CR 77:63-76 (2019)  -  DOI: https://doi.org/10.3354/cr01543

ABSTRACT: Understanding local patterns and large-scale processes in past climate necessitates a detailed network of temperature reconstructions. In this study, a merged temperature inference model using fossil chironomid (Diptera: Chironomidae) datasets from Finland and Poland was constructed to fill the lack of an applicable training set for East European sites. The developed weighted averaging partial least squares (WA-PLS) inference model showed favorable performance statistics, suggesting that the model can be useful for downcore reconstructions. The combined calibration model includes 212 sites, 142 taxa, and a temperature gradient of 11.3-20.1°C. The 2-component WA-PLS model has a cross-validated coefficient of determination of 0.88 and a root mean squared prediction error of 0.88°C. We tested the new East European temperature transfer function in chironomid stratigraphies from a Finnish high-resolution short-core sediment record and a Polish paleolake (Zábieniec) covering the past ~20000 yr. In the Finnish site, the chironomid-inferred temperatures correlated closely with the observed instrumental temperatures, showing improved accuracy compared to estimates by the original Finnish calibration model. In addition, the long-core reconstruction from the Polish site showed logical results in its general trends compared to existing knowledge on the past regional climate trends; however, it had distinct differences when compared with hemispheric climate oscillations. Hence, based on these findings, the new temperature model will enable more detailed examination of long-term temperature variability in Eastern Europe, and consequently, reliable identification of local and regional climate variability of the past.

KEY WORDS: Chironomidae · Climate reconstruction · Finland · Holocene · Late Glacial · Paleoclimate · Poland · Training set · Transfer function