Merging chironomid training sets: implications for palaeoclimate reconstructions

Abstract

Icelandic and Norwegian chironomid calibration or training sets were merged to investigate whether a larger combined training set would be useful to apply to subfossil chironomid data from Iceland for periods such as the early Holocene, the Holocene Thermal Maximum and the Little Ice Age, when temperatures can be expected to be outside the current temperature range of the Icelandic training set. Following taxonomic harmonisation, the Icelandic and Norwegian data sets were compared before being merged to form a combined Norwegian-Icelandic training set. Analyses showed that it was biologically and statistically valid to merge the two data sets. The resulting combined inference model for mean July air temperature had improved performance statistics (r 2 jack = 0.87; RMSEP jack = 1.13) when compared to the best performing Icelandic model (r 2 jack = 0.61; RMSEP jack = 0.83), due to the longer environmental gradient covered (Icelandic 6–11 °C; combined 3.5–16 °C), and to the increased number of samples (Icelandic = 53 lakes; combined = 207 lakes) and taxa (Icelandic = 47 taxa; combined = 133 taxa) present within the combined training set. The inference models were applied to an early Holocene chironomid sequence from Vatnamýri, north Iceland, and a 450-year recent record from Myfluguvatn, north-west Iceland, to compare the reconstructions produced. The various inference models produced similar trends and patterns of temperature reconstruction, but the inference model based on the combined training set produced a larger range of reconstructed temperatures than the Icelandic model. It was found that different inference models produced more variation in the reconstruction than when different training sets were used. A comparison of the Myfluguvatn reconstructions with meteorological observations showed that the combined Norwegian–Icelandic inference model produced more reliable results than the Icelandic or Norwegian inference models alone.