Preprints
https://doi.org/10.5194/egusphere-2024-3253
https://doi.org/10.5194/egusphere-2024-3253
04 Nov 2024
 | 04 Nov 2024
Status: this preprint is open for discussion.

Reask UTC: a machine learning modeling framework to generate climate connected tropical cyclone event sets globally

Thomas Loridan and Nicolas Bruneau

Abstract. In the early 1990s, the insurance industry pioneered the use of risk models to extrapolate Tropical Cyclone (TC) occurrence and severity metrics beyond historical records. These probabilistic models rely on past data and statistical modelling techniques to approximate landfall risk distributions. By design such models are best fit to portray risk under conditions consistent with our historical experience. This poses a problem when trying to infer risk under a rapidly changing climate, or in regions where we do not have a good record of historical experience. We here propose a solution to these challenges by rethinking the way TC risk models are built, putting more emphasis on the role played by climate physics in conditioning the risk distributions. The Unified Tropical Cyclone (UTC) modelling framework explicitly connects global climate data to TC activity and event behaviours, leveraging both planetary scale signals and regional environment conditions to simulate synthetic TC events globally. In this study we describe the UTC framework and highlight the role played by climate drivers in conditioning TC risk distributions. We then show that, when driven by climate data representative of historical conditions, the UTC is able to simulate a global view of risk consistent with historical experience. Additionally, the value of the UTC in quantifying the role of climate variability on TC risk is illustrated using the 1980–2022 period as a benchmark. 

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.
Thomas Loridan and Nicolas Bruneau

Status: open (until 16 Dec 2024)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
Thomas Loridan and Nicolas Bruneau
Thomas Loridan and Nicolas Bruneau

Viewed

Total article views: 72 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
52 17 3 72 0 0
  • HTML: 52
  • PDF: 17
  • XML: 3
  • Total: 72
  • BibTeX: 0
  • EndNote: 0
Views and downloads (calculated since 04 Nov 2024)
Cumulative views and downloads (calculated since 04 Nov 2024)

Viewed (geographical distribution)

Total article views: 72 (including HTML, PDF, and XML) Thereof 72 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 06 Nov 2024
Download
Short summary
Tropical Cyclone (TC) risk models have been used by the insurance industry to quantify occurrence and severity risk since the 90s. To date these models are mostly built from backward looking statistics and portray risk under a static view of the climate. We here introduce a novel approach, based on machine learning, that allows sampling of climate variability when assessing TC risk globally. This is of particular importance when computing forward looking views of TC risk.