Physics-informed detection of tropical cyclonesbased on dynamical systems metrics

Even though tropical cyclones are well documented from the moment they materialise to the moment they evanesce, many physical processes governing them are not resolved by global circulation models (GCMs). Thus, the simulation and the projection of future changes in tropical cyclones characteristics remain challenging. In particular, due to the low spatial resolution of most GCMs, studying simulated tropical cyclones is difficult because they are hard to detect. Some cyclone detection schemes are now available but their results are subject to the spatial resolution of the simulations and  usually need numerous variables that are not always available in the models. In this study, the ability of dynamical system metrics to detect atmospheric situations related to cyclones in coarse-resolution data-sets, where maximum wind speed and minimum sea-level pressure may not be accurately represented, is explored.

Our analysis is based on 613 cyclones occurring in Tropical Atlantic over the 1980-2021 period and extracted from the IBTrACS database. Three dynamical system metrics, the local dimension d, local persistence θ and the co-recurrence ratio α, are computed from mean sea level pressure and the total column water of the ERA5 reanalysis dataset. They respectively characterise the number of degrees of freedom of the atmospheric flow, its stability and the  coupling between different atmospheric variables. In the preliminary results, atmospheric situations corresponding to cyclones stand out from other situations. This preliminary result shows potential for detection and attribution studies in climate change context.