This repository consists consists of R code for the paper titled "Predictions of damages from Atlantic tropical cyclones: a hierarchical Bayesian study" over for data spanning 1960-2019. The code for Chapter 1 is in R Code Part 1 and for Chapter 2 is in R Code Part 2.
- R Code Part 1 consists of R code for modeling the relationship between maximum wind speed and minimum central pressure across basins using weighted linear models and generalized extreme value models using data from 1960-2020. Damages caused by the hurricanes are integrated in the bivariate GEV model towards the end. This provides a baseline understanding of the underlying phenomenon which motivates using a Bayesian model as in Part 2.
- R Code Part 2 consists of R code for hierarchical Bayesian modeling of frequency, landfall and damage capabilities of North Atlantic hurricanes from 1960-2019. It begins with a preliminary analysis of modeling frequency, landfall and damages with 6 covariates, namely, SOI, NOA, AMO, Nino 3.4 Anomaly, Atlantic SST and Sunspots, using GLMs. Then a fully Bayesian hierarchical model is fit and a predictive analysis is conducted for years 2016-2017 to validate the model fit. For comparison sake, Empirical Bayes HM using hyperparameter choices from the preliminary analysis is also fit with all and selected covariates to compare the results with the fully Bayesian HM with all and selected covariates, respectively.
- R Code Part 3 consists of R code for modeling the trivariate relationship between maximum wind speed, minimum central pressure and damages for yearly data on the Atlantic tropical cyclones from 1960-2019. We use three different Bayesian models, namely 1) trivariate non-stationary extreme value logistic model, 2) hierarchical extreme value distribution model, and 3) hierarchical extreme value distribution model with EVD models for minCP and maxWS and with log-normal model for damages. The folder Posterior Prediction includes code for posterior prediction for years 2016-2017 using these Bayesian models.