Aim and Location: Our aim is to develop pollen-climate inference models for southern Europe and to test their performance and inference power by cross-validation with modern climate data. Surface sediments collected from lakes along a climate gradient from the winter-cold/summer-wet Alps to winter-wet/summer-dry Sicily were analysed for modern pollen assemblages.
Methods: For each lake, mean monthly temperatures, seasonal precipitation and site-specific climate uncertainties have been estimated. Pollen-climate relationships were studied using numerical analyses, and inference models were derived by partial least squares (PLS) and weighted-averaging PLS (WA-PLS) regressions for January and July temperatures (T ), and for winter, spring and summer precipitation (P). In order to assess whether these variables are also of ecological importance for vegetation in the subregions, we split the data set into an Alpine and a Mediterranean subset.
Results: Low bootstrap cross-validated root mean square errors of prediction (RMSEP) for January T (1.7 °C), July T (2.1 °C) and summer P (38 mm), as well as low RMSEPs expressed as a percentage of the gradient length (8 -9%), indicate a good inference power. Models revealed excellent to good performance statistics for January T, July T and summer P (r2 = 0.8), and for winter and spring P (r 2 = c. 0.5). We show that the variables with the highest explanatory power differ between the two subregions. These are summer T and P for the Alpine set, and January T, winter P and July T for the Mediterranean set.
Main conclusions: The study reveals the influence of climatic conditions during the growing season on modern pollen assemblages and indicates the potential of pollen data for long-term climate reconstructions of parameters such as winter precipitation and temperature, which seem to be the main factors having an influence on the variability of Mediterranean climate. These models may therefore provide important information on past regional climate variability in southern Europe.
