hide
Free keywords:
transfer functions; weighted-averaging partial-least-squares; summer temperatures; surface sediments; modern training-sets; Switzerland
Abstract:
Diatom, chrysophyte cyst, benthic cladocera, planktonic cladocera, and chironomid assemblages were studied in the surface sediments of 68 small lakes along an altitudinal gradient from 300 to 2350 m in Switzerland. In addition, 43 environmental variables relating to the physical limnology, geography, catchment characteristics, climate, and water chemistry were recorded or measured for each lake. The explanatory power of each of these predictor variables for the different biological data-sets was estimated by a series of canonical correspondence analyses (CCA) and the statistical significance of each model was assessed by Monte Carlo permutation tests. A minimal set of environmental variables was found for each biological data-set by a forward-selection procedure within CCA. The unique, independent explanatory power of each set of environmental variables was estimated by a series of CCAs and partial CCAs. Inference models or transfer functions for mean summer (June, July, August) air temperature were developed for each biological data-set using weighted-averaging partial least squares or partial least squares. The final transfer functions, after data screening, have root mean squared errors of prediction, as assessed by leave-one-out cross-validation, of 1.37 degrees C (chironomids), 1.60 degrees C (benthic cladocera), 1.62 degrees C (diatoms), 1.77 degrees C (planktonic cladocera), and 2.23 degrees C (chrysophyte cysts)
