We studied the relationships between earlywood/latewood width, stable carbon isotope ratio (δ13C) of cellulose, and soil moisture at a dry and a wet site in Yakutsk, eastern Siberia, which differed considerably in soil water conditions. Recharge of soil water by snowmelt in spring and subsequent drought in summer provided a marked seasonal contrast in soil water conditions between the earlywood and latewood formation period. Ring index was calculated by dividing each earlywood/latewood width by the 5-year averaged width for each individual. In order to determine whether drought influenced the ring index-δ13C relation, the ring index time series were compared with δ13C time series. We collected wood samples from eight Larix gmelinii (Rupr.) Rupr. and four Pinus sylvestris L. trees from the two sites and measured the earlywood and latewood widths and δ13C of earlywood and latewood formed during the years 1996–2000. At the dry site, seasonal soil water content variation corresponded to seasonal δ13C variation of tree rings. We found negative ring index-δ13C correlations in latewood for both species at the dry site mainly dominated by Pinus but not in latewood of Larix at the wet site dominated by Larix. Decrease and/or early cessation of latewood growth and increase in δ13C under drought conditions possibly explain this negative correlation. This suggests the growth limitation of trees in this region by drought and the prospects of reconstructing past drought with latewood δ13C of the dry site.
 In eastern Siberia, Larix and Pinus forest accounts for 62.8% (215 million ha) and 13.2% (44.4 million ha) of the total forested area, and its wood reserves are 21.9 and 7.6 billion m3, respectively [Tseplyaev, 1965]. Because of their vast distribution and large wood reserves, they play an important role as a carbon reservoir in the global carbon cycle and as timber production forests. Climatic factors limiting their growth have come to attract great attention [Brooks et al., 1998; Jarvis and Linder, 1999; Vaganov et al., 1999]. A larger temperature increase has been expected for continental regions at high latitude, because the decreasing albedo due to snow reduction is expected to cause a positive feedback on temperature rise [Intergovernmental Panel on Climate Change (IPCC), 2001]. Corresponding changes in spatial distribution of these species and the size of their carbon reservoir is also expected. Thus clarifying the effect of water stress on tree growth in this region is necessary. Compared to other boreal regions such as North America or Scandinavia, eastern Siberia is characterized by very low precipitation, or dry climate. Differences in drought tolerance often determine the distribution of tree species within an ecosystem. Among the two dominant tree species in eastern Siberia, Pinus sylvestris L. (Scots pine) is distributed in relatively drier areas than Larix gmelinii (Rupr.) Rupr. (Dahurian larch) due to its higher drought tolerance. In this region, the beginning of the growing season depends on the timing of surface soil thawing [Vaganov et al., 1999]. The end of the growing season might be determined by the soil water status, since this region receives little precipitation (236 mm/year), and summer temperature sometimes exceeds 30°C. Plants therefore usually experience drought conditions in the latter part of the growing season. Cell enlargement, or growth, is very sensitive to water deficit [Abe and Nakai, 1999; Kozlowski, 1982] and drought may thus limit latewood formation.
where δ13Cp and δ13Ca are the carbon isotope ratios of plant tissue and atmospheric CO2, respectively, a is the isotope fractionation associated with diffusion (4.4‰), b is the net fractionation associated with carboxylation (27‰), and Ca and Ci are the mole fractions of CO2 in the atmosphere and the intercellular spaces, respectively. Ca and Ci relate to the rate of assimilation, A, and the CO2 conductance of the boundary layer and stomata pores, g, as follows.
The two expressions can be combined to
 According to this model, a positive deviation of δ13Cp is either caused by enhanced photosynthetic rate (A) or reduced stomatal conductance (g), when δ13Ca and Ca are constant. Under water stress, decrease in stomatal conductance (g) reduces preferential fixation of 12CO2 against 13CO2 and results in elevated δ13Cp values. Under such conditions where water stress can limit tree growth, δ13Cp usually increases, thus showing negative correlation between earlywood/latewood width and δ13Cp. For example, Mazany et al.  studied Pinus ponderosa and Abies concolor from New Mexico and concluded that warm and dry years cause narrow rings and high δ13C values. Such a negative correlation has been observed in many studies [Dupouey et al., 1993; Livingston and Spittlehouse, 1993; Livingston and Spittlehouse, 1996; Ponton et al., 2001]. McNulty and Swank  found a negative correlation between δ13C and basal area growth for Pinus strobus when soil water potential was below −1 MPa, even though the study area received a high rate of precipitation (1800 mm per year).
 An increased rate of carbon fixation (A) by the Rubisco enzyme also causes an increase of δ13Cp value, and it causes an increase of tree growth at the same time. Under such conditions, a positive correlation can be expected between ring width and δ13Cp. In fact, there are a few reports of positive correlations in regions such as Yakushima Island of Japan and the western coast of America, where soil moisture is unlikely to be a limiting factor for growth [Kitagawa and Matsumoto, 1993; Stuiver et al., 1984]. Thus by measuring both tree ring δ13C and ring width, and seeing correlation between the two variables, it may be possible to see if water stress can be a limiting factor of tree growth in eastern Siberia.
Leavitt  demonstrated that seasonal tree ring δ13C patterns of Pinus strobus and Acer saccharinum corresponded to soil moisture conditions and that δ13C from as few as four co-dominant trees of the same species can provide a drought record of a stand. Four is now considered the number of trees required to produce a representative stable isotope record [Leavitt and Long, 1984; Robertson et al., 1997; Saurer et al., 1995]. However, responses to soil moisture conditions are known to differ among species.
 Because of this close relationships between tree ring δ13C and soil water condition, tree ring δ13C is expected to be a good indicator for drought reconstruction, regardless of time-consuming δ13C analysis. The objective of this study was to examine whether ring index - δ13C relations change among sites or seasons with extreme differences in soil moisture and to see whether water stress can be a limiting factor of tree growth in this region. Furthermore, we wanted to see if there are prospects of reconstructing past drought with tree ring δ13C at these sites by measuring short-term δ13C series.
2.1. Site Description
 Trees for δ13C analysis were sampled at a Larix gmelinii (wet site) and a Pinus sylvestris stand (dry site) in the Spasskaya Pad experimental forest of the Institute for Biological Problems of Cryolithozone, located about 20 km north of Yakutsk city (62°15′N, 129°37′E) in a continuous permafrost region. The wet site was located on relatively moist and clayey soils with mixed sandy layers, where L. gmelinii was the only dominant species. The stand density was 836 (trees/ha), and the total basal area was 27.1 (m2/ha). The dry site, where ca. 70% of total basal area is accounted for by Pinus and ca. 30% by Larix, was located on well-drained sandy soils at a slightly higher ground level than the wet site, resulting in less water availability than at the wet site. The stand density and total basal area of L. gmelinii at the dry site were 624 (trees/ha) and 5.1 (m2/ha), respectively, while those of Pinus were 736 (trees/ha) and 11.9 (m2/ha). The two sites were located 400 m from each other, at about 220 m above sea level with the forest floors mostly covered with Vaccinium spp.
 Temperature and precipitation data obtained at Yakutsk city were used for this study [National Climate Data Center, 2001]. The years 1996–2000 experienced various climatic conditions. Growing season precipitation was lower in 1996 and 1998 and higher in 1999, while it was moderate in 1997 and 2000 (Figure 1). During 1996–2000, the average annual and growing season precipitation (June–August) at the study sites were 284 mm and 110 mm, respectively. There was an observation tower of 30 m height at the wet site and transpiration activity at the site was estimated [Ohta et al., 2001] during the period of the field campaign of GAME/Siberia (GEWEX Asian Monsoon Experiment/Siberia).
2.2. Wood Sampling
 Four dominant Larix trees were sampled at the wet site (the wet site Larix, or WSL) and four co-dominant trees each of Larix and Pinus were sampled at the dry site (the dry site Pinus, or DSP, and the dry site Larix, or DSL). Because of very narrow tree rings of these trees (0.4–0.8 mm) and the laborious sample preparation process, we had to limit the number of trees used. However, four trees per site were proved to be enough to represent the stand δ13C chronology [Leavitt and Long, 1984]. Tree age, height and DBH ranged 50–115 years, 8–13 m and 8–11 cm at the dry site, and 110–125 years, 16–23 m and 15–20 cm at the wet site, respectively. The size of the dry site trees was generally smaller than the wet site trees of the same age, probably due to lower water availability at the dry site. Trees at the dry site had open crowns while those at the wet site had semi-open ones, and all the trees sampled at the wet site reached crown top and were not shaded significantly by neighboring trees. Because of circumferential and vertical isotope variability in a tree ring [Leavitt and Long, 1984, 1986], wood blocks were extracted from each tree at the four cardinal compass points, at 1.3 m height above the forest floor in June 2001. Young trees were avoided because of the so-called “juvenile effect” [Francey and Farquhar, 1982]. Although trees at the wet site had less open crown than those at the dry site, the tower δ13Ca profile of CO2 taken at an observation tower at the wet site has shown that the circumambient air at the wet site was well mixed with the atmosphere above the canopy during the daytime (Naito, personal communications). Inputs of CO2 from decomposing organic material were therefore assumed to have had minimal influence on the δ13Ca. At several depths and locations at each site, soil moisture (volumetric soil water content) was obtained with time domain reflectometry (TDR; Moisture Point, Environmental Sensors Inc., Canada) manually at 0–15, 15–30, 30–60, 60–90, and 90–120 cm. Three and five probes were installed at the wet and dry site, respectively [Sugimoto et al., 2003]. Growing season soil water content (SWC) was measured from June 1998 to Aug. 2000 at the wet site and from Aug. 1999 to Aug. 2000 at the dry site.
 Parts of the sampled blocks were trimmed to 5 mm in tangential and 4 mm in longitudinal length. Then extractives were removed with a soxhlet extractor. The flask was first filled with a 2:1 toluene/ethanol mixture and run for three days. Then it was replaced with 100% ethanol and run for another three days. After the extraction, the samples were mounted into grooved wooden sticks with vinyl acetate emulsion and cut into 1.00 mm thick cross sections for soft X-ray densitometry. The sections were conditioned at 20°C and 60% RH before taking X-ray negatives. The negative films were scanned using a Dendro2003 instrument (Walesch Electronic, Effretikon, Switzerland), which has a sampling interval of ±0.01 mm, and the earlywood/latewood boundary was set at the half value of the minimum and maximum density for each ring. To remove width difference between individuals, earlywood and latewood ring indices (RI) were calculated as (earlywood or latewood width in a given year)/(average earlywood or latewood width during 1995–2000).
2.4. Sample Preparation for Carbon Isotope Analysis
 After cross dating the sample using the narrow rings formed in drought years 1994, 1996 and 1998 (which had characteristic very narrow latewood) and wide latewood formed in 1994, radial sections of 100 μm thickness were prepared from each block sample with a sliding microtome. Because of the narrow tree rings of the sample trees, we used thin radial sections to ensure accurate recognition and separation of earlywood and latewood. From the radial sections, we carefully subdivided each Larix tree ring into two parts (earlywood and latewood) and each Pinus tree ring into three parts (the first half of earlywood, the latter half of earlywood and latewood) under a stereomicroscope with a razor knife. Because of wider earlywood of Pinus than Larix, we could separate Pinus earlywood into half. Subdivisions from four directions were pooled into a single sample for each year to obtain an adequate amount for analysis and to ensure representation of the full circumference [Borella et al., 1998]. Samples were then purified to holocellulose according to the method described by Leavitt and Danzer . The holocellulose samples were then left in 17.5% NaOH solution for 45 min to produce α-cellulose.
2.5. Carbon Isotope Analysis of Earlywood/Latewood
 One-milligram subsamples of the cellulose were introduced to the combined system of an elemental analyzer (EA1108, Carlo-Erba, Italy) and an isotope ratio mass spectrometer (Delta S, Thermo Finnigan, Bremen, Germany) for δ13C measurement [Boutton, 1991]. Repeated combustion and analysis of identical samples yielded a standard deviation of 0.07(‰). The δ13C of wood tissue was calculated as follows;
which is expressed in the δ notation in per mil units (‰) with respect to the Peedee belemnite (PDB) standard. Because of the short span of the tree ring years (1996–2000), no correction was made to the tree ring δ13C data for changing δ13C of atmospheric CO2.
3. Results and Discussion
3.1. Annual Variation
 Before comparing ring width data with carbon isotope data, earlywood and latewood ring indices (RI) were calculated as (earlywood or latewood width in a given year)/(average earlywood or latewood width during 1995–2000). The trees generally showed similar annual δ13C variations regardless of sites or species, while RI trends were less coherent (Figure 1). Such similarity between sites and species was also reported by Leavitt . The δ13C values of earlywood were maximum in the years 1996 and 1998, medium in 1997 and 1999, and minimum in 2000. Latewood δ13C showed the same annual fluctuation except for 2000, when δ13C was not at a minimum. The latewood δ13C variations corresponded better to growing season precipitation than earlywood δ13C (Figure 1), and only latewood δ13C of DSL (the dry site Larix) was significantly correlated with the growing season precipitation (r = −0.90, P < 0.05).
 This negative correlation can be explained by the fact that trees at the dry site usually experience drought because of low soil water content during the latewood formation period [Sugimoto et al., 2002]. Other studies also found close relationships between growing season precipitation and δ13Cp [Dupouey et al., 1993; Leavitt and Long, 1988; Saurer et al., 1995]. Both the annual δ13C and RI trends fluctuated more in the latewood than in the earlywood and intersite variability of latewood δ13C (DSL - WSL) seemed to become bigger in the drought years (1996 and 1998, Figure 1), probably reflecting the extreme drought condition at the dry site in these latewood formation periods.
3.2. Seasonal δ13C and Soil Water Content
 Soil water content (SWC) over the growing season at the wet site decreased in 1998, increased or stayed constant in 1999, and decreased again in 2000 during the growing seasons (Figure 2). In 1998 and 2000, growing season precipitation was low (43 mm) and intermediate (109 mm), respectively (Figure 1), decreasing SWC during the growing seasons. In 1999, however, there was unusually high growing season precipitation (180 mm), resulting in higher SWC in the late summer. The average and standard deviation of precipitation over recent 100 years were 124 ± 41 mm [National Climate Data Center, 2001]. At the dry site, SWC was continuously lower than the wet site because of its sandy soil texture and deeper active layer (Figure 2). In spite of the difference in absolute SWC values, seasonal trend in SWC variation was similar between the two sites in 2000, where SWC at both sites decreased gradually over the course of the growing season (Figure 2), and this downward trend seemed to be typical in this region, according to the observational results by Sugimoto et al.  and the past climate record [National Climate Data Center, 2001]. SWC at the dry site was expected to have decreased in 1998 and increased in 1999 due to the growing season precipitation trend [Sugimoto et al., 2003, 2002].
 The δ13C of the atmospheric CO2 (δ13Ca) shows seasonal variation, and this variation can also affect the seasonal δ13C of wood (equation (1)). However, the differences between the average δ13Ca of the earlywood and latewood formation periods and between δ13Ca of the two sites were small (Naito, personal communications) and it was negligible compared to the seasonal δ13C variation of wood cellulose (Figure 2). We therefore did not correct the seasonal δ13C data of wood for the seasonal δ13Ca change. Intraring δ13C of the dry site trees (DSL, DSP) had become positive in 1998 and 2000, and stayed constant in 1999 during the growing seasons, corresponding inversely to SWC trend.
3.3. Effects of Drought on Seasonal δ13C
 Seasonal δ13C indicated the drought these trees experienced in the late summer of 1998 and 2000. However, unlike the dry site trees, the wet site trees showed less seasonal variation, with little or no change in 1998 and 2000, becoming more negative in 1999. WSL δ13C did not correspond well to the SWC trend (Figure 2). Even under the severest drought in 1998, SWC at 30–60 cm depth at the wet site remained relatively higher. According to the observation of seasonal xylem water δ18O variation at the wet site by Sugimoto et al. , WSL trees used meltwater from permafrost during drought summer. From the thawing layer of permafrost, water was continuously supplied to the upper soil layer during the drought at the wet site, to the depth where the deepest Larix roots were distributed in small amount. At the two sites, more than 80% of fine roots of the trees were distributed at 0–20 cm depth, however, fine roots were also observed up to 60 cm depth (N. Yanagisawa et al., manuscript in preparation, 2002). In the late summer in 1998, the thaw depth of permafrost at the wet site (1.2 m) was shallower than that of the dry site (more than 2.0 m), creating the difference in water supply from the permafrost and resulting in drier soil condition at the dry site [Sugimoto et al., 2003].
 According to the study on Pinus ponderosa in Arizona by Leavitt and Long , seasonal fluctuation of tree ring δ13C was highly and negatively correlated with measured soil moisture. Another study of seasonal δ13C changes in tree rings of Pinus strobus and Acer saccharum also showed good correlation between seasonal δ13C and drought index [Leavitt, 1993]. However, a study on Pinus strobus in western North California found tree ring δ13C was no longer affected by soil water conditions at higher SWC [McNulty and Swank, 1995]. The less sensitive response of seasonal δ13C to SWC at the wet site would be due to the higher SWC observed at the wet site than the dry site.
 Although subannual δ13C variation at WSL was small or sometimes absent, annual δ13C variation at WSL (Figure 1) was similar to DSL and DSP. Even when SWC is high, trees experience mid-day water stress because the transpiration flux usually exceeds that of water uptake [Waring et al., 1980; Kuwada, 2002]. For example, growing season temperature in 1998 was the highest of 1996–2000 and this high temperature may have contributed to stomatal closure and increased δ13C. Annual variation of δ13C in tree ring cellulose reflects not only SWC, but also other environmental variables such as annual variation of VPD, temperature or solar radiation [Hanba et al., 1997; Porté and Loustau, 2001]. The similarity in such environmental variables between the canopies of the two sites might be the reason for the similar annual variation of tree ring δ13C.
3.4. Ring Index-δ13C Relation
 Negative correlation between RI and δ13C was observed only for latewood formed at the dry site (r2 = 0.63 for Larix and 0.70 for Pinus, P < 0.005 for both; Figure 3) and a weak positive RI-δ13C correlation in WSL earlywood was observed at the wet site (r2 = 0.21, P < 0.05). We did not observe significant RI-δ13C correlation in earlywood formed at the dry site or latewood formed at the wet site.
 Since soil usually became dry in the latewood formation period, latewood of the dry site was formed under the most water-stressed conditions. In gymnosperms, water deficits affect the timing of the start of latewood formation. Low water supply usually causes latewood cells to begin forming earlier in the growing season [Kozlowski, 1982] resulting in decreased earlywood width and potentially increasing latewood width. However, water deficits decrease the cell expansion rate and can cause earlier cessation of latewood formation, decreasing latewood width. When the soil became driest in 1996 and 1998, extremely narrow latewood was observed at the dry site (Figure 1), hence the latter scenario seemed to explain the latewood formation in this region. δ13Cp also became positive in these drought years. Since water stress causes stomatal closure, it leads to decreased Ci and increased δ13Cp. One possible explanation for the negative correlation is increased δ13Cp and decreased latewood growth under drought conditions. Such correlation difference between sites is also reported by Martin and Sutherland , who found negative RI-δ13C correlation in SO2 stressed trees. SO2 exposure to leaves also cause stomatal closure, resulting in increased δ13Cp and decreased photosynthetic activity and growth. We also found negative correlation between tree ring δ13C and density for the dry site Pinus (A. Kagawa et al., manuscript in preparation, 2003).
3.5. Potential Cause of the Positive Correlation
 The WSL earlywood segments were formed under the least water-stressed conditions. According to equation (3), high photosynthetic activity, or high carbon assimilation rate (A), can also increase δ13Cp. Since there is always plenty of water available for plants at the beginning of the growing season in this region, stomatal closure due to low SWC is unlikely. In fact, both the tower observation and the porometer measurement proved that transpiration and photosynthetic activity became highest in June at the wet site, coinciding with the earlywood formation period in this region [Fujita et al., 1998; Ohta et al., 2001]. In 1998, the widest earlywood was formed (Figure 1) and there was least rainfall (very few cloudy days) in the earlywood formation period. Broadmeadow and Griffiths  found concurrent increases in δ13Cp with increasing photosynthetic photon flux density in a laboratory experiment with Picea abies trees, and increased photosynthetic activity usually leads to increased ring width. The weak positive correlation observed at the wet site may be explained by increased δ13Cp and earlywood growth under high photosynthetic activity. Increased δ13Cp and earlywood growth may also be explained by the use of stored photosynthates with higher δ13C value from the latter part of previous year's growing season. Earlywood in the dry site and latewood in the wet site were formed under mildly water-stressed conditions. McNulty and Swank  found negative relation between basal area growth and tree ring δ13C when soil water potential (SWP) was lower. However, the negative relation disappeared when SWP was over −0.1 MPa, where other factors like solar radiation seemed to affect growth. The absence of correlation could also be caused by the higher SWC.
3.6. Effects of Water Stress on Radial Growth
 A substantial drop of photosynthetic activity was observed [Fujita et al., 1998] under the driest conditions at the dry site (i.e., latewood formation period in 1998) and latewood formed in this period at the dry site was very narrow. Increased δ13C values were observed at the same time. The photosynthetic activity at the wet site did not show such a drop and latewood widths were at a medium level. Taking these observational results at each site (RI, earlywood and latewood δ13C, photosynthetic activity, and SWC) into consideration, water stress seemed to limit latewood formation at the dry site, while such limitation was not apparent at the wet site.
 Latewood δ13C corresponded to soil water conditions better than earlywood δ13C, probably because of the drought in the latter half of the growing season and possible use of stored photosynthates from previous growing season for earlywood growth, and all the trees showed similar annual variation of earlywood and latewood δ13C. Soil water content (SWC) at the wet site was constantly higher than the dry site. The SWC continuously decreased throughout the growing season in 1998 and 2000, while it stayed constant in 1999. Seasonal δ13C variation of tree rings at the dry site corresponded well to the variation of soil water content, while that at the wet site was less pronounced, probably due to higher SWC at the wet site than at the dry site. Latewood δ13C at the dry site, which experienced the driest conditions, was correlated negatively to ring index (RI), while earlywood δ13C at the wet site, which experienced the wettest conditions, showed a weak positive correlation to RI. Early cessation and/or limitation of latewood formation and increased δ13Cp under drought conditions may explain the negative correlation. Under relatively wet conditions, increased photosynthetic activity can increase the water use efficiency, or δ13C. Increased carbon assimilate also tends to increase radial growth. Increased δ13C and radial growth therefore may be the cause of the positive correlation. Contribution of stored photosynthates with higher δ13C values from the latter part of previous year's growing season may also explain this positive correlation. We could not observe any significant RI-δ13C correlations in earlywood formed at the dry site and latewood formed at the wet site. Thus the RI-δ13C relation appeared to change in response to soil water conditions at each site and season. Taking every observational result into consideration, water stress appeared to limit the latewood growth of trees at the dry site. These results suggest the prospects of reconstructing past soil moisture condition with latewood δ13C of the dry site.
 This study was supported by Grant-in-Aid 11554017 from the Ministry of Education, Culture, Sports, Science and Technology, Japan. The authors thank staff of the Institute for Biological Problems of Cryolithozone for helping with the field observations. We appreciate the cooperation with all members of GAME/Siberia.