Dendroclimatic Reconstructions

The following sections provide selected examples of how tree rings have been used to reconstruct climatic parameters in different regions of the world. This is not by any means an exhaustive review and for further information the reader should examine the sections on dendroclimatology in Bradley and Jones (1995), Jones et al. (1996), and Dean et al. (1996).

10.3.1 Temperature Reconstruction from Trees at the Northern Treeline

Many studies have shown that tree growth at the northern treeline is limited by temperature; consequently, both tree-ring widths and density provide a record of past variations in temperature. A 1500-yr long summer temperature reconstruction using Scots pine (Pinus sylvestris) from northern Scandinavia has already been discussed

(see Fig. 10.12). Even longer records may be possible because in this region logs of Holocene age have been dredged from lakes and bogs at or beyond the present tree-line. By cross-dating these samples, it should eventually be possible to construct a well-replicated chronology extending back over most of the Holocene (Zetterberg et al, 1995, 1996). Similarly, in the northern Ural mountains of Russia, the dendrocli-matic record of living trees has been extended back over 1000 yr by overlapping cores from dead larch trees (Larix sibirica) found close to or above present treeline (Graybill and Shiyatov, 1992; Briffa et al, 1995) (see Section 8.2.1). Ring-width and density studies of these samples show warm conditions in the fourteenth and fifteenth centuries, declining to uniformly low summer temperatures in the sixteenth and seventeenth centuries. There is little similarity between Fennoscandian and northern Urals data prior to -1600, after which time both records show a gradual increase in temperatures through the 20th century. In fact, in the northern Urals, the 20th century (1901-1990) is the warmest period since (at least) 914 AD, although this does not appear to be the case in northern Scandinavia. Care is needed in the interpretation of these apparent long-term changes because of the number of cores and the individual core segment lengths contributing to the overall record vary over time (Briffa et al, 1996). In the northern Urals data set, the replication (sample depth) and mean core length are at a minimum from -1400-1650 and -1500-1700, respectively. For similar reasons, reliability of the mean chronology is poor before -1100. The overall summer temperature reconstruction is therefore very sensitive to the standardization procedure used to correct for growth effects in each core segment; quite different temperature reconstructions can be produced depending on whether cores are individually standardized (by a cubic spline function) or if a regional curve standardization method is employed (Fig. 10.21). This again points to the dangers implicit in long-term paleotemperature reconstructions built up from multiple short cores when sample replication and mean record length are limited.

Shorter records, derived only from living trees have been recovered all along the Eurasian treeline by Schweingruber and colleagues (Schweingruber and Briffa,

Dendroclimatic

FIGURE 10.21 May-August mean temperature reconstruction for the northern Ural Mountains, Russia, based on tree-ring widths standardized by a cubic-smoothing spline (thin line) and maximum latewood density data standardized by deriving a regional curve for all samples (see Fig. 10.13) (thick line). Data are shown in °C anomalies from the 1951-1970 mean and are smoothed with a 25-yr lowpass filter.The large differences between the two reconstructions indicate that low frequency information is lost in the reconstruction in which only ring widths are used (Briffa et al., 1996). Note that both approaches give comparable reconstructions in the calibration and verification periods, thus providing no a priori warning that there might be a problem with the spline-standardized data set.

FIGURE 10.21 May-August mean temperature reconstruction for the northern Ural Mountains, Russia, based on tree-ring widths standardized by a cubic-smoothing spline (thin line) and maximum latewood density data standardized by deriving a regional curve for all samples (see Fig. 10.13) (thick line). Data are shown in °C anomalies from the 1951-1970 mean and are smoothed with a 25-yr lowpass filter.The large differences between the two reconstructions indicate that low frequency information is lost in the reconstruction in which only ring widths are used (Briffa et al., 1996). Note that both approaches give comparable reconstructions in the calibration and verification periods, thus providing no a priori warning that there might be a problem with the spline-standardized data set.

1996; Vaganov et al., 1996). Both ring-width and density studies of these samples enable a picture of regional variations to be built-up, from Karelia in the west to Chukotka in the east. This shows that the temperature signal is not uniform across the region; for example, from -80-150° E the early 19th century was uniformly cold, but this cool period was far less pronounced or as persistent farther west, from -50-80° E. Such variations point to possible shifts in the upper level Rossby wave circulation, which may have been amplified or displaced during this interval. Building up networks of tree-ring data in this way will eventually lead to continental-scale maps of Eurasian temperature anomalies through time (see Fig. 10.19; Schweingruber et al., 1991).

Ring widths in spruce (mainly Picea glauca) from the northern treeline of North America contain a strong record of mean annual temperature (Jacoby and D'Ar-rigo, 1989; D'Arrigo and Jacoby, 1992). Indices averaged over many sites along the treeline, from Alaska to Labrador, reveal a strong correlation with both North American and northern hemisphere temperatures over the last century (Jacoby and D'Arrigo, 1993). Based on these calibrations, long-term variations in temperature have been reconstructed (Fig. 10.22). This indicates low temperatures throughout the seventeenth century with a particularly cold episode in the early 1700s. The eighteenth century was relatively warm, followed by very cold conditions (the coldest of the last 400 years) in the early to mid-1800s. Temperatures then increased to the 1950s, but have since declined. This record is broadly similar to that derived from trees in the northern Ural mountains of Russia, but has less in common with the Fennoscandian summer temperature reconstruction discussed earlier.

Reconstructed Annual Temperatures for Northern North America

Dendroclimatic
calibration period

1600 1650 1700 1750 1800 1850 1900 1950 2000

Years

FIGURE 10.22 Reconstruction of annual temperature across northern North America for 1601-1974, based on a set of 7 tree-ring width chronologies from Alaska to Quebec; 5-yr smoothed values also shown by darker line (D'Arrigo and Jacoby, 1992).

Using a different network of trees sampled for maximum latewood density, Schweingruber et al. (1993) and Briffa et al. (1994) identify distinct regional signals along the North American treeline; by grouping records with common signals they reconstructed summer (April-September) temperatures back to A.D. 1670 for the Alaska/Yukon, Mackenzie Valley, and Quebec/Labrador regions. These series do not capture low frequency variability very well, but do show significant anomalies associated with major volcanic eruptions known from historical records (Jones et al., 1995). However, not all regions are affected in the same way; for example, 1783 (the year that Laki, a major Icelandic volcano erupted) was a very cold summer in Alaska (also noted by Jacoby and D'Arrigo, 1995) but it was relatively warm in the Mackenzie valley. By contrast, 1816 (following the eruption of the Indonesian volcano Tambora in 1815) was exceptionally cold in Quebec and Labrador, but was warm in Alaska/Yukon. Such variations indicate that volcanic aerosols do not always produce uniform climatic responses across the globe and may, in fact, produce strong meridional temperature gradients (Groisman, 1992).

10.3.2 Drought Reconstruction from Mid-latitude Trees

Drought frequency is of critical importance to agriculture in the central and eastern United States. Several dendroclimatic studies have attempted to place the limited observational record in a longer-term perspective by reconstructing precipitation or Palmer Drought Severity Indices (PDSI) for the last few centuries. Cook et al. (1992b) used a network of over 150 tree-ring chronologies from across the eastern United States to reconstruct summer PDSI for each of 24 separate regions. On average, 46% of the summer PDSI variance was explained by the tree-ring data. The entire set of reconstructed PDSI for all 24 regions was then subjected to principal components analysis to identify the principal patterns of drought distribution across the entire eastern United States. Six main patterns were identified, with each one corresponding to a different subregion affected by drought (Fig. 10.23). By examining the time series of each principal component, the drought history of each region could be assessed (Fig. 10.24). This revealed that while drought is common in some areas, it is quite rare for drought to extend over many different regions simultaneously. One exception was the period 1814-1822, when severe drought devastated the entire northeastern quadrant of the United States, from New England to Georgia and from Missouri to Wisconsin. This may have been related to circulation anomalies resulting from the major eruption of Tambora in 1815 and/or to four moderate-to-strong El Nino events in quick succession (1814, 1817, 1819, and 1821) (Quinn and Neal, 1992).

Further analysis of drought history in Texas (see Factor 4 in Fig. 10.24) using ring-width chronologies from post oak trees (Quercus stellata) has enabled the PDSI to be reconstructed separately for northern and southern Texas (Stahle and Cleave-land, 1988). From these series, the recurrence interval of different levels of drought severity can be calculated (Fig. 10.25). This shows that while the probability of moderate drought (a PDSI of -2 to -3) is high every decade in both north and south Texas, there is a >50% chance of severe drought (PDSI<-4) once a decade in south Texas; in north Texas, however, such droughts are less likely (once in 15 yr).

Dendroclimatic

FIGURE 10.23 Major drought patterns across the eastern United States, based on principal components analysis. Shaded areas are where 49% or more of the variance of each drought factor (I to 6) is explained.The temporal history of pattern 2, for example, will primarily reflect drought in New England, whereas pattern 4 will reflect conditions in Texas (Cook et al„ 1992b).

Drought Factor #1 Scores

Drought Factor #4 Scores

Drought Factor #1 Scores

Drought Factor #4 Scores

1925 1950 1975

Drought Factor #2 Scores

Drought Factor #5 Scores

Drought Factor #2 Scores

Drought Factor #5 Scores

Drought Factor #3 Scores

Drought Factor #6 Scores

Drought Factor #3 Scores

Drought Factor #6 Scores

FIGURE 10.24 Reconstructed drought history in the eastern United States. Each series corresponds to a drought pattern shown in Fig. IO.23.The drought in New England in the 1960s (drought factor 2) was unprecedented in the last 270 yr (Cook et al., 1992b).

In the southeastern United States, bald cypress trees {Taxodium distichum) growing in swamps have provided a surprisingly good record of precipitation and drought history extending back over 1000 years in some places (Stahle et al., 1985; 1988; Stahle and Cleaveland, 1992). Apparently, tree growth is affected by the water level and water quality in the swamps (oxygenation levels, pH etc.) so tree-ring records show excellent correlation with spring and early summer rainfall in the region. In fact, individual tree-ring chronologies in Georgia and the Carolinas can explain almost as much of the variance in state-wide rainfall averages as a similar network of individual rainfall records (Stahle and Cleaveland, 1992). Long-term reconstructions (>1000 yr) show that non-periodic, multidecadal fluctuations from predominantly wet to dry conditions have characterized the southeastern United States throughout the last millennium (Fig. 10.26). However, since 1650, rainfall in North Carolina appears to have systematically increased, with the period 1956-1984 being one of the wettest periods in the last 370 yr. This trend was

North Texas

North Texas

Return Interval (yr)

South Texas

South Texas

Return Interval (yr)

FIGURE 10.25 Return intervals for various levels of drought (solid line) and wetness (dashed line) in north and south Texas.Values shown are for June Palmer Drought Severity Indices ranging from ±1 to ±6, from top left to bottom right. Moderate droughts (second line from the left) have a return interval of a decade or less in both regions, whereas extreme drought (4th line from the left) can be expected with a 50% probability once every 15 years in north Texas, and once every 10 years in south Texas (Stahle and Cleaveland, 1988).

Return Interval (yr)

FIGURE 10.25 Return intervals for various levels of drought (solid line) and wetness (dashed line) in north and south Texas.Values shown are for June Palmer Drought Severity Indices ranging from ±1 to ±6, from top left to bottom right. Moderate droughts (second line from the left) have a return interval of a decade or less in both regions, whereas extreme drought (4th line from the left) can be expected with a 50% probability once every 15 years in north Texas, and once every 10 years in south Texas (Stahle and Cleaveland, 1988).

abruptly ended by two remarkable, consecutive drought years (1986 and 1987); only five other comparable two-year droughts have been registered by the bald cypress trees since A.D. 372 (Stahle et al., 1988).

Many meteorological studies have noted the relationship between El Niño/ Southern Oscillation (ENSO) events in the Pacific (or their cold event equivalents, La Niñas) and anomalous rainfall patterns in different parts of the world. These teleconnections result from large-scale displacements of major pressure systems and consequent disruptions of precipitation-bearing storm systems (Ropelewski and Halpert, 1987, 1989; Diaz and Kiladis, 1992). Several attempts have been made to identify an ENSO signal in tree-ring data in order to reconstruct a long-term ENSO index (Lough and Fritts, 1985; Lough, 1992; Meko, 1992; Cleaveland et al., 1992; D'Arrigo et al., 1994). The strongest regional signal in North America is in the

900 1100 1300 1500 1700 1900

FIGURE 10.26 Reconstructed state-wide rainfall for North Carolina (NC) in April-June and South Carolina and Georgia (SC and GA) in March-June.The upper diagram shows the annual values reconstructed; the lower diagram shows the same data smoothed to emphasize low frequency variations more clearly (with periods >30 yr) (Stahle and Cleaveland, 1992).

900 1100 1300 1500 1700 1900

YEAR

FIGURE 10.26 Reconstructed state-wide rainfall for North Carolina (NC) in April-June and South Carolina and Georgia (SC and GA) in March-June.The upper diagram shows the annual values reconstructed; the lower diagram shows the same data smoothed to emphasize low frequency variations more clearly (with periods >30 yr) (Stahle and Cleaveland, 1992).

southwestern U.S. and northern Mexico, where warm events tend to be associated with higher winter and spring rainfall, which leads to increased tree growth (Fig. 10.27). This led Stahle and Cleaveland (1993) to focus on trees from that area to reconstruct a long-term South Oscillation Index (SOI) back to 1699. Although their results showed considerable skill in identifying many major ENSO events in the past (in comparison with those known from historical records) they estimate that only half of the total number of extremes were clearly defined over the last 300 yr. Similar problems were encountered by Lough and Fritts (1985) using a network of aridsite trees from throughout the western United States. The principal SOI extremes they identify for the last few centuries are not the same as those selected by Stahle and Cleaveland's analysis. This points to the problem of characterizing ENSOs from teleconnection patterns, which are spatially quite variable in relation to both positive and negative extremes of the Southern Oscillation. Consequently, although

Dendroclimatic

FIGURE 10.27 Tree-ring anomalies across western North America associated with the high and low phases of the Southern Oscillation.The pattern on the left is associated with El Niños; teleconnections lead to heavier winter and spring rainfall in northern Mexico and the U.S. Southwest. By contrast, cold events in the Pacific (La Niñas) are associated with drier conditions and reduced tree growth in the same region (Lough, 1992).

FIGURE 10.27 Tree-ring anomalies across western North America associated with the high and low phases of the Southern Oscillation.The pattern on the left is associated with El Niños; teleconnections lead to heavier winter and spring rainfall in northern Mexico and the U.S. Southwest. By contrast, cold events in the Pacific (La Niñas) are associated with drier conditions and reduced tree growth in the same region (Lough, 1992).

there may be an overall signal related to ENSO in one region, precipitation patterns vary enough from event to event that precise, yearly ENSO reconstructions are very difficult (Lough, 1992; D'Arrigo et ai, 1994).

One additional factor related to drought is the frequency of fire in some areas (Swetnam, 1993). Fire scars damage the cambium of trees and are clearly visible in tree sections. By building up a chronology of fire history from non-contiguous regions, the frequency of large-scale fires affecting wide areas (related to regional drought episodes) can be identified. In California, fires affecting widely separated groves of giant Sequoia are associated with significant negative winter/spring precipitation anomalies. Over the last 1500 yr, fire frequency was low from A.D. 500-800, reached a maximum ~A.D. 1000-1300, then generally declined. Interestingly, the relationship with temperature in the region is not significant on an annual basis, but over the long-term fire frequency and temperature are positively correlated. Swetnam (1993) attributes this to long-term temperature fluctuations controlling vegetation changes on decadal to century timescales, whereas fire activity from year to year is more related to fuel moisture levels, which are highly correlated with recent precipitation amounts. In the southwestern United States, dry springs and extensive fires are associated with "cold events" in the Pacific (La Niñas) as a result of related circulation anomalies that block moisture-bearing winds from entering the region (Swetnam and Betancourt, 1990). Thus, tree-ring widths and fire occurrence are manifestations of large-scale teleconnections linking sea-surface temperatures in the tropical Pacific to rainfall deficits in the arid Southwest.

10.3.3 Paleohydrology from Tree Rings

Tree rings can be used to reconstruct climate-related phenomena that in some way integrate the effects of the climate fluctuations affecting tree growth. In particular, much work has been devoted to paleohydrological reconstructions involving streamflow. Stockton (1975) was interested in reconstructing long-term variations in runoff from the Colorado River Basin, where runoff records date back only to 1896. As runoff, like tree growth, is a function of precipitation, temperature, and evapotranspiration, both during the summer and in the preceding months, it was thought that direct calibration of tree-ring widths in terms of runoff might be possible. Using 17 tree-ring chronologies from throughout the watershed, eigenvectors of ring-width variation were computed. Stepwise multiple regression analysis was then used to relate runoff over the period 1896-1960 to eigenvector amplitudes over the same interval. Optimum prediction was obtained using eigenvectors of ring width in the growth year (tQ) and also in years 12, t v and t v each of which contained climatic information related to runoff in year tQ. In this way an equation accounting for 82% of variance in the dependent data set was obtained; the reconstructed and measured runoff values are thus very similar for the calibration period (Fig. 10.28). The equation was then used to reconstruct runoff back to 1564, using the eigenvector amplitudes of ring widths over this period (Fig. 10.29). The reconstruction indicates that the long-term average runoff for 1564-1961 was -13 million acre-feet (~16 X 109 m3) over 2 million acre-feet (-2.47 X 109 m3) less than during the period of instrumental measurements. Furthermore, it would appear that droughts were more common in this earlier period than during the last century, and the relatively long period of above average runoff from 1905 to 1930 has only one comparable period (1601-1621) in the last 400 yr. Stockton argues that these estimates, based on a longer time period than the instrumental observations, should be seriously considered in river management plans, particularly in regulating flow through Lake Powell, a large reservoir constructed on the Colorado River. In this

1900

1910

1920

1930

1940

1950

I960

FIGURE 10.28 Runoff In the Upper Colorado River Basin. Reconstructed values (-----) are based on tree-

ring width variations in trees on 17 sites in the basin. Actual data, measured at Lee Ferry, Arizona, are shown for comparison (-). Based on this calibration period, an equation relating the two days sets was developed and used to reconstruct the flow of the river back to 1564 (Fig. 10.29) (Stockton, 1975).

1900

1910

1920

1930

1940

1950

I960

FIGURE 10.28 Runoff In the Upper Colorado River Basin. Reconstructed values (-----) are based on tree-

ring width variations in trees on 17 sites in the basin. Actual data, measured at Lee Ferry, Arizona, are shown for comparison (-). Based on this calibration period, an equation relating the two days sets was developed and used to reconstruct the flow of the river back to 1564 (Fig. 10.29) (Stockton, 1975).

O-l-i-.-1-1-,-,-1-.-1-.-1-1-,-1-.-1-1-.-1-1-■-.-.-,-1-1-.-.-1-.-1-1-1-1-1-1-1-■-1-

1564 1600 1600 1700 1750 1800 1850 1900 1950

Year

FIGURE 10.29 Annual virgin runoff of the Colorado River at Lee Ferry, as reconstructed using ring-width index variation, calibrated as shown in Fig. 10.28. Growth for each year, and the three following years, was used to estimate water flow statistically. Smooth curve (below) represents essentially a 10-yr running mean. Runoff in this period ~ 1905-1925 was exceptional when viewed in the context of the last 400 yr (Stockton, 1975).

O-l-i-.-1-1-,-,-1-.-1-.-1-1-,-1-.-1-1-.-1-1-■-.-.-,-1-1-.-.-1-.-1-1-1-1-1-1-1-■-1-

1564 1600 1600 1700 1750 1800 1850 1900 1950

Year

FIGURE 10.29 Annual virgin runoff of the Colorado River at Lee Ferry, as reconstructed using ring-width index variation, calibrated as shown in Fig. 10.28. Growth for each year, and the three following years, was used to estimate water flow statistically. Smooth curve (below) represents essentially a 10-yr running mean. Runoff in this period ~ 1905-1925 was exceptional when viewed in the context of the last 400 yr (Stockton, 1975).

case, dendrohydrological analysis provided a valuable long-term perspective on the relatively short instrumental record. Similar work has been accomplished by Stockton and Fritts (1973), who used tree-ring eigenvectors calibrated against lake-level data to reconstruct former levels of Lake Athabasca, Alberta, back to 1810 (Figure 10.30). Their reconstruction indicated that although the long-term average lake level is similar to that recorded over the last 40 yr, the long-term variability of lake levels is far greater than could be expected from the short instrumental record. To preserve this pattern of periodic flooding, essential to the ecology of the region, the area is now artificially flooded at intervals that the dendroclimatic analysis suggests have been typical of the last 160 yr.

A few studies have attempted to reconstruct streamflow in humid environments, but they have been less successful than dendroclimatic reconstructions in more arid areas (Jones et al., 1984). In humid regions, periods of low flow are generally more reliably reconstructed than high flows, as trees are more likely to register prolonged drought than heavy precipitation events that may lead to high streamflow and flooding. Cook and Jacoby (1983) reconstructed summer discharge of the Potomac River near Washington, DC, back to 1730 using a set of 5 tree-ring width chronologies; their results indicate that there was a shift in the character of runoff around 1820. Before this time, short-term oscillations about the median flow were common, generally lasting only a few years. After 1820 more persistent, larger amplitude anomalies became common. For example, runoff was persistently below the long-term median from 1850-1873. If such an event were to be repeated in the future with all of the modern demands on water from the Potomac River, the consequences would be extremely severe. Such a perspective on natural variability of the hydrological system is thus invaluable for water supply management and planning.

indicate that prior to 1935 there was greater variability in lake levels during May and July, but there was less variability in lake levels for September than during the recent calibration period. Dots indicate actual lake levels used for calibration. Lines connect the three estimates from tree rings, representing mean lake level for May 21 -30, July I I -20, and September 21-30. Points are not connected over the winter season, as calibrations of levels for the frozen lake could not be made (Stockton and Fritts, 1973).

indicate that prior to 1935 there was greater variability in lake levels during May and July, but there was less variability in lake levels for September than during the recent calibration period. Dots indicate actual lake levels used for calibration. Lines connect the three estimates from tree rings, representing mean lake level for May 21 -30, July I I -20, and September 21-30. Points are not connected over the winter season, as calibrations of levels for the frozen lake could not be made (Stockton and Fritts, 1973).

0 0

Post a comment