• Open Access

Reconstruction of soil moisture for the past 100 years in eastern Siberia by using δ13C of larch tree rings


Corresponding author: A. Sugimoto, Faculty of Environmental Earth Science, Hokkaido University, Kita 10, Nishi 5, Kitaku, Sapporo 060-0810, Japan. (atsukos@ees.hokudai.ac.jp)


[1] A stable carbon isotope ratio (δ13C) chronology for the past 100 years was developed from larch tree rings in eastern Siberia (near Yakutsk, 62°14′N, 129°37′E), to reconstruct past soil moisture water equivalent (SWE). Based on the correlation analyses between SWE and tree ring δ13C, we developed a linear regression model for SWE in the late growing period (LGP: 15 July to 31 August) using annual tree ring δ13C, which was calculated from the combination of latewood in a current year and earlywood in the following year, and then reconstructed SWE (LGP) for 1908–2007. Reconstructed SWE was compared with factors such as the output of the land surface model, annual precipitation, and Palmer Drought Severity Index for July. From the results, the reconstructed SWE appears reasonable and shows a large variation, including repeated occurrences of severe drought and an unprecedented high soil moisture event in 2006–2007 during the past 100 years. The reconstruction also captured a past documented record of severe drought in the 1940s. Despite the generally good performance of the reconstruction, by the 1930s the estimated SWE was higher than that expected from the annual precipitation. Tree ring width and δ13C were negatively correlated in most periods. However, the negative correlation was weaker for the period from 1919 to 1925, when relatively low air temperature was observed. This result suggests that the rate of photosynthesis, together with the degree of stomata opening, also affected the tree ring δ13C during cool periods.

1 Introduction

[2] Boreal forests can greatly affect the global climate through their energy, water, and carbon cycles [Bonan et al., 1992]. Recent flux observations have reported that boreal forests may become either a sink or source of carbon, depending on the forest conditions, e.g., water and nutrient availability [Lindroth et al., 1998; Valentini et al., 2000]. In high-latitude continental regions, a rapid increase in air temperature has been observed, and a positive feedback to the global climate change has been reported [Lemke et al., 2007].

[3] The permafrost is closely related with the water cycle in eastern Siberia. The boundary between the active layer and permafrost plays an important role as source water storage [Sugimoto et al., 2003]. A large increase in water content of the surface soil with an increasing trend of active layer depth in recent years is probably related with not only the amount of annual precipitation but also inflow from the deeper thawing layer [e.g., Ohta et al., 2008]. Furthermore, thawing or warming of permafrost has been frequently reported in the observations during international polar year (IPY) [Romanovsky et al., 2010, Smith et al., 2010]. Degradation of the permafrost system triggered by global warming is expected to change soil moisture conditions, then, as a result to affect vegetation and carbon reservoir [e.g., Lopez et al., 2007].

[4] Eastern Siberia is characterized by small amount of annual precipitation (200–300 mm) and continuous permafrost. In drought summers, plants in this region tend to use ice meltwater in the soil rather than rain water, and their physiological activities are closely related to soil moisture conditions [Sugimoto et al., 2002]. In addition, it has been observed that plant transpiration greatly contributes to the atmospheric moisture [Ueta et al., 2013a, 2013b]. Therefore, soil moisture is of great importance in order to better understand the water and carbon cycles in eastern Siberia. However, a lack of long-term soil moisture observation in eastern Siberia makes it difficult to evaluate the processes related to the variation in soil moisture.

[5] In eastern Siberia, efforts have been made to accumulate accurate hydrological and meteorological data since 1998 in our study site under the GAME/GEWEX project (http://www.gewex.org/game.html) and soil moisture data are available for the past 14 years. However, much longer records of soil moisture are necessary to place the current moisture variability for this climatically sensitive region in a long-term context.

[6] Tree ring reconstruction of the past soil moisture can be a promising approach to extend the short-term record. In this region, early dendroclimatic studies focused largely on warm-season temperatures [e.g., D'Arrigo and Jacoby, 1993]. A number of more recent studies have demonstrated a reduced sensitivity of tree growth to rising temperatures (now referred to as “divergence problem”) at least since the 1960s [e.g., Briffa et al., 1998; Vaganov et al., 1999; Yonenobu and Eckstein, 2006; Wilson et al., 2007; D'Arrigo, et al., 2008]. From a plant physiological point of view, both soil moisture condition and temperature can greatly affect tree growth. Barber et al. [2000] reported that temperature-induced drought may limit growth of trees, using white spruce in North America.

[7] Tree ring reconstructions of Palmer Drought Severity Index (PDSI) have recently given fruitful insight on past droughts [e.g., Davi et al., 2010; Fang et al., 2010; Tei et al., 2013]. Cook et al. [2010] reconstructed the past PDSI for the summer period in monsoon Asia using a network of more than 300 tree ring chronologies. They presented cycles of dry and wet extreme events over the area and suggested a demise of the Khmer civilization due to a megadrought.

[8] Stable carbon isotope compositions in tree rings provide information on plant physiological conditions [e.g., McCarroll and Loader, 2004]. Under a warm and dry condition, the ratio (δ13C) reflects seasonal and annual variations in moisture condition [Leavitt and Long, 1989; Treydte et al., 2001; Sass-Klaassen et al., 2005; Sidorova et al., 2008]. Previous studies in eastern Siberia reported negative correlations between larch tree ring δ13C and hydrological conditions such as precipitation, relative humidity, and soil moisture [Kagawa et al., 2003; Kirdyanov et al., 2008; Sidorova et al., 2008; Tei et al., 2013]. It was also presented that tree ring δ13C negatively correlates with ring width. Consequently, they suggested that tree growth is predominantly limited by soil moisture.

[9] While the effect of drought on tree growth has been reported, many trees died due to over-wetting in our study site [Iwasaki et al., 2010; Ito et al., 2012]. Obviously, soil moisture (drought and also over-wetting) is one of the most important parameter, especially in such a dry climate region as eastern Siberia, to understand not only water but also carbon cycle. Data set on soil moisture is essential for various data analysis and modeling works as forcing or validation data. However, long-term observation data on soil moisture are not available in this region.

[10] Despite the importance of the soil moisture in various scientific fields, very few attempts for soil moisture reconstruction have been done [e.g., Anderson et al., 2012]. On the other hand, PDSI, which is the most commonly used drought index for agriculture, has been reconstructed with statistical approach in many studies using tree ring variables [Cook et al., 2010; Davi et al., 2010; Fang et al., 2010]. At our study site, Tei et al. [2013] reconstructed summer PDSI using larch tree ring δ13C with a statistical approach. Because of a property of PDSI, which is a comprehensive drought index for plant growth, it seems easier to reconstruct using tree ring variables. Unlike the PDSI, available period for soil moisture data set is also too short to reconstruct with a statistical approach. The PDSI, however, is not always a good measure of soil moisture [Dai et al., 2004]. Unquestionably, long-term reconstruction of soil moisture is required.

[11] In this study, we reconstructed soil moisture water equivalent (SWE) for the past 100 years using larch tree ring δ13C in eastern Siberia, which can be used as a forcing or validation data set in modeling works. In order to reconstruct SWE, we generated four sets of tree ring width and δ13C chronologies and directly compared with observed SWE data set from 1998–2008 to find most robust calibration model for past SWE and then compared the reconstructed SWE values with various data sets for validation. We also carefully evaluated accuracy of our reconstruction from not only a statistical approach but also plant physiological points of view.

2 Methods

2.1 Observations and Samplings

2.1.1 Site Description

[12] Sampling was conducted at a forest site in the Spasskaya Pad Scientific Forest Station (62°14′N, 129°37′E) of the Institute for Biological Problems of Cryolithozone, 20 km north of Yakutsk, Republic of Sakha, Russia. The detail of site information is described in Tei et al. [2013].

2.1.2 Tree Sampling

[13] Crosscut disks from eight larch (Larix cajanderi) trees were taken at 1.3 m above the ground in July 2009. Because the study site was a semi-open canopy forest, sample trees were not shaded significantly by neighboring trees. Ages, heights, and diameters at breast height were 182–223 years, 15–23 m, and 20–35 cm, respectively.

2.2 Ring Width and Carbon Isotope Analysis

2.2.1 Ring Width Measurement

[14] Standard techniques of dendrochronology were employed in sample processing and chronology development [e.g., Baillie and Pilcher, 1973; Cook and Kairiukstis, 1990]. These include measurement of earlywood and latewood (hereafter EW and LW) widths for two radii of a sample at a precision of 0.01 mm, followed by visual cross dating of the total ring (hereafter TR) width series using the PAST4 program (SCIEM, Inc., Vienna, Austria). The quality of cross dating was later checked by the COFECHA program [Holmes, 1983]. During the process of cross dating, missing rings were detected in 2003 for four of the eight samples. Finally, the raw ring width chronology (mean curve) was successfully cross-dated with a reference chronology derived from raw measurements (russ112.rwl) in the International Tree Ring Data Bank (http://www.ncdc.noaa.gov/paleo/treering.html).

2.2.2 Carbon Isotope Ratio of Tree Rings

[15] We used four samples for carbon isotope analysis. Each tree ring was separated into two parts, i.e., earlywood and latewood, with a surgical knife. Resin and oils were then removed from the wood using ultrasonic bath with acetone for the isotope analysis, using the same method described in Tei et al. [2013].

[16] We used bulk wood with acetone treatment for isotope analysis, since its δ13C showed satisfactorily high correlation with that of extracted wood cellulose (r = 0.96, p < 0.001), as previously pointed out [Loader et al., 2003; Cullen and Grierson, 2006]. Only acetone was used for extraction. We confirmed that our acetone treatment without hot water extraction produces the δ13C identical to that with hot water extraction (r = 0.96, p < 0.001).

[17] Carbon isotope ratio was analyzed with a ConFlo system (FLASH EA1112 with Delta V, Thermo Finnigan, Bremen, Germany). The isotope ratio was expressed by the delta notation,

display math(1)

where Rsample is the isotope ratio (13C/12C) of the sample and Rstandard is that of VPDB (Vienna PeeDee Belemnite). Repeated analyses of a sample yielded a standard deviation of less than 0.08‰.

[18] According to the model of plant carbon isotope discrimination by Farquhar et al. [1982], plant δ13C depends on a stomatal conductance (g) and photosynthetic rate (A), which are affected by environmental factors such as light intensity, relative humidity, temperature, and soil moisture. We will use the response of the δ13C to the change in soil moisture which is the most important controlling factor, as described in the introduction.

[19] Before the reconstruction of the soil moisture, raw δ13C values were corrected based on two factors. One is a change in δ13C of atmospheric CO2 [McCarroll and Loader, 2004], and the other is a change in plant physiological response to increasing atmospheric CO2 concentration [Treydte et al., 2001]. It has been confirmed that the latter correction makes the correlations between tree ring δ13C and climate variables better at our study site [Tei et al., 2013]

2.3 Chronology Building

[20] The series of ring widths of EW (earlywood), LW (latewood), and TR (total ring), which include 182–223 years old larch trees, were detrended by 128 year spline fits to generate standard chronologies using the ARSTAN program [Cook, 1985]. For the δ13C series of EW, LW, and TR, ensemble means were used as standard chronologies, since there is no age-related trend in δ13C chronology generally.

[21] To assess the reliability of the chronologies, the average of correlation coefficients between trees (RBAR) and the 51 year running expressed population signal (EPS) [Wigley et al., 1984] were calculated. The values of EPS for all above mentioned chronologies were greater than 0.88 for the analyzed span of 1907–2008 in this study, i.e., being well above the 0.85 critical level suggested by Wigley et al. [1984].

[22] Adding to the sets (width and δ13C) of chronologies EW, LW, and TR, in this study, we newly generated a set of chronology of width and δ13C, LE (latewood and earlywood), to find most robust calibration model for past soil moisture water equivalent (SWE). The time series of LE annual ring width was calculated from LW of a year and EW of the following year. The series of LE δ13C was obtained as a weighted average of LW and EW, using the ring width observed and average bulk wood densities for EW and LW, as follows:

display math(2)
display math(3)

where EW δ13C, ERW, and ERD are the δ13C, ring width, and ring density of EW, respectively, and LW δ13C, LRW, and LRD are those of LW. Subscripts of t and t + 1 mean a current year and the following year.

[23] For EW and LW densities, average values previously observed in our study site (0.393 g/cm3 for EW and 1.031 g/cm3 for LW) (A. Kagawa, unpublished data, 2000) were used.

2.4 Climate Data

[24] A combined monthly precipitation record was used for the analysis (1907–2008). For the period of 1907–1949, a precipitation record (62.5°N, 131°E) was extracted from a 2.5° longitude by 3.75° latitude degree gridded dataset at the Climatic Research Unit (CRU), East Anglia University (http://www.cru.uea.ac.uk/cru/data/precip/) [Hulme et al., 1998]. For the period 1950–2008, Baseline Meteorological Data in Siberia (BMDS) Version 5.0 (Yakutsk Weather Station: 62.017°N, 129.717°E) [Yabuki et al., 2011], which is considered to be more reliable but shorter, was concatenated to the CRU record.

[25] However, it should be noted that these observed data sets on precipitation contain uncertainties. Figure 1 shows variations of three precipitation records from CRU, Hydromet Yakutsk station, and BMDS. Average annual precipitation defined from previous August to the following July for these records was calculated for the period from 1950 to 1998, in which data of three records are available. The average values were slightly different (243 ± 51 mm for CRU, 249 ± 56 mm for Hydromet Yakutsk station, and 231 ± 48 mm for BMDS). Difference was also observed in the trend of these records. The CRU precipitation data set shows a gradual increasing trend over the twentieth century, whereas that of Hydromet Yakutsk station showed a slight decreasing trend. The BMDS showed no clear trend in the period. If we investigate the precipitation records in the period from 1950 to 1998, Hydromet Yakutsk station data showed significant decreasing trend (p < 0.05) and other two records showed no statistically significant trend (p > 0.10). Although year to year variations in the annual precipitation agree among these three data sets, longer time scale trends do not always agree.

Figure 1.

Annual precipitation calculated for the period from August in the previous year to July in the current year by using (a) Baseline Meteorological Data in Siberia (BMDS) Version 5.0 (Yakutsk Weather Station: 62.017°N, 129.717°E) for the period 1951–2008, (b) Climatic Research Unit (CRU) data set (gridded by 2.5° in latitude, 3.75° in longitude) for 62.5°N, 131°E, and (c) observed data of Yakutsk Hydromet Observatory.

[26] Monthly temperature and relative humidity for 1998–2008 were obtained from the monitoring data by the Automatic Climatic Observational System (ACOS) installed at our study site in a larch forest and maintained by Japan Marine-Earth Science and Technology [Iijima et al., 2010]. A monthly temperature anomaly data set (1907–2008) obtained from CRU (http://www.cru.uea.ac.uk/cru/data/temperature/) was also used for validation.

[27] Soil moisture water equivalent (SWE) for 1998–2008 was estimated by interpolating manual measurements by automatic measurements in ACOS using the same method developed by Sugimoto et al. [2003]. Manual observations of soil moisture content were conducted at three to five sites in the forest with time domain reflectometry (TDR; Moisture Point, Environmental Sensors Inc., Canada) at the depths of 0–15, 15–30, 30–60, 60–90, and 90–120 cm. On the other hand, automatic observations were also made with TDR in ACOS at 0, 10, 20, 40, 60, 80, and 120 cm. In this study, the soil moisture in the soil layer 0–60 cm which can be used by trees is considered for the analysis. The SWE data in August 2003 were obtained only from the manual measurements, because of a failure of the automatic observation system.

[28] A 2.5° × 2.5° gridded data set of monthly Palmer Drought Severity Index (PDSI) was obtained from the National Center for Atmospheric Research (NCAR). PDSI [Dai et al., 2004] is an index for comprehensive severity of drought in the hydrological system, in which precipitation is the primary variable and surface air temperature provides a secondary contribution to account for the effect of evaporation [Palmer, 1965]. A grid for 61°25′N, 128°75′E that covers the study site was obtained and used for the analysis.

3 Results and Discussion

3.1 Chronologies

[29] Figure 2 shows the standard δ13C and width chronologies for earlywood (EW) (top panel) and latewood (LW) (bottom panel) for the period of 1907–2008. The EW δ13C ranged from −24.3 to −22.3‰ with an average of −23.3‰ and the standard deviation ranged from 0.0 to 0.8‰. The LW δ13C also showed similar δ13C values ranging from −24.3 to −21.7‰ with an average of −23.1‰ and the standard deviation from 0.0 to 0.6‰. The EW and LW δ13C chronologies showed larger annual variations and relatively smaller variation between trees.

Figure 2.

Year to year variations in tree-ring δ13C (solid line) and ring-width index standardized by ARSTAN (dotted line) from 1907 to 2008 for (a) earlywood and (b) latewood. Tree-ring δ13C chronology shows average values for four trees with standard deviations (error bars).

[30] The EW δ13C chronology showed a similar trend to that for LW. However, significant difference of up to 1.1‰ in δ13C value was occasionally observed between EW and LW. In this region, EW and LW of larch tree rings are usually formed from the beginning of June to mid-July and from mid-July to the beginning of September, respectively. Large seasonal and interannual variations in SWE (soil moisture water equivalent) have been observed at our site [Sugimoto et al., 2003] and can cause significant differences in δ13C value between EW and LW. Therefore, further analysis was performed separately for EW and LW in this study.

[31] From 1907 to 2008, δ13C of EW and LW correlated with each ring width (Figures 2a and 2b). The correlation coefficient was higher for LW (r = −0.40, p < 0.01) than that for EW (r = −0.22, p < 0.05). The negative relationship between ring width and δ13C is generally expected when tree growth is limited by moisture conditions [Livingston and Spittlehouse, 1996; Ponton et al., 2001]. High negative correlation between ring width and δ13C for LW was also observed in the previous study [Kagawa et al., 2003], indicating that drought conditions cause a decrease or early termination of latewood growth and an increase in δ13C.

3.2 Climate/Tree Ring Parameters Relationships

3.2.1 Soil Moisture Water Equivalent

[32] Table 1 summarizes the correlation analysis between climate variables with tree ring parameters for the period from 1998 to 2008. EW and LW δ13C correlated negatively with soil moisture water equivalent (SWE) of the current summer (June to August) and the previous August. LW widths correlated positively with SWE both for current and previous summers. On the other hand, no significant correlations were found between summer SWE and EW widths at the 95% level. The explained variances (r2) for summer SWE accounts ~40–70% of the total variance of EW and LW δ13C, showing higher values than those of LW widths (~32–52%). This agrees well with a theory [McCarroll and Loader, 2004] as to that controlling factors of annual growth increments are more complicated than those of δ13C. It is concluded that latewood growth and physiological conditions of larch trees are predominantly controlled by summer soil moistures.

Table 1. Results of Correlation Analysis Between Tree Ring Parameters and Climate Variables for the Period From 1998 to 2008a
  Previous YearCurrent Year
Tree Ring ParameterClimate VariableJuneJulyAugustJJALGPJuneJulyAugustJJALGP
  1. aEW: earlywood, LW: latewood, TR: total ring, LE: combination of latewood in a current year and earlywood in the following year. JJA: June-July-August, LGP: late growing period (15 July to 31 August). Climate variables are SWE observed for the depth of 0–60 cm, and RH, air temperature and precipitation data from ACOS [Iijima et al., 2010]. Levels of significance are as follows: *, p < 0.01; blank, not significant at the 95% level; other values, p < 0.05.
δ13CEWSoil moisture water equivalent (SWE)  −0.81*−0.67−0.80*−0.80*−0.76*−0.63−0.77* 
LW  −0.76*   −0.76*−0.78*−0.81* 
TR  −0.79*   −0.78*−0.76*−0.80*−0.80*
LE      −0.80*−0.84*−0.83*−0.90*
TR  0.680.610.68   0.58 
δ13CEWRelative humidity (RH)  −0.72 −0.66     
LW  −0.64    −0.64−0.57−0.66
TR  −0.69 −0.58  −0.61 −0.60
LE0.64      −0.70−0.59−0.67
WidthEW  0.57       
LW  0.670.61      
TR  0.700.58      
δ13CEWAir temperature 0.76*        
LW      0.68   
TR 0.58        
LE 0.59    0.71   
WidthEW  −0.73* −0.74*0.59  −0.59 
LW −0.69        
TR  −0.60−0.60−0.70     
LE       −0.65 −0.62
δ13CEWPrecipitation  −0.67−0.61−0.66     
LW  −0.69 −0.68  −0.58−0.58 
TR  −0.70 −0.69  −0.60  
LE  −0.67 −0.66  −0.66−0.65−0.64
LW  0.710.670.73*     
TR  0.62       

[33] The significant correlations for previous summers are probably attributable to a carryover effect of late summer soil moisture kept in the active layer above the permafrost [Sugimoto et al., 2003]. Soil moisture in late summer may affect photosynthetic activity in the next growing season. It is also known that carbohydrates accumulated in previous summers are utilized to produce earlywood cells as demonstrated by a tracer experiment using larch trees in this region [Kagawa et al., 2006].

[34] Interestingly, the highest correlation was found between SWE in the late growing period (LGP: 15 July to 31 August) and LE δ13C. The explained variance accounts for 81% of the total variance. Although the timing of transition from EW to LW formation could vary due to many environmental factors [e.g., Vaganov et al., 1999], this roughly corresponds to the seasonal change in SWE. In June, more water is usually available for plants because of an infiltration of snowmelt water, and then gradually decreases from July to August (Figure 3). As a result, the effect of stomatal conductance on soil moisture is low in early summer, and then increases in late summer. In addition, carryover of carbohydrates photosynthesized in previous summers to the next growing season also contributes this highest correlation between SWE in LGP and LE δ13C.

Figure 3.

Soil moisture water equivalent (SWE) at a depth of 0–60 cm averaged for June (square), July (triangle), and August (circle) for the period 1998–2008.

3.2.2 Other Climate Variables

[35] Although the δ13C and ring width were less sensitive to climate variables than soil moisture water equivalent (SWE), some characteristic pattern can be seen in Table 1. Relative humidity in previous August showed significant correlations with all tree ring parameters except for LE (combination of latewood of the current year and earlywood of the following year) ring width and δ13C. Precipitation in previous August showed the similar correlation pattern except for EW width. Relative humidity and precipitation in current August and summer (June-July-August) correlated with LW δ13C. However, these are less responsible to the fluctuation of tree ring parameters, compared to those for SWE. It is thus reasonable to conclude that larger amount of precipitation in summer may contribute to higher growth rate and increase in carbon uptake; however, soil moisture is rather an essential factor to control physiological condition of larch trees which in turn controls growth and δ13C.

[36] It is not easy to fully explain the observed relationships between temperatures and the tree ring parameters. However, July temperature in previous or current year showed significant positive correlation with δ13C series and the temperature in previous year tend to show negative influence on tree growth.

3.3 Reconstruction of the Past Soil Moisture Water Equivalent

[37] As the number of data available for modeling is small (1998–2007, n = 10), we chose a simple regression model using the relationship with the highest correlation coefficient (r = −0.90) in Table 1 between LE δ13C and soil moisture water equivalent (SWE) (0–60cm depth) for the latter half of growing period (LGP: 15 July to 31 August):

display math(4)

[38] The scatterplot is shown in Figure 4a. The explained variance of the model after adjusted for the degree of freedom was 79.0% (p < 0.001), and the intercept and slope were statistically significant (p < 0.001).

Figure 4.

(a) Relationship between LE δ13C (combination of latewood in the current year and earlywood in the following year) and soil moisture water equivalent (SWE) (0–60cm) in the late growing period (LGP: 15 July to 31 August), (b) SWE estimated from LE δ13C (solid line), observed SWE (dotted line) and SWE estimates with a one dimensional land surface model (thin line with circle) [Yamazaki et al., 2007], and (c) reconstructed SWE (solid line) and the changes in mean state estimated by STARS method [Rodionov, 2004] (dotted line).

[39] An output of a physical model supports our reconstruction (Figure 4b). The open circles in Figure 4b indicate simulated SWE using a one-dimensional land/surface model [Yamazaki, 2001; Yamazaki et al., 2004, 2007] at the same site, which consists of three submodels for heat balance of vegetation, snow cover, and physical properties of surface soil. The reconstructed SWE agrees quite well with the independent model estimates after 1974 (r = 0.67, p < 0.001). The apparent deviations over the period 1966–1973 are possibly due to uncertainty of the initial condition.

[40] Figure 4c shows the reconstruction over the period 1908–2007. The reconstructed SWE mostly fell within the range where the regression is valid (−22.3‰ < LE δ13C < −24.3‰) with a few exception of −22.2‰ in 1947 and −22.1‰ in1980 over the period 1908–2007 (see Figure 4c). Dotted line in Figure 4c shows the changes in mean state of the SWE detected statistically with a method developed by Rodionov. [2004], in which sequential t test analysis is conducted to detect a regime shift (STARS) (cutoff length = 5, p < 0.1) by finding a multiple change points in a chronology. Severe drought periods at the study site were detected for 1947–1950 and 2002–2004. The former was well known severe drought described in local documents, and growth decrement of larch and pine trees has also been demonstrated by the tree ring analysis in this region [Nikolaev et al., 2009]. It can be said that trees in this region have experienced such a severe drought like in 2002–2004 in the past, of which amount of annual (previous August to current July) precipitation was 203 mm in average. After the drought in 2003, SWE has increased very rapidly. Over the past 100 years, the years 2006–2007 were the extreme wet event, of which amount of annual precipitation was 327 mm in average. In 2007, as a result, many larch trees in this region died [Ito et al., 2012]. In addition, the SWE time series is characterized by the changes in mean state and variance, showing the relatively high mean state with less variance until 1930 and then low mean state with large year-to-year fluctuations. This will be discussed in the next section.

3.4 Comparison With Other Environmental Records

[41] Figure 5 illustrates comparisons between the reconstructed soil moisture water equivalent (SWE) in LGP and annual precipitation from August in the previous year to July in the current year and July PDSI. Means and variances are adjusted to the same levels on vertical axes.

Figure 5.

(a) Soil moisture water equivalent (SWE) estimated from LE δ13C (solid line) (combination of latewood in the current year and earlywood in the following year) and annual precipitation for the period from August in previous year to July in the following year (dashed line), (b) reconstructed SWE (solid line) and July Palmer Drought Severity Index (PDSI) (dashed line).

[42] After 1930 the variation of SWE agrees well with those of the total precipitation (Figure 5a) and PDSI (Figure 5b), reasonably indicating that in situ soil moisture (LGP) is largely influenced by hydrometeorological conditions. On the other hand, for the period 1907–1930, SWE showed deviations from the precipitation and PDSI records in spite of the harmonic year-to-year variation. During this period, summer temperature was characteristically low and constant condition (Figure 6a). There is a possibility that a factor other than the SWE such as low radiation condition could also affect the δ13C. Therefore, relationship between ring widths and δ13C was examined to know the temporal change in sensitivity of larch tree ring δ13C to soil moisture depending on the change in physiological condition of larch trees.

Figure 6.

(a) June-July-August (JJA) temperature anomalies (thin line) and an 11 year running mean (thick line), (b) deviation of observed LE δ13C (combination of latewood in a current year and earlywood in the following year) from that expected by linear regression between LE δ13C and ring width index (LE δ13C = −1.06 × LE ring width index −22.15, r2 = 0.24), expressed as the difference in standard scores (ΔZδ13Crw-ob): math formula where δ13Crw and δ13Cob are LE δ13C estimated from the regression line and observed value, respectively. SDrw and SDob mean standard deviations of 13Crw and δ13Cob, and upper bar indicates the mean value for the whole period of each parameters. ΔZδ13Crw-ob (thin line) is shown with a 5 year running mean (thick line).

[43] The relationship between ring width and δ13C for LE was expressed by a simple linear regression (r2 = 0.24, p < 0.001) as follows:

display math(5)

[44] We calculated deviations of observed LE δ13C from that of estimated, by a difference in standard scores:

display math(6)

where δ13Cob and δ13Crw are the δ13C observed and that estimated from the equation (5), respectively, and the upper bar represents the average for the whole period. SDrw and SDob are standard deviations of δ13Crw and δ13Cob.

[45] Figure 6b shows the deviation of LE δ13C (ΔZδ13Crw-ob). When the observed LE δ13C is lower than that estimated from ring width (i.e., relatively low ring width and low δ13C), ΔZδ13Crw-ob shows positive value. This may happen when δ13C was influenced not by stomatal conductance (g) but by photosynthetic rate (A), according to the model of plant carbon isotope discrimination [Farquhar et al., 1982] as follows:

display math(7)

where δ13Cp and δ13Ca are the δ13C values of plant and atmospheric CO2, respectively, and the coefficients a and b are isotope discriminations during diffusion (4.4‰) [Craig, 1954] and CO2 fixation by RuBisCO (29‰), respectively, including the effect of CO2 dissolution in cell sap [Farquhar and Richards, 1984].

[46] Variations in the ΔZδ13Crw-ob series showed significant positive values during the 1920s. Long-term variation of ΔZδ13Crw-ob obtained by a 5 year running mean was higher than 1.0 for the period. We therefore concluded that low air temperatures generally caused by low radiation (i.e., cloudy) condition caused a decrease in the photosynthetic rate (A), resulting in decrease of tree ring δ13C. Thus, the anomaly (ΔZδ13Crw-ob) indicates a contribution of air temperature and radiation, together with the soil moisture, to the tree growth. Similarly, a positive ΔZδ13Crw-ob was observed during the early 1970s, when the temperature was low. Investigation on physiological condition of trees using the relationship between ring width and δ13C as described above is useful when the reconstructed soil moisture is utilized as a validation data for modeling, because accuracy of the reconstructed data is not equal for whole period of the reconstruction.

[47] Soil moisture observation over the last decade in our study site indicated that in situ soil moisture was very low in 1998, 2002, and 2003 and very high in 2006 and 2007 (Figure 3). Our reconstruction indicates that soil moisture in 2006–2007 was the highest for the past 100 years, while trees frequently experienced such severe drought as in 1947–1950 and 2002–2004 (Figure 4c).

[48] In this study, we used a tight relationship between larch tree physiology and soil water condition in eastern Siberia where moisture condition is the most important limiting factor for tree growth because of the small amount of precipitation under the continental climate. Undoubtedly, soil moisture is the most important controlling factor also for the δ13C of tree ring, and this results in the remarkable correlation between LE δ13C and soil moisture in LGP. In addition, such a high correlation as obtained here resulted from the regression model which was formulated from the physiological and hydrometeorological points of view, based on the observational results for C carryover and allocation and physiology of larch [e.g., Kagawa et al., 2003, 2006; Sugimoto et al., 2002] and hydrometeorology of permafrost system [e.g., Sugimoto et al., 2003; Ohta et al., 2008]. However, even if the correlation between LE δ13C and SWE in LGP was very high, it does not always ensure the same condition in the past, because various parameters might affect the tree growth. Testing a reliability of the reconstruction as shown in this study offers useful information when the reconstruction is utilized.

[49] It should be also noted that our reconstruction is based on in situ observations, while geographical characteristic in eastern Siberia, where larch-dominating forest covers the vast flat terrain with low spatial heterogeneity, provides regionally representative reconstruction.

[50] Obviously, such relationship as observed here is expected only under a dry climate. However, similar results have been reported not only in eastern Siberia [Kagawa et al., 2003; Kirdyanov et al., 2008; Sidrova et al., 2008; Nikolaev et al., 2009; Tei et al., 2013] but also in the forest step zone in north [Sidorova et al., 2009] and south [Knorre et al., 2010] of central Siberia where water availability for plants is low because trees are growing on a plateau. Although the method developed here is not applicable for whole region in Siberia, it may be possible to use for reconstruction of past soil moisture in dry regions.

4 Conclusion

[51] Soil moisture water equivalent (SWE) in the late growing period (LGP: 15 July to 31 August) was reconstructed for the past 100 years using larch tree ring δ13C in eastern Siberia, which was calculated using a combination of latewood (LW) of the current year and earlywood (EW) of the following year. The reconstructed SWE was supported by a physical model output using a one-dimensional land surface model [Yamazaki, 2001; Yamazaki et al., 2004, 2007] and reasonably agreed with the hydrometeorological records of annual precipitation and the Palmer Drought Severity Index (PDSI).

[52] Our reconstruction provides a high-resolution, long-term record of soil moisture in Yakutsk region of eastern Siberia. The reconstructed SWE shows large variations and repeated occurrence of severe drought over the past 100 years including the documented record of a severe drought in the 1940s. Our reconstruction revealed that soil moisture observed in 2006–2007 was the highest and it was an extreme event in the past 100 years.

[53] Although the reconstruction exhibited a generally good performance, it is also important to show reliability. Our reconstruction of soil moisture is based on a theory assuming that tree ring δ13C is exclusively controlled by stomatal conductance. In that case, the δ13C negatively correlates with tree ring width. We investigated this correlation between the δ13C and ring width to evaluate the reliability. Those were negatively correlated as we expected in most periods. However, during a period of relatively low air temperature from 1919 to 1925, negative correlation was weaker and the reconstructed SWE was larger than expected from the amount of annual precipitation. Low temperature condition possibly associated with low radiation affected the photosynthetic rate (A), as a result caused a decrease in δ13C in addition to the effect of stomata opening (g) controlled by soil moisture. Evaluation of reliability of the reconstruction is important because accuracy of the reconstructed soil moisture is not equal for the whole period.

[54] Soil moisture is obviously one of the most important factors controlling the water and energy balance between the land and atmosphere. Climate change in high-latitude regions could greatly affect the water cycle in the boreal forests and cause significant changes. Since our reconstruction is made independently from the precipitation records that include large uncertainties, our results are expected to greatly contribute a future analysis of the water and carbon cycle in eastern Siberia.


[55] We thank A. Maximov, A. Kononov and the other members of IBPC for their support for field works at the Spasskaya Pad station. We also thank A. Ueta, A. Popova, and other members of our laboratory. We also appreciate Y. Hoshino for her technical support in the laboratory works. This study was partly supported by grant-in-aid for scientific research (KAKENHI) (21403011, 21101002, and 23240116), IFES-GCOE, Hokkaido University (Global Center of Excellence Program), research project C-07 of the Research Institute for Humanity and Nature (RIHN), entitled “Global Warming and the Human–Nature Dimension in Siberia: Social Adaptation to the Changes of the Terrestrial Ecosystem, with Emphasis on Water Environments” (PI: Tetsuya Hiyama) and JSPS Research Fellowships for Yong Scientists.