Master Thesis defense by Master Thesis defense by Tinus Claus Blæsbjerg Lodahl

Title: Hurricane damage in a changing climate: A Bayesian Analysis of Hurricane Damage.

Abstract:

Tropical cyclones are the costliest and deadliest natural disasters in US history, being responsible for $1.5 trillion in economic costs and 7211 deaths from 1980 to 2024. This thesis tries to investigate what physical and socioeconomic drivers best explain hurricane damage, whether a climate change signal is detectable in the data, and how much damage could have been avoided without climate change. Variable selection using Projective Predictive Variable Selection and BART variable importance reveals exposed wealth and central pressure drop in the hurricane as the strongest predictors of damage and maximum observed storm surge and trend as less important predictors.

Using Bayesian inference and Markov Chain Monte Carlo sampling, parametric regression models and Bayesian Additive Regression Tree (BART) models are fitted using damage estimates from 219 landfalling US hurricanes in the period from 1900 to 2017. Expected Log Pointwise Predictive Density (ELPD) model comparison ranks a log-student’s t BART model as the best model, and the partial dependence plots for the model indicate positive relationships between damage and all four predictors and, furthermore, show a possible saturation relationship between damage and both pressure and surge at high intensity levels.

A climate change signal could not be quantified from the data, however the trend predictor is included in the best models. Finally, climate change damage attribution was estimated by adjusting surge and pressure measurements to reflect a no-climate-change scenario, which gave median climate change damage attribution estimates of 50-58%, corresponding to $314B-$362B of the total observed damage of $620B in the period from 1900 to 2017. However, these estimates rely on several assumptions and should be seen as speculative more than exact numbers.

Supervisor: Aslak Grinsted
Censor: Martin Drews (DTU)