Introduction

Using a collection of geographically distributed tree-ring records for climatic study requires the assumption that top-down climate processes and patterns influence annual growth ring formation locally (Cook, 1992). The strength of this climatic expression can be gauged by assessing the common growth signal between trees in the same local environment, as well as trees growing several hundred kilometers apart. In continental regions where climate variability exhibits pronounced seasonality and rhythmic behavior such as the desert southwest (Swetnam and Betancourt, 1992; Meko et al., 1993), south-central plains (Stahle, 1990), or the central Pacific Coast (St. George and Ault, 2011), tree-ring records share a high proportion of common growth variance, and faithfully record interannual to decadal variability. Trees growing in mountainous regions near upper- and lower-forest zones also share significant common variance because either cold temperatures or moisture deficits frequently limit growth (LaMarche, 1974; Fritts, 1976; Hughes and Funkhouser, 2003). However, outside of these cardinal dendroclimatic regions, non-climatic factors often complicate the extraction of climate variability from tree-ring records. For example, temperate regions commonly lack pronounced seasonality in precipitation, high-elevation environments that contain temperature-limited trees, and relatively open stands devoid of competition for available resources. Thus, conducting dendroclimatological research in temperate ecosystems primarily yields a weakened common growth signal that reflects increased local-scale noise.

In order to obtain tree-ring records that capture decadal climate variability over eastern North America, efforts must be made to locate open-canopy stands within a region dominated by dense closed-canopy forest (Crawford and Kennedy, 2009). Trees growing within this closed-canopy structure typically experience higher inter-tree competition—an endogenous factor that dampens the common signal (Cook, 1985). This local-scale ecological effect, in the form of tree-level growth suppressions, can significantly override the influence of macro-scale climate variability on ring-width variations (Cook and Briffa, 1990). In contrast, trees growing within a low-density open-canopy forest are more likely to have stronger climate signals because ecological competition is minimal (Fritts et al., 1965; Pilcher, 1990). Open-canopy forests across eastern North America are rare, and this fact has restricted development of dendroclimatic records capable of recording regional climate patterns at decadal time scales. Still, numerous tree-ring records exist for eastern North America (see International Tree Ring Databank contributors,  http://www.ncdc.noaa.gov/paleo/treering.html) thanks to advancements in closed-canopy tree-ring standardization (Cook and Peters, 1981).

Northern red oak, a deciduous species that occupies low- to high-elevation sites across the Southern Appalachian mountain range, has received considerable attention in dendroclimatic research (e.g., Tainter et al., 1990; Pan et al., 1997; Fekedulegn et al., 2003; Speer et al., 2009); however, these studies were based on tree-ring records from lower elevation closed-canopy stands where climate is less variable and often markedly differs from higher elevation regions (Konrad, 1996). In this paper, a collection of six open-canopy high-elevation northern red oak (Quercus rubra L.) ring-width records are evaluated for the presence of a common climate-driven growth signal across the Southern Appalachian mountain range in the southeastern United States. The first stage of analysis determines whether high-elevation northern red oak ring-width records share a spatially coherent common growth signal in time and frequency domains, and the second stage test for significant surface temperature and precipitation association.

This study differs from previous dendroclimatic studies on northern red oak from two standpoints. First, living tree-ring records, supplemented by remnant wood, have been collected from open-canopy high-elevation forest environments above 1600 m along a 250 km southwest–northeast transect. Second, these records have been developed at northern red oak's elevational range limit, and comprise the longest set of multi-century northern red oak tree-ring records ever developed from the southeastern United States. This common signal approach allows the determination to be made whether open-canopy high-elevation northern red oak tree-ring records along the Southern Appalachian mountain range record regional climate variability.

Tree-Ring Data and Methods

The Southern Appalachian mountain range is located in the southeastern United States between 34–37°N and 81–84°W (Fig. 1). High-elevation forest stands across the mountain range are reminiscent of the high-latitude boreal-deciduous ecotone (Whittaker, 1956), and include a northern hardwood species assemblage with isolated spruce-fir forest zones. The high-elevation open-canopy northern red oak stands under investigation in this study tend to occur on west and southwest aspects in and around Southern Appalachian grass/heath balds (Crawford and Kennedy, 2009). This southeastern sub-region has a predominantly humid subtropical climate with cooler and wetter conditions occurring with increasing elevation. In general, the winter season (December–February) tends to be cold and snowy with mild-moist summers (June–August). Total annual precipitation can exceed 150 cm across higher elevation regions, but is highly variable across the range resulting from differences in windward (leeward) slope exposure and proximity to Gulf of Mexico (Konrad, 1995, 1996). Annual snowfall varies with valleys receiving 20 cm or less and higher mountain slopes receiving 80+ cm from orographic effects (Perry and Konrad, 2006).

Figure 1

Southern Appalachian mountain range, high-elevation northern red oak ring-width records, and CRU TS3.00 temperature and precipitation grid points (0.5° × 0.5°). The Big Bald Craggy (BBC) chronology represents ring-width data combined from two nearby sites to lengthen and increase sample depth.

Living tree and remnant wood specimens from six high-elevation (>1600 m) open-canopy northern red oak stands were collected along a southwest–northeast gradient in the Southern Appalachian mountain range (Fig. 1). Between 25 and 40 trees were sampled at each site, and either two cores or one cross section were taken from each tree using an increment borer or chainsaw. Each sample was prepared and processed according to standard dendrochronological techniques (Stokes and Smiley, 1996). Skeleton plotting was used to visually crossdate samples, and each series was measured to the nearest 0.001 mm. Visual crossdating was statistically verified with COFECHA (Holmes, 1983).

Age-related geometrical growth trends for each tree were removed with a deterministic growth curve using ARSTAN (Cook, 1985; Cook and Kairiukstis, 1990). Before detrending, an adaptive power-transformation was applied to each ring-width series to account for any ring-width population skew, improve curve fits, and minimize trend propagation in the final mean chronology calculation (Cook and Peters, 1997). In addition, the variance was stabilized with a 67% n spline and Briffa's r-bar method (Briffa and Jones, 1990). The 67% spline is necessary to appropriately stabilize the radial growth variance in ring-porous species because as Quercus spp. mature, their earlywood vessels increase the proportional ring-width variability unrelated to environmental conditions. This tendency has been identified as a “trend in variance” that is biologically dependent and must be removed for confident interpretation of ring-width variation towards the end of time series (e.g., Shiyatov et al., 1990).

A conservative detrending approach was applied to each ring-width series by fitting a horizontal mean line, a negative exponential curve, or a linear regression line with a negative slope. Ring-width chronologies were constructed by dividing the expected ring-width by the actual ring-width to produce a dimensionless index of average growth across all series. Each chronology was estimated by using a bi-robust weighted mean to ensure against the inclusion of outlying ring-width values (Cook, 1985). Autoregressive modeling was used to remove serial autocorrelation and obtain a residual chronology without lower frequency growth persistence, unlike the standard chronology produced from the same series.

Correlation and principal components analysis (PCA) were used to assess the common growth variance between each standardized ring-width chronology in the time domain (e.g., Peters et al., 1981). The period 1800–2005 was selected for comparison because the sample size for each chronology was deemed sufficient using an expressed population signal (EPS) cutoff of 0.85 (e.g., Wigley et al., 1984). Both residual and standard chronologies were subject to this analytical step because each time-series reflects a distinct difference in autocorrelation structure. For the same period, spectral analysis was used to estimate the relative concentration of frequency-dependent variance within each ring-width chronology, followed by a comparison between frequency-stratified variance (LaMarche, 1974). Spectral estimates were computed using the Blackman-Tukey method (Meko, 1992; Chatfield, 2004) and only site-specific standard chronologies were used for this analytical step because the autocovariance structure was preserved during standardization.

Using the Matlab function seascorr (e.g., Meko et al., 2011), Monte-Carlo correlation analysis (1000 simulations) was used to assess the significance between northern red oak ring-width variability and surface temperature and precipitation. Mean monthly temperature and total monthly precipitation were obtained from the University of East Anglia Climate Research Unit TS3.00 0.5° × 0.5° gridded data set (e.g., Mitchell and Jones, 2005; CRU, 2008), and the four grid points surrounding each site chronology were selected for comparison. In addition, principal component one (hereinafter PC-1) was compared with the twenty-six TS3.00 grid points covering the spatial domain of the Southern Appalachian mountain range (Fig. 1). Monthly and seasonal (2–4 month averages) coefficients for temperature and precipitation were computed across a 14-month window from prior August to current September during 1903–2005. Monthly and/or seasonally significant ring-width-climate associations were tallied for p < 0.05 and p < 0.01 levels (two-tailed). The temporal stability of each significance coefficient was tested using a split-period approach that separated 1903–2005 into two distinct early (1903–1947) and late (1948–2005) sub-periods. The difference between each sub-period correlation coefficient was calculated and subject to a significance test to determine whether sub-period coefficients were statistically different (e.g., Meko et al., 2011). Although the East Anglia Climate Research Unit's TS3.00 gridded data set accounts for changes in climate station density over time (Mitchell and Jones, 2005), early and late periods were selected because station density was noted to have increased substantially across the Southern Appalachian region after 1948 (e.g., Menne et al., 2009). This pre- and post-1948 change in meteorological observation offered a logical break across 1903–2005 to evaluate climate signal stability.

Results

High-elevation northern red oak ring-width records along the Southern Appalachian mountain range (Table 1 and Fig. 2) are significantly correlated across the past two centuries (Table 2). Slightly stronger correlations were observed between residual variance than autocorrelated variance, and in part reflects site-level southwest–northeast differences in mean sensitivity and biological persistence (Table 1). PCA between standardized ring-width records revealed a strong spatially coherent common growth signal along the mountain range. Each standard ring-width chronology loaded positively with PC-1, and cumulatively, accounted for 50% of the common variance (Table 3). Although PC-2 and PC-3 contributed an additional 34%, the sign of the loadings were different and indicate possible southwest–northeast differences between ring-width records (Table 3); however, because only six sites were evaluated, and the interpretation of lower order components becomes increasingly difficult without post-hoc reasoning, PC-2 and PC-3 were excluded from further analysis. PCA on residual variance (not shown) yielded very similar results and contributed no new meaningful information.Table 3

Figure 2

Open-canopy high-elevation northern red oak Whitetop Mountain (WTM), Sandy Mush Bald (SMB), Rich Mountain Bald(RMB), Bob-Stratton Bald (BSB), Burningtown Gap (BTG), and principal component one (PC-1) ring-width records with an 11-year moving average to visually emphasize decadal variance.

Table 1

High-elevation northern red oak ring-width chronology statistics.

Table 2

Time domain correlations between high-elevation northern red oak ring-width records (1800–2005). (A) Residual chronologies; (B) Standardized chronologies.

Table 3

Principal component statistics for high-elevation northern red oak ring-width records.

Spectral analysis of ring-width records revealed that over the past two centuries, the relative stratification of frequency-dependent ring-width variance has been concentrated at lower frequencies, particularly at decadal time scales. Ring-width variance at frequencies less than 5 years consistently fell below the white noise floor (i.e., null continuum) and indicated no significant variance is present at higher frequencies. Correlation analysis between the frequency-stratified variance (5–20 years) of each ring-width chronology yielded high coefficients (0.66–0.95), and suggests spatial coherence between frequency-dependent variance. Spectral estimates on PC-1 indicated that the common growth variance also exhibited significance variance at decadal time scales (Fig. 3).

Figure 3

Spectral estimates of the common ring-width variance (PC-1) (heavy black line) with 95% confidence intervals (thin dotted lines). The straight line indicates the white noise floor.

High-elevation northern red oak ring-width records were significantly correlated with mean temperature and total precipitation for 1903–2005 (Table 4). Individual monthly correlations varied by site, but the overall pattern of significance was consistent for prior fall–winter and current summer seasons. For mean temperature, prior October–prior December (positive) and current June–July (positive) were significant, and for total precipitation, prior September–current January (negative) and current June–July (negative) were significant (Table 4). Correlations for standard and residual chronologies did not differ substantially, although standard ring-width exhibited a wider seasonal response window with slightly higher coefficients (Table 4).

Table 4

Significant high-elevation northern red oak ring-width, temperature, and precipitation correlations.

Ring-width records showed time-dependent correlations with each monthly and/or seasonally significant temperature and precipitation coefficient for early and late sub-periods, but aside from a few notable exceptions, these differences were not statistically significant (Table 4). Examining the seasonal correlations between each site-specific temperature and precipitation CRU TS3.00 grid point (i.e., closest grid to each site by distance) for the 14-month period from prior August to current September revealed a significant seasonal change between temperature-precipitation covariance. For all ring-width sites across the mountain range, two significant seasonal peaks in temperature-precipitation covariance occur during winter (December–March) and summer (June–July) seasons with intervening fall and spring seasons exhibiting weak association (Fig. 4). During the winter season, the covariance between temperature and precipitation is positive, whereas, the relationship is negative for the summer season. This seasonality in covariance indicates a tendency towards warm-wet (cool-dry) winters, and warm-dry (cool-wet) summers. Though synchronous temporal changes in ring-width-climate signals do not occur across early and late periods, there appears to be a possible subtle shift from a prior fall precipitation response at a few sites during the early period to a current summer season precipitation response during the late period (Table 4). For standard ring-width chronologies, only one site (SMB) showed a statistically significant change in prior September precipitation sensitivity, whereas, for residual ring-width chronologies, three sites (WTM, SMB, and BSB) demonstrated a significant change in prior November temperature, prior September precipitation, and current June–July precipitation, respectively (Table 4).

Figure 4

Fourteen-month (prior August–current September) seasonal covariance between mean temperature and total precipitation.Correlation coefficients reflect the association between seasonal temperature average and seasonal precipitation sum for the 3 prior months. The dotted line represents the 95% confidence interval.

PC-1 demonstrated significant seasonal temperature and precipitation correlations with each CRU TS3.00 grid point across the Southern Appalachian mountain range for 1903–2005 (Fig. 5). Prior October–prior December mean temperature was significant for all grid points, and exhibited a stronger spatial pattern of significance towards the southern portion of the mountain range (Fig. 5, part a). Current June–July mean temperature was only significant for 12 of the 26 grid points, but revealed a spatial pattern of significance towards the northern portion of the Southern Appalachian range (Fig. 5, part b). Prior September–current January total precipitation was significant for 23 of the 26 grid points, and showed a strong spatial pattern of significance for both northern and southern portions of the mountain range (Fig. 5, part c). For PC-1, a significant temporal change in prior fall, winter, and summer season climate signals did not occur between early and late sub-periods.

Figure 5

Significant correlations between high-elevation northern red oak PC-1 and seasonal mean temperature and seasonal total precipitation for the Southern Appalachian mountain range.

Discussion

Conclusion

Open-canopy high-elevation northern red oak ring-width records share a significant common growth signal along the Southern Appalachian mountain range that exhibits variance at decadal time scales. The primary regional climate signal entrained within these ring-width records corresponds to prior fall–early winter mean temperature and prior late summer–winter total precipitation, though regional climate-growth signal strength appears contingent on the seasonal covariance between temperature and precipitation during winter and summer seasons. Despite a breakdown in climate signal stability during the instrumental period, the temporal change is not significant. Rather, it indicates that both temperature and precipitation can limit growth during the prior fall, winter, and current summer seasons, but that these climate-growth controls may operate jointly or in opposition over time to produce a mixed signal. Equally important to consider is that since a significant proportion of northern red oak's ring-width variance is concentrated at decadal time scales, a frequency-dependent climate response may exist, complicating the goal of developing a stable climate-growth calibration at interannual time scales.

This study provides compelling evidence that in addition to other Quercus spp. tree-ring records developed from temperate and Mediterranean regions, open-canopy high-elevation northern red oak tree-ring records hold the potential for evaluating decadal climate variability over the past four centuries across the Southern Appalachian region. Disentangling the apparent time-dependent mixed temperature-precipitation signal retained in high-elevation northern red oak ring-width represents a challenge that will likely require an intra-seasonal examination of vesselwood-latewood variance, especially relationships to seasonal climate variability and/or possible frequency-dependent climate responses contained therein (e.g., Gonzalez and Eckstein, 2003; Fonti et al., 2009). It is important to note that although future hydroclimate change remains uncertain for the southeastern United States in response to climatic warming (e.g., Seager et al., 2009), this study points out that temperature change across the Southern Appalachian region may have differential effects on winter and summer precipitation seasonality, and if so, tree-ring records may have difficulty resolving this change clearly.