Tree-ring cellulose δ18O records similar large-scale climate influences as precipitation δ18O in the Northwest Territories of Canada

Stable oxygen isotopes measured in tree rings are useful for reconstructing climate variability and explaining changes in physiological processes occurring in forests, complementing other tree-ring parameters such as ring width. Here, we analyzed the relationships between different climate parameters and annually resolved tree-ring δ18O records (δ18OTR) from white spruce (Picea glauca [Moench]Voss) trees located near Tungsten (Northwest Territories, Canada) and used the NASA GISS ModelE2 isotopically-equipped general circulation model (GCM) to better interpret the observed relationships. We found that the δ18OTR series were primarily related to temperature variations in spring and summer, likely through temperature effects on the precipitation δ18O in spring, and evaporative enrichment at leaf level in summer. The GCM simulations showed significant positive relationships between modelled precipitation δ18O over the study region and surface temperature and geopotential height over northwestern North America, but of stronger magnitudes during fall-winter than during spring–summer. The modelled precipitation δ18O was only significantly associated with moisture transport during the fall-winter season. The δ18OTR showed similar correlation patterns to modelled precipitation δ18O only during spring–summer when water matters more for trees, with significant positive correlations with surface temperature and geopotential height, but no correlations with moisture transport. Overall, the δ18OTR records for northwestern Canada reflect the same significant large-scale climate patterns as precipitation δ18O for spring–summer, and therefore have potential for reconstructing past atmospheric dynamics in addition to temperature variability in the region.


Introduction
Tree rings have been used to reconstruct climate, particularly temperature, over northwestern North America prior to the instrumental period, primarily using tree-ring width (TRW) and maximum latewood density (MXD) data for the past millennium (Anchukaitis et al. 2012;Briffa et al. 2004;D'Arrigo et al. 2014). Such records have also been used to generate indices of patterns of large-scale atmosphericocean circulation, such as the Aleutian Low, or the Pacific Decadal Oscillation (PDO) for the north Pacific sector (e.g. D' Arrigo et al. 2001;Gaglioti et al. 2019;Villalba et al. 2011). Tree-ring density proxies such as MXD have been shown to have a more stable and robust temperature signal than ring-width chronologies from the same trees, and thus have been used to generate temperature reconstructions for a variety of northern sites, e.g. in British Columbia (Wilson and Luckman 2003), at latitudinal treeline at Firth River in Alaska (Anchukaitis et al. 2012;Andreu-Hayles et al. 2011a) and in the Yukon (Morimoto 2015), as well as for the Northern Hemisphere Wilson et al. 2016). Blue intensity (BI), a novel proxy for density, has been used to produce reconstructions in Yukon (Wilson et al. 2019) and the Gulf of Alaska (Wilson et al. 2017), among other locations.
The isotopic composition of stable oxygen (δ 18 O, ratio of 18 O-16 O relative to a standard) measured in cellulose from tree rings can be used as another climate proxy and can provide complementary and unique information relative to TRW and MXD/BI data. This isotopic information includes, for example, physiological insights into tree response to environmental changes in boreal and other terrestrial ecosystems (e.g. Andreu-Hayles et al. 2011b; Barber et al. 2000;Levesque et al. 2017), information about the source water used by the tree (e.g. Gessler et al. 2014;McCarroll and Loader 2004), and climate variability (e.g. Andreu-Hayles et al. 2017;Gennaretti et al. 2017). The δ 18 O signature recorded in tree rings mostly results from (1) the isotopic composition of the source water that is taken up by the roots; (2) the isotopic enrichment occurring due to leaf transpiration; and (3) the isotopic exchange of oxygen atoms between cellulose and xylem water when cellulose is formed. The first and the third contributions are related to the water source signal of precipitation δ 18 O and isotopic balance in the soil (Dansgaard 1964). Precipitation δ 18 O can vary regarding the trajectory of the air masses, the distance from the original source and their exposure to warmer/colder atmospheric conditions that will determine the amount of moisture that can be held and the number of rainouts (i.e. depleting the original δ 18 O signature) before arriving to the studied trees. The δ 18 O of source water can also vary due to the use of water pools from different soil depths (Barbeta et al. 2020;Brinkmann et al. 2018). The second contribution, the enrichment due to leaf transpiration, is associated to physiological response of the plant to changes in relative humidity and temperature, both determining vapor pressure deficit (VPD). Finally, post photosynthetic fractionation can also occur (Gessler et al. 2014) modulating the recorded δ 18 O signal in tree rings. Although the mechanisms described above are well accepted, there are strengths and caveats on their physiological interpretation (Barbour 2007;Cernusak et al. 2016;Gessler et al. 2014). Disentangling the dominant signal in cellulose δ 18 O, leaf water enrichment or source water can be challenging.
Because tree-ring δ 18 O (δ 18 O TR ) records are affected by climate variability, they can be a powerful additional proxy for reconstructing atmospheric circulation patterns for centuries prior to the instrumental period (Balting et al. 2021;Szejner et al. 2016). Understanding the linkages between δ 18 O TR and precipitation δ 18 O is also important for improving past climate reconstructions in these high-latitude boreal regions D'Arrigo et al. 2014;Wilson et al. 2016). In the extratropical regions of the Northern Hemisphere, precipitation δ 18 O is strongly related to temperature (Birks and Edwards 2009). This relationship is in part reflected in the positive correlations typically seen between δ 18 O TR and temperature, and in theory may be attributable to large-scale atmospheric circulation patterns that prevail in this area. For example, summer temperatures were reconstructed over the last millennium using δ 18 O TR records (Naulier et al. 2015), and annual temperatures and δ 18 O meteoric water values were estimated from Pleistocene subfossil wood from Bylot Island, Canada (Csank et al. 2013). In the southern Yukon, an atmospheric general circulation model equipped with stable water isotope tracers demonstrated that high δ 18 O values in meteoric water were associated with an intensified Aleutian Low pressure cell, bringing stronger southerly moisture flow to eastern Alaska and the southern Yukon (Field et al. 2010). Such general circulation models can provide an idealized picture of the climatic influence on local precipitation δ 18 O (Field et al. 2010;Porter et al. 2014) in the absence of long-term precipitation δ 18 O records.
Determination of the climate signal in δ 18 O TR requires comparisons with observed climate variables, typically obtained from nearby meteorological stations. In prior work, Begin et al. (2015) and Naulier et al. (2015) identified summer maximum temperature and VPD influences on black spruce δ 18 O TR for a site in north-central Quebec using weather station data from three stations 100-300 km away, each with data available for roughly 50 years. Holzkamper et al. (2012) reported a robust relationship between spring temperature and precipitation with white spruce δ 18 O at a site in Nunavut over the 1986-2004 interval for a weather station roughly 300 km away. Csank et al. (2016) documented spring-summer climatic controls on δ 18 O using Global Historical Climate Network (GHCN) stations within 100 km of sampling sites in south-coastal Alaska between 1949 and 2011.
Station records are typically few, very limited across space and have short or incomplete records in remote regions such as those studied herein (e.g. Holzkamper et al. 2012), making it difficult to identify robust local or large-scale climatic influences. Gridded observational products and meteorological reanalysis are, in theory, an alternative, and can also help to identify regional influences on tree-ring signals. For example, gridded climate products have been used for reconstructing summer temperatures (Gennaretti et al. 2017) and streamflow (Brigode et al. 2016) in northern Quebec.
Here, our objective is to assess the climate signal and atmospheric circulation patterns associated with inter-annual variations in a newly-developed alpha-cellulose-derived δ 18 O TR chronology for a site located in the Northwest Territories, Canada, and thus determine the potential of these records to reconstruct large-scale climate variability in the region. We focus specifically on how these relationships are detected in several different types of datasets, namely: (i) homogenized station records and 'raw' station records with additional parameters, (ii) two gridded temperature datasets estimated from meteorological station data but using different interpolation techniques, and (iii) two meteorological reanalyses. We also use an isotopically-equipped general circulation model (GCM) to understand climatic controls on local precipitation δ 18 O, and to help interpret the different seasonal relationships identified between δ 18 O TR and the climate observations. Overall, we aimed to determine the potential of using tree-ring δ 18 O to reconstruct large-scale climate indicators interpreting what climatic signals might be detectable using a broad range of data and GCM simulations. Our interest in this region is motivated by the need to provide context prior to the instrumental period for north Pacific climate variability and high-latitude climate change.

Tree-ring data
The tree-ring samples were collected from white spruce (Picea glauca [Moench]Voss) located at 1145 m a.s.l near Tungsten (61.98° N; − 128 to 25° W; Fig. 1), Northwest Territories (NWT), Canada in the year 2003 (Morimoto 2015). The sampled stand consisted of isolated tall spruce trees growing from an underbrush of willow (Salix spp.) and alder (Alnus spp.) on an irregular, 10-15° north-east facing slope about 100 m below the contiguous treeline. A total of 26 tree-ring samples (5 mm-cores) from 25 trees were selected from the "Western Collection", a tree-ring data set that was The samples were scanned at a resolution of 3200 dpi using a color calibrated Epson V850 Pro scanner and the Sil-verFast Ai IT8 imaging software (Version 8) and the TRW were measured using the software Coorecorder 9.3 (Cybis Electronik 2019). Ring width was measured to 0.001 mm (0.0038px) precision and cross-dated against the original chronology (Morimoto 2015) to ensure accurate calendar dating using dendrochronological methods (Stokes and Smiley 1968). The 26 individual ring-width timeseries were standardized using a 200-year spline (Cook and Peters 1981) after applying a power transformation to stabilize the variance (Cook and Peters 1997). An autoregressive model was then applied to the individual standardized ring-width series to create residual ring-width timeseries. These were averaged using a robust mean with the software Arstan (Cook and Kairiukstis 1990) resulting in TRW residual chronology that emphasizes high-frequency variability.
The δ 18 O TR records were generated at LDEO following the procedures described in Andreu-Hayles et al. (2019) for cellulose extraction and the measurement of δ 18 O using hightemperature pyrolysis in a High Temperature Conversion Elemental Analyzer (TC/EA) coupled to a Thermo Delta plus mass spectrometer. Five trees were analyzed from 1900 to 2003, a period that overlaps partially with the climate data. We selected one core sample from five individual trees mostly based on the following criteria: (1) high correlations with the master TRW chronology to be sure that they were representative of the stand; (2) trees older than 200 years to avoid potential juvenile effects; (3) visually adequate samples for wood preparation (e.g. wide rings for cutting, no locally absent rings or signs of reaction wood). Each ring was separated under a stereomicroscope using a scalpel and was analyzed individually. The resulting annual timeseries from the five individual trees were normalized and the resulting z-scores were averaged to compute a mean chronology (Fig. 2a). The Expressed Population Signal (EPS) metric was also calculated as a metric of the level of agreement among the individual trees. An EPS value exceeding the widely used threshold value of 0.85 (Wigley et al. 1984) indicates a high level of agreement among trees.

Climate data and the NASA ModelE2
isotopically-equipped climate model We used climate data from different sources, including individual station records nearest to the study site and gridded products, each constructed differently, and which allow us to identify regional relationships between climate parameters and the tree-ring data beyond what can be detected for a single weather station. The gridded products were:  (Rohde et al. 2013) is a gridded product based on different station data going back to 1850, also including GHCN. The underlying data are subject to sophisticated quality control and cross-checking and there are separate estimates of mean daily maximum temperature (TMAX), daily minimum temperature (TMIN) and daily average temperature (TAVG) estimates. The BEST temperature fields are smoother than GISTEMP because of broader spatial interpolation over regions of missing station data. 5. UDEL precipitation: The University of Delaware global gridded precipitation product (Legates and Willmott 1990) is a spatially interpolated dataset derived from various sources of gauge data, starting with the GHCN and supplemented from other sources where GHCN data are sparse. The version 3.01 version used here is described at http:// clima te. geog. udel. edu/ ~clima te/ html_ pages/ Globa l2011/ README. Globa lTsP2 011. html 6. Atmospheric Reanalyses: Reanalyses products provide a complete estimate of the state of the atmosphere by combining a numerical weather prediction model and observations from different sources. This allows us to examine metrics other than surface variables such as large-scale circulation features. In our case, we examine relationships with horizontal moisture flux, defined as the product of specific humidity (q) and the vector wind field < u,v > in the mid troposphere to identify possible source water pathways, sea-level pressure, and geopotential height (Z) in the mid-troposphere to identify possible large-scale circulation influences. Reanalyses are less suitable for analyzing relationships between local climate and tree-ring records but are the only practical means of identifying large-scale circulation influences. We used two reanalysis products to guard against product-specific interpretation of our analysis. The National Center for Environmental Prediction / National Center for Atmospheric Research reanalysis (NCEP/NCAR, Kalnay et al. 1996) is a mature, coarseresolution reanalysis going back to 1948, providing coverage for approximately half the tree-ring record, and which assimilates a broad range of surface, upper air and satellite data. For comparison, we also used the Twentieth Century Reanalysis System version 3 product (20CRv3, Slivinski et al. 2019). 20CRv3 provides coverage for the entire tree-ring record but is constrained only by surface pressure observations.
The ModelE2 GCM (Schmidt et al. 2014) is one of several GCMs equipped with stable water isotope tracers. The simulations are forced by observed, interannually-varying Sea Surface Temperatures (SSTs). Model output can be used to identify idealized climate controls on the isotopic composition of precipitation δ 18 O over a region of interest and examine idealized relationships between climate patterns and precipitation δ 18 O for seasons outside of the growth season.

Data analyses
We examined correlations between the Tungsten δ 18 O TR data and the aforementioned climate datasets. All of them span different periods and are constructed differently. We filtered the climate data seasonally, with the expectation that climate relationships would be most strongly affected by interannual variability during the growing season. The δ 18 O TR could also be influenced by climate during the previous winter due for example to snow δ 18 O, spring runoff, and consequently soil moisture available for the growth season, considering that during winter the climatic influence on high-latitude precipitation δ 18 O is more pronounced (Birks and Edwards 2009;Field et al. 2010). Large-scale climatic influences were also expected to vary seasonally because of their distinct strengths during different seasons, for example the Aleutian Low which is most strongly expressed in winter (Hartmann and Wendler 2005).
We compared the relationships of precipitation δ 18 O with large-scale climate provided by the ModelE2 GCM versus the relationships of δ 18 O TR with the same climate variables from reanalyses products. This comparison can help us to determine the prevailing signal in δ 18 O TR that results from strong climate influences on the δ 18 O composition of soil water without the influence of tree physiological processes.

Tree-ring chronologies
The Tungsten TRW chronology spans from 1584 to 2002, although replication is lower for the earlier period. EPS values in the ring width (N = 25 trees, 26 timeseries) and δ 18 O TR (N = 5 trees) chronologies (Fig. 2) exceed 0.85 from 1900 to 2002, suggesting that both tree-ring chronologies can be considered reliable for the studied period. The average of the Pearson correlation coefficient values (r) between each δ 18 O TR tree timeseries was 0.601 (p < 0.05; 1900-2003), 0.68 (p < 0.05; 1900-1969) and 0.513 (p < 0.05; 1970-2003), while the mean of the δ 18 O TR values was 19.02 ± 0.77‰ , 19.08 ± 0.67‰  and 18.88 ± 0.98‰ . Thus, lower correlations among trees and higher Standard Deviation (SD) were found during the period 1970-2003 (r = 0.513,  p < 0.05 and SD = 0.98‰) than during the period 1900-1969 (r = 0.68, p < 0.05 and SD = 0.67‰). This less common variance and higher variability among trees is also shown by the higher SD of the δ 18 O TR chronology (z-scores) in the period 1970-2003 (SD = 0.58) than in the period 1900-1960 (SD = 0.49). Table 1 lists the Pearson's correlation coefficients between the residual TRW chronology and several climate variables for different seasons and periods of observational data availability. Over the 1977-2002 period common to both the ISD and GHCN gridded datasets, TRW was negatively correlated to spring (MAM) minimum temperature using both ISD (r = − 0.48, p < 0.05) and GHCN (r = − 0.49, p < 0.05), and positively correlated to snow depth (SNDP, r = 0.64, p < 0.05). The relationship of TRW with TMIN and SNDP were not significant over the longer 1938-2002 period, although there was a weak positive correlation (r = 0.34, p < 0.05) with summer (JJA) TMAX, and weak negative correlations with precipitation for seasons prior to the growing period, peaking at r = − 0.39 for winter-spring-summer (previous DJFMAMJJA). Table 2 lists the Pearson's correlation coefficients between the δ 18 O TR chronology and these same climate variables. VPDMAX and TMAX were positively correlated during the MAMJJA period (r = − 0.86, p < 0.05) and both agreed with δ 18 O TR z-scores fluctuations (Fig. 3). The δ 18 O TR z-scores showed a strong correlation with average spring-summer TMAX (Fig. 3a) from the GHCN dataset for 1938-2002 (MAMJJA, r = 0.67, p < 0.01), but lower correlation over the 1977-2002 period using TMAX from the ISD dataset (MAMJJA, r = 0.49, p < 0.05). This correlation with the spring-summer ISD TMAX was lower than when δ 18 O TR Table 1 Pearson's correlation coefficients (r) between Tungsten treering width (TRW) residual and seasonal "Watson Lake A" maximum temperature (TMAX), minimum temperature ( 1977-2002 1977-2002 1938-2002 Fig. 3b, for spring-summer (MAM-JJA, r = 0.55, p < 0.05). While the correlations between δ 18 O TR and summer VPDMAX were significant (JJA, r = 0.44, p < 0.05), non-significant correlations were found with spring VPDMAX (MAM, r = 0.39, p = 0.08). There were also weaker positive relationships between δ 18 O TR and GHCN TMIN (MAM, r = 0.28, p < 0.05); JJA, r = 0.27, p < 0.05), although higher with ISD TMIN but only for summer (JJA, r = 0.50, p < 0.05) and over a shorter period. Weak negative relationships between δ 18 O TR and SNOW during spring (MAM, r = − 0.31, p < 0.05) slightly higher during spring-summer (MAMJJA, r = − 0.33, p < 0.05), as well as between δ 18 O TR SNDP (MAM, r = − 0.33, p < 0.05; MAM-JJA, r = − 0.34, p < 0.05) were also found. The GHCN TMAX correlation in summer for 1977-2002 was lower (r = 0.41, p < 0.05) than for the ISD data (r = 0.48, p < 0.05) likely because most GHCN data were missing for 1993 and 1994. Overall, the strongest correlation with δ 18 O TR was found with GHCN TMAX for spring-summer (r = 0.67, p < 0.01) for the whole 1938-2002 period (Fig. 3a, Table 2). The strong correlation during this season is related mainly to lower frequency changes in δ 18 O and TMAX (Fig. 3a). Higher δ 18 O TR from 1938 until the late 1950s was associated with warmer temperatures, followed by a decrease in both from 1960 until the early 1970s, and then an increase in δ 18 O TR and summer TMAX in the late 1970s, which persisted until the early 2000s.

Observed climate relationships of TRW and δ 18 O records
Based on the strength of spring and summer TMAX controls on δ 18 O TR , we examined the correlations between δ 18 O TR and different gridded climate fields. Figure 4 shows the spatial field correlations between δ 18 O TR and seasonal surface temperature anomalies from BEST TMAX, GIS-TEMP T surf and, for reference, the UDEL precipitation, for the period 1938-2002. For GISTEMP (Fig. 4a), there was a positive correlation pattern centered over northern British Columbia during spring-summer (MAMJJA) and extending across most of Canada. The BEST correlation field (Fig. 4c) is similar but is smoother and with higher correlations over the study site. For both GISTEMP (Fig. 4b) and BEST (Fig. 4d), there were no coherent patterns of correlation during autumn-winter (SONDJF), consistent with the analysis performed at the weather station scale. For UDEL precipitation, there were no coherent correlation patterns over the study site for either MAMJJA (Fig. 4e) or SONDJF (Fig. 4f), showing only weak negative, albeit significant, correlation in continental Canada.
To identify large-scale circulation influences, we also examined δ 18 O TR correlation fields for selected variables from the NCEP/NCAR Reanalysis I over 1948-2002 (Fig. 5). Temperature correlation maps were similar to GISTEMP and BEST and displayed a coherent region of positive correlation in western Canada during spring-summer (Fig. 5a, MAM-JJA), but no coherent pattern in autumn-winter (Fig. 5b,  SONDJF). During either season, there was no coherent correlation pattern between δ 18 O TR and precipitation amount Fig. 3 a The δ 18 O TR chronology (z-scores, blue) and average spring-summer (MAMJJA) TMAX at GHCN and ISD Watson Lake A weather station (bold and dashed orange lines, respectively), and b The δ 18 O TR chronology (z-scores, blue) and average spring-summer VPD-MAX from ISD at the same station (dashed orange line). The r indicates the Pearson correlation coefficient between the δ 18 O TR chronology and the climate timeseries (Fig. 5c, d). The precipitation amount correlation field was included for completeness, though we note that the NCEP reanalysis precipitation estimates are only weakly constrained by observations (Kalnay et al. 1996), unlike the corresponding UDEL precipitation used in Fig. 4e, f. There was also no apparent moisture pathway signature (Fig. 5e, f) which would have appeared as a coherent vector field in the vicinity of the study site. No clear correlation pattern was found between δ 18 O TR and SLP during either season (Fig. 5g, h). There were, however, strong correlations between δ 18 O TR and geopotential height at 500 hPa during MAMJJA (Fig. 5i), capturing the basic association between warmer temperatures at the site and pronounced high-pressure ridging over western Canada. Individual correlation maps for spring, summer, autumn and winter were similar to the maps using 6-month season definitions for Figs. 4 (Figures S1 and S2) and 5 ( Figures S3 and S4). We also compared the NCEP/NCAR 6-month correlation fields (Fig. 5) to those for the 20CRv3 reanalysis ( Figure S5) for the same period 1948-2002. The region of positive correlation for MAMJJA ( Figure S5a) to the southwest of the site is consistent with that for NCEP/NCAR, but it is smaller in its extent and weaker in magnitude, which was also the case for the 500 hPa geopotential height patterns ( Figure S5g). The weaker patterns in both cases are presumably due to the 20CRv3 product having far fewer observational constraints. Using the 20CRv3 reanalysis over the full 1900-2002 length of the δ 18 O TR record ( Figure S6), there were also positive temperature ( Figure S6a) and 500 hPa geopotential height patterns south of the site ( Figure S6g), but which were more diffuse in their extent compared with patterns obtained using NCEP/NCAR reanalyses ( Fig. 5a and 5i).

Climate-precipitation δ 18 O relationships in GISS ModelE2 GCM
Distinct seasonal ModelE2 correlation fields of large-scale circulation features with the modelled precipitation δ 18 O over the study site reveal different seasonal influences on source water δ 18 O by 6-month periods (Fig. 6) and by seasons ( Fig.  S7 and S8). During spring-summer, the correlation pattern showed a positive relationship between precipitation δ 18 O at the study site and temperature over northwestern North America, although with no apparent relationship in eastern Canada (Fig. 6a). These features were similar to patterns observed between δ 18 O TR and the TMAX field for the GISTEMP (Fig. 4a), BEST (Fig. 4c) and NCEP reanalysis (Fig. 5a) temperature fields. There was also pronounced positive correlation in MAMJJA between modelled precipitation δ 18 O and TMAX in the Gulf of Alaska (Fig. 6a), seen somewhat in the GISTEMP MAMJJA correlation map with δ 18 O TR (Fig. 4a). For autumn-winter (SONDJF), there were also pronounced patterns of positive correlation between modelled precipitation δ 18 O and TMAX (Fig. 6b), unlike the lack of relationships between δ 18 O TR and TMAX for that season when looking at the study site point scale (Fig. 4b).
During spring-summer (MAMJJA), there was no apparent relationship between precipitation amount (Fig. 6c) or moisture pathway (Fig. 6e) and modelled precipitation δ 18 O over Tungsten, consistent with the absence of any patterns in the δ 18 O TR correlation maps (Fig. 5c, e). During autumn-winter (SONDJF), there were more coherent patterns showing a positive relationship between modelled precipitation δ 18 O and precipitation amount in the Gulf of Alaska and negative relationship to the southeast (Fig. 6d). During the same season, higher precipitation δ 18 O was also associated with southwesterly moisture origin (Fig. 6f). For the SLP field, no significant correlation pattern was observed during spring-summer (MAMJJA, Fig. 6g), but a strong pattern was found during autumn-winter (SONDJF, Fig. 6h) with a negative center over Alaska and the Bering Sea and a positive center over the US Great Plains. The correlation between MAMJJA precipitation δ 18 O and 500 hPa geopotential height (Fig. 6i) were consistent with that observed for Tungsten δ 18 O TR (Fig. 5i). The correlation patterns in MAMJJA (Fig. 6i) are similar to SONDJF (Fig. 6j), but they are stronger in SONDJF.

Discussion
In this section, we discuss the signal and stability of the relationship between climate and the tree-ring proxies, the strong imprint of temperature in δ 18 O TR and its potential for reconstructing large-scale atmospheric patterns. For the variables considered, the climatic information contained in the TRW was weaker than in the δ 18 O TR records, but several significant relationships were identified.

Instability in the relationship between TRW and climate variables
We found a relatively unstable relationship between climatic and TRW records in some locations. For example, no significant correlation was found between TRW and summer TMAX (June to August, JJA) for the period 1977-2002, while a significant positive correlation was observed over the longer 1938-2002 period with the same dataset (GHCN). This may be related to the divergence-type phenomenon which has been well documented for a number of boreal forest sites (D'Arrigo et al. 2008; and references therein). Local site conditions, recent warming and hydroclimatic trends may also complicate the understanding of the relationship between climate and tree-ring parameters (Gedalof and Smith 2001;Li et al. 2020;Wendler et al. 2017). In the Yukon region for example, almost all of the 111 TRW chronologies from a white spruce network lost their positive relationship with summer temperatures, with one third of them showing negative responses after ~ 1950 (Morimoto 2015). A weakening of the precipitation signal and subsequent strengthening of temperature sensitivity in white spruce has been recorded in various tree-ring sites from the Alaskan and Canadian interior (Chavardes et al. 2013;Lange et al. 2020). Despite the relative instability of TRW responses in such regions, many TRW records still show strong relationships with local (Jacoby and Cook 1981), as well as largerscale temperatures and serve for north hemispheric climate reconstructions across a network of sites (D'Arrigo et al.

and references therein).
Considering that cold temperatures in spring may delay the start of the growing season and reduce the period for xylogenesis in northern environments (Rossi et al. 2008), the negative correlations that we found between TRW and spring TMIN (1977TMIN ( -2002 are difficult to interpret. In contrast, the negative correlations between TRW and spring precipitation  may reflect a detrimental effect on growth because spring precipitation falling as snow could delay the start of the growing season. Vaganov et al. (1999) suggested that more abundant snow accumulation may delay snowmelt and induce a delay in cambial activity and a reduction in growth. However, other authors suggested a positive role of spring snowpack on growth related to an increase in moisture availability with higher snowmelt water when the growing season starts in late spring/early summer (Yarie 2008) and/or thermal soil insulation by snow that could enhance growth (Grippa et al. 2005). In moisture-limited sites in western North America forests positive snow-growth relationships have been also reported (Coulthard et al. 2021). In the white spruce forest studied here, a positive effect of larger snowpack on tree-ring growth may be occurring although this relationship was only found in the most recent period .

Temperature as the dominant signal in the δ 18 O TR records
The δ 18 O TR relationships with temperature identified here were stronger and more stable throughout time than those for TRW, and broadly consistent with other studies at highlatitude North American sites. In northeastern Canada, Alvarez et al. (2018) found stronger positive correlations between δ 18 O TR and TMAX than between δ 18 O TR and TMIN. Naulier et al. (2014) found a June-July correlation of r = 0.55 (p < 0.05) with TMAX and black spruce δ 18 O TR over 1949-2005 in Quebec, slightly higher than the correlation of r = 0.45 for VPD. Using data from this site combined with a process-based model (MAIDENiso), Lavergne et al. (2017) found that the temperature signal recorded in δ 18 O TR more likely reflects the effect of temperature on isotopic enrichment of the leaf water than on the isotopic composition of the source water. Begin et al. (2015) found higher correlations between δ 18 O TR and summer VPD than for summer TMAX (r = 0.64 versus r = 0.55, p < 0.05). In Alvarez et al. (2018), δ 18 O TR was negatively correlated with river discharge. Similarly, δ 18 O TR was negatively correlated with summer precipitation in Naulier et al. (2014) and Begin et al. (2015). We found a negative relationship between precipitation and δ 18 O TR , but which was weaker and only significant during the combined winter-spring-summer seasonal period (Table 2). By contrast, Holzkamper et al. (2012) found that δ 18 O TR in white spruce was positively correlated with spring temperatures, and negatively correlated with precipitation amount at a site in north-central Canada. In northwestern North America, Porter et al. (2009) found a weak negative correlation between δ 18 O TR and April precipitation amount. Positive relationships were found between δ 18 O TR and early-spring to mid-summer minimum temperatures and summer relative humidity, attributing the former to the temperature dependence of source water δ 18 O and the latter to evaporative δ 18 O enrichment. This is in agreement with our findings showing that the δ 18 O TR record was influenced by temperature variations in spring and summer. While physiological processes have most likely influenced δ 18 O TR in summer through evaporative enrichment at leaf level, spring climate also imprints δ 18 O TR potentially via temperature effects on precipitation δ 18 O signatures (e.g. Treydte et al. 2014). Snow plays an important role in these latitudes. A possible influence of winter snowpack on δ 18 O TR is also suggested by the negative relationships with the snow variables for the 1938-2002 period. Along with a positive correlation observed between δ 18 O TR and TMAX, Csank et al. (2016) found a negative correlation between δ 18 O TR for a site in southern Alaska and prior winter snow amount. This was consistent with our results, and possibly explained by the effects of snowpack as a moisture source during the growing season. Snow has a lower δ 18 O than rain (Kurita et al. 2004), therefore years with greater snow accumulation would contribute to source water in the soil having lower δ 18 O (Beria et al. 2018). However, processes operating in the opposite directions could also be taking place, weakening this negative relationship depending on the snow accumulation and residence time of the snow before melting. For example, with more snow, more snowmelt (freeze/unfreeze events), sublimation and other kinetic processes can take place leading to more 18 O enrichment (Beria et al. 2018;Ebner et al. 2017).
For all tree-ring and climate parameters considered during the full 1938-2002 period of analysis, δ 18 O TR had the highest correlation (r = 0.67, p < 0.05) with spring and summer TMAX. Since δ 18 O TR is influenced by the precipitation δ 18 O at the site via soil water, this relationship can be explained, in part, by the large-scale co-variation between temperature and δ 18 O precipitation in northern latitudes. This occurs through a Rayleigh distillation of the water vapor that is transported by the air masses (Araguas- Araguas et al. 2000;Gat 1996). Air masses arriving with a colder history will have undergone more rainout, during which the heavier isotopologues (i.e. molecules of a particular element which differ only in the neutron number) will be removed preferentially through fractionation, leading to lower precipitation δ 18 O at the sampling site. In addition, during condensation from water vapour to rain, more fractionation of δ 18 O occurs under colder conditions than warmer conditions (Clark and Fritz 1997) leading to even more depleted precipitation δ 18 O under colder conditions. Therefore, the yearly isotopic signature of precipitation δ 18 O is the result of the variation between rainout occurrence due to colder (warmer) air masses that experience more (less) rainout events and rainouts Fig. 5 Spatial field correlations between the δ 18 O TR chronology and NCEP surface temperature (T surf ), precipitation (Precip), moisture transport at 500 hPa (< qu,qv >), sea-level pressure (SLP), and geopotential height at 500 hPa (Z 500 ) for spring-summer (March-August, MAMJJA, left) and autumn-winter (September-February, SONDJF, right), over 1948-2012. Correlations with p-values < 0.05 have been excluded ◂ with more (less) 18 O fractionation associated with lower (higher) temperatures during condensation. This is manifested interannually, with lower δ 18 O TR in years with lower precipitation δ 18 O related to colder upstream conditions (more rainout events and more 18 O fractionation during condensation), while higher δ 18 O TR in years with higher precipitation δ 18 O may be related to warmer upstream conditions (less rainout events and less 18 O fractionation during condensation). Note that at high latitudes, there are not strong direct relationships between precipitation amount and precipitation δ 18 O in the GISS GCM (Schmidt et al. 2007), measurements from the Global Network of Isotopes in Precipitation (Risi et al. 2010) or in an isotopic atmospheric water balance model (Zhang et al. 2015), even when there are positive relationships between temperature and precipitation δ 18 O. In our study, this was seen by only weak negative correlations with GHCN precipitation during the winter-spring-summer period (Table 2) and a lack of significant correlations between the Tungsten δ 18 O TR and both the instrumental precipitation (Fig. 4e, f) and the reanalysis precipitation fields (Fig. 5c, d). In agreement, our GCM results show a lack of relationship between spring-summer modelled precipitation δ 18 O and precipitation and over the Tungsten site (Fig. 6c, d). Note that in fall-winter higher modelled precipitation δ 18 O was associated with more precipitation over the Gulf of Alaska and southwesterly moisture transport, which we interpret as primarily as covariation with warmer, more moist air masses arriving from the south to the Gulf of Alaska.
During the 1977-2002 period, δ 18 O TR was related to VPDMAX during the annual and combined spring / summer periods. However, the strength of these correlations was mostly driven by summer VPDMAX because non-significant correlations were found in spring alone. Thus, VPD increase may be driven by warmer summers and may induce evaporative 18 O enrichment at leaf level during transpiration (Barbour 2007;Gessler et al. 2014). This is consistent with other studies in high-latitude forests of Quebec (e.g. Lavergne et al. 2017). Our GCM results relate to the idealized source water signal unaffected by tree physiological isotopic fractionation. Higher correlation between modelled δ 18 O precipitation and temperatures were found in spring compared to summer (Fig. S7a cf. S7b). This indicates that the source water signal is stronger in spring than in summer, illustrated by the GCM diagnosis where physiological enrichment is not present but where other processes such as summer post-condensation exchange can weaken the temperature signal. Additionally, as reflected in results from observations with higher correlations between δ 18 O TR and temperature in summer than in spring (Fig. S1b cf. Fig. S1a), the temperature signal in the δ 18 O TR is higher in summer when both the source water signal and the VPD-induced 18 O enrichment at leaf level are present. The strength of the δ 18 O TR -VPDMAX relationships suggests that annual isotopic measurements in tree rings could be potentially good proxies for reconstructing summer temperature and VPD, but that analyzing earlywood and latewood isotopic measurements independently may be a better option for distinguishing the seasonality effect of source water and VPD in the δ 18 O TR signatures at a higher temporal resolution (e.g. Belmecheri et al. 2018;Levesque et al. 2017). Finally note that in the case of the study site, these data are limited by the short length of ISD data over which VPD could be calculated (compared to the longer GHCN records, for example, but which had no humidity records).

The δ 18 O TR records as a proxy for large-scale atmospheric circulation fields
The positive association between δ 18 O TR and TMAX timeseries reported is driven by both inter-annual and decadal variations in spring-summer temperature (Fig. 3). In this context, can δ 18 O TR serve as a proxy for temperature variations or even for large-scale atmospheric circulation fields?
The strong δ 18 O TR -TMAX relationship was clearly seen regionally in correlation maps with GISTEMP and BEST temperature fields, with areas of higher correlation centered in northwestern North America. This was, in turn, related to a region with positive correlation with 500 hPa geopotential height centered over the study site, which we interpret as a signature of the relationship between high temperature and stronger meridional (southerly) atmospheric flow. This was similar to the patterns seen in the composite relationships between 500 hPa geopotential height and precipitation δ 18 O at three sites in central Canada (Birks and Edwards 2009). These relationships between δ 18 O TR , surface temperature and geopotential height were also seen in those between modelled precipitation δ 18 O over the study site, surface temperature and geopotential height in the NASA GISS ModelE2 simulations.
In the model simulations, these relationships were seen for both the spring-summer (MAMJJA) and fall-winter (SONDJF), unlike the δ 18 O TR for which positive correlations were only seen during the MAMJJA period more associated with the growing season. Similarly, for precipitation amount and moisture transport, coherent positive associations between precipitation δ 18 O and southwesterly moisture transport were only seen in model simulations for the SONDJF period; their absence in the δ 18 O TR can  (Barnston and Livezey 1987). Over southern North America, Liu et al. (2014) found a positive to negative dipole correlation pattern between the PNA index and modeled winter precipitation δ 18 O (Yoshimura et al. 2008), oriented southeastward from western Canada which is consistent with the spatial pattern observed in Fig. 6j. This suggests that previous reconstructions of the PNA using tree-ring width records (Liu et al. 2017) could be enhanced with isotopic measurements to further understand hydroclimatic relationships and external forcing over North America throughout the last millennium. Our results also suggest that combining δ 18 O TR , which is most sensitive to summertime circulation, with other isotopic archives more sensitive to wintertime circulation such as ice cores (Field et al. 2010) have potential for annual or seasonally-varying reconstructions of atmospheric circulation.

The potential role of the Pacific Ocean forcing
Changes in the relationships between climate and both TRW and δ 18 O TR records over the 1938-2002 period can be in part driven by a regional climate shift in the mid-1970s. After 1977 TRW became insensitive to the previous positive role of summer temperatures, negatively influenced by TMIN and positively by snow depth, while δ 18 O TR became more strongly linked to TMAX and insensitive to the previously negative influence of snow depth. We also observed less common shared variance and higher variability in δ 18 O TR among the trees after 1970 (Fig. 2). The increase in temperature and δ 18 O TR variability during this period is consistent with an abrupt shift towards higher mean annual observed temperature in interior Alaska (Hartmann and Wendler 2005) and a broad range of environmental changes (Ebbesmeyer et al. 1991;Mantua et al. 1997). These were concordant to the well-known regime shift of the Pacific Decadal Oscillation (PDO) in 1976/77 from its negative (cold) to positive (warm) phase (Ebbesmeyer et al. 1991;Mantua et al. 1997;Trenberth and Hurrell 1994) and of the PNA Pattern index to its positive phase (Minobe and Mantua 1999;Overland et al. 1999), both associated with a strengthening of the Aleutian Low. Such apparent readjustment of largescale mode of climate variability was also seen in δ 18 O TR records for the Mackenzie Delta, NWT (Porter et al. 2014), and in δ 18 O data from the Mt. Logan ice core (Field et al. 2010). These observations are consistent with the broader regional footprint of TMAX in our δ 18 O TR chronology, seen in the correlation patterns with the GISTEMP and BEST gridded temperature products (Fig. 4). It is also interesting to note that similar weakness in the strength of the relationship between δ 18 O TR and temperatures have also been observed after 1970 in the extra-tropics in Patagonia, South America (Lavergne et al. 2016), reinforcing the hypothesis that our observations may be related to changes in the PDO and its impact in driving inter-hemispheric ocean-atmospheric connections across both of the Western Americas (Villalba et al. 2011).

Conclusions
Here, we investigated the potential of tree-ring isotopic and ring-width measurements of white spruce at the boreal forest treeline in the Northwest Territories, Canada to record local to regional climate and reconstruct atmospheric circulation patterns. Among the relationships examined, the strongest was a temperature signal imprinted in δ 18 O TR cellulose at the Tungsten site over 1938-2002, likely driven by the precipitation δ 18 O signature (i.e. source water). This was seen consistently comparing δ 18 O TR with temperature data from different sources, i.e. a weather station, two gridded temperature products, and two reanalyses. The imprint of temperature on δ 18 O TR is likely associated to the temperature impact on fractionation processes during the condensation of water vapor to rainwater expected in this high latitude (i.e. colder upstream conditions, more rainout events and more 18 O fractionation during condensation leading to lower precipitation δ 18 O). Evaporative enrichment of 18 O at leaf level could also contributing to the final δ 18 O TR signature, but mainly during summer. We also found a weak but significant negative relationship between snow accumulation and δ 18 O TR over the 1983-2002 period; a deeper snowpack leads to a greater supply of soil water with lower δ 18 O values. Diagnosis with an isotopically-equipped climate model contributed to our understanding of seasonal differences in the influence of temperature and circulation patterns on the δ 18 O TR without the influence of tree physiology. No significant relationships were found between the δ 18 O TR and reanalysis moisture transport for either fall-winter or springsummer, but appeared during fall-winter, if unevenly, for the modelled precipitation δ 18 O at the sampling site. Our interpretation is that the fall-winter circulation controls on precipitation δ 18 O are not strong enough to influence tree uptake of isotopically depleted water during the spring and summer growth season, despite a possible relationship between winter snow depth and δ 18 O TR .
We conclude that the δ 18 O TR records for northwestern Canada reflect the spring-summer atmospheric circulation patterns in this region. The broad consistency of the positive relationships between δ 18 O TR and temperature observed in this study and across northern North America demonstrates the potential of using stable oxygen isotopes measured in tree rings for reconstructing temperature, but also other large-scale climate indicators as a novel aspect. Combining the isotopic and other climate signals gleaned from various tree-ring parameters (e.g., MXD, BI) we could produce more robust climate reconstructions. As with dendroclimatological studies from tree-ring width at multiple sites, we therefore expect further gains in reconstructions using a multispecies network (Pederson et al. 2013) of tree-ring isotopic records at regional, continental and eventually hemispheric scales. Forward modeling of tree-ring parameters, as in Lavergne et al. (2017), will further help to understand the relative contributions of 'site-level' factors such as source water uptake and leaf-level processes to better isolate past signals of temperature and moisture variability.