The first quantitative palaeoclimate reconstructions from microfossil assemblages were low resolution, with a sample spacing of hundreds or thousands of years and focused on seasonal climate. Reconstructions focusing on a single month are now published at annual resolution. Can resolution be further improved? Few would argue that seasonal reconstructions are impossible (notwithstanding a sensible choice of season); only the foolish would believe a reconstruction focused on 4th July temperatures (even though it might have impressive transfer function performance statistics).
LT15 reconstruct August air temperature from chironomid (non-biting midges) assemblages from Lake Żabińskie in eastern Poland using a transfer function calibrated on chironomid assemblages in lakes in Canada and Poland. As the lake sediments are varved, LT15 are able to make a reconstruction on a near-annual basis for the last century.
The correlation between the reconstruction and instrumental data from local weather stations is remarkably good at 0.74. The mean absolute difference between the instrumental and the inferred August air temperatures is only 0.75 °C, much smaller than the root mean squared error of prediction of the transfer function (2.3 °C; the original paper erroneously reported the RMSEP as 1.3 °C, which the corrigendum persists in using). This correlation is not the result of a shared trend in both time series – the trend in the instrumental data is slight (r2 = 0.05) and the residuals from this trend have low autocorrelation (ar1 = 0.14).
To put this correlation into context, the 99th percentile of positive correlations between tree-ring widths and summer temperature is 0.63 (St. George 2014). Even though tree-rings are a much simpler proxy than chironomid assemblages, few ring-width series have a better correlation with climate than the Lake Żabińskie chironomid reconstruction.
What are the physical, ecological and taphonomic limits to reconstruction resolution?
Transfer functions are (usually) an example of space-time substitution: spatial patterns in biotic assemblages and climate are used to estimate climatic changes in time from fossil assemblages. The problem is that the space-time substitution is imperfect: the strength of the correlation between different climatic variables in space and in time differ. For example, the correlation in space between June and August mean temperature across northern Europe is 0.99, whereas the year-on-year correlation in time between June and August temperatures at Vilnius (close to Lake Żabińskie) is only 0.16.
This is not in the least surprising. Places where the climatological mean temperature is warm in June inevitably have warm August mean temperatures. Conversely, if June is warm in Vilnius this year there is no reason to expect August to be warm as well. In short, this is the difference between climate (long-term mean) and weather.
This difference between climate and weather has serious implications for transfer-function development and application. Transfer functions trained on August, June or the whole summer will have very similar performances regardless of which month or months are ecologically important for the biota as they depend on the climatological means. In contrast, if August temperature is reconstructed, but June temperature is the ecologically important month, the correlation between the reconstruction and instrumental records will be woeful. This is a particular problem for high-resolution reconstructions – at decadal and longer scales the effect of weather will be averaged out. An identical problem occurs with the correlation between environmental variables such as dissolved organic carbon and climate; it might be strong in space but cannot be expected to correlate as well on a year-on-year basis.
I don’t know which months chironomids are most sensitive to, but, if they are temperature sensitive it is reasonable to assume that they respond to the whole summer rather than only August. The correlation at Vilnius between August temperature and the whole summer (JJA) temperature is 0.76. If this whole summer assumption is correct, the correlation reported by LT15 between the reconstruction and the instrumental data is on the edge of what could physically be expected if chironomids were perfect thermometers.
The ideal proxy for highly focused reconstructions would be directly sensitive to the environmental variable we want to reconstruct; only be sensitive to this variable during the time of year we want to make the reconstruction; not be sensitive to variability in other environmental variables; have a short generation time so they can quickly respond to the environmental conditions; and be sufficiently abundant that large counts can be made.
Some algal proxies might meet most of these criteria. Chironomids don’t. The larval phase of chironomids is indirectly sensitive to air temperature via water temperature and oxygen availability; they are probably sensitive (and producing head-capsules) to temperature throughout the summer; they are sensitive to anoxia; the larval phase can last over a year; and often only small counts are possible (one of the issues in the corrigendum).
Taphonomic processes will tend to degrade the signal in biotic assemblages. In anoxic lakes, like Żabińskie, bioturbation will be minimised, but sediment reworking before its final deposition in the centre of the lake could blur the record. Reworking might be a minor problem for planktonic proxies where the majority of the organisms are deposited without reworking, but large in littoral proxies that have to be reworked to reach the core site.
Lake Żabińskie is 44.4 m deep yet the chironomid assemblages are dominated by littoral taxa, probably because few chironomids can survive in the anoxic hypolimnion in the deep central part of the lake where the core was collected. The littoral chironomids must have been transported into the deep water. Water currents when the lake’s thermal stratification breaks down in autumn can transport the chironomids. The taphonomic problem is that to retain the annual signal, the currents need to just transport the chironomids that lived that year. If the currents transport chironomids that lived in different years, the assemblages will include chironomids from different years, smoothing the stratigraphy and weakening the correlation with instrumental records. This is only a problem for high resolution records. It is possible to imagine lakes where autumn currents would only transport the chironomids that lived in the preceding summer, for example one where all the sediment deposited in shallow water was scoured every year, but that does not seem possible for anything other than a lake in a rock basin.
High-resolution reconstructions depend on the chronology. The varve-based chronology for Żabińskie is excellent with an uncertainty of a couple of percent. However, even an error of one year could ruin the correlation between the reconstruction and the instrumental record. An error near the top of the core would be worst, affecting more of the record. An error in the lower resolution section at the bottom of the record would have much less impact.
Excellent sites are needed for very high-resolution palaeoecological studies. Lake Żabińskie with its excellent chronology seems to be an ideal site.
Identifying biotic proxies that are sensitive to a single month’s weather will prove difficult. Planktonic algal proxies are probably best as they might reduce taphonomic problems and have short generations times.
Chironomids would not appear to be ideal for very high-resolution analyses of monthly climate variables. Despite this, the reconstruction of August air temperatures from Żabińskie is remarkably good, at the limit of what is physically possible and with an error substantially lower than expected given the uncertainty of the transfer function.
Fortunately, the authors have archived some data. In a subsequent post, I will take a look at these data to test if the results can be reproduced and if the remarkably good performance can be explained.