Spatial and temporal variability in the stable isotope systematics of modern precipitation in China: Implications for paleoclimate reconstructions

: The stable isotopic composition of materials such as glacial ice, tree rings, lake sediments, and speleothems from low-to-mid latitudes contains information about past changes in temperature (T) and precipitation amount (P). However, the transfer functions which link δ 18 O p to changes in T or P, dδ 18 O p /d T and dδ 18 O p /d P , can exhibit significant temporal and spatial variability in these regions. In areas affected by the Southeast Asian monsoon, past variations in δ 18 O and δD of precipitation have been attributed to variations in monsoon intensity, storm Abstract The stable isotopic composition of materials such as glacial ice, tree rings, lake sediments, and speleothems from low-to-mid latitudes contains information about past changes in temperature ( T ) and precipitation amount ( P ). However, the transfer functions which link N 18 O p to changes in T or P , d N 18 O p /d T and d N 18 O p /d P , can exhibit significant temporal and spatial variability in these regions. In areas affected by the Southeast Asian monsoon, past variations in N 18 O and N D of precipitation have been attributed to variations in monsoon intensity, storm tracks, and/ or variations in temperature. Proper interpretation of past N 18 O p variations here requires an understanding of these complicated stable isotope systematics. Since temperature and precipitation are positively correlated in China and have opposite effects on N 18 O p , it is necessary to determine which of these effects is dominant for a specific region in order to perform even qualitative paleoclimate reconstructions. Here, we evaluate the value of the transfer functions in modern precipitation to more accurately interpret the paleorecord. The strength of these transfer functions in China is investigated using multiple regression analysis of data from 10 sites within the Global Network for Isotopes in Precipitation (GNIP). N 18 O p is modeled as a function of both temperature and precipitation. The magnitude and signs of the transfer functions at any given site are closely related to the degree of summer monsoon influence. N 18 O p values at sites with intense summer monsoon precipitation are more dependent on the amount of precipitation than on temperature, and therefore exhibit more negative values in the summer. In contrast, N 18 O p values at sites that are unaffected by summer monsoon precipitation exhibit strong relationships between N 18 O p and temperature. The sites that are near the northern limit of the summer monsoon exhibit dependence on both temperature and amount of precipitation. Comparison with simple linear models ( N 18 O p as a function of T


Introduction
Since 1964, when Dansgaard [1] ¢rst reported that the N 18 O of precipitation (N 18 O p ) at a given locale re£ects environmental factors such as surface temperature (T), amount of precipitation (P), elevation, and source composition, numerous studies have attempted to infer paleoclimate information from the isotopic composition (N 18 O and ND) of paleoprecipitation preserved in natural archive materials, such as glacial ice, tree rings, lake sediments, freshwater mollusks, pedogenic carbonates, £uid inclusions in speleothems, and speleothem calcite. The particularly strong correlation between N 18 O p and temperature (V0.55x/ ‡C) at mid-to high-latitude sites has led to the common utilization of N 18 O p as a proxy for continental paleotemperature. However, this correlation is much weaker or non-existent in low latitudes, where di¡erent mechanisms of moisture transport predominate [2]. The goal of this study is to investigate the systematics of stable isotopes in precipitation in the Asian monsoon region of China, to improve interpretations of isotopic paleoclimate records obtained from archives in this region. Oxygen isotopes in precipitation at low-to midlatitudes may be signi¢cantly a¡ected by the amount of precipitation in addition to temperature, leading to di⁄culties in interpretation of oxygen and hydrogen isotopic data from these regions. A negative relationship between N 18 O p and amount of precipitation has been observed at lower latitudes (known as the 'amount e¡ect'). This e¡ect is important in convectively active regions where air mass trajectories have a large vertical component, leading to progressive depletion of 18 O and D in precipitation at a given site. The amount e¡ect is analogous to the continental effect, in which horizontally transported air masses are progressively depleted in the heavy isotopes with distance from the source region. This e¡ect can be ampli¢ed by the fact that isotopic exchange with vapor and evaporative enrichment of raindrops are both greatly reduced with heavy rains.
Ho¡mann and Heimann [3] determined that the N 18 O p^P relationship in the Asian monsoon region, while weaker than that in tropical islands, is still a signi¢cant factor in£uencing N 18 O p in China. The stable isotope systematics in this region are highly complex, however, due to the combined in£uence of temperature, precipitation, and circulation changes. Both temperature and precipitation are highest in the summer in the Asian monsoon region; therefore the temperature e¡ect (positive dN 18 O p /dT) and the amount e¡ect (negative dN 18 O p /dP) partially cancel each other out. In order to use N 18 O p or ND in precipitation as a paleoproxy, it is necessary to determine the relative importance of each e¡ect at a given site. In other words, we must investigate whether N 18 O p is highest during the summer (T e¡ect dominant) or lowest during the summer (P e¡ect dominant). Previous paleoclimate studies from Southern and Central China have interpreted N 18 O p records in speleothems [4] or ND records in tree rings [5] as proxies of paleomonsoon intensity or paleoprecipitation, while stable isotope records from more northern or western sites are interpreted in terms of paleotemperature [6^8].
The use of N 18 O p or ND as a quantitative paleothermometer has been called into question in recent years, even at mid to high latitudes, where it was previously thought to be fairly straightforward. Recent studies have suggested that the transfer function, dN 18 O p /dT, may exhibit signi¢cant temporal and spatial variability. Cu¡ey et al. [9,10], for example, used borehole temperature pro¢les from ice cores in Greenland as an independent paleothermometer to demonstrate that the temporal N 18 O p /T s relationship may be as much as a factor of two weaker than the modern spatial relationship commonly used in paleotem-perature reconstructions. In addition, Fricke and O'Neil [11] have shown that the slope of the N 18 O p^T relationship is somewhat dependent on climate mode. Thus, the use of a single relation for all time periods, encompassing di¡erent climate regimes, can lead to erroneous paleotemperature estimates. Hendricks et al. [12] found that between 45 ‡N and 45 ‡S, the N 18 O p^T relationship is signi¢cantly weaker than at higher latitudes, due to the greater degree of water vapor recycling in this region.
In the monsoon regions, where circulation patterns are particularly complicated, extreme caution must be used in interpreting past variations in N 18 O p . The East Asian monsoon system leads to seasonal changes in air pressure, wind, precipitation, and temperature gradients over East and Southeast Asia (Fig. 1). During the winter, cold air over the continent and warm air over the South China Sea cause a pressure gradient, producing dry northeasterly winds that £ow over Southern Asia. This pressure gradient is reversed during the summer, as the air over the continent becomes warmer than the air over the South China Sea. Therefore, during the summer monsoon, moisture-bearing winds sweep over the continent. The presence of the Qinghai-Tibetan Plateau in-tensi¢es the uplift and cooling of these warm air masses, leading to increased precipitation over the continent.
Over long timescales, changes in solar insolation, albedo, glacial ice volume, continental snow cover, sea surface temperatures, and the Southern Oscillation all contribute to monsoon variability [13]. In order to understand the complex interactions between these forcing mechanisms, it is necessary to obtain and accurately interpret high-resolution proxy records of monsoon history from multiple sources. To date, continental paleoclimate records from the region include ice core records [6], loess records [14], speleothem records [4], lake records [15], and tree ring records [5]. These proxy records suggest that the intensity of the Asian monsoon has undergone dramatic changes throughout the Pleistocene and Holocene, with a much drier climate during glacial times due to a decrease in summer monsoon intensity, and an increase in the intensity of the dry winter monsoon [16]. Higher-frequency oscillations in monsoon intensity have also been observed during the late Pleistocene and Holocene through both instrumental and proxy records [17], and have been correlated with Heinrich and Dansgaard^Oeschger events in the North Atlantic region [18]. Interpretation of many of these proxy records relies on our understanding of the stable isotope systematics of precipitation in China.
In previous studies of areas a¡ected by the Southeast Asian monsoon, past variations in N 18 O and ND of precipitation have been attributed to variations in monsoon intensity, storm tracks, and/or variations in temperature. In such studies, the modern relationship between T, P, N 18 O, and ND has been used to infer paleotemperatures [6] and paleoprecipitation [5]. In this paper, we reassess the validity of using continental N 18 O p and ND records as paleotemperature or paleoprecipitation records in monsoon regions, by determining the approximate values of dN 18 O p /dT and dN 18 O p /dP in modern precipitation near this region. To separate the temperature e¡ect from the amount e¡ect on N 18 O p , we performed a number of multiple regression analyses on isotopic and meteorological data from the Global Network for Isotopes in Precipitation (GNIP). Although we focus this investigation on precipitation in China, the results may have implications for the entire Asian monsoon region, as well as tropical and equatorial regions.
Accurate interpretation of paleomonsoon activity has important societal implications. Long-term records of monsoon activity will aid in the prediction of anomalous monsoons. The economic, environmental, and societal impacts of anomalous monsoon activity can be severe and widespread. The high population density in the agriculturally based societies found in many monsoon regions, such as Southeast Asia, has created a strong dependence on monsoon rainfall for producing adequate crop yields. Even small variations in the timing and intensity of the summer monsoon can have signi¢cant consequences, such as low crop yields during weak summer monsoons, or devastating £oods during exceptionally strong summer monsoons [19].

Data and methods
To assess the suitability of stable isotope records of precipitation from China for use in paleoclimate reconstructions, we utilized data from the International Atomic Energy Agency-World Meteorological Organization GNIP database [20]. This database includes mean monthly N 18 O p , ND, precipitation amount, surface temperature, and vapor pressure values for 29 sites in China, although many of these records are discontinuous. To prevent inter-annual variability in stable isotope systematics from biasing our results, we chose to focus our study on 10 (Fig. 2).
Assuming that at any given site, surface temperature and amount of precipitation are the main environmental variables correlated with N 18 O p , these data are ideally suited for multiple regression analysis. Multiple regression analysis is a convenient method for viewing stable isotopes in precipitation in three dimensions, treating N 18 O p , for example, as a function of two variables : surface temperature and precipitation amount. This is analogous to linear regressions, such as those that are commonly performed between N 18 O p and T at higher latitudes, but through the use of multiple regression, we separate out the confounding e¡ects of precipitation that may obscure the underlying N^T relationship at low latitudes and in the Asian Monsoon region. Our hypothesis is that N 18 O p (or ND) at any given site, in regions where the amount e¡ect is important, can be described as: where L 0 is the y-intercept, L t and L P are the partial regression coe⁄cients for temperature and precipitation respectively, and O is the residual scatter that is not described by T or P. The residual scatter, O, represents the sum of all in£uences, other than T and P, on the stable isotopic composition of precipitation. In China, these other e¡ects may be signi¢cant and are related to such things as orography, circulation changes, source of precipitation, and relative humidity. In order to most accurately model N 18 O p in the Asian monsoon region, these additional e¡ects should be included. This is best done using General Circulation Models (GCM). However, since P and T are more strongly correlated with N 18 O p than these other factors, we believe that large-scale patterns can be elucidated using much simpler multiple regression models, which are easier and cheaper to run than GCMs. We therefore model N 18 O p as a function of only T and P as: We performed several checks to ensure that multiple regression is the appropriate analysis to use for our data. First, we investigated whether our dependent and independent variables, N 18 O p , and T and P, respectively, exhibit a Gaussian distribution. This was accomplished by plotting the distributions of N 18 O p (x), mean monthly T ( ‡C), and P (mm/month) in order to check for any evidence of non-normality (Fig. 3). The distributions for all N 18 O p and T are similar in shape and are close to normally distributed. The skew-ness values are 31.12 and 30.47 for N 18 O p and T, respectively. Likewise, the kurtosis values are 1.47 and 30.54. The P distribution is positively skewed. To correct this, we chose to log transform this parameter, resulting in the following revised model: The skewness and kurtosis values decreased from 2.77 and 11.44, respectively, for P, to 30.53 and 30.41 for log P. While the skewness and kurtosis values indicate that these distributions are not perfectly normal, they are close enough that they should not a¡ect the overall reliability of this method.
Second, we had to consider the e¡ects of serial correlation in which residuals from adjacent measurements in a time series are not independent of one another. In this type of data, there is often serial correlation that results from seasonal variations in T, P, and N 18 O p . We performed checks for serial correlation by plotting the multiple regression model residuals, O i , against their lags, O i31 and calculating the correlation coe⁄cients, r. We set the critical value of r = 0.32 (95% con-¢dence level), as the acceptable level for serial correlation. None of the models in this study required correction for serial correlation.
The third check we needed to make was for multicollinearity, or a correlation between our two independent variables, T and log P. Due to the fact that the climate in China is highly seasonal, and that the warmest season is also the rainiest season, there is a positive correlation between T and log P in the GNIP data (Fig. 4). This multicollinearity can lead to very large uncertainties in the partial regression coe⁄cients. To check if this is a problem for these data, we calculated variance in£ation factors (VIF) for the T^log P relationship at each site. The VIFs for all sites were all less than 2, so the correlations were not strong enough to warrant corrections for multicollinearity, however this e¡ect does slightly increase uncertainties of the model results.
Finally, we compare the results of our multiple regression model with those of simple linear regression models to test whether the more complex model successfully explains more of the variance in the data than the simpler models.

Results
The deuterium excess values, mean annual values of temperature (MAT) and precipitation (MAP), and the precipitation weighted annual means of N 18 O p and ND for the period 19881 990 are shown in Table 1 for the 10 GNIP sites used in this study. A plot of N 18 O p^T and N 18 O pP illustrates the lack of a signi¢cant relationship between N 18 O p and T or P (Fig. 5a,b) 6) for three of the GNIP sites: Hong Kong, which is dominated by the summer monsoon, Hetian, which is dominated by the winter monsoon, and Baotou, which is near the northern boundary of the summer monsoon.
The seasonal N 18 O p pattern in Hong Kong consists of lower N 18 O p values in the summer and higher values in the winter (ranging from about 310x to about 0x, respectively), suggesting that the dominant factor in£uencing stable isotopes in precipitation here is the amount e¡ect. In contrast, the highest N 18 O p values in Hetian occur in the summer and the lowest in the winter, suggesting that temperature is the dominant factor controlling stable isotopes in precipitation at  Based on the results of the multiple regression of N 18 O p against T and log P (Fig. 7), the 10 GNIP sites can be divided into four general regions: Southern China (Fuzhou, Guilin, Guiyang, Hong Kong, and Zunyi), Northeastern China (Baotou and Yantai), Northcentral China (Zhangye), and Western China (Hetian and Lhasa). The N 18 O p values at sites within each region exhibit similar dependence on temperature and the amount of precipitation (Table 2), due to the dominance of common atmospheric circulation patterns.

Southern China
The sites in this region are strongly in£uenced by summer monsoon precipitation. The dominant air mass source for these ¢ve sites is the Western Equatorial Paci¢c, due primarily to the southern location, south of the intertropical convergence zone for much of the year [21]. The multiple regression model results show that N 18 O p is nega-tively correlated to both T and log P at these sites ( Fig. 7), although the slopes and statistical signi¢cance vary. The fact that the partial regression slope of the N 18 O p^T relationship is negative, even after the precipitation e¡ect has been removed, is an artifact of the correlation between T and P. Since this artifact masks any underlying positive N 18 O p^T relationship, we assume that N 18 O p at these sites is most strongly related to the amount of precipitation.
Araguas-Araguas et al. [22] also reported the dominance of the amount e¡ect at sites in this region; however, the estimates of the slopes may be improved through use of a multiple regression model as opposed to a simple linear regression model. For example, in Hong Kong, Araguas-Araguas et al. reported a N 18 O p^P slope of 30.01 (R 2 = 0.76) and a N 18 O p^T slope of 30.39 (R 2 = 0.89), while the multiple regression model yields partial regression slopes of 31.78 and 30.19 for L logP and L T respectively (R 2 = 0.55), suggesting a greater in£uence of P on N 18 O p . While the R 2 for the multiple regression model is lower than for the linear regression models, the slopes are likely to be more accurate due to the separation of the confounding e¡ects of T and P. The di¡erence between the model results may also be due somewhat to the di¡erent time periods used in the two studies.

Northeastern China
The sites in Northeastern China are located near the northern limit of the summer monsoon precipitation. Although most of the rain falls during the summer months, it is far less than in the south, and a larger proportion of precipitation may be due to small, locally derived convective storms as opposed to the summer monsoon circulation. The multiple regression models for these sites all show positive correlations of N 18 O p with T (Fig. 7), suggesting that temperature is an important control on N 18 O p . N 18 O p is negatively correlated with log P at these sites, however, suggesting that the stable isotopic composition of precipitation is also related to the amount of precipitation in this region.

Northcentral China
The site in Northcentral China, Zhangye, is located north of the summer monsoon limit. N 18 O p  at Zhangye is positively correlated with both T and log P. This suggests that the N 18 O of precipitation in this region is strongly related to temperature. The positive N 18 O p^P relationship at Zhangye suggests that the multicollinearity artifact is larger than the amount e¡ect at this site.

Western China
The majority of the annual precipitation at the two sites in Western China falls during the summer monsoon months. Precipitation at Hetian does not originate from the monsoon, while precipitation at Lhasa does. While these two sites are both characterized by very strong positive relationships between N 18 O p and temperature and negative relationships between N 18 O p and precipitation (Fig. 7), these common relationships are re£ecting di¡erent controls on N 18 O p . The large positive L T (0.75) at Hetian is highly signi¢cant (1c = 0.07), while L logP is only slightly signi¢cant. The majority of precipitation at Hetian is derived during the summer from small convective storms, with temperature being the dominant control on N 18 O p . In Lhasa, where the annual N 18 O p pattern is similar to that of Hong Kong (Fig. 6), and L logP is much more signi¢cant than L T (Table 2), N 18 O p is controlled largely by the amount e¡ect. This suggests that the dominant moisture source for Lhasa is the summer monsoon precipitation. The N 18 O p systematics at Lhasa may be further complicated by interaction of the Southeast Asian monsoon and the Indian monsoon circulation.

Deuterium excess
The D-excess values are also somewhat distinct within each region. In Southern China, D-excess values are all greater than the global value of 10x, and range from 11.3x at Zunyi to 21.7x at Guiyang. In Northeastern China, Dexcess values are negative, with values of 36.9x at Baotou and 30.7x at Yantai. The value in Northcentral China at Zhangye is 4.4x. Values in Western China range from 4.5x at Lhasa to 10.7x at Hetian. This spatial variability of Dexcess in China suggests that the stable isotope systematics here are a¡ected by factors other than T and P. The D-excess values may re£ect site-to-site variations in the origin of atmospheric moisture (e.g. Indian monsoon vs. Southeast Asian monsoon precipitation), the relative humidity in the oceanic source areas, and secondary processes, such as the re-evaporation of raindrops below the cloud base [22].

Multiple regression versus simple linear regression models
To test whether a simpler model is more appropriate, we performed linear regressions of N 18 O p versus T and P for each site (Fig. 9), as in Araguas-Araguas et al. [22]. The dN 18 O p /dT slopes (L T ) and the dN 18 O p /dlog P (L logP ) slopes for all sites in Southern China exhibit the same sign as the partial regression slopes, L T and L logP , although the magnitudes vary signi¢cantly. The L logP values have opposite signs for Baotou and Hetian, while L T has the opposite sign for Lhasa. Table 3 shows the di¡erence between the 19881 990 mean N 18 O p for the 10 sites and the N 18 O p values modeled only as a function of temperature (v meanÀT model ) and only as a function of precipitation (v meanÀP model ). If these values are compared to the v meanÀmodel values in Table 3, it is evident that the multiple regression model does a better job of predicting N 18 O p in the Southern China sites, which are most heavily in£uenced by the summer monsoon precipitation. This suggests  that N 18 O p variations in precipitation in the Asian monsoon region cannot be used solely as a paleothermometer or a paleomonsoon proxy, but must be used as a combination of these.

Geographic model vs. multiple regression model
Bowen and Wilkinson [23] demonstrated that N 18 O p can be modeled quite accurately as a function of solely latitude and elevation. We compare the N 18 O p values calculated using this geographic model with those from the multiple regression model ( Table 3). The multiple regression model more accurately predicts N 18 O p for all ¢ve sites in the Southern China Region, while the simple latitude/altitude model is best in the other three regions, which are less a¡ected by the Asian Monsoon. Bowen and Wilkinson point out, however, that regional characteristics of the atmospheric circulation a¡ect N 18 O p patterns and hence, N 18 O p can be better represented via inclusion of an interpolated ¢eld of latitude/altitude model residuals. This signi¢cantly improves the geographic model's performance in Southern China.

Conclusions
Records of past variations of N 18 O and ND in precipitation, whether preserved in ice cores or £uid inclusions, or inferred from proxy data from geologic materials, such as speleothem calcite, can provide a wealth of useful information about paleotemperatures, paleoprecipitation, and paleocirculation patterns. The Asian monsoon region has one of the most interesting and complex climatic histories in the world; unfortunately, it is this very complexity that presents major obstacles to paleoclimatologists and climate modelers alike. A thorough understanding of the stable isotope systematics in this region is key for improving interpretations of multiple paleoenvironmental records. The simple multiple regression model presented here, which interprets N 18 O p as a function of only T and P, provides some useful insights.
The multiple regression results demonstrate the high degree of spatial variability in N 18 O p^T^P relationships in this region. The signs of L T and L logP are clearly related to the dominant atmospheric circulation patterns at any given site,  although the magnitude of these slopes is more di⁄cult to interpret. N 18 O p values at sites in Southern China, which are greatly a¡ected by the summer monsoon precipitation, are somewhat dependent on the amount of precipitation. Inclusion of temperature in the model, however, greatly improves our ability to reproduce the modern mean N 18 O p values. Sites in Northern and Western China may not be a¡ected as much by the amount e¡ect, yet inclusion of precipitation in the model may also improve the accuracy of the calculated slopes.
This high degree of spatial variability, which is somewhat related to the intensity of the summer monsoon, suggests that caution should be used in any quantitative, or even qualitative, interpretation of N 18 O p or ND paleorecords. It is well known that the intensity of the summer monsoon has varied greatly between glacial and interglacial periods, and also exhibits high-frequency decadal to annual variations in intensity. The fact that the degree of dependence of N 18 O p on T and P is related to monsoon intensity, which is highly variable, leads to the conclusion that the L T and L logP values at any given site may vary as monsoon intensity varies, and thus it may not be justi¢ed to use modern values in interpretation of paleorecords. Interpretation of paleorecords which lie near the northern limit of the summer monsoon may be especially di⁄cult, since the dominant control on N 18 O p may switch back and forth between P and T as the strength of the monsoon increases and decreases, respectively.
The site-to-site variability of the partial regression slopes may be re£ecting more complex factors such as the dependence of N 18 O p on elevation, latitude, source of moisture, topography, degree of water vapor recycling, and atmospheric transport pathways. The relatively low R 2 values for the ¢t of the multiple regression model to the data ( Table 2) suggest a large portion of the variance is unexplained by either T or P and may be attributed to one or more of these factors. The model could be signi¢cantly improved through inclusion of some of these variables, although this is obviously better accomplished using a GCM. This study is also intrinsically limited by the short duration of the data set, which is characteristic of the GNIP database in China. Inclusion of all the available data may slightly improve the ¢t of the model, but the use of discontinuous time series could lead to systematic biases in the resulting model, for example, through inclusion of only summer precipitation.
Uncertainty may also arise when these modeled seasonal N 18 O p^T^P trends are applied to interannual trends that are observed in the paleorecord. Unfortunately, the short duration of the GNIP data set makes direct multiple regression analysis of interannual trends impossible. Therefore, we assume that the best estimate of how N 18 O p varies as a function of MAT and MAP is derived through our investigation of how N 18 O p varies throughout the year as a function of mean monthly temperature and precipitation. A change in the relative dominance of the precipitation amount e¡ect and the temperature e¡ect on N 18 O p in the seasonal cycle would lead to a similar change on the interannual cycle.
The complex stable isotope systematics in China certainly present many challenges to isotope paleoclimatologists. If caution is used in interpreting the paleorecords, however, there is still great potential to obtain high-resolution records of monsoon variability over the last several glacialî nterglacial cycles, as well as further increase our understanding of the systematics of stable isotopes in precipitation in the Asian monsoon region. For example, seasonally or annually resolved ice core or speleothem N 18 O records may be useful for extending the record of N 18 O in precipitation. If independent records of temperature and/or the amount of precipitation are available, multiple regression analyses can be performed to investigate N 18 O p^T^P relationships over longer timescales than is possible from the GNIP database alone. Speleothems in particular hold great potential for providing high-resolution continental records of monsoon variability. They preserve multiple types of paleoclimate proxy data (N 18 O p , trace elements, growth band thickness, etc.) in highly resolvable growth bands and therefore, interpretation of past N 18 O variations can be checked against independent proxy data within individual samples. Studies such as this could sig-ni¢cantly improve our understanding of stable isotopes in precipitation, as well as improving the paleomonsoon record.