Improving the statistical method can raise the upper tail of sea-level projections

Open Access
Ruckert, Kelsey Leigh
Graduate Program:
Master of Science
Document Type:
Master Thesis
Date of Defense:
November 17, 2015
Committee Members:
  • Klaus Keller, Thesis Advisor
  • Chris Eliot Forest, Thesis Advisor
  • James Kasting, Thesis Advisor
  • Sea-level rise
  • Autocorrelated projections
  • Calibration methods
  • Overconfidence
Anthropogenic greenhouse gas emissions have caused increases in temperatures and sea-levels. Currently, many large urban cities reside within meters of present-day sea-level (e.g., Miami, New York City, New Orleans, Tokyo, Amsterdam, and Mumbai). To protect people and infrastructure against catastrophic events, many cities implement strategies to adapt to sea-level rise. These strategies are often designed for low annual flooding probabilities (i.e., one in 50 to one in 10,000). As a result, the design of these strategies hinges on the upper tails of sea-level rise projections. Over the past decades, studies have used various calibration methods including frequentist bootstrap and Bayesian inversion to provide probabilistic projections. However, these studies often do not estimate explicitly the upper tail and neglect the effect of known properties of observations. Sea-level observations often have autocorrelated (interdependent) residuals (model minus data) and typically have time-dependent (heteroskedastic) observation errors. In this thesis, we analyze the impact of neglecting such known observed properties on sea-level projections. Specifically, we compare the output of a semi-empirical sea-level model calibrated with different statistical estimation methods: (i) a frequentist bootstrap, (ii) a Bayesian inversion neglecting heteroskedastic residuals, and (iii) a Bayesian inversion considering heteroskedastic residuals. We show how implementing a more appropriate calibration method (in terms of accounting for known observed properties of the residuals) leads to a sizeable increase in the upper tail of sea-level rise projections. The results indicate that the choice of the calibration method can have considerable implications for the design of sea-level rise adaptation strategies.