are also appreciated.