Quantitative paleoclimate reconstructions based on multiple proxies in southeastern Europe, Turkey, and the Levant

Several techniques are commonly used for quantitative climate reconstructions based on plant information from lake sediments. One of these is the creation of transfer functions that link current climate to current plant distributions and apply the results to pollen information from sediment cores. However, one challenge is to combine this information with that of other proxies. Therefore, we have developed a new algorithm that automatically creates a compromise from all available proxy information. This technique has already been successfully tested on over 500 sediment cores throughout Europe and the Middle East. Moreover, it allows us to reconstruct not only the climate of the last centuries, but also that of several glacial-interglacial cycles. In this context, the following proxies were used: Plant information from lake sediment cores, isotopic information from speleothems, marine cores, and ice cores.

To achieve this, we need to solve four major tasks. First, we conducted a machine learning competition to find the best generalized transfer functions linking current climate to vegetation data. Second, we developed a Bayesian-based age-depth/distance transformation that allows us to construct a regular temporal grid not only for sediment cores but also for speleothems. This computationally fast and easy-to-implement transformation allows us to include all age uncertainties in our reconstruction method. Third, if the proxies are to be considered in spectral space rather than temporal space, this can be achieved by our method using wavelet power spectra. Fourth, we have designed a fast algorithm that combines all inserted proxy information. This was made possible through the use of a Markov chain Monte Carlo method that derives specific weights for each taxon based on the included proxy information. In addition, this flexible technique can be enhanced by incorporating direct climate information, e.g., from instrumental meteorological measurements, from results of paleoclimate simulations, and/or from historical records. If the human influence on a proxy in a given period is not negligible, the algorithm ignores it and focuses on the remaining proxies to minimize this effect. Therefore, the procedure can be easily extended with further proxy information such as CO2 and/or solar insolation.

In summary, our new method provides quantitative paleoclimate reconstructions that approximate proxies not only in temporal space but also in spectral space, and can be further constrained by climate anchor points.