Global Signatures and Dynamical Origins of the Little Ice Age and Medieval Climate Anomaly

See allHide authors and affiliations

Science  27 Nov 2009:
Vol. 326, Issue 5957, pp. 1256-1260
DOI: 10.1126/science.1177303

Patterns of Change

The global climate record of the past 1500 years shows two long intervals of anomalous temperatures before the obvious anthropogenic warming of the 20th century: the warm Medieval Climate Anomaly between roughly 950 and 1250 A.D. and the Little Ice Age between around 1400 and 1700 A.D. It has become increasingly clear in recent years, however, that climate changes inevitably involve a complex pattern of regional changes, whose inhomogeneities contain valuable insights into the mechanisms that cause them. Mann et al. (p. 1256) analyzed proxy records of climate since 500 A.D. and compared their global patterns with model reconstructions. The results identify the large-scale processes—like El Niño and the North Atlantic Oscillation—that can account for the observations and suggest that dynamic responses to variable radiative forcing were their primary causes.


Global temperatures are known to have varied over the past 1500 years, but the spatial patterns have remained poorly defined. We used a global climate proxy network to reconstruct surface temperature patterns over this interval. The Medieval period is found to display warmth that matches or exceeds that of the past decade in some regions, but which falls well below recent levels globally. This period is marked by a tendency for La Niña–like conditions in the tropical Pacific. The coldest temperatures of the Little Ice Age are observed over the interval 1400 to 1700 C.E., with greatest cooling over the extratropical Northern Hemisphere continents. The patterns of temperature change imply dynamical responses of climate to natural radiative forcing changes involving El Niño and the North Atlantic Oscillation–Arctic Oscillation.

Considerable progress has been made over the past decade in using climate “proxy” data to reconstruct large-scale trends in past centuries, and in using climate models to assess the roles of natural and anthropogenic forcing in those trends (1). Owing in part to the sparseness of the available proxy data, less progress has been made in identifying the underlying spatial patterns of those changes, let alone the causal factors behind them. Yet a better understanding of past patterns of climate change and their causes (e.g., the role of past changes in the El Niño–Southern Oscillation, or ENSO) may be even more important for validating the regional-scale projections, which are paramount in assessing future climate change impacts.

Patterns of past climate change can be estimated through the simultaneous analysis of multiple spatially distributed proxy records. Such analyses have been performed via statistical reconstruction (28) and model assimilation approaches (9), but available proxy networks have been insufficient for estimating spatially resolved large-scale temperature reconstructions beyond the past few centuries (2, 4, 7).

Here, we employ a diverse multiproxy network previously used to estimate global and hemispheric mean annual temperature trends (10) to reconstruct global patterns of surface temperature changes over the past 1500 years. We use a climate field reconstruction (CFR) approach (11) that has been rigorously tested with synthetic “pseudoproxy” networks generated from forced climate model simulations (12). We interpret the resulting reconstructions in the context of results from climate model simulations forced by estimated past changes in natural (solar and volcanic) radiative forcing.

We employ the global proxy data set used by (13) comprising more than a thousand tree-ring, ice core, coral, sediment, and other assorted proxy records spanning the ocean and land regions of both hemispheres over the past 1500 years. The surface temperature field is reconstructed by calibrating the proxy network against the spatial information contained within the instrumental annual mean surface temperature field (14) over a modern period of overlap between proxy and instrumental data (1850 to 1995) using the RegEM CFR procedure (12) with additional minor modifications. Further details of the reconstruction procedure, associated statistical validation and skill assessments, uncertainty estimation procedures, data used, and MATLAB source codes for the analysis procedures are provided in the Materials and Methods.

Earlier proxy-based large-scale surface temperature reconstructions (2, 15) resolved only a single statistical degree of freedom before the 15th century, precluding the possibility of investigating spatial patterns of surface temperature variation in earlier centuries. By contrast, the current reconstructions resolve multiple degrees of freedom in the surface temperature field back through the 6th century, allowing us to meaningfully interpret spatial features in the reconstructions. Nonetheless, certain caveats must be kept in mind in interpreting the proxy-based surface temperature reconstructions. Before 1600 C.E., the low-frequency component of the surface temperature reconstructions is described as a linear combination of just two leading patterns of temporal variation, so that regional features in the temperature field are represented by a spatiotemporally filtered approximation. Moreover, as decadal-resolution proxy data were used in addition to annual-resolution data, only interdecadal and longer-term variations are meaningfully resolved; i.e., the details of individual years and even individual decades should not be emphasized. Thus, it is the longer-term, and larger-scale, variations resolved by the reconstructions that are most meaningful.

The large-scale surface temperature reconstructions, when spatially averaged, e.g., over the Northern Hemisphere, yield a long-term history very similar to the hemispheric mean reconstructions of (13) (Fig. 1A). However, the spatial reconstructions can also be averaged to yield other indices of interest [Fig. 1, B to E; other regional average series are shown in the Supporting Online Material (SOM) Text]. Though there are relatively few distinct patterns of variation resolved by the reconstructions, particularly before 1600 C.E., there are notable differences of behavior among the various diagnosed indices. The Atlantic Multidecadal Oscillation (AMO) series, for example, is marked by substantial multidecadal variability, consistent with previous proxy studies of North Atlantic variability [e.g., (16)]. The high-frequency fluctuations of the Niño3 series are consistent with the oscillatory nature of ENSO. The Niño3 index suggests strong and persistent La Niña conditions around 1000 years ago, as discussed further below.

Fig. 1

Decadal surface temperature reconstructions. Surface temperature reconstructions have been averaged over (A) the entire Northern Hemisphere (NH), (B) North Atlantic AMO region [sea surface temperature (SST) averaged over the North Atlantic ocean as defined by (30)], (C) North Pacific PDO (Pacific Decadal Oscillation) region (SST averaged over the central North Pacific region 22.5°N–57.5°N, 152.5°E–132.5°W as defined by (31)], and (D) Niño3 region (2.5°S–2.5°N, 92.5°W–147.5°W). Shading indicates 95% confidence intervals, based on uncertainty estimates discussed in the text. The intervals best defining the MCA and LIA based on the NH hemispheric mean series are shown by red and blue boxes, respectively. For comparison, results are also shown for parallel (“screened”) reconstructions that are based on a subset of the proxy data that pass screening for a local temperature signal [see (13) for details]. The Northern Hemisphere mean Errors in Variables (EIV) reconstruction (13) is also shown for comparison.

Our reconstructions span two climatologically interesting periods, the so-called Little Ice Age (LIA) and Medieval Climate Anomaly (MCA). For the purpose of investigating the associated spatial patterns (Fig. 2), we defined the LIA and MCA in terms of distinct three-century-long intervals (1400 to 1700 C.E. and 950 to 1250 C.E., respectively), which both correspond to relative cold and warm hemispheric conditions, respectively (Fig. 1), and are distinct with regard to the estimated external radiative forcing of the climate (1, 17). The observed patterns are not, however, sensitive to the precise time intervals used to define these periods (fig. S9). The MCA pattern is based on a smaller number of predictors than the LIA pattern (Fig. 2) and, accordingly, on fewer resolved spatial degrees of freedom (SOM Text).

Fig. 2

Reconstructed surface temperature pattern for MCA (950 to 1250 C.E.) and LIA (1400 to 1700 C.E.). Shown are the mean surface temperature anomaly (left) and associated relative weightings of various proxy records used (indicated by size of symbols) for the low-frequency component of the reconstruction (right). Anomalies are defined relative to the 1961–1990 reference period mean. Statistical skill is indicated by hatching [regions that pass validation tests at the P = 0.05 level with respect to RE (CE) are denoted by / (\) hatching]. Gray mask indicates regions for which inadequate long-term modern observational surface temperature data are available for the purposes of calibration and validation.

The reconstruction skill diagnostics suggest that the MCA and LIA reconstructions are most reliable (Fig. 2) over the Northern Hemisphere and tropics, and least reliable in the Southern Hemisphere, particularly in the extratropics. To assess if the larger-scale features of the earlier MCA pattern are robust, we used only the more restricted network of proxy data available back through the beginning of the MCA interval to reconstruct temperatures for the LIA interval. This analysis gave a reconstruction very similar to the LIA reconstruction based on the full data set (fig. S10).

The reconstructed MCA pattern is characterized by warmth over a large part of the North Atlantic, Southern Greenland, the Eurasian Arctic, and parts of North America, which appears to substantially exceed that of the modern late–20th century (1961–1990) baseline and is comparable to or exceeds that of the past one-to-two decades in some regions. This finding is consistent with that of a recent tree-ring–based study of high-latitude Eurasian temperatures (18). Relative warmth in the central North Pacific MCA is consistent with the expected extratropical signature of the strong observed La Niña–like pattern in the tropical Pacific (strong cooling in the east and warming in the west). Certain regions, such as central Eurasia, northwestern North America, and (with less confidence) parts of the South Atlantic, exhibit anomalous coolness. The LIA pattern is characterized primarily by pronounced cooling over the Northern Hemisphere continents, but with some regions—e.g., parts of the Middle East, central North Atlantic, Africa, and isolated parts of the United States, tropical Eurasia, and the extratropical Pacific Ocean—displaying warmth comparable to that of the present day. In some places, e.g., northern Labrador, apparent LIA warmth is a product, at least in part, of the relatively cool nature of the 1961–1990 reference period in the region.

For comparison with model simulation results, it is useful to eliminate the influence of the choice of modern reference period by examining the pattern of the MCA-LIA difference itself (Fig. 3). The MCA-LIA pattern highlights the extent to which the MCA is both more “La Niña–like” [e.g., (17, 1921)] and, with enhanced warmth over interior North America and the Eurasian Arctic, and cooling over central Eurasia, suggestive of the positive phase of the North Atlantic Oscillation (NAO) and closely related Arctic Oscillation (AO) sea-level pressure (SLP) pattern (17, 22), as discussed further below.

Fig. 3

Spatial pattern of MCA-LIA surface temperature difference in reconstructions and model simulations. (A) Proxy-based temperature reconstructions, (B) GISS-ER (using the same solar forcing difference used in the NCAR simulation—shown is the ensemble mean; see the SOM for example results from one of six realizations), and (C) NCAR CSM 1.4 simulation (using the same MCA and LIA time intervals as defined above). The observational mask has been applied to both model patterns for ease of comparison. Statistical skill for (A) is indicated with the same conventions as in Fig. 2 (statistical significance here indicates that the particular test statistic independently passed during both the MCA and LIA intervals).

We examined results for two different coupled model simulations of the past millennium, driven with those factors (solar irradiance changes and stratospheric aerosols from explosive volcanic eruptions) that can most plausibly explain the climate changes of the past millennium (17): (i) the National Center for Atmospheric Research (NCAR) Climate System Model (CSM) 1.4 coupled model driven with estimated solar plus volcanic forcing over the past millennium [see (23) for details]; and (ii) the Goddard Institute for Space Studies–ER (GISS-ER) coupled model with solar (but no volcanic) forcing (SOM Text), scaled for an MCA-LIA solar radiative forcing at the tropopause of 0.37 W/m2 (equivalent to the MCA-LIA solar forcing difference used in the NCAR simulation). Both simulations give very similar estimates of the global mean MCA-LIA temperature difference (0.16° and 0.24°C for NCAR and GISS, respectively; the latter is identical to the proxy reconstructed mean surface temperature difference of 0.24°C). The spatial patterns of response for the two models (Fig. 3), however, are quite different, as discussed further below.

The La Niña–like nature of the MCA-LIA pattern is not reproduced in either of the two different coupled model simulations analyzed. On the other hand, such a pattern is reproduced in simulations (19) using the low-order Cane-Zebiak (24) model of the tropical Pacific coupled ocean-atmosphere system. The discrepancy in the model responses may arise because the tropical Pacific “thermostat” mechanism (25) is not active in either the NCAR or GISS simulations. In (19), this mechanism is responsible for the La Niña–like response to the positive tropical radiative forcing of the MCA that arises from a combination of relatively high solar irradiance and inactive tropical volcanism. Although there is still a vigorous debate regarding the nature of the response of the tropical Pacific to anthropogenic radiative forcing [e.g., (26)], paleoclimate evidence examined here, as elsewhere [e.g., (19, 27)], appears to support a thermostat-like response, at least for natural radiatively forced climate changes in past centuries.

The NCAR simulation also does not reproduce the enhanced warming over the Eurasian Arctic, high-latitude North Atlantic, and North American region evident in the reconstructed MCA-LIA pattern. As discussed previously, this surface temperature pattern is consistent with a relative positive (negative) NAO-AO atmospheric circulation anomaly during the MCA (LIA), associated with annular bands of positive (negative) SLP anomalies in the subtropics and mid-latitudes, and negative (positive) SLP anomalies in the subpolar latitudes. Such a pattern has been inferred in paleoclimate studies of the past millennium (5, 17, 22, 28, 29), and the negative phase of this pattern has been produced as a dynamical response to decreased solar radiative forcing during the LIA using a previous version of the NASA GISS model that incorporates the effects of ozone photochemistry on the vertical structure of the atmosphere (28, 29). These effects are not accounted for in the NCAR simulation, which is limited to 36 km in vertical extent. The GISS-ER model used here extends to ~80 km and does incorporate these processes and, indeed, reproduces roughly the observed pattern of enhanced North American, high-latitude North Atlantic, and Arctic Eurasian warming, as a dynamical response to the imposed radiative forcing. These surface temperature changes are, in turn, associated with an annular atmospheric circulation response (Fig. 4) reminiscent of the positive phase of the NAO-AO pattern, though with some differences [in particular, (i) the high- and low-pressure regions in the North Atlantic sector are somewhat asymmetric and geographically shifted relative to the conventional pattern—hence, for example, the relative absence of warming in western Europe; and (ii) there is a positive SLP anomaly over Northern Greenland and part of the Eurasian Arctic Ocean that is absent in the conventional pattern]. Comparisons over the Pacific sector and neighboring regions, by contrast, are of limited utility, given the inability of the GISS-ER model to reproduce the aforementioned La Niña–like feature of MCA-LIA pattern, which strongly affects the Pacific basin. There is no evidence of a positive NAO-AO response in the NCAR simulation (Fig. 4).

Fig. 4

Spatial pattern of MCA-LIA sea-level pressure difference in model simulations. (A) NCAR CSM 1.4 and (B) GISS-ER. For the NCAR model, a single run was available (23). For the GISS-ER coupled model, we show the ensemble mean of six realizations; see SOM section 5 for further details.

The observed patterns of change, even when averaged over multicentury intervals, are unlikely to be entirely forced in nature, as there is also a potentially important role for purely internal, natural variability (9). Consistent with this view, we find that individual realizations of the GISS-ER transient response to the MCA-LIA solar forcing difference yield patterns that differ modestly in their details. For at least one realization, for example, the reconstructed warm anomaly over Western Europe is reproduced. In most cases, the basic features discussed above are nonetheless evident (SOM Text).

The paleoclimate reconstructions presented here hold important implications for future climate change. For example, if the tropical Pacific thermostat response suggested by our analyses of past changes applies to anthropogenic climate change, this holds profound implications for regional climate change effects such as future drought patterns. Continued refinement of paleoclimate reconstructions through expanded proxy databases and refinements of CFR methodology, improved estimates of past radiative forcing, and a better understanding of the influence of radiative forcing on large-scale climate dynamics should remain priorities as we work toward improving the regional credibility of climate model projections.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S11

Tables S1 to S5


SOM Data

References and Notes

  1. See Dataset S1 in the Materials and Methods.
  2. The synthetic proxy data in these tests are constructed to have noise characteristics similar to those estimated for actual proxy data. The calibration process, as with real-world reconstructions, is performed over a modern interval that is subject to anthropogenic forcing. The ability of the method to reproduce the earlier variations is then objectively assessed. See (12) for further details.
  3. M.E.M. and Z.Z. gratefully acknowledge support from the ATM program of the National Science Foundation (grant ATM-0542356). R.S.B. acknowledges support from the Office of Science (BER), U.S. Department of Energy (grant DE-FG02-98ER62604). M.K.H. and F.B.N. were supported by the National Oceanic and Atmospheric Administration (grant NA16GP2914 from CCDD). D.T.S. and G.F. acknowledge support from NASA’s Atmospheric Chemistry, Modeling, and Analysis Program.

Stay Connected to Science

Navigate This Article