Data-driven discovery of Spatiotemporal Causal Precursors to Tropical Weather Extremes

This proposed thesis aims to understand how data-driven statistical modelling and machine learning (ML)

can yield new physical insights into what causes tropical weather extremes. All proposed projects in this

thesis will focus on tropical cyclones (TCs). In the three projects proposed in this thesis, we will explore

utilizing different data-driven methodologies to address two questions regarding TCs that improve the slowest

in operational forecasts: the genesis of TCs and extreme precipitation associated with TCs of different

intensities.

The overarching objective of this thesis is to better understand the causal pathways that (I) cause a few

tropical cloud clusters to undergo a sudden regime change from regular mesoscale convective systems (MCSs)

to intensifying tropical cyclones, and (II) cause a TC to produce more heavy and sustained precipitation.

What combination of processes or chain of events results in the greatest likelihood of these destructive

phenomena? That is what we hope to clarify in this thesis.