SEARCH

SEARCH BY CITATION

Keywords:

  • δ13C chronology correction;
  • Qaidam Basin;
  • high-elevation trees;
  • temperature variations;
  • tree ring δ13C;
  • lake

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] Improved understanding of climate influences on tree ring stable carbon isotope (δ13C) ratios for Qilian juniper (Sabina przewalskii Kom.) will improve prospects for long climate reconstructions in northwestern China's Qaidam Basin, where weather stations are widely scattered with relatively short records. Here, we developed an annual-resolution δ13C series from 1800 to 2005 for trees in this extremely arid, high-elevation area. As expected, a significant decline in δ13C (of about 3.5‰) occurred from 1850 to 2005 in response to increasing atmospheric CO2 concentrations and decreasing atmospheric δ13C. High-frequency correlation analysis based on comparison of the tree ring δ13C chronology with recorded weather parameters revealed that mean temperature during the current growing season (April–August) most strongly influenced tree ring δ13C discrimination from 1956 to 2005. To clarify the climatic implications of the long-term trend, we systematically compared four previously published approaches to remove the effects of decreasing atmospheric δ13C from the climate signals. The optimal correction, which accounted for the decline in atmospheric δ13C (δ13Ccor) and for a discrimination rate of about 0.016‰ ppmv−1 for the CO2 partial pressure, captured the strongest temperature signal (r = 0.75, P < 0.001). The historical mean April–August temperatures inferred from the correlations of tree ring δ13C with climate data revealed a persistent warming trend during the past two centuries, especially since the 1980s. Our results therefore reveal a high potential for reconstruction of growing season temperatures on a millennial scale in the northeastern Tibetan Plateau.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] The Qinghai-Tibetan Plateau strongly influences the atmospheric circulation patterns over Asia, and especially the summer monsoon circulation, as a result of its large area (2.4 × 106 km2) and its high average altitude of more than 4000 m above sea level (asl) [Li and Yanai, 1996]. Evidence suggests climate on the plateau has also been responding to recent global climate change [X. D. Liu and Chen, 2000]. To obtain a better understanding of long-term climatic change on the Plateau, more climatic data over long time scales are needed. However, the observed meteorological data set is sparse and limited in length to less than a century (i.e., it began in 1950).

[3] During recent decades, many high-resolution palaeoclimatic records on the Qinghai-Tibetan Plateau have been derived from ice cores [Thompson et al., 2000; Wang et al., 2006], tree rings [Sheppard et al., 2004; Y. Liu et al., 2009; Shao et al., 2010], and lake sediments [Herzschuh et al., 2006; Zhao et al., 2008]. The temperature variations over the past two millennia have been established by integrating multiple proxies [Yang et al., 2003; J. A. Holmes et al., 2009], and the results indicate that the plateau has experienced long-term climatic episodes such as the Medieval Warm Period, the Little Ice Age, and the recent warming period. These studies have offered important insights into the patterns of climate change at interannual to centennial timescales.

[4] Among these records, tree rings have the advantages of covering wide areas, providing high temporal resolution (seasonal to annual), permitting exact dating, and providing high overall reliability [Esper et al., 2002; Cook et al., 2010]. Tree ring widths are therefore the easiest and most commonly used indices, and their use has provided many significant achievements [e.g., Esper et al., 2002; X. H. Liu et al., 2005; Shao et al., 2010]. Numerous tree ring width studies have been performed in the Qaidam Basin of the Qinghai-Tibetan Plateau and have achieved some significant findings. For example, Zhang et al. [2003] developed a 2326 year tree ring chronology for the Dulan area and found that the annual growth rings reflected variations in regional spring precipitation; similar results have been reported for nearby areas [Sheppard et al., 2004; Shao et al., 2005]. In addition, many scientists have identified temperature variations in and around the Qaidam Basin [X. H. Liu et al., 2005; Zhu et al., 2008; Y. Liu et al., 2009]. Yin et al. [2008] reconstructed the soil moisture conditions based on a water-balance model, revealing a general trend toward wetter conditions during the most recent 300 years. Recently, Shao et al. [2010] created the longest tree ring chronology in China, which covered the past 3585 years, using living and archeological Qilian juniper (Sabina przewalskii Kom.) trees. However, tree ring width data in this area suffers from several uncertainties, including trends related to the age of the trees and complicated interactions among environmental factors, and the results of tree ring analyses have therefore generated significant differences of opinion; for example, some results suggest that precipitation is the dominant factor affecting ring width [Shao et al., 2005], whereas others emphasize temperature [Y. Liu et al., 2009].

[5] Compared to tree ring widths, tree ring stable carbon isotope (δ13C) values have certain advantages, including an apparent lack of age-related trends after a short juvenile phase [Gagen et al., 2008], and often provide a higher signal-to-noise ratio [McCarroll and Pawellek, 1998]. The isotopic ratio can also potentially reflect past fluctuations in atmospheric CO2 [Feng and Epstein, 1995], large-scale atmospheric circulation patterns [Tardif et al., 2008], and local and regional climatic change [X. H. Liu et al., 2008]. Therefore, tree ring δ13C values have been widely used in dendroclimatology [X. H. Liu et al., 2008; Tardif et al., 2008; Young et al., 2010].

[6] However, tree ring δ13C records are widely considered to show a prominent downward trend unrelated to climatic but driven by increasing atmospheric CO2 concentrations and decreasing atmospheric δ13C since the industrial revolution [Waterhouse et al., 2004; Treydte et al., 2009; Young et al., 2010]. This trend is a direct result of atmospheric inputs of emissions of 13C-depleted CO2 from the combustion of fossil fuel and, as such, needs to be corrected prior to any discussion of the climatic significance of δ13C isotopic data.

[7] During the past few decades, a variety of correction methods have been used to reduce the noise and enhance the climatic signal. Early research simply fitted a polynomial or exponential equation to the raw tree ring δ13C series to detect the high-frequency climatic variations [Leavitt, 1994; Feng and Epstein, 1995]. Later, the incremental decline in atmospheric δ13C since the industrial revolution (around A.D. 1850) was added to the raw tree ring δ13C data [Treydte et al., 2001; McCarroll and Loader, 2004]. Recently, two newer approaches have considered both the influence of decreasing atmospheric δ13C and the simultaneous physiological responses of trees to increasing atmospheric CO2 concentrations [McCarroll et al., 2009; Treydte et al., 2009]. The main difference between these two approaches lies in the method used to capture the low-frequency climatic trend. McCarroll et al. [2009] mainly depended on properties inherent in the tree ring δ13C series, whereas Treydte et al. [2009] stressed the importance of calibrating the climatic signals. The methods differ in how they correct for the long-term effects of increasing CO2 concentration on tree ring δ13C.

[8] Tree ring δ13C values usually showed significant responses to climatic signals in the previous research. Typically, relative humidity and soil moisture are more likely to be a dominant factor in dry environments, for which stable carbon isotope ratios will be correlated with atmospheric humidity and precipitation during previous periods and the current growing season. In contrast, water stress is lower in moist environments, so the rate of photosynthesis is likely to be the dominant factor, leading to strong correlations with summer temperatures [McCarroll and Loader, 2004; Young et al., 2010]. However, in dry, high-elevation regions, temperature often becomes the dominant factor that controls tree ring carbon isotope discrimination [Kirdyanov et al., 2008; Treydte et al., 2009; Xu et al., 2011a]. Loader et al. [2010] reported that tree ring δ13C was strongly correlated with summer temperatures in the Altai Mountains of southern Siberia. Similar results were reported by other researchers near the same study area [Kirdyanov et al., 2008; Sidorova et al., 2011]. Treydte et al. [2009] found that tree ring δ13C was a sensitive proxy for summer temperatures at high elevations in the Karakorum Mountains of the southern Qinghai-Tibetan Plateau. In addition, Xu et al. [2011a] investigated the relationship between tree ring δ13C and climatic parameters in the source region of the Yangtze River and found that tree ring δ13C was a sensitive proxy for summer temperature.

[9] Although studies of tree ring δ13C responses to climate have resulted in rapid progress in this field around the world, there has been little investigation of annual-resolution data for the Qaidam Basin [X. H. Liu et al., 2007b]. To better understand the past climatic trends in this region, it is necessary to establish tree ring δ13C chronologies of local species such as Qilian juniper at a high resolution. In the present study, we developed an annually resolved tree ring δ13C time series for the eastern Qaidam Basin. We then systematically used the above-mentioned correction methods to correct our raw data for the tree ring δ13C series to account for anthropogenic increases in atmospheric CO2 concentrations and their δ13C depletion of fossil fuel. We then compared the four corrected tree ring δ13C series with local weather parameters to identify the isotopic responses to the regional climate and to determine which of the correction methods was best suited to our study area. On the basis of the results of this analysis, we reconstructed mean April–August temperatures from 1800 to 2005. We compared our temperature reconstructions with spatially gridded data sets (CRU TS 3) from around the study area and other temperature-sensitive proxies to confirm that our results truly reflected regional temperature trends. Our results have important implications for millennial-scale climatic reconstruction in China on the basis of tree ring δ13C.

2. Study Area and Materials

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1. Study Area and Climate

[10] Our study site (37°26′N, 98°03′E; 3670 m asl) was situated near the city of Delingha, in the northeastern part of the Qaidam Basin (Figure 1). Delingha is located at the margin of alluvial fans with elevations ranging between 2900 and 3000 m asl, but the mountains in the region can reach elevations higher than 4400 m. The area's climate has a strong continental influence, characterized by long, cold, dry winters; mean annual temperature is around 3.7°C at the Delingha meteorological station (37°22′N, 97°22′E, 2981.5 m asl) based on data from 1956 to 2005, and total annual precipitation averaged only 165.5 mm during the same period, with up to 60% falling in summer (June–August) (Figure 2a). From 1956 to 2005, the mean temperature and total precipitation have increased significantly (r = 0.85 and 0.67, respectively, with P < 0.001), at rates of 0.05°C yr−1 and 2.49 mm yr−1, respectively. However, atmospheric relative humidity has not changed significantly during this period (Figure 2b). Data from other weather stations in the region have recorded similar climatic variations [Xu et al., 2011b], and other researchers have found a strong warming trend in the Qinghai-Tibetan Plateau [X. D. Liu and Chen, 2000].

image

Figure 1. Map of our study area and locations of the sampling site near the city of Delingha in the Qaidam Basin.

Download figure to PowerPoint

image

Figure 2. (a) Climate diagram for the meteorological station in Delingha, 1956–2005. (b) Trends in the mean annual temperature and total annual precipitation, 1956–2005. Horizontal dashed lines represent the mean value for this period.

Download figure to PowerPoint

[11] The region is characterized by a desert steppe landscape [Zheng, 1996], with the natural vegetation consisting of various desert and dry grassland plants, such as Artemisia spp., Haloxylon ammodendron, and Sympegma regelii. However, in the mountain areas, sparse Qilian juniper trees are distributed on sunny and semi-sunny slopes, where above-average precipitation is received due to orographic effects [Du and Sun, 1990]. The crown density (the percentage of the total light that is blocked by the tree's crown; dimensionless) was generally lower than 0.2, and tree heights ranged from 3 to 6 m. Soils in the study area are primarily loess on 20° slopes, with a thickness of 20 to 50 cm, but they are thin and may even be absent on steeper and eroded slopes [Shao et al., 2005].

2.2. Tree Ring δ13C Chronology Development

[12] We collected one or two cores per living Qilian juniper tree from 24 trees using 12 mm-diameter increment borers at the DLH4 site on 20 April 2009. The DLH4 site [Shao et al., 2005] is a key site from which China's longest tree ring width chronology (3585 years) was developed. After air-drying and sanding the cores in the laboratory, we measured the ring widths to a precision of 0.01 mm and confirmed their dates using the COFECHA software [R. L. Holmes, 1983] in comparison with the previous ring width chronology created by Shao et al. [2005]. We used negative exponential or linear regression models to detrend the age series for each tree and developed our tree ring width chronology using the ARSTAN software [Cook, 1985]. We then used simple correlation to compare our tree ring width chronology with that of Shao et al. [2005] and found a strong correlation (r = 0.88, n = 203, P < 0.001) for the period from 1800 to 2002.

[13] In the Qaidam Basin, Qilian juniper grows very slowly, often by less than 0.1 mm in diameter annually, and the percentage of missing rings is high [Shao et al., 2003]. Previous research has indicated that there are no significant differences between pooling the samples from the same year regardless of their mass or width and averaging the individual trees in tree ring stable isotope research [Leavitt, 2007, 2008; Dorado Liñán et al., 2011], and this pooling method has been valid applied in the Qaidam Basin [Xu et al., 2011b]. In the current research, this pooling was made necessary by the fact that we also analyzed oxygen isotope ratios and will analyze hydrogen isotope ratios in the core samples, which required the use of larger samples than any individual core could provide. We selected 17 tree cores (one from each of 17 different trees). The cores were selected because they were well cross-dated with those of Shao et al. [2005] chronology, had few missing rings, and had regular ring boundaries that facilitated sample separation and subsequent tree ring isotope measurement. Each core was carefully separated into individual annual rings using a dissecting scalpel under a binocular microscope. We then pooled the annual rings from the samples by mixing all tree rings from the same year and storing them in a microcentrifuge tube labeled with the year [Treydte et al., 2001; Leavitt, 2008]. Wood samples were ground in a sample mill, and α-cellulose was extracted using a modified version of the method used by Loader et al. [1997, 2008] and Green [1963]. To better homogenize the tree ring cellulose, we used an ultrasound machine (JY92–2D, Ningbo Scientz Biotechnology Co., Ningbo, China) on the cellulose immersed in water to break down the cellulose fibers [Laumer et al., 2009]. The cellulose was then freeze-dried prior to the isotope analysis.

[14] The δ13C values were determined using a Flash EA 1112 Elemental Analyzer coupled with a Finnigan Delta Plus mass spectrometer (Thermo Electron Corporation, Bremen, Germany) at the Key Laboratory of Western China's Environmental Systems, Lanzhou University. By convention, the ratio of 13C to 12C is expressed in delta (δ) notation with reference to the Vienna Pee Dee Belemnite (VPDB) standard, for which the isotopic ratio is known. The carbon isotope ratio (δ13C) is expressed, in parts per thousand (‰), as:

  • equation image

where Rsample and Rstandard are the 13C/12C ratios in the sample and in the VPDB standard, respectively. During analysis, laboratory reference standards were periodically used to calibrate the analytical results. The analytical error of the isotope measurements (the standard deviation) was less than 0.07‰ for repeated samples.

3. Methods to Correct the Noise Resulting From Nonclimatic Factors

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[15] De-trending raw isotopic data to account for the atmospheric δ13C (δ13Ca) decline since the start of the industrial revolution is a critical step in exploring the response of tree ring δ13C to climate change. Since approximately 1850, the atmospheric CO2 concentration has been increasing and δ13Ca (air δ13C) has been decreasing due to burning of fossil fuels and deforestation [Keeling et al., 1979; Francey et al., 1999]. It is clear that δ13Ca directly controls plant δ13C (δ13Cp), so we attempted to correct the δ13Cp values using measurements of the atmospheric CO2 concentration and its δ13CO2 derived from ice cores as well as from direct measurements [McCarroll and Loader, 2004 and references therein]. Since we lacked data on the atmospheric δ13CO2 value in 2004–2005, we obtained this data from Prof. Danny McCarroll (Swansea University, personal communication). The annual values of atmospheric CO2 concentration in 2004 and 2005 were based on linear extrapolation from the atmospheric δ13C values from 1973 to 2003, which is similar to the approach used by McCarroll and Loader [2004] to estimate the 1998 to 2003 atmospheric CO2 values. The estimated atmospheric CO2 concentration generally accord with the observed values at Mauna Loa (http://cdiac.ornl.gov/trends/co2/sio-mlo.html).

[16] To clarify the climatic signals, scientists have used different approaches to correct for the decreasing trend in δ13Cp. Early studies used a quadratic polynomial or an exponential equation to fit the raw tree ring δ13C series (defined as the δ13Cpoly method in this paper). This approach assumes that δ13Cp can be simply and well explained by one of the two equation forms. The high-frequency variation in the δ13Cpoly values is then considered to represent climatic signals that are revealed after removing the low-frequency trends represented by the fitted δ13Cp values [Leavitt, 1994; Feng and Epstein, 1995].

[17] We chose to use a mathematical correction for the effects of changing δ13Ca using the following equation [McCarroll and Loader, 2004]:

  • equation image

[18] Since the carbon dioxide concentration of the atmosphere changed very slowly (rising from about 260 ppm at 8 ka B.P. to 285 ppm by A.D. 1850), we assumed that the atmospheric δ13C was relatively stable during this period and that the value of −6.4‰ can be the preindustrial baseline value [Saurer et al., 1997; McCarroll and Loader, 2004; Leuenberger and Filot, 2007]. This correction method ignores the physiological response of trees to elevated atmospheric CO2, and may suppress some of the low-frequency signal caused by nonclimatic trends in the records.

[19] We also employed some other correction approaches to account for effects of the increasing availability of atmospheric CO2 on the physiological behavior of trees. Farquhar et al. [1989] determined that the carbon isotope response depends on the relative increase in the intercellular CO2 concentration (ci) compared with the increase in the atmospheric CO2 concentration (ca); if ca increased at the same rate as ci (i.e., caci = a constant), this would result in a strong increase in the discrimination and would thus produce lower δ13C values. If ci increased more slowly than ca but in a proportional way (i.e., ci/ca = a constant), there would be no change in the discrimination. Many studies have shown that trees showed a switch from near-constant ci/ca toward near-constant caci, and have therefore revealed a sharp decline in δ13Cp in recent decades as a result of the effect of elevated CO2 on plant growth and related plant physiological properties [e.g., Feng and Epstein, 1995; Kürschner, 1996; Waterhouse et al., 2004].

[20] McCarroll et al. [2009] developed a correction procedure to correct the stable carbon isotope chronologies for increase in the atmospheric CO2 concentration. This method produces a “preindustrial corrected” δ13C series, and is defined here as the δ13Cpin method. There are two constraints in this method: first, a unit increase in ca cannot result in more than the same unit increase in the internal concentration of CO2 (ci); second, increases in water-use efficiency as a result of increased ca are limited to maintain a constant ci/ca ratio. This approach combines the high-frequency δ13Cp residuals achieved using nonlinear regression with low-frequency δ13Cp values since A.D. 1850. This procedure has been automated using the MATLAB software and the code from the R statistical programming language provided by McCarroll et al. [2009] (http://www.ldeo.columbia.edu/∼kja/access/code/pin.m). This approach was used to remove nonclimate trends that could be attributed to the increasing atmospheric CO2 concentration.

[21] Another approach for correction that we tested was to add a standard δ13C value per unit increase in the CO2 content of the atmosphere. This approach assumes that the response of trees to changes in atmospheric CO2 variation has been linear and uniform. Different correction values have been proposed, ranging from 0.007 to 0.020‰ ppmv−1 of CO2 [Feng and Epstein, 1995; Kürschner, 1996; Treydte et al., 2001, 2009]. Treydte et al. [2009] systematically tested a range of potential discrimination changes under elevated CO2 by increasing the correction factor stepwise from −0.05‰ to 0.05‰ ppmv−1 and found an “optimum” correction value with the strongest statistical correlation. On the basis of their approach, we first calculated the high-frequency signal (e.g., residuals from the linear trend fit) derived from the tree ring carbon isotope data and weather parameters (the mean temperature, total precipitation, and relative humidity), to avoid the low-frequency signals that would produce spurious correlations with the climate signals. Next, we correlated the high-frequency δ13Cp signal with high-frequency climate information to detect the main climate signals captured by the δ13Cp data. Third, we selected the optimal correction factor from a comparison of these results with the low-frequency variations. We then used a linear equation to fit the observed climatic signal identified from the high-frequency δ13Cp data and fit the corrected δ13Ccor time series by adding a continuous range of values from 0–0.05‰ at intervals of 0.001‰, and the low-frequency climatic signal had the same slope obtained from the fitted linear equations. We refer to this henceforth as the δ13Ccor + optimal correction-factor method.

4. Results and Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

4.1. Description of the Raw δ13C Chronology

[22] The raw δ13C series from the Delingha site was characterized by decreasing δ13C values since 1850, reflecting the impacts of industrialization and deforestation on atmospheric δ13CO2 (Figure 3a). The raw δ13C values ranged from −20.5‰ to −17.0‰, which is two times the decline in atmospheric δ13C (approximately 1.8‰) that occurred during this period; the overall average was −18.4‰ (standard deviation = 0.74). From 1850 to 2005, a linear regression of δ13C against time yielded a slope of −0.011‰ yr−1; since about 1950, the rate of decrease in δ13C accelerated to −0.034‰ yr−1. It is clear that the long-term declining trend in the raw δ13C chronology was not caused by climatic changes, thus to extract the climatic signals, especially the low-frequency variations, appropriate correction is required. The mean δ13C value is similar to that in our previous study (−18.9‰) in an arid basin site [X. H. Liu et al., 2007a], but is much greater (i.e., enriched) than the value at a semi-arid mountain site (−22.4‰) in China's Qilian Mountains [X. H. Liu et al., 2007a].

image

Figure 3. (a) The δ13C time series based on the raw data. The time series was then corrected using (b) a second-order polynomial (δ13Cpoly) [Leavitt, 1994], (c) the δ13Ccor method of McCarroll and Loader [2004], (d) the δ13Cpin method of McCarroll et al. [2009], and (e) the δ13Ccor + 0.016‰ ppmv−1 method of Treydte et al. [2009]. Horizontal dashed lines represent the mean value for this period.

Download figure to PowerPoint

4.2. Tree Ring δ13C Residuals Respond to the High-Frequency Climatic Signals

[23] To reveal the influence of climatic conditions on carbon isotope discrimination, we correlated the high-frequency carbon isotope residuals with interannual climatic variations (relative humidity, precipitation, and mean temperature from the previous August to the current September). Correlation analyses indicate that tree ring δ13C residuals sensitively respond to the mean temperature during the growth period, but the relationships with precipitation and relative humidity are relative weak (Table 1). The δ13C series were significantly (P < 0.05) correlated with relative humidity in the previous November and the current February and September, and positively correlated with precipitation in the current September, but significantly (P < 0.05) negatively correlated with the total precipitation during the April–August period of the present year. The tree ring δ13C series showed a persistent significant (P < 0.05) positive correlation with mean temperatures in the current April–August, and the correlation with mean April–August temperature was even stronger.

Table 1. Correlation Coefficients Between the δ13C Chronologies and the Weather Parameters in High-Frequency Variationsa
MonthCorrelation coefficient (r)b
TemperaturePrecipitationRelative Humidity
  • a

    The high-frequency variation was represented by the residuals between the raw data series and the fitted linear regression. The weather parameters were obtained from the Delingha weather station and are based on data from 1956 to 2005. The letter p before a month name indicates the previous year. “AMJJA” represents the mean for April, May, June, July, and August.

  • b

    A single asterisk (*) indicates a correlation significant at P < 0.05 level (two-tailed test), and a double asterisk (**) indicates a correlation significant at P < 0.01 level (two-tailed test).

pAug0.278−0.266−0.12
pSept0.190.035−0.107
pOct−0.144−0.095−0.016
pNov0.0140.282*0.321*
pDec−0.265−0.1940.055
Jan−0.075−0.061−0.144
Feb0.044−0.262−0.279*
Mar0.103−0.228−0.178
Apr0.332*−0.0740.028
May0.326*−0.165−0.133
Jun0.430**−0.223−0.193
Jul0.570**−0.231−0.217
Aug0.289*−0.0550.194
Sep0.0350.336*0.304*
AMJJA0.702**−0.334*−0.111

[24] The tree ring δ13C residuals appear to primarily reflect the temperature signal in the Qaidam Basin of the Northeastern Tibetan Plateau. The dominant environmental controls on the δ13C values in tree rings should be those that control the rate of stomatal conductance and the rate of photosynthesis [Farquhar et al., 1982; McCarroll and Loader, 2004]. In the arid Qaidam Basin, the precipitation or relative humidity is often the primary limiting factors for tree growth [Shao et al., 2005]; however, in mountainous areas, the cold temperatures found at high elevations, which strongly affect physiological processes such as photosynthesis, explain why the mean growing season temperature (April–August, mean temperature = 12.7°C) is an important influence. The strongly positive correlations with temperature suggest that the effect of photosynthesis rate is exerted on leaf internal CO2 concentrations (ci) during the growing season, when the stomatal conductance is low because of the dry climate. From April to August, as the mean temperature increases, the photosynthetic rate increases, resulting in decreased ci and increased δ13Cp. In addition, as the temperature increases, there is a simultaneous decrease in soil moisture content [Yin et al., 2008], and stomatal conductance decreases, resulting in decreased ci and increased δ13Cp. The effects of both increased photosynthetic rate and decreased stomatal conductance lead to the strong correlation with mean temperature. Similar results were reported for other areas of the Qinghai-Tibetan Plateau [Helle et al., 2002; Treydte et al., 2009; Xu et al., 2011a]. Relative humidity and precipitation also play a part in tree ring δ13C discrimination in the previous November and current September (Table 1), but the relationship was barely significant. Overall, the mean air temperature during the growing season was the dominant climatic factor that controlled tree ring carbon isotope discrimination in our study area. Since several interrelated climatic factors influence photosynthesis, at least part of the temperature signal recorded in the tree carbon isotope values may be indirect and may result from the existence of a correlation between temperature and other controlling factors, such as irradiance, sunshine duration, or vapor-pressure deficit [McCarroll and Loader, 2004; Young et al., 2010].

4.3. Corrected Tree Ring δ13C Chronologies Capture Regional Temperature Information

[25] Our correlation analysis revealed that the tree ring carbon isotope residuals in the Qaidam Basin mainly responded to the mean growing season temperature in the high-frequency domain. To clarify this hypothesis and capture the low-frequency trend, we created the four tree ring δ13C chronologies using the methods described in section 3 and then analyzed how well the different chronologies captured the temperature signals from 1956 to 2005.

[26] The δ13Cpoly series (Figure 3b) is a purely mathematical correction that has no physiological significance [McCarroll and Loader, 2004; McCarroll et al., 2009], and it cannot distinguish trends produced directly by changes in atmospheric CO2 isotope values and concentrations from trends caused by changes in climate or other environmental controls [McCarroll and Loader, 2004]. It therefore only reveals the high-frequency variations in δ13Cp. Even when the correction for changes in δ13Ca has been made, however, the δ13Ccor (Figure 3c) series, particularly for recent decades, continue to show a declining trend for which there is no evidence of a climatic cause. This decline has been verified around the world [Treydte et al., 2001; Waterhouse et al., 2004; Gagen et al., 2007], and this trend is often site- and species-specific [X. H. Liu et al., 2007a]. The two methods ignore the physiological response of trees to elevated atmospheric CO2 and may underestimate some of the low-frequency temperature signal.

[27] The δ13Cpin approach (Figure 3d) attempts to remove nonclimate trends that could be attributed to the increasing atmospheric CO2 concentration. However, the long-term fitting applied in the first step of this approach (the nonlinear regression) directly affects the slope and the direction of the trend revealed by the record corrected based on the preindustrial value. The span of this function can be widely varied to fit the low-frequency behavior of the data, but it remains unclear which fit is most appropriate to describe this low-frequency pattern [McCarroll et al., 2009; Treydte et al., 2009].

[28] To gain the optimal factor, we systematically compared the tree ring stable carbon isotope series to the mean April–August temperature. First, we standardized the observed mean April–August temperature as Z-scores and used the linear regression to fit it (a = 0.027). Next, we used the same method to fit the δ13Cp series corrected using correction factors ranging from 0 to 0.05‰ at intervals of 0.001‰. In the last step of our analysis, we found that the corrected δ13Cp chronology at a discrimination rate of about 0.016‰ ppmv−1 atmospheric CO2 concentration had the same slope (a = 0.027) as that of temperature. We therefore selected the correction factor of 0.016‰ ppmv−1 for the δ13Ccor chronology (Figure 3e). This approach highlights the quality of the related meteorological data. Moreover, the value of the factor may often be site related (e.g., the “optimal” factor is 0.012‰ ppmv−1 in the Karakorum Mountains, western high Asia [Treydte et al., 2009]). In addition, the assumption of linear trend may be not always effective as the concentration of atmospheric CO2 increased sharply since 1960s [McCarroll et al., 2009; Treydte et al., 2009].

[29] The δ13Cpoly and δ13Ccor chronologies still displayed a decreasing trend after the primary correction (Figures 3b and 3c). However, when we accounted for the physiological effect of higher CO2 levels on plant carbon isotope discrimination, the δ13Cpin and δ13Ccor + 0.016‰ ppmv−1 series showed a steady trend and a slight increasing trend, respectively (Figures 3d and 3e), even though the physiological correction is to some extent subjective and results in very different corrections [McCarroll et al., 2009; Treydte et al., 2009]. Based on a comparison of the corrected results mentioned above, we attempted to choose an optimal correction method that was best suited to the climate parameters of our study area, and used it to study the regional climate variations in the Qaidam Basin.

[30] We found significant (P < 0.01) correlations between mean temperature and stable carbon isotope ratios in most months (Figure 4). δ13Cpoly was significantly negatively correlated with the mean temperature in the previous October–December and the current January–February, and was also positively correlated with mean temperature in the present July. δ13Ccor was significantly negatively correlated with mean temperature in the previous December and significantly positively correlated with mean temperature in the current July and in the April–August period. δ13Cpin was significantly positively correlated with mean temperatures in the current July and in the current April–August period. δ13Ccor + 0.016‰ ppmv−1 was significantly positively correlated with mean temperatures in the previous August, in the current April, June, July, and August, and with mean temperature in the current April–August period.

image

Figure 4. Correlation coefficients between the instrumental mean temperature variations and the four tree-ring-corrected δ13C series from 1956 to 2005. Horizontal dashed lines represent the 99% confidence interval. The letter p before a month name indicates the previous year. “AMJJA” represents the mean for April, May, June, July, and August.

Download figure to PowerPoint

[31] Climate response to the four corrected series differed in the signal strength mainly caused by the low-frequency variations, i.e., the different correction methods. The high-frequency variations of the δ13C residuals results in similar climate relationships as the different approaches only corrected the long-term declining trend. The high correlation coefficients between δ13Cpoly and the mean temperature from the previous October to the current February may be spurious correlations; on one hand, the δ13Cpoly chronology still showed a declining trend, especially in the most recent few decades, but on the other hand, the winter mean temperature increased significantly over the period of instrumental records in the Tibetan Plateau [X. D. Liu and Chen, 2000]. These two low-frequency trends caused the negative correlations between δ13Cpoly and mean temperature of the winter half year. Thus, the δ13Cpoly series cannot reflect, at least as indicated by the weak correlation, the low-frequency climatic variability in the Qaidam Basin. The negative relationship between δ13Ccor and mean temperature in the previous December may also be caused by the low-frequency variation as the δ13Ccor chronology showed a slightly decreasing trend. The δ13Cpin series seem to reflect the growing season temperature information, but its correlation coefficient with April–August mean temperature is weaker than that with the δ13Ccor + 0.016‰ ppmv−1 series. Therefore it is not the most optimal correction method in the Qaidam Basin, though it has been employed in recent studies [Young et al., 2010]. The δ13Ccor + 0.016‰ ppmv−1 series positively correlates to the mean temperature, which is opposite to the other corrected δ13C chronologies and the high-frequency variations in the winter half year. This paradox is also rooted in the low-frequency, as the δ13Ccor + 0.016‰ ppmv−1 series exhibits a slightly increasing trend. However, this does not matter, because we mainly focused on the growing season temperature changes as the δ13Ccor + 0.016‰ ppmv−1 chronology correlated with the average April–August mean temperature. During the growing season, the four corrected δ13C chronologies correlated positively with the temperature variation, the δ13Ccor + 0.016‰ ppmv−1 series exhibiting the highest correlation coefficient with the temperature data.

[32] To investigate the validity of the δ13Ccor + 0.016‰ ppmv−1 correction method, we correlated the four corrected δ13C series with spatially gridded land surface temperatures from the CRU TS 3 time series [Mitchell and Jones, 2005] covering the Qaidam Basin (Figure 5). We found highly significant positive correlations between the gridded land temperature data and the instrumental mean temperature data from April to August 1956–2005 (Figure 5a). We then correlated the mean April–August spatially gridded temperatures with the four corrected tree ring δ13C series (Figures 5b5e). The results suggest that the tree ring δ13C chronology corrected using the δ13Ccor + 0.016‰ ppmv−1 method captured more of the temperature signal than the other methods, with correlation coefficients reaching around 0.7 (n = 50, P < 0.01) during the calibration period. Based on these analyses, we believe that the δ13Ccor + 0.016‰ ppmv−1 correction method is the most appropriate correction method for the tree ring δ13C time series (and to remove the declining δ13C trend) in the Qaidam Basin.

image

Figure 5. Spatial correlations (r values in color, P < 0.1) between the mean April–August (AMJJA) temperature data from a regional grid (the CRU TS 3 temperature time series, 1.0° × 1.0° resolution) and the (a) observed AMJJA mean temperature at the Delingha station, (b) tree ring δ13Cpoly series, (c) tree ring δ13Ccor series, (d) tree ring δ13Cpin series, and (e) tree ring δ13Ccor + 0.016‰ ppmv−1 series during the period 1956–2005. The analyses were accomplished using the KNMI Climate Explorer software (Royal Netherlands Meteorological Institute, http://climexp.knmi.nl). The star represents the location of the sampling site in the present study.

Download figure to PowerPoint

4.4. Reconstruction of the Mean Temperature From April to August

[33] Based on the analysis in section 4.3, we believe that in the extremely arid and high-altitude Qaidam Basin, the tree ring carbon isotope characteristics in Qilian juniper were more sensitive to the mean temperature from April to August of the current growing season than they were to precipitation. We plotted the δ13Ccor + 0.016‰ ppmv−1 values and the mean current growing season temperature both in the recorded level and high-frequency variation (Figures 6a and 6b), and found the carbon isotope values and temperature are linearly related. Then, using the temperature data for the period from 1956 to 2005, we established a regression model that accounted for 56.3% of the variance (adjusted R2 = 55.4%, P < 0.001). The DW (Durbin-Watson) test value was 2.15, which is close to the optimal value of 2, and this indicates that there was little autocorrelation in the regression model's residuals. Furthermore, the leave-one-out test yielded a positive Reduction of Error (RE = 0.53), indicating that the regression model provides a useful prediction, and the Sign test value (38+/12−, P < 0.001) was also statistically significant. We also found the instrumental mean April–August temperature matched the reconstructed data during the period 1956–2005 (Figure 6c). In addition, we split the mean temperature into two periods (1956–1980 and 1981–2006) and tested the temporal stability of the reconstructed data using calibration/verification statistics such as r, DW, Sign test, RE and CE (coefficient of efficiency) [Cook et al., 1994]. The calibration/verification tests indicated that while correlation was relatively low for the first period of the series, there appeared little reason to doubt the reliability of the early part of the reconstruction, as the RE and CE statistics remained positive and the Sign test was significant. These results suggested that the reconstruction formula was sufficiently reliable that it could be used to reconstruct past temperature variability. We therefore used the model to reconstruct the mean temperature from April to August (TAMJJA) from 1800 to 2005 in the Qaidam Basin.

image

Figure 6. Scatterplots with linear regression lines for the relationship between δ13Ccor + 0.016 ppmv−1 and the mean April–August temperature (a) observed values and (b) linear detrended values. (c) Observed mean April–August temperature (solid line), plotted with reconstructed mean April–August temperature (dashed line) during the period 1956–2005. Inset table shows the calibration/verification statistics for the reconstructed growing season temperature.

Download figure to PowerPoint

[34] Figure 7a shows the reconstructed temperature variations, with the smoothed line representing the 10 year Fast Fourier Transform (FFT) smoothing curve to emphasize decadal-scale fluctuations. The mean temperature from 1800 to 2005 was 12.25°C, and the standard deviation (σ) was 0.57°C. In contrast to the decrease in temperature that occurred from 1800 to 1850, a slight warming trend was obvious from 1850 to 2005 (Figure 7a). We defined significantly warmer years as years in which the temperature was greater than the mean + σ (12.82°C) and significantly colder years as years in which the temperature was less than the mean − σ (11.68°C). During the 206 year study period, the years with extremely low temperature were A.D. 1847–1859, 1864–1867, 1870, 1888, 1889, 1892–1907, 1915, and 1944; and the years with extremely high temperature were A.D. 1811–1814, 1877, 1884–1885, 1918–1920, 1963, 1966, 1980, and 1995–2004. The FFT smoothing curve revealed obvious cold periods from 1836 to 1872 and from 1886 to 1915, and a recent warm period from 1980 to 2005. This indicates that during the twentieth century, and especially since the 1980s, the mean growing season temperature has been increasing significantly in the Northern Tibetan Plateau. However, we also find a remarkable phenomenon that the warming trend reached a peak in the year of 1998. Since then, the growing season temperature shows an obvious declining trend, though it remains higher than the average temperature of the past 206 years.

image

Figure 7. Comparisons among different temperature reconstructions at both regional and hemisphere scale: (a) the reconstructed mean temperature for the April–August period based on the δ13Ccor + 0.016‰ ppmv−1 correction in the Delingha region, 1800–2005 (present study); (b) reconstruction of mean temperatures from the previous September to the current April using tree ring width data from a site near Wulan in the Qinghai-Tibetan Plateau for the period 1800–2004 [Zhu et al., 2008]; (c) the reconstructed mean May–July temperature using tree ring stable carbon isotope values in the source region of Yangtze River for the period 1850–2002, central Tibetan Plateau [Xu et al., 2011a]; (d) the tree ring δ13C recorded summer (May–August) temperature variations in the western part of High Asia [Treydte et al., 2009]; and (e) the Northern Hemisphere temperature variations derived from tree ring width for the period 1800–1992 [Esper et al., 2002]. The gray horizontal dashed lines in Figure 7a represent the mean ±σ (standard deviation) for this period. The black horizontal dashed lines represent the mean value for this period in all graphs; the bold curve represents the smoothed results using a 10 year Fast Fourier Transform (FFT) filter to emphasize long-term (decadal) variations in the graphs.

Download figure to PowerPoint

[35] A comparison of our temperature-sensitive tree ring records with published temperatures developed using tree ring records from Wulan (Figure 7b; 37°02′N, 98°39′E, 73 km distant from our study site) [Zhu et al., 2008], the central Tibetan Plateau (Figure 7c; 33°48′N, 96°08′E, 4,060 m asl) [Xu et al., 2011a], the western part of High Asia (Figure 7d) [Treydte et al., 2009], and the Northern Hemisphere (Figure 7e) [Esper et al., 2002] provides a reference to validate our low-frequency variation reconstruction. Our temperature reconstruction had a significant warming trend during the past two centuries intermittent cool periods, which compares well with the other previous reconstructions. The epochs of the obvious warm and cold intervals, such as the warm periods during the 1880s and 1920s, were recorded in the central and eastern part of the Tibetan Plateau but were not recorded in the western part of the Tibetan Plateau. This indicates that the temperature variations were not homogenous around the vast Plateau. The significant cold periods from 1840 to 1860 and in the 1890s and 1940s were generally recorded in all of the results. However, the 1970s cold period was only partially recorded in the Wulan area, possibly because winter temperature had increased faster than the summer temperature in the Qinghai-Tibetan Plateau [Liu and Chen, 2000]. The cooling trend since 1998, which was found in the Wulan area and central Tibetan Plateau, indicates that this phenomenon may be common in High Asia [Zhu et al., 2008; Xu et al., 2011a], and we will address this phenomenon in the future using temperature data from a longer timespan.

[36] However, from 1810 to 1823, there was a strong discrepancy between our temperature reconstruction and the other temperature reconstructions (Figure 7), with our results showing a temperature increase and the other simulations showing a decrease. The reason for this discrepancy is not clear and should be investigated in future work.

5. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[37] In the extremely arid Qaidam Basin, the tree ring δ13C chronology revealed a significant decrease in raw δ13C values since A.D. 1850 (by about 3.5‰), presumably as a consequence of changes in atmospheric CO2 resulting from industrialization. Our high-frequency correlation results suggested that the mean temperature from April to August had the strongest effect on tree ring carbon composition. We used four approaches from the literature to correct tree ring stable carbon isotope values for the effects of increasing atmospheric CO2, and compared the results with both local weather data and with the CRU TS 3 spatially gridded data set. We found that adding a constant correction factor of 0.016‰ ppmv−1 to δ13Ccor could extract more reliable climatic signals, especially for the low-frequency signals, and the mean temperature from April to August had the strongest correlation with the δ13Ccor + 0.016‰ ppmv−1 chronology from 1956 to 2005 (r = 0.75, P < 0.001). Using this tree ring stable carbon isotope series, we reconstructed the historical mean April–August temperatures and found an obvious warming period since the 1980s. Our temperature reconstruction generally corresponded well with previous regional temperature reconstructions, though we found some differences in certain periods that will require further study to explain. Our results therefore have some value for inferring historical interannual growing season temperature variations and reveal a high potential for reconstruction on a millennial scale temperature changes in the northeastern Tibet Plateau.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[38] This research was supported by the Global Change Research Program of China (2010CB951401), by the National Natural Science Foundation of China (40871002), by the Knowledge Innovation Project of the Chinese Academy of Sciences (KZCX2-YW-QN308), and by the Self-Determination Project of the State Key Laboratory of Cryospheric Sciences (SKLCS09–03). We thank the journal's anonymous reviewers and the journal's editor, whose comments and suggestions were helpful in improving the quality of this paper. We thank Prof. Danny McCarroll (Swansea University) for providing the atmospheric δ13CO2 values from 2004 to 2005 that we used in our analysis.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study Area and Materials
  5. 3. Methods to Correct the Noise Resulting From Nonclimatic Factors
  6. 4. Results and Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jgrg819-sup-0001-t01.txtplain text document1KTab-delimited Table 1.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.