Improving the Representation of the Statistical Properties of the Data-Model Discrepancies Can Increase the Upper Tail of Sea-Level Projections
Understanding the driving mechanisms and uncertainties of sea-level changes pose nontrivial geophysical and statistical challenges. A team of geoscientists and statisticians supported through a Stanford University-led, multi-institutional Cooperative Agreement analyzed how improving the method to confront a simple model of sea-level changes with observations affects the model hindcasts and projections. They found that improving the statistical method increases the uncertainty and the upper tail of the sea-level projections in the considered example.
Understanding the uncertainty surrounding sea-level projections can be important to inform the design of reliable infrastructures. The analysis also provides insights into the importance of choosing appropriate statistical tools on hindcasts, projections, and mechanistic insights.
The researchers quantified the effects of common methodological choices on parameter as well as projection uncertainties using the example of sea-level changes. They implemented three commonly used approaches to data-model fusion: a simple bootstrap method and two implementations of a Bayesian method that accounts for (or neglects) the time-varying nature of the observation errors. They applied these approaches to a published data set and model structure. These approaches were then assessed in terms of hindcasts and projections as well as the inferences on the model parameters. The results showed that choosing a more appropriate statistical method can have nontrivial impacts on the uncertainty surrounding the projections.