Solar insolation changes, resulting from long-term oscillations of orbital configurations1, are an important driver of Holocene climate2, 3. The forcing is substantial over the past 2,000 years, up to four times as large as the 1.6 W m−2 net anthropogenic forcing since 1750 (ref. 4), but the trend varies considerably over time, space and with season5. Using numerous high-latitude proxy records, slow orbital changes have recently been shown6 to gradually force boreal summer temperature cooling over the common era. Here, we present new evidence based on maximum latewood density data from northern Scandinavia, indicating that this cooling trend was stronger (−0.31 °C per 1,000 years, ±0.03 °C) than previously reported, and demonstrate that this signature is missing in published tree-ring proxy records. The long-term trend now revealed in maximum latewood density data is in line with coupled general circulation models7, 8 indicating albedo-driven feedback mechanisms and substantial summer cooling over the past two millennia in northern boreal and Arctic latitudes. These findings, together with the missing orbital signature in published dendrochronological records, suggest that large-scale near-surface air-temperature reconstructions9, 10, 11, 12, 13 relying on tree-ring data may underestimate pre-instrumental temperatures including warmth during Medieval and Roman times.
Over recent millennia, orbital forcing has continually reduced summer insolation in the Northern Hemisphere5. Peak insolation changes in Northern Hemisphere high latitudes, at ~65° N between June–August (JJA), have been identified as the prime forcing of climate variability over the past million years1. Together with long-term CO2 variability resulting from biogeochemical feedbacks of the marine and terrestrial ecosystems14, these insolation cycles have initiated the interplay between glacial and interglacial periods15.
State-of-the-art coupled general circulation model (CGCM) simulations and high-resolution climate reconstructions rarely extend beyond the past few hundred years, limiting possibilities to evaluate low-frequency temperature fluctuations beyond broad assessments (and debate) of the Medieval Warm Period and Little Ice Age4. In fact, most high-resolution temperature reconstructions16 including tree-ring width (TRW) records, the most widespread and important late-Holocene climate proxy17, have never even been compared with orbital forcing. However, limitations related to the necessary removal of biological noise and the questioned ability of TRW records to reliably track recent (and past) warm episodes18 may not make this proxy suitable to investigate the role of orbital forcing on climate. Indeed an evaluation of long-term temperature reconstructions, even over the past 7,000 years from across northern Eurasia, demonstrates that TRW-based records fail to show orbital signatures found in low-resolution proxy archives and climate model simulations (Supplementary Fig. S1). These discrepancies not only reveal that dendrochronological records are limited in preserving millennial scale variance, but also suggest that hemispheric reconstructions, integrating these data, might underestimate natural climate variability.
We here address these issues by developing a 2,000-year summer temperature reconstruction based on 587 high-precision maximum latewood density (MXD) series from northern Scandinavia (Fig. 1). The record was developed over three years using living and subfossil pine (Pinus sylvestris) trees from 14 lakes and 3 lakeshore sites >65° N (Methods), making it not only longer but also much better replicated than any existing MXD time series (for example, the widely cited Tornetraesk record contains 65 series19). We carried out a number of tests to the MXD network and noted the robustness of the long-term trends, but also the importance of including living trees from the lakeshore to form a seamless transition to the subfossil material preserved in the lakes (Methods). Calibration/verification with instrumental data is temporally robust and no evidence for divergence20 was noted. The final reconstruction (N-scan) was calibrated against regional JJA temperature (r1876–2006 = 0.77) and spans the 138 BC–AD 2006 period.
N-scan shows a succession of warm and cold episodes including peak warmth during Roman and Medieval times alternating with severe cool conditions centred in the fourth and fourteenth centuries (Fig. 2). AD 21–50 (+1.05 °C, with respect to the 1951–1980 mean) was the warmest reconstructed 30-year period, ~2 °C warmer than the coldest AD 1451–1480 period (−1.19 °C) and still ~0.5 °C warmer than maximum twentieth-century warmth recorded AD 1921–1950 (+0.52 °C). Twentieth-century Scandinavian warming is relatively small compared with most other Northern Hemisphere high-latitude regions4.
Superimposed on this interannual to multicentennial variability, N-scan reveals a long-term cooling trend of −0.31 °C per 1,000 years (±0.03 °C) over the 138 BC–AD 1900 period (the dashed red curve in Fig. 2) in line with evidence from low-resolution Holocene proxies2, 3. The cooling trend, representing a −0.34 °C temperature difference between the first and second millennium AD (−0.36 °C excluding the twentieth century from the second millennium mean), is however not preserved in the TRW data from the same temperature-sensitive trees (Fig. 3). Similarly, no evidence for a long-term cooling trend is observed in a previous Fennoscandian TRW-based temperature reconstruction spanning the past 2,000 years21. Such a trend was found in only low-resolution lake sediment and ice-core data of a circum-Arctic proxy network6. The high-latitude TRW data included in ref. 6 also suggested no sign of a long-term cooling trend, demonstrating that variance at this lowest frequency is absent in existing tree-ring time series and thus most of the large-scale reconstructions produced so far16. Removal of the tree-ring data from the Arctic-wide network6 results in an increased cooling trend substantially larger than the −0.21 °C per 1,000 years estimated using all proxy archives (Fig. 3c). Other reconstructions (besides the records shown here) that would be long enough and contain a clear and calibrated temperature signal to assess orbital signatures in tree-ring data are not available from the Northern Hemisphere at present.
As suggested previously2, 6, we propose that the millennial scale cooling trend retained in N-scan is forced by JJA insolation changes of ~ −6 W m−2 over the past 2,000 years5, as other potential forcings, including volcanic eruptions, land use and greenhouse gas changes, are either too small or free of long-term trends4. We tested this theory by analysing the two CGCMs that were run over several millennia (ECHO-G and ECHAM5–MPIOM; refs 7, 8), and compared the modelled temperature trends with and without orbital forcing (Methods). The CGCMs revealed similar JJA temperature patterns including a long-term cooling trend over the past two millennia centred in Northern Hemisphere high latitudes (Supplementary Fig. S12). The cooling trends are stronger in the ECHO-G model (−0.19 °C per 1,000 years for the 60°–70° N latitudinal band) compared with the ECHAM5–MPIOM model (−0.10 °C per 1,000 years), but diminish in both CGCMs towards lower latitudes. A smaller trend in orbital forcing and reduced albedo-driven feedbacks from high-latitude terrestrial snow and sea-ice cover contribute to this latitudinal gradient (Supplementary Fig. S13; ref. 22). Both models reveal stronger cooling over the continents in comparison with the oceans and indicate trends of −0.17 °C (ECHO-G) and −0.10 °C (ECHAM5–MPIOM) per 1,000 years in the grid boxes closest to northern Scandinavia. Whereas the absolute values from the simulations might not be as reliable as the empirically derived estimates reported here (> −0.3° per 1,000 years in northern Scandinavia and the Arctic zone), the long-term CGCM runs are valuable for the spatial assessment of orbital signatures.
The missing millennial scale trends in existing TRW records as well as the increased cooling trend after removal of this proxy type from the Arctic-wide estimates6 both suggest that the widely cited hemispheric reconstructions9, 10, 11, 12, 13, 14 underestimate pre-instrumental temperatures to some extent. This hypothesis seems to be important as most of the annually resolved, large-scale records are solely composed of or dominated (on longer timescales) by TRW data16, and their spatial domain encompasses the Northern Hemisphere extratropics including northern boreal and Arctic environments. Inclusion of tree-ring data that lack millennial scale cooling trends, as revealed here (Fig. 3 and Supplementary Fig. S1), thus probably causes an underestimation of historic temperatures. In line with the course and magnitude of the underlying orbital forcing (Supplementary Fig. S13), this underestimation ought to build up back in time, for example from the Medieval back to Roman times. Impacts on large-scale reconstructions from the omitted long-term trend in tree-ring data should, however, diminish towards lower Northern Hemisphere latitudes, as the forcing and radiative feedbacks5, 22 decrease towards equatorial regions.
Further calculation of these effects based on the missing cooling trends in TRW data and the spatial CGCM temperature patterns is, however, not yet possible, as the existing large-scale reconstructions include changing regional proxies, represent varying fractions of Northern Hemisphere (full Northern Hemisphere, Northern Hemisphere extratropics, Northern Hemisphere high latitudes), and are composed and calibrated using an array of methods16. The implications of missing orbital signatures in tree-ring records are generally related to the overall variance of temperature variations retained in the large-scale reconstructions (Supplementary Fig. S14). Whereas most of these time series show a similar course of long-term temperature change—including warmth during Medieval times, cold during the Little Ice Age and subsequent warming—the magnitude of reconstructed temperatures differs considerably among the hemispheric records16, 23. Some records show decadal scale variations of the order of ~ 0.4 °C over the past millennium12, whereas others indicate temperature variability up to 1.0 °C over the same period9. As a consequence, any adjustment for the omitted millennial scale temperature trends in dendrochronological records would have stronger implications, if the calibrated amplitude of past temperature variations was indeed small, say <0.5 °C, but would be less significant if the high-variance reconstructions turn out to be correct. The missing orbital signature in tree-ring records is also less significant to reconstructions containing variance increases (owing to reduced proxy coverage) back in time9, 13. These results reinforce the need to better constrain the pre-instrumental temperature variance structure10 and amplitude23.
Whereas our results on orbitally forced climate trends in a 2,000-year MXD chronology seem to be in line with coarse-resolution Holocene proxies2, 3 and are supported by CGCM evidence7, 8, little attention has been paid to the lack of these trends in long-term tree-ring records and implications thereof. The JJA temperature reconstruction presented here closes this gap, a finding that largely stems from the exceptionally strong and temporally stable climate signal, and the unprecedented length and replication of the new N-scan MXD chronology. The ability of MXD data to retain millennial scale temperature trends seems to result from a number of properties, including a reduced age trend24 and biological persistence25 resulting in less distortion of retained trends through regional curve standardization26 (RCS), the ability of tree populations to develop cell walls of continuously changing thickness over millennia and the non-plastic response of the termination of cell-wall lignification with respect to the integrated heat over the high and late summer seasons27. It is the combination of these properties that seems to enable the retention of a millennial scale trend in the MXD record and the lack of this lowest frequency variance in existing TRW records. These findings together with the trends revealed in long-term CGCM runs suggest that large-scale summer temperatures were some tenths of a degree Celsius warmer during Roman times than previously thought.
It has been demonstrated4 that prominent, but shorter term climatic episodes, including the Medieval Warm Period and subsequent Little Ice Age, were influenced by solar output and (grouped) volcanic activity changes, and that the extent of warmth during medieval times varies considerably in space. Regression-based calculations over only the past millennium (including the twentieth century) are thus problematic as they effectively provide estimates of these forcings that typically act on shorter timescales. Accurate estimation of orbitally forced temperature signals in high-resolution proxy records therefore requires time series that extend beyond the Medieval Warm Period and preferably reach the past 2,000 years or longer6.
Further uncertainty on estimating the effect of missing orbital signatures on hemispheric reconstructions is related to the spatial patterns of JJA orbital forcing and associated CGCM temperature trends. First, the simulated temperature trends, indicating substantial weakening of insolation signals towards the tropics, can at present be assessed in only two CGCMs (refs 7, 8). More long-term runs with GCMs to validate these hemispheric patterns are required. Whereas the large-scale patterns of temperature trends seem rather similar among the CGCMs, the magnitude of orbitally forced trends varies considerably among the simulations. Additional uncertainty stems from the weight of tree-ring data and varying seasonality of reconstructed temperatures in the large-scale compilations. Although some of the reconstructions are solely composed of tree-ring data, others include a multitude of proxies (including precipitation-sensitive time series) and may even include non-summer temperature signals. Some of these issues are difficult to tackle, as the weighting of individual proxies in several large-scale reconstructions is poorly quantified. The results presented here, however, indicate that a thorough assessment of the impact of potentially omitted orbital signatures is required as most large-scale temperature reconstructions include long-term tree-ring data from high-latitude environments. Further well-replicated MXD-based reconstructions are needed to better constrain the orbital forcing of millennial scale temperature trends and estimate the consequences to the ongoing evaluation of recent warming in a long-term context.