A 1179-yr (417–1595 CE) tree-ring oxygen isotope chronology for northern Japan validated using the 774–775 CE radiocarbon spike

Abstract

We present an annual-resolution, millennium-long tree-ring chronology for northern Japan. The chronology is based on 5309 measurements of tree-ring δ¹⁸O from 37 samples of Hiba arbor-vitae (Thujopsis dolabrata var. hondae). Although the exact geographical origin of 27 of the samples is unknown because they were extracted from excavated archaeological material, pattern matching of the tree-ring δ¹⁸O variations was robust among all 37 samples. The floating chronology constructed using all samples was cross-dated against a previously published δ¹⁸O chronology from central Japan, yielding a correlation coefficient of 0.26 (t = 9.0; p < 0.01), resulting in a temporal coverage of 417–1595 CE (i.e., 1179 yrs). The global ¹⁴C spike event at 774–775 CE was clearly recorded in the annual ¹⁴C data, which provides independent support for the dating of tree rings using oxygen isotopes. Furthermore, this δ¹⁸O chronology from northern Japan was used to successfully cross-date a wood sample buried during the “Millennium Eruption” of Baitoushan, which is located on the border between China and North Korea.

1. Introduction

Oxygen isotope dendrochronology has developed rapidly in Japan over recent decades. The method is particularly useful because tree-ring δ¹⁸O levels are primarily governed by two climatic factors, the δ¹⁸O in precipitation and relative humidity (Ramesh et al 1986; Robertson et al 2001), both of which vary with the hydroclimate (i.e., wet–dry conditions) during the summer monsoon season in Japan (Nakatsuka et al 2020). Therefore, variations in tree-ring δ¹⁸O are closely correlated among different trees, irrespective of the species. Three millennial-scale δ¹⁸O chronologies have been produced for Japan, covering the past 4354 yrs (2349 BCE to 2005 CE). Two chronologies cover central Japan and the periods 612 BCE to 2005 CE (Nakatsuka et al 2020; Sano et al 2022) and 2349–1009 BCE, and the other covers northern Japan and the period 1412–466 BCE (Sano et al 2023). These chronologies have been widely used to date archaeological wood samples excavated in Japan (e.g., Sano et al 2022).

However, the available tree-ring δ¹⁸O dataset has spatial and temporal limitations. In particular, tree-ring δ¹⁸O data covering the last 2500 yrs—during which numerous radiocarbon age data have been derived from excavated wood—rely only on the single chronology from central Japan. Therefore, samples from sites distant from the origin of this chronology often cannot be dated. Spatial correlations of tree-ring δ¹⁸O series between any two samples distant from each other do not simply weaken depending on their geographical distance, but are also modulated by the large-scale climatic conditions. The spatial pattern of summer precipitation is expressed as loadings of the first empirical orthogonal function (EOF; Figure 1). The signs (positive or negative) of the loadings, which represent correlations against the first EOF, split meridionally, highlighting the north–south contrast associated with the rainfall pattern across Japan. This pattern, which is enhanced by the Meiyu–Baiu front (a zonally oriented rain band), means that tree-ring δ¹⁸O series from two samples from a similar latitude (but distant from each other) will be relatively well-correlated. In fact, modern tree-ring δ¹⁸O data obtained from South Korea are closely correlated with our chronology from central Japan (Seo et al 2019), as both sites (ca. 1000 km apart) are located at a similar latitude. In contrast, two δ¹⁸O chronologies originating from different latitudes (ca. 400 km apart) in central and northern Japan show a significant but relatively weaker correlation (Sano et al 2023). These findings indicate that the development of a meridionally distributed tree-ring network is required if we wish to improve the performance of tree-ring dating in Japan, and possibly also across Korea and China.

Figure 1. Map of Japan showing the locations of the Sarugamori and Aomori sites (green triangles), where dead trunk (Sarugamori) and archaeological (Aomori) samples were collected for this study; the Baitoushan site (yellow triangle), where sample C5 dated using the 774–775 CE radiocarbon spike event was collected (Hakozaki et al 2018a); and central Japan (oval), where 67 samples were collected for the 2617-yr master chronology that extends to the present day (Nakatsuka et al 2020; Sano et al 2022). Background colors represent first EOF loadings based on June–August precipitation data for the period 1891–2019 CE. We also used the Global Precipitation Climatology Centre (GPCC) Full Data Monthly Product Version 2020 (Schneider et al 2020) as our precipitation dataset (0.25° × 0.25°).

High-resolution temporal records of radiocarbon content often allow us to determine the exact calendar year of a wood sample without using conventional tree-ring dating if a sharp spike in ¹⁴C concentration of precisely known age is present within a sequence of tree rings. Rapid changes in the ¹⁴C content of tree rings dated to 774–775 and 993–994 CE are good examples of this approach and have been observed in wood samples worldwide (Büntgen et al 2018; Miyake et al 2012; Park et al 2017). Such ¹⁴C spike events have been widely used for single-year radiocarbon dating (Hakozaki et al. 2018a, Kuitems et al 2022; Oppenheimer et al 2017; Meadows et al 2023; Philippsen et al 2022; Wacker et al 2014).

In this study, we developed a 1179-yr tree-ring δ¹⁸O chronology for northern Japan, using a total of 37 samples collected from two regions (Hakozaki et al 2011, 2016, 2017, 2018b). The chronology covers the period 417–1595 CE and was cross-dated against another δ¹⁸O chronology from central Japan (Nakatsuka et al 2020; Sano et al 2022). One of our cross-dated samples from the period 770–780 CE was also used for annual-resolution radiocarbon dating to verify our δ¹⁸O-based dates by reproducing the sharp ¹⁴C spike observed globally at 774–775 CE (Büntgen et al 2018). The utility of our newly developed chronology was evaluated by cross-dating a wood sample, which was buried by the “Millennium Eruption” of the Baitoushan volcano located on the border between China and North Korea. The tree-ring record from the master chronology is archived as Supplementary Material accompanying this article.

2. Materials and Methods

2.1. Study Sites and Sampling

The tree-ring samples used in this study were obtained from two regions in northern Japan (Figure 1). The first site was Sarugamori, where old, dead trunks of Hiba arbor-vitae (Thujopsis dolabrata var. hondae) remain within a planted pine forest. These trees have been buried owing to the deposition of sand dunes, induced partly by past human activities (Okamoto et al 2000). A total of 10 samples were collected from trunks that were decaying in situ at Sarugamori. The second site was located in Aomori, which is ca. 80 km from Sarugamori. A total of 27 samples of Hiba arbor-vitae were collected from wooden material recovered from five archaeological sites, all of which were contained within a circle with a radius of ca. 15 km. As will be discussed later, the tree-ring δ¹⁸O series obtained from the samples from Sarugamori and Aomori are closely correlated, indicating that the geographical origin of these archaeological samples is not likely to have been far from Sarugamori. All 37 samples from the two regions were used to develop a master chronology for northern Japan. However, although we also measured the tree-ring widths of these samples, we were unable to successfully cross-date most of these series because they lacked common variations (see Supplementary Figure 1). We believe that the sampled trees had been growing at sites with little climate-related stress. Consequently, we excluded the ring-width series obtained from these samples from our study.

We also used one Korean pine sample (C5) from the Baitoushan (Changbaishan) volcano, which is located on the border between China and North Korea, to test the utility of tree-ring dating based on our master chronology. Tree rings from the C5 sample have been dated using the ¹⁴C spike at 774–775 CE (Hakozaki et al 2018a). The outermost ring of the sample with exterior bark was dated to 946 CE, which corresponds to the date of the “Millennium Eruption” (Oppenheimer et al 2017; Hakozaki et al 2018a). As the last 14 rings (933–946 CE) of the sample have been carbonized, the remaining inner part consisting of 298 rings (635–932 CE) was used for oxygen isotope analysis.

2.2. Tree-Ring Oxygen Isotope Data

Following the standard protocol for isotope dendrochronology (Kagawa et al 2015), cellulose was isolated directly from 1-mm thick wood plates while preserving the anatomical structures during chemical treatment. Each annual ring in the cellulose plates was manually separated from adjacent rings using a razor blade under a microscope. We loaded the annual ring samples (100–250 μg) into silver foil and then determined their oxygen isotopic compositions (¹⁸O/¹⁶O) using a continuous flow mass spectrometer (Thermo Fisher Scientific Delta V Advantage) coupled to a pyrolysis-type elemental analyzer (Thermo Fisher Scientific High Temperature Conversion Elemental Analyzer). The oxygen isotope ratios are reported as δ¹⁸O (‰), with reference to the international Vienna Standard Mean Ocean Water (VSMOW). To reduce the analytical uncertainty, we used a nearly-zero-blank autosampler sealed against the atmosphere. Some 600 tree-ring samples were continuously measured in a single sequence that spanned 3 days. A laboratory standard material (Merck cellulose) was added every eight samples throughout the measurement procedure. The analytical uncertainty (i.e., 1 standard deviation) associated with the standard material was <0.20‰.

Pattern matching of the interannual δ¹⁸O variations was carried out to assign relative years to every ring of the 37 tree-ring series produced in this study. Correlation analysis was conducted using paired series to identify the relative years showing the highest correlation coefficients. However, the raw tree-ring δ¹⁸O series contain low-frequency variability (Figure 2), which may result in the inflation of correlation coefficients for incorrect dates. To avoid this, we used the high-pass filtered values of the tree-ring δ¹⁸O series for pattern matching. Relatively dated tree-ring δ¹⁸O series were then merged to form a chronology, which was used for pattern matching to identify the relative years of the remaining samples. This stepwise procedure of pattern matching was continued until every ring from all 37 series was relatively cross-dated.

Figure 2. Plots of cross-dated raw tree-ring δ¹⁸O series from the Sarugamori and Aomori sites for the periods (a) 417–800, (b) 801–1200, and (c) 1201–1595 CE. The different colors of the plots designate the different δ¹⁸O series that were analyzed. Enlarged plots, together with sample legends, are presented in Supplementary Figure 2.

A master chronology was developed by merging all 37 tree-ring δ¹⁸O series. The statistical properties of tree-ring δ¹⁸O differ from those of ring-width series. To account for this, Loader et al. (2019) proposed the use of a rectangular filter to standardize the raw tree-ring δ¹⁸O series. Specifically, a rectangular filter is easy to apply and enabled us to: 1) remove long-term trends; 2) minimize the loss of degrees of freedom; and 3) retain a low absolute autocorrelation (Loader et al 2019). Although we scrutinized rectangular filters ranging from 5 to 21 yrs in length, the choice of filter length was insensitive to the pattern matching, as was also found by Sano et al. (2022). To minimize autocorrelation and the loss of degrees of freedom, our master chronology was constructed using an 11-year rectangular filter, which is the same as previously applied to δ¹⁸O data from Japan (Sano et al 2022, 2023). Each of the raw tree-ring δ¹⁸O series was filtered and subtracted to produce anomalies with a mean of zero, and the resulting 37 series were then averaged to build the master chronology. The strength of the common variations in the standardized tree-ring δ¹⁸O series between different samples was evaluated using the mean inter-series correlation (Rbar) and the expressed population signal (EPS; Wigley et al 1984). Rbar is calculated by averaging the correlations derived from all pairs in a given time window, and the EPS indicates how well the chronology estimates a theoretically infinite population. The higher the sample size and/or Rbar, the closer the EPS approaches 1. An EPS value of 0.85 or more is a widely accepted threshold for the development of a robust chronology. We calculated the Rbar and EPS statistics using 50-yr windows, lagged by 25 yrs.

To assign the calendar years, we cross-dated our master chronology from northern Japan with a master chronology from central Japan (Figure 1; Nakatsuka et al 2020; Sano et al 2022). The latter was based on 67 samples consisting of Japanese cypress (mainly Chamaecyparis obtuse) and covers the past 2617 yrs (612 BCE to 2005 CE). Finally, a 298-yr tree-ring δ¹⁸O series from a Korean pine sample was compared with the two master chronologies from northern and central Japan to explore the utility of tree-ring oxygen isotope dating. The pattern-matching results were evaluated using a statistical test of correlation. Specifically, Student’s t-test was applied using correlation coefficients and degrees of freedom that were corrected for autocorrelation and the statistical cost of the filter, so that they conformed to Student’s t distribution (Loader et al 2019). The t values were converted into one-tailed probabilities (expressed conventionally as 1/p) that were corrected for multiple tests of significance using the Bonferroni correction (Dunn 1961). We applied the critical threshold of 1/p ≥ 100, proposed by Loader et al. (2019), to the results of the pattern matching. Our analysis was conducted using dplR (Bunn 2008, 2010) and other packages in the R environment (R Core Team 2020). The statistical data associated with the individual tree-ring δ¹⁸O series are provided in Supplementary Table 1.

2.3. Radiocarbon Measurements

To verify the tree-ring dates obtained via pattern matching of the oxygen isotope data, we conducted annual-resolution tree-ring ¹⁴C measurements for the period 770–780 CE, which covers the 774–775 CE radiocarbon spike event (Miyake et al 2012). Each of the 11 rings covering the period 770–780 CE from one of the Aomori samples (AONT001) was separated from its adjacent rings. Cellulose was extracted from small pieces of each annual ring using acid–alkali–acid and sodium chlorite treatments (Nakamura et al 2013). The graphite extraction and radiocarbon measurements were conducted at the Center for Chronological Research, Nagoya University (Nakamura et al 2007). We then compared our tree-ring ¹⁴C data with measurements from southern Japan (Miyake et al 2012) and from Baitoushan (Hakozaki et al 2018a).

3. Results and Discussion

The cross-dated raw and standardized tree-ring δ¹⁸O series from our 37 samples are shown in Figures 2 and 3, respectively (see Supplementary Figures 2 and 3 for enlarged plots with sample legends). Offsets from the mean δ¹⁸O values are evident among the samples, because our sampled trees were growing over a wide region. Correlation matrices calculated using all pairs of overlapping periods of ≥50 yrs are presented in Supplementary Figures 4 and 5 for the raw and standardized series, respectively. Even though the mean δ¹⁸O values vary among the samples, all pairs of raw series show a statistically significant correlation (p < 0.05), with a mean inter-series correlation coefficient of 0.58. We therefore consider the relative variations in annual δ¹⁸O to be well matched among the samples. Removing low-frequency variations using the 11-yr rectangular filter from the raw tree-ring δ¹⁸O series resulted in enhanced common variations among the samples (all pairs yielding p < 0.01), with a mean inter-series correlation coefficient of 0.63. Our cross-dating was further tested using the leave-one-out principle. Specifically, standardized tree-ring δ¹⁸O series from one sample were correlated against a master chronology that was developed using the remaining samples. This showed that all 50-yr segments from any one sample are significant at p < 0.01 (Supplementary Figure 6).

Figure 3. As for Figure 2, but for standardized tree-ring δ¹⁸O series. Standardization was conducted using an 11-yr rectangular filter to extract the high-frequency variability component. Enlarged plots, together with sample legends, are presented in Supplementary Figure 3.

Overall, variations in tree-ring δ¹⁸O are closely correlated among the different samples, indicating the pattern matching is robust. The 1179-yr-long master chronology is presented in Figure 4, together with the number of samples and the Rbar and EPS values calculated using 50-yr windows. The Rbar values exceed 0.51 throughout the 1179 yrs, which statistically supports the reliability of our pattern matching. However, the EPS values (0.78–0.95) often do not attain the generally accepted threshold value of ≥0.85, and this is largely because of the limited number of samples. Accordingly, a robust estimate of mean tree-ring δ¹⁸O is not achieved over the entire period of the chronology. On the other hand, as noted above, the pattern matching of our samples worked well; therefore, we consider the dating results to be reliable. In addition, our tree-ring dating based on this chronology is consistent with previously reported dendrochronological dates that are independent from our samples, as discussed below.

Figure 4. (a) The 1179-yr master chronology derived by averaging all standardized tree-ring δ¹⁸O series. An enlarged version of this plot is presented in Supplementary Figure 7. (b) Number of samples used for the master chronology. (c) Rbar and EPS statistics calculated for 50 yrs and lagged by 25 yrs.

The results of the pattern matching of our chronology against another chronology from central Japan (Nakatsuka et al 2020; Sano et al 2022) are presented in Figure 5. A high match in the t value of 9.0 (r = 0.26, 1/p > 10⁶) was observed when the last year of our chronology was placed at 1595 CE, which corresponds to the period 417–1595 CE. Remarkably high matches were also found using the early (417–1006 CE, t = 6.8, r = 0.28, 1/p > 10⁶) and late (1007–1595 CE, t = 5.8, r = 0.24, 1/p > 8 × 10⁴) half-segments of our chronology. All of the pattern matching passes the critical threshold of 1/p ≥ 100, indicating that our tree-ring δ¹⁸O patterns were successfully cross-matched against the chronology from central Japan.

Figure 5. Distribution of Student’s t values through time for all possible dates of our tree-ring δ¹⁸O chronology, using the periods (a) 417–1595, (b) 417–1006, and (c) 1007–1595 CE, against the master chronology from central Japan.

Our tree-ring chronologies from northern and central Japan for the common period 417–1597 CE are presented in Figure 6, together with sample sizes and running correlations. The running 51-yr correlations between the two chronologies show that the correlation tests alternated cyclically (period of 50–150 yrs) between significant and non-significant. As presented in Figure 1, the contrasting spatial distribution of hydroclimatic variability is seen with first EOF loadings based on June–August precipitation data for the period 1891–2019 CE. For the instrumental period, northern Japan (including our study region) is considered part of the Asian continental domain, whereas central Japan is part of the Japanese archipelago domain. Both master chronologies originate near the boundary between these two domains, and this boundary is likely to have shifted in the past in response to large-scale climatic conditions. This shifting of the boundary position seems to be at least partially responsible for the cyclical trends in periods showing significant or non-significant correlations between the two chronologies. We further scrutinized how the spatial correlations changed temporally during the instrumental period by analyzing the observed climate data from the Aomori weather station, which is located near our study sites, and also the ERA5 dataset (Hersbach et al 2020). Specifically, spatial correlations of precipitation and relative humidity for the June–August period from the ERA5 dataset were calculated against the corresponding Aomori station data for three periods (i.e., 1940–1967, 1968–1995, and 1996–2022 CE). Precipitation in Aomori was not significantly correlated with that in central Japan, but was correlated with that within a certain latitude range (ca. 37°–45°N) for the three periods (Figure 7). Aomori precipitation shows a positive correlation with central Japan precipitation for the period 1968–1995 CE, although not statistically significant. Interestingly, this correlation was negative for the other two periods (1940–1967 and 1996–2022 CE). The shift in boundary position, as proposed above, is partially reproduced in terms of the spatial correlations of precipitation. On the other hand, the correlations of relative humidity between Aomori and central Japan are much more temporally stable when compared with those for precipitation. Specifically, relative humidity in Aomori is significantly correlated with that in central Japan for the 1940–1967 and 1968–1995 CE periods. A temporal change in the spatial patterns of precipitation seems to be one of the main factors contributing to the unstable correlations between the two chronologies from northern and central Japan.

Figure 6. (a) Master chronologies for northern (this study) and central (Nakatsuka et al 2020; Sano et al 2022) Japan. An enlarged version of this plot is presented in Supplementary Figure 8. (b) Number of samples used for each master chronology. (c) Running 51-yr correlations between the two master chronologies.

Figure 7. Spatial correlations between (a, b, c) precipitation and (d, e, f) relative humidity from the Aomori station (cyan circle) for the June–August period, and also data from the ERA5 dataset, for the periods (a, d) 1940–1967, (b, e) 1968–1995, and (c, f) 1996–2022 CE. Green lines enclose areas with significant correlations (p < 0.05).

As shown in Figure 8, our annual-resolution radiocarbon record from northern Japan is closely correlated with that from southern Japan (Miyake et al 2012) and Baitoushan (Hakozaki et al 2018a). The sharp increase in ¹⁴C content observed globally at 774–775 CE (Büntgen et al 2018) is clearly reproduced in our measurements. Therefore, our tree-ring dates, derived from the pattern matching of oxygen isotope data, are independently verified by this ¹⁴C spike matching.

Figure 8. Comparison of annual-resolution tree-ring Δ¹⁴C time-series between our samples and those from southern Japan (Miyake et al 2012) and Baitoushan (Hakozaki et al 2018a).

Our results of dating the Baitoushan sample against the master chronology from northern Japan are presented in Figure 9. The last year of the C5 sample was dated to 932 CE, which is consistent with the dating result derived independently from the ¹⁴C spike matching of the sample at 774–775 CE (Hakozaki et al 2018a). More specifically, the 298-yr series from Baitoushan is cross-dated against our master chronology from northern Japan for the period 635–932 CE, as clearly seen with a prominent peak in Student’s t value of 6.45 (r = 0.37; Figure 9). The probability (1/p) of this pattern matching exceeds 10⁶, which passes the critical threshold of 1/p ≥ 100. However, patten matching of this sample against the master chronology from central Japan failed, with no prominent t value observed over the past 2617 yrs. The t value was 0.97 at the correct date (932 CE), indicating that the master chronology from central Japan was unsuitable for dating sample C5 from Baitoushan. Figure 10 shows the running 31-yr correlations for the Baitoushan δ¹⁸O series against the master chronologies from northern and central Japan. As the two chronologies are correlated (Figure 6), variations in their r values through time are similar (Figure 10b). Significant correlations are observed with the chronology from northern Japan, with non-significant correlations appearing in some periods, such as around 700, 750, and 880 CE. In contrast, the chronology from central Japan is less well correlated, with significant correlations appearing only in the period around 850 CE. These findings indicate that the northern Japan chronology exhibits certain advantages over the central Japan chronology for dating samples originating at higher latitudes on the Asian continent. Again, these contrasting results from tree-ring dating of the Baitoushan sample are related to the large-scale climatic controls on tree-ring δ¹⁸O levels. Our results highlight the importance of developing a spatially distributed tree-ring δ¹⁸O network over Japan, in particular for tree-ring data from different latitudes.

Figure 9. Distribution of Student’s t values through time for all possible dates of the Baitoushan tree-ring δ¹⁸O series against the master chronology from northern Japan.

Figure 10. (a) Cross-dated tree-ring δ¹⁸O series from the Baitoushan site and the corresponding segment of the master chronologies from northern (this study) and central (Nakatsuka et al 2020; Sano et al 2022) Japan. (b) Running 31-yr correlations of the Baitoushan δ¹⁸O series against the master chronologies from northern and central Japan. Enlarged versions of these plots are presented in Supplementary Figure 9.

Conclusions

We have developed a 1179-yr-long tree-ring δ¹⁸O chronology for northern Japan using 37 samples collected from dead trunks and unearthed wood from Hiba arbor-vitae. Our chronology was successfully cross-dated against another master chronology from central Japan. Our tree-ring dates were independently verified by reproducing the radiocarbon spike event at 774–775 CE. One sample from the Baitoushan site, which is located on the border between China and North Korea, was successfully cross-dated using our master chronology, clearly indicating the utility of tree-ring dating between samples from different countries. Continued efforts to develop tree-ring δ¹⁸O chronologies in and around Japan will contribute significantly to future tree-ring dating of wood samples.