Correlated compositional and mineralogical investigations at the Chang′e-3 landing site

The chemical compositions of relatively young mare lava flows have implications for the late volcanism on the Moon. Here we report the composition of soil along the rim of a 450-m diameter fresh crater at the Chang′e-3 (CE-3) landing site, investigated by the Yutu rover with in situ APXS (Active Particle-induced X-ray Spectrometer) and VNIS (Visible and Near-infrared Imaging Spectrometer) measurements. Results indicate that this region's composition differs from other mare sample-return sites and is a new type of mare basalt not previously sampled, but consistent with remote sensing. The CE-3 regolith derived from olivine-normative basaltic rocks with high FeO/(FeO+MgO). Deconvolution of the VNIS data indicates abundant high-Ca ferropyroxene (augite and pigeonite) plus Fe-rich olivine. We infer from the regolith composition that the basaltic source rocks formed during late-stage magma-ocean differentiation when dense ferropyroxene-ilmenite cumulates sank and mixed with deeper, relatively ferroan olivine and orthopyroxene in a hybridized mantle source.


Supplementary
Two types of rock, light-toned, coarse-grained rock and dark, fine-grained rock, as seen by the Panoramic Camera at the CE3-0006 site (a) and -0008 site (b). See Supplementary Note 1 for more discussion.  Table 1 Elemental peak-area ratios to Si by CE-3 APXS measurements. Analysis 0006_1 was in the test mode, i.e., with a surface-to-detector separation of ~5 cm. Analysis 0006_2, 0006_3, and 0008 were measured at ~2 cm distance between detector and regolith surface. Analysis 0006_1 was in the test mode. Mean_0006 is the average value of CE3-0006_2 and CE3-0006_3.
** Plag An is defined as An/(An+Ab+Or)×100.    Figure 1 shows the Panoramic view of four compositional sites. In general, CE-0005 and CE-0008 are more blocky with more rocks than the other two sites. Especially at the CE-0007 site, rover tracks and the deep trench made by the rover suggests a thicker regolith than at the other sites. Two types of rocks were observed by Yutu during its traverse: one is the dominant light-toned, coarse-grained basalt (similar to Outer Fence, which is ~4 m long by 1.5 m high), and the other is darker, fine-grained basalt. The CE3-0006 regolith surface appears to be smoother than the CE3-0008 site, which is expected, owing to the proximity of CE3-0008 to the Zi Wei crater. The CE3-0008 regolith has more coarse rock fragments from crater ejecta. The chemical composition may vary but the color may be due to the texture (grain size) of the rock. Both factors may account for the observations, but we prefer the latter on the basis of the false-color images and Panoramic Camera data. The upper parts of the lava flow should be parts of the flow that underwent rapid cooling, leading to a fine-grained rock texture, whereas the coarse-grained rock would come from a deeper level of the flow and from deeper in Zi Wei crater. Given the limitation of only two soil samples being measured by APXS (CE3-0006 and -0008), if we assume that the soils derive only from the local basalts and that these basalts are all from the same (upper-most, Eratosthenian) flow unit, then perhaps the analyses reflect a simple fractional crystallization trend of slightly more magnesian and olivine-rich (corresponding to the light-toned coarse-grained rock, Supplementary Figure 2) to more iron and pyroxene-rich lava (corresponding to the CE3-0006 soil and related to the dark, fine-grained rock, Supplementary Figure 2).

Estimation of plagioclase abundance by photo-image analysis
We estimated the abundance of plagioclase in one of the nearby boulders, Outer Fence (Supplementary Figure 3), at the landing site, which was likely excavated from the ~450 m diameter Zi Wei crater west of the landing site. This boulder is ~1.5 m tall and dust appears to be minor or absent, thus it is conducive to image analysis. As shown in Supplementary Figure 3, the bright grains likely represent plagioclase with high albedo. The reddish, bluish, or greenish-colored grains are most likely the mafic minerals of these rocks. We determined the pixels' DN values and the threshold of each image. The statistics of the RGB channels are shown in Supplementary Figure 3. The data show a single-mode distribution of image DN values, suggesting the DN variation of the rock is not abrupt. However, we still find a step of ~20 DN for the bright grains compared with surrounding pixels. For Image 1, the threshold of the blue channel DN is 130, indicated by the red pixels in the lower image. For Image 2, the threshold of the blue channel DN is 125. The red pixels of both images make up ~17% (14.7±1.3% and 18.4±1.6%, respectively) of the images. This percentage provides a lower limit estimate of volume percentage of plagioclase in the rock. In order to accommodate the micro-topography of the rock surface, we subdivided the two images into A-D areas respectively, and we also lowered the thresholds to 125 for Image 1 and 120 for Image 2. The added pixels are shown as yellow in Supplementary Figure 3 and the results are shown in Supplementary  Table 5)). However, we cannot estimate the fine grains of plagioclase in the rock by this method. Note that the values of DN are also influenced by the roughness of the rock and the observation/illumination geometry, thus one cannot fully capture the mineral mode of plagioclase in this way. In any case, this image-based analysis provides a constraint on the mineral mode of plagioclase. Our APXS data-reduction results suggest ~33 vol.% of plagioclase in the soils of the landing site. This value is in general agreement with the photo-based results here (~20±4 vol.% on average, as a lower limit), considering that the undetermined abundance of fine-grained plagioclase must be added to this value. In summary, our photo-image analysis suggests that plagioclase phenocrysts make up ~20 vol.% of the rock. If the rocks are richer in Al 2 O 3 than indicated by our APXS data reduction (e.g., ~12.6 wt.% 4 ), then the plagioclase content would be over ~40 vol.%.

Albedo variations of the lunar soils near CE-3 landing site
We examined reflectance variations of the lunar regolith near the CE-3 landing site by using LROC NAC image data. We compared the average reflectance of the regional basaltic regolith with reflectance values for different areas at the landing site that are relatively flat and thus comparable.
Supplementary Figure 4 a and b shows variations in reflectance of soils near the CE-3 landing site. The image shown in (a) is centred at 44.1223ºN and 340.5000ºE, and extracted from NAC image M1142582775R, map in sinusoidal projection (incidence = 77º, emission = 2º, phase = 75º). Most of the variance in reflectance is associated with impact craters, but there is also an increase of reflectance on approach to the rim of Zi Wei crater. The mature surrounding basaltic regolith surface has an average I/F value of 0.0082 whereas the rim of Zi Wei crater rises to ~0.010. We selected 11 sites near the CE-3 landing and traverse sites, as shown in Supplementary Figure 4 c. Typical 8×8 pixel homogeneous areas were selected to average the reflectance values in order to eliminate small-scale topographic effects, except for the two CE-3-landing sites (CE-3-landing and CE-3-landing2, 10×10) and the four traverse sites (3×3). The values more-or-less occur within the range of 0.008 to 0.010.
Those sites near the Zi Wei crater rim (CE-3 landing, E1, S1, as shown in Supplementary Figure 4 c,d) are very similar to the soils ~200 m farther away (S3 and E2). The darkening effect of space weathering on the surrounding regolith away from Zi Wei crater is subtle in the NAC image, indicating that the ejecta, from which the soils derived (e.g., CE-3-landing, S1) near the crater rim is compositionally similar to the regolith derived on top of the Eratosthenian lava flow away from the crater (e.g., S3, E2). If the soils at the CE-3 site had derived from a deeper, more aluminous basalt 4 , there would likely be a more obvious increase in reflectance between the ejecta and surrounding regolith.

MGM analysis
We applied the Modified Gaussian Model (MGM) 5,6 to deconvolve the spectral bands as measured by the CE-3 VNIS. We have attempted to use different mineral assemblages(e.g., HCP, HCP+LCP, HCP+OL, HCP+LCP+OL) and various numbers of Gaussian bands (e.g., 4, 5, 6, 7, 8, and 9, etc.) to deconvolve the overlapping absorption features of the CE-3 VNIS spectra. Figure S6 shows examples of different numbers of Gaussian bands used in our MGM modelling. Note that an instrumental artefact peak (~1450 nm) is required for better spectral deconvolution.
One The precise location of the M2 band (~1050 nm) of olivine in the VNIS spectrum is difficult to determine via MGM deconvolution. Three factors can be influential: (1) overlap with the 1 μm (M1) band (~1030 nm) of HCP; (2) Fe-rich olivine would have an M2 band with low intensity and a shifted central position (Sunshine and Pieters, 1998) 6 to 1080 nm that can cause overlap with the M2 band shoulder (~1150 nm) of Fe-rich pyroxene; and (3) pyroxene M1 bands would dominate the 1 μm band from a mineral mixture whose olivine content is less than 50 vol.% 7 .
Note that there are substantial uncertainties 8 (e.g., band positions and widths) in MGM modelling of VNIS spectra if different parameters (e.g., wavelength range, No. of bands, and/or band parameters as shown in Supplementary Table 6) are chosen. In order to find the best MGM modelling results with the lowest uncertainties, we tested many different mineral mixtures and spectral parameters (different numbers of bands, band positions, band widths, and band strengths), guided by chemical and mineralogical constraints from APXS results. The results shown in Supplementary Figure 7 have the lowest uncertainties among all attempted combinations, and at the same time, the best matches with standard spectra of mafic mineral endmembers were achieved.
The MGM analysis of pyroxene mixtures with laboratory and OMEGA data determined that the relative pyroxene abundance (LCP/(HCP+LCP)) can be constrained to ±10% (Kanner et al., 2007) 8 . For MGM analysis of laboratory olivine, compositional estimation can be accurate to within ~5-10% (5-10 Fo units) (Sunshine and Pieters, 1998; Isaacson and Pieters, 2010) 6,9 . When the remotely-sensed M 3 data with lower signal-to-noise is applied, the MGM composite error would be on the order of 20 Fo units (Isaacson et al., 2011) 10 . As mentioned above, the uncertainties of our MGM analysis of CE-3 VNIS data have been constrained based on the correlated chemical data from the CE-3 APXS. Therefore, the composite errors of MGM analysis in this study, although difficult to quantify, would be the same range or even smaller if considering the higher quality data obtained in-situ by CE-3 VNIS compared to remotely acquired data with increased noise level and decreased spectral resolution by lunar orbital spectrometers.