Paleoclimate records are essential tools for contextualizing 20th century climate changes. By discerning natural patterns and ranges of climate change in the centuries preceding any possible anthropogenic forcing we will refine our understanding of the climate system’s present and future response to hypothesized global warming. In particular, well-dated, monthly to annually-resolved proxy climate records such as corals, tree-rings, and select marine sediments offer multi-century perspectives on the most pressing climate questions of today, including decadal to centennial climate variability. Many coral-based climate reconstructions spanning the last few centuries present coherent pictures of tropical climate on interannual and decadal time-scales, while tree-ring and other paleoclimate records from the last millennium point to episodes of severe, prolonged drought in North America that might have implications for future climate change. High-resolution records that allow glimpses of climate beyond the last millennium are rare yet provide valuable opportunities to address the impact of mean climate state changes on modes of natural variability.
Tropical climate variability is dominated by ENSO, which strongly affects extratropical climate patterns around the globe. The interaction of the ENSO mode with longer-term decadal to centennial fluctuations of climate, including proposed increases of the global mean temperature during the late twentieth century, is key to our understanding of the mechanisms and consequences of these low-frequency modes of variability. In this regard, coral proxy climate records of 100 - 300 years in length are a valuable supplement to the instrumental data of the tropical Pacific, which are incomplete before the mid-twentieth century. Long-lived coral can be found in most areas of the tropical oceans, providing near-monthly resolution of oceanic properties locked into the chemistry of the coral's skeleton.
Numerous geochemical proxies exist to extract environmental information from the coral skeleton, which yield records of SST, upwelling variability, salinity/precipitation, and riverine runoff. For instance, oxygen isotopic variability is commonly used to reconstruct SST, as the two isotopes of oxygen, 16O and 18O, are incorporated into the skeleton as a function of temperature and salinity; in most regions of the tropical surface oceans the latter is strongly correlated to precipitation. This dual dependence on temperature and salinity is especially useful in reconstructing tropical climate variability, as elevated SST anomalies usually drive anomalous convection, and thus the signal is amplified in the coral record.
A 160 year oxygen isotopic record from a Maiana coral (1oN, 173oE) contains a dual signature of increased SST and increased rainfall associated with ENSO, decadal-scale variability, and a strong 20th century trend towards a warmer/wetter climate (Urban et al. 2000). Evolutive spectral analysis of this record suggests that while interannual variability is present throughout, decadal variability is more dominant during the 19th Century. Also, the period between 1920 and 1960 was characterized by reduced variance (Cole et al. 1993). The coral record supports the hypothesized regime shift of 1976, and hints at a similar, but less abrupt, climate transition during 1900-1915.
Comparing a composite of western equatorial Pacific coral SST reconstructions to instrumental ENSO indices from the 20th century (NINO3.4 (Kaplan et al. 1998) and the multi-variate ENSO index (Wolter and Timlin 1998)) show that the corals are reliable recorders of regional climate, with correlation coefficients of 0.6 to 0.9. High coherence in ENSO-decadal bands of cross spectral estimates using Pacific corals, from Maiana and Galapagos (Dunbar et al. 1994), and Indian Ocean corals from Seychelles (Charles et al. 1997) and Malindi (Cole et al. 2000), suggest strong teleconnections between the Pacific and Indian Oceans on both ENSO and decadal timescales. If the Indo-Pacific domain behaves similarly on ENSO and decadal timescales, then a Pacific based forcing for both modes of variability is plausible.
A proposed player in ENSO and decadal mechanisms of climate variability in the tropics is the depth of the seasonal thermocline, and its movements can be approximated by changes in upwelling intensity, which corals record in the form of radiocarbon concentrations. During periods of high upwelling, corals record anomalously low 14C , as the deeper, older waters brought to the surface contain less 14C . In turn, periods of reduced upwelling shut off this supply of older waters, and surface waters, and the corals growing therein, will record relatively higher 14C . Radiocarbon records from corals in upwelling-sensitive areas, such as the Galapagos Islands, show fluctuations associated with El Niño (reduced upwelling) and La Niña (enhanced upwelling) as expected (Guilderson and Schrag 1998). Super-imposed on this high-frequency signal is a jump in radiocarbon values that suggests a reduction in upwelling after 1976, consistent with the theory of a global climate shift during this time (Graham 1994).
Corals that pre-date the last few centuries are rare, but some fossil sequences have been recovered that date from the last millennium to 124 kybp (thousand years before present). These fossil corals, typically comprising 30-70 years of record, provide windows on tropical climate characteristics during periods in which other proxy records, (deep-sea sediments, ice cores, and tree rings) show that the global mean climate was different than that of today. For example, at 124 kybp sea level and perhaps global temperatures were higher than at present, yet a 65 year coral record from the Western tropical Pacific yields a power spectrum on interannual timescales that is very similar to that of NINO3, a 20th century tropical Pacific SST index that captures ENSO variability (Hughen et al. 1999).
In extratropical regions the annual growth-rings of trees provide long proxy records of climate variability. Insights into the long-term relationships between tropical climate variability and extratropical drought can be gained by the comparison of tree-ring records with both instrumental and coral records of ENSO variability. A network of ~440 precisely dated tree-ring chronologies from across the coterminous United States was used to develop a gridded annual database of mean summertime Palmer Drought Severity Index (PDSI) spanning the last 300 years (Cook et al. 1994; Cook et al. 1999). The PDSI is an index of moisture availability that is estimated from observations of precipitation and temperature, which allows for persistence from previous seasons (Palmer 1965). A cross-spectral analysis of the reconstructed PDSI and the Maiana d18O record shows that the two series have similar spectra, with very good coherence at interannual and decadal wavelengths.
Correlation analysis of the Southern Oscillation Index (SOI) and the observed PDSI record for the period from 1895 to 1978 reveals centres of action in the United States where enhanced or reduced drought conditions are associated with El Niño or La Niña events (Cole and Cook 1998). In particular, there is a tendency for the American southwest and Great Plains regions to be wet (dry) in response to El Niño (La Niña) events, and the mid-Atlantic states to be dry (wet). A correlation map of the coral-derived ENSO index with the tree-ring derived PDSI for the same interval results in very similar patterns in the southwest and mid-Atlantic regions, but underestimates the strength of the relationship in the Great Plains. These results suggest that, at least in these regions, the paleorecord provides a robust measure of the relationship between ENSO and drought in the continental United States.
Changes in the nature of this relationship can be seen when the correlation maps are calculated for 30 year intervals, overlapped by 15 years, for the complete duration of the record. Two dominant patterns emerge from this analysis. First, the northeastward penetration of ENSO related moisture anomalies was more extensive prior to ca. 1935 and after ca. 1965 than it was during this interval. Second, at ca. 1920 the association between ENSO and drought in the mid-Atlantic reverses sign, revealing a bipolar structure in the maps for the latter portion of the record. Two potential causes for these changes are (1) changes in the character of ENSO episodes over time; and (2) changes in extratropical influences.
The first of these scenarios is supported by analysis of the variance in the instrumental SOI and coral reconstructions, which shows that there is a general weakening of ENSO at ca. 1925, and stronger variability since ca. 1955 (Trenberth and Shea 1987; Cole et al. 1993). This interval corresponds well to the period of reduced penetration of ENSO-related moisture anomalies into the United States, and supports the results of modeling studies which theorize that stronger ENSO anomalies will generate more consistent and stronger teleconnection patterns (Kumar and Hoerling 1997).
In the North Pacific, decadal variations in SST and the overlying atmosphere, generally termed the Pacific Decadal Oscillation (PDO; Zhang et al. 1997), have significant impacts on climate over much of North America and may constrain the range of ENSO teleconnections (Gershunov and Barnett 1998). Differentiating the effects of ENSO from those of the PDO is complicated by the strong correlation that exists between indices of the two processes. In order to independently evaluate the influence of ENSO and the PDO on drought, two new grids of the PDSI were generated – one with the linear relationship with ENSO removed from it; and one with the linear relationship with the PDO removed from it. These two residual series were then regressed against the index of interest to identify the association that is unique to either the ENSO or PDO. This analysis shows that ENSO is most strongly associated with southwest moisture anomalies with a weak northeastward extension, but has virtually no influence in the mid-Atlantic region of the United States. Conversely, the PDO drought signature is strongest in the mid-Atlantic and Pacific Northwest, but has virtually no impact on the southwest. An analysis of the variance in the PDO index shows that the bipolar structure in the mid-Atlantic region emerges during intervals of enhanced PDO strength – notably after ca. 1920. Curiously, the baseline shifts in the PDO which occurred in 1925, 1947 and 1976 (Mantua et al. 1997) do not seem to have an effect on the PDO - drought relationship.
Although the association between ENSO events and drought in the continental United States does provide cause for optimism in the area of forecasting, the paleorecord also shows that there are exceptions to the observed associations. For instance, the Maiana coral record shows evidence of two extended La Niña events – one lasting from 1855 - 1865, and one from 1879 - 1887. Although the first of these events corresponds to a period of prolonged drought in the American southwest, the second does not. This difference may reflect changes in the behavior of the PDO between these two intervals.
Additional insight into the frequency and persistence of drought events can be derived from the analysis of multiple proxy data sources. Woodhouse and Overpeck (1998) review paleoclimate analyses of tree-rings, lake levels, varved sediments, historical documents, archaeological remains and geomorphic events. These data provide coherent evidence of pre-instrumental drought events over the last 2000 years. In this context, the droughts of the 1930s and 1950s are not unprecedented in either their duration or magnitude, and at least two events occurred which were longer, more severe, and more extensive than anything seen in the twentieth century. The more recent of these "megadroughts" occurred in the western United States in the latter part of the 1500s. Tree-ring evidence suggests that it began at ca. 1565 in the southwest, spread throughout the entire western United States by 1585 (Fritts 1965), and ended at ca. 1600. The timing and extent of these findings are supported in other tree-ring studies of precipitation, drought and river flow, lake sediments, the timing of tree stand establishment and sand dune activation. Another drought occurred at ca. 1276 and lasted about 40 years (Woodhouse and Overpeck 1998). This drought is sometimes referred to as the "Great Drought" and coincided with the abandonment of Anasazi settlements in the Four Corners region, redistribution of populations, and societal disruptions. Tree stumps rooted in modern lakes in the Sierra Nevada, California, have been radiocarbon dated to this time, suggesting that lake levels were substantially lower than anything observed in the recent record and that low lake levels persisted for many decades.
The severity and extent of the sixteenth century megadrought can be compared to the droughts of the twentieth century independently using the tree-ring network of Stahle et al. (2000). This network of drought sensitive trees from northwestern Mexico, the Great Plains, western United States, and southwestern Canada, shows evidence of the “Dustbowl” drought of the 1930s in western North America, and the Texas drought of the 1950s. The drought of the late 1500s, however, is both longer and more extensive than either of these droughts, lasting up to 40 years at some locations, and influencing the entire study area for at least a decade in the 1570s.
The paleoclimate record provides a context in which to place twentieth century climate variability, but it also often presents scenarios that defy explanation or have no modern analogue. These discoveries challenge the paleoclimate community to verify their findings at other locations and using other proxy indicators. When these findings are found to be robust they challenge the scientific community at large to explain their occurrence.
Sediment records from the Cariaco Basin, in the southern Caribbean, provide a millennial record of ocean-atmosphere variability in the North Atlantic Basin (Black et al. 1999). These records provide evidence of a strong 12 to 13 year mode of variability, as well as century-scale variability and episodes of abrupt change in circulation. These findings corroborate results from analyses of North American lake sediments, tree-rings, and from Greenland ice cores. While the decadal variability is also found in the instrumental record, both the lower frequency mode of variability and the abrupt regime shifts are unique to paleoclimate records.
Sediment records from Lake Naivasha, Kenya, provide an 1100 year proxy record of rainfall and drought in east Africa (Verschuren et al. 2000). This record shows that, for instance, the Medieval Warm Period was significantly drier than today, whereas the Little Ice Age was wetter. The study also showed that abrupt transitions can occur in the rainfall trends. Although the Little Ice Age was generally wetter than modern conditions, it was interrupted on three occasions by episodes of drought more severe than any recorded drought of the twentieth century. The intervals of prolonged drought identified in this record correspond to periods of drought-related political upheaval recorded in oral tradition. Extended wet periods are associated with "ages of prosperity"that are described in oral histories. Corroborating evidence can be found in records of the Nile River discharge.
These examples of pre-instrumental climate illustrate how the range of natural variability is often well beyond what we know from the twentieth century, but also how the character of that variability is often different from what we currently see. The paleorecord is still not particularly well described, though. There is a need for more and better observations of pre-instrumental climate variability, including better replication of existing proxy records, new techniques for interpreting proxy data, and new approaches to proxy climate analysis. Coupled with developments in paleoclimate analysis there is a need for a better understanding of the natural variability of the climate system, including the mechanisms and feedbacks that give rise to the observed spatial-temporal patterns. Lastly, there is a need for improved comprehension of societal vulnerabilities to dec-cen variability.