4.1. Signal Strength of Tree Ring δ18O Data
 The means (and standard deviations) of δ18O values for the 6 trees, calculated for the common period 1905–2004, range from 22.4‰ (0.86‰) to 23.6‰ (1.08‰). The mean δ18O values are 1.0‰–1.3‰ less than those obtained for 4 trees of Fokienia hodginsii from northern Laos [Xu et al., 2011]. Because of the relatively higher latitude of our sampling site (see Figure 1), δ18O values of precipitation (source water) are expected to be more depleted at our site than at the Laos site, reflecting the ‘latitude effect’ of precipitation δ18O [Dansgaard, 1964]. The standard deviations of the δ18O values for the 6 trees for each year, calculated after adjusting the mean δ18O values to 0, range from 0.15‰ to 1.32‰ (mean, 0.56‰), which is comparable to values of 0.08‰ to 1.42‰ (mean, 0.65‰) obtained from the 4 trees from northern Laos [Xu et al., 2011].
 Our tree ring δ18O data from 6 trees show significant inter-series correlations over the past 300 years, as determined by running Rbar statistics (0.59–0.80) calculated using 50-year windows and a lag-time of 25 years (Figure 3). We found relatively low Rbar values during the last 100 years, and the tree ring δ18O data show overall increasing trends during this time. Nevertheless, even the lowest Rbar value of 0.59 is highly significant, suggesting that climatic signals are well recorded in the δ18O data. In contrast, ring width-index series derived from the same 6 trees show relatively weak inter-series correlations (0.10–0.36) [Sano et al., 2009], which is attributed to the effects of endogenous disturbances on radial growth, such as competition with neighboring trees in a closed-canopy forest site [Fritts, 1976]; δ18O values, on the other hand, are less influenced by such disturbances. The EPS for the δ18O data attains the generally accepted threshold value of 0.85 or greater over the entire period of the last 300 years (Figure 3), although the sample size is smaller in the earlier portions of the chronology. Therefore, the chronology can be considered to provide robust estimates of mean tree ring δ18O in the region.
Figure 3. Tree ring δ18O data with sample size, and the running expressed population signal (EPS) and mean inter-series correlation (Rbar) calculated using 50-year windows and a lag-time of 25 years.
Download figure to PowerPoint
 It should be noted that age-related long-term trends in tree ringδ18O have been reported [Esper et al., 2010; Nakatsuka, 2007; Treydte et al., 2006], although the physiological mechanisms underlying these trends have yet to be fully analyzed. For example, Treydte et al. identified an age-dependent decrease in theδ18O chronology of juniper by analyzing biologically younger and older tree rings for the same time period; however, the age dependency had little effect on their millennium-long climate reconstruction. Other studies have also noted a decrease inδ18O series with increasing tree age, a trend which is observed in different species and at different locations. In contrast, our δ18O chronology shows no significant changes in δ18O during the period 1705–1804 (a possible slight increase of 0.021‰ per decade). In this study, the 3 tree cores selected for analyses of the period 1705–1804 have rings dating to 1560, 1596 and 1624, and none of these rings contain pith; therefore, these trees were at least 100 years old in 1705. Although we did not measure younger and older tree rings for the same time period, and therefore cannot directly evaluate the effect of tree age on δ18O, our findings suggest that the age effect does not significantly influence our δ18O chronology. Young et al.  have recently reported that age effects are not observed in tree rings whose cambial ages exceed ca. 50 years. The results of Young et al.  indicate that the aging effect does depend on species and location.
4.2. Climate Signals in Tree Ring δ18O and the PDSI Reconstruction
 The δ18O chronology shows significant positive correlations with temperatures from May to August, and negative correlations with precipitation in May and August (Figure 4). Overall, the May–October monsoon season optimizes the temperature (r = 0.64, p < 0.001, effective degrees of freedom (edf) = 54) and precipitation signals (r = −0.59, p < 0.001, edf = 55) in the chronology. As recorded in the meteorological data from the monsoon season, higher temperatures usually correlate with lower precipitation and relative humidity levels, both of which lead to enriched δ18O in tree rings. More specifically, an inverse relationship between the δ18O of precipitation and the amount of precipitation is observed at lower latitudes, which is referred to as the ‘amount effect’ [Araguás-Araguás et al., 1998; Dansgaard, 1964]. Lower levels of precipitation correlate with higher δ18O values of the precipitation, resulting in δ18O enrichment of source water taken up by trees. On the other hand, lower humidity causes higher vapor pressure gradients between leaf interstitial spaces and the atmosphere, resulting in preferential loss of lighter isotopes and consequent enrichment of δ18O in leaf water [Roden et al., 2000]. In addition, higher temperatures stimulate evaporation of soil water, leading to soil-waterδ18O enrichment, thus affecting the composition of source water taken up by trees. Overall, our results are consistent with the model of tree ring cellulose δ18O, as explained in section 2.
Figure 4. Correlations between tree ring δ18O and monthly (prior-year November to current-year October) and seasonal (May–October) temperature, precipitation, and PDSI for the period 1948–2004. To account for autocorrelation, the correlation coefficients were tested for significance using the number of effective degrees of freedom.
Download figure to PowerPoint
 The strength of the tree ring δ18O response to climate is consistent across a variety of climate parameters; moreover, as climate parameters are significantly correlated with one another, it is difficult to identify a dominant factor. Generally, lower precipitation levels correlate with higher δ18O values of the precipitation (amount effect); at the same time, higher temperatures (usually associated with lower precipitation levels) also lead to δ18O enrichment of soil water; acting together, these factors cause δ18O enrichment of source water. In addition, lower precipitation and higher temperature usually result in lower relative humidity levels, leading to increased leaf water δ18O. Importantly, these associated climatic parameters contribute to tree ring δ18O enrichment or depletion in a similar way (direction), indicating that tree ring δ18O values can be used as a pure proxy of past hydroclimate (wet–dry conditions); this interpretation is supported by relatively high correlations between δ18O and the monthly PDSI during the period May–October, with the highest correlation occurring in the May–October season (r = −0.66, p < 0.001, edf = 50). The PDSI is a dry–wet metric based upon a water balance model that considers precipitation, temperature, and soil characteristics [Palmer, 1965]. Although the mechanisms underlying variations in tree ring δ18O can generally be attributed to δ18O of source water and evaporative enrichment of leaf water δ18O, in pluvial monsoon regions these effects seem to be well represented by the PDSI, a simplified climatic variable. Therefore, PDSI values during the monsoon season are considered to be a strong predictor of the variance in tree ring δ18O values.
 The correspondence of our δ18O chronology with climatic factors is consistent with the findings of a 52-year tree ringδ18O chronology of F. hodginsii from Phu Leuy (PL) in northern Laos [Xu et al., 2011]. Furthermore, a direct comparison between the Vietnam and Laos δ18O chronologies is highly significant (r = 0.77, p < 0.001, edf = 48) (Figure 5), even though these sampling sites are 150 km distant from each other, indicating that regional hydroclimatic signals are well recorded in our δ18O chronology.
 We scrutinized the fidelity of a linear regression model used to reconstruct May–October PDSI, by conducting split-period calibration-verification tests commonly used in dendroclimatology [Cook and Kairiukstis, 1990]. More specifically, the PDSI reconstructed by regression upon one half of the climate data (calibration period) was tested against the actual PDSI of the remaining half of the data (verification period) withheld from regression. This test was repeated by switching the calibration and verification periods. As shown in Table 1, the reduction of error (RE) and the coefficient of efficiency (CE) tests are both positive, indicating the validity of the regression model. Thus, we were able to reconstruct summer monsoon PDSI back to 1705, based on a regression model that accounts for 44% of the PDSI variance (Figures 6a and 6c).
Table 1. Calibration and Verification Statistics for the Period of 1948–2004
|Calibration Period||R2||Verification Period||r||RE||CE|
|Early half (1948–1975)||0.252a||Late half (1976–2004)||0.687a||0.557b||0.327b|
|Late half (1976–2004)||0.471a||Early half (1948–1975)||0.502a||0.502b||0.151b|
Figure 6. (a) Actual and reconstructed PDSI values and linear regression trends. (b) Same as Figure 6a but for the Multivariate ENSO Index (MEI), including running 21-year correlations between the data sets. (c) Reconstructions of the May–October PDSI (left axis) and the November–October MEI (right axis) for the period 1705–2004. The red and blue lines represent 30-year splined values and the mean values of the reconstructions, respectively. Note that the right axis for the MEI is inverted because of the inverse PDSI-MEI correlation.
Download figure to PowerPoint
 The reconstructed PDSI values reveal that the 18th century was generally wet, followed by a dry period in 1820–1870 and a wet period in 1870–1900 (Figure 6c). The dry epoch in the early to mid 19th century is also suggested by pre-monsoon (March–May) PDSI reconstruction based on ring width records originating from the same site as that of this study [Sano et al., 2009]. For other periods, on the other hand, the δ18O reconstruction agrees less well with the ring width record. Perhaps the most notable difference between the δ18O and ring width records is the absence of evidence for a multidecadal drought in the mid-eighteenth century in theδ18O-based reconstruction. This is not surprising because of differences in the reconstructed seasons. Recently,Kajikawa et al. showed a seasonal dependence of various features of the Asian monsoon over the last 30 years, such as trends toward earlier onset of the monsoon and increase in May precipitation (but no long-term changes in July and August precipitation). These findings point out that PDSI reconstructions for different seasons usingδ18O and ring width data are both important for analyses of seasonal changes in Asian monsoon patterns over long time periods.
 Another noteworthy feature of our δ18O-based reconstruction is a trend in decreasing PDSI during the latter half of the 20th century, which parallels similar trends in meteorologically based PDSI (Figure 6a). Singhrattna et al.  also observed a weakening of monsoon precipitation in central Thailand over the last 50 years. In addition, tree ring δ18O from the Himalaya [Sano et al., 2012] and Tibet [Grießinger et al., 2011], and varve-thickness data from lake sediments in Tibet [Chu et al., 2011], also show a decrease in monsoon rainfall over the past two centuries, indicating that the summer monsoon has weakened across wide areas of South and Southeast Asia. A modeling study reveals that a weakening trend of monsoon precipitation found in northern India and the eastern Tibetan Plateau over the latter half of the 20th century is deducible from the atmosphere's response to an increase in observed SSTs over the tropical Pacific and Indian Ocean [Zhou et al., 2008]. These findings suggest that elevated SSTs may also be responsible for the recent decrease observed in our reconstructed PDSI values.
 We further examined the reconstructed PDSI for the period of 1705–2004 in terms of its spectral power, using the multitaper method [Mann and Lees, 1996]. Significant peaks that fall within the range of ENSO variability occur at 3.8 and 4.2–4.4 years (see Figure 7). On the other hand, multidecadal cycles, which are prominent in ring width records from the Indochina Peninsula [Buckley et al., 2007, 2010; Sano et al., 2009], lack significant power in our data set. Overall, a persistent multicentury trend is the most notable feature of our δ18O chronology; this trend may be related to an increase in tropical SSTs, as a warming trend over the last 200 years has been observed in a reconstruction of large-scale tropical SST patterns [Wilson et al., 2006].
Figure 7. Multitaper method power spectra for the reconstructed PDSI (AD 1705–2004). Peaks above the red line are significant at the p < 0.01 level.
Download figure to PowerPoint
4.3. Links With the El Niño-Southern Oscillation
 Our δ18O chronology shows a significant positive correlation with annual SSTs over most of the tropical Pacific and Indian Ocean, in particular the central Pacific (Figure 8), indicating that El Niño and La Niña phases result in dry and wet conditions, respectively, in the study region. The δ18O chronology is also significantly correlated with annual (previous November to current October) ENSO-related indices during the 1871–2004 period, especially annual NINO3.4 SST (r = 0.67, p < 0.001, edf = 114), the SOI (r = −0.60, p < 0.001, edf = 117), and the MEI (r = 0.70, p < 0.001, edf = 110). The long-term linear trends of increasing tree ringδ18O and decreasing reconstructed PDSI since the late 19th century is also consistent with that of increasing NINO3.4 SST and MEI, and of decreasing SOI. Furthermore, the influence of tropical Pacific climates on tree ring δ18O is temporally stable, as revealed by running 21-year correlations with MEI data (significance level,p < 0.01) (Figure 6b). Surprisingly, the explained variance for the MEI (49%) is higher than that for the local PDSI (44%). We attribute this, at least in part, to the necessity of using a PDSI data set based on records mostly from Thailand. Overall, our results indicate the potential of tree ring δ18O for reconstructing ENSO. For further investigation, we established a linear regression model to estimate the annual MEI over the last 300 years (Figures 6b and 6c), which shows variations that are virtually identical to those of the PDSI reconstruction.
Figure 8. Spatial correlations between tree ring δ18O and annual (prior-year November to current-year October) sea surface temperatures (HadISST1 data set) for the period 1871–2004. All time series were individually detrended using linear regressions to remove long-term (common) trends in the series before calculating correlation coefficients. Correlations of >|0.23| are significant at thep < 0.01 level, assuming 120 effective degrees of freedom.
Download figure to PowerPoint
 Over the last decade, considerable effort has been devoted to reconstructions of ENSO variability using tree rings and other proxy records, based mainly on data from North America and/or the tropical Pacific [e.g., Braganza et al., 2009; Cook et al., 2008; D'Arrigo et al., 2005; Li et al., 2011; Mann et al., 2000; Wilson et al., 2010]. While existing ENSO reconstructions are in good agreement for the period since 1871 (the time at which instrumental records commenced), there is relatively limited agreement about ENSO variability prior to this time [Wilson et al., 2010]. As presented in Table 2, our reconstructed MEIs also show markedly weakened associations with published ENSO reconstructions before the instrumental period. Among the ENSO reconstructions, the ‘center of action’ (COA) reconstruction [Wilson et al., 2010] is perhaps the most reliable, as its calibration model accounts for over 80% of the instrumental variance. The importance of the COA reconstruction is further emphasized by the fact that all of the original proxy (coral) records are located in the central and eastern Pacific (the core region where ENSO develops), while other reconstructions rely largely on proxy records located in regions that are teleconnected with the central Pacific. Unfortunately, the COA reconstruction is not considered robust prior to 1850, because of lack of replication [Wilson et al., 2010]; this lack of sample replication may be responsible for reduced inter-series correlations before the mid-19th century.
Table 2. Correlation Matrix Calculated for 7 ENSO Reconstructions From the 1871–1980 and 1705–1870 Periodsa
|Source Proxy Location||Reference||This Study||Wilson COA||Wilson TEL||Cook||Mann||Li|
|Central/Eastern Pacific||Wilsonb (COA)||0.57|| || || || || |
|Indo-Pacific Basina||Wilsonb (TEL)||0.60||0.80|| || || || |
|Southwest N America||Cookc||0.46||0.60||0.48|| || || |
|Near Global||Mannd||0.49||0.64||0.53||0.82|| || |
|North America||Lie||0.44||0.34||0.28||0.69||0.67|| |
|Pacific Basin||Braganzaf (R5)||0.52||0.55||0.54||0.71||0.72||0.60|
|Central/Eastern Pacific||Wilsonb (COA)||0.09 ns|| || || || || |
|Indo-Pacific Basina||Wilsonb (TEL)||0.13 ns||0.03 ns|| || || || |
|Southwest N America||Cookc||0.22||0.00 ns||0.16 ns|| || || |
|Near Global||Mannd||0.25||0.15 ns||0.14 ns||0.48|| || |
|North America||Lie||0.12 ns||−0.06 ns||0.06 ns||0.61||0.39|| |
|Pacific Basin||Braganzaf (R5)||0.26||−0.02 ns||0.22||0.66||0.59||0.48|
 It should be noted that the ENSO reconstructions of Mann et al. , Cook et al. , Braganza et al. , and Li et al. are in moderate agreement prior to 1871. This is due partly to the shared tree ring data from southwest North America for these reconstructions. The cross correlations, however, do indicate some degree of coherence among these reconstructions. On the other hand, the ‘teleconnection’ (TEL) reconstruction, based on 1 ice core and 11 coral records from the Indo-Pacific basin (but out of the region used for the COA reconstruction) [Wilson et al., 2010], is less well correlated with those reconstructions [Braganza et al., 2009; Cook et al., 2008; Li et al., 2011; Mann et al., 2000] for the period 1705–1870, although the TEL record is considered as robust since 1727. These results, taken together with the fact that the TEL series is entirely independent of the reconstructions produced by Cook et al.  and Li et al. , indicate that the teleconnected relationship between the tropical central/eastern Pacific and the regions where proxy records are located may not be temporally stable [see Wilson et al., 2010].
 The significant time-stable correlation between tree ringδ18O and ENSO records over the 135-year period of available instrumental data is the most striking feature of our analysis. Our results imply that northern Indochina may be a ‘hot spot’ firmly teleconnected with the tropical Pacific, although whether the teleconnected relationship was stable before the instrumental period remains unclear. In this context, the reason that the hydroclimate of northern Indochina is closely linked to ENSO has yet to be resolved by numerical modeling experiments. Recently,Liu et al.  revealed a significant correlation between a tree ring δ18O chronology from southwestern China and the SOI for the periods 1902–1930 and 1980–2004. However, no significant correlation was observed during the period 1930–1980, indicating a complex association between the ENSO and regional climate. Evidence of the non-stationarity between ENSO and the local climate (and thus tree ringδ18O values) indicates the importance of investigating long-term changes in atmospheric circulation patterns based on widely spaced proxy records in monsoon Asia. Nevertheless, our overall results indicate the significant potential for generating independent and robust reconstructions of ENSO variability using tree ringδ18O from mainland Southeast Asia, and for investigating the geographical influence of ENSO before the instrumental period.