Deducing Mineralogy of Serpentinized and Carbonated Ultramafic Rocks Using Physical Properties With Implications for Carbon Sequestration and Subduction Zone Dynamics

Serpentinization of ultramafic rocks is fundamental to modern plate tectonics and for volatile (re‐)cycling into the mantle and magmatic arcs. Serpentinites are also highly reactive with CO2 such that they are prime targets for carbon sequestration. Serpentinization and carbonation of ultramafic rocks results in changes in their physical properties such that they should be detectable using geophysical surveys; this could provide constraint on the reactivity of rocks without extensive sample characterization. We constrain the physio‐chemical relationships in altered ophiolitic ultramafic rocks using petrographic observations, major‐element chemistry, quantitative X‐ray diffraction, and physical properties on a suite of >400 samples from the Canadian Cordillera. Serpentinization results in a systematic decrease in density that reflects the increase in serpentine abundance and carbonation results in an increase in density, mostly reflecting the formation of magnesite; based on these data, we present two formulations for determining extent of serpentinization: one based on major‐element chemistry and the other on density. Magnetic susceptibility is variable during serpentinization; most harzburgitic samples show a 100‐fold increase in magnetic susceptibility, whereas most dunitic samples and a minor proportion of harzburgitic samples show very little change in magnetic susceptibility. We use quantitative mineralogy and physical properties of the samples to constrain a model for using density and magnetic susceptibility to approximate the mineralogy of ultramafic rock. Although further work is required to understand the role of remanence in applying these models to geophysical data, this presents an advancement and opportunity to prospect for the most reactive ultramafic rocks for carbon sequestration.

in the oceanic lithosphere (abyssal peridotites) occurs in the geologic record as the lowermost structural portion of ophiolitic sequences that have been obducted during orogeny (Bodinier & Godard, 2014;Juteau, 2003). Ophiolitic ultramafic rocks and their hydrated (serpentinites) and carbonated equivalents are observed in all major Phanerozoic orogenic belts (Bodinier & Godard, 2014 and references therein) and may host Ni-Fe-Cu-PGE mineralization and, thus, have historically been, and continue to be common targets for resource exploration (e.g., Britten, 2017;Hulbert, 1997). Recent work has demonstrated that (serpentinized) ultramafic rocks are particularly reactive with carbon dioxide to the extent that they are prime targets for carbon sequestration (e.g., Kelemen & Matter, 2009;Kelemen et al., 2018Kelemen et al., , 2020Mervine et al., 2018;Power et al., 2013;Snaebjörnsdóttir et al., 2020). Mineral carbonation may be achieved using either ex situ or in situ techniques; the former involves exposure of fine-grained material, such as mine tailings, to either atmospheric CO 2 or from a concentrated source, whereas the latter involves injecting CO 2 to a depth of >2 km. Ex situ techniques take advantage of the increased reactive surface area of tailings and the high reactivity of minerals such as brucite (e.g., Power et al., 2013Power et al., , 2018Power et al., , 2020Vanderzee et al., 2019), whereas in situ techniques exploit the elevated temperatures and pressures at depth to accelerate reactions and fully carbonate the rock (e.g., Kelemen & Matter, 2009;Kelemen et al., 2018). As resource extraction invariably requires the consumption of energy and emission of CO 2 , whether it be for transportation of materials and personnel or for ore processing/extraction, any deposit hosted in ultramafic rocks should explore their potential to sequester CO 2 and so offset or prevent emissions and promote an environmentally sustainable mining industry (e.g., Hamilton et al., 2020;Power et al., 2013Power et al., , 2018Power et al., , 2020Wilson et al., 2014).
Highly serpentinized ultramafic rocks are particularly susceptible to carbonation by CO 2 -bearing fluids and follow three simplified reactions (e.g., Hansen et al., 2005): Carbonation is also commonly considered to behave isochemically with respect to the major-element cations (Falk & Kelemen, 2015;Hansen et al., 2005;Menzel et al., 2018), resulting primarily in the formation of (hydro-)magnesite (e.g., Hansen et al., 2005), except for at high degrees of alteration in which some Ca may be added to the system (e.g., Bideau et al., 1991;Falk & Kelemen, 2015). These carbonation reactions serve as a proxy for the reactions that are commonly used in experimental carbon sequestration models (e.g., Power et al., 2020).
The physical properties associated with the serpentinization and carbonation of ultramafic rocks have been studied for decades in various capacities (e.g., Bina & Henry, 1990;Burch, 1968;Li et al., 2020;Maffione et al., 2014;Saad, 1969;Toft et al., 1990;Whiteford & Lumb, 1975). However, most studies have focused on one or the other process (cf. Hansen et al., 2005;Toft et al., 1990) and typically focus on the behavior of the volumetrically dominant harzburgitic rocks (e.g., Maffione et al., 2014;Toft et al., 1990). Additionally, most studies focus on either the geochemistry or physical properties of serpentinites. In this contribution, we use >400 samples of variably serpentinized and carbonated ophiolitic ultramafic rocks collected from the Cache Creek/Atlin terrane in the western Canadian Cordillera ( Figure 1) and combine petrographic observations, major-element chemistry, estimates of mineral abundance through quantitative X-ray diffraction (qXRD) and thermogravimetric analysis (TGA), and physical properties (magnetic susceptibility, density, porosity, natural remanent magnetization [NRM]) to constrain the relationships and changes in physical properties that occur during alteration and to determine if and how these may vary as a function of location, protolith, and degree of alteration. Using these results, we present several formulations for determining the %serpentinization of a given sample using chemistry and physical properties and we present a model that uses the physical properties to estimate-at a first order-the mineralogical variability in ophiolitic ultramafic rocks. Due to the carbonation potential of highly serpentinized rocks, we suggest applications of our results to geophysical survey analysis with the goal of remotely identifying the most prospective ultramafic rocks for carbon sequestration.

Study Sites
The 441 samples in this study were collected from the western Canadian Cordillera by various research groups (e.g., Hansen et al., 2004Hansen et al., , 2005McGoldrick et al., 2017McGoldrick et al., , 2018Milidragovic & Grundy, 2019;Steinthorsdottir et al., 2020;Zagorevski et al., 2018 and references therein) from the traditional territories of the Taku River Klingit, Kaska Dena, Tāłtān Konelīne, Carcross/Tagish, Teslin Tlingit, Tl'azt'en, Binche Whut'en, Yekooche, and Takla First Nations. All localities comprise rocks that are part of the Atlin (equivalent to the undivided Cache Creek in southern B.C.) terrane (Figure 1), which represents Middle Permian to Middle Triassic discontinuous, dismembered ophiolitic massifs that may have formed either as ocean core complexes or in supra-subduction zone settings . The samples were not subjected to greater than greenschist-facies conditions post-emplacement Milidragovic & Grundy, 2019;Steinthorsdottir et al., 2020;Zagorevski et al., 2018) and, thus, should preserve their serpentinization-and carbonation-related physical properties (Shive et al., 1988). The northern segment is subdivided into the ophiolitic Atlin terrane and sedimentary overlap assemblages of the Cache Creek complex, whereas the southern segment remains undivided . The sample set is dominated by rocks from the Decar, Atlin, and Nahlin areas with complementary samples from King Mountain, Hogem, and South Yukon. A summary of the sample suites and data sets used as part of this study are provided in Table S1 and all sample location information and data are given in Table S2.

Materials and Methods
We mostly relied on published major-element chemistry data sets (Hansen, 2005;McGoldrick et al., 2017McGoldrick et al., , 2018Milidragovic & Grundy, 2019;Steinthorsdottir, 2021;Zagorevski, 2020). Samples from the Hogem area did not have corresponding published major-element chemistry and these were sent to Geoscience Laboratories for lithogeochemical analysis using the XRF-M01 and IRC-100 packages (https:// www.mndm.gov.on.ca/sites/default/files/2021_geo_labs_brochure.pdf). Samples that did not have published CO 2 estimates were analyzed for their total inorganic carbon at the Department of Earth, Ocean and Atmospheric Sciences, University of British Columbia (EOAS-UBC) using a UIC Inc. CM5130 acidification module and a CM5014 carbon dioxide coulometer.
Quantitative mineralogy for all whole-rock powders was determined by qXRD at the Electron Microbeam and X-ray Diffraction Facility, EOAS-UBC; Atlin samples of Hansen (2005) were analyzed in 2004, whereas all other samples were analyzed in 2019-2020; for all instrumentation details and running conditions, see Table S3. Qualitative results for Atlin samples are reported in Hansen (2005) and quantitative estimates for the samples were then determined as part of this study. As qXRD estimates for minerals occurring in low abundance (∼1%) are relatively uncertain, brucite abundances from Atlin and Decar samples were further characterized by TGA. The robust correlations between the qXRD and TGA data sets from Atlin and Decar was used to correct the qXRD results, for which there are a greater abundance of data across the sample set ( Figure S1). The correlations between qXRD and TGA data from the two localities differed, which is expected given the different instrumentation and analytical settings; qXRD of Atlin samples were corrected using their respective correction factor, whereas the qXRD of all other samples were corrected according to the Decar correction factor. All weight% (wt%) qXRD results have been converted to volume% (vol%) abundances using typical densities for each mineral as this more directly relates to the physical properties.
The magnetic susceptibility and density of all samples were first measured at EOAS-UBC. Magnetic susceptibility (κ) was measured using a ZH Instruments SM30 and applying thickness and demagnetization corrections (Table S4 and Figure S2). The reported values reflect the mean of five measurements and the uncertainty reflects the 2 SD. Densities were determined using the masses of samples in air and in water using specific gravity mode of an A&D Company Ltd. EJ-6100 balance. To assess the accuracy of our magnetic susceptibility and density measurements, and to provide additional information on the porosity and NRM, a subset of samples that reflect the full diversity of rock types and degrees of alteration were sent to the Paleomagnetism and Petrophysics Laboratory (PPL) at the Geological Survey of Canada-Pacific (GSC-P) in Sidney, British Columbia; all reported porosity and NRM values reflect analyses done at the PPL. Measurements at PPL were done on a 2.5 cm diameter core of unweathered material using the methodology and instrumentation as described in Enkin et al. (2020). Comparison of physical property results from EOAS-UBC and GSC-P show a near 1:1 linear relationship with an R 2 of 0.92 for density ( Figure S3a) and a near-linear relationship of log (magnetic susceptibility) with an R 2 of 0.97. Both data sets show slight offset to higher values for the GSC data ( Figure S3b) and, as such, EOAS-UBC measurements were corrected based on the relationships of the two data sets. The final physical property values that are reported reflect GSC-P physical property measurements when available, otherwise, the corrected EOAS-UBC were used.

Petrographic Observations
Relatively fresh dunitic rocks consist dominantly of 1-6-mm-sized equigranular olivine and primary spinel ( Figure 2a). Serpentine in relatively fresh samples is primarily localized along grain boundaries and fine intragrain fractures (Figure 2a). With increasing degrees of serpentinization, olivine grains are cut by serpentine veins of increasing thickness ( Figure 2d) until they are fully replaced by mesh texture (Figure 2g); serpentine occurring as massive veins and as mesh texture are likely lizardite (e.g., Viti, 2010;Wicks & Whittaker, 1977). Brucite increases in abundance with increasing degrees of serpentinization; in fresher rocks, it occurs in close association with relict olivine grains, whereas in highly serpentinized rocks it occurs as discrete or aggregate grains intergrown with mesh-serpentine or in veins with magnetite. Magnetite is typically absent in the freshest dunitic rocks.
Fresh harzburgitic/lherzolitic rocks occur in all study areas and consist of anhedral orthopyroxene (2-6-mmsized) and clinopyroxene (0.5-3-mm-sized), anhedral to euhedral olivine, and subhedral to euhedral equant Cr-spinel ( Figure 2b) (McGoldrick et al., 2018;Milidragovic & Grundy, 2019). Serpentinization in the relatively fresh samples affects olivine almost exclusively, with orthopyroxene, clinopyroxene, and spinel remaining relatively unchanged with the exception of minor magnetite occurring along grain boundaries (Figures 2b and 2e) and minor orthopyroxene alteration to bastite (McGoldrick et al., 2018). Early serpentinization of olivine in harzburgitic/lherzolitic rocks is similar to dunitic rocks in that it is primarily restricted to cross-cutting lizardite veins ( Figure 2b) that increase in thickness with progressive serpentinization. In the more pervasively serpentinized harzburgitic/lherzolitic rocks, olivine and pyroxene are progressively pseudomorphed by lizardite or antigorite, which occurs as blocky to tabular interlocking crystals in the matrix, as anastomosing vein networks, and as bastite ( Figure 2h). Although the textures are typical of lizardite and antigorite, the specific serpentine variety was not ascertained. Some pervasively serpentinized samples also exhibit evidence for a third stage of serpentinization involving the formation of chrysotile in veins (e.g., Steinthorsdottir, 2021). Brucite in harzburgitic/lherzolitic rocks that are >60 %serpentinized occurs in close association with olivine, as grains in the lizardite groundmass, or as aggregates or veins spatially associated with serpentine and/or magnetite ( Figure 2h). Rocks in which antigorite dominates rarely contain brucite.
Magnetite occurs both as a pseudomorph of primary spinel and as newly formed grains in the serpentine matrix, as veinlets along former grain boundaries, as larger cross-cutting veins and, locally intergrown with awaruite and sulphides (Britten, 2017;Milidragovic & Grundy, 2019); magnetite clearly increases in grain size and abundance with greater degree of serpentinization ( Figure 2h).
Pyroxene-rich rocks, which include olivine-websterites and websterites/pyroxenites (<10% olivine) are typically relatively fresh (Figure 2c), with hydration resulting mainly in the replacement of any olivine that is present by serpentine (Figures 2f and 2i). As with the other ultramafic protoliths, early serpentinization is localized along olivine intragrain fractures with progressive serpentinization fully replacing the grains. Brucite is typically not observed in pyroxene-rich lithologies and magnetite occurs in relatively low abundances and is restricted to areas in which olivine has been serpentinized. Pyroxene-rich rocks commonly contain talc, chlorite, and/or tremolite.
Early carbonation consists of thin magnesite-calcite veins cross-cutting serpentine veins, as small grains within a matrix of interlocking antigorite, or within bastite grains. Such rocks are termed ophicarbonates and are characterized by mineral assemblages dominated by serpentine with minor magnesite and talc (<20 vol% each) ( Figure 3a). Progressive carbonation results in the formation of soapstone, which is characterized by a mineralogy dominated by magnesite, with >10 vol% talc, <10 vol% quartz (Figure 3b), and the gradual disappearance of serpentine from the assemblage. The most highly carbonated rocks are termed listwanite (also referred in the literature to as listvenite, listvanite, or listwaenite: e.g., Menzel et al., 2018) and these consist primarily of magnesite and >10 vol% quartz with minor talc, and local fuchsite (Cr-mica) ( Figure 3c). Brucite in carbonated rocks is restricted to uncarbonated sub-domains in a relatively minor number of ophicarbonates. Magnetite is restricted to relict serpentinite sub-domains in ophicarbonate and soapstone.

Protolith Discrimination and Changes in Volatile Content
The changes in major-element composition during the alteration of ultramafic rocks is generally thought to result from passive dilution by volatiles (e.g., Deschamps et al., 2013;Hansen et al., 2005;Niu, 2004). In such a system, the molar ratio [(Mg + Fe tot + Ca)/Si] (herein referred to as the OPE: olivine-pyroxene elemental ratio: Steinthorsdottir, 2021) of the samples should provide a reasonable approximation of the olivine/pyroxene in the protolith with OPE >1.93 = opx-poor dunite (95-100 wt% olivine), OPE 1.42-1.93 = opx-rich dunite (90-95 wt% olivine) and harzburgite/ lherzolite/wehrlite (50-90 wt% olivine), OPE 1.15-1.42 = olivine websterite (10-50 wt% olivine), and OPE <1.15 = websterite/pyroxenite (<10 wt% olivine); opx-rich dunite was combined with harzburgites as there were not many samples in this category. At a given OPE (isolines in Figure 4a), the samples show a systematic increase in loss on ignition (LOI) with decreasing major-element cation concentrations, which would support the inference that any changes in major-element chemistry are due to passive volatile dilution. Using the OPE as a protolith discriminant, most samples have OPE between 1.52 and 1.87, which corresponds to harzburgitic/lherzolitic protoliths ( Figure 4a). For uncarbonated samples (<1 wt% CO 2 ), these geochemical discriminations are consistent with petrological and textural observations and agree well with various normalizations schemes ( Figure S4). The same approach could not be used for carbonated samples due to the extensive overprinting of primary textures; instead, these are subdivided on the basis of their mineral assemblages as outlined in Section 3.1.
Uncarbonated samples are identified here as having <1 wt% CO 2 and have LOI of up to 14.5 wt%, reflecting the progressive incorporation of H 2 O during serpentinization (R1: Figure 4b). For carbonated samples, changes in their volatile composition involve (a) an increase in CO 2 and decrease in H 2 O contents along a linear trajectory (R2-R4: Figure 4b) from ∼12 wt% H 2 O and <1 wt% CO 2 to 0 wt% H 2 O and 28.99 wt% Although some samples show small deviations from the expected volatile regression, significant loss or gain of the principal cation constituents would have resulted in deviations from such a trend and more scatter in the regression. This provides additional evidence that the systems appear to have behaved broadly isochemically with respect to the major elements and involved mainly passive dilution due to the addition of volatiles. The continued gain in CO 2 in H 2 O-free samples likely reflects some nonisochemical behavior given that many of these samples also plot to the left of the OPE = 2.00 isoline ( Figure 2a); this may be in the form of Mg, Ca, or Fe gain or Si loss (e.g., Back et al., 2004;Malvoisin, 2015;Paulick et al., 2006;Snow & Dick, 1995).

Determining %Serpentinization From Major-Element Chemistry
To calibrate LOI as a proxy for degree of serpentinization, the LOI of uncarbonated samples were normalized to their OPE and this was compared to the vol% of nonrelict minerals as determined by qXRD ( Figure 5). This normalization accounts for the tendency for olivine-rich samples to yield  greater abundances of brucite (e.g., , which would have a higher bulk-rock LOI at a given %serpentinization due to the greater water content of brucite relative to serpentine (e.g., D'Antonio & Kristensen, 2004). A regression of the data yields a linear relationship (R 2 = 0.91) for determining %serpentinization: where LOI is in wt%, OPE is the olivine-pyroxene elemental ratio, and %serpentinization refers to 100%-(vol% olivine + pyroxene); uncertainties were determined by quadratic addition of slope and intercept uncertainties. We note that qXRD results have typical relative errors of ±10%-15% Wilson et al., 2009) although these effects may be relatively minimal as the data are remarkably consistent despite having been analyzed on two different instruments using different analytical procedures and 15 years apart. Additional uncertainty in this formulation may be introduced due to the potential occurrence of primary spinel, secondary metamorphic olivine, or dehydration due to subsequent metamorphism; however, this too we consider to have minimal effect as primary spinel is typical at abundances of <2 vol% and metamorphic olivine is relatively uncommon in the studied samples. From here on, the %serpentinization of any uncarbonated samples use the value determined by Equation 1.

Changes in Mineralogy and Physical Properties During Serpentinization
Weakly serpentinized (<25 %serpentinized) samples exclusively have OPE <1.42 (Figures 6 and 7) requiring significant pyroxene content; this would dilute the amount of olivine available for serpentinization and is consistent with the higher resistance of pyroxene-particularly clinopyroxene-to hydration relative to olivine. Moderately to pervasively serpentinized (>25 %serpentinized) samples typically have OPE >1.42 and CaO <1 wt%, indicating that they are dominated by olivine and orthopyroxene. For such rocks, the increasing serpentinization (as indicated by LOI) is also reflected in the quantitative estimates of decreasing relict mineral (olivine + pyroxene) and increasing serpentine abundances (Figure 6a). Brucite and total spinel (i.e., spinel + magnetite) abundances are variable at a given LOI (0-10 vol% and 0-6 vol%, respectively); however, they occur in rocks with as low as 3-4 wt% LOI (30 %serpentinized) and generally increase in abundance with increasing LOI (Figure 6a-inset). Brucite abundances increase for samples with higher OPE (Figure 7a). Spinel abundances also increase for samples with higher OPE but peak at values of ∼1.65 after which abundances decrease; above OPE of 1.93 (>95 vol% olivine) spinel abundances sharply decrease (Figure 7b). Rocks with very high brucite contents (>4 vol%) typically have low spinel contents (<2 vol%) (Figure 7).
Samples typically have porosities of <1% regardless of the degree of serpentinization and OPE, with the exception of samples with an intermediate OPE of 1.42-1.63 (equivalent to 40-60 vol% olivine) and LOI >10 wt%, which have porosities of up to 3.25% (Figure 6c). Density and brucite contents linearly correlate only for samples with OPE >1.93 (>95 vol% olivine) and appears locality specific: samples from Atlin and Nahlin, which occur immediately along strike of each other  show lower brucite contents for a given density compared to the other localities (Figure 9b).
At a given LOI, magnetic susceptibility varies by over two orders of magnitude. Rocks that are <25 %serpentinized have magnetic susceptibilities <3 × 10 −3 SI for ( Figure 8) where %serpentinization is defined by Equation 1, and E  is magnetic susceptibility in SI × 10 −3 . Although this formulation is by no means robust, the data show that magnetic susceptibilities of >20 SI × 10 −3 rarely occur in samples that are <60 %serpentinized with the exception of those with OPE >1.93. Magnetic susceptibility correlates strongly with spinel contents for the various OPE intervals (R 2 > 0.85: Figure 9d).
The samples show a wide range of NRM values, spanning ∼4 orders of magnitude and these broadly increase with increasing serpentinization (Figures 6e and 8c). NRM does not appear to vary systematically as a function of OPE except for samples that are >75 %serpentinized for which it generally decreases with increasing olivine content. The samples hold relatively high Koenigsberger ratios (Q-ratio) (Figure 8d) with 48% having values greater than 1. Samples that are <50 %serpentinized show a wide range in values that do not vary systematically with increases in %serpentinization or OPE (Figure 8). For rocks that are >50 %serpentinized, there is significant overlap in the Q-ratios across the various OPE categories; however, in general, for rocks that are 50-75 %serpentinized, the median Q-ratio values broadly increase with increasing olivine content, whereas for rocks that are >75 %serpentinized, median values show the reverse relationship (Figure 8d).

Changes in Mineralogy and Physical Properties During Carbonation
The first phase of carbonation (R2) involves a gradual decrease in serpentine and increase in magnesite abundance, to form ophicarbonate rocks. This is reflected in samples with LOI as low as 8-10 wt% but is most common beginning at ∼12-13 wt% LOI; ophicarbonate rocks have LOI up to ∼20 wt% (Figure 6a). The first stage of carbonation is also associated with a disappearance of brucite from the assemblage (Figure 6a-inset). The physical properties of ophicarbonate samples are indistinguishable from those of highly serpentinized samples with OPE 1.42-1.93 (Figure 8).
The second phase of carbonation (R3) is identified by a sharp drop in serpentine abundance concomitant with a more pronounced increase in magnesite and talc abundance to form soapstone (Figure 6a). The median density of soapstone is higher than that of ophicarbonate at ∼2.86 g/cm 3 (Figures 6b and 8a) and  The final stage of carbonation (R4) involves the final consumption of serpentine-group minerals, a continued increase in magnesite abundance, a decrease in talc content, and an increase in quartz abundance to form listwanite (Figure 6a). Such rocks may also locally contain the Cr-mica, fuchsite. Relative to soapstone, listwanite samples show a further increase in the median density to ∼2.91 g/cm 3 (Figures 6b and 8a), have variable porosity (up to 2.5%: Figure 6c), and show a continued decrease in median magnetic susceptibility and NRM to <10 SI × 10 −3 (Figures 6d and 8b) and <1 A/m (Figures 6e and 8c), respectively. The Q-ratios for listwanite samples are typically <1, which is similar to that of soapstones (Figure 8d).  Figure 6 with the exception of the blue boxes, which represent all samples with OPE <1.42. Colors for carbonated samples also reflects the symbology shown in Figure 6.

Co-Variation of Density and Magnetic Susceptibility
On the Henkel plot (density vs. log magnetic susceptibility) the serpentinite samples follow two main trajectories ( Figure 10a). The first (T1: Figure 10a) involves a decrease in density and increase in magnetic susceptibility with serpentinization from ∼3.2 g/cm 3 and 0.3 SI × 10 −3 converging to values of 2.6-2.7 g/cm 3 and 100 SI × 10 −3 ; this is followed by most harzburgitic samples and a minor proportion of dunitic samples. The second trajectory (T2: Figure 10a) involves decreasing density with only slight increases in magnetic susceptibility from ∼3.2 g/cm 3 and 0.3 SI × 10 −3 converging to values of 2.6-2.7 g/cm 3 and 0.5-20 SI × 10 −3 ; this is commonly observed for dunitic samples and a relatively minor number of harzburgitic samples. The high susceptibility trajectory is widely reported (e.g., Li et al., 2020;Maffione et al., 2014;Toft et al., 1990); however, the low susceptibility trajectory, although proposed by Toft et al. (1990) on the basis of typical serpentinization reactions; is rarely observed and documented. Toft et al. (1990) proposed that harzburgitic samples would follow the first trajectory regardless of the volumetric proportions of olivine and orthopyroxene (e.g., Equations 25-32 in Figure 4D of Toft et al., 1990) and this is clearly reflected in our data set that show samples with variable OPE converging to a relatively consistent point. The first trajectory also appears to reflect the majority of samples globally (e.g., Li et al., 2020;Maffione et al., 2014;Oufi et al., 2002;Toft et al., 1990). The second low susceptibility trajectory, although not commonly observed or documented, has been proposed to occur for both dunitic and harzburgitic rocks in a high fluid/rock environment (e.g., Equations 19-21 in Figure 4C of Toft et al., 1990).
For carbonated rocks (Figure 10b), the physical properties of ophicarbonate rocks are identical to the majority of highly serpentinized samples (high magnetic susceptibility and low density). The most prominent change occurs with subsequent carbonation of ophicarbonate samples and is reflected in increasing density and decreasing magnetic susceptibility observed in soapstone and listwanite samples; the main difference between soapstone and listwanite samples is in a slight increase in density.

Magnetic Susceptibility Alone Is Not an Accurate Predictor of Extent of Serpentinization
Most prior studies investigating the physical property changes occurring during serpentinization have been focused on rocks with harzburgitic protoliths (e.g., Maffione et al., 2014;Toft et al., 1990) as they volumetrically comprise the bulk of ophiolitic complexes (Bodinier & Godard, 2014). Our data on rocks spanning the full range of ultramafic protoliths have added new insight into the behavior of rocks during serpentinization and corroborate the previous works on the stability of magnetite.
Due to the mineralogical simplicity of ultramafic rocks, their chemistry and physical properties vary relatively predictably during serpentinization. We have presented three new formulations (Equations 1-3) for determining the %serpentinization of a given ultramafic rock on the basis of whole-rock major-element chemistry (LOI and OPE), density, or magnetic susceptibility, which were calibrated using quantitative mineralogical estimates ( Figure 4); the major-element chemistry-and density-based formulations are far more robust than that based on magnetic susceptibility as has been noted previously (e.g., Li et al., 2020;Maffione et al., 2014;Oufi et al., 2002). The principal requirement for using these formulations is that it must be ascertained that a given sample is minimally carbonated (<1 wt% CO 2 ). In uncarbonated rocks, increased volumes of brucite will tend to result in higher LOI, while in ophicarbonate rocks, carbonation may not always be obvious petrographically as it can be highly domainal and fine-grained, which could also potentially skew LOI to higher levels. These two factors could account for inconsistencies observed in using LOI in previous studies (e.g., Deschamps et al., 2013). Density-based calibrations have been previously calibrated (e.g., Miller & Christensen, 1997;Oufi et al., 2002) and yield similar results to Equation 2; ours differs in that the %serpentinization of the rock has been robustly constrained using quantitative mineralogical determinations and whole-rock chemistry on >200 samples. We have also shown here (Figure 9b) that for dunitic rocks, density should be a relatively accurate predictor of brucite content but that this may be site specific.
Estimating %serpentinization using magnetic susceptibility is far less reliable than density and chemistry as indicated by the regression statistics in Equation 3 and as shown in Maffione et al. (2014). This inconsistency may be a product of many factors, including that our samples reflect the full diversity of ultramafic of protoliths and role of mineral chemistry (Mg/Fe in the primary silicates) and the extent and conditions of serpentinization (temperature, pH, aSiO 2 , and oxygen fugacity) in controlling the stability of magnetite and that of the specific serpentine polymorphs and brucite (e.g., Evans, 2008;Evans et al., 2009;Huang et al., 2017;Klein et al., 2014;Li et al., 2020;McCollum, Klein, Solheid, & Moskowitz, 2020;Miyoshi et al., 2014;O'Hanley & Dyar, 1993). A two-stage serpentinization process for ophiolitic harzburgite is commonly inferred involving: (a) initial replacement of olivine to form Fe-rich lizardite serpentine and brucite at temperatures <200°C; (b) orthopyroxene breakdown and Si release, which leads to the formation of more magnesian antigorite serpentine and brucite, magnetite, and hydrogen at temperatures of 200-300°C (e.g., Bach et al., 2006;Beard et al., 2009;Frost et al., 2013;Klein et al., 2014Miyoshi et al., 2014). As dunitic rocks lack or contain only minimal orthopyroxene, only the first serpentinization stage would be expected for such lithologies and magnetite production would not be expected. The data set presented herein demonstrates this well in that highly serpentinized dunitic rocks typically contain similar spinel contents to their less-serpentinized equivalents ( Figure 9) and consequently show relatively low magnetic susceptibilities (<10 SI × 10 −3 ) compared to harzburgitic-lherzolitic rocks (Figure 7), which show consistently high values (>50 SI × 10 −3 : Figure 8). This would indicate that, while magnetic susceptibility may be able to accurately identify highly serpentinized harzburgitic rocks, it may not be able to do so for dunitic or pyroxenite-rich lithologies or to distinguish if the rocks are slightly carbonated (ophicarbonates rocks). We do note; however, that the reverse temperature trend has been shown for some abyssal peridotites (e.g., Su et al., 2015); although this seems to be in the minority. Additionally, this two-stage process for the generation of magnetite may only apply to ultramafic rocks originating in the oceanic crust (ophiolitic or abyssal) as highly magnetic dunitic rocks have been observed in komatiitic units and mafic-ultramafic intrusions (ter Maat et al., 2019;Williams, 2009).
Although harzburgitic-lherzolitic rocks, in general, behave more predictably in terms of their increased magnetite production with serpentinization, there is a subset of samples with OPE between 1.42 and 1.74 that shows high degrees of serpentinization, but very low magnetic susceptibilities (<10 SI × 10 −3 ) (Figure 6). These are dominated by samples from the Nahlin locality with a minor number of samples from Decar. The Decar samples exhibiting low magnetic susceptibilities are relatively uncommon and are distinct in that they show evidence for a third stage of serpentinization comprising mainly of chrysotile that overprints the more typical pervasive lizardite and/or antigorite serpentinization (e.g., Milidragovic & Grundy, 2019;Steinthorsdottir, 2021). Since rocks nearby do not show evidence for this third serpentinization phase and do contain abundant magnetite, it is most likely that these samples simply reflect localized regions of transient permeability that destabilized magnetite. Interpretation of the Nahlin samples is more problematic due to a lack of access to the samples or detailed textural documentation of serpentinization. Nevertheless, among the 38 harzburgitic-lherzolitic Nahlin samples, only five of these show magnetic susceptibilities >10 SI × 10 −3 despite being >50 %serpentinized. There is evidence that the melt-depleted rocks of the Nahlin ophiolite were re-fertilized prior to serpentinization to form intergranular base metal sulphides, clinopyroxene, and Cr-spinel (Lawley et al., 2020). Such an event may be the reason that magnetite formation was inhibited. It would, thus, seem that although magnetic susceptibility in general tends to behave predictably for harzburgitic rocks, which comprise the bulk of ophiolitic ultramafic rocks, there are locality-specific features or events that may inhibit the use of magnetic susceptibility as a proxy for estimating %serpentinization.

Identifying Ultramafic Carbon Sinks Using Physical Properties
Mineral carbonation of ultramafic rocks can be done through either in situ or ex situ methods; the former involves injection of pressurized CO 2 at depths of >2 km (e.g., Kelemen & Matter, 2009), whereas the latter would involve introduction of CO 2 into finely crushed material at the surface or passive exposure to the atmosphere (e.g., Power et al., 2013). For ex situ carbonation, the occurrence of brucite is key due to its high reactivity at surface pressure and temperature conditions, whereas in situ carbonation models invoke the use of olivine as the prime mineral target for carbonation (Kelemen & Matter, 2009). The CO 2 and H 2 O contents of the rocks in this study clearly define two linear arrays (Figure 4b) documenting an initial serpentinization (hydration) event followed by a later carbonation event. No samples have volatile contents that plot between these two arrays suggesting that carbonation only occurred where ultramafic rocks were previously highly serpentinized. We suggest several explanations for this observation: 1. Although serpentinization and carbonation reactions are both thermodynamically favored at low temperatures (Kelemen et al., 2020), there may be major kinetic barriers to the direct carbonation of olivine and pyroxene (olivine + pyroxene→carbonate + quartz), such that carbonation may only proceed after hydration reactions have created layered silicates (serpentine) and hydroxides (brucite). If serpentinization is a prerequisite for the carbonation of ultramafic rocks, then both ex situ and in situ carbon sequestration plans (e.g., Kelemen & Matter, 2009;Power et al., 2013) should consider targeting serpentinized rather than fresh mantle rock.
2. The restriction of carbonation to previously serpentinized rock may reflect the evolution of fluid pathways in these initially dry ultramafic rocks whereby initial serpentinization caused by the infiltration of H 2 O-rich fluids along fractures would increase permeability (e.g., Kelemen & Hirth, 2012), resulting in preferred pathways for fluid flow which were later used by CO 2 -bearing fluids. In this scenario, the carbonated portions of the ultramafic massifs in this study all experienced a similar two-stage fluid infiltration history with early H 2 O-rich fluids followed by later CO 2 -bearing fluids, with the latter only infiltrating fully serpentinized fracture zones. 3. Carbonation is driven by the infiltration of H 2 O-rich, CO 2 -bearing fluids (e.g., Paukert et al., 2012) that drives serpentinization fronts ahead of carbonation fronts as a natural consequence of the relative abundances of the two chemical components and the fluid composition being buffered by carbonation reactions. If this reaction pathway is a natural consequence of the infiltration of H 2 O-rich, CO 2 -bearing fluids into peridotite, then in situ direct carbonation of olivine without prior serpentinization (Fo + CO 2 →Mgs + Qz) would presumably require the injection of CO 2 -rich solutions; none of the localities in this study provide a geologic record of such a process.

Modeling Ultramafic Rock Mineralogy
Maximum serpentine and brucite contents occur in samples with 10-14 wt% LOI (>75 %serpentinized) (Figure 6a). In the absence of characterizing the mineral abundance and reactivity of samples in great detail, the physical properties of ultramafic rocks should be effective at predicting-at a first order-mineral content and, thus, a rough estimate of the reactivity of a given volume of rocks. Such an approach has been used to identify the boundaries and extent of carbonation within a serpentinite body (e.g., Hansen et al., 2005;Tominaga et al., 2017) and scanning magnetic microscopy has been used to identify magnetic sources at the micron-scale (Pastore et al., 2018); geophysical techniques have yet to be applied for identifying highly serpentinized ultramafic rocks at the outcrop-scale. Two approaches for doing so for serpentinites include: (a) taking physical property measurements of hand-samples or drill-core and/or (b) using physical property models to inform geophysical survey inversions. We follow the methods outlined in Enkin et al. (2020), in which the relationship between physical properties and mineral abundances were used as a forward model to calibrate the Henkel plot ( Figure 10) and then to inverse model the mineralogy based on physical properties. In this approach, minerals with similar physical properties and that behave in a similar way are grouped together into mineral endmembers and mixing lines with volumetric proportions are calibrated. Three endmembers are used to construct two models: one for (uncarbonated) serpentinites and another for carbonated rocks. In both, end-members and mineral mixing lines are constructed by considering both the theoretical published values for various minerals and the vol% abundance of constituent minerals in the samples as constrained by qXRD.

Forward Modeling
For the serpentinites (Figure 10a), the three mineralogical end-members that we consider are: ultramafic silicates (UM), serpentine and brucite (SB), and magnetite (M). For UM, we use a density of 3.224 g/cm 3 and magnetic susceptibility of 0.3 SI × 10 −3 ; the former is based on the intercept of Equation 2, while the latter was chosen to encompass all minimally serpentinized samples. For SB, we use a density of 2.558 g/cm 3 and magnetic susceptibility of 0.03 SI × 10 −3 ; the density was determined from using a 100 %serpentinized value applied to Equation 2 and then assuming 3 vol% magnetite as this would be consistent with the magnetic susceptibility of the samples. The magnetic susceptibility of the SB component was chosen to encapsulate all uncarbonated samples with minimal porosity. A major unknown for ultramafic rocks is the proportion of primary spinel to magnetite; the former is most common in unserpentinized rocks and cannot be distinguished using XRD. Consequently, for the M endmember, we use a density and magnetic susceptibility of 5.20 g/cm 3 and 3000 SI × 10 −3 , respectively, which is based on empirical studies that suggest these values to be representative of magnetite regardless of grain size (Heider et al., 1996;C. Peters & Dekkers, 2003). On the Henkel plot for carbonated rocks (Figure 10b), the same SB and M end-members are used for serpentinites but UM is replaced with an endmember reflecting the physical properties of magnesite (MS); although quartz and talc may also be present in carbonated rocks, magnesite is by far the most volumetrically significant. For MS, we use a density of 3.00 g/cm 3 and magnetic susceptibility of 0.3 SI × 10 −3 ; the former is based on the density of magnesite, while the latter was chosen to fully encapsulate all samples.

Inverse Modeling
For rocks that plot within the curves defined by the forward modeling, we can invert the model to provide estimated volumetric mineral abundances. Following the same framework as in Enkin et al. (2020), for the serpentinite model we have three relationships with three unknowns: where d = density K = magnetic susceptibility, and UM, SB, and M, refer to the volumetric mineral abundances of ultramafic minerals, serpentine + brucite, and magnetite. These are combined as follows: where d x and K x refer to the values assigned to the end-member mineral compositions as defined in the forward model, d and K refer to the bulk rock measured density and magnetic susceptibility, respectively. A similar relationship can be constructed for carbonated rocks but substituting the properties for UM with those of MS. When the inverse modeling is applied to the measured physical properties of the serpentinites, the results show a near 1:1 correlation ( Figure 11) for the calculated abundances of hydrous (mostly serpentine and brucite; SB) and primary igneous (olivine + pyroxene; UM) minerals and those determined Figure 11. Modeled mineralogy of serpentinites (a-c) and carbonated ultramafic rocks (d-f) compared to mineral abundances as determined by quantitative X-ray diffraction (qXRD). Hydrous minerals encompasses any minerals formed by the hydration of ultramafic rocks and includes serpentine, brucite, actinolite, and chlorite. Carbonation minerals includes carbonate minerals (magnesite, dolomite, and calcite), talc, and quartz. All regressions (solid lines) were done after omitting outliers, which were identifying by their residuals being outside of 2 SD of the mean.
using qXRD. The model magnetite contents are consistently lower than the total spinel as determined by qXRD, which is incapable of separating out the magnetite concentration; this result is to be expected and the model results are considered an accurate estimate of the abundance of magnetite based on there being no (or only negligible quantities) of other highly magnetic minerals present in these rocks (e.g., magnetic sulphides and alloys).
When the inverse model is applied to the carbonated samples, the carbonation minerals (mostly magnesite + talc + quartz) are consistently overestimated for ophicarbonates and underestimated for listwanites and the opposite is true for hydrous minerals (mostly serpentine). This discrepancy can be explained by the model assumptions that the MS end-member is approximated by the physical properties of magnesite. Overestimation of MS in ophicarbonates could be caused by the presence of relict primary ultramafic minerals, which are still present in such rocks at abundances of up to 15 vol%. Overestimation of MS in listwanites likely reflects the presence of quartz in the assemblage at abundances of up to 20 vol%. The correlation between calculated and qXRD abundances is shown in Figure 11 and can be used to correct for the systematic bias. In carbonated samples, the calculated magnetite contents of carbonated rocks are highly variable but cluster around the 1:1 line with qXRD spinel abundances. We note that several samples of our samples fall outside the field defined by the mixing lines and would thus yield negative modal abundances according to the inverse model. Samples with densities lower than the SB-M mixing line, likely reflect increased porosity; porosities of 1%-4% are observed in our samples.

Using Geophysics to Identify and Quantify Ultramafic Carbon Sinks
For a given identified ultramafic body, whose mapped boundaries are relatively well known, the ultimate goal would be to identify and quantify the volumes of serpentinized rocks using remote sensing (i.e., geophysical surveys). The most accurate interpretations of geophysical data sets, such as aeromagnetic or gravity surveys, will involve integration of robust physical property-lithology models. Dunitic protoliths (OPE >1.93) tend to generate the highest brucite content and, thus, will be most reactive, however, harzburgitic rocks (OPE 1.42-1.93) can contain relatively high brucite contents and are volumetrically the most significant in ophiolite massifs and, thus, their physical properties will dominate any geophysical signature.
The systematic drop in density with increased serpentinization across all protoliths indicates that gravity surveys should be the most accurate means for assessing the degree of alteration of a given ultramafic body. If available, and at a sufficiently high resolution, an upper limit of ∼2.8 g/cm 3 would be effective at identifying most highly serpentinized rocks while excluding fresher and moderately to highly carbonated rocks. However, high-resolution gravity surveys are expensive. Although using magnetic susceptibility to predict the degree of serpentinization is subject to more unpredictable physical property behavior, magnetic surveys are more widespread. In general, highly serpentinized harzburgitic rocks show values of >20 SI × 10 −3 with median values of >50 SI × 10 −3 . A lower cut-off of 20 SI × 10 −3 should identify all highly serpentinized harzburgitic samples and would exclude many rocks that are <60 %serpentinized and those that are moderately to highly carbonated (soapstones and listwanites).
We note that weakly carbonated rocks (ophicarbonates) have identical physical properties to harzburgitic rocks (Figures 7 and 10) and, thus, cannot be discriminated using physical properties and, by extension, geophysical surveys. While the above physical properties limits will result in the identification of false positives-mostly commonly in the form of weakly carbonated rocks-it will rarely result in false negatives. For ex situ carbonation, which relies on highly reactive minerals such as brucite that are no longer stable in the presence of any CO 2 , the inclusion of ophicarbonates will dilute the overall reactivity of a given volume as ophicarbonate rocks are commonly in close proximity to, or intercalated with, uncarbonated rocks. On the other hand, for in situ carbonation, which could also make use of serpentine minerals, ophicarbonates are still a viable option given the abundance of such minerals in these rocks.
A complication of applying magnetic susceptibility models to magnetic survey data is that the total magnetization of a given anomaly will reflect a combination of its remanent and induced magnetic fields (Gee & Kent, 2007). Any samples with Koenigsberger ratios (Q-ratio) <1 will have total magnetization signatures dominated by their induced magnetization (magnetic susceptibility). Samples with Q-ratio >1 will have total magnetization signatures dominated by their NRM and the strength of a given anomaly will depend on its coherence (direction). The high Q-ratios recorded in many harzburgitic samples that are <75 %serpentinized (Figure 8d) likely reflects the lack of high volumes of serpentinization-related magnetite and, thus, relatively low observed magnetic susceptibilities (Figures 6d, 8a and 8b). In contrast, the observation of a large range in NRM values (Figures 6e, 8a and 8b) but decreasing median Q-ratios (Figure 8d) in more serpentinized samples likely reflects the increased magnetite production during serpentinization. In the absence of information on the direction of the preserved NRM, we refrain from calculating hypothetical total effective susceptibilities; future work will include detailed analysis using oriented samples. Nevertheless, the power of physical property models of ultramafic rocks to inform geophysical inversions with the goal of prospecting for serpentinites was demonstrated by Mitchinson et al. (2020). More detailed analysis of NRM will refine such models for a more accurate assessment of the carbon sequestration capacity of a given volume of ultramafic rock.

Implications for Imaging Hydration and Carbonation in Subduction Zones
At depths >30-50 km in subduction zones, H 2 O-CO 2 fluids liberated from the downgoing plate infiltrate and react with the overlying forearc mantle wedge, which in turn provides a record of slab-derived fluid infiltration integrated over millions of years (e.g., Hyndman & Peacock, 2003). Experiments by Yardley et al. (2014) demonstrate that hydration reactions are rapid (months to weeks) confirming the rate of serpentinization is likely controlled by the H 2 O flux to the reaction front (Macdonald & Fyfe, 1985). The hydration and carbonation history of ultramafic rocks detailed in this study provides insight into similar reactions expected to occur in the mantle wedge. Rock density systematically decreases with increasing serpentinization, but later carbonation increases rock density (Figure 6b) suggesting that gravity data cannot be simply inverted to determine the degree of mantle-wedge serpentinization. Similarly, strong, deep magnetic anomalies have been cited as evidence of extensive forearc serpentinization in Cascadia (Blakely et al., 2005) and other subduction zones. The results presented here suggest serpentinization dramatically increases the magnetic susceptibility of harzburgite, but not dunite. If these results can be extrapolated to the elevated P-T conditions of mantle-wedge serpentinization, this implies that the hydrated Cascadia mantle wedge is made up mostly of harzburgite.
This study documents the sequential replacement of serpentinites during CO 2 infiltration to form magnesite, talc, and ultimately quartz, which occurred, in part because the ultramafic rocks were interleaved with, and obducted onto, carbonate-bearing sedimentary rocks of the Cache Creek terrane (e.g., Zagorevski et al., 2018). In the case of ultramafic rocks occurring in the mantle wedge in subduction zones worldwide, although the infiltration of CO 2 -bearing fluids likely occurs as a result of the subduction of carbonate sediments and carbonate-bearing oceanic crust, the reactions and resulting mineral assemblages would be expected to be similar to those observed here and the resulting rheological changes to subduction zone dynamics could be profound. Specifically, deformation experiments (Escartín et al., 2008;Moore & Lockner, 2008) demonstrate that talc is an extremely weak mineral over a wide range of crustal P-T conditions, and serpentinite is dramatically weakened by the presence of minor amounts of talc (Moore & Lockner, 2011). If such minerals were to form along the base of the mantle wedge, this would affect the rheology of the subduction plate interface.

Conclusions
We present petrological observations, major-element chemistry, quantitative mineralogy, and physical properties from >400 samples of variably serpentinized and carbonated ophiolitic ultramafic rocks collected from the Cache Creek (Atlin) terrane in British Columbia to constrain the physio-chemical changes that occur during the alteration of ultramafic rocks. The samples show a systematic decrease in density during serpentinization that largely reflects a drop in relict ultramafic mineral abundances and an increase in serpentine minerals and an increase in density during carbonation that mostly reflects the formation of magnesite. The samples show two magnetic susceptibility trends during serpentinization: one involves a 100fold increase in magnetic susceptibility and is followed by most harzburgitic samples, whereas the second involves very little change in magnetic susceptibility and is followed by most dunitic samples and a minor proportion of harzburgitic samples. Using the quantitative mineralogy and physical property relationships, we provide a model that can be used to estimate the mineralogy of variably altered ultramafic rocks (at a first-order). These results can be used as a basis for interpreting gravity or magnetic geophysical surveys in order to prospect for the most prospective ultramafic rocks for carbon sequestration and, potentially, to aid in imaging subduction zone lithological heterogeneity.

Data Availability Statement
All data forming the basis for this paper are available at http://doi.org/10.5281/zenodo.5213140. ing access to samples from the Decar and Hogem areas. Alex Zagorevski is thanked for providing samples and data from the King Mountain, S. Yukon, Nahlin, and parts of Atlin; these samples were collected as part of the Geo-Mapping for Energy and Minerals-2 (GEM-2) program. The authors also acknowledge that all samples used in this study were collected from the traditional and unceded territories of Taku River Klingit, Kaska Dena, Tāłtān Konelīne, Carcross/Tagish, Teslin Tlingit, Tl'azt'en, Binche Whut'en, Yekooche, and Takla First Nations. This manuscript was greatly improved by constructive reviews and comments from Juan Carlos de Obeso and an anonymous reviewer and the editorial handling of Joshua Feinberg. Funding is provided by Natural Science and Engineering Research Council (NSERC) discovery grants to G. M. Dipple (RGPIN-2015-05036) and S. M. Peacock (RGPIN-2020-07066), and Natural Resources Canada Clean Growth Program grant (CGP-17-0739), and Geoscience BC grant (2018-038) to G. M. Dipple.