On the Robustness of Bayesian Fingerprinting Estimates of Global Sea Level Change

Global mean sea level (GMSL) over the twentieth century has been estimated using techniques that include regional averaging of sparse tide gauge observations, combining satellite altimetry observations with tide gauge records in empirical orthogonal function (EOF) analyses, and most recently the Bayesian approaches of Kalman smoothing (KS) and Gaussian process regression (GPR). Estimated trends in GMSL over 1901–90 obtained using the Bayesian techniques are 1.1–1.2 mm yr −1 , approximately 20% lower than previous estimates. It has been suggested that the adoption of a less restrictive subset of records biased the Bayesian-derived estimates. In this study, different subsets of records are used to demonstrate that GMSL estimates based on the Bayesian methodologies are robust to tide gauge selection. A method for determining the resolvability of individual sea level components estimated in a Bayesian framework is also presented and applied. It is found that the incomplete tide gauge observations result in posterior correlations between individual sea level contributions, making robust separation of the individual components impossible. However, various weighted sums of these components, as well as the total sum (i.e., GMSL), are resolvable. Finally, the KS and GPR methodologies allow for the simultaneous estimation of sea level at sites with and without observations. The first KS and GPR global maps of sea level change over the twentieth century are presented. These maps provide new estimates of twentieth-century sea level in data-sparse regions.