[Paper Review] Statistical modelling of tropical cyclone tracks: modelling cyclone lysis

Tim Hall, Stephen Jewson|ArXiv.org|Dec 10, 2005

Tropical and Extratropical Cyclones Research5 references3 citations

TL;DR

This paper improves a statistical tropical cyclone track model by introducing a spatially varying, non-parametric lysis model that estimates the probability of cyclone dissipation based on historical track data using cross-validated Gaussian weighting. The new lysis model significantly reduces systematic errors in track simulations, particularly the overrepresentation of tracks in the eastern subtropical Atlantic and improved realism in the high-density track region off the U.S. East Coast.

ABSTRACT

We describe results from the fifth stage of a project to build a statistical model of tropical cyclone tracks. The previous stages considered genesis and the shape of tracks. We now consider in more detail how to represent the lysis (death) of tropical cyclones. Improving the lysis model turns out to bring a significant improvement to the track model overall.

Motivation & Objective

• Improve the realism of basin-wide tropical cyclone track simulations by refining the representation of cyclone lysis (dissipation).
• Address systematic errors in prior models where simulated hurricanes persisted too long in regions where real hurricanes typically dissipate.
• Develop a non-parametric, spatially adaptive lysis model that reflects observed lysis patterns without assuming a parametric distribution.
• Enhance overall track model performance by integrating a more accurate lysis mechanism that better captures regional lysis frequency and spatial structure.

Proposed method

• Use a non-parametric, spatially adaptive lysis model based on historical HURDAT data (1950–2003) from the Atlantic basin.
• For each simulated hurricane at each time step, compute the lysis probability $ p $ as the ratio of Gaussian-weighted historical lysis points to all historical points within a spatial window of lengthscale $ L $.
• Apply a two-dimensional Gaussian kernel with lengthscale $ L $ centered at the current hurricane location to weight historical points.
• Optimize the lengthscale $ L $ using a jack-knife cross-validation scheme that maximizes out-of-sample likelihood across years.
• Ensure robustness by excluding the current year’s data when estimating $ p $ for each hurricane point, thus avoiding overfitting.
• Validate the model by comparing simulated lysis distributions and track densities with observations, particularly focusing on coastal landfall rates and track density patterns.

Experimental results

Research questions

• RQ1 How can the lysis process of tropical cyclones be modeled more realistically to reduce systematic errors in track simulations?
• RQ2 What spatial structure in historical lysis patterns can be captured using a non-parametric, kernel-based approach with cross-validated lengthscale selection?
• RQ3 To what extent does improving the lysis model reduce biases in simulated track density, particularly in the high-density region off the U.S. East Coast?
• RQ4 How does the inclusion of a realistic lysis model affect the overall performance and realism of basin-wide track simulations?
• RQ5 What role does the coupling between track dynamics and lysis distribution play in shaping the fidelity of simulated hurricane tracks?

Key findings

• The optimal lysis lengthscale $ L $, determined via cross-validation, is 360 km (±10 km), balancing sampling noise and spatial smoothing.
• The new lysis model significantly reduces systematic errors in track simulations, particularly the overrepresentation of tracks in the eastern subtropical Atlantic.
• The model captures the overall spatial distribution of historical lysis events, though it underestimates lysis off Newfoundland and overestimates it north of the Canary Islands—likely due to residual track model biases.
• The improved lysis model reduces the artificial eastward curvature of simulated tracks, bringing them closer to observed patterns.
• The model’s simulated landfalling rates show strong agreement with observations in key regions, with low variability in high-frequency zones and higher uncertainty in low-frequency regions.
• The ensemble standard deviation of simulated landfalling rates across 20 realizations reflects the influence of sampling variability, indicating that observed rates in low-hurricane regions may be less reliable estimators of true risk.