The carbon and oxygen isotopic compositions of tree ring cellulose were examined for trees along a precipitation gradient in western Oregon, United States. Two years of cellulose from four sites dominated by coniferous forests ranging in precipitation from 227 to 2129 mm were sampled in conjunction with studies that measured the δ18O and δ13C of ecosystem respiration. The mean tree ring cellulose δ13C varied from −22.1 to −26.3‰ among sites and showed enrichment with decreasing water availability across the transect. The δ13C in cellulose varied across the precipitation transect in a similar pattern to the δ13C of leaf and root tissues as well as ecosystem respiration, although tree ring cellulose was enriched in 13C by over 3‰ compared to other organic matter components. The mean tree ring cellulose δ18O varied from 28.1 to 30.3‰. However, trends of cellulose δ18O change with water availability were obscured by differences in stem water δ18O. When calculated as deviation from stem water (δ18Ocellulose − δ18Ostem water) the differences in evaporative enrichment between sites was more pronounced (range of 9.6‰). The limited observed variation in tree ring cellulose δ18O of field grown trees despite large site difference in stem and leaf water δ18O across the transect agreed with predictions from a mechanistic model. Tree ring records of cellulose δ18O may provide useful proxy information regarding humidity and site water balance especially if combined with δ13C records that also vary with plant water status.
 The isotopic compositions of tree ring cellulose are widely used to study climate variability, since individual tree rings can be accurately dated (with sub-annual resolution) and cellulose once deposited is immobile. The three isotopic tracers in cellulose (carbon, hydrogen, and oxygen) have been correlated with climatic drivers, often leading to robust conclusions that allow climatic reconstruction in a region [Anderson et al., 1998; Leavitt and Long, 1991; McCarroll and Pawellek, 2001; Robertson et al., 2001; Saurer, 2003], but occasionally studies have been at odds with each other especially with regard to whether or not a specific climatic parameter is recorded in cellulose. For example, some studies have concluded that cellulose isotopes recorded interannual variations in atmospheric humidity [Edwards and Fritz, 1986], whereas other studies have concluded that cellulose recorded water source without influence of humidity [DeNiro and Cooper, 1989]. Roden et al.  provided a model that quantitatively explained the extent to which environmental drivers influenced the hydrogen and oxygen isotopic composition of cellulose.
 While it is evident that both δ13C and δ18O in plant organic matter can be used as interrelated climate proxies, it is not evident how these parameters will vary along the complex precipitation gradients [MacFarlane et al., 2004] that typify the coastal-to-interior gradients found on all continents. This is because oxygen isotope patterns in tree rings need not exhibit a unidirectional pattern if both water source values and humidity exhibit contrasting patterns along the transect. Similarly, carbon isotope values need not exhibit a monotonic relationship with precipitation if offsets in phenology and seasonality of primary growth periods are restricted in their duration at some sites and extended throughout the entire year at other sites.
 This project was conducted along one such climatic gradient, the North American Oregon Transect Ecosystem Research (OTTER [Peterson and Waring, 1994]) from coastal to central Oregon, USA. Along this coastal-to-interior transect, there is a substantial moisture gradient (differences of ∼2500 mm in mean annual precipitation over 250 km). Previous studies along the Oregon transect have established clear patterns relating site water balance to primary productivity [Running, 1994; Runyon et al., 1994; Peterson and Waring, 1994; Anthoni et al., 1999] and to the stable isotopic composition of ecosystem respiration [Bowling et al., 2002, 2003a; McDowell et al., 2004b]. Each site along this transect was dominated by a single species of coniferous tree, but the dominant species of the forest differed across sites (Figure 1).
 Our first objective was to test the hypothesis that both δ13C and δ18O in tree ring cellulose increase with aridity along the precipitation transect owing to the effects of reduced soil moisture and humidity at the dry sites on both reducing stomatal conductance and the ratio of leaf-internal to atmospheric CO2 (ci/ca effects on δ13C [Farquhar et al., 1989]) and in enhancing evaporative enrichment in δ18O of leaf and soil waters. The Oregon transect provides an ideal opportunity to evaluate how much the δ18O in source water and humidity alter the δ18O in tree ring cellulose because of large differences in meteoric water δ18O, humidity and precipitation (Figure 2). The coastal sites have a milder climate than the inland sites, especially during winter months when the majority of precipitation falls. Site differences in precipitation and humidity suggest that evaporative enrichment should differentially modify leaf water δ18O across the transect (Figure 2).
 Our second objective was to determine whether variation in tree ring cellulose δ18O across the transect can be predicted using an existing model developed by Roden et al. . Environmental and isotopic information collected across the Oregon transect was used to field test this mechanistic model that describes the factors affecting the δ18O in organic matter. This model has been tested previously under controlled environmental conditions [Roden and Ehleringer, 1999b; Helliker and Ehleringer, 2002] and field conditions using angiosperms [Roden and Ehleringer, 2000]. This study provides an opportunity to test this model under field conditions using gymnosperms, which are widely used in tree ring studies. Data collected across this transect from other studies [Bowling et al., 2002, 2003a, 2003b; McDowell et al., 2004a, 2004b] were used to test model predictions of tree ring cellulose δ18O values.
 Previous studies [Bowling et al., 2002, 2003a, 2003b; McDowell et al., 2004a, 2004b] utilized these same field sites to study the effects of site water balance and vapor pressure deficits on both the carbon and oxygen isotopic composition of ecosystem respiration (δ13CR and δ18OR respectively). They found that the δ13C of carbon stocks (leaves, fine roots, soil organic matter etc.) as well as δ13CR was correlated with precipitation and vapor pressure deficit through stomatal regulation of gas exchange [McDowell et al., 2004b]. In addition, the site differences in vapor pressure deficits affected δ18OR through evaporative enrichment that modified the primary effects of meteoric water δ18O inputs [Bowling et al., 2003a, 2003b]. This study was performed during the same study period and on the same sites as that of Bowling et al. [2002, 2003a, 2003b] and McDowell et al. [2004a, 2004b] in order to link the tree ring results with existing data on the carbon and oxygen isotopic composition of ecosystem respiration and organic matter, as well as to use their measurements of microclimate and δ18O of ecosystem waters to parameterize the Roden et al.  cellulose model.
2.1. Site Descriptions
 The coastal “spruce site” was dominated by a relatively young stand of even-aged Picea sitchensis (Table 1). The “Douglas-fir site” in the Tum Tum tree farm located in the central Oregon Coast Range was also a young stand of even-aged Pseudotsuga menziesii. The “pine site” in the Metolius Research Natural Area on the eastern side of the Cascade mountain crest was dominated by Pinus ponderosa of varying ages. The pine site is an AmeriFlux site, where eddy fluxes and soil and plant process rates were also measured and used to interpret the Bowling et al. [2002, 2003a, 2003b] and McDowell et al. [2004a, 2004b] studies. The “juniper site” in central Oregon was dominated by stands of Juniperus occidentalis of varying ages (Table 1). These four sites were a subset of the six used by Bowling et al. [2002, 2003a, Figure 1] and McDowell [2004a, 2004b].
Sites correspond to those of Bowling et al.  and differ from sites on the original Oregon transect. The pine site is an Ameriflux long-term CO2 flux study site (Metolius Research Natural Area).
Picea sitchensis Tsuga heterophylla
44°07′N, 124°07′W, 300 m
44°35′N, 123°35′W, 290 m
44°30′N, 121°37′W, 941 m
44°18′N, 121°20′W, 930 m
2.2. Sample Processing
 In November of 2001, six healthy-looking trees were selected in the proximity of the air sampling towers at each site [see Bowling et al., 2002]. Four cores (12 mm diameter) were obtained from the four cardinal directions at breast height of each tree. Owing to small growth rings for juniper trees, six cores were obtained for each tree at this site to provide sufficient material for analysis. Since the spruce and Douglas-fir sites were plantations, the variability in ages and growth rates between trees at these sites was minimal. The ages in the pine and juniper sites varied substantially more, and trees were selected if they were either a canopy dominant or subdominant, but not if they were very large (and old). Very old trees tend to have narrow tree rings [Fritts, 1976] making it difficult to obtain enough material when subdividing the annual growth (i.e., latewood and earlywood).
 The last two years of growth (2000, 2001) were subdivided into earlywood and latewood on the basis of position and visual assessments of wood density (under a 20x dissecting microscope). The latewood portion of the ring was distinctly darker than earlywood in these conifers. A portion of the ring between the subdivisions was discarded (>1/3 of the ring) to ensure distinctiveness of sub-sampling. The amount of sample obtained from each tree ring varied with width. The spruce and Douglas-fir sites had trees with particularly large growth increments (>10 mm) and some juniper rings were less than 0.5 mm in width. The samples from the four cardinal directions on a single tree were pooled and cut into small pieces with a razor blade. The samples were dried at 70°C for 48 hours and ground to a fine powder with a ball mill (Wig-L-Bug, Crescent, Elgin, Illinois).
 Approximately 50–100 mg (if available, some juniper samples were as small as 15 mg) of ground sample was loaded into fiber filter bags (ANKOM, Fairport, New York) and heat-sealed for cellulose extraction as described in detail by Leavitt and Danzer . Briefly, the filter bags were placed in a Soxhlet apparatus to reflux a 2:1 solution of toluene:ethanol for a 24-hour period followed by a period of drying and another 24-hour period of extraction (for lipids and resins) with 95% ethanol. The bags were dried and then boiled in water for 1 hour to extract soluble sugars and low molecular weight polysaccharides. The samples were then “bleached” using a strong sodium chlorite/acetic acid solution that was periodically replaced over a 4-day extraction (to extract lignin and other N containing compounds). Following these treatments the extracted material was in the form of holocellulose. To obtain pure α-cellulose the sample was exposed to a strong (17% w/v) NaOH solution followed by an acetic acid solution to neutralize the pH with each step followed by extensive rinsing with double distilled water. The α-cellulose was dried at 70°C for 48 hours.
2.3. Analysis of Stable Carbon and Oxygen Isotope Ratios
 For δ18O measurements, 90–110 μg of α-cellulose was loaded into silver capsules and converted to CO by pyrolysis [Saurer et al., 1998] in a hot (1400°C) alumina/glassy carbon reactor (Thermo-Finnigan TC/EA) and separated from other gases in a 0.6-m molecular sieve 5A gas chromatography (GC) column connected to a Finnigan MAT deltaPlus XL isotope ratio mass spectrometer. Mass spectrometry was performed at the Stable Isotope Ratio Facility for Environmental Research at the University of Utah. For δ13C measurements, approximately 1 mg of α-cellulose was loaded into tin capsules and combusted on an elemental analyzer (Carlo-Erba 1110, Milan) coupled to a Finnigan MAT delta Plus isotope ratio mass spectrometer. The precision of α-cellulose standards run with the samples for δ18O was 0.23‰ (standard deviation, n = 36) and for δ13C was 0.16‰ (n = 13).
2.4. Leaf, Stem, and Atmospheric Water Sampling
 To obtain source water values from trees at each site, stems between 5 and 10 mm in diameter and 5- to 7-cm long [see Bowling et al., 2003a], were collected and stored in screw top vials covered in wax film, and refrigerated or frozen until the water could be extracted by cryogenic distillation under vacuum [Ehleringer et al., 2000]. At each site, stems were sampled approximately 4 times during the growing season (April to September) in both 2000 and 2001. During August 2001, sun and shade needles were sampled in the morning and mid-day over a 2-day period at each site. Needles were stored, frozen and extracted for water in the same manner as stems. Water vapor was collected cryogenically at each site during the August 2001 field campaign using the method of Helliker et al. . All water samples were analyzed for δ18O as described by Fessenden et al. .
2.5. Micrometeorological Measurements
 At each site during 2001, measurements of air temperature and relative humidity (HMP45A, Vaisala, Inc. Woburn, Massachusetts) at the top of the canopy were measured every 5 s and stored as hourly averages with a data logger (CR23X, Campbell Scientific, Inc. Logan, Utah).
 In our analysis we used both the mean environmental values (split between early season (April–June) and late season (July–September)) and the range of possible modeled outcomes of cellulose δ18O (split between earlywood and latewood) based on the range of measured environmental values. Only daytime values of temperature and humidity were used in the model since that was considered the period of active photosynthesis. Although model predictions have no variance, the input parameters can vary widely and so we used the range of possible predictions as a way to provide some estimate of model output sensitivity.
3.1. Carbon Isotopes
 The δ13C values of tree ring cellulose were more negative (depleted in 13C) at sites with consistently greater precipitation (Figure 3). There were no significant differences (p > 0.05) between samples collected in 2000 and 2001 and so the data for both years were pooled in Figure 3 (and subsequent figures). The similarity in δ13C for these 2 years was reasonable since environmental conditions (air temperature and precipitation) were similar for both 2000 and 2001 at locations near each site (data (not shown) from nearby weather stations). Differences between sites were highly significant (p < 0.05, ANOVA using a Tukey multiple comparison test, SPSS Inc.) with the exception of the earlywood δ13C values for the pine and Douglas-fir sites. Data for earlywood δ13C were significantly different (more depleted in 13C) than latewood at the pine and juniper sites only. The coastal spruce site and the inland juniper sites differed in latewood cellulose δ13C by over 4.4‰.
 The δ13C of fine root organic matter, tree ring cellulose and ecosystem respiration were correlated with δ13C of sun needles (Figure 4, organic matter and respiration data from Bowling et al. ). Latewood cellulose was consistently more enriched in 13C than other components of tree organic matter or respiration (Figure 4). All components of tree organic matter δ13C followed the same general trends with site water status, with more enriched δ13C observed at the drier sites.
3.2. Oxygen Isotopes
 There were no significant differences (p > 0.05) in cellulose δ18O between samples collected in 2000 and 2001 and so the data for both years were pooled in Figure 5. Tree ring cellulose δ18O differed by as much as 2‰ between sites. Differences in latewood δ18O between sites were significantly different (p < 0.05) with the exception of the spruce and juniper site comparisons. Very few site differences were significant for earlywood δ18O except when compared to the pine site. Although earlywood tended to be more enriched in δ18O than latewood the differences were rarely significant. There were no clear trends in tree ring cellulose δ18O with differences in site water availability.
 The lack of trends between sites in terms of δ18O is odd considering hypothetical expectations. At less humid sites stomatal conductance should be reduced, causing a decrease in ci/ca and the observed enrichment in 13C (Figure 3), but there should also be enhanced evaporative enrichment in 18O for leaf water. However, differences in source water δ18O (measured as stem water) will also affect cellulose δ18O values. Expressing oxygen isotope ratios as deviations from stem water can clarify the effects of evaporative enrichment (δ18Ocellulose − δ18Ostem water). A plot of δ18Ocellulose − δ18Ostem water versus δ13Ccellulose (Figure 6) provides a way to relate the effects of humidity and water deficit on fractionation for both δ18O and δ13C. Whereas the juniper and spruce sites were indistinguishable in terms of cellulose δ18O, the deviations in δ18Ostem water at these two sites were distinct (Figure 6). All sites were significantly different (p < 0.05) in terms of δ18O deviation.
 Site comparisons also benefit from using only a single portion of the ring. The pine and Douglas-fir sites have substantially more overlap on a δ18Ocellulose − δ18Ostem water versus δ13Ccellulose graph of earlywood alone or combined data (data not shown) than latewood alone (Figure 6). The range in δ18Ocellulose − δ18Ostem water between sites (10‰) was similar to site differences in the oxygen isotopic composition of ecosystem respiration (δ18OR) estimated at the same sites and at the same time as the tree rings sampled in this study were produced (2000 and 2001; refer to Bowling et al. [2003a] for δ18OR information) in contrast to minimal site variation in cellulose δ18O (2‰, Figure 5).
3.3. Model Predictions Regarding Cellulose δ18O
 One of the striking features of this data set is the similarity in δ18O values from sites with very different moisture availability (e.g., spruce and juniper sites). We collected environmental information to parameterize a mechanistic model [Roden et al., 2000] to determine if this model was able to explain these observed tree ring δ18O values. Daytime meteorological data used for modeling cellulose δ18O for each site are presented in Figure 7. During the growing season, coastal sites were cooler and more humid than inland sites. Pine and juniper sites had similarly high vapor pressure deficits. The four sites had distinct source water δ18O values (Figure 8). Mean stem water δ18O values over an entire year were −6.0, −7.8, −13.6 and −10.8‰ (standard deviations ranged from 0.9 to 1.3‰) for the spruce, Douglas-fir, pine and juniper sites respectively. The pattern of source water variation relate to both the proximity to the ocean and subsequent rainout effects (a Rayleigh process) and temperature during precipitation events associated with elevation. Stem water δ18O was more negative at the pine site than the juniper site, despite their relative distance from the coast (Figures 1 and 2). Atmospheric water vapor δ18O was measured during the August 2001 field campaign. The mean atmospheric vapor δ18O values measured regularly 12 times over a 24-hour period were −14.4, −14.1, −15.1 and −15.4‰ (standard deviations ranged from 0.8 to 1.2‰) for the spruce, Douglas-fir, pine and juniper sites respectively. Estimates of leaf diffusive conductance (stomatal + boundary layer) were derived from conductance/vapor pressure deficit relationships for coniferous trees [Sanford and Jarvis, 1986; Grieu et al., 1988; Kolb and Stone, 2000].
 The cellulose model accurately predicted that tree ring δ18O should be less variable between sites (≈2‰, Figure 5) than between-site variation in either stem water (≈8‰, Figure 8) or leaf water (≈23‰, Figure 9b). The model underestimated both earlywood and latewood tree ring δ18O for most of the sites except the juniper site (Figure 9a). However, the range of measured values and the range of model predictions generally overlapped. Bulk leaf water δ18O was only measured during the August 2001 field campaign so no comparisons were possible with early season model predictions. The leaf water model [Flanagan et al., 1991] predicted bulk leaf water quite well (when corrected for the fraction of unevaporated water) except for leaves at the juniper site. The range of possible leaf water values modeled was quite large (as much as 11‰) depending on the environmental parameters input, with humidity being the variable that had the largest impact on model outcomes.
 One issue with transects that span substantial distance is species turnover and inter-specific differences in physiology and function. Although surveys of tree ring δ13C in various species of conifer can encompass the range of δ13C found in this study (>8‰ [see Warren et al., 2001]), comparisons are more valuable under common environmental conditions. Marshall and Monserud  found no significant difference in tree ring δ13C for three co-occurring conifers (Pseudotsuga menziesii, Pinus ponderosa and Pinus monticola). Carbon isotope discrimination values were similar between a juniper (Juniperus osteosperma) and pine (Pinus edulis) species growing in similar environments [Williams and Ehleringer, 1996]. Panek and Waring  measured tree ring δ13C of a single species (Douglas-fir) over a moisture gradient similar in magnitude to this study (>2000 mm). They observed significant site differences in δ13C (>4‰) with drier sites more enriched in 13C associated with humidity deficit. Flanagan et al.  showed that carbon isotope discrimination was correlated with life form and that wood δ13C was similar for two conifer species (Picea mariana and Pinus banksiana) growing at the same boreal forest sites. Although similar life forms (conifers only) were sampled in this study, we still cannot partition all differences in δ13C between sites as environmental.
 There was a striking lack of difference in cellulose δ13C between the Douglas-fir and pine sites despite large difference in precipitation (Figure 3). Previous studies have established that ponderosa pine at the Metolius site access groundwater [James et al., 2000; Anthoni et al., 1999; McDowell et al., 2004b]. Thus precipitation inputs at the pine site may not accurately reflect tree water status. The Douglas-fir site was a commercial plantation and was fertilized for stemwood production. Increased nitrogen inputs can increase δ13C values owing to increased photosynthetic capacity [Livingston et al., 1999; MacFarlane et al., 2004]. The similarity in tree ring δ13C for trees growing at the pine and Douglas-fir sites may be the result of both of these factors.
 This study was carried out concurrently with other projects [Bowling et al., 2002, 2003a, 2003b; McDowell et al., 2004a, 2004b] that describe the effect of water status on the variability in δ13C and δ18O of ecosystem respiration. Bowling et al.  and McDowell et al. [2004b] found that sites with higher precipitation had more negative δ13C values in plant and organic matter and that isotopic composition of respired CO2 (δ13CR) followed the same trends. Pataki et al.  extended this result across a wide range of biomes. The δ13C of tree ring cellulose, fine roots and ecosystem respiration were all strongly correlated with sun leaf δ13C (Figure 4) indicating that current photosynthate may be the predominant substrate for respiratory flux within these systems. Pataki et al.  also found that sun leaf δ13C was correlated with the δ13C of ecosystem respiration. Although the δ13C of tree ring cellulose was more enriched by about 3‰ as compared to sun needles and fine roots, similar patterns across the OTTER transect were observed and the difference in δ13C from the wettest to driest sites was comparable (4 to 5‰). These results highlight the importance of timescales of carbon storage in different ecosystem pools and their potential influence on atmospheric fluxes. In general, drier sites exhibited more enriched δ13C in plant tissues, soil organic matter, ecosystem respiration [Bowling et al., 2002; McDowell et al., 2004b] and tree ring cellulose (this study [see also Panek and Waring, 1997]) than wetter sites. In our results, the cellulose δ13C values were less variable than ecosystem δ13CR observations, which is likely due to temporal integration of isotopic inputs in organic matter. The tree ring data support the conclusions of Bowling et al.  that differences in photosynthetic carbon isotope discrimination across the OTTER transect cause the observed variation in δ13C of carbon stocks and fluxes.
 Few studies have linked the oxygen isotopic composition of tree ring cellulose with moisture or humidity. There were no distinct trends in δ18O values of tree ring cellulose across the OTTER transect. Although meteoric and stem water δ18O were more depleted at the arid sites than the wet sites (owing to rainout and temperature effects), evaporative enrichment associated with vapor pressure deficits overshadowed the δ18O inputs from precipitation. In contrast to other studies [Burk and Stuiver, 1981; Roden and Ehleringer, 2000; Barbour et al., 2001], there was a significant negative relationship (p < 0.05) between source water δ18O and cellulose δ18O. In contrast to our study, Saurer  and Saurer et al.  found that the δ18O in tree rings was positively correlated with precipitation amount as well as precipitation δ18O. In their study (across the northern tree line in Eurasia encompassing 150° of longitude), differences in tree ring δ18O were related to variation in meteoric water δ18O inputs associated with continental temperature variation [Saurer et al., 2002]. Since precipitation amounts also decreased with continentality [Saurer, 2003], the positive correlation of tree ring δ18O and precipitation likely resulted from the covariance of precipitation amount and meteoric water δ18O. A transect across the northern tree line may present minimal variation in vapor pressure deficit between sites, reducing the impact of humidity in modifying the relationship between source water δ18O and tree ring δ18O (in contrast to our results across the OTTER transect).
 Two questions arise regarding tree ring δ18O observations from the pine and juniper sites. First, why was the stem water more depleted at the pine site than the juniper site? A trend of enriched cellulose δ18O with aridity might have been stronger had the pine site had less negative source water δ18O. As stated above, the pine at the Metolius site had access to groundwater. This groundwater has a different isotopic composition from that expected from seasonal precipitation inputs. At no other site does it appear that trees have access to groundwater. Thus trees at the Metolius site are accessing a water source that does not fully reflect the expected inputs, leading to a more negative water source than available to juniper trees. Second, why did the δ18O results indicate less evaporative enrichment for trees at the most arid site (juniper)? The δ18O deviations from source water for juniper trees were much lower than for pine trees (Figure 6). We speculate that trees at the juniper site were active early in the growing season when soil moisture was high owing to snowmelt and temperatures were cool. This would lead to δ18O values that indicate less evaporative enrichment than those in late summer. By later in the season the trees may be far less active owing to very dry soils. It is unknown when latewood is laid down for this species, but it could well represent a much different time of year than assumed for other species. Some studies [Moore et al., 1999; Law and Waring, 1994; Leffler et al., 2002] have demonstrated that water availability can severely limit leaf level gas exchange of juniper species, creating seasonal variation in carbon assimilation and transpiration. However, this interpretation must be tentative since the latewood δ13C values for juniper trees do not likewise indicate reduced ci/ca values that would be indicative of reduced stomatal conductance.
 Our results do not support the prediction that tree ring cellulose δ18O will necessarily increase with aridity because no significant correlation between precipitation and cellulose δ18O was observed (Figure 5). However, when δ18O values of cellulose are plotted as deviations from stem water δ18O (Figure 6) the effects of water availability and humidity became apparent. Bowling et al. [2003a] measured the oxygen isotopes of CO2 respired (δ18OR) by these same ecosystems and found more enriched δ18OR values at the drier sites, indicating that evaporative enrichment associated with site vapor pressure deficits overshadowed the precipitation δ18O input as a first order control on the δ18O of ecosystem respiration. Our cellulose δ18O data further confirm the importance of evaporative enrichment at these sites.
 Using δ18O alone to infer site differences in water status may be problematic as shown in this study. Tree ring cellulose δ18O for the juniper and spruce sites were indistinguishable, yet those sites differed dramatically in terms of water status and potential for evaporative enrichment. The similarities across sites are a result of counteracting influences of source water δ18O and evaporative demand. However, using cellulose δ18O deviations from stem water δ18O captures the evaporative effect and allows for partitioning between sites (Figure 6). However, to derive useful climate information from ancient tree ring records, an independent estimate of stem water δ18O would be required to calculate δ18Ocellulose − δ18Ostem water from cellulose data, which is seldom available. Using a second isotope (δ13C) that is also influenced by vapor pressure deficit may allow better interpretation of δ18O values recorded in tree rings. Each isotope provides different though related information regarding water status, and together, they provide a powerful data set to probe tree water status and its impact on carbon fluxes.
 Mechanistic models that predict cellulose δ18O from environmental parameters [Sternberg et al., 1986; Yakir and DeNiro, 1990; Luo and Sternberg, 1992; Roden et al., 2000] have been successfully field tested on sites that differed in source water δ18O, humidity and temperature [Roden and Ehleringer, 2000]. The cellulose model used in the present study tended to underestimated δ18O in tree ring cellulose (except for juniper trees), although some of the predictions based on mean environmental information were excellent. Our cellulose model was robust in predicting that differences in tree ring δ18O between sites would be smaller than the differences between meteoric waters or leaf waters across the transect. Clearly, site differences, such as vapor pressure deficit, play a substantial role in modifying the primary δ18O input of source water recorded in cellulose δ18O.
 Average environmental information may not necessarily be representative of conditions when photosynthesis is occurring and when cellulose is being constructed. Even though annual rings were subsampled, the slices represented an integrated sample over many weeks of metabolism. For example, the spruce and Douglas-fir sites experience high humidities and large amounts of precipitation, however, owing to low light intensities, very cloudy periods may be less important for photosynthesis than sunny and warm days. Periods of bright sunlight likely correspond to periods of lower humidity and thus greater evaporative enrichment in leaf water δ18O. Therefore mean environmental data may skew model predictions to periods of less photosynthetic activity and less evaporative enrichment. Our results support this hypothesis because the model underestimated spruce and Douglas-fir cellulose δ18O (Figure 9). The ranges of possible outcomes (the error bars in Figure 9) were based on the range of measured environmental inputs and tended to overlap with observed values of cellulose δ18O. If sunny periods are more critical for carbon gain at the wet sites then the upper range of model predictions may better represent the true environmental influence on cellulose δ18O.
 Model predictions of bulk leaf water δ18O [Craig and Gordon, 1965; Flanagan et al., 1991; Roden and Ehleringer, 1999a] were reasonably similar to observed values. This may imply that under-estimations of tree ring cellulose δ18O were more a function of the cellulose model rather than the leaf water model. However, the leaf water observations were made during one field campaign in August of 2001, which may not necessarily be representative of integrated leaf water δ18O over the entire summer period. No leaf waters were sampled during spring and, owing to limited sampling, broad conclusions regarding how well the leaf water model fits the data are tenuous at best. The Craig and Gordon  [see also Flanagan et al., 1991] model for evaporative enrichment assumes steady state conditions which may not apply during daily temperature and humidity fluctuations and thus a dynamic model may be more applicable [Cernusak et al., 2002]. Unfortunately, the data to parameterize the Cernusak et al.  leaf water model were unavailable (leaf water concentration, mole fraction of water vapor in the leaf intercellular spaces, and transpiration rates). In addition, using a dynamic model may not be critical since Cernusak et al.  found that steady state models gave reasonable predictions during the daytime periods (when photosynthesis is occurring) and that long-lived, perennial plants may not be as influenced by diurnal fluctuations as crop plants. One interesting feature is how far off the model predictions for juniper leaves were compared to the measured August leaf water δ18O values. Even with the depleted late-season leaf water δ18O values predicted by the leaf water model, the cellulose model overestimated juniper latewood δ18O by over 2‰. When the measured August leaf water δ18O values were used instead, the model overestimated latewood cellulose δ18O by 7‰. Although more leaf water sampling would be required to make any conclusions, this does support the idea, stated above, that latewood cellulose for juniper may be laid down much earlier in the season. August conditions may have been so dry that little or no wood was produced and the trees were simply involved in maintenance metabolism. At the other arid site (pine), latewood predictions of cellulose δ18O were much better than for the juniper site. Since the pine trees on this site may have access to groundwater [Anthoni et al., 1999; McDowell et al., 2004b], late season environmental information may be more relevant for modeling leaf water and latewood cellulose δ18O values than for juniper trees (Figure 9). Law and Waring  found that drought and vapor pressure deficit were equally limiting to carbon uptake in juniper, while vapor pressure deficit had a larger effect than drought on seasonally integrated carbon uptake in pine. Model predictions would also be enhanced if reasonable assumptions could be made regarding what time of the year each portion of the ring is constructed and what carbon sources are utilized (stored versus current). For example, if we assume that latewood cellulose for juniper trees was produced from carbon sources produced earlier in the growing season, then model predictions would better match observations (data not shown). More research is needed that can provide information about tree ring construction for a variety of species.
 Isotope models are unlikely to be useful for climate reconstruction of ancient tree ring records since the variables needed to parameterize these models would seldom be available. However, as we test and confirm, using field-based and controlled experiments, that these models are robust in their predictive ability we learn more about the controls of δ18O variation in tree ring records, which will enhance our interpretation of those records. This study demonstrates that useful information on site water balance can be gleaned from tree ring cellulose δ13C and δ18O. Thus climate change events that cause shifts in site water balance could be studied by using long-term tree ring records. Many other studies [Leavitt and Long, 1991; Loader and Switsur, 1995; Robertson et al., 1997; Anderson et al., 1998; Saurer, 2003] have also demonstrated that stable isotope dendrochronology can be a powerful tool for climate analysis and reconstruction.
5. Conclusions and Implications
 Carbon isotopic composition of tree ring cellulose increased with aridity along the precipitation transect, supporting our hypothesis that differences in soil moisture and vapor pressure deficits between sites modify stomatal conductance, gas exchange and the δ13C composition of organic matter. The δ13C in tree ring cellulose varied in a similar pattern to ecosystem respiration δ13C. Minimal differences in tree ring δ18O across sites indicate that site differences in source water δ18O were overshadowed by vapor pressure deficit effects on evaporative enrichment. Thus historical records of cellulose δ18O may provide useful proxy information regarding humidity and site water balance especially if combined with the complementary information located in the δ13C of the same sample and if reasonable assumptions regarding source water δ18O values can be made. A mechanistic model was capable of predicting the δ18O in tree ring cellulose (to a first approximation) in field grown trees. These observations and model predictions demonstrate that environmental influences on evaporative enrichment can have profound effects on cellulose δ18O even across a transect with substantial site differences in source water δ18O.
 We thank Bev Law for meteorological data and access to the Ameriflux site at the Metolius Research Natural Area and Starker Forests Inc., for access to the Douglas-fir plantation. We thank Cameron Parker, Corey Weber, Julia Morales, Justin Soares, Claire Lunch, Mike Lott, Craig Cook, Alice May, and S. B. Bill who provided technical assistance. This study was supported by the United States Department of Agriculture (grant 99-35101-7772).