Supporting Information (SI) For

Supporting Information (SI) for

Volcanic history of the Imbrium basin: A close-up view from the lunar

rover Yutu

Jinhai Zhang, Wei Yang, Sen Hu, Yangting Lin, Guangyou Fang, Chunlai Li, Wenxi

Peng，Sanyuan Zhu, Zhiping He, Bin Zhou, Hongyu Lin, Jianfeng Yang, Enhai Liu, Yuchen Xu, Jianyu Wang, Zhenxing Yao, Yongliao Zou, Jun Yan, Ziyuan Ouyang

S1. Topographical features of the Chang’e-3 landing site ...... 2 S2. APXS experimental method ...... 3 S3. Geochemical features of Chang’e-3 landing site...... 8 S4. VNIS spectra decoding method...... 9 S5. Lunar Penetrating Radar data processing...... 15

S6. SI References...... 29

● Supplementary Figures S1 to S21

● Supplementary Tables S1 to S3

S1. Topographical Features of the Chang’e-3 Landing Site Figure S1 shows the landing site of Chang’e-3 in the northeastern Imbrium basin. It locates on the young and high-Fe, high-Ti lava flow that overlying on the old and low-Ti basalt unit emerging about 10 km north. The young lava was dateed 2.0-2.3 Ga (1), and the old unit was dated 3.5 Ga (2). Another basalt unit, with intermediate FeO and

TiO2 contents (3) or referred to as higgh-Al (4), can be recognized (Fig. S1c). However, age of this intermediate basalt unit is indistinguishable from the old low-Ti basalt. From Apollo orbital photography, Schaber (5) outlined three main lava flows extended about 1200 km, 600 km, 400 km from the venting region near the Euleer crater 700 km southwest to the landing site, and they overlie on the old low-Ti basalt unit (Fig. S1b). Chang’e-3 landed on the youngest lava flow that extended 1200 km. The landing site has abundant rocky ejecta (Fig. S2), which cover about 5.7% of the surface.

Fig. S1. The geological map of Mare Imbrium, showing Chang’e-3 landed on the young high-Ti basalt unit that probably overlies on other two units. (a) The image of Mare Imbrium, taken by Chang’e-1; (b) Disstribution of three main lava flows originated from a venting area near the crater Euler, modified from (5); (c) Compositional map of basalt units, after (3); (d) Modal age map after (1), except for 3.5 Ga from (2).

Fig. S2. Typical landscape at the Chang’e-3 landing site, showing abundant rocky ejecta.

S2. APXS Experimental Method The Active Particle-induced X-ray Spectrometer (APXS) was equipped on the robotic arm of Yutu, and it consists of a dual radioactive source of 55Fe (half-life of 2.73 year, 470 mCi) and 109Cd (half-life of 1.27 years, 45 mCi) and a Si-drift detector with an energy resolution of 135 [email protected] keV (6). The surfaces of targets were irradiated by the X-ray from decay of both 55Fe and 109Cd, and the fluorescent X-ray spectrum was counted by the Si-drift detector. The data used in this study are 2B level, which have been corrected for energy calibration, working distance, effect of temperature and dead time (6). Twelve elements, including Mg, Al, Si, Ca, Ti, K, Cr, Fe, Sr, Y, Zr and Nb, have been detected in the lunar soil at the landing site (Fig. S3). The K and K lines of Cu were from the device material, and K lines of Mn and Fe overlapped by Compton scattering peaks of the 55Fe source. Two APXS analyses of the lunar soil have been carried out, together with the measurements of the onboard basaltic working reference. All analyses have been corrected for background and peak overlapping. The compositions of the lunar soil were calculated from the net counts, calibrated with the onboard basalt reference and a set of standards that were measured in laboratory. S2.1 Measurements of the standards and working references in laboratory In order to calibrate the in-situ analyses of the lunar soil at the landing site, a total of 10 standard materials and working references have been measured in laboratory within a period from April 24, 2013 to July 2, 2014. These analyses were carried out in three

3 analysis sessions, using the onboard APXS or its duplicate (Table S1). The standards used in this study include one China National Standards basalt (GBW07105, also labeled as GSR-3 (7)), 2 Lunar Soil Simullants (CLRS-1, CLRS-2), 2 rock chips of lunar meteorites (NWA 2995, NWA 4734) and 5 terrestrial basalts (DC13-16R, NAO rock, DC13-08, 03JG-2, HBJ4-3). These standards and working references cover the compositional ranges of the lunar soil. The powder sample surfaces (60 mm in diameter) were strickled flat, and the measurements were carried out in a vacuum chamber. The working distance is about 10 mm. Each sample was counted for 30 minutes.

Fig. S3. The X-ray spectra of the lunar soil and the onboard basaltic reference analyzed by APXS on the rover Yutu. Ka and Kb lines of Cu are from the device itself, and Ka lines of Mn and Fe overlapped by Compton scattering peaks of the 55Fe source.

S2.2 Data processing Background of the X-ray spectra: The APXS data were processed using PyMca software (8). The continuum background was fitted using a linear function. For the dual radioactive sourrce, it is difficult to evaluate the spectral background theoretically, which is mainly contributed from elastic and inelastic peaks of the Fe and Cd sources, Compton scattering in the targets and the detector, and fluorescence of the major elements of the targeets. In order to find the smooth and best fitting background, sseveral empirical functions including none background, constant background, linear function, linear polynomial function and exponential polynomial function were applied, and the results are compared in Fig. S4. The major elements, including Al, Si, K, Ca, Ti and Cr, show nearly the same results for the different modes of background; whereas, Mg shows differences between the connstant or none background and the other background functions.

In order to assess the fitting of background, we plotted the net peak areas of the elements versus the recommended concentrations of the standards and working references. It is found that the data can be best fitted by the linear model of background (Fig. S5).

Table S1. The information of the standards and working references. Chemical Analysis Sample Name Rock Type Form Composition session GBW07105 Basalt Powder (1) (a)(c) NAO powder Basalt Powder (2) (a) NAO rock Basalt Rock Chip (2) (a) NWA2995 Lunar Breccia Rock Chip (3) (b) NWA4734 Lunar Basalt Rock Chip (3) (b) DC13-16R Basalt Rock Chip (4) (b) DC13-16P Basalt Powder (4) (c) DC13-08 Basalt Powder (4) (c) 03JG-3 Basalt Powder (5) (c) HBJ4-3 Basalt Powder (5) (c) CLRS-1 Lunar Soil Simulant Powder (6) (c) CLRS-2 Lunar Soil Simulant Powder (7) (c) The experiments were carried out in 3 analysis sessions in ground laboratory: (a) April 24, 2013, with the onboard APXS; (b) December 20, 2013, with the duplicated APXS, and (c) July 2, 2014, with the duplicated APXS. (1) GBW07105 is a basalt of China National Standards; (2) NAO powder and NAO rock are the same basalt in different form, which was collected from Hannuoba, northeastern China, by National Astronomical Observatory (Report for CE-3 APXS scientific verification experiments, File Number: CE3-GRAS-CSSY-004-F2); (3) NWA2995 is a lunar feldspathic breccia, and NWA4734 is a lunar basalt (9); (4) DC13-08 and DC13-16 are low-Ti and high-Ti Emeishan flood basalts, respectively, collected from Dongchuan, southwestern China. DC13-16P and DC13-16R are powder and rock chip of DC13-16, respectively; (5) 03JG-3 and HBJ4-3 are Cenozoic basalts from western Liaoning, northeastern China (10); (6) CRLS-1 is China low-Ti Lunar Soil Simulant standard (Report for low-Ti basaltic lunar soil simulant standard CLRS-1); (7) CRLS-2 is the China high-Ti Lunar Soil Simulant standard (Report for high-Ti basaltic lunar soil simulant standard CLRS-2).

Peak overlapping correction: After background removal, the peak area fitting was performed within the range of channel 50 and channel 1300 (corresponding to energy from 0.70 keV to 16.80 keV) with Pseudo-Viogt functions. The peak area of each element was determined from the deconvolved peak. The uncertainties of the peak areas were estimated < 1-3% for Si, K, Ca, Ti and Fe, <6% for Al, <15% for Mg, < 10% for Sr, Zr and Cr, < 30-50% for Nb and Y. Calibration of decay of the radioactive sources: Since 109Cd source (half-life of 1.27 years) decays quickly than 55Fe source (half-life of 2.73 year), the relative peak counting rates of Fe, Sr, Y, Zr and Nb to those of other elements (the X-ray energy lower 5 than K line of Fe) vary with time. Therefore, the peak area of each element was calibrated to the same intensity of the radioactive source on April 24, 2013, assuming that the low energy X-ray lines of Mg, Al, Si, K, Caa, Ti and Cr were excited only by 55Fe source. Calibration by the standards and working references: After calibration of decay of the radioactive source, the net peak areas of the elements were normalized to the internal reference element, i.e., Mg, Al, K, Ca, Ti, Fe and Cr normalized to Si, whereas Sr, Zr, Y and Nb normalized to Fe. Figure S5 plots the abundance ratios versus the peak area ratios of these elements of the standards and working references. All of the elements show linear correlations, with the regression lines through the origin of coordinates. The two in-situ analyses of the lunar soil at the landing site were plotted in Fig. S5. It is noticed that both analyses are nearly identical to each other, indicative of homogeneity of the lunar soil. Only Y/Fe peak area ratios of both analyses show a significant difference. This is likely due to heterogeneity of phosphates (Y- and REE-bearing phases) in the lunar soil, instead of the analytical uncertainty. Using the calibration lines in Fig. S5, the in-situ analyses of the lunar soil at the landing site can be converted to the elemental abundance ratios. The composition of the lunar soil was then calculated by normalizing the analysis total to 100 wt%, and the results are listed in Table 1.

Fig. S4. Net peak areas of the elements, using different spectral background functions. The sample is the lunar soil LS2.

Fig. S5. Calibration lines of the standards and working references. The peak area fitting was performed with Pseudo-Viogt functions, after linear function background removal. The vertical lines are the measured peak area ratios of the lunar soils.

Fig. S6. FeO vs. Al2O3 plot of the lunar soil at Chang’e-3 landing site, in comparison with: (a) the Apollo lunar soil samples (11), and (b) lunar meteorites (9).

Fig. S7. Histogram of the TiO2 contents of the rims and proximal ejecta of small craters (0.4–4 km). The smaller craters (0.4–1 km) display a peak at ~5 wt% TiO2, and with a nearly flat tail at the lower side. The low TiO2 peak is attributed to the underlying low-Ti basalt excavated by the larger craters. The TiO2 content of the lunar soil at the landing site is indicated by arrow. The TiO2 data of the ejecta of small craters are from (4).

S3. Geochemical Features of Chang’e-3 Landing Site

The lunar soil at Chang’e-3 landing site contains higher FeO and lower Al2O3 than those of Apollo mare soils, plotting within the range of mare basalts (Fig. S6). This result suggests that the lunar soil at Chang’e-3 landing site contains little feldspathic ejecta, which could represent the beneath lava flow. The FeO and TiO2 contents of the lunar soil fill the gap between the high-Ti and low-Ti basaltts (Fig. 3b) (12). The trace incompatible lithophile Y and Zr deviate from Apollo basalts, but can be explained by assimilation of the KREEP component (Fig. 3a) (4).

The TiO2 contents of the rims and proximal ejecta of small craters (0.4–4 km) on the young lava flow have been determined with compositional remote sensing data from

Lunar Prospector Gamma Ray Spectrometer and Clementine (4). The TiO2 content distribution patterns are related with the sizes of the craters (Fig. S7). The smallest craters

(0.4–1 km) show a high-TiO2 peak at about 5 wt% with a flat and low-TiO2 tail. This distribution pattern can be decoded into a high-TiO2 peak and another low-TiO2 peak. The high-TiO2 peak is similar to the in-situ analysis by Yutu and to the whole surface TiO2 distribution (13), suggesting that the smaller craters have not penetrated the most upper high-TiO2 basalt unit. In other words, the thickness of the high-TiO2 basalt unit should be significantly larger than 120 m, the depth of a crater with a diameter of 0.4 km.

The low-TiO2 peak indicates that the other craters have excavated the beneath low-Ti basalt unit. This is confirmed by the low-TiO2 content distribution of the large craters (1– 4 km), which penetrated even deeper and excavated the underlying low-TiO2 basalt unit emerging about 10 km north from the Chang’e-3 landing site (Fig. S1c). The depth of 270 m was calculated for a crater with a diameter of 1 km using the lunar crater mode (http://www.lpi.usra.edu/lunar/tools/lunarcratercalc/), which can be referred to as the maximum depth of the high-TiO2 basalt unit.

S4. VNIS Spectra Decoding Method The mineral modal composition of the lunar soil can be decoded using the modified Gaussian Model (MGM) method from the four VNIS spectra, and the results are summarized in Table S2. The average lunar soil contains 16.4 vol% plagioclase and 17.9 vol% pyroxenes. A modal abundance of 6.3-8.8 vol% ilmenite can be estimated from the

TiO2 content measured with APXS, assuming most TiO2 in ilmenite. This modal composition is consistent with that of the average Apollo mare soils. The FeO contents of the lunar soil was determined from the correlation between FeO contents and Fe values of the Apollo soils, where Fe= -arctan((R945/R750-1.22)/(R750-0.04)) (14, 15), ranging from 18.7 wt% to 19.5 wt% with an average of 18.9 wt%. The decoded FeO content of the lunar soil is consistent with the APXS results within the analytical uncertainties. The

TiO2 contents of the lunar soil can also be determined from the spectra, based on the correlation between the TiO2 contents and the Ti values, where Ti = arctan ((R415/R750-0.42)/(R750-0.00)) (14, 15). The TiO2 contents vary from 5.3 wt% to 9.0 wt%, with an average of 6.6 wt%. The TiO2 contents are higher than the APXS analyses, which might be due to the rough surface of the landing site, which leads to part of the analysis areas darker due to shadow (Fig. S8).

S4.1 Correction of the VIS/NIR and SWIR spectra The onboard VNIS was installed at the front of the rover Yutu, and it consists of a VIS/NIR imaging spectrometer (450–945 nm) and a shortwave IR (SWIR) spectrometer (900–2395 nm) (16). The VIS/NIR data were recorded as images of 256×256 pixels, with a spatial resolution of ~0.6 mm. The SWIR data were integrated from a round area with a radius of 54 pixels, whose center corresponds to the pixel (128, 96) of the VIS/NIR

image. Four analyses have been carried out, and the analysis positions are labeled in Fig. 2. Each of the VIS/NIR image was integrated for 40 min (except for CD006: 18 min, and CD007: 31 min) and each of the SWIR spectrum was integrated for 32 min (except for CD006: 8 min), respectively. The spectrum resolution is 2–7 nm for VIS/NIR and 3–12 nm for SWIR spectrometer. The data have been corrected for dark current, the effect of temperature, radiometric and geometric calibrations, and released as 2B level (17). The 2B level VIS/NIR data were sequentially reduced to repair bad lines (pixel gray-scale slope threshold method) and bad points (discretely bright pixels), to make flat field corrections (global histogram equalization) (18), and then the data were converted to reflectance. The SWIR radiance was directly converted to reflectance. Both VIS/NIR and SWIR reflectances have been calibrated for incident angles of light and solar irradiances (19). The corrected VIS/NIR images are shown in Fig. S8. As VIS/NIR and SWIR spectra were separately measured with different spectrometers, there was discontinuity between them (900–945 nm). An offset was applied to the SWIR spectrum, which shifted the spectrum continuously to the VIS/NIR spectrum. The offset value was determined by minimizing the standard deviation (equation 1) between the overlapping wavelengths.

Table S2. The decoded results from the VNIS spectra of the lunar soils. * Distance R Ti TiO2 Fe FeO OMAT Plagioclase Pyroxene 450 415 750 950 (m) (nm) (nm) (nm) (nm) (wt %) (wt %) (vol %) (vol %) CD005 19.79 0.043 0.039 0.070 0.0611 1.175 5.8 1.487 18.8 0.312 15.0 20.6 CD006 32.06 0.033 0.030 0.057 0.0594 1.154 5.3 1.482 18.7 0.157 16.6 20.3 CD007 38.72 0.027 0.025 0.047 0.0489 1.195 6.4 1.534 19.5 0.158 17.5 17.8 CD008 40.89 0.032 0.029 0.051 0.0559 1.274 9.0 1.489 18.8 0.098 16.3 13.0 Avg. 6.6 18.9 0.181 16.4 17.9 *Distance from the lander; OMAT: optical maturity parameter.

Fig. S8. Micrographs of the lunar soil at the Chang’e-3 landing site, showing a rough surface with shadow areas. The width of the field is ~150 mm.

945 2 error   R   offset R   / N 1 （1）   VIS/NIR   SWIR    900 Where error is the standard deviation between the overlapping wavelengths of VIS/NIR and SWIR spectra, N is the band number of the overlap (N=10), is the wavelength in nm with 5 nm interval, RVIS/NIR() is the reflectivity of VIS/NIR at , and RSWIR() is the reflectivity of SWIR at . The best offsets of CD005–008 are -0.00459, 0.00987, 0.0094 and -0.00655, respectively.

Fig. S9. The calibration curves for TiO2 and FeO contents, derived from the spectra of Apollo soils measured in laboratory (20, 21).

S4.2 TiO2 and FeO contents achieveed from the VNIS spectra The TiO2 content of the lunar soil correlates with the reflectance ratio between bands at

415 and 750 nm, R415/R750, via Ti parameter, Ti (Fig. S9a). The parameter Ti is defined by Ti=arctan[(R415/R750-y0Ti)/(R750-x0Ti)] after Lucey et al. (15), and the R415 was 2 determined from R450 by R415=0.95R450-0.0013 (R =0.99) derived from the spectra of Apollo lunar soils (20), where y0Ti=0.40 and x0Ti=0.0 were optimized to maximize the correlation coefficient between TiO2 contents and Ti of Apollo soils with grain size < 45µm (20, 21). The TiO2 contents of the lunar soil were 5.3–9.0 wt% with an average of 5.44 6.6 wt%, calculated using the best fit curve of TiO2 (wt%)=2.416Ti (Fig. S9a). The calculated TiO2 contents are somewhat higher than the APXS analyses, likely due to shadow effects of the rough surface of the landing site (Fig. S8). Similarly, tthe FeO contents of the lunar soils can be determined from the best fit curve of FeO (wt%)=16.103Fe-5.18 (Fig. S9b), where Fe is defined by Fe=-arctan[(R950/R750-y0Fe)/(R750-x0Fe)] after Lucey et al. (15), and y0Fe=1.23 and x0Fe=0.04 were optimized to maximize the correlation coefficient between the FeO contents and Fe of Apollo lunar soils with grain size < 45µm (20, 21). The R950 value is nearly identical to R945, based on the spectra of Apollo lunar soils (20). The FeO contents of the lunar soil are 18.7–19.5 wt% with an average of 18.9 wt%. These calculated FeO contents are consistent with the APXS measurements within the analytical uncertainties.

S4.3 Major constituent mineral abundance deconvolution The mineral compositions of the lunar soils were deeconvolved from the VNIS spectra using the Modified Gaussian Model (MGM) (22). The possible constituents of the lunar mare soils include impact-induced agglutinate glasses, high-Ca and low-Ca pyroxenes and plagioclase, with minor ilmenite, olivine and volcanic glass (21). Both

12 agglutinate glass and ilmenite have no absorption around 1 m and 2 m bands, and olivine is a minor phase in the lunar mare basalts (<5 vol%) (21). These constituents were not considered in the MGM fitting calculation. In this work, four of the absorption peaks of pyroxenes (orthopyroxene and clinopyroxene)), the absorption peak of plagioclase at 1.25 m and one more peak below 0.5 m were used as the starting parameters for MGM fitting. Figure S10 shows the results of the deconvolution, with all four spectra well fitted.

Fig. S10. MGM deconvolution of the lunar soil spectra at the landing site.

In order to determine the mineral contents of the lunar soils, we processed the spectra of the Apollo soils reported by (20) with the known mineral compositions (21), using the same MGM. Figure S12 plots the deconvolved absorption band strength of pyroxene and plagioclase versus their abundances, respectively. The calibrated spectra are similar to the laboratory measurements of Apollo mare soil samples, showwing absorption at 1 µm and 2 µm responding to the presence of pyroxene and plagioclase (Fig. S11). Based on the correlations of the Apollo soils (Fig. S12), the mineral compositions of the lunar soil at the Chang’e-3 landing site were determined with 17.9 vol% pyroxene (13.0–20.6 vol%) and 16.4 vol% plagioclase (15.0–17.5 vol%) (Table S2).

Fig. S11. Reflectance specttra of the lunar soil meaasured in situ by the rover Yutu.

Fig. S12. Plot of mineral abundances of the Apollo soils versus its natural log band strengths deconvolved from the spectra using MGM method. (a) pyroxene as suum of orthopyroxene and clinopyroxene at 1 m; (b) plagioclase at 1.2 m. The Apollo data from (20, 21).

S4.4 The maturity of the lunar soil The optical maturity (OMAT) of the lunar soil was calculated from the equation of

OMAT= 0.08 ⁄ 1.19 defined by (23). The R950 is nearly

14 identical to R945, as mentioned above. The optical maturity of CD005–CD008 varies from 0.098 to 0.181 with exception of 0.312 for CD005 (Table S2). The unusually high OMAT of CD005 is likely due to blowing the top dusts off the surface by the rocket during the landing process, because CD005 locates most close to the lander (Fig. 2).

S5. Lunar Penetrating Radar Data Processing The Lunar Penetrating Radar (LPR) on the Yutu rover has two channels that work with different dominant frequencies (24, 25): Channel 1 with 60 MHz for detecting of the subsurface structure and Channel 2 with 500 MHz for detecting of the lunar regolith, respectively. Channel 1 antenna has two broadband monopoles on the two sides of the rover; Channel 2 antenna was on the bottom of the rover, which is ~30 cm from the ground (24). The depth resolutions of Channel 1 and 2 are <10 meters and <30 cm, respectively, based on the ground experiments (24). A total length of ~114 m of the LPR profile has been carried out (Fig. 2). Seismic exploration has developed many advanced techniques to detect terrestrial reservoirs buried under complex substructures, such as signal processing and seismic migration that are essential for extracting weak signals and high-accuracy imaging of substructures (26, 27). We applied these advanced methods to process the LPR data obtained by the Yutu. The basic differences between the seismic exploration and the LPR data are: (1) the seismic exploration works with the reflected elastic wavefields, whereas the LPR works with the reflected electromagnetic wavefields; (2) the seismic exploration adopts various offsets between the source and the receivers, whereas the LPR adopts only the zero-offset observation system. However, the wave propagation of either the elastic or the electromagnetic wavefields are basically the same; in addition, the observation system adopted by the LPR is a special case for that adopted by the seismic exploration. Therefore, there is no apparent barrier for extending the advanced methods from the seismic exploration to the LPR profile.

S5.1 Data processing of Channel 2B profile Channel 2 has two receiving antennas (2A and 2B) but traces recorded by Channel 2A are too noisy; thus, we do not use Channel 2A data in this paper. In contrast, Channel 2B has recorded 14634 high-quality traces in total. We show only one continuous segment, which contains 1500 traces, to illustrate the data processing procedure. As shown in Fig. S13a, the raw data have extremely unbalanced amplitudes between small, middle and big travel times, mainly due to spherical spreading, transmission loss, and dissipation loss; thus, we have to use an extreme clip to clearly present those small amplitudes. There are many direct currents in big travel times, as indicated by white arrows in Fig. S13a. We set the zero-wavenumber components of the 2D forward Fourier transform of raw data to be zeros (27), and those direct currents in big travel times have

15 been perfectly removed, comparing Figs. S13a and S13b. We applied spherical spreading amplitude compensation to balance the amplitudes for different travel times. After numerous trials, we found that an amplitude compensation factor by t ∗ 0.1 works well, as shown in Fig. S13b, which has both relatively balanced amplitudes and acceptable background noise. In addition, we applied a second-order Butterworth filter with a bandpass between 160 MHz and 1.28 GHz, and the low frequency direct currents, indicated by arrows in Fig. S13b, have been greatly reduced, as shown in Fig. S13c. As the Yutu rover was moving, the LPR obtained a continuous profile of reflected electromagnetic waves from the lunar regolith with heterogeneous relative dielectric constants; whereas, while the Yutu rover stopped, nearly identical reflected electromagnetic waves were recorded at the same point, which appears as absolutely flat segments in the profile. These segments are redundant and should be removed from the data set. The coordinates of traces (other than navigating points) are not available; thus, we could not identify redundant traces simply based on their coordinates. Instead, we picked up valid traces manually by leaving those absolutely flat segments. Unfortunately, the Yutu rover travels at a nonuniform speed, as shown in the left part of Fig. S13d, since the reflected waves seem to be highly flattened compared with the adjacent segments of the valid traces. We tried to keep only one trace for every two, three or four traces by checking whether the pattern of the reflected waves is consistent with its adjacent segments of the valid traces. The spatial interval between two traces is not constant; thus, we should estimate an average spatial interval for the conveniences of trace plotting and data processing. We took 0.05 m as the average spatial interval, after checking the ratio of the distance between two adjacent navigation points over the total number of the valid traces within the corresponding segment.

S5.2 Time-frequency analysis of LPR data We performed time-frequency analyses to obtain some basic characteristics of the lunar regolith. Figure S14 shows 16 traces selected from four typical segments with an interval of 5 traces. Obviously, the signals can be divided into three domains according to the spectrogram patterns. For the first domain (i.e. the early arrivals, within 0–40 ns), the dominant frequency for most selected traces is about 300–600 MHz; for the second domain (i.e. the middle arrivals, within 40–70 ns), the dominant frequency is significantly lower, about 200–300 MHz; and for the third domain (>70 ns), there is almost no visible energy. The rapid decrease of dominant frequency is probably due to the strong loss tangent of the lunar regolith. The loss is shown to be strongly dependent upon the (TiO2 + FeO) content (11), which is up to ~25 wt% at the landing site (Table 1). The sudden disappearance of the reflected waves is probably due to the strong wave scattering between steep boundaries of abundant large blocks and the nearly full transmit into the bedrock (i.e. almost without any reflection towards the lunar surface).

Fig. S13. Illustration of key steps of LPR data processing for Channel 2B. (a) raw data shown using 99.9% clip; (b) intermediate data after removing weak-amplitude direct currents for later arrivals and applying spherical spreading amplitude compensation; (c) data after bandpass filtering; (d) zoom in of the dashed rectangle shown in (c).

Fig. S14. Time-frequency analyses of four typical segments from the LPR profile. The short-time Fourier transform computations are based on the 128 samples long Hamming window with the overlap of 122 samples. Only 4 traces are shown for each segment with an interval of 5 traces. The black curves over panels are LPR traces extracted from the whole LPR profile. Dashed rectanglee ranges from 40 to 70 ns in time and from 200 to 300 MHz in frequency, respectively.

S5.3 Verifying the function of migration by synthetic data Migration is one of the key parts for imaging heterogeneous media with complex structures using the reflected seismic waves (26-28). Generally, it is great difficult to identify substructures directly from the seismic profile of a complex media, since each scattering point in the media would be mapped into a hyperbola in the seismic profile due to the propagation effects. Furthermore, an anticline model would be exxhibited as a much wider but gentle anticline in the seismic profile; in contrast, a syncline model would be exhibited as a much narrower but steeper syncline or even a bowknot in the seismic profile. Therefore, seismic migration has gained a lot of attentions in geophysical

18 research communities since 1970 (26, 28). The main function of the migration is to reduce the propagation effects from the seismic profile and make the migrated profile much closer to the true structures, which is essential for the successful survey of geological structures and for reducing the risk of oil drilling. For the traditional data processing of ground penetrating radar (29), one usually uses two axes along the left and right sides of the profile, with one axis (usually the left one) for the travel time and the other for the depth of the media. In fact, this is only valid for a homogeneous media or laterally homogeneous media with a linear vertical-velocity gradient. The real media may have far more complex structures and even sudden velocity contrasts, which would cause significant errors for both the shape and depth of structures. Therefore, the migration is crucial for the high resolution imaging of the lunar regolith using the LPR data. Figure S15a shows a theoretical lunar regolith model of relative dielectric constant, which contains several typical structures: syncline, anticline, dipping structures, circular and rectangle blocks, and the background has a linear increasing gradient with depth. Based on this model, we simulate a reflected LPR profile by numerically solving the electromagnetic waves equations using finite-difference modeling in time domain (30, 31). Both the dominant frequency of the source (we use a Ricker wavelet with 500 MHz dominant frequency) and the observation system are consistent with those used by the Yutu rover. Figure S15b shows the synthetic profile. Obviously, it is difficult to identify most structures directly from the synthetic profile, since we can only observe a lot of crossed hyperbolas due to propagation effects. We perform migration by the one-way wave equation method (32, 33) to reduce propagation effects and make those hyperbolas move back towards the true positions of the corresponding scattering points or reflecting interfaces. A model of background is required for the migration; whereas, the true model is hardly known for real cases. Instead, we can build up a reasonable background model (or macro model) based on some priori information. For example, although we used only the laterally homogeneous background of the model shown in Fig. S15a for the migration, the imaging results emerge many structures with high reliability compared with the true model, especially for those horizontal structures and reflectors with gentle dipping angles, as shown in Fig. S15c.

Fig. S15. Verifying the function of migration by synthetic data. (a) a theoretical lunar regolith model of relative dielectric constant, which contains several typical structures: syncline, anticline, dipping structures, circular and rectangle blocks, and the background has a linear increasing gradient with depth; (b) synthetic zero-offset profile using finite-difference modeling in time domain by a Ricker wavelet with 500 MHz dominant frequency; (c) migration result of the synthetic data using one-way wave equation method by the background model.

It should be pointed out that the migration fails to image those highly dipping structures and the nearly vertical reflectors; this is caused mainly by the limited aperture of the zero-offset observation system used by the LPR. It is well known that the computed tomography works well for various structures, which can stimulate signals and collect signals at all angles around the target. However, the electromagnetic waves (i.e. the source) can only be stimulated over the lunar surface; meanwhile, the electromagnetic reflected waves from the subsurface structures can only be received near the source. Therefore, the LPR has difficulties in receiving the reflected waves from those nearly vertical structures, and we would not image these structures since we have too weak or even no signal from them. Due to the same reason, the migration can only image the upper and bottom surfaces of blocks and could not image their vertical boundaries. This means it is difficult to pick up stones directly from the migration results, since we can seldom obtain a closed imaging boundary of a block. Nevertheless, the migration still presents us solid supports in interpreting the subsurface structures, since it can greatly reduce the artifacts by focusing the scattering waves and make the depth of the reflectors much closer to their true positions.

S5.4 Scanning the depth of the lunar soil sublayer by migration A thin sublayer can be recognized on the top of the lunar regolith profile, which has few reflected curving structures distinct from the main part of the regolith (Fig. S16). This top sublayer is likely fine-grained soil with low abundance of blocks and/or other heterogeneous clasts that appear as the curving reflectors. This is consistent with the presence of a homogeneous sublayer (~60 cm thick) in the deep drilled cores of Apollo regolith (11). We scanned the depth of the sublayer of the lunar regolith using migration. Given a reasonable background velocity model (11), we can focus the reflected waves close to the true positions of the corresponding reflectors by the migration method described above. To determine the depth of the lunar soil sublayer, we scan all possible depths using several well-established models of relative dielectric constant. When the trial depth consists well with the actual depth of the lunar soil sublayer in the migration result, we regard it as the best estimation. Partial LPR profile is shown in the right part of Fig. S13d, to illustrate the scanning procedure using the migration. The distance between the antenna and the lunar surface is 0.3 m as mentioned above. The lunar surface reflections exactly locate at the depth of 0.3 m in the migration results, as indicated by the yellow lines shown in Fig. S16. This demonstrates that the migration is correct for electromagnetic waves propagating in the vacuum. We scanned depth of the lunar soil sublayer from 0.1 to 1.9 m with an interval of 0.1 m, and the results between 0.3 m to 0.8 m are listed in Fig. S16. The relative dielectric constant is dominantly controlled by the bulk density and is nearly independent of frequency above 1 MHz and temperature within the range of the lunar surface (11). The relative dielectric constant for the lunar regolith is approximately

Fig. S16. Scanning depth of the lunar soil sublayer by migration. The yellow line indicates the lunar surface, and red dashed line indicates the scanning depth of the lunar soil sublayer. The panels with symbol “√”represents the best consistence betweeen the scanning depth and the migration result. Three lunar regolith models (i.e. the miniimum, middle and maximum cases) used by the migration are shown in Fig. S17a.

ε 1.9 , where ρ is the bulk density in g/cm, which varies with depth z (in centimeter) in a hyperbolic form (11) . ρ 1.92 . Olhoeft et al. (34) suggested a similar model but with some perturbations ε 1.93 0.17 . 3 Whereas, the smallest value of ε 1.93 0.17 1.76 is ~2.1 for ρ=1.3 g/cm at the lunar surface (11), which is apparently bigger than the smallest measured value 1.90– 1.94 (11, 35); thus, insteadd, we use ε 1.65 , which has the smallest value of ~1.92. Figure S17a shows these three models of relative dielectric constant: ε 1.65 , ε 1.9 and ε 2.1 , wwhich have a wide range and could cover most real cases.

We know that the velocity of the electromagnetic waves propagating in homogeneous media is , √ where is the speed of light in vacuum. The lunar-soil depth can be obtained by /2, if we know the travel time . We define the positive percentage of velocity variation as below:

⁄ 100% ⁄ 1 100% and the negative percentage of velocity variation as below:

⁄ 100% ⁄ 1 100%, where the superscript min, mid and max means the minimum, middle and maximum value of relative dielectric constant at a given depth, respectively. We took 1.9 to evaluate the maximum range of the lunar soil sublayer depth. Table S3 shows four typical cases of relative dielectric constant and their perturbations on the velocity. Obviously, the positive variation is less than 20% and the negative is less than -10%. The lunar soil sublayer depth using 1.9 is ~0.7 m, as shown in the middle row in Fig. S16. The maximum depth (using ε 1.65 ) is 0.70.7 16.28%=0.81 m; meanwhile, the minimum depth (using ε 2.1 ) is 0.70.7 8.41%=0.64 m. In other words, the possible depth of the lunar soil sublayer ranges from 0.64 to 0.82 m; thus, a reasonable estimation of the lunar soil sublayer depth is 0.73±0.09 m, for the selected part of the LPR profile. Figure 4 shows the whole profile after migration, where the depth of the lunar soil sublayer shows gentle variation and is generally ~0.7 m.

Table S3. Four typical cases of ε and their perturbations on the velocity. h (m) 0.002 1.8434 2.3054 2.6261 -06.30% 11.83% 0.5 2.2824 3.0862 3.6791 -08.41% 16.28% 1.0 2.3584 3.2274 3.8744 -08.73% 16.98% 3.0 2.4253 3.3530 4.0492 -09.00% 17.58%

S5.5 Estimating depth of the lunar regolith bottom by migration From the high resolution images of the landing camera (Fig. S2), the abundance of ejecta blocks on the surface at the landing site was determined 5.7%. We assume that the volume abundance of blocks in the top soil sublayer (0–0.7 m) of the lunar regolith is the same and constant, and then increases linearly up to 100% at 10 m deep. The relative dielectric constant of the basaltic block is 6.5 at 60 MHz (11), and the part of the lunar soil is still 1.9 . Thus, the bulk relative dielectric constant of the lunar regolith at a depth z can be expressed as below:

Fig. S17. Three lunar regolith models of relative dielectric constant. (a) ε 1.65 , ε 1.9 and ε 2.1 ; (b) modified lunar regolith model after taking basaltic block into account. The three models shown in (a) have a wide range and could cover most real cases of the lunar regolith.

1.9 1 6.5, where is the volume percentage of basaltic blocks

0 1 1000 1000 and is the minimum volume percentage of the blocks and is the depth of the lunar soil sublayer. Our model, with 5.7% and 70 , has a slightly higher relative dielectric constant than 1.9 , as shown in Fig. S17b. If the basaltic blocks, which have high relative dielectric constant, are not considered in the model for migration, we would obtain a larger depth of the lunar regolith, since a model with low relative dielectric constant would present the depth of the lunar regolith at deeper positions, as shown in Fig. S16. We perform migration with the modified model (Fig. S17b) for all valid traces between navigation points No. 105 annd 208 (Fig. 4). Figure S18 compares the original LPR profile and the migration result. Obviously, the migration results show much clearer and more continuous reflectors than the original LPR profile, as indicated by ellipses and arrows. This means that the migration is necessary for high resolution imaging of the complex lunar regolith structures. Dipping structures between two verrtical dashed lines show apparent movement after migration. This is reasonable and correcct according to the theoretical analyses by synthetic data shown in Fig. S15. In addition, these two dipping

24 structures are parallel and have reversal phases, which means they could be the upper and bottom boundaries of a block. We perform time-frequency analyses, as done above, on several key traces in the migration result, as shown in Fig. S19. We pick up the bottom of the lunar regolith manually for these key traces, where time-frequency analysis results at the corresponding depth positions show apparent changes of spectrum pattern. Then, we mark these depth positions as bllack circles on the colorful migration profile (Fig. 4). Obviously, a continuous bottom boundary picked up manually (indicated by the pinnk curve) is well controlled by those depth positions picked up independently on the key traces. The bottom of the lunar regolith is rough: most part of the profile has a depth of 5 m, with the minimum of ~2.2 m and the maximum of ~5.4 m. In addition, we can see that the bottom of the lunar soil (at ~0.7 m) varies gently, and the patterns over and blow the lunar soil sublayer bottom are quite different. It seems that there are two nearly horizontal structures along the whole profile within the lunar soil sublayer (0–0.7 m); however, they are residual errors of direct currents in fact rather than meaningful signals.

Fig. S18. Comparison between the original LPR profile (a) and the migration result (b). The migration result shows much clearer and more continuous reflectors than the LPR profile, as indicated by ellipses and arrows. The migration results within the dashed rectangle show apparent artifacts due to absurd spatial truncations.

Fig. S19. Time-frequency analyses on several key traces after migration. The short-time Fourier transform computations are based on the 128 samples long Hamming window with the overlap of 122 samples. The black curves over panels are from migration result shown in Fig. S18b. Pink arrow indicates the boundary picked up manually from waveforms and spectrums.

S5.6 Data processing of Channel 1 profile Channel 1 works with a dominant frequency of 60 MHz, which can penetrate deeper than Channel 2 (with a dominant frequency of 500 MHz). Figure S20a shows all traces, where we can’t identify any meaningful signal from the extremely noisy background. First, we tried to pick up valid traces at the same positions used for Channel 2B, but observed no significant reflectors even though various bandpass and median filters were applied. This is due to the extremely low signal to noise ratio of Channel 1. In order to

26 increease the signal to noise ratio and to identify the weak reflected waves, all traces of Channel 1 were used. In seismic exploration, we usually extract weak signals from noisy background by stacking the traces under the assumption that the signal is coherent but the noise is random (27). When the reflector is strictly or nearly horizontal, its reflected waves would show great consistency thus can be identified from the noisy background if a large enough number of records are obtained. This is the case for the profile of Channel 1, where the signals are consistent or coherent along horizontal directions for all records but the noise is generally random and inconsistent. After numerous trials, we found the best band-pass filteer between 4 and 30 MHz for each trace, and several reflected waves from the deep reflectors appear iindistinct at those positions indicated by horizontal arrows (Fig. 20b). The depth is converted from travel time using a relative dielectric constant 6.5, which is suggested by Heiken et al. (11) for lunar mare rock at 60 MHz. Note that the low-cutoff frequency of 4 MHz is farr lower than the working frequency of 60 MHz, which is usually out of the bandpass. Whereas, a too high low-cutoff frequency (say 15 MHz) would miss or ruin those weak signals.

Fig. S20. LPR data of Channel 1 without removing redundant traces. (a) The raw data profile, without any meaningful signals; (b) the profile after a bandpass filter between 4 and 30 MHz for each trace, showing three obvious reflectors (indicated by arrows).

Fig. S21. Comparison between the raw data (a) and the processed data (b) for the third layer in Channel 1.

However, the reflectors at lower depth (3000–4000 ns) are still ambiguous. We further applied a 2D median filter with a window width of 3 and 40 points along temporal and trace indices, respectively. The spatial continuity of reflectors, especially lower reflectors, has been greatly enhanced, as shown in Fig. 5. Figure S21b zooms in Fig. 5, clearly illustrating the necessity of careful and proper selection on the parameters for both the bandpass and 2D median filters; otherwise, such weak signals are easily neglected and would not be extracted out from the noisy background. Three reflectors have been clearly identified from the noisy background, which show gentle variation in horizontal dimension and are distinct from the straight horizontal lines of noise. The depths of these reflectors were determined via processing all traces together, without picking valid traces up and migration as done for Channel 2B, which are 195 m,

215 m, and 345 m, respectively. In order to compare with the results carried out by the Lunar Radar Sounder on the Kaguya spacecraft, which reported additional data set using

r=1, the appearance depths of these three reflectors for the same r=1 have also been calculated, and they are 500 m, 550 m, and 880 m, respectively.

S6. SI References 1. Morota T, et al. (2011) Timing and characteristics of the latest mare eruption on the Moon. Earth Planet Sci Lett 302(3–4):255-266. 2. Hiesinger H, Jaumann R, Neukum G, & Head III JW (2000) Ages of mare basalts on the lunar nearside. J Geophys Res 105:29239-29275. 3. Bugiolacchi R & Guest JE (2008) Compositional and temporal investigation of exposed lunar basalts in the Mare Imbrium region. Icarus 197(1):1-18. 4. Kramer GY, Jolliff BL, & Neal CR (2008) Searching for high alumina mare basalts using Clementine UVVIS and Lunar Prospector GRS data: Mare Fecunditatis and Mare Imbrium. Icarus 198(1):7-18. 5. Schaber GG (1973) Lava flows in Mare Imbrium: Geologic evaluation from Apollo orbital photography. Proc Lunar Planet Sci Conf 4:73. 6. Fu X-H, et al. (2014) Data processing for the Active Particle-induced X-ray Spectrometer and initial scientific results from Chang'e-3 mission. Research in Astronomy and Astrophysics 14(12):1595-1606. 7. Xie X, Yan M, Li L, & Shen H (1985) Usable values for Chinese standard reference samples of stream sediments, soils, and rocks: GSD 9-12, GSS 1-8 and GSR 1-6. Geostandards Newsletter 9(2):277-280. 8. Solé VA, Papillon E, Cotte M, Walter P, & Susini J (2007) A multiplatform code for the analysis of energy-dispersive X-ray fluorescence spectra. Spectrochim Acta Part B At Spectrosc 62(1):63-68. 9. Righter K (2013) The lunar meteorite compendium. http://curatorjscnasagov/antmet/lmc/indexcfm. 10. Yang W & Li SG (2008) Geochronology and geochemistry of the Mesozoic volcanic rocks in Western Liaoning: Implications for lithospheric thinning of the North China Craton. Lithos 102(1-2):88-117. 11. Heiken G, Vaniman D, & French BM (1991) Lunar sourcebook: A user's guide to the Moon (Cambridge University Press, Houston). 12. Kramer GY, Jolliff BL, & Neal CR (2006) Distinguishing high-alumina mare basalts using Clementine UVVIS and Lunar Prospector GRS data: Mare Moscoviense and Mare Nectaris. J Geophys Res 113(E1):E01002. 13. Giguere TA, Taylor GJ, Hawke BR, & Lucey PG (2000) The titanium contents of lunar mare basalts. Meteorit Planet Sci 35:193-200. 14. Lucey PG, Keil K, & Whitely R (1998) The influence of temperature on the spectra of the A-asteroids and implications for their silicate chemistry. J Geophys Res Planets 103(E3):5865-5871. 15. Lucey PG, Blewett DT, & Jolliff BL (2000) Lunar iron and titanium abundance algorithms based on final processing of Clementine ultraviolet-visible images. J Geophys Res 105(E8):20297-20305. 16. He Z-P, et al. (2014) Operating principles and detection characteristics of the Visible 29

and Near-Infrared Imaging Spectrometer in the Chang'e-3. Research in Astronomy and Astrophysics 14(12):1567-1577. 17. Liu B, et al. (2014) Data processing and preliminary results of the Chang'e-3 VIS/NIR Imaging Spectrometer in-situ analysis. Research in Astronomy and Astrophysics 14(12):1578-1594. 18. Hu S & Lin YT (2011) Modified calibration method of the chang’E-1 IIM images (in Chinese). SCIENTIA SINICA Phys, Mech & Astron 41(7):879-888. 19. Gueymard CA (2004) The sun’s total and spectral irradiance for solar energy applications and solar radiation models. Solar energy 76(4):423-453. 20. Pieters CM, et al. (2000) Space weathering on airless bodies: Resolving a mystery with lunar samples. Meteorit Planet Sci 35(5):1101-1107. 21. Taylor LA, Pieters CM, Keller LP, Morris RV, & McKay DS (2001) Lunar mare soils: Space weathering and the major effects of surface-correlated nanophase Fe. J Geophys Res Planets 106(E11):27985-27999. 22. Sunshine JM, Pieters CM, & Pratt S (1990) Deconvolution of mineral absorption bands: An improved approach. J Geophys Res 95(B5):6955-6966. 23. Lucey PG, Blewett DT, Taylor GJ, & Hawke BR (2000) Imaging of lunar surface maturity. J Geophys Res Planets 105(E8):20377-20386. 24. Fang G-Y, et al. (2014) Lunar penetrating radar onboard the Chang'e-3 mission. Research in Astronomy and Astrophysics 14(12):1607-1622. 25. Su Y, et al. (2014) Data processing and initial results of Chang'e-3 lunar penetrating radar. Research in Astronomy and Astrophysics 14(12):1623-1632. 26. Claerbout JF (1985) Imaging the earth's interior (Blackwell Scientific Publications, Inc, Palo Alto). 27. Yilmaz Ö (1987) Seismic data processing (Society of Exploration Geophysicists, Tulsa). 28. Etgen J, Gray SH, & Zhang Y (2009) An overview of depth imaging in exploration geophysics. Geophysics 74(6):WCA5-WCA17. 29. Daniels DJ (2004) Ground penetrating radar (Institution of Engineering and Technology, London) 2nd revised edition Ed. 30. Yee KS (1966) Numerical solution of initial boundary value problems involving Maxwell’s equations in isotropic media. IEEE Trans Antennas Propag 14(3):302-307. 31. Kunz KS & Luebbers RJ (1993) The finite difference time domain method for electromagnetics (CRC press, Boca Raton) first edition Ed. 32. Stoffa PL, Fokkema JT, de Luna Freire RM, & Kessinger WP (1990) Split-step Fourier migration. Geophysics 55(4):410-421. 33. Wu RS (1994) Wide-angle elastic wave one-way propagation in heterogeneous media and an elastic wave complex-screen method. J Geophys Res 99(B1):751-766. 34. Olhoeft GR & Strangway DW (1975) Dielectric properties of the first 100 meters of the Moon. Earth Planet Sci Lett 24(3):394-404. 35. Gold T, Bilson E, & Baron RL (1977) Electrical properties at 450 MHz of Apollo 15 and 16 deep drill core samples and surface soil samples at the same sites. Proc Lunar Planet Sci Conf 8:1271-1275.