Minimizing the edge-effect in environmental reconstructions by trimming the calibration set: Chironomid-inferred temperatures from Spitsbergen

Abstract

A core from Lake Skardtjørna on western Spitsbergen was analysed for subfossil chironomids with the aim of inferring past temperatures. The core spans the last 1760 years and has a low concentration and low diversity of chironomids. Three taxa dominate the record: Oliveridia tricornis, Micropsectra radialistype, and M. insignilobus-type, the latter not previously recorded on Svalbard. Compared with the full Norwegian modern climate—chironomid calibration data set, Skardtjørna is at the coldest end of the modern temperature gradient. In an attempt to decrease potential bias caused by the numerical edge-effect and to improve the taxon response functions, the training set was trimmed by excluding lakes dissimilar to the study site. Two trimming approaches are tested. In the first, lakes at the warm end of the gradient are excluded. In the second, the calibration data-set lakes are partitioned by TWINSPAN on the basis of their modern chironomid assemblages, and lakes most dissimilar to Skardtjørna are excluded. According to the training-set performance diagnostics, a WA-PLS inference model excluding lakes >9°C has the most promising results (component 1, R ² = 0.85, maximum bias = 0.58°C, RMSEP= 0.73°C). The performance diagnostics based on the trimmed inference model out-performs the performance diagnostics from the full model, and the reconstructed temperatures indicate different trends. According to the <9°C inference model, the long-term temperature trend has been decreasing, with cold periods 1630—1450, 1150—1050, and around 250 cal. yr BP. Warm periods occurred between 1760 and 1650, 1420 and 1180, and 1000 and 830 cal. yr BP. Temperatures have been increasing during the last 250 years.