The study of past climate is crucial for us to understand how climate changes naturally through time and thus define the background context against which anthropogenic changes take place. Variations in regional temperature have widespread implications for society, but our understanding of the amplitude and origin of long-term natural variability is insufficient for accurate regional projections. This is especially the case for terrestrial temperature variability, which is currently thought to be weak over long timescales. 

The traditional view of climate variability across timescales is that the variability consists of noise that quickly converges to a stable mean when averaged over bigger time windows. Concretely, this means that the typical 1℃ fluctuations observed in local temperature at the decadal timescale would decrease to 0.4℃ at the centennial timescale, and further to 0.2℃ at the millennial timescale, i.e. local temperature would be extremely stable outside of the response to external forcing. In addition to this variability, one expects variability as a direct response to forcing, mainly due to cycles in solar insolation: the daily and annual cycles on shorter timescales, and the deterministic Milankovitch cycles resulting from variations in the Earth’s orbital parameters over multi-millennial timescales.

Most paleoclimate investigations have focused on reconstructing the slow changes in the mean state, but few have attempted to characterize the natural variability. This is particularly important as the magnitude of the natural variability at the regional scale is of similar magnitude as that of the forced anthropogenic warming on multi-decadal scales, and therefore contributes significant uncertainty to expected climate change at the local scale felt by societies. Understanding the statistical properties of natural variability in the late Holocene (the last 8000 years in this study) is particularly important precisely because its relative stability extends up until today and can help us better interpret current climate changes and better understand the ramifications of human impacts on future climate.

There are several climate proxies, i.e. indirect recorders of past climate variations, which can be used to study long-term temperature variability over land such as for example speleothems, ice cores, tree ring records and other archives extracted from lake and marine sediments. Plants leave behind in sediments traces of their presence in the form of pollen grains, which once compiled provide us snapshots of past ecological composition, thus tracking vegetation changes along with climate changes. Over the past decades, thousands such records have been compiled over the world, particularly in the northern hemisphere extratropics, providing the largest spatio-temporal coverage of past terrestrial vegetation and climate changes. In this study, we took advantage of the most extensive compilation of pollen records compiled in the ERC project GlacialLegacy of Ulrike Herzschuh. This allowed us to reconstruct past summer temperatures over the last 8000 years in North America and Eurasia and, combined with instrumental data, we quantified the average behaviour of local temperature variability from annual to multi-millennial timescales.

We found that while in climate models the average tends towards a constant value with smaller and smaller variations for longer and longer periods, the pollen-based temperature reconstructions we studied support that temperature fluctuations longer than multi-decadal are about 1℃ independent of the span of the averaging period, such that a typical century scale fluctuations is of similar magnitude than a millennial scale one. This hints at the existence of mechanisms driving long-term variability that are missing or not well represented in the climate models. 

Our study did not only characterize the mean behaviour, but also provided spatially resolved estimates of millennial scale variability. This revealed meaningful and spatially coherent patterns intrinsically linked to the propagation of oceanic influence inland. The relationship we identified supports a dual role of oceans in influencing temperature variability. We know that oceanic climates are generally more stable, at least on annual to decadal timescales, and this was quantified with recent instrumental data. What was surprising however, is that the same regions became the most variable on millennial timescales. The converse was also true such that very continental regions such as the interior of North America were more variable than other regions on sub-decadal timescales, but less on millennial timescales. This relationship between the spatial patterns of sub-decadal and millennial variability is particularly significant as it comes from two completely independent sources of data, namely instrumental data and pollen records, and thus provides us with additional confidence in the robustness of the result.

Therefore, while oceanic influence stabilizes short term climate, it appears to be the main driver of long-term variability. The oceans of the Earth have a strong role in buffering warming and releasing it later due to their large thermal inertia and therefore it may not be so surprising that they drive long-term variability, but this is not observed in global climate models. We thus think that model development should focus on the oceanic component in order to foster the emergence of long-term natural variability in agreement with proxy data. This also means that we need to develop data products based on multi-proxy data which can be used as targets for evaluating the  variability in climate models across timescales.