Tree Ring–Width Climate Reconstruction

Tree cores for climate reconstruction were taken at breast height, with replicate cores extracted from opposite sides of trees. These trees were mainly located along ridges surrounding the catchments and rock outcrops within the catchments. Identification of old trees relied on branch width, needle loss in the upper portion of the trees, and tree size. Cross-dating for long chronologies was completed using a 32-year cubic smoothing spline with a 50% wavelength cutoff for filtering. To maximize the common signal within each chronology, only the most highly inter-correlated radii were retained, and complacent tree ring records (showing benign growth sites or an invariable limiting growth factor) were discarded (Biondi, 2001). Detrending of growth variations associated with age was accomplished using ARSTAN software (Holmes, 1992). Each series was detrended using a negative exponential curve, a linear regression line of negative slope, or a horizontal line through the mean. This detrending method is relatively conservative and was used to preserve low frequency, climate-related trends in the chronology (Cook et al., 1995).

The standard chronologies were correlated with monthly values of mean temperature, total precipitation, and the mean Palmer Drought Severity Index (PDSI; Palmer, 1965) for California Climate Division 3 (Great Basin;  http://www.ncdc.noaa.gov). Pearson correlation coefficients were calculated for the interval of overlap (1895–2002) between the standardized chronologies and the monthly instrumental data for the 16-month interval beginning June of the previous year and ending September of each growth year (Blasing et al., 1981).

After it was determined that California Climate Division 3 winter PDSI was the best predictor of annual changes in the ring-width series (see Results section), the monthly PDSI indices were used to run a principal components analysis (PCA) and stepwise regression to form a transfer function to predict ring-width measurements. For a thorough review of the PCA and stepwise regression techniques used for this study see Hidalgo et al. (2000). The resulting best-fit model was applied to tree ring–width variations that predate station records, establishing a long-term winter PDSI reconstruction.

The reconstruction was run through a Morlet wavelet analysis to determine the dominant modes of variability within the time series and how those modes vary through time (Torrence and Compo, 1998). This analysis offered information on the dominant periodicities within the record, and can indicate the underlying causal mechanisms by highlighting patterns of variation (Gray et al., 2003). Finally, tree ring chronologies were correlated to a 400-year record of reconstructed North Pacific PDO (Biondi et al., 2001) to assess if the reconstructions are capturing regional PDO patterns.

Transect Data

Each transect was begun 5 m from the lakeside, so as not to leave empty quadrants at the first sampling point. Along each 100-m transect, there were six sampling points spaced 20 m apart (from Bonham, 1989). At each point, the area around the point was divided into four 90° quadrants, the tree closest to the point in each quadrant was identified and cored, and distance from node was measured. In addition, the nearest stand-dominant tree, as determined by species, tree height in relation to canopy, and surrounding trees, was identified and cored to ensure that a representative sample of dominant trees was studied. Trees were cored or cross-sectioned as close to the ground as possible, usually within 20 to 30 cm of ground level for larger trees and at ground level for seedlings and small saplings. If coring missed the pith, the number of rings to the pith was estimated geometrically (Baker, 1992). Ring-width series were cross-dated with the master chronologies where possible. A correction was made for time to coring height by determining the age of a representative 20- to 30-cm seedling at each site. There were 5 years added to P. contorta and P. jeffreyi ages, and 9 years added to J. occidentalis ages.

Age structure data were aggregated into 5-year age classes, and histograms were constructed for each of the three dominant species. Age structure histograms were further separated by lake catchment and proximity to lakeside (mostly to account for recent lakeside recruitment). The resulting histograms were compared to two models that have commonly been used to depict the age structure of tree populations: the exponential function (constant natality and mortality) and the power function (constant natality) (Hett and Loucks, 1976; Légère and Payette, 1981). As the exponential function consistently displayed a higher significance level when compared with the histograms, this model was used for the remainder of the study.

Deviations of the actual age structures from the modeled exponential curves were used to reflect deviations from the expected constant rates in recruitment over time (Ågren and Zackrisson, 1990; Szeicz and MacDonald, 1995). When constructing the age structure models, the oldest 2% of the individuals in each histogram were eliminated to avoid the disproportional influence of outliers on the age model. Furthermore, a value of 1 was added to each age class to permit inclusion of empty age classes in the model (Szeicz and MacDonald, 1995).

Statistical relationships between recruitment and winter PDSI were tested by calculating Pearson correlation coefficients between the above residuals and the aforementioned reconstruction of winter PDSI (averaged over 5-year intervals to correspond with age classes). The local reconstruction was used over regional climate indicators as it represents a local manifestation of regional climate signals. Since age class patterns are partially determined by post-establishment mortality, particularly when young (Kozlowski et al., 1991), recruitment residuals may show a significant relationship with PDSI for a period following establishment. To test this, aggregated PDSI data were averaged forward in time (10-, 15-, 20-, and 25-year aggregates). Pearson correlation coefficients were then calculated between recruitment residuals and each of these forward mean classes.

Lakeside Mortality Event

Cross-dating of lakeside snags, mostly P. contorta, with a tree ring chronology from living trees proved problematic, as dry years might not result in narrow rings along the lakeside. Four steps were taken to overcome these cross-dating difficulties. First, the general timing of the mortality event was determined by comparing the decay state of the lakeside snags with an existing snag decay model for P. contorta in the eastern Sierra Nevada (Raphael and Morrison, 1987; Morrison and Raphael, 1993). Second, living lakeside species were grouped into 5-year age classes and a histogram was constructed to determine when the bulk of young, shade-intolerant P. contorta (Bartolome, 1983) seedlings and saplings established themselves along the lakeside. Third, lakeside ring-width series were cross-dated with each other using the skeleton plot method and COFECHA (non-splined) to obtain a fully cross-dated floating dead tree chronology (Holmes, 1986). Finally, this chronology was matched with several old living trees using fire scars and marker rings to determine exact dates of the mortality event.

Tree Ring–width Chronologies

After eliminating complacent records, rotted cores, and those cores that were not cross-datable, there were 25 trees and 49 ring-width series from P. contorta, 26 trees and 44 ring-width series from P. jeffreyi, and 20 trees and 33 ring-width series from J. occidentalis (Table 1). Although the oldest trees date back further, only the statistically viable portions of the tree ring chronologies (those with a subsample signal strength [SSS] of ≥0.75; Wigley et al. 1984) were used for analysis in this study. Series intercorrelation is at similar levels for all three species, although intercorrelation was highest for J. occidentalis.

Table 1

(A) ARSTAN chronology statistics. SSS  =  subsample signal strength. (B) Statistics for best fit regression for the winter Palmer Drought Severity Index (PDSI) reconstruction using the standardized P. jeffreyi ring-width chronology. CVSE  =  cross-validation standard error.

Although all three standardized chronologies tend to follow each other closely over the past 200 years, the J. occidentalis and P. jeffreyi chronologies follow less closely prior to 1750 (Fig. 2a). During this time, the J. occidentalis chronology becomes unstable, as seen by the increasing sample-size-adjusted (Osborn et al., 1997) 30-year standard deviation (Fig. 2b). The P. jeffreyi chronology regularly exhibited higher correlations with monthly indicators than the J. occidentalis chronology (monthly correlations not shown here). The P. jeffreyi chronology also showed higher monthly correlations with PDSI than with temperature or precipitation. Furthermore, PDSI records from the winter prior to the growth season exhibited the highest correlation with P. jeffreyi ring width. Therefore, the standardized P. jeffreyi chronology was used to reconstruct winter (October to February) PDSI for California Climate Division 3 from 1482 to 2002.

Figure 2

(A) 11-year running average standard chronology for each species. (B) 30-year running standard deviations for the chronologies. Variance was adjusted for sample size (Osborn et al., 1997). (C) Sample depth for standard chronologies.

The observed and estimated PDSI data follow each other closely, as the model captured year-to-year changes and general trends in the instrumental data (Fig. 3). Through Morlet wavelet analysis, the reconstructed winter PDSI chronology shows variability similar to ENSO frequencies, exhibited both by the significant contour lines between 4 and 8 years periodicity in the wavelet power spectrum (Fig. 4a) and the small but significant spike in the global wavelet centered at ∼7 years periodicity (Fig. 4c) (Diaz and Markgraf, 2000). There is also a 60–80 year periodicity from the mid-1700s to the end of the record that is marked by darker colors in the wavelet power spectrum (Fig. 4b) and a peak in the global wavelet centered at ∼70 years (Fig. 4c). The 10% significance contour line (as used by Gray et al., 2003) shows that this signal was strongest from the mid-1800s to the mid-1900s.

Figure 3

Reconstructed October to February Palmer Drought Severity Index (PDSI) using standardized P. jeffreyi chronology. (A) Reconstructed versus observed PDSI (1895–2002). (B) Reconstructed winter PDSI with an 11-year moving average.

Figure 4

(A) Reconstructed winter PDSI. (B) The Morlet 6.00 wavelet power spectrum. The contour levels are chosen such that 75%, 50%, 25%, and 5% of the wavelet power is above each level, respectively. Black contour is the 10% significance level, using a red-noise (autoregressive lag1) background spectrum. (C) The global wavelet power spectrum (black line). The dashed line is the significance for the global wavelet spectrum, assuming the same significance level and background spectrum as in (B).

The standardized P. jeffreyi chronology is significantly correlated with a 400-year reconstructed North Pacific PDO record (Fig. 5; Biondi et al., 2001; Pearson correlation coefficient  =  0.208, p ≤ 0.001), and even more highly correlated with the PDO record after 1750 (0.223, p ≤ 0.001). Radial growth trends in J. occidentalis are also significantly correlated to the PDO through the entire record (0.142, p ≤ 0.01) and after 1750 (0.139, p ≤ 0.01). Growth trends for P. contorta are not significantly correlated to the PDO. None of the chronologies exhibited a direct correlation with ENSO through time.

Figure 5

Comparison of reconstructed North Pacific Decadal Oscillation (PDO) (black; Biondi et al., 2001) and standard tree ring chronologies (gray). Graphs cover the time span of reconstructed PDO (1661–1991).

Recruitment Patterns

Recruitment of the living P. contorta trees showed an increase in successful establishment from 1840, and a peak in establishment from 1970 to 2000 (Fig. 6a). The residuals from the entire catchment exhibit a greater dip in recruitment during the early 20th century, and a spike in recruitment during the 1990s (Fig. 6b). The residuals from the entire catchment model generally have a negative correlation with reconstructed winter PDSI, with a significant correlation for the 20- and 25-year forwarded averages (Table 2). The relationship between P. contorta recruitment patterns and winter PDSI was weak away from the lakeside.

Figure 6

Recruitment patterns compared to the winter PDSI reconstruction. Figure shows 5-year age groups for the entire catchment, as well as 5-year age groups for those nodes away from the lakesides. (A) Age structure histograms and the corresponding exponential model to predict age structure given constant mortality and recruitment. (B) Standardized residuals from the exponential model. ** p ≤ 0.01, *** p ≤ 0.001.

Table 2

Pearson correlation coefficients between recruitment residuals from exponential modeling and reconstructed PDSI.

Recruitment of living J. occidentalis showed an increase in successful establishment starting ca. 1850, with a subsequent increase in recruitment beginning ca. 1940 (Fig. 6a). The residuals for each model exhibit decreased recruitment during the mid-1800s and the 1930s and increased recruitment during the 1940s and 1970s (Fig. 6b). Residuals for the entire catchment also show a spike in recruitment during the late 20th century. The residuals from the models show a positive correlation with reconstructed winter PDSI, with a significant correlation for the 15-, 20-, and 25-year forwarded averages for both models (Table 2).

The P. jeffreyi histograms were subdivided by lake catchment, as the lake catchments exhibited differing age structure patterns. Recruitment patterns of living P. jeffreyi showed an increase in successful establishment at 1900 for both lake catchments (Fig. 6a). The Hidden Lake catchment also exhibited a secondary recruitment spike during the 1960s and 1970s. Hidden Lake residuals showed a depression in recruitment from the 1930s to the 1950s, while the Roosevelt Lake residuals showed an increased recruitment during this time (Fig. 6b). Recruitment patterns in the Hidden Lake catchment displayed a significant positive correlation with reconstructed winter PDSI for the 10-, 15-, 20-, and 25-year forwarded climate averages away from the lakeside, and for the 10- and 15-year forwarded averages away for the entire catchment (Table 2). There were no significant correlations within the Roosevelt Lake catchment.

Lakeside Mortality Event

On average the dead snags along the shoreline of area lakes exhibit mostly decay stage 4 and stage 5 characteristics (absent needles, absent twigs, 1–19 limbs) within the snag decay model designed by Raphael and Morrison (1987). As these trees were likely killed by water immersion and are mostly 20 to 50 cm diameter, it is likely that the snags decayed to stage 4 within 10–20 years and to stage 5 by 20–30 years. After they reach these decay stages, they can remain standing for 20–30 years (Raphael and Morrison, 1987; Morrison and Raphael, 1993), placing the mortality event at some point in the latter half of the 20th century. Living tree recruitment at the lakeside nodes offers a clearer picture of when this event likely occurred. Successful recruitment began in the 1950s (Fig. 7). Recruitment increased by ca. 1970, especially for shade-intolerant P. contorta seedlings (Bartolome, 1983). Recruitment of P. contorta again increased by 1980.

Figure 7

Mortality patterns for dead trees around the lake edges. Length of horizontal lines show lifespan of trees. Dashed lines indicate rotted tree core and uncertain recruitment date. Numbers in parentheses indicate whether more than one radius was measured from each tree. Histogram shows 5-year age classes of living trees at lakeside nodes using combined Roosevelt Lake and Hidden Lake catchment populations. (HDL  =  Hidden Lake, RTL  =  Red Top Lake, LLL  =  Lane Lake).

The series intercorrelation for the cross-dated dead tree COFECHA chronology was 0.354 (p ≤ 0.01), and mean sensitivity was 0.234 using 31 of the most well-preserved series. Using marker rings and fire scars, this floating chronology was matched with several living trees to place the mortality event between 1952 and 1978 (Fig. 7). The snags do not show a clear pattern in recruitment date.

Ring-width Chronology

The P. jeffreyi ring-width chronology was favored over the J. occidentalis chronology for climate reconstruction for a number of reasons. The P. jeffreyi chronology regularly exhibited higher correlations with all climate variables. Furthermore, the J. occidentalis chronology became unstable toward the beginning of the record (Fig. 2b). As the record becomes older, sample-size-adjusted 30-year running standard deviation creeps upward and the chronology becomes less stable. Finally, J. occidentalis ring-width series are known to be a poor predictor of climate patterns, as their annual growth can be heavily influenced by other factors (Grissino-Mayer et al., 1997).

The P. jeffreyi ring-width chronology proved to be a good predictor of PDSI values for the previous winter (October to February). However, annual ring growth patterns only accounted for 33.6% of the variance in the winter PDSI reconstruction, meaning that other factors also influence radial growth, and the reconstruction is prone to some uncertainty. These confounding influences include other climatic factors, fire, pathogens, competition while in the understory, and possible human impacts (McBride, 1988). Furthermore, typical of many dendroclimatic reconstructions, the tree-ring model does not capture the full amplitude of annual variations (Loaiciga et al., 1993).

Although the P. jeffreyi ring-width series only explains a third of the variation in winter PDSI, the reconstruction can be used to ascertain intervals of large or persistent droughts over the past 520 years. There were substantial droughts in the record centered at ca. 1490, ca. 1580, ca. 1620, ca. 1645, ca. 1700, mid 1800s, and the 1930s, with lesser droughts from 1510 to 1545, the late 1750s, and 1780–1800. The drought in the late 16th century generally coincides with a mega-drought that has been recorded in tree ring records over much of North America (Stahle et al., 2000). Other major droughts in the record coincide with the lowest flows in a tree ring–based Sacramento River flow reconstruction (e.g. 1580s, 1650s, 1840s, and 1930s) (Meko et al., 2001). Correlation with this record is 0.225 (Pearson correlation coefficient; p ≤ 0.01). The reconstruction also matches some general trends of the bristlecone pine precipitation reconstruction from the White Mountains in eastern California and western Nevada: a prolonged drought during the first half of the 16th century, a more intense drought centered at ca. 1580, a highly fluctuating record through most of the 18th century, and a dry interval during the 1930s (Pearson correlation coefficient  =  0.100, p ≤ 0.05; Hughes and Graumlich, 1996). Finally, the reconstruction correlates (Pearson correlation coefficient  =  0.247, p ≤ 0.001) with the nearest grid-point (37.5°N, 120°W) of a 2000-yr tree ring reconstructed summer PDSI record (Cook et al., 2004) when the summer PDSI record is lagged one year.

As the wavelet analysis has shown, the chronology has both a ∼7 year periodicity throughout the record and a 60–80 year periodicity after ca. 1750. This could be due to the influence of both ENSO and PDO on winter precipitation in the region. The PDO has been shown to have a frequency domain of 60 to 70 years starting at 1650 (Mann et al., 1995), with a stronger signal along western North American beginning in the late 1700s (MacDonald and Case, 2005; Benson et al., 2003; Gedalof et al., 2002; Gedalof and Smith, 2001; Biondi et al., 2001). Although ENSO exerts an influence on winter PDSI through the reconstruction, the PDO has been shown to have an overarching influence on ENSO patterns for the multi-decadal to century time scales (McCabe and Dettinger, 1999). This is supported by the significant correlation of the P. jeffreyi and J. occidentalis chronologies with long-term PDO values. The lack of correlation with ENSO could be due to the overriding influence of the PDO in the region, or to local stand dynamics dampening the signal.

Recruitment Patterns

Although it has been shown that recruitment patterns in this area are influenced by variations in winter PDSI, this is not the only controlling factor in determining establishment and survival of the three conifers. Competition in the more dense stands can lower the variance of the recruitment/climate correlations. This is especially true in the dense P. contorta and closed canopy P. jeffreyi stands, where shade-intolerant species recruitment can be thwarted during times when climate is otherwise conducive to survival (Bartolome, 1983). Insect outbreaks and fire can also have a large influence on recruitment and survival of each species (Laudenslayer, 1988; McBride, 1988; Bartolome, 1988). Finally, other climate factors including summer heat and spring rains may influence recruitment patterns (Laudenslayer, 1988). Given all of this possible dampening of the recruitment residuals versus winter PDSI correlation, many relationships still proved significant.

The recruitment patterns of P. contorta have an inverse correlation with winter PDSI, so that this species shows positive pulses in recruitment during extended time intervals (15 to 25 years) with dry winters. This may be because P. contorta thrives after fire, which can be promoted during extended dry climatic interludes (Critchfield, 1980). It may also be because P. contorta is mostly found on flat land and poorly drained soils within the catchments, and would thus benefit from more favorable moisture conditions during droughts than those species that are found on more open slopes.

The recruitment patterns of J. occidentalis showed a strong positive correlation with winter PDSI. Because J. occidentalis is mostly found on dry open slopes within the lake catchments, water is likely the limiting factor to their survival (Royce and Barbour, 2001). Thus, during persistently wetter conditions, this species shows a pulse in hillside recruitment. Additionally, they do not compete well when in close association with other conifers (Grissino-Mayer et al., 1997). Therefore, young J. occidentalis that are found within the P. contorta forest around Hidden Lake may be out-competed during dry periods.

As with J. occidentalis, P. jeffreyi recruitment showed a positive correlation with winter PDSI within the Hidden Lake catchment. All but one of the P. jeffreyi in the Hidden Lake catchment are on south-facing slopes, and may thus be under more water stress than those that surround Roosevelt Lake. However, when the Roosevelt Lake population was separated by slope aspect, there was no discernable pattern for the south-facing slope. Intraspecific competition may be dampening the signal in this catchment, as most of the trees on the south-facing slope are in a dense, closed-canopy P. jeffreyi stand. As with the closed-canopy P. contorta stand at Hidden Lake, these trees may be benefiting from poorly drained soil.

Lakeside Mortality Event

Timing of mortality was dependent upon the lake catchment. Trees surrounding Red Top Lake died the earliest (1952–1964), as this is a small, shallow lake that likely overflowed more easily than the other lakes. Trees at Lane Lake died in the middle of this event (1960–1973), and mortality occurred around Hidden Lake during the later part of this event (1960–1978). No dead trees from Roosevelt Lake were preserved well enough for accurate dating, due to tree clearing for recreational users.

The mortality event that occurred along the shores of area lakes between 1952 and 1978 may be related to early snowmelt over much of the western United States and Canada during the last 50 years. Wilby and Dettinger (2000) have shown that in the Sierra Nevada, earlier rains on warmer snowpack could result in more flooding events dependent on the shape of the water basin. This earlier snowmelt could be due to global warming (Mote, 2003; Stewart et al., 2004) or natural climate variability, including the PDO (Cayan et al., 2001, 2004). The mortality event could also be due to generally wetter conditions in the Sierra Nevada after the 1930s (Haston and Michaelsen, 1994; Graumlich, 1993). These dead trees are likely still becoming inundated with water annually or during wetter years, evidenced by watermarks on all of the snags and hydrophytic vegetation around the snags' bases. P. contorta (the dominant lakeside snag species) can survive short intervals in waterlogged conditions, but has been shown to suffer in a more consistently waterlogged environment (Coutts and Philipson, 1978). Finally, since snags do not exhibit any discernable pattern in recruitment date, the sites were likely conducive to tree establishment for many years (from at least the mid-19th century) until the 1950–1978 mortality event.