Main

Water abundance in the lunar mantle provides insights into the giant impact formation models for the Moon6,7,8 and plays a crucial role in the crystallization of the lunar magma ocean10,11 and subsequent magmatism and long-lived volcanism12. In the past two decades, substantial advances in secondary ion mass spectrometry have improved our knowledge of water in the lunar mantle, causing a shift from the once traditional ‘bone dry’ concept13 to a relatively ‘wet’ Moon14. Extensive studies of various lunar samples derived from the melting of lunar mantle have shown a heterogeneous distribution of water in the lunar interior with abundances ranging from about 1 μg g−1 to 200 μg g−1 (refs. 9,15,16,17,18,19). However, most of these lunar water estimates derive from returned samples from the Procellarum KREEP Terrane on the nearside of the Moon (Fig. 1). The crust exposed at the surface of the Procellarum KREEP Terrane notably exhibits a higher thorium (Th) abundance than the other two main lunar geochemical provinces, the Feldspathic Highlands covering most of the Moon and the South Pole–Aitken (SPA) Basin, the oldest and largest impact basin of the Moon located on the lunar farside5.

Fig. 1: Return sample missions with lunar water estimates and the thorium distribution of the Moon.
figure 1

The Apollo (prefix A), Luna (prefix L), and Chang’e-5 (CE5) missions collected samples in and around the Procellarum KREEP Terrane which exhibits notably higher thorium (Th) than the Feldspathic Highlands and the SPA Basin. The Chang’e-6 (CE6) mission returned the first lunar samples from the lunar farside from the SPA Basin. The terrane boundaries are from ref. 5 and the Th base map is adapted from ref. 45. The sizes of the circles are scaled to the estimated maximum water abundance1,9,33 in the mantle source beneath the sampling sites.

Full size image

Both Th and water behave like incompatible elements in magmatic processes, preferring to stay in the melt rather than getting incorporated in the solidifying phase, potentially suggesting that the lunar mantle in the Th-poor Feldspathic Highlands and the SPA Basin may have lower water content. This possibility, if true, could critically inform the spatiotemporal distribution of water in the mantle of the Moon, as well as the giant impact formation models6,7,8 and subsequent magmatic evolution histories of the Moon12,20. However, little is known about the water abundance in the mantle beneath the Feldspathic Highlands and the SPA Basin. On 25 June 2024, Chang’e-6 (CE6), China’s lunar sample return mission landing within the SPA Basin, successfully returned the first lunar farside samples2,3,4, providing the samples required to investigate the water abundance of the farside lunar mantle.

Water in CE6 apatite and melt inclusions

A total of 578 particles varying from 0.1–1.5 mm in size were sieved and hand-picked from one CE6 soil sample (CE6C0200YJFM001 of 5 g). Mare basalt fragments account for around 28% of the selected particles characterized by scanning electron microscopy and electron probe microanalysis (EPMA; Extended Data Table 1 and Supplementary Table 1). The mare basalts, derived from the partial melting of lunar mantle rocks, are the main targets selected for this study. The dominant hydrous mineral in mare basalts—apatite and entrapped melt hosted by olivine and ilmenite—were carefully characterized using scanning electron microscopy (Extended Data Fig. 1 and Supplementary Figs. 1–3) for further in situ chemical and isotopic analysis of water.

The studied CE6 mare basalt fragments display ophitic, subophitic, porphyritic and poikilitic textures, mainly composed of pyroxene, plagioclase, ilmenite and silica with minor olivine, mesostasis and sulfide and trace apatite (Supplementary Fig. 4), similar to a survey conducted with a large number of lithic clasts (Zhang, Q. W. L. et al., manuscript in preparation; Zhou, Q. et al., manuscript in preparation). Parallel studies conducted on the CE6 chronology (Zhang, Q. W. L. et al., manuscript in preparation) and petrogenesis (Zhou, Q. et al., manuscript in preparation) have shown that these low-Ti mare basalts probably derive from the same lava flow, or at least the same sequence of lava flows. Therefore, we sum up all the studied clasts to calculate the modal abundance of apatite (<0.07 vol%; Supplementary Table 2). Most melt inclusions hosted by olivine and ilmenite display post-entrapment crystallization of 31–52% (Extended Data Table 2). The chemical compositions of pyroxene and plagioclase in CE6 mare basalt fragments are similar to those from the nearside, such as those of Chang’e-5 (CE5; Extended Data Figs. 2 and 3). The apatite grains are F-rich, with compositions in the range of known mare basalts (Extended Data Fig. 4). Two CE6 mare basalt fragments have MgO-rich olivine (Fo = 45–62 mol%) in which three melt inclusions were found (Extended Data Fig. 1, Supplementary Fig. 1 and Supplementary Table 3). Even more melt inclusions were found in ilmenite (Supplementary Fig. 2). Olivine-hosted melt inclusions have notably higher MgO (3.76–16.8 wt%) than ilmenite-hosted melt inclusions (<0.3 wt% MgO and 9.9–12.8 wt% Al2O3; Extended Data Fig. 5 and Supplementary Table 3). Tens of apatite grains with sizes of 5–30 μm and several melt inclusions hosted by olivine and ilmenite from 17 basalt fragments were also measured for their water abundance and hydrogen isotopic composition.

The CE6 basaltic apatite grains have water abundances ranging from 345 ± 4 μg g−1 (2σ) to 3,529 ± 30 μg g−1 (average 1,511 ± 748 μg g−1, 1 s.d.) with δD (δD = 1,000 × ([D/Hsample]/[D/Hstandard] − 1), using Vienna standard mean ocean water as the standard) values ranging from 657 ± 72‰ to 974 ± 109‰ (average 826 ± 94‰, 1 s.d.; Fig. 2 and Extended Data Table 3). The ilmenite-hosted melt inclusions contain water contents of 21 ± 3 μg g−1 to 238 ± 21 μg g−1 (2σ), with a wide range of δD values from −183 ± 261‰ to about 1,000‰ (Fig. 2 and Extended Data Table 2). By contrast, the olivine-hosted melt inclusions have water abundances ranging from 28 ± 8 μg g−1 to 46 ± 2 μg g−1 with δD values of −358 ± 610‰ to 100 ± 474‰ (2σ; Extended Data Table 2). All data were corrected for cosmic ray spallation using a cosmic ray exposure (CRE) age of 108 million years ago and those spots with H2O less than twice the instrument background were excluded (Methods).

Fig. 2: Water abundance and δD of melt inclusions and apatite from CE6 mare basalt fragments.
figure 2

Olivine-hosted melt inclusions have notably lower water abundances than ilmenite-hosted melt inclusions and apatite, as well as mostly lower δD values. Most melt inclusions have a lunar mantle-like δD value (yellow band)9,23,24,40. Three ilmenite-hosted melt inclusions have elevated δD values that are unresolvable from those of apatite within analytical uncertainties. Melt inclusions from mare basalts (MI in basalt) are from refs. 9,23,42, high- and low-Ti mare basalts are referred to the datasets compiled by refs. 1,9. The CE6 datasets have been corrected for a nominal CRE of 108 million years (Methods). The arrow represents the enrichment of water in the residual melt during crystallization of the parent melt. The dashed line is the degassing trend from a water abundance of 200 μg g−1 with a δD value of −200‰. Note the horizontal axis is a log scale. The error bars are 2σ. Ol-MI, olivine-hosted melt inclusion; Ilm-MI, ilmenite-hosted melt inclusion; LT, low-titanium; and HT, high-titanium.

Full size image

Water in parent melt and mantle source

Two methods were adopted to estimate water abundance in the parent melt of the CE6 basalts. The first method is based on apatite, which is the dominant water-bearing phase in lunar magmatic samples21. The bulk rock of CE6 basalts is estimated to have a water abundance of 1.1 ± 0.5 μg g−1 based on an average water abundance of 1,511 ± 748 μg g−1 (1 s.d.; Extended Data Table 3) of apatite and a modal abundance of ~0.07 vol% (Supplementary Table 2). By contrast, apatite grains in CE6 mare basalts show notably elevated δD values (657 ± 72‰ to 974 ± 109‰; Extended Data Table 3) compared with the low δD (average at −96 ± 230‰, 1 s.d.; Extended Data Table 3) of olivine-hosted melt inclusions and the recommended value of −200 ± 200‰ for the lunar mantle22,23,24. This difference indicates a notable loss of water by degassing in the form of H2 before the crystallization of apatite under a reduced oxygen fugacity environment24,25. This observation is consistent with the much lower hydrogen isotope compositions in mineral-hosted melt inclusions (Fig. 2). This feature is identical to that of high-Ti and low-Ti mare basalts collected from the nearside of the Moon (Fig. 2), suggesting that it resulted from a common degassing process incurred during mare volcanism. Thus, a degassing extent of 97.99–99.35% based on Rayleigh fractionation (Methods) yields a water abundance of 55 ± 25 μg g−1 to 168 ± 77 μg g−1 (1 s.d.) for the parent melt of CE6 basalts, assuming an initial δD value of about −200‰ based on the analyses of D-poor and H2O-rich CE6 melt inclusions hosted by olivine and ilmenite (Methods). Meanwhile, based on apatite–melt hygrometry1,26,27, we estimated the parent magma to have a water abundance of 76 ± 41 μg g−1 (Methods and Supplementary Table 6). Thus, both calculations on apatite yield comparable results within uncertainties.

The second method for estimating water in the parent melt uses the melt entrapped in minerals of CE6 basalts. Two types of mineral inclusions, hosted by olivine and ilmenite, were measured in this study (Fig. 2 and Extended Data Table 2). The analytical results for the melt inclusions depict an evolutionary pathway of the parent melt of the CE6 basalts, in which the entrapment of melts in olivine occurred in a relatively earlier stage of magmatic processes than that of ilmenite, consistent with the crystallization sequence of other mare basalts28 (Fig. 3). Olivine is the first-crystallized phase in the CE6 basalt and thus its melt inclusions may preserve the initial hydrogen isotope composition of the CE6 parental melt. Therefore, olivine-hosted melt inclusions are used to estimate the water abundance for the parent melt, yielding a range of 15–22 μg g−1 after correction for post-entrapment crystallization (Extended Data Table 2). These two methods yield a water abundance of about 15–168 μg g−1 for the CE6 basalt parent magma. This estimate is comparable to some values from some nearside samples (Extended Data Fig. 6).

Fig. 3: Evolutionary pathways of magmatic water in CE6 mare basalt.
figure 3

a, CE6 mare basalt was derived from the partial melting of lunar mantle rocks. The melt aggregates at depth in networks to form the source magma and intrudes upwards into shallow depths. The ambient rocks are much colder than the hot melt, resulting in fractional crystallization of the source magma. In certain circumstances, the residual melt (parent magma of CE6) from the source magma flowed up to the lunar surface, forming the rocks of CE6 mare basalt. The water abundances and hydrogen isotopic compositions measured on melt inclusions and apatite from the CE6 mare basalt depict an evolutionary pathway for water in the parent melt. b, Rare MgO-rich olivine (Fo 45–62 atom%; Supplementary Table 3) in CE6 mare basalt should crystallize much earlier than the main constituent phases (pyroxene and plagioclase) in the crystallization processes of the parent melt, characterized by a low water abundance and a lunar-mantle-like δD value (Fig. 2) in the melt trapped in the olivine grains, as well as a notably high MgO content of the melt (Extended Data Table 2 and Extended Data Fig. 5, Supplementary Table 3). c, Onset of ilmenite formation in low-Ti mare basalts occurred at a late crystallization stage of the parent melt28 accompanying enrichment of water in the residual melt followed by degassing of water when the ambient pressure is inadequate to trap the water dissolving in the melt. Thus, ilmenite-hosted melt inclusions exhibit variability in both water concentration and δD (Fig. 2). d, Apatite is one of the last-crystallized minerals in CE6 basalts, recording notable higher δD values than olivine-hosted melt inclusions. Some residual melt from which apatite crystallized may have been trapped by late-crystallized ilmenite (Fig. 2). Ol, olivine; Pl, plagioclase; Px, pyroxene; Ilm, ilmenite; and Ap, apatite.

Full size image

The parent melt of CE6 mare basalts originated from the partial melting of a depleted mantle source, subsequently experienced fractional crystallization according to the study of its petrogenesis (Zhou, Q. et al., manuscript in preparation) and potential degassing during magmatic processes (Fig. 3). Early-formed melt inclusions hosted by olivine could be more reliable than the late-formed minerals and melt (Fig. 3). Thus, olivine-hosted melt inclusions were used to calculate the water abundance in the lunar farside mantle. Both Ce and H2O exhibit similar incompatibilities, whereas Ce is not a volatile species29, thus the resulting H2O/Ce ratio of the parent melt can serve as a useful index to estimate water abundance for the mantle source1,30,31,32. We calculated a H2O/Ce ratio of 0.58–0.85 for the CE6 olivine-hosted melt inclusions using a bulk Ce abundance of 25.9 μg g−1 measured in Zhou, Q. et al. (manuscript in preparation). The mantle source of the CE6 mare basalt was estimated to have a water abundance of approximately 1–1.5 μg g−1 (Fig. 4 and Supplementary Table 7), considering a Ce content of 1.728 μg g−1 for the primitive lunar mantle31. This estimate may represent an overestimate if considering the ultra-depleted nature for the mantle source of the CE6 mare basalt (Zhou, Q. et al., manuscript in preparation). The CE6 estimate represents the first result for the water abundance of the mantle from the farside of the Moon.

Fig. 4: Spatiotemporal variation of lunar mantle water abundance estimates.
figure 4

The water abundance in the mantle source of CE6 mare basalt is estimated to be 1–1.5 µg g−1, plotting within the lowest end of the ranges from nearside samples. It is thus apparent that the lunar mantle may exhibit a hemispheric dichotomy in terms of water contents. The mantle water abundances estimated from Apollo and CE5 lunar samples with definite (non-ejecta) geographic locations are plotted for comparison9,33. All data are plotted as average values with the error bars representing the ranges ((max − min)/2) of the estimates (Supplementary Table 7). The approximately 2.8 Ga age of the CE6 mare basalt is provided in Zhang, Q. W. L. et al. (manuscript in preparation). Note the vertical axis is the log scale. Red (1.25 µg g−1), light blue (7.5 µg g−1) and orange (70.3 µg g−1) lines are the median H2O abundances for the mantle sources estimated on the melt inclusions from the farside CE6 mare basalt, all nearside mare basalts and the nearside pyroclastic glasses.

Full size image

Distribution of lunar mantle water

A water abundance of 1–1.5 μg g−1 estimated for the mantle source of CE6 basalts plots among the lowest end of the ranges (about 1–200 μg g−1) reported from the nearside samples (Fig. 4), suggesting that the mantle source of CE6 mare basalt is drier than that of most nearside samples. A potential drier mantle in the farside of the Moon carries important implications for understanding the distribution of water in the interior of the Moon. This finding favours a heterogeneous distribution of water in the lunar mantle, consistent with the dichotomy distribution of Th concentration on the surface of the Moon5. If the whole farside mantle is similar to the mantle source of CE6 mare basalts and drier than the nearside counterpart, then the water abundance for the bulk silicate Moon may have been overestimated to some extent1,33. Therefore, a lower water content in the bulk silicate Moon is expected, yielding a result more in line with the giant impact formation model of the Moon6,7,8.

The dating of a large number of CE6 basalt clasts indicates that 99% of them belong to the local basalt erupted at about 2.8 × 109 years ago (Zhang, Q. W. L. et al., manuscript in preparation). The spatiotemporal evolution of water in the lunar mantle indicates that the water abundances may exhibit a spatial hemispheric contrast rather than a temporal trend based on the formation age of the samples (Fig. 4). A scenario to explain this lower water content is that the mantle source of CE6 basalts was affected to some extent by the SPA giant impact event, which may have driven the flux of water and other incompatible elements (Th, K and H2O) enriched in the mantle of the Procellarum KREEP Terrane towards the lunar nearside and became simultaneously depleted in its hemispheric antipode in the SPA Basin34,35. Another potential interpretation is a vertical distribution of water in the lunar mantle36,37. The CE6 basalt originated from an extremely depleted mantle source, which may have a deeper depth of melting than most other lunar samples (Zhou, Q. et al., manuscript in preparation). Lunar mantle could have relatively less water for earlier and deeper crystallized cumulates than later and shallower cumulates formed from the lunar magma ocean because of the gradual enrichment of incompatible elements in the residual melt38,39.

Origin of lunar mantle water

The analyses of olivine- and ilmenite-hosted melt inclusions consistently point to a parent melt of the CE6 basalts that had a δD value of −123 ± 167‰ (1 s.d.; Methods). This value is indistinguishable from that (−200 ± 200‰) derived from nearside samples9,23,24,40 within analytical uncertainties, suggesting that the mantle of the Moon could have a relatively homogeneous hydrogen isotopic composition, which does not show a spatiotemporal variation. This uniform hydrogen isotopic composition for the mantle source from various types and ages of lunar samples in this study and previous works9,23,24,40 may have been inherited from the lunar magma ocean from which the mantle rocks crystallized41. Furthermore, the hydrogen isotopic composition of the farside mantle of the Moon favours a chondritic origin for the indigenous water of the Moon40,42.

The first water estimate retrieved from the lunar farside indicates that water abundance in the lunar mantle may exhibit a hemispheric dichotomy (Fig. 4), similar to the well-known overlying crustal asymmetry of the nearside and the farside of the Moon5. Given the limited sampling of the lunar farside, it is still uncertain whether the potential drier mantle documented here is widespread beneath the SPA Basin or the whole farside hemisphere. New sample return missions scheduled by Artemis43 and post-Chang’e programs44 aim to collect more samples from the farside of the Moon, thus potentially testing and further unravelling the extremely heterogeneous distribution of the abundance of water on the Moon.

Methods

Sample preparation

Five grams of scooped Chang’e-6 (CE6) lunar soil (CE6C0200YJFM001) allocated by the China National Space Administration (CNSA) were used in this study. The allocated samples were sieved into two dividers in an ultraclean room at the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS) in Beijing, China. One separate was >355 μm and the other was <355 μm. Eighty-eight particles >355 μm were selected and prepared into 7 Sn–Bi polished mounts following the protocol in ref. 46. The mounts were labelled as CE6C02,MGP01–MGP07 (Extended Data Table 1 and Supplementary Table 1). Another 490 particles with grain sizes ranging from 100 μm to 350 μm were hand-picked from the sieved soils. These particles were placed in a tunnel (depth about 300 μm, width about 400 μm and length about 1.8 mm) of silica glass cut by low-speed diamond saw (Supplementary Fig. 5). Then the tunnel was filled up with epoxy. These particles were finally prepared into 14 polished thick sections labelled as CE6C02,WGP01–WGP14 (Extended Data Table 1 and Supplementary Table 1). The prepared sections were cleaned up using anhydrous ethanol before drying at 50 °C in a baking oven. All the mounts were coated with carbon for petrographic observation, chemical analysis and isotopic analysis. Statistically, the particles used in this study consist of 163 mare basalt fragments (28%), 97 monomineralic fragments (17%), 50 agglutinates (9%), 101 breccias (18%), 52 highlands fragments (9%), 76 impact melt fragments (13%) and 39 glass beads and fragments (7%) (Extended Data Table 1). Representative petrography of these components in the CE6 lunar soils is shown in Supplementary Fig. 6.

Scanning electron microscopy

Petrographic observations and elemental mapping were carried out using field emission scanning electron microscopes using the FEI Nova NanoSEM 450 and the Thermo Fisher Apreo instruments at IGGCAS, using electron beam currents of 2–3.2 nA and an acceleration voltage of 15 kV. Energy dispersive spectroscopy (EDS) X-ray maps were collected for each basaltic clast to locate P-bearing phases, identical to the protocol used in the study of CE5 lunar soils9,47,48. The phosphates were then observed at higher magnification in back-scattered electron (BSE) images. The modal abundance of apatite from various CE6 basalt fragments was counted by the exposed surface areas (Supplementary Table 2). The apatite in the studied samples has a modal abundance of 0.067 vol% calculated from 163 basalt fragments using Image J software (Supplementary Table 2 and Supplementary Fig. 7).

Electron probe microanalysis

The major and minor elemental abundances in phosphates, glassy melt inclusions and associated minerals (olivine, pyroxene, plagioclase and ilmenite) were measured using a JEOL JXA-8100 electron probe microanalyser at IGGCAS after the NanoSIMS measurements to avoid possible H loss due to bombardment by the electron beam49. For silicate, ilmenite and apatite analysis, the operating accelerating voltage was 15 kV and the beam current was 20 nA. The standards used were albite for Na and Al, diopside for Si, Mg and Ca, magnetite for Fe, bustamite for Mn, sanidine for K, rutile for Ti, Cr2O3 for Cr, NiO for Ni, apatite for P, tugtupite for Cl and synthetic fluorite for F. The F Kα X-rays were counted with a LDE1 crystal to avoid drift with time (ref. 50). F and Cl were measured first to minimize the loss of volatiles by electron beam irradiation. The counting time for F was 20 s. The detection limits (1σ) were 0.01 wt% for K2O, P, F, and Cl; 0.02 wt% for Cr2O3, Al2O3, MnO, CaO, FeO, TiO2 and NiO; 0.03 wt% for SiO2 and Na2O. For analysis of melt inclusion glass, the accelerating voltage was 15 kV and the beam current was 10 nA, with a beam size of 5 μm in diameter to obtain the composition of the glass within a melt inclusion. A natural obsidian glass was used as a monitor. The detection limits for MI were 0.02 wt% for K2O, CaO, FeO and NiO; 0.03 wt% for SiO2, MgO, Cr2O3, Al2O3, MnO and TiO2; and 0.04 wt% for Na2O. The EPMA data obtained for apatite, melt inclusions and the coexisting silicates are listed in Supplementary Table 3.

In situ water abundance and hydrogen isotope analysis

A CAMECA NanoSIMS 50L at IGGCAS was used to measure the hydrogen isotopes and water abundance of apatite and melt inclusions from the CE6 basaltic fragments. The samples were loaded in sample holders together with the standards, and were baked overnight at about 60 °C in the airlock chamber. The holders were then stored in the NanoSIMS vessel chamber to improve the vacuum quality and minimize the H background51,52. The vacuum pressure in the analysis chamber was 2.1 × 10−10 torr to 3.0 × 10−10 torr during analysis facilitated by a liquid nitrogen cold trap53. Each 15 μm × 15 μm analysis area was pre-sputtered for 2 min with a Cs+ ion beam current of 5 nA to remove the surface coating and potential contamination. A 4 μm × 4 μm region of interest on targets was selected for analysis. The secondary anions 1H−, 2D−, 12C− and 18O− were simultaneously counted by electron multipliers from the central 3 μm × 3 μm areas using the electronic gate technique of NanoSIMS (about 50% blanking). The ion beam current (Fco) was around 0.5 nA for analysis with a beam size of about 500 nm in diameter. The charging effect on the surface of the samples was compensated by an electron gun during analysis. More details are provided in refs. 52,54.

A series of apatite and glass standards were used for calibrating the water abundances and hydrogen isotopes of the samples: Durango apatite (H2O = 0.0478 wt% and δD = −120 ± 5‰) (refs. 16,55) and Kovdor apatite (H2O = 0.98 ± 0.07 wt% and δD = −66 ± 21‰) (ref. 56), the SWIFT MORB glass (H2O = 0.258 wt% and δD = −73 ± 2‰), and two basaltic glasses, ALV-519-4-1 (H2O = 0.17 wt%) (ref. 51) and ALV-1833-11 (H2O = 1.2 wt%) (ref. 51) (Supplementary Table 4). The water content calibration lines established on apatite and glass standards were used for calibrating the samples (Extended Data Fig. 7). Instrument mass fractionation (IMF) on hydrogen isotopic composition are the same between apatite and silicate glass within analytical uncertainties (Extended Data Fig. 7). Instrumental mass fractionation on H isotopic compositions of both apatite and melt inclusions were conducted using the Kovdor apatite standard and monitored by analysing both the Durango apatite and SWIFT MORB glass standards during the whole analytical session (Extended Data Fig. 7). Hydrogen isotopic compositions are given using the delta notation, δD = ((D/H)sample/(D/H)SMOW) − 1) × 1,000‰, where SMOW is the standard mean ocean water with a D/H ratio of 1.5576 × 10−4 (ref. 57).

The instrument H background was monitored by standards of San Carlos olivine58 (H2O = 1.4 μg g−1) and a synthetic anhydrous quartz glass Suprasil 3001 (personal communication with Erik Hauri). The H background correction of ref. 59 was used, which follows the relationship: H/Obg = (Hcounts − Hbg)/Ocounts and D/Hmeasured = (1 − f) × D/Htrue + f × D/Hbg, where f is the proportion of H emitted from the instrumental background. Owing to the inadequate baking and storing time in the instrument, the first 2 days have a slightly higher instrument background at 26–53 μg g−1 (D/Hbg was (1.40 ± 0.25) × 10−4 to (1.13 ± 0.41) × 10−4 and Hbg was 1546 ± 266 counts per second (cps) to 786 ± 81 cps (Supplementary Table 4). The instrument H2O background was about 10 μg g−1 (D/Hbg was (1.19 ± 0.26) × 10−4 to (1.14 ± 0.49) × 10−4 and Hbg was 374 ± 170 cps to 311 ± 37 cps) in the following days (Supplementary Table 4). After background subtraction, the water abundances of apatite and melt inclusions were quantified by the slope of the calibration line (Extended Data Fig. 7), which was determined by measuring apatite and glass standards.

All data are reported with their 2σ uncertainties that include reproducibility of D/H measurements on the reference materials, uncertainty of H2O background subtraction, internal precision on each analysis and uncertainty of corrections of spallation effects (Supplementary Table 5). The raw measured D/H ratios were corrected for the background, followed by IMF and spallation in sequence (Supplementary Table 5).

Correction of water abundances and D/H ratios for spallation effects

The measured D/H ratios have been corrected for the potential effects of spallation by CRE, using a D production rate of 2.17 × 10−12 mol D/g/Ma (ref. 60) for melt inclusions and 9.20 × 10−13 mol D/g/Ma (ref. 61) for apatite, where Ma signifies millions of years ago. The correction errors induced by D spallation are around 50% for δD and negligible for water content42. The CRE age for the CE6 lunar soils derives from a cosmogenic 21Ne age of 146 ± 27 Ma (T21; Zhang, X. et al., manuscript in preparation), which was converted to an exposure age (TCRE) of 108 ± 20 Ma using an empirical relationship (T21 = 1.35 × TCRE) between T21 and TCRE based on the study of Apollo samples60. An uncertainty of 50% (2σ) in D and H production rates raised by the spallation process was combined to calculate its contribution to the δD errors (ref. 42; Supplementary Table 5).

Petrography of CE6 mare basalt

Approximately 28% of the lithic clasts in a total of 578 fragments studied in this work are mare basalt fragments (Supplementary Table 1), which display subophitic, poikilitic and porphyritic textures (Supplementary Fig. 4), similar to those in Zhang, Q. W. L. et al. (manuscript in preparation) and Zhou, Q. et al. (manuscript in preparation). Lath-shaped plagioclase in the basalts exhibits sharp boundaries, whereas pyroxene is anhedral with compositional zonation in BSE images (Supplementary Fig. 4). These basalt fragments are mainly composed of pyroxene, plagioclase and ilmenite with minor Si–K-rich mesostasis, fayalite and troilite, and trace apatite, merrillite, tranquillityite and baddeleyite (Supplementary Fig. 4). Olivine is very rare and absent in most CE6 mare basalt except two basalt fragments (CE6C02,WGP09,G17 and CE6C02,WGP14,G3), as shown in this study and other parallel studies (Zhang, Q. W. L. et al., manuscript in preparation; Zhou, Q. et al., manuscript in preparation). These two olivine grains have several melt inclusions with diameters of about 5–50 μm (Extended Data Fig. 1 and Supplementary Fig. 1). Melt inclusions have experienced post-entrapment crystallization (31–52%) (Extended Data Fig. 1, Extended Data Table 2 and Supplementary Fig. 1). Ilmenite intergrows with plagioclase but has a smaller size and irregular edge, occurring as laths partially enclosed by pyroxene (Supplementary Fig. 1). Nine melt inclusions in ilmenite were identified in 7 CE6 mare basalt fragments with diameters mostly less than 30 μm (Supplementary Fig. 2). Ilmenite-hosted melt inclusions have experienced post-entrapment crystallization (0–31%) (Extended Data Table 2 and Supplementary Fig. 2). Phosphorus- and zirconium-bearing minerals mainly occur in the fine-grained interstitial areas. Several apatite grains are observable in the margins of plagioclase and pyroxene (Supplementary Fig. 3). Most apatite grains have euhedral shape and rare ones are lath-shaped (Supplementary Fig. 3). CE6 mare basalts have notably less apatite (0.067 vol%; Supplementary Table 2) than CE5 basalts (0.4 vol%) (ref. 9). Meanwhile, apatite grains in CE6 mare basalt fragments have smaller sizes (mostly <5 μm; Supplementary Fig. 3).

Mineral chemistry of CE6 mare basalt

Plagioclase from CE6 basalt clasts is relatively homogeneous with a composition of An75.2–92.5Ab5.3–20.1Or0.1–5.0 (Extended Data Fig. 3 and Supplementary Table 3). The pyroxene displays compositional zonation from Mg-rich cores (En42.1–56.1Wo9.1–38.1Fs20.7–49.5) to Fe-rich rims (En0.3–25.9Wo11.6–47.3Fs47.8–87.6) (Extended Data Fig. 2 and Supplementary Table 3). Ilmenite shows no chemical difference between various basalt fragments, containing about 52.9 wt% TiO2 and about 46.2 wt% FeO, with minor other elements (Supplementary Table 3). Most CE6 basalt fragments contain trace fayalite (Mg# < 1). Two relatively large olivine grains having melt inclusions identified in this study are quite homogeneous in chemistry with average Mg# of 49.3 ± 2.7 and 59.1 ± 2.2 (1σ, n = 5; Extended Data Fig. 3 and Supplementary Table 3). Interstitial mesostasis is SiO2-rich (65.2–76.9 wt%) and K2O-rich (2.40–7.65 wt%; Supplementary Table 3). Olivine- and ilmenite-hosted melt inclusions have 38.0–79.9 wt% SiO2, 0.61–7.13 wt% TiO2, 5.43–17.5 wt% Al2O3, 0.68–15.6 wt% CaO and 1.44–33.1 wt% FeO (Extended Data Fig. 5 and Extended Data Table 3). Concentrations of MgO in olivine-hosted melt inclusions range from 3.76 wt% to 16.8 wt%, whereas those of melt inclusions in ilmenite are less than 0.30 wt%. Melt inclusions in olivine have lower SiO2 contents (38.0–60.6 wt%) compared with ilmenite-hosted ones (44.5–79.9 wt%), indicating an earlier entrapment for the former. Apatite grains in this study contain 2.54–4.08 wt% F and 0.09–0.43 wt% Cl (Supplementary Table 3), close to the fluorapatite end-member in the F–Cl–OH ternary diagram (Extended Data Fig. 4). The evaluated OH contents of apatite range from 0 wt% to 0.22 wt%, assuming the volatile site only contains F, Cl and OH. Apatite grains in CE6 mare basalt fragments are usually less than 5 μm, rendering it unavoidable to sample some surrounding minerals during EPMA and yielding an EPMA total usually less than 98 wt% (Supplementary Table 3).

Estimate of water abundance for the parent melt of CE6 mare basalt

Melt inclusions

Melt inclusions are trapped melt in the minerals during crystallization of the parent melt. Three melt inclusions were identified in olivine from two CE6 mare basalt fragments (CE6C02,WGP09,G17 and CE6C02,WGP14,G3; Extended Data Table 2 and Supplementary Fig. 1). These olivine-hosted melt inclusions have a water abundance of 28–46 μg g−1, slightly higher than the instrument background, with large variation in δD values (−358 ± 610‰ to 100 ± 474‰) after correction of spallation effects (Extended Data Table 2). The correction of spallation can markedly affect δD for those samples with water abundances <50 μg g−1 (ref. 42). The hydrogen isotopic compositions for analyses with H2O < 20 μg g−1 may not be reliable because of the extremely low measured D counts1. Thus, those spots with H2O < 20 μg g−1 (that is, two times the instrument background) are not discussed or used in deriving the water abundance for the parent magma and for the mantle source (Extended Data Table 2). An average δD of −96 ± 230‰ (1 s.d.) in olivine-hosted MIs is comparable to some early trapped melt hosted by ilmenite, which have water abundances of 153–184 μg g−1 and δD values of −183‰ to −140‰ with an average value of −158 ± 23‰ (1 s.d.; Fig. 2 and Extended Data Table 2). The low-δD nature of the melt inclusions hosted by olivine and by early-formed ilmenite are comparable to the mantle δD value (−200 ± 200‰) recommended in previous studies9,23,40,62, indicating that these MIs plausibly represent the parent melt in the terms of water abundance and hydrogen isotope signature. This prediction is also supported by the high MgO content in olivine-hosted melt inclusions compared with previous work23,30,32.

The host olivine grains of melt inclusions have forsterite content of 45.0–61.6 atom% (Supplementary Table 3), indicative of early precipitation phases crystallized from the parent melt of the CE6 mare basalt. In comparison, CE6 mare basalt has a slightly lower bulk Ti abundance (4.2 wt%) (Zhou, Q. et al., manuscript in preparation) than that of CE5 (5.3 wt%) (ref. 63), which would result in the late precipitation of ilmenite from the parent melt as shown by modelling28. A gap in water abundance between olivine- and ilmenite-hosted melt inclusions could be the result of the crystallization sequence (Fig. 2). It is also noted that some ilmenite-hosted melt inclusions have comparable values to apatite, suggesting a late entrapment probably accompanying the onset of apatite at a quite late crystallization stage. Therefore, we provide an illustration diagram depicting the potential evolutionary pathways of magmatic water for CE6 mare basalt (Fig. 3). Finally, the δD-poor spots of olivine-hosted melt inclusions are regarded as a target for estimating the water abundance of the parent melt. We estimate a water abundance of 15–22 μg g−1 for the parent melt of the CE6 mare basalt based on the olivine-hosted melt inclusions (Extended Data Table 2). Excluding the notable high δD ilmenite-hosted melt inclusions probably formed accompanying apatite in the late crystallization stages, we estimate the parent melt of the CE6 mare basalt as having a δD value of −123 ± 167‰ (1 s.d.) (Extended Data Table 2).

Apatite

Apatite is the major OH-bearing phase in lunar igneous rocks1. A modal abundance and an average water abundance of apatite were used to quantify water in bulk CE6 basalts. Statistically assessed using the surface areas of apatite grains in all basaltic clast, the modal abundance of apatite in CE6 basalts is found to be less than 0.07 vol% (Supplementary Table 2). The average water content and δD value of the CE6 apatite measured by NanoSIMS 50L are 1,511 ± 748 μg g−1 and 826 ± 94‰ (1 s.d., n = 16), respectively (Extended Data Table 3). Hence, the water abundance of the bulk CE6 basalt is 1.1 ± 0.5 μg g−1 (1 s.d.). Apatite is a late-crystallizing mineral in mare basalts; its notably higher δD values (826 ± 94‰) than the low-δD (−200 ± 200‰) nature of lunar mantle9,23,40,62 and low δD of CE6 melt inclusions indicate strong degassing and loss of water in the form of H2 (refs. 24,25). The hydrogen isotope fractionation during volatile loss into a vacuum is given by α2 = M1/M2, where M1 and M2 are the masses of the volatile phase isotopologues. The change of the isotopic composition of H during volatile loss by Rayleigh fractionation is given by R = R0 × f(α−1), where R0 and R are the initial and final D/H ratios for a fraction f of remaining hydrogen. Degassing of H2 (M1 = 2 for H2 and M2 = 3 for HD) yields an α value of about 0.8165 (ref. 25). Degassing modelling indicates that the crystallization of apatite grains in CE6 mare basalt started at the time when 97.99–99.35% of initial water in the parent melt has been degassed. After degassing loss correction, the water abundance of the parent melt is about 55 ± 25 μg g−1 to 168 ± 77 μg g−1 (1 s.d.).

In another viable approach1,26,27, the authors have developed apatite-based melt hygrometry methodologies, to estimate parent magma water abundance using apatite. These methods are designed to estimate parent melt water abundances from apatite regardless of how much H-degassing has occurred after apatite crystallization. The measured F and H2O abundance in apatite from each fragment was used to derive the parental melt H2O, using a 0.07 vol% modal abundance of apatite (Supplementary Table 2). The results yield H2O abundance of 59–99 μg g−1 for the parental magma (Supplementary Table 6), plotting within the approximately 55–168 μg g−1 range of the other method presented earlier. Considering the early entrapment of olivine-hosted melt inclusions17, these targets were used to estimate the water abundance for the mantle source.