Investigating the crystallization history of Apollo 15 mare basalts using quantitative textural analysis

Mare basalts collected at the Apollo 15 landing site are classified as belonging to either the quartz‐normative suite or the olivine‐normative suite, based on differences in whole‐rock major element chemistry. A wide range of textures are displayed within samples from both suites, which provide insight into eruption processes on the Moon. Here we use crystal size distribution (CSD) analysis and spatial distribution pattern (SDP) analysis of pyroxene, olivine, and plagioclase crystals in eight Apollo 15 mare basalt samples to investigate the crystallization and emplacement of the quartz‐normative and olivine‐normative suites. In general, our results show similarities between the CSDs and SDPs for both mare basalt suites. However, we also report two distinct groups of pyroxene CSD trends that likely represent samples with common cooling histories, originating from comparable depths within respective olivine‐normative and quartz‐normative lava flows. We use our results to determine the relative depths of samples within the lava flows at the Apollo 15 landing site.


INTRODUCTION
Mare basalts provide valuable insights into the chemistry of the lunar interior, magmatic, and volcanic systems, and the thermal history of the Moon through time Shearer et al., 2006;Shearer & Papike, 1999;Wieczorek et al., 2006). The Apollo missions returned 134 individual mare basalt samples >40 g from six landing sites on the lunar nearside ( Figure 1a). A diverse range of chemical and petrological characteristics exist within the mare basalt samples due to differences in lunar mantle source compositions and partial melting conditions (e.g., Hallis et al., 2010;Snape et al., 2019;Snyder et al., 1991Snyder et al., , 1992Tartese et al., 2013), eruption processes (e.g., Lo et al., 2021;Wieczorek et al., 2001;Head, 2017, 2018 andreferences therein), and crystallization histories (e.g., Grove & Krawczynski, 2009;. A vital method for understanding mare basalt petrogenesis and lunar magmatic processes is the analysis of mare basalt mineral chemistries and textures. The development of quantitative petrological techniques, such as crystal size distribution (CSD) analysis (e.g., Cashman, 2020;Cashman & Marsh, 1988;Higgins, 2000Higgins, , 2006Marsh, 1988Marsh, , 1998Morgan & Jerram, 2006), has greatly improved our knowledge of magmatic processes and has enabled the nondestructive analysis of small and precious samples (Day & Taylor, 2007;Fagan et al., 2013;Hui et al., 2011;Neal et al., 2015). Although many Apollo samples were extensively studied upon their return, the application of new image analysis techniques means that much can still be learned from Apollo mare basalts, even 50 years on, while ensuring the samples remain preserved for future generations (e.g., Pernet-Fisher et al., 2019).

Apollo 15 Mare Basalts Background
On July 30, 1971, the Apollo 15 lunar lander touched down on the eastern edge of the Imbrium Basin (26.13222°N latitude, 3.63386°E longitude), close to Hadley Rille and the Apennine Mountain front (ALGIT, 1972a;Figure 1b). A third of all samples collected on the Apollo 15 mission were mare basalts, the majority of which were collected at stations 1, 4, and 9 (ALGIT, 1972b; Figure 1c). Examination of these basalts in hand specimen reveals a wide range of mineral textures and grain sizes, and differing degrees of vesicularity: some samples have >50% voids and vuggy textures (ALGIT, 1972b). Categorization of the Apollo 15 mare basalts into two suites was made based on the analysis of whole-rock major element abundances (Chappell & Green, 1973;Helmke et al., 1973;Rhodes, 1972). The two chemically distinct suites were named the quartznormative basalts and the olivine-normative basalts (Chappell & Green, 1973;Rhodes & Hubbard, 1973). The quartz-normative basalts have comparatively lower FeO (19-20 wt%) and TiO 2 (1-2 wt%) and higher SiO 2 (47-49 wt%) bulk rock compositions (Chappell & Green, 1973;Rhodes & Hubbard, 1973). The olivine-normative basalts have comparatively higher FeO (22-23 wt%) and TiO 2 (2-3 wt%) and lower SiO 2 (44-46 wt%) bulk rock compositions (Chappell & Green, 1973;Rhodes & Hubbard, 1973). The quartz-normative basalts have average crystallization ages of 3371 AE 21 Ma and therefore were erupted first, followed by the olivine-normative basalts at 3287 AE 21 Ma (Snape et al., 2019). Although the selection of samples dated by Snape et al. (2019) does not directly overlap with the samples used in this study (some of which have never been dated), these values provide the most recent estimate of the average ages of the two suites.
The petrogenetic relationship and cause of the compositional variations between the quartz-normative and the olivine-normative suites has been debated (Chappell & Green, 1973;Rhodes & Hubbard, 1973;Schnare et al., 2008;Snyder et al., 2000). Some studies attribute the variation to differences in mantle source composition and conclude that the two suites cannot be related by simple fractional crystallization processes alone (Chappell & Green, 1973;Rhodes & Hubbard, 1973;Snape et al., 2019;Snyder et al., 2000). However, others suggest that trace element analyses of major mineral phases in the two suites are not sufficiently different to exclude the possibility that both suites originated by partial melting of the same source but experienced different degrees of fractional crystallization (Schnare et al., 2008).
We have used CSD and spatial distribution pattern (SDP) analyses to quantify the textures within a selection of quartz-normative and olivine-normative Apollo 15 mare basalts. Our aim is to identify textural differences or similarities between the two Apollo 15 mare basalt suites that could provide information as to their crystallization histories. This will provide further insight into the eruption processes operating on the Moon through time and allow for comparisons with mare basalts collected from other landing sites.

Samples
A total of eight Apollo 15 mare basalts were chosen for this study, across both quartz-normative and olivinenormative suites. One of the benefits of CSD analysis is that it can be conducted on relatively small samples, although a minimum number of crystals is required to accurately assess a crystal population within a rock. Morgan and Jerram (2006) suggest that this threshold should be n = 75 for crystals with tabular shapes and n > 250 for crystals with acicular shapes. We therefore only analyzed thin sections that appeared to be able to meet the threshold number of crystals needed for statistically robust CSD analysis.
The olivine-normative basalt thin sections ( Figure 2) selected were 15555,209; 15105,6; and 15536,7. Sample 15555, known as "Great Scott," is one of the largest mare basalt samples collected on the Moon and is thought to represent a primitive magma composition due to its high MgO content of 11 wt% (Chappell et al., 1972;Chappell & Green, 1973;Mason et al., 1972). Conversely, 15105 is FIGURE 2. Images of olivine-normative (ON) thin sections 15555,209, 15105,6; and 15536,7 taken using plane-polarized light (a, d, and g), cross-polarized light (b, e, and h) and backscattered electron (BSE) mapping (c, f, and i). The scale is the same for all images. Examples of olivine (ol), pyroxene (pyx), and plagioclase (plag) crystals are labeled in the BSE images. Black areas in the BSE image correspond to vugs, vesicles, or holes in the sample. For full thin section images, see Data SA. (Color figure can be viewed at wileyonlinelibrary.com) a small 5.6 g rake fragment (Ma et al., 1976;Swann et al., 1972) with higher SiO 2 and lower MgO than 15555. Chipped from a boulder at the edge of Hadley Rille, 15536 is representative of several samples collected in that area (Mason et al., 1972) and is chemically similar to 15555, but with slightly higher FeO and MgO contents. All of the olivine-normative thin sections selected contained enough olivine and pyroxene crystals (>250) to produce CSD plots of both phases.
The quartz-normative basalt thin sections ( Figure 3) selected for this study were 15597,12 and 15597,18; 15125,6; 15475,15; and 15076,68. The bulk composition of 15597 is thought to represent a primary undifferentiated melt that was quenched from a high temperature, and has been the subject of many experimental studies (Grove & Bence, 1977;Grove & Raudsepp, 1978;Grove & Walker, 1977;Lofgren et al., 1975). Two different thin sections of 15597 were included to help assess the reproducibility of our results. Although considered analogous to 15597 (Dowty et al., 1973), 15125 is included as it contains olivine phenocrysts as well as pyroxene phenocrysts, which is uncommon for quartz-normative samples. Sample 15475 was selected as it has been the subject of experimental petrology studies (Grove & Walker, 1977;Lofgren et al., 1975) and is thought to be one of the slowest cooled Apollo 15 mare basalts (Takeda et al., 1975). Sample 15076 has higher MgO and lower FeO contents that 15597 and 15125, and is thought to be part of a sequence of basalts that represent different depths from within a lava flow excavated by Elbow Crater (Brett, 1975;Grove & Walker, 1977;Lofgren et al., 1975;Onorato et al., 1979). We also considered sample 15065, collected close to the rim of Elbow Crater at station 1 (Figure 1c), which is thought to have sampled some of the deepest parts of the lava flow in this area (Grove & Walker, 1977;Lofgren et al., 1975;Onorato et al., 1979). As first noted in the original Apollo 15 sample catalog (Butler, 1971), 15065 is composed of two distinct regions with a main portion of the sample and more vuggy/mafic portion ( Figure 4). We analyzed thin sections of both the main portion (sections 74, 81, 196, and 197) and the vuggy portion (sections 91, 198, and 199) of 15065 ( Figure 4).

METHODS
Thin sections were prepared at the NASA Johnson Space Center Apollo Curatorial labs. Samples were first imaged optically with a petrographic microscope using plane-polarized and cross-polarized light at the University of Manchester, prior to carbon coating. We used the FEI QUANTA 650 field emission gun scanning electron microscope (SEM) at the University of Manchester to collect backscattered electron (BSE) maps of each sample using a dwell time of 10 μs at an accelerating voltage of 15 kV. All BSE images were mapped at a resolution of 0.98 μm per pixel and saved as .tif files (see Data SA). Compression or rotation was not applied to any of the BSE maps at any stage.

CSD Analysis
CSD analysis trends are conventionally plotted as the natural logarithm of the population density against crystal length ( Figure 5). The population density (mm À4 ) is the volumetric number density of crystals within a given crystal length bin width. Crystal length (mm) refers to measured data that have been converted from 2D to 3D values. The slope of a CSD graph can be described as: where G is the crystal growth rate (mm s À1 ) and τ the crystal residence time (s). In petrological studies, the crystal residence time is akin to a measure of the average time a crystal within the sample has been growing. The crystal residence time does not provide an estimate of how long a crystal remained in a particular magmatic environment. This is because the crystal residence time relates to the full range of crystal sizes within the sample population, and it is unreasonable to assume that the smallest crystals within a sample remained in the magma for an identical period of time to the largest crystals within the sample. The y-intercept of a CSD graph is equal to the nucleation density (n0). The CSD slope equation is only applicable for linear portions of a CSD trend. Linear CSD trends ( Figure 5) imply that the crystals formed during steady-state crystallization (Marsh, 1988) with a constant cooling rate (Higgins, 2000). Curved CSD trends ( Figure 5) can indicate magmatic processes such as the loss of crystals due to fractionation or the addition of crystals due to accumulation (Figure 5, Marsh, 1988, 1998. For example, the fractionation of crystals would be expressed as a concave-down curve with a lower-than-expected population density (in the largest crystal size bins) compared to a normal distribution (i.e., a straight line). The opposite is true for crystal accumulation which results in concave-up trends. Other processes such as crystal coarsening can also result in curved CSD plots. Due to the growth of larger crystals at the expense of smaller crystals, coarsening trends are concave-up with a downturn in slope at the smallest grain sizes (e.g., Higgins, 2006). Kinked CSD plots ( Figure 5) can indicate that the system experienced two (or more) different cooling rates (e.g., Higgins, 2006), or that magma mixing occurred (e.g., Higgins, 1996aHiggins, , 1996bHiggins, , 2006Marsh, 1988Marsh, , 1998. To collect the required crystal size data, outlines of plagioclase, olivine, and pyroxene crystals were manually FIGURE 3. Images of quartz-normative samples 15597,12; 15597,18; 15125,6; 15475,15; and 15076,68 taken using planepolarized light (a, d, g, j, and m), cross-polarized light (b, e, h, k, and n) and backscattered electron (BSE) mapping (c, f, i, l, and o). The scale is the same for all images. Examples of olivine (ol), pyroxene (pyx), and plagioclase (plag) crystals are labeled in the BSE images. Black areas in the BSE image correspond to vugs, vesicles, or holes in the sample. For full thin section images, see Data SA. (Color figure can be viewed at wileyonlinelibrary.com) traced from the BSE maps using CorelDraw 2018. The boundaries between touching crystals were determined using the optical images taken of each thin section. Crystals that intersected the edge of the thin section or large cracks/fractures within the sample were not included as they do not represent true crystal lengths. Images of the traced crystal boundaries were imported into ImageJ version 1.51 (Abràmoff et al., 2004). We used the Analyse Particles ImageJ function to calculate the major and minor axis length (mm), coordinates (X and Y), area (mm 2 ), and roundness (indexed on a scale from 0 to 1.0) of each of the traced crystals. A crystal minor axis size cutoff of <0.03 mm was used for all samples. Although crystals with minor axes <0.03 mm were readily identifiable using the high-resolution (0.98 μm per pixel) BSE maps, this cutoff is used to ensure our results are comparable with previous CSD studies of lunar samples which have used optical images to trace crystal boundaries (e.g., Neal et al., 2015). For CSDs determined using optical images, the threshold of <0.03 mm is imposed because this is the average thickness of a standard thin section, meaning that crystals smaller than this are possible projections rather than true crystals within the sample (Higgins, 2000). Data analysis including crystal minor axis values <0.03 mm can be found in Bell et al. (2020) for samples 15597,18; 15125,6; 15555,209; and 15475,15, where the data were used to assess the plausibility of using QEMSCAN (Quantitative Evaluation of Minerals by SCANing electron microscopy) as a semi-automated method of CSD data collection. For the remaining thin sections in this study, data including minor crystal axis values of <0.03 mm and associated CSD plots are found in Data SA, but will not be discussed further in this manuscript.
CSD analysis requires 2D data collected from the images to be converted into 3D crystal data. Major and minor crystal lengths from ImageJ were entered into the CSDslice software package (Morgan & Jerram, 2006) which uses raw 2D measurements to calculate best-fit short (x), intermediate (y), and long (z) axis values of 3D crystal habits within a sample. For each 3D crystal habit, an estimate of the R 2 value (a statistical measure of how close the 2D data lie to the fitted regression line) is calculated. An R 2 value of >0.8 is considered a reliable estimate of 3D crystal habit (Morgan & Jerram, 2006). The 3D crystal habit parameters were then used in CSDcorrections v1.6 (Higgins, 2000) to stereologically convert the imaged 2D crystal lengths into 3D shape estimates. To avoid errors arising from too few crystals in each size bin, crystals lengths were sorted using the recommended five bins per decade (Higgins, 2000). Size bins containing <3 crystals were not used to produce the CSD plots or in subsequent calculations. Error bars for the population density are calculated in CSDcorrections based on the statistical counting error which is equal to the square root of the number of crystals in that size bin (Higgins, 2000). Other sources of error arise from the stereological conversion of 2D to 3D crystal shapes, shape probability parameters being calculated based on idealized crystal shapes (Higgins, 2000), and the assumption that crystal shapes are constant and do not vary with crystal size (e.g., Mangler et al., 2022). It is not possible to accurately quantify the additional contribution of these sources, and so error bars calculated in CSDcorrections represent a minimum error.
The equation of a CSD trend (Equation (1)) is only valid for linear portions of a CSD plot (Marsh, 1988), so it is important to assess the linearity (or otherwise) of CSD trends. In CSDcorrections, there is an option to calculate the Q value of a CSD plot (Higgins, 2000). This provides a measure of the goodness of fit of CSD data to a straight line and enables CSD plots to be evaluated as linear (Q > 0.1), sublinear (0.001 < Q < 0.1), or curved (Q < 0.001) Higgins, 2006). Based on the Q value, any plots that were considered to be curves were split into two portions at the point of maximum curvature, and each portion fitted as a linear segment allowing us to calculate two values for CSD intercept and slope. The first portion includes the smallest crystals size bins within that sample, and is usually the steepest gradient; this is referred to as Slope 1. The second portion encompasses the largest crystal size bins, and is usually the shallowest gradient; this is referred to as Slope 2. Slope 2 is measured from the point of the maximum change in slope for the whole CSD plot, excluding minor changes between individual bins. Intercept values for Slopes 1 and 2 are referred to in the results as Intercepts 1 and 2, respectively.
Using Equation (1), residence times (τ) were calculated using growth rates (G) of 1.309× 10 À8 mm s À1 for plagioclase (Burkhard, 2005), 1.797 × 10 À8 mm s À1 for pyroxene (Burkhard, 2005), and 3.162 × 10 À8 mm s À1 for olivine (Borell & Kilinc, 2012). Although our calculations assume constant growth rates, dynamic crystallization experiments on natural samples indicate that crystal growth rates are likely to vary (e.g., Dowty et al., 1974;First & Hammer, 2016;Grove & Walker, 1977;Hammer, 2009). The constant growth rates used here were calculated from 1-atm experiments and CSD analysis of natural terrestrial basalts from Hawaii under terrestrial crystallization conditions, and it is unclear to what extent they are representative of crystal growth rates under lunar magmatic conditions. Crystal growth rates are affected by factors such as continuous undercooling (e.g., Cashman, 1990), dissolved volatile content, and magma viscosity (e.g., Bianco & Taylor, 1977). It is well established, for example, that mare basalts have higher Fe/ Mg ratios and lower SiO 2 than terrestrial basalts (Papike & Bence, 1978), meaning mare basalt melts are less viscous (Bottinga & Weill, 1972), and the diffusivity of components within the melt is higher. As a consequence, crystal growth rates in lunar magmas have the potential to be faster than in terrestrial magmas. The selected crystal growth rate impacts the calculated residence time and, in turn, the calculated cooling rate (e.g., Brugger & Hammer, 2010). For example, increasing the crystal growth rate by a factor of 10 would reduce the calculated residence time by the same factor. The crystal growth rate values we selected have previously been used in CSD analyses of mare basalts (Bell et al., 2020;Neal et al., 2015), and we have chosen these values so that the outcomes of this study can easily be compared with previous work. The residence times we report allow our CSDs to be compared with previous studies. For CSD plots subdivided into two segments, we report residence times for both portions (i.e., using Slopes 1 and 2) giving a maximum and minimum range. We note that the calculated τ values carry significant uncertainty due to our assumptions regarding crystal growth rates, and that CSD-derived residence times do not necessarily reflect the duration a crystal has spent within a particular magmatic environment.

SDP Analysis
SDP analysis is a quantitative measure of the packing arrangement of crystals within a rock (Jerram et al., 1996). It can be used to understand how igneous textures are influenced by growth rates, crystal nucleation, and the formation of clusters or crystal frameworks. Previous studies have used SDP analysis to determine the relative position of igneous samples within intrusions and lava flows (Day & Taylor, 2007;Jerram et al., 2003Jerram et al., , 2010. Following the method of Jerram et al. (1996), SDP analysis is based on the ratio between the mean nearest neighbor distance for a random distribution of points and the mean nearest neighbor distance for crystals within a sample. This ratio is referred to as R, and is defined as: where ρ is the number density of the sample distribution in crystals per unit area, r the the nearest neighbor distance, and N the number of crystals measured. To understand the homogeneity or potential clumping of crystals, the R value for crystals within a sample is plotted against porosity (% melt and/or other minerals) ( Figure 5). The R value and phase abundance (%) can be calculated using the Big-R software in CSDcorrections at the same time as CSD analysis (Higgins, 2000;Jerram et al., 1996). The spatial distribution parameter is calculated using X and Y crystal coordinates and other parameters collected during 2D crystal analysis in ImageJ.
An SDP plot is divided into domains of ordered and clustered crystals by the random sphere distribution line (RSDL; Jerram et al., 1996Jerram et al., , 2003. The RSDL was determined using computer modeling of spheres to see how the R value for a random arrangement varied with porosity (Jerram et al., 1996). The touching framework line (TFL) is used to distinguish between touching frameworks and non-touching crystal arrangements .
Variations in R value can be used to identify geological processes, such as crystal overgrowth, compaction, or sorting (Jerram et al., 1996). Studies have been conducted on intrusive igneous rocks such as drill cores from a rhyolitic laccolith in the Halle Volcanic Complex, Germany, in which crystal clustering revealed flow and shear processes during emplacement (Mock et al., 2003). Other studies have used SDP analysis to determine the relative position of igneous samples within magmatic intrusions and lava flows (Day & Taylor, 2007;Jerram et al., 2003Jerram et al., , 2010. For example, R values and porosity have been shown to vary systematically with respect to position within komatiite flows due to the gravitational settling and compaction of olivine crystals (Jerram et al., 1996). As such, SDP analysis may be useful in determining the relative position of Apollo 15 mare basalt samples from within the lava flow.

Olivine-Normative Basalts: Sample Petrology
Sample 15555,209 is porphyritic with phenocrysts of pyroxene and olivine set in a poikilitic plagioclase matrix ( Figure 2). The sample is coarse-grained with equant olivine crystals up to 1.5 mm in length and subhedral pyroxene crystals up to 3 mm in length. Both olivine and pyroxene crystals show chemical zoning. The sample contains minor chromite, ilmenite, silica, and millimetersized vugs (see also Bell et al., 2020;Ryder, 1985). Thin section 15105,6 has a porphyritic texture and contains olivine phenocrysts in a fine-grained matrix of pyroxene, plagioclase, and smaller olivine crystals ( Figure 2). Olivine phenocrysts are up to 3 mm in size and are chemically zoned. Pyroxene are generally lath-shaped, <0.5 mm in size, and also show chemical zoning (see also, Dowty et al., 1973). Several vesicles ( ∼ 1 mm diameter) are also present as well as minor amounts of chromite and ilmenite (Dowty et al., 1973). Like 15555,209, thin section 15536,7 is also coarse-grained with olivine and pyroxene in a poikilitic matrix of plagioclase ( Figure 2). Olivine and pyroxene crystals are generally equant, ranging in size from 0.05 to 3 mm, and showing chemical zoning (see also Shervais et al., 1990). It contains minor ilmenite, silica, and small ( ∼ 1 mm) vugs.

Quartz-Normative Basalts: Sample Petrology
Sample 15597 is vitrophyric, dominated by pyroxene phenocrysts in a brown glassy matrix (Figure 3). Both thin sections of 15597 (12 and 18) show the same texture with acicular pyroxene crystals up to 1 mm in length (excluding one 5 mm fragment of a pyroxene crystal in 15597,12). Pyroxene crystals are strongly zoned with some showing skeletal shapes and hollow cores (see also Weigand & Hollister, 1973). Minor chromite is also present (Bell et al., 2020). Sample 15125,6 is dominated by zoned pyroxene phenocrysts 0.05-3 mm in length (Figure 3). It also contains skeletal olivine phenocrysts ( ∼ 3%) and chromite crystals <200 μm in length (see also Bell et al., 2020;Dowty et al., 1973;Ryder, 1985). Phenocrysts are set in a microcrystalline matrix of needle-like pyroxene and plagioclase crystals with a feathery texture. Thin section 15475,15 has a porphyritic vuggy texture with phenocrysts up to 5 mm in length and vugs less than 1 mm in size ( Figure 3). Large zoned pyroxene phenocrysts are set in a matrix of sub-ophitic plagioclase and anhedral pyroxene. It also contains minor amounts of ilmenite, chromite, silica, and K-rich glass (see also Bell et al., 2020;Ryder, 1985;Schnare et al., 2008). Sample 15076,68 is coarse-grained and dominated by pyroxene phenocrysts 3-4 mm in size which show strong chemical zoning and twinning ( Figure 3). Pyroxene also forms a sub-ophitic texture with hollow plagioclase laths in between larger pyroxene phenocrysts (see also Ryder, 1985).

Sample Petrology
Thin sections of sample 15065 had the coarsest crystal size of all samples in this study, with pyroxene crystals up to 8 mm in length ( Figure 4). Pyroxene crystals display both zoning and twinning and are subhedral, surrounded by an ophitic to sub-ophitic plagioclase matrix. Thin sections from the vuggy portion show the same textures but with a higher modal proportion of pyroxene and less plagioclase, as well as several vugs several mm in length. All thin sections of 15065 contain minor ilmenite, chromite, spinel, silica, and olivine, in variable modal proportions (see also Ryder, 1985).

CSD Results
CSD data are presented for both quartz-normative and olivine-normative samples in Table 1. The majority of samples showed CSD trends that were assessed as curved on the basis of Q values. These CSD trends were split into two linear to sublinear portions at the point of maximum curvature (see Data SB). All but one of the samples have two values for CSD intercept and slope, yielding an upper and lower range for residence times (Table 1). Crystal residence times are akin to average crystal growth times and are used in the Relative Position of Samples in Quartz-Normative and Olivine-Normative Lava Flows Section where we consider the relative cooling rates of the samples and in the Comparison with Other Apollo Mare Basalts and Lunar Meteorites Section when comparing our residence time values with other studies.

Olivine-Normative CSDs
For each of the three olivine-normative samples, CSD analysis was conducted on both pyroxene and olivine crystals ( Figure 6). The number of crystals exceeded the threshold of 250 crystals required for accurate 3D crystal habit estimates (Morgan & Jerram, 2006) in all cases except for olivine in 15105,6, where only 211 crystals were identified.
Crystal shape estimates for olivine in each of the three samples show intermediate axis values ranging from 1.2 to 1.4 and long axis values from 1.5 to 1.8 (Table 1). The R 2 values for all olivine crystal shape estimates are >.8. The Q values of the olivine CSD plots are many orders of magnitude lower than one, indicating the CSD trends are all curved to varying degrees ( Figure 6a). Each trend has a curved appearance, which could suggest accumulation, crystal coarsening, or multi-stage cooling. The curves for olivine in 15555,209 and 15536,7 plot within error of each other for crystal lengths over 1.2 mm. The olivine CSD trend for 15105,6 has the most pronounced kink of the three samples (Figure 6a), and the Q values for Slopes 1 and 2 indicate linear trends suggesting this is a true kink rather than a continuous curve.
The pyroxene crystal shape estimates also show comparable intermediate (1.3-1.4) and long (1.9-2.8) axis values for all three thin sections and the R 2 values are all >0.8. The CSD trends for pyroxene in the olivinenormative samples show a much less pronounced concaveup curve than for olivine, suggesting that the pyroxenes record reasonably constant cooling rates with only minor variabilities (Figure 6b). The CSD trends for pyroxene in 15105,6 and 15536 plot within error of each other from crystals lengths of ∼ 0.5 to 1.1 mm with Slope 2 gradients of À10.10 and À9.21, respectively. In comparison, the slope of 15555,209 is much shallower indicating a relatively slower cooling rate that also explains the presence of larger, 1.5-3 mm long, crystals.

Quartz-Normative CSDs
Pyroxene is the main phenocryst phase in all the quartz-normative samples. Pyroxene CSD analysis was conducted on all quartz-normative samples, except 15475,15 for which plagioclase was analyzed because there were too few pyroxene crystals in this thin section (Figure 7). Olivine is present in 15125,6; however, there were not enough olivine crystals for CSD analysis (see also n/a n/a n/a n/a n/a n/a n/a n/a h Ratio between the mean nearest neighbor distance for a random distribution of points and the mean nearest neighbor distance for crystals within a sample calculated using Big-R in CSDcorrections for SDP analysis. i Calculated using Big-R function in CSDcorrections. For SDP plots, the porosity is considered to be the remaining portion (%) of the sample. Bell et al., 2020). In addition, olivines in 15125,6 have skeletal textures crystals and are therefore unsuitable for CSD analysis. In all cases, the total number of pyroxene (15597,12; 15597,18; 15125,6; and 15076,7) or plagioclase (15475,15) crystals was >250 per thin section (Table 1). Pyroxene crystal shape estimates for 15597,12 and 15597,18 are consistent with each other (1:1.25:1.8) and are comparable to the shape estimated for pyroxene in 15076,68 (1:1.3:2.5). Pyroxene in 15125,6 had a larger long axis estimate of 4.5. All pyroxene crystal shape estimates had R 2 values of >0.8. The Q values for pyroxene in the quartz-normative thin sections were all <0.001 but showed varying degrees of curvature. CSD plots for 15597,12 and 18 plot within error of each other and show slight concave-up curves (Figure 7). Both pyroxene in 15125,6 and 15076,68 show CSD trends with shallower slopes than 15597, suggesting they experienced relatively slower cooing rates. Thin sections 15125,6 and 15076,68 also show Slope 1 values of À3.35 and À3.29, but the offset in these two FIGURE 6. Crystal size distribution (CSD) plots for (a) olivine in olivine-normative samples and (b) pyroxene in olivine-normative samples. Error bars are calculated in CSDcorrections and in most cases the symbol is larger than the error bar. The x-axis varies in size from frame to frame. Points shown with no fill represent where the curved CSD plots were segmented into Slopes 1 and 2 (see Table 1). (Color figure can be viewed at wileyonlinelibrary.com) FIGURE 7. Crystal size distribution (CSD) plots for (a) pyroxene in quartz-normative samples and (b) plagioclase in quartz-normative samples. Error bars are calculated in CSDcorrections and in most cases the symbol is larger than the error bar. The x-and y-axes vary in size from frame to frame. Points shown with no fill represent where the curved CSD plots were segmented into Slopes 1 and 2 (Table 1). (Color figure can be viewed at wileyonlinelibrary.com) trends is due to differences in population density (Figure 7a). In the longest crystal bins sizes for 15076,68, there is a more pronounced concave-up trend which could be due to crystal accumulation and/or crystal coarsening (Higgins, 2011).
The shape estimate calculation for plagioclase in 15475,15 yields a long axis value of 7.0. This estimate has a low R 2 value of 0.73, which indicates a less reliable shape estimate. Reasons for this could possibly be due to subhedral crystal shapes, subsets of the crystal population having slightly different crystal habits, or the fact that in some places the plagioclase crystals in 15475,15 are intergrown with pyroxene. The small Q value for plagioclase in this sample reflects the concave-up trend of the CSD plot (Figure 7b). A change in slope is seen at the 6 mm length crystal mark from À1.48 to À0.73, due to a single data point at 9.39 mm. This data point is most likely an artifact of the shape estimate, which is supported by our observation that the longest plagioclase crystal in the sample is <5 mm. Because of the downturn in the CSD trend at the smallest crystal sizes, it is possible that the curvature in this CSD trend could be due to crystal coarsening (Higgins, 2011), or alternatively by accumulation of larger crystals.

CSDs
Pyroxene CSD analysis was conducted using samples from both the main and vuggy portions of 15065 ( Figure 4). Due to the coarse-grained nature of this rock, it was not possible to trace enough crystal boundaries from one thin section alone to perform CSD analysis. Table 1 details the CSD results for each individual thin section, as well as results obtained by combining crystals from multiple thin sections to produce overall CSD trends for the main and vuggy portions of the sample. Only a small number of crystals were available in 15065,91 (n = 43) and this resulted in very poor crystal shape estimates (R 2 = 0.36). Consequently, the crystals in 15065,91 were not analyzed as an individual sample, but are still included in the combined CSD plot for samples from the vuggy portion of 15065. Of the remaining thin sections, 15065,196 and 197 were the only ones to have >250 crystals. We have reported analyses of those thin sections that had >100 crystals as it allows us to compare CSDs obtained from individual thin sections with the CSD obtained by combining multiple thin sections together.
Crystal shape long axis estimates for 15065 range from 2.6 to 8.0 in the main portion thin sections and from 2.8 to 4.0 in the vuggy portion thin sections (Table 1). Low R 2 values (<0.7) reflect the small number of crystals in some of the thin sections, and are ∼ 0.8 for thin sections with >250 crystals. The variability in crystal shape estimates could be due to the anhedral nature of the pyroxene crystals in these samples. Using crystals from multiple samples increases the R 2 values for the combined 15065 Main from <0.82 to 0.85 and 15065 Vuggy from <0.7 to 0.76.
The Q value for the majority of samples is <0.001; however, 15065,81 has a Q value of 0.3 which indicates a linear CSD trend. The curved trends of the remaining thin sections are shown in Figure 8, which shows CSD trends for each individual thin sections as well as the combined 15065 Main and 15065 Vuggy portion CSDs. The main difference between 15065 CSDs and the rest of the samples in this study is that the plots extend to much longer crystal lengths of up to 30 mm, which is still large despite the grain size of 15065 being the coarsest of all the samples in this study (the longest measured pyroxene crystal from 15065 is 9.6 mm in length). Several of the large size bins for 15065 Main contain only four crystals, which is only one more than the required number of crystals to include a particular size bin. Therefore, the difference in the main and vuggy CSD plots at crystals lengths >10 mm reflects only a small percentage of the overall crystal population. The plots for individual thin sections all show concave-up curves but extend to larger crystal lengths for those in the main portion compared with the vuggy section. Variation between CSDs of thin sections from the same portion of the sample is likely due to the coarse grain size and difference in thin section area (i.e., number of crystals that can be analyzed).
Compared to the CSD plots of other Apollo 15 basalts, 15065 Main has a much shallower gradient and, in turn, must have experienced a slower cooling rate, consistent with the coarse texture of the sample. The maximum observed pyroxene crystal length in a 15065 Main thin section is 8 mm and so it is likely that the extension of the CDS plot to crystal lengths of >10 mm is due to the large value for the long crystal axes (8.0) in the crystal habit estimate. Even though the lengths of these >10 mm crystals my not be representative of the lengths we see in the thin sections, they still represent a (small) proportion of the crystal population that is generally larger than the majority of other crystals within the sample. The concave-up trend for 15065 Main could indicate the accumulation of these larger crystals. Crystal coarsening is another possible cause for the curved shape of the CSD plot as there is a downturn and positive slope in the smallest crystal size bins which is an indication that larger crystals were growing at the expense of smaller crystals (Higgins, 2011). Equally, the sample could have undergone grain coarsening and/or accumulation.
The CSD plot for 15065 Vuggy and 15065 Main suggests much slower cooling rates than the other Apollo 15 mare basalts. However, the vuggy region experienced slightly quicker cooling rates than 15065 Main, with steeper slopes yielding shorter residence times of 350-1132 days. The shape of the 15065 Vuggy plot is complex with a transition from a concave-up to convex-up profile around 4 mm. The initial concave-up shape could suggest the accumulation of crystals in the ∼ 2.5-4 mm range. The change in slope and convex-up profile of the slope for crystals larger than 4 mm in length could be the result of crystal fractionation. This may provide a further reason as to why we do not observe the 15065 Vuggy CSD plot extending to the same large crystal lengths as the 15065 Main.

SDP Analysis
The SDP plots for olivine and pyroxene crystals in quartz-normative and olivine-normative samples are shown in Figure 9. The R values for olivine in the olivine-normative basalts are all within 0.67-0.89, and the R values for pyroxene in samples from both suites (excluding 15065) range from 1.03 to 1.23 (Table 1). The olivine and pyroxene values form two distinct groups with plagioclase in 15475,15 plotting in between the two clusters ( Figure 9). All pyroxene analyses plot below the RSDL and TFL, meaning they form clustered touching networks. Olivine R values are all below the RSDL and are clustered, but they span across the TFL. The olivine in 15105,6 are non-touching, those in 15536,7 lie close to the boundary between touching and non-touching, and the olivine in 15555,209 form a touching framework. Apollo 15 SDP values for olivine and pyroxene plot within the ranges obtained for Apollo 17 mare basalts. The SDP plot for thin sections of 15065 is shown in Figure 10 with pyroxene values for other quartznormative and olivine-normative samples shown for reference. We cannot calculate an overall SDP for 15065 Main and 15065 Vuggy because it is not possible to calculate the nearest neighbor distances for crystals in separate thin sections. All the individual thin sections plot below the RSDL and TFL lines, indicating that the pyroxene forms a clustered touching network like that of the other Apollo 15 mare basalts.

DISCUSSION
Are There Textural Similarities or Differences Between the Quartz-Normative and Olivine-Normative Suites?
We can use pyroxene CSD trends to investigate potential differences in crystallization histories between the quartz-and olivine-normative suites as this mineral phase is present in all the samples. When plotted together (Figure 11), two groups of CSD trends emerge, which do not correspond to the two respective sample suites. The first group, comprising 15105,6 (ON), 15536,7 (ON), and 15597 (QN), shows relatively steeper gradients, indicating faster cooling rates, and maximum pyroxene crystal lengths of <1.5 mm. The second group, comprising 15555,209 (ON), 15125,6 (QN), and 15076,68 (QN), has shallower gradients, indicating relatively slower cooling rates, and larger corresponding maximum crystal lengths of ∼ 4 mm. We note that the crystal shape estimates for pyroxene in both olivine-and quartz-normative samples have comparable intermediate and long axis values, and all shape estimates had R 2 > 0.8 (Table 1). Therefore, FIGURE 9. Spatial distribution plot (after Jerram et al., 1996Jerram et al., , 2003 showing R-value versus porosity (% matrix) for Apollo 15 olivine-normative (shades of green) and quartz-normative (shades of blue) mare basalts. The random sphere distribution line (RSDL) and the touching framework line (TFL) are shown as solid and dashed lines, respectively. Included for reference are the fields for olivine in Martian meteorite nakhlites (after Balta et al., 2015;Udry & Day, 2018) and shergottites (after Rahib et al., 2019). Literature values for pyroxene, olivine, and plagioclase in Apollo 17 mare basalts are from Neal (2015, 2018). Pyroxenes and plagioclases in this study fall in the range of clustered touching frameworks (R = 0.9-1.2, porosity = 51%-74%), while olivines span the region between non-touching and touching clusters (R = 0.9-1.2, porosity = 88%-93%). (Color figure can be viewed at wileyonlinelibrary.com) FIGURE 10. Spatial distribution plot (after Jerram et al., 1996Jerram et al., , 2003 showing R-value versus porosity (% matrix) for pyroxene from the main (yellows) and vuggy (reds) portions of 15065. Pyroxene values for other quartz-normative (blue) and olivinenormative (green) samples in this study are also shown. Included for reference are Apollo 17 literature values for pyroxene (Donohue & Neal, 2018). There is a larger spread in the data for 15065 than between the other Apollo 15 samples; however, this may be attributed to the coarse-grained nature of 15065 meaning fewer crystals could be analyzed per thin section. (Color figure can be viewed at wileyonlinelibrary.com) differences or similarities in the pyroxene CSD trends between suites cannot be attributed to poorly defined or different crystal shape estimates.
The two different groups of pyroxene cooling rates could actually be providing information about the thermal history of the two suites and be indicative of the relative positions of the samples within respective quartznormative and olivine-normative lava flows. The cooling rates of the samples would provide a proxy for the relative depth within the lava flow from which samples originated. In highly simplified terms, samples that cooled slower are likely sourced from relatively deeper and closer to the center of the flow than those with faster cooling rates (e.g., Jaeger, 1961). This is discussed more in the Relative Position of Samples in Quartz-Normative and Olivine-Normative Lava Flows Section. As the two groups of pyroxene CSD trends contain samples from both the quartz-normative and olivine-normative suites, these findings suggest that there are commonalities between cooling rates and the thermal histories of pyroxene crystals across the quartz-and olivine-normative suites.
Similarities between quartz-normative and olivinenormative basalts are also seen in SDP analysis of pyroxene. Pyroxene is the dominant phase in most thin sections accounting for 34%-49% by area and forming clustered touching frameworks (Figure 9). Conversely, olivine in olivine-normative samples fall closer to the TFL, with 15536,7 showing a non-touching framework. This is consistent with olivine crystallizing before pyroxene within these samples. In studies of terrestrial komatiites, the relative position of samples within lava flows could be discerned from compaction trends . With the Apollo 15 data, we are unable to make similar inferences as the samples do not show a clear compaction trend and are instead more clustered. This could be due to the smaller number of samples used in this study.
Although olivine was only present in measurable amounts in the olivine-normative samples, the CSD trends for all three samples did display a curve/kinked appearance (Figure 6), showing clear evidence of a cooling history more complex than simple linear cooling rates or crystal coarsening. To investigate this further, we used olivine size and compositional data from QEMSCAN mineral phase maps of 15555,209 from Bell et al. (2020) to consider how olivine composition varied with crystal length. Individual olivine crystals were manually sorted based on the forsterite number (Fo # ) of the crystal core. Figure 12 shows that the largest olivine crystals are those with cores of Fo 70 , suggesting an initial period of slow cooling to produce these large crystals. This is followed by a trend of decreasing crystal size and decreasing number of crystals over the range of Fo 60 to Fo 20 . There is a spike in the number of olivine with Fo 10 cores, which are likely to have nucleated rapidly in the final stages of crystallization. From the size data (Figure 12b), there is no evidence to suggest the mixing of two crystal populations. The data collected from QEMSCAN mineral phase maps correspond with what is seen in the CSD plot for 15555,209 ( Figure 6) and support our interpretation of two-stage cooling to form these olivine crystals. Based on their CSD trends, we suggest that olivine-normative samples 15536,7 and 15105,6 also experienced a two-stage cooling history. The process which generated this twostage cooling could have been initial growth of more forsteritic cores (Fo 70 ) in the subsurface during magma transport followed by growth of more fayalitic cores (Fo 10 ) in lava flows on the lunar surface.
Plagioclase CSD analysis was only conducted on quartz-normative sample 15475,15. The CSD trend indicates relatively linear cooling with a kink in the largest crystal size fraction (Figure 7) which could be an indication of crystal coarsening or plagioclase accumulation. Sample 15475 has one of the lowest bulkrock MgO concentrations of all the quartz-normative samples in this study. Despite this, there is little textural evidence of crystal fractionation, which suggests that early-formed phenocrysts must be efficiently separated from eruptible magma in the lunar subsurface. The slope and residence times for 15475,15 are some of the shallowest and longest of the samples analyzed, which is to be expected if this sample experienced one of the slowest cooling rates of all the Apollo 15 mare basalts (Takeda et al., 1975).

Origin of Sample Heterogeneities in 15065
The CSD and SDP patterns in sample 15065 are different from any of the other quartz-or olivinenormative basalts (Figures 8 and 10). The large crystal sizes in the CSD plots and the extremely shallow gradients are consistent with 15065 being the most slowly cooled sample in this study. This is further supported by the curved shape of the 15065 Main plot (Figure 8c) which suggests either crystal coarsening or accumulation of large pyroxene crystals. Previous studies have suggested that the vuggy, more mafic portion of the sample, may be due to the accumulation of pyroxene crystals compared with the main portion of 15065 (Ryder, 1985). The trend for 15065 Vuggy is complex; although it may show signs of accumulation of crystals <4 mm and a broad overall concave-up trend, we also see a change slope to convex-up at crystal sizes of >4 mm which is indicative of fractionation.
The SDP analysis of 15065 is also different from the other Apollo 15 mare basalts in this study ( Figure 10). Although some of the variation between 15065 samples may be due to the small number of crystals (>250) in some thin sections, it is still clear from those samples with a larger number of crystals (>250) that they do not plot within the same region as pyroxenes from the other Apollo 15 mare basalts. Pyroxene R values for 15065 samples <1, suggesting decreased sorting relative to the other samples ( Figure 10). This suggests accumulation of pyroxene phenocrysts  which is consistent with the curved (accumulation) shapes of the CSD graphs and with 15065 being slowly cooled and possibly originating from deep within a lava flow (Grove & Walker, 1977;Lofgren et al., 1975;Onorato et al., 1979).
The coarse grain size of 15065 poses a barrier to CSD and SDP analysis, as crystals are more likely to intersect the edge of the thin section and be excluded from further FIGURE 12. Analysis of how crystal size varies with composition using QEMSCAN mineral phase maps of olivine in 15555,209 (after Bell et al., 2020). (a) Crystals were separated based on forsterite core compositions which ranged from Fo 70 to Fo 10 , as determined via QEMSCAN analysis. The most probable core composition is Fo 60 the least abundant core composition is Fo 20 . (b) The max major axis length calculated using ImageJ for each core composition is also shown. In each plots, we see a late spike in the Fo 10 range both in the number of crystals and the crystal size. (c) QEMSCAN images showing color-coded olivine compositions only, in a progressive crystallization sequence from Fo 80 to Fo 10 . The gray region corresponds to the area of other minerals and the extent of the thin section. (Color figure can be viewed at wileyonlinelibrary.com) analysis. As a result, there is a high possibility that any thin section made of 15065 would incur the same issues and the largest crystals would always be excluded from further analysis. Inferences about this sample and the possible origins of the vuggy portion should be made with this factor in mind. On a macroscopic scale, the vuggy and main portions of 15065 are the only visible evidence in our samples to suggest any process more complex than simple two-stage cooling. Whereas CSD analysis of 15065 may be of limited use due to its coarse grain size, it is clear that further analysis of the vuggy portion is needed to fully understand this sample as many previous studies have only considered the main portion of 15065.

Petrogenic Relationship Between the Quartz-Normative and Olivine-Normative Basalts
The two main opposing hypotheses as to the origin of the difference in bulk rock chemistries between the quartznormative and olivine-normative are that (1) the two suites reflect magma derived from different mantle source compositions or melting systematics (Chappell & Green, 1973;Rhodes & Hubbard, 1973;Snape et al., 2019;Snyder et al., 2000) and (2) the two suites are from the same source but the difference in composition is the result of differences in magmatic history with the quartznormative basalts undergoing multiple stages of opensystem fractional crystallization compared with the olivine-normative basalts (Schnare et al., 2008).
We found no systematic textural differences between the quartz-normative and olivine-normative suites and, as such, a shared cooling history (and by extension perhaps a shared mantle source) cannot be ruled out on the basis of our data. Through CSD analysis alone, we also cannot exclude the possibility that the quartz-normative basalts experienced more fractional crystallization than the olivine-normative basalts, despite the lack of fractionation signature in the CSD analyses. This suggests that crystal fractionation in magma chambers within the lunar crust allowed large crystals to settle out early in the magma's history to form non-eruptible crystal-rich bodies, long before the crystal population observed in the samples began to grow. If none (or very few) of the early-formed crystals that grew in subsurface magma chambers are reentrained and erupted, then evidence of sub-surface crystal fractionation cannot not be captured by CSD analysis.

Relative Position of Samples in Quartz-Normative and Olivine-Normative Lava Flows
As well as providing information about the magmatic histories of the samples, CSD analysis can also provide estimates of the relative position of mare basalts within a lava flow on the lunar surface (Day & Taylor, 2007;. Figure 13 shows an estimate of the relative position of samples included in this study in respective quartz-normative or olivine-normative lava flows based on calculated cooling rates. The schematic lava flow model is based on an inflated pahoehoe lobe split into three main sections: an upper crust, core, and lower crust (Self et al., 1998). For the purposes of comparison, we plot the cooling rates and normalized height of quartznormative samples and olivine-normative samples within their respective lava flows. Cooling rates were determined using the maximum residence time for pyroxene in each sample (except for 15475,15 as plagioclase was the only mineral for which we collected CSD data for in that sample) and an average mare basalt crystallization interval of 200°C (Anand et al., 2006;Day & Taylor, 2007). For reasons outlined in the methods section, the calculation of residence times from CSD analysis is, at best, a relative approximation and so the resulting cooling rates are also caveated with the same uncertainties.
Cooling rates across the two suites varied from 0.13°C h À1 (15105) to 0.0035°C h À1 (15065 Main). The quartz-normative samples extend to slower cooling rates (0.08-0.0035°C h À1 ) than the olivine-normative basalts (0.13-0.031°C h À1 ). The vitrophyric quartz-normative sample 15597 and olivine-normative sample 15105 have the two fastest cooling rates and would likely both be from the chilled top or bottom of respective quartz-normative and olivine-normative flows in the upper lava crust. Samples with progressively slower cooling rates are found at increasing depths toward the core of the flows, with 15065 Main being the slowest cooled sample across both suites. Olivine-normative samples follow the same pattern as quartz-normative samples, but there are no samples in this study from the same depths within the olivinenormative flow as 15065 Vuggy, 15475, and 15065 Main are found in the quartz-normative flow. As expected, the sample position based on relative cooling rate broadly correlates with the grain size and textures of samples ( Figure 13), with the fastest cooling samples having the finest grain size (15597 and 15105) and the slowest cooling basalts having the coarsest grain size (15065). This is also true when comparing between flows: for example, quartznormative sample 15076 and olivine-normative sample 15555 have similar textures and cooling rates of ∼ 0.03°C h À1 . Therefore, assuming similar absolute flow thicknesses, these samples may originate from the same depths within the two separate quartz-normative and olivine-normative flows.

Comparison with Other Apollo Mare Basalts and Lunar Meteorites
Previous studies have conducted CSD analysis on Apollo 12 (Fagan et al., 2013;Neal et al., 2015), Apollo 14 (Fagan et al., 2013;Hui et al., 2011;Neal et al., 2015), Apollo 16 , and Apollo 17 Fagan et al., 2013;Neal et al., 2015) mare basalt samples. Work by Hui et al. (2011) and Donohue and Neal (2015) used CSD analysis to determine the petrogenesis and cooling histories of basalt groups within Apollo 14 and Apollo 17 samples, respectively. In addition, other studies (Fagan et al., 2013;Neal et al., 2015) focused on using CSD analysis as a nondestructive method for determining whether a sample was a crystalline impact melt or an endogenous basalt produced by lunar volcanism.
In all cases, our analysis is comparable to other Apollo samples in terms of both crystal shape estimates and crystal size ranges. The olivine, pyroxene, and plagioclase CSD profiles for other Apollo mare basalts vary from linear/sublinear to curved concave-up, which is to the same as what we see for the Apollo 15 samples. In situations where samples have curved CSD profiles, crystal coarsening has often been suggested as a possible cause Fagan et al., 2013;Hui et al., 2011;Neal et al., 2015). Figure 14 shows a plot of CSD intercept vs CSD slope for olivine crystals in this study combined with results from Neal et al. (2015). This plot was used by Neal et al. (2015) to distinguish basaltic impact melts from endogenous basalts. Our Apollo 15 olivine-normative samples plot within the endogenous basalt range and have CSD slope and intercept values comparable to mare basalts from the Apollo 12 and Apollo 17 landing sites (Figure 14).
Crystal residence times, often reported in CSD studies, correspond to an average crystal residence time FIGURE 13. Model of a mare basalt lava flow (after Day & Taylor, 2007;Self et al., 1998) showing the relative position of samples within this study. The inferred stratigraphy (a) includes a chilled margin and lower crust, a core region with vesicle cylinders (VCs), vesicle sheets (VSs) and pipes (P), and an upper crust with vesicular zones (VZs). The lava flow thickness estimate of 10 m is an approximation based on photographs of outcrops in the walls of Hadley Rille (Howard et al., 1972). Cooling rates are calculated from crystal residence times of pyroxene (and plagioclase for 15475) and were used to infer the sample location within the lava flow. Relative quartz-normative sample positions in a hypothetical quart-normative flow are shown in (b) and the same is shown in (9c) for olivine-normative samples in a hypothetical olivine-normative flow. A schematic cooling rate trend is also shown in both instances (Day & Taylor, 2007). Cross-polarized light images showing the textures of the thin sections are shown in their relative order in (b) and (c). The scale bar is the same for all thin section images. Sample textures correspond with what would be expected based on relative cooling rates, with degree of crystallinity and grain size increasing with depth within the flow. (Color figure can be viewed at wileyonlinelibrary.com) and is not an absolute time value applicable to all crystals in that system (Marsh, 1988). Straight line portions of a CSD graph, from which crystal residence times are calculated, can represent a wide range in crystal sizes and, in turn, a wide range in values for individual crystal residence times. In addition, crystal residence time calculations assume a constant growth rate for all crystals within the sample (Cashman, 2020). Although the significance of crystal residence timescales should be treated with caution, they do provide a way to roughly compare the magnitudes of timescales (i.e., days, years, months) between different samples. Crystal residence times in this study were calculated using growth rates that are of the same order of magnitude as those used in other Apollo mare basalt CSD studies Neal et al., 2015). The residence times discussed here show that those determined for the Apollo 15 data in this study are on the same timescales (i.e., months to years: Table 1) as other Apollo landing site mare basalts. Olivine residence times for Apollo 15 low-Ti mare basalts (26-125 days) are at the lower end of the range of 65 days to ∼ 5 years found in other studies of Apollo mare basalts Neal et al., 2015). Pyroxene residence times for high-Ti Apollo 17 samples are in the range of 46-247 days compared to 26-370 for Apollo 15 low-Ti mare basalts (excluding the much longer residence times of 15065 of up to 2430 days or 6.7 years; . A wide range of mare basalt plagioclase residence times can be found in the literature from 51 days  to 29.2 years , and our Apollo 15 plagioclase CSD results are at the lower end of the scale at ∼ 1.5 to 3 years. In addition, lunar meteorites, such as low-Ti North West Africa (NWA) 032 and the La Paz (LAP) mare basalt meteorites (LAP 02205/02224/02226/02436) have also been analyzed using CSD and SDP analyses to determine a potential lava flow geometry based on cooling histories (Day & Taylor, 2007). Results for crystal shape, size range, and CSD profiles from these samples are all comparable to our Apollo 15 mare basalts (this study) and those from other Apollo missions Fagan et al., 2013;Hui et al., 2011;Neal et al., 2015;Xue et al., 2021).

Recommendations for Processing of Future Returned Samples
A limitation of CSD analysis (particularly of Apollo and other meteorite samples) is the number of crystals in thin sections of coarse-grained samples, such as 15065. With samples of <250 crystals, it is challenging to draw robust conclusions from the data. This situation is analogous to that faced when determining the bulk chemistries of coarse-grained samples from only a few grams of material which may not be representative of the sample as a whole (e.g., Ryder & Schuraytz, 2001). We therefore recommend that future thin sections be made using standard rectangular glass slides where possible (as opposed to 1inch round slides), especially for coarse-grained samples, to maximize the number of crystals in each section. We acknowledge that for some coarse-grained samples this still may not provide enough crystals for robust textural analysis. However, this is not an issue for the majority of samples and does not detract from the value and potential of CSD analysis of extraterrestrial materials. In this study, over 50% of the CSD analysis included a data set size of >1000 crystals, and aside from thin sections of 15065 there was only one instance out of 11 analyses where <250 crystals were present (olivine in 15105,6). Most importantly, the wealth of textural information gathered by CSD analysis is nondestructive, making it a powerful tool for the analysis of current mare basalt samples (i.e., Apollo, Chang'e 5, Luna, and lunar meteorites), and those samples yet to be returned on future missions to planetary bodies with volcanic surfaces (Li et al., 2019;Muirhead et al., 2020;Qian et al., 2021). Values for Apollo 14 impact melts and Apollo mare basalts (other than Apollo 15) are for crystals <0.4 mm . Slope 1 values (Table 1) for Apollo 15 samples were plotted as they are comparable to the 0.4 mm cutoff in crystal size of the other samples. The three olivine-normative Apollo 15 mare basalts in this study plot within the endogenous basalt region and are within the same range as other Apollo 12 and Apollo 17 mare basalts. (Color figure can be viewed at wileyonlinelibrary.com)

CONCLUSIONS
We conducted CSD and SDP analyses on crystals in three olivine-normative and five quartz-normative thin sections to determine their crystallization and emplacement histories. We also examined seven thin sections of sample 15065 to explore the origins of textural heterogeneities within this unusual Apollo 15 mare basalt.
We find evidence of two-stage cooling histories in olivine from olivine-normative basalts and we relate this to how crystal size varies with composition. We find two groups of sublinear pyroxene CSD trends, which relate to differences in cooling rate. Both groups contain both olivine-normative and quartz-normative samples, suggesting that there are similarities in cooling rates between the two suites. We attribute differences in pyroxene-derived cooling rates to relative positions within a typical mare basalt lava flow, with samples in different groups coming from comparable depths within their respective lava flows. Some CSD profiles for pyroxene and plagioclase display curved trends, which is likely due to crystal coarsening, although we cannot rule out accumulation. The main and vuggy portions of 15065 have different pyroxene CSD trends, but the coarse grain size of this sample means that we have limited confidence in interpretations of these CSDs in terms of understanding the cause of the sample heterogeneity. However, cooling rates and overall curved CSD trends for 15065 Main further confirm that 15065 is likely to have formed comparatively deeper within the lava flow.
With the rapid development of new analytical and quantitative petrological techniques, it is important to consider what more we are able to learn about lunar magmatic process from existing Apollo mare basalts. Nondestructive, quantitative textural analysis techniques such as CSDs and SDPs provide a method by which the magmatic histories of these rare and precious samples can be further understood while also preserving them for future studies.