Partitioning uncertainty components in climate projections using smoothing splines

A critical issue in climate change impact studies is the assessment of uncertainties associated with future projections. Various methods have been proposed for partitioning uncertainty sources, usually based on an Analysis of Variance (ANOVA). In this paper, we show how Smoothing-Spline ANOVA approaches (SS-ANOVA) can be used to estimate the total uncertainty and its partition in climate projection ensembles. A Bayesian framework is proposed to handle heteroscedastic and autocorrelated residual errors between the climate change responses and the main additive effects modelled with cubic smoothing splines