Towards the end of the last glaciation, as solar insolation at high latitudes slowly increased as the Earth’s orbit changed, the climate over Greenland suddenly tipped from a glacial climate to an interglacial climate — a warming of over 10°C — in a decade or so. The following warm period, known as the Bølling in northern Europe was short lived, no more than a flicker of warmth, before the climate cooled first into the tepid Allerød and then the frigid Younger Dryas. The next time the climate abruptly warmed, the following warm period, the Holocene, was more enduring.
The rapid warming into the Bølling and Holocene are critical transitions between alternative stable climate states. Critical transitions can be either noise induced or the result of forcing pushing the system until one climate state is no longer stable and the system tips to the other at a bifurcation point. The latter type of critical transition is at least in principle possible to predict due to early warning signals as the system approaches the bifurcation. The three classic early warning signals are increases in autocorrelation, variance and kurtosis as the basin of attraction becomes shallower (See here for an explanation of these indicators).
Several papers have looked for early warning signals before the transitions into the Bølling and Holocene. The usual procedure is to detrend the record and then use a running variance and running autoregressive model over the data. The choice of the bandwidth for the detrending and the window size for the running analysis can be critical.
Dakos et al (2008) find autocorrelation in the GISP2 isotope-temperature record increases weakly and non-significantly before the Bølling transition, and increases strongly and significantly before the Holocene transition in the Cariaco basin grey-scale record. However, variance increases in GISP2 and decreases before the transition in Cariaco (Lenton et al 2012).
Ditlevsen and Johnsen (2010) in the excellently titled paper “Tipping points: Early warning and wishful thinking” argue that both autocorrelation and variance need to increase before a critical transition to be considered an early warning signal. In an analysis of the NGRIP isotope record, they do not find any evidence of early warning signals before the abrupt warming of any of the Dansgaard-Oeschger cycles (including the Bølling transition). This implies that these transitions are noise-induced rather than bifurcations.
Praetorius & Mix (2014) look for evidence of synchronisation between the North Atlantic and Pacific prior to the Bølling transition as an alternative early warning signal, but I think their results are an artefact of their methods.
The evidence for early warning signals before the Bølling transition is underwhelming. Despite this, I am going to argue that the results of a recent paper in Nature indicate that there could have been early warning signals before the Bølling transition but that existing records are not in the right place to detect them.
Thiagarajan et al (2014) use my favourite proxy, clumped isotopes, to reconstruct temperature from cold-water corals across a depth gradient in the North Atlantic. Clumped isotopes rely on the temperature-dependent tendency for heavy isotopes of 13C and 18O to co-occur in carbonate molecules more often than expected by chance. It is a fairly new proxy, and the uncertainties on individual measurements are large, but corals are large enough to support multiple replicates, driving down the uncertainty on the mean.
They date the coral using uranium/thorium dating and use the difference between the expected and observed radiocarbon date – the reservoir age – as a metric of the time elapsed since the water mass was last in contact with the atmosphere.
They find that in the period before the Bølling transition, the thermal structure of the ocean suddenly changes with warm, 14C-depleted water at depth overlain by cooler water. This inverts the normal temperature stratification and is presumably maintained by a salinity gradient (unfortunately the authors do not try to use 18O and clumped isotopes together to estimate salinity – perhaps the uncertainties are too large). The sudden change in temperature indicates that the warm water has been advected rather than being warmed in situ.
This thermal stratification would be inherently unstable even if supported by a salinity gradient because of a property of water known as thermobaricity. Water is compressed, becoming denser, when subjected to high pressure at depth in the ocean. Warm water is compressed more by high pressures than cold water. So if cold fresh (in an oceanographic sense) water overlies slightly denser warm salty water, any disturbance that brings the warmer water nearer to the surface will reduce the pressure and reduce the density, reversing the density gradient. The large reservoir of warm saline water will then punch its way to the surface, releasing the heat to the atmosphere, as recorded by the Greenland ice cores. As the warm water cools, it will become dense due to its high salinity and sink, drawing subtropical water north and kickstarting the Atlantic meridional overturning circulation.
Before the warm water punched its way to the surface, there may well have been a host of early warning signals at the boundary between the cold fresh and warm saline waters. As the density difference declined, disturbances that were not quite large enough to cause overturn would cause increasing variance and autocorrelation. It is doubtful that any palaeoceanographic proxy will have the temporal resolution and sensitivity to record these possible early warning signals. Proxies records from Greenland and elsewhere, away from the thermocline, will not show early warning signals.
The inverted thermal gradient did not reoccur at the end of the Younger Dryas: some other process was responsible. So even if the last of the mammoths had identified the early warning signals before the Bølling transition and sought to use this information, they would have been taken unaware by the Holocene transition. So much for the utility of early warning signals.
Oceanographically-aware mammoths should have been aware that the thermal structure of the pre-Bølling Atlantic was unstable and, regardless of early warning signals, taken remedial action. My feeling is that today, we should take a physics based approach to identifying tipping points rather than relying on statistical tools.
Thiagarajan, N. et al 2014. Abrupt pre-Bølling–Allerød warming and circulation changes in the deep ocean. Nature, 511, 75–78 doi:10.1038/nature13472