Temporal changes in Angkorian ironmaking technology in Northern Cambodia during the 10th–14th centuries Authors: Keishiro Azami, Etsuo Uchida, Aoi Mizumori, Yoshimitsu Kobayashi, Kaito Kuriyama, Tamaki Nakamura, Yasuhiro Kato Published: 2025 ABSTRACT The Khmer people constructed numerous monuments from the 9th to 15th centuries (Angkor period) using iron tools. Ancient ironmaking sites have been identified as slag mounds near Khmer monuments. However, details of the ironmaking technology used during the Angkor period remain unknown. Herein, we show the temporal changes in ironmaking technology in northern Cambodia, which was a major iron-producing region of the Angkorian Empire. The chemical and Nd-Sr-Pb isotopic compositions of slag and ore samples indicate that Angkorian iron producers used diverse ores and employed at least two distinct ore formulations. From the late 10th to early 13th centuries, laterite and umber were used as iron ores with the addition of Mn oxide and sulphide ores, resulting in high iron yield. This technique may have evolved in response to increased iron demand during the peak of the Angkorian Empire. However, after the late 13th century, ironmaking became less efficient, relying solely on laterite and magnetite ores without additives, which coincided with the decline of the Angkorian Empire. The efficient ironmaking method was lost with only a part of the technique possibly being transferred to another area. These results provide new insight into how technological changes in iron production were linked to the socio-economic dynamics and eventual decline of the Angkorian Empire.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-27581-x. Introduction The Angkorian Empire expanded across much of mainland Southeast Asia between the 10th and 13th centuries, integrating upland and lowland regions through a state road network, temple foundations, and hydraulic works1,2. Today, Khmer monuments constructed in Angkor period (9th–15th centuries) are primarily located in Cambodia, with some monuments also being found in Thailand, Laos, and Vietnam. Architectural iron (e.g., clamps/crampons and internal armatures) is documented at these Khmer monuments, suggesting that iron production was an essential industry supporting temple construction, infrastructure, and tool manufacture throughout the Angkorian Empire3–5. The Preah Khan of Kompong Svay (hereinafter referred to as Preah Khan) was a political and religious centre on the frontier of the Angkorian Empire; it served both as an eastern outpost against the Cham incursions2,6, and as an entrepôt for iron produced in the eastern hinterlands1,2. Preah Khan, located approximately 100 km east of Angkor—the capital of the Angkorian Empire near present-day Siem Reap— occupied a strategic position along the state road system linking the capital with ore-rich regions in the eastern provinces1. Phnom Dek (“Iron Mountain”), located approximately 30 km southeast of Preah Khan, supplied magnetite ore to the Preah Khan region7 and constituted one of mainland Southeast Asia’s largest premodern iron‑production zone8. In this area, at least three distinct smelting traditions have been identified based on tuyere gauge (broad, medium, and thin). Broad-gauge tuyeres correspond to the early Angkor period, medium-gauge tuyeres to the expansion phase (10th − 13th centuries) through the early post-Angkor period, and thin-gauge tuyeres to post-Angkor smelting by the Kuay people8. Geochemical and microstructural studies of Angkorian iron products also suggest that beginning in the 11th century, changes in smelting practice led to improved iron quality in some areas of the Preah Khan region3. By the 14th century, Preah Khan was occupied by Ayutthaya, and ironmaking became active within the walled complex9. Although the Kuay people are known to have conducted ironmaking around Phnom Dek after the post-Angkor period, the identity of iron producers during the Angkor period itself remains uncertain8. Moreover, little is known about how temporal changes in iron production—thought to be linked to the rise and decline of the Angkorian Empire—correlate with shifts in ore formulations and supply sources. Previous studies at Preah Khan reported Mn-bearing slags containing up to 8.9 wt% Mn in fayalite (ferrosilicate mineral), produced from iron ore with approximately 4 wt% Mn in magnetite7. However, our recent field survey, in approximately 90 km north of Preah Khan in northern Cambodia, has revealed exceptionally Mn-rich slags—exceeding 24 wt% Mn in bulk composition—together with Mn ores (Fig. 1). Manganese-rich slags have been reported in pre-industrial ironmaking sites worldwide, such as in Germany10, Uganda11, France12, and Japan13. In bloomery iron smelting, which is the predominant pre-industrial ironmaking method, ferrosilicate slag forms through the reaction of SiO2 and FeO at approximately 1200 ℃. Because MnO is substitutes for FeO in ferrosilicate slag, the addition of Mn-rich ores can enhance the iron yield14. In the study area, Mn-poor slags, similar to those found at other Khmer ironmaking sites, were also encountered in several slag mounds (Table 1). As slag chemistry records ironmaking process14, compositional variation among slags is expected to reflect differences in ore formulation and ironmaking technology. Therefore, the chemical compositions and Nd-, Sr-, and Pb-isotope ratios of slags and ores were analysed to reveal the origin of the ores, ore formulations, and the evolution of ironmaking technology in the study area during the Angkor period. Study area and sampling Geological background The samples were collected in and around the Techo Sen Russey Treb Park (within an area of approximately 18 km) in the Preah Vihear Province, northern Cambodia (Fig. 1). The study area is located approximately 90 km northeast of Preah Khan. Although a detailed geological survey has not been conducted in the study area, Jurassic, Permian-to-Triassic, and Devonian-to-Carboniferous sedimentary rocks and Carboniferous-to-Triassic volcanic rocks (andesite to rhyolite) have been identified around the study area15. Several traces of oceanic crustal subduction have been observed in Cambodia (Fig. 1). The Siem Reap-Stung Treng structural zone, where the study area is located, was formed as a result of the subduction and closure of the Paleo-Tethys Ocean crust in the Late Permian to Middle Triassic periods16,17. The Jurassic, Permian-to-Triassic, and Devonian-to-Carboniferous sedimentary rocks around the study area have been reported include marine sediments such as chert, shale, and limestone15. Accordingly, the diversity of geological formations of different ages and origins around the study area suggests the potential occurrence of a wide variety of mineral resources derived from magmatic, sedimentary, and weathering processes. Samples Most samples were collected from slag mounds at ancient ironmaking sites (Table 1; Fig. 1). These ancient ironmaking sites were generally located near lowlands or ponds. The slag mounds consisted of a vast number of slag fragments exposed at the surface (Supplementary Fig. S1). In addition to slags and ores, furnace debris was found in some slag mounds. In addition to Mn-rich slags and Mn-poor slags, slags that were glassier than typical were observed in some slag mounds (sites I10–I13). Glassier slags are a major component of the slag mounds at these sites. Based on visual inspection, hardness, and preliminary chemical analyses performed with a handheld X-ray fluorescence (XRF) analyzer, the slags at each site were judged to be almost homogeneous. Therefore, three to four representative samples were collected from each site. However, because typical and glassier slags were observed at sites I10 and I11, both types of slag were collected. Charcoal fragments were discovered within slags at five sites (I2, I3, I9, I11, and I15) and were used for 14C dating. The slag samples I1-S1 and I1-S2, collected from site I1, contained metal iron, which was subsequently hand-picked as a subsample during the pulverisation of the slags. Several ore samples were collected from country roadsides (sites O1–O6). Because the Fe–Mn ore samples (I10-O2 and I12-O1) contained red chert, they were cut into two subsamples, namely, ore- and chert-rich portions. The chemical composition of the Mn-poor slags collected in a previous study7 from Preah Khan region were also analysed for comparison with the slag samples used in this study. Results Slag samples The Mn-rich slags exhibited high Mn and Cu contents (24.3–38.5 wt% and 4882–9123 ppm, respectively), high Mn/Fe mass ratios (1.81–7.91), and negative Eu anomalies in the C1-chondrite-normalised rare-earth elements and yttrium (REY) pattern (Fig. 2 and Supplementary Table S1). Although most of the Mn-rich slags exhibited negative Ce anomalies, some showed positive Ce anomalies and negative Y/Ho ratios. The Mn-rich slags contained olivine and sulphide Cu-Fe-Mn (Supplementary Tables S1 and S2 and Supplementary Fig. S2). In contrast, the Mn-poor slags exhibited high Fe contents (56.6–83.2 wt%), low Mn/Fe mass ratios (2.34 × 10− 3–6.40 × 10− 3), positive Ce anomalies, and light-REY enriched pattern (Fig. 2 and Supplementary Table S1). They consisted of olivine and wüstite. Their Nd- and Pb- isotope ratios tend to be higher than those of the Mn-poor slags (Fig. 3). No significant isotopic fractionation was observed between the metal iron and surrounding Mn-poor slag (Fig. 3 and Supplementary Fig. S3), indicating that Nd-Sr-Pb isotopic compositions of the slag samples reflect those of the materials introduced into furnace. Some slag samples exhibit high Al + Si and low Fe + Mn contents and different Nd-Sr-Pb isotopic trends (Supplementary Figs. S3 and S4); therefore, they were determined to be furnace-wall molten and furnace-wall mixed slags. Thus, they were excluded from the ore formulation estimation. The 14C age of charcoal collected from Mn-poor slag at site I2 was from the late 13th to the 14th century (Table 1). In contrast, the age of charcoal from Mn-rich slag (site I3, I9, I11, and I15) were from the late 10th to the 13th century. Ore samples The ore samples were classified as Fe-rich, ferromanganese (Fe–Mn), Mn-rich, laterite, and sulphide-bearing. Furthermore, the Fe-rich ore was subdivided into magnetite and hematite ores. The Mn/Fe mass ratios of the Fe-rich, Fe–Mn, and Mn-rich ores were 7.3 × 10− 5–1.2 × 10− 3, 0.25–0.37, and 5.3–1210, respectively (Supplementary Table S1). The sulphide-bearing ores were rich in Cu (up to 1.62 wt%). The magnetite-ore samples are almost entirely composed of pure magnetite with a minor hematite (Supplementary Table S1 and Supplementary Fig. S2). Hematite was contained in the hematite ore, Fe–Mn ore, and laterite. The Mn-rich ores contained Mn oxide minerals (e.g., braunite). The sulphide-bearing ore samples were volcanic and weathered rocks containing pyrite and minor amounts of chalcopyrite. Quartz was contained in the almost ore samples. The hematite-ore samples were characterised by negative Ce anomalies (Fig. 2). In contrast, the magnetite-ore samples exhibited no Ce anomalies, excepting for two samples. The Fe–Mn and Mn-rich ore samples were characterised by large negative Ce and Eu anomalies. Among the ore samples, only laterite samples exhibited large positive Ce anomalies and negative Y/Ho ratios. The Sr-isotope ratios and 208Pb/204Pb ratios of ore samples other than laterite generally fell within the ranges of 0.7064–0.7074 and 37.72–38.15, respectively (Fig. 3). In contrast, the laterite samples exhibited lower Sr-isotope ratios and higher 208Pb/204Pb ratios. The Mn-rich and Fe–Mn ores tended to show lower Nd-isotope ratios than that of the Fe-rich ores. In addition, the Mn-rich and Fe–Mn ores can be distinguished from each other based on their 206Pb/204Pb and 207Pb/204Pb ratios. Discussion Origin of ores In pre-industrial ironmaking, ores should have been supplied from the vicinities of ironmaking sites. Therefore, ores found in slag mounds as well as those collected from country roadsides in the study area are potential ores that Angkorian iron producers used. To reveal the diversity of ores they used, we identified the origin of the ore samples. Negative Ce anomaly indicates that the samples incorporated REY from a Ce-depleted material in oxic environment such as deep-sea seawater18. Therefore, the hematite, Fe–Mn, and Mn-rich ores likely formed in submarine environment. Moreover, the hematite, Fe–Mn, and Mn-rich ore samples are classified as having a hydrothermal origin in the geochemical discrimination diagram for submarine Fe–Mn oxides (Supplementary Fig. S5)19. The Mn/Fe mass ratios and REY patterns of the Fe–Mn and Mn-rich ore samples are similar to those of Fe- and Mn-oxide deposits in accretionary complexes, respectively (Fig. 2). Additionally, the Fe–Mn ore samples are accompanied by chert. Iron-oxide deposits in accretionary complexes, known as umbers, are commonly associated with mid-ocean ridge (MOR) basalts20–24, while Mn-oxide deposits are associated with oceanic island basalts24,25. This suggests that both deposit types formed through hydrothermal activity occurring at MORs and oceanic islands, respectively24,25. However, the hematite-ore samples exhibit lower Mn/Fe mass ratios (7.3 × 10− 5–1.9 × 10− 4) than that of umbers. Rocks with the same characteristics as the samples are known to be ironstones (Mn/Fe = 2.2 × 10− 3–1.9 × 10− 5) resulting from hydrothermal activity in seamounts26,27. Notably, an iron mine, likely composed of ironstone, has been reported in Cambodia15. Sedimentary rocks (chert, limestone, and marine shales) associated with the subduction of the Paleo-Tethys Ocean are distributed in and around the study area (Fig. 1), which is consistent with the estimated origins of the hematite, Fe–Mn, and Mn-rich ores . Marine sedimentary rocks distributed around the study area formed from Devonian to Jurassic. The Sr-isotope ratios of the hematite ores (ironstone), Fe–Mn ores (umber), and Mn-rich ores (hydrothermal Mn oxides) are primarily within the range of the Devonian-to-Jurassic marine Sr-isotope ratios28 (Fig. 3). The low Sr-isotope ratios observed in some hematite and Mn-rich ore samples may be owing to the influence of volcanic components from oceanic islands, which generally have low Sr-isotope ratios29. Therefore, the ores resulting from submarine hydrothermal activity were likely supplied from the vicinity of the study area. The Mn-rich ores are divided into those with high and low Nd-isotope ratios. The Nd-isotope ratio of marine Mn oxides reflects that of the ambient water mass at the time of formation30; therefore, the Mn-rich ores may have been supplied from multiple mines with different formation sites. The magnetite-ore samples are almost entirely composed of pure magnetite and are considered to be skarn magnetite based on the geochemical discrimination diagram for magnetite (Supplementary Fig. S6)31,32. This is consistent with the fact that most iron mines in Cambodia are classified as skarn deposits15,33. However, two magnetite-ore samples exhibit negative Ce anomalies, similar to those of the hematite-ore samples (Fig. 2a). Since magnetite in skarn deposits does not generally exhibit Ce anomalies34,35, these samples may have been metamorphosed ironstone. Although the origin of the sulphide-bearing ores is unknown, sulphide minerals are common in MOR and skarn deposits. Laterite, widely distributed in the study area, forms through intense chemical weathering and leaching of parent rocks in tropical settings, which concentrates Fe–Al (oxyhydr)oxides and develops thick weathering profiles36. Ironmaking using laterite during the Angkor period has been reported in Thailand37. In summary, Angkorian iron producers in the study area potentially used skarn magnetite, ironstone, umber, hydrothermal Mn oxides, sulphides, and laterite. Ore formulation of ironmaking The ore formulations for ironmaking in the study area were estimated from the chemical and Nd-Sr-Pb isotopic compositions of the samples (Supplementary Table S1). Cerium anomalies in Mn-rich slags are positively correlated with Eu anomalies (r = 0.67) and negatively correlated with the Y/Ho ratio (r = − 0.93) (Fig. 2). In addition, the Mn-rich slags have an average Cu content 153 times higher than the Mn-poor slags. These features indicate that the Mn-rich slags originated from a mixture of laterite with positive Ce anomalies and negative Y/Ho ratios, hydrothermal Mn oxides (Mn-rich ore) with negative Ce and Eu anomalies and positive Y/Ho ratios (Fig. 2), and sulphide-bearing ores with high Cu contents. This finding is also supported by the Nd-Sr-Pb isotopic compositions. The Nd- and Sr-isotope ratios of the Mn-rich slags can be explained by a mixture of laterite, umber, hydrothermal Mn oxides, and sulphide-bearing ore (Fig. 3). The Mn-rich slag from site I4 exhibits a higher Nd isotopic ratio than the other Mn-rich slags, most likely owing to the high Nd-isotope ratio of the Mn-rich ore collected at site I4. In contrast, the Pb-isotope ratios of Mn-rich slags are mainly controlled by the mixing of laterite and umber, with the Pb contents being approximately five times higher than those of other ores. The marginal shift to a high 206Pb/204Pb ratio shown by some Mn-rich slags may be owing to the influence of hydrothermal Mn oxides and sulphide-bearing ore. As described above, no geochemical traces of the Fe-rich ores (skarn magnetite and ironstone) were observed in the Mn-rich slags. Furthermore, the Fe-rich ores were not observed in the Mn-rich slag mounds (Table 1). Therefore, the Mn-rich slags result from ironmaking using laterite, umber, hydrothermal Mn oxides, and sulphides. The large dispersion in the Pb-isotope ratios of the hydrothermal Mn oxides precludes a reliable quantitative estimation of the mixing proportions among these ores. Nevertheless, the Nd-Sr-Pb isotopic compositions of the Mn-rich slags indicate that umber, rather than laterite, was predominantly employed as the iron ore. Because the collected sulphide-bearing ores exhibit low Fe contents owing to the high proportion of host rocks, their use as iron ore or an additive cannot be completely determined. However, Cu, which is rich in sulphide-bearing ores, is often added to steel in modern steelmaking practices along with Mn to increase its strength and weather resistance38,39. Therefore, sulphide-bearing ores may have been added to improve the quality of the iron products. Because Mn ores are added as a desulphurising agent in modern steelmaking, Angkorian iron producers may have added Mn-rich ore not only to improve the iron yield but also to desulphurise the sulphide-bearing ore. The Cu-Fe-Mn sulphide particles observed in the Mn-rich slag (Supplementary Fig. S2) were likely caused by desulphurisation. The positive correlation between Mn and Cu in the Mn-rich slags (r = 0.56) suggests that the added ratio of sulphide-bearing ore to Mn-rich ore was approximately constant. Furnace wall-molten slag and ore molten (Mn-rich) slag did not coexist at the same site, except for site I11. At site I11 as well, they appeared to be separated. The Angkorian iron producers might intentionally separate the slag derived from the ores and furnace walls for reasons such as to reuse Mn-rich slag as an MnO additive which was reported in Uganda11. The positive Ce anomalies and enriched light-REY compositions of the Mn-poor slags indicate the influence of laterite (Fig. 2). The Ca/Al mass ratios of the Mn-poor slags (4.72 × 10− 1–8.60 × 10− 1) are higher than those of laterite (1.42 × 10− 2 on average). Because the Ca/Al ratio between the ore mixture and slag is considered to be constant14, the Mn-poor slags must originate from a mixture of laterite and materials with high Ca/Al and low Mn/Fe ratios. Skarn magnetite exhibits higher Ca/Al mass ratios than ironstone (2.04 × 10− 1–1.05 and 9.65 × 10− 2–5.28 × 10− 1, respectively). Moreover, ironstone was not identified in the slag mounds but was collected from the country roadside. Therefore, skarn magnetite and laterite likely have been used as iron ores. The Nd- and Sr-isotope ratios of the Mn-poor slags are also consistent with the mixing of laterite and skarn magnetite (Fig. 2). The mixing proportions between skarn magnetite and laterite, calculated from the Nd- and Sr-isotope ratios of the Mn-poor slags, showed wide 95% confidence intervals (CI) due to the large dispersion of the skarn magnetite data (Supplementary Fig. S7). The 95% CI ranges for the estimated proportion of skarn magnetite overlapped at 30.5%–61.2% (site I1), 22.1%–79.2% (site I2), and 46.6%–98.0% (site I14). Although the Pb-isotope ratios of the Mn-poor slags from sites I1 and I2 are similar to those of the laterite samples, the Mn-poor slags from site I14 exhibit higher Pb-isotope ratios, which may be owing to the differences in the laterite supply source, which is widely distributed in the study area. Since laterite results from the weathering of rocks and soils36, its Pb-isotope ratios can vary depending on the Pb-isotope composition and U/Th ratio of the source materials. In summary, at least two ore formulations are considered to have been used in the study area, namely, formulation I—ironmaking from iron ores consisting of laterite and umber with the addition of hydrothermal Mn oxides and sulphides (at sites I3, I4, I5, I6, I7, I8, I9, I10, I11, I15, and I16); and formulation II—ironmaking from iron ores consisting of laterite and skarn magnetite (at sites I1, I2, and I14). Temporal changes in ironmaking technology The 14C ages of charcoal in the slags indicate that ironmaking based on formulation I was conducted from at least the end of the 10th to early 13th centuries, whereas ironmaking based on formulation II was conducted from the late 13th to 14th centuries. We inferred how temporal variations in ore formulation affected the efficiency of ironmaking. The reducible iron index (RII)14 indicates the amount of reducible iron remaining in the ferrosilicate slag and is defined based on the Si, Fe, and Mn contents (wt%) in slags (Eq. 1): 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{*{20}{c}} {RII=~\frac{{2.39 \times {\mathrm{Si}}{{\mathrm{O}}_2}}}{{{\mathrm{FeO}}+{\mathrm{MnO}}}},~} \end{array}$$\end{document} where 2.39 is a value used to convert the FeO/SiO2 mass ratio to that of the molar ratio in fayalite (Fe2SiO4). An RII < 1 indicates that excess FeO remains in the slag without being reduced; therefore, slags with an RII > 1 indicate a high iron yield. The RII of Mn-rich slags is approximately 1 (1.06–1.51), while that of Mn-poor slags is < 1 (0.17–0.61). The Mn-poor slags contain wüstite as excess FeO (Supplementary Fig. S2 and Supplementary Table S1). The low Ca content (0.71–1.72 wt%) and RII of the Mn-poor slags suggest that no attempts were made to improve the iron yield by adding Ca-rich materials. However, RII is unsuitable for evaluating Mn-rich slags in this study because it cannot evaluate the increase in iron yield due to the substitution of MnO, CaO, and MgO for FeO in the ferrosilicate slag. Therefore, we propose a new index, the iron-substitution rate (ISR), which is based on the Si and Fe contents (wt%) in the slags (Eq. 2): 2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{*{20}{c}} {ISR=~\left( {1 - \frac{{{\mathrm{FeO}}}}{{2.39 \times {\mathrm{Si}}{{\mathrm{O}}_2}}}{\text{}}} \right) \times 100.} \end{array}$$\end{document} The ISR indicates the quantity of FeO replaced in the slag and is expressed as a percentage. The absolute value of a negative ISR corresponds to the amount of excess FeO in the slag (in %). Whereas the Mn-poor slags exhibit negative ISRs (− 62.3% to − 495%), the Mn-rich slags show positive ISRs ranging from 69.1% to 90.6% (Table 1). The ISRs of the Mn-rich slags indicate that 69.1%–90.6% of the FeO that would normally have remained in ferrosilicate slags was substituted predominantly by MnO, demonstrating that metal iron was extracted from ores with exceptionally high efficiency. Therefore, the addition of hydrothermal Mn oxides improves the iron yield. The temperature at which the slag melts in coexistence with metal iron can be estimated based on the contents of the major elements. The estimated melting temperature of the Mn-rich slags (1150–1300 °C) was almost same with that of the Mn-poor slags (1200–1350 °C) (Fig. 4). Therefore, the addition of hydrothermal Mn oxides is unlikely to have required special measures to raise and maintain furnace temperature. The Mn-rich slags in the 12th and 13th centuries (site I15) exhibit a higher ISR than the Mn-rich slags in the end of the 10th century (sites I3 and I11), suggesting the modification of formulation I over time. The Mn-rich ores exhibit average Si/Al mass ratios and larger negative Ce anomalies than the other ore samples; however, the range of their Si/Al mass ratios is wide (1.60–103). Because the Si/Al ratio between the ore mixture and slag is considered to be constant14, the positive correlation between the ISR and Si/Al mass ratio (r = 0.84) and the negative correlation between the ISR and Ce anomalies (r = − 0.55) suggest that the increased addition of hydrothermal Mn oxides improves the iron yield. Therefore, Angkorian iron producers may have adjusted the amount of hydrothermal Mn oxides and modified formulation I to improve efficiency. The period during which ironmaking based on formulation I was conducted corresponds to the peak of the Angkorian Empire. Therefore, the ironmaking technology may have been refined because of the high demand for iron due to the construction of Khmer monuments and foreign expeditions. Formulation I may have been developed by adopting laterite, which is readily available but exhibit less Fe content than other iron ores, as the iron ore and adding hydrothermal Mn oxides to increase the iron yield to meet the high demand for iron. In some areas of Preah Khan region, higher-quality iron production compared to previous iron products is suggested to start from the 11th century based on the chemical composition and structure of iron crampons used in Khmer monuments3. However, the specific elemental concentrations of these irons have not been published. Because the iron produced using formulation I might be of high-quality owing to the addition of Cu and Mn, the study area was possibly one of the source of high-quality iron during the late 10th century to the 13th century. In contrast, simple and inefficient ironmaking based on formulation II was conducted from at least the 14th century, less than 100 years after the efficient ironmaking technology based on formulation I. This sudden decline in efficiency in ironmaking coincided with the decline of the Angkorian Empire owing to internal strife and the rise of Ayutthaya from approximately the 13th century. Therefore, it is more plausible that the discontinuation of ironmaking based on formulation I resulted not from the depletion of its ore sources, but rather from the social upheaval associated with the decline of the Angkorian Empire. Preah Khan that was an entrepôt for iron and an eastern outpost of the Angkorian Empire was occupied by the 14th century1,6. After that, ironmaking by the Kuay people became active around Preah Khan9. Skarn magnetite has been found in slag mounds around Preah Khan (11th − 16th centuries)7, and the slags in these mounds exhibit positive Ce anomalies and negative Y/Ho ratios (Supplementary Fig. S8). Therefore, iron was produced in this area using skarn magnetite and laterite, as in formulation II. Slag mounds exhibiting a positive ISR (11.4%–22.6%) first appeared at the end of the 13th century (Supplementary Fig. S9), and the Mn content of the slags increased to 2.34–5.88 wt%. The increase in Mn content in the slags is believed to reflect the use of Mn-bearing magnetite ores7, suggesting possibility that the concept of improving iron yield through MnO addition had been transmitted from the study area to this region by the late 13th century. However, conventional slag mounds with negative ISRs coexisted with slag mounds with positive ISRs around Preah Khan until at least the early 15th century. Therefore, it is also possible that, after the occupation of Preah Khan, the parts of ore sources simply shifted by chance to mines containing Mn-bearing magnetite. In any case, based on the ISR of the slags, the iron yield of ironmaking around Preah Khan was not as high as that of ironmaking based on formulation I in the study area, indicating that most of the technological knowledge for enhancing iron yield through the addition of Mn oxides had been lost. In summary, the technological change in ironmaking is suggested to have occurred through the following process. The social turmoil associated with the decline of the Angkorian Empire and the invasion by surrounding powers may have led to the sudden loss of efficient ironmaking based on formulation I in the study area between the 13th and early 14th centuries. Part of the ironmaking technology, which involves the addition of MnO, possibly spread to the Kuay people around Preah Khan from the end of the 13th century onwards, leading to improved iron yields. Perhaps, to mitigate the impact of the decreased iron production capacity owing to the loss of efficient ironmaking technology in the study area, ironmaking became active around Preah Khan in the 14th century9. However, because this study estimates the efficiency of ancient ironmaking based on the geochemistry of slags, practical experiments are required to verify the effectiveness of the addition of hydrothermal Mn oxide for improving iron yields. Although the study area belongs to the major iron production region of the Angkorian Empire, more ancient ironmaking sites must be investigated to assess the extent to which the ironmaking technological change observed in the early 14th century in the study area was prevalent in the Angkorian Empire. The geochemical approach used in this study, based on the chemical compositions (especially REY) and Nd-Sr-Pb isotope compositions of slags, will improve our understanding of the evolution of ancient ironmaking technologies linked to social situations. Conclusions This study revealed the temporal changes in the ore formulation of ironmaking during the Angkor period for the first time. In the study area, between the late 10th and 13th centuries, iron was produced using umber and laterite as iron ores, with hydrothermal Mn oxides and sulphides as additives. The addition of hydrothermal Mn oxides enhanced the iron yield, and iron producers perhaps adjusted the proportion of these Mn oxides to further improve smelting efficiency. However, by at least the 14th century, this efficient ironmaking technology had been lost, and less efficient smelting using only umber and skarn magnetite became prevalent. The period of this technological change coincides with the decline of the Angkorian Empire, suggesting that social upheaval—such as invasions and political instability—may have led to the disappearance of the advanced ironmaking technology. Fig. 1Sample locations. (a) Simplified geological map modified from previous studies15,40. The shaded topological map in the background was drawn using GMT software41 using SRTM15 + topological data42. (b) Satellite view of the study area. (Modified from Google Earth, ©2024 Google). Red stars indicate the sample locations. Sample locations. (a) Simplified geological map modified from previous studies15,40. The shaded topological map in the background was drawn using GMT software41 using SRTM15 + topological data42. (b) Satellite view of the study area. (Modified from Google Earth, ©2024 Google). Red stars indicate the sample locations. Fig. 2C1-chondrite-normalised REY pattern of the samples. (a) REY patterns of the Fe-rich ore samples shown with those of skarn magnetite34,35 and ironstones26,27. (b) REY patterns of the Mn-rich and Fe–Mn ore samples shown with those of umbers21,43 and hydrothermal Mn oxides24,44. (c) REY patterns of the slag samples. (d) REY patterns of the laterite, chert, sulphide-bearing ore, and metallic-iron samples. (e) Cerium anomalies versus Eu anomalies of the samples. (f) Cerium anomalies versus Y/Ho ratio anomalies of the samples. The REY compositions of the samples and reference materials were normalised by C1-chondrite45. C1-chondrite-normalised REY pattern of the samples. (a) REY patterns of the Fe-rich ore samples shown with those of skarn magnetite34,35 and ironstones26,27. (b) REY patterns of the Mn-rich and Fe–Mn ore samples shown with those of umbers21,43 and hydrothermal Mn oxides24,44. (c) REY patterns of the slag samples. (d) REY patterns of the laterite, chert, sulphide-bearing ore, and metallic-iron samples. (e) Cerium anomalies versus Eu anomalies of the samples. (f) Cerium anomalies versus Y/Ho ratio anomalies of the samples. The REY compositions of the samples and reference materials were normalised by C1-chondrite45. Fig. 3Nd-Sr-Pb isotope ratios of the samples. (a) Nd isotope ratio versus Sr isotope ratio. Dotted blue vertical line indicates the range of marine Sr-isotope ratios from the Devonian to Jurassic periods28. (b) 208Pb/204Pb ratio versus 207Pb/204Pb ratio. (c) 206Pb/204Pb ratio versus 208Pb/204Pb ratio. Mixing curves were calculated from the weighted mean values of the ore samples; the samples used to calculate each weighted mean are enclosed by hatched areas of the same colour. Mixing curves are marked with crosses at 20% increments in the mixing ratio. The pairs of metal iron and the surrounding slag are indicated by dotted circle. Nd-Sr-Pb isotope ratios of the samples. (a) Nd isotope ratio versus Sr isotope ratio. Dotted blue vertical line indicates the range of marine Sr-isotope ratios from the Devonian to Jurassic periods28. (b) 208Pb/204Pb ratio versus 207Pb/204Pb ratio. (c) 206Pb/204Pb ratio versus 208Pb/204Pb ratio. Mixing curves were calculated from the weighted mean values of the ore samples; the samples used to calculate each weighted mean are enclosed by hatched areas of the same colour. Mixing curves are marked with crosses at 20% increments in the mixing ratio. The pairs of metal iron and the surrounding slag are indicated by dotted circle. Fig. 4Liquidus of slags. (a) Liquidus in the SiO2-FeO-Al2O3 system46. Since MnO exhibits a similar behaviour to FeO in ferrosilicate slag, the Mn-rich slags were applied to the FeO + MnO content instead of the FeO content. (b) The liquidus in the SiO2-FeO-MnO system was calculated using the FactPS and FToxide databases in FactSage (ver. 8.1)47. In both diagrams, oxide phases are in equilibrium with metal iron and the colour lines indicate the liquidus at each temperature. Liquidus of slags. (a) Liquidus in the SiO2-FeO-Al2O3 system46. Since MnO exhibits a similar behaviour to FeO in ferrosilicate slag, the Mn-rich slags were applied to the FeO + MnO content instead of the FeO content. (b) The liquidus in the SiO2-FeO-MnO system was calculated using the FactPS and FToxide databases in FactSage (ver. 8.1)47. In both diagrams, oxide phases are in equilibrium with metal iron and the colour lines indicate the liquidus at each temperature. Table 1Details of study sites.LocationSite No.Lon. (N)Lat. (E)SamplesOre formulation*14C age (2σ) of charcoalISR**Ironmaking siteI113° 50’ 48.4”105° 26’ 00.8”Mn-poor slag and Fe-rich ore (magnetite)II−62.0% and − 464.3%I213° 50’ 48.6”105° 26’ 02.4”Mn-poor slag and Fe-rich ore (magnetite)II1289–1325 cal AD (41.00%)1354–1393 cal AD (54.45%)−91.3% and − 495.7%I313° 50’ 47.9”105° 26’ 06.6”Mn-rich slag and Mn-rich oreI906–908 cal AD (0.19%)977–1038 cal AD (95.26%)75.4% and 81.0%I413° 50’ 45.7”105° 26’ 01.5”Mn-rich slag and Mn-rich oreI69.1% and 76.5%I513° 50’ 52.7”105° 25’ 40.9”Mn-rich slag and sulphide-bearing oreI72.0% and 74.5%I613° 50’ 57.4”105° 25’ 40.5”Mn-rich slagI77.7%, 84.0%, and 90.6%I713° 50’ 57.6”105° 25’ 41.4”Mn-rich slagI79.5%I813° 50’ 57.4”105° 25’ 43.2”Mn-rich slagII913° 50’ 59.8”105° 25’ 43.2”Mn-rich slagI1042–1087 cal AD (46.82%)1090–1107 cal AD (6.28%)1116–1179 cal AD (37.34%)1191–1207 cal AD (5.02%)74.8%I1013° 50’ 48.7”105° 25’ 41.8”Mn-rich slag, Mn-rich ore, and Fe-Mn oreI77.4%I1113° 53’ 48.8”105° 23’ 25.8”Mn-rich slag, furnace-wall molten slag, Mn-rich ore, and sulphide-bearing oreI993–1029 cal AD82.1%I1213° 53’ 50.2”105° 23’ 24.3”Furnace-wall molten slag and Fe-Mn oreI1313° 53’ 48.1”105° 23’ 24.8”Furnace-wall molten slagI1413° 50’ 51.7”105° 24’ 51.8”Furnace-wall molten slag−189.5%I1513° 44’ 36.2”105° 27’ 01.7”Mn-rich slag and Mn-rich oreI1162–1221 cal AD83.5% and 85.2%I1613° 44’ 33.5”105° 27’ 07.5”Mn-rich slag and Mn-rich oreI75.3% and 79.2%RoadsideO113° 45’ 15.9”105° 26’ 00.0”Fe-rich ore (hematite)O213° 45’ 09.9”105° 26’ 00.0”Fe-rich ore (magnetite and hematite)O313° 53’ 41.3”105° 25’ 31.4”LateriteO413° 45’ 15.6”105° 25’ 03.1”Fe-rich ore (magnetite)O613° 44’ 44.3”105° 25’ 38.2”Fe-rich ore (magnetite and hematite)*Ore formulation I: laterite and umber with the addition of hydrothermal Mn oxides and sulphides; ore formulation II: laterite and skarn magnetite.**Iron-substitution rate indicating smelting efficiency represented by Eq. (2). Table 1Details of study sites.LocationSite No.Lon. (N)Lat. (E)SamplesOre formulation*14C age (2σ) of charcoalISR**Ironmaking siteI113° 50’ 48.4”105° 26’ 00.8”Mn-poor slag and Fe-rich ore (magnetite)II−62.0% and − 464.3%I213° 50’ 48.6”105° 26’ 02.4”Mn-poor slag and Fe-rich ore (magnetite)II1289–1325 cal AD (41.00%)1354–1393 cal AD (54.45%)−91.3% and − 495.7%I313° 50’ 47.9”105° 26’ 06.6”Mn-rich slag and Mn-rich oreI906–908 cal AD (0.19%)977–1038 cal AD (95.26%)75.4% and 81.0%I413° 50’ 45.7”105° 26’ 01.5”Mn-rich slag and Mn-rich oreI69.1% and 76.5%I513° 50’ 52.7”105° 25’ 40.9”Mn-rich slag and sulphide-bearing oreI72.0% and 74.5%I613° 50’ 57.4”105° 25’ 40.5”Mn-rich slagI77.7%, 84.0%, and 90.6%I713° 50’ 57.6”105° 25’ 41.4”Mn-rich slagI79.5%I813° 50’ 57.4”105° 25’ 43.2”Mn-rich slagII913° 50’ 59.8”105° 25’ 43.2”Mn-rich slagI1042–1087 cal AD (46.82%)1090–1107 cal AD (6.28%)1116–1179 cal AD (37.34%)1191–1207 cal AD (5.02%)74.8%I1013° 50’ 48.7”105° 25’ 41.8”Mn-rich slag, Mn-rich ore, and Fe-Mn oreI77.4%I1113° 53’ 48.8”105° 23’ 25.8”Mn-rich slag, furnace-wall molten slag, Mn-rich ore, and sulphide-bearing oreI993–1029 cal AD82.1%I1213° 53’ 50.2”105° 23’ 24.3”Furnace-wall molten slag and Fe-Mn oreI1313° 53’ 48.1”105° 23’ 24.8”Furnace-wall molten slagI1413° 50’ 51.7”105° 24’ 51.8”Furnace-wall molten slag−189.5%I1513° 44’ 36.2”105° 27’ 01.7”Mn-rich slag and Mn-rich oreI1162–1221 cal AD83.5% and 85.2%I1613° 44’ 33.5”105° 27’ 07.5”Mn-rich slag and Mn-rich oreI75.3% and 79.2%RoadsideO113° 45’ 15.9”105° 26’ 00.0”Fe-rich ore (hematite)O213° 45’ 09.9”105° 26’ 00.0”Fe-rich ore (magnetite and hematite)O313° 53’ 41.3”105° 25’ 31.4”LateriteO413° 45’ 15.6”105° 25’ 03.1”Fe-rich ore (magnetite)O613° 44’ 44.3”105° 25’ 38.2”Fe-rich ore (magnetite and hematite)*Ore formulation I: laterite and umber with the addition of hydrothermal Mn oxides and sulphides; ore formulation II: laterite and skarn magnetite.**Iron-substitution rate indicating smelting efficiency represented by Eq. (2). Details of study sites. Mn-poor slag and Fe-rich ore (magnetite) −62.0% and − 464.3% Mn-poor slag and Fe-rich ore (magnetite) 1289–1325 cal AD (41.00%) 1354–1393 cal AD (54.45%) −91.3% and − 495.7% Mn-rich slag and Mn-rich ore 906–908 cal AD (0.19%) 977–1038 cal AD (95.26%) 75.4% and 81.0% Mn-rich slag and Mn-rich ore 69.1% and 76.5% Mn-rich slag and sulphide-bearing ore 72.0% and 74.5% 77.7%, 84.0%, and 90.6% 1042–1087 cal AD (46.82%) 1090–1107 cal AD (6.28%) 1116–1179 cal AD (37.34%) 1191–1207 cal AD (5.02%) Mn-rich slag, Mn-rich ore, and Fe-Mn ore Mn-rich slag, furnace-wall molten slag, Mn-rich ore, and sulphide- bearing ore Furnace-wall molten slag and Fe-Mn ore Mn-rich slag and Mn-rich ore 83.5% and 85.2% Mn-rich slag and Mn-rich ore 75.3% and 79.2% Fe-rich ore (magnetite and hematite) Fe-rich ore (magnetite and hematite) *Ore formulation I: laterite and umber with the addition of hydrothermal Mn oxides and sulphides; ore formulation II: laterite and skarn magnetite. **Iron-substitution rate indicating smelting efficiency represented by Eq. (2). Materials and methods Sample preparation The surfaces of the cut sample slabs were polished and then ultrasonically cleaned in ultrapure water for at least 10 min. The samples were then thoroughly rinsed with ultrapure water and dried. Finally, the samples were pulverised using an agate mortar and pestle. The powdered samples were dried at 110 °C for 12 h before undergoing geochemical analyses. However, the metal-iron samples were difficult to powder; thus, they were dissolved in 0.5 N HCl at 80 ℃ to minimise the dissolution of slag fragments adhering to the metal-iron samples. Dissolved metal-iron samples were filtered through a 0.45-µm mesh syringe filter. 14C dating All charcoal detected in the slags was collected for 14C dating. 14C dating was performed by Paleo Labo Co., Ltd48. However, because the measured 14C ages indicate the formation age of the wood from which the charcoal was made, ironmaking was potentially performed at a period slightly younger than the measured 14C ages. Chemical composition analyses Major (Na, Al, Si, P, Ca, Ti, Mn, Fe) and trace elements were analysed using XRF spectroscopy (ZSX Primus II, Rigaku, Japan) and inductively coupled plasma mass spectroscopy (ICP-MS; iCAP Q, Thermo Fisher Scientific, USA), respectively, at the Department of Systems Innovation, University of Tokyo. The XRF and ICP-MS analyses and correction of the spectral interference during the ICP-MS analyses were performed according to methods used in previous studies20,49–51. The analytical results for the geological reference material (JB-2) differed from the certified values52 by typically less than 1% for major elements and less than 5% for trace elements (Supplementary Table S1). Nd-Sr-Pb isotopic analysis The Nd- and Sr-isotope ratios (143Nd/144Nd and 87Sr/86Sr, respectively) of the samples were analysed using thermal ionisation mass spectrometry (Triton XT, Thermo Fisher Scientific, USA) at the Ocean Resources Research Center for Next Generation (ORCeNG), Chiba Institute of Technology. The Pb-isotope ratios (206Pb/204Pb, 207Pb/204Pb, and 208Pb/204Pb) of the samples were analysed using the multi-collector ICP-MS (Neptune plus, Thermo Fisher Scientific, USA) at the ORCeNG. Amount of the aliquot for isotopic analysis was typically adjusted to contain at least 500 ng of Sr, 200 ng of Nd, and 150 ng of Pb. The Fe-rich and Fe–Mn ore samples were decomposed in aqua regia at 80 ℃. The remaining samples were decomposed in a mixed acid (8 N HBr:20 N HF = 1:1 in volume ratio) at 110 ℃. After decomposition, the sample solutions were dried and then dissolved in 0.5 N HBr at 50 ℃. Column separation and isotopic analyses were performed for Nd, Sr, and Pb following the procedures described in a previous study53. The averaged procedural blank for Pb was 22 pg (n = 4, 2SD = 12 pg), which is negligible compared with the Pb content of the samples (10–200 ng). The relative deviation between the measured and reference isotopic ratios for JB-2 was approximately 5 × 10− 3% for Nd and Sr and 0.1% for Pb (Supplementary Table S1). The analytical precision of the isotopic analyses is comparable to that reported in recent geochemical studies54,55. Bayesian estimation of ore mixing proportions The mixing proportion (f) of skarn magnetite in the Mn-poor slags was estimated using a Bayesian isotope-mixing model implemented in R. The model considers both the isotope ratios and concentrations of Sr and Nd for end-member A (skarn magnetite) and B (laterite), and accounts for their uncertainties by assuming multivariate normal priors for each end-member. The model was formulated as follows: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{*{20}{c}} {{R_A}}&{\sim \mathcal{N}\left( {{\mu _A},{{\mathrm{\boldsymbol{\Sigma}}}_A}} \right),} \\ {{R_B}}&{\sim \mathcal{N}\left( {{\mu _B},{{\mathrm{\boldsymbol{\Sigma}}}_B}} \right),} \\ {{R_{{\mathrm{mix}},i}}}&{=\frac{{{f_i}\,{C_A}\,{R_A}+\left( {1 - {f_i}} \right)\,{C_B}\,{R_B}}}{{{f_i}\,{C_A}+\left( {1 - {f_i}} \right)\,{C_B}}},} \\ {{R_{{\mathrm{obs}},i}}}&{\sim \mathcal{N}\left( {{R_{{\mathrm{mix}},i}},{\sigma _{{\mathrm{obs}},i}}} \right),} \end{array}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_A}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_B}$$\end{document} are vectors of the Sr- and Nd-isotope ratios for end-members A and B, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C_A}$$\end{document}and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C_B}$$\end{document} are their corresponding concentrations, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{{\mathrm{mix}},~~i}}$$\end{document} is the predicted isotopic composition of the mixture for sample i, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{{\mathrm{obs}},i}}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma _{{\mathrm{obs}},i}}$$\end{document} are the observed isotope ratios and their measurement uncertainties (standard errors). Prior uncertainties for the end-members were based on their isotopic means and covariance matrices derived from repeated analyses. The model was implemented in Stan56 via the cmdstanr package in R, using four MCMC chains (2000 sampling iterations after 1000 warm-up iterations). The R script used for model execution is provided as Supplementary Code S1. Convergence of the Markov Chain Monte Carlo sampling was confirmed by ensuring that all R-hat values were below 1.01 and that effective sample sizes exceeded 400 for all parameters. Mineral composition analyses The mineral compositions of the samples were determined using X-ray diffraction (XRD; Ultima IV, Rigaku, Japan) at the Department of Systems Innovation, University of Tokyo and a secondary electron microscope with an energy dispersive X-ray spectrometer (JMS-6360; JEOL Ltd., Japan) at the Faculty of Science and Engineering, Waseda University. XRD analyses were conducted using a tube voltage of 40 kV, tube current of 40 mA, scan range of 5°–65° (2θ), and scan rate of 4°/min. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1