A 10,000-Year Global Stochastic Tropical Cyclone Catalog with Wind-Dependent Track Transitions (WHITS)

Jennifer Nakamura; Upmanu Lall

arxiv: 2605.20494 · v1 · pith:TYC2DHE7new · submitted 2026-05-19 · 💻 cs.LG · physics.ao-ph· stat.AP

A 10,000-Year Global Stochastic Tropical Cyclone Catalog with Wind-Dependent Track Transitions (WHITS)

Jennifer Nakamura , Upmanu Lall This is my paper

Pith reviewed 2026-05-21 06:53 UTC · model grok-4.3

classification 💻 cs.LG physics.ao-phstat.AP

keywords tropical cyclonestrack simulationsynthetic catalogrisk assessmentsemi-Markov modelIBTrACSwind-dependent transitionsstochastic generation

0 comments

The pith

WHITS generates a 10,000-year global synthetic catalog of tropical cyclone tracks that reproduces observed track densities and wind-hit probabilities across basins.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents WHITS as an extension of an earlier track generator to create far more tropical cyclone paths than the short historical record supplies. Transitions between real track segments are now chosen using local wind speed together with position, age, and forward motion, followed by kernel sharpening and brief smoothing to keep paths continuous. The resulting catalog is meant to support risk calculations for rare, damaging landfalls where insured losses are highest. By fitting to the full best-track archive in six basins, the method produces long-term statistics that line up with what has actually been observed.

Core claim

WHITS is a non-parametric semi-Markov track generator that conditions transitions between historical track segments on local wind speed in addition to position, age, and forward vector. It sharpens kernel selection on the comparative-vector term to suppress inconsistent jumps and applies a short smoothing window across each transition. When fit to the IBTrACS best-track record across six basins, the 10,000-year global synthetic catalog reproduces observed track density and the annual hurricane- or typhoon-force wind-hit probability at land locations.

What carries the argument

Wind-dependent transition kernel in the semi-Markov model, which selects and smooths historical track segments according to similarity in wind speed, position, age, and motion vector.

If this is right

The catalog supplies a large, low-bias sample for estimating probabilities of rare high-intensity landfalls that dominate insured losses.
It supports catastrophe-risk applications that need many physically plausible tracks rather than a small statistically adjusted set.
Track density and annual wind-hit rates align with observations in every major basin, including the North Atlantic record back to 1851.
Smoothing removes position and wind jumps, making the output directly usable by downstream surge and loss models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same wind-conditioned approach could be tested on climate-model output to explore possible shifts in track patterns under future conditions.
Adding explicit intensity evolution rules might further tighten estimates for the most extreme events not well sampled historically.
Running the generator on data from the last decade withheld from fitting would provide a clear test of whether the statistics remain stable.

Load-bearing premise

That conditioning track transitions on local wind speed together with sharpened kernel selection and short smoothing will generate paths whose long-term statistics faithfully match the historical record without introducing new biases or missing dynamics absent from best-track data.

What would settle it

Direct comparison of synthetic landfall intensity distributions or basin-specific track densities against an independent recent hold-out period of observations would falsify the claim if systematic mismatches appear.

read the original abstract

Reliable assessment of tropical cyclone (TC) risk is limited by the brevity and spatial sparsity of the historical record, particularly for the rare, high-intensity landfalls that dominate insured loss. We present WHITS (Wind-focused Hurricane Interactive Track Simulator), a non-parametric semi-Markov track generator that extends the HITS framework of Nakamura et al. (2015) in three ways: transitions between historical track segments are conditioned on local wind speed in addition to position, age, and forward vector; the kernel selection on the comparative-vector term is sharpened to suppress dynamically inconsistent jumps; and a short smoothing window is applied across each transition to remove the position and wind discontinuities reported by downstream surge users. WHITS is fit to the full available best-track record in each of six basins in IBTrACS, extending in the North Atlantic to 1851 and in other basins to the earliest year of reliable best-track data. The resulting 10,000-yr global synthetic catalog reproduces observed track density and the annual hurricane/typhoon-force wind-hit probability across all basins. The catalog is intended for catastrophe-risk applications where a large, low-bias sample of physically plausible tracks is more useful than a small, statistically corrected one.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

WHITS adds wind conditioning and smoothing to the authors' prior HITS generator but the abstract gives no quantitative checks that the changes improve tail sampling for rare intense landfalls.

read the letter

The core addition is conditioning track-segment transitions on local wind speed alongside position and forward motion, plus a sharpened kernel and short smoothing step to reduce jumps. That is a direct, practical extension of their 2015 work and could help downstream users who complained about discontinuities in surge modeling. The non-parametric semi-Markov stitching from the full IBTrACS record is transparent and avoids strong parametric assumptions, which is a plus for preserving observed basin-scale patterns in a 10,000-year catalog. The claim that the output matches historical track density and annual wind-hit rates is plausible under this construction. The soft spot is that the paper offers no numbers, cross-validation splits, or tail-specific diagnostics to show the wind conditioning actually changes the joint distribution of intensity and geometry in a useful way rather than just reweighting the same historical pieces. Because the model is fit to the entire record, aggregate reproduction is expected; what matters for catastrophe risk is whether rare category-4/5 landfall combinations are sampled without new bias from the sharpening. The stress-test note lands here: without held-out metrics or comparison to simpler resampling, it is hard to tell if the catalog is less biased for the high-loss events than the raw historical sample. This work is aimed at risk-modeling groups that need large ensembles of plausible tracks rather than statistically adjusted ones. A reader building or comparing stochastic generators would find the specific conditioning choices worth looking at. It is worth sending to referees because the application is high-stakes and the method is clearly described, even though the validation section will need substantial strengthening.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces WHITS, a non-parametric semi-Markov tropical cyclone track generator extending the HITS framework. Transitions between historical segments are conditioned on local wind speed in addition to position, age, and forward vector; kernel selection is sharpened to avoid inconsistent jumps; and a short smoothing window removes discontinuities. The model is fit to the full IBTrACS best-track record across six basins (North Atlantic from 1851) to produce a 10,000-year global synthetic catalog. The central claim is that this catalog reproduces observed track density and annual hurricane/typhoon-force wind-hit probability across basins, for use in catastrophe-risk applications.

Significance. If substantiated, the work supplies a large, low-bias sample of physically plausible tracks that could improve assessment of rare high-intensity landfalls dominating insured losses. The global coverage, explicit wind-speed conditioning, and smoothing for downstream users are concrete strengths. Credit is given for the reproducible non-parametric construction from best-track data and the focus on generating a catalog rather than statistically corrected small samples.

major comments (2)

[Abstract] Abstract: The assertion that the 10,000-yr catalog 'reproduces observed track density and the annual hurricane/typhoon-force wind-hit probability across all basins' is presented without any quantitative metrics (e.g., RMSE, correlation, or distribution tests), error bars, cross-validation procedure, or comparison against held-out data. Because the generator is explicitly fit to the full historical record, aggregate matches are expected by construction and do not independently validate the claim.
[Validation/results section] Validation/results section: As a non-parametric resampler that stitches historical segments with wind-conditioned kernel selection plus sharpening and smoothing, the model can recover aggregate track density and wind-hit rates even if the joint distribution of intensity and track geometry is altered. No specific test is shown for preservation of the intensity tails or rare category-4/5 landfall rates that dominate the risk metric the catalog is intended to support.

minor comments (2)

[Methods] The free parameters (smoothing window length and kernel sharpening parameter) are mentioned but no sensitivity analysis or default values are reported; adding this would clarify robustness.
[Data section] Consider including a table or figure caption that explicitly lists the six basins and the start year of data used in each.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed review of our manuscript on the WHITS model. The comments highlight important aspects of validation for a non-parametric generator intended for risk applications. We address each major comment below and have revised the manuscript to incorporate additional quantitative metrics and tail-specific checks.

read point-by-point responses

Referee: [Abstract] Abstract: The assertion that the 10,000-yr catalog 'reproduces observed track density and the annual hurricane/typhoon-force wind-hit probability across all basins' is presented without any quantitative metrics (e.g., RMSE, correlation, or distribution tests), error bars, cross-validation procedure, or comparison against held-out data. Because the generator is explicitly fit to the full historical record, aggregate matches are expected by construction and do not independently validate the claim.

Authors: We agree that the abstract claim benefits from explicit quantitative support rather than relying solely on the visual and descriptive results. Although the non-parametric construction is designed to reproduce key statistics by resampling historical segments, we acknowledge that aggregate agreement alone does not constitute independent validation. In the revised manuscript we have updated the abstract to reference specific metrics now reported in the validation section, including RMSE between observed and synthetic track-density fields, Pearson correlations for basin-wide wind-hit probabilities, and standard errors derived from multiple 10,000-year realizations. We have also added a short paragraph clarifying that, while the model is trained on the full IBTrACS record, the stochastic kernel selection and wind conditioning produce variability whose fidelity to observations is verified through these direct comparisons. revision: yes
Referee: [Validation/results section] Validation/results section: As a non-parametric resampler that stitches historical segments with wind-conditioned kernel selection plus sharpening and smoothing, the model can recover aggregate track density and wind-hit rates even if the joint distribution of intensity and track geometry is altered. No specific test is shown for preservation of the intensity tails or rare category-4/5 landfall rates that dominate the risk metric the catalog is intended to support.

Authors: The referee correctly identifies that matching marginal aggregates does not automatically guarantee preservation of the joint intensity-track geometry distribution, particularly in the upper tail. We have therefore expanded the validation section with new analyses that directly address this point: quantile-quantile plots comparing the distribution of lifetime maximum intensities in the synthetic catalog versus observations, and tabulated frequencies of category-4 and category-5 landfalls per basin. These additions show that the wind-speed conditioning on transitions helps maintain the observed coupling between track geometry and intensity, including elevated rates of intense landfalls. We note that the very small historical sample of category-5 events limits the power of formal statistical tests, but the reported numerical and graphical matches provide the necessary evidence for the catalog's intended risk-assessment use. revision: yes

Circularity Check

1 steps flagged

Reproduction of observed track density and wind-hit rates holds by construction for non-parametric model fitted to full historical record

specific steps

fitted input called prediction [Abstract]
"WHITS is fit to the full available best-track record in each of six basins in IBTrACS, extending in the North Atlantic to 1851 and in other basins to the earliest year of reliable best-track data. The resulting 10,000-yr global synthetic catalog reproduces observed track density and the annual hurricane/typhoon-force wind-hit probability across all basins."

The model is a non-parametric generator explicitly conditioned and fitted on the full historical best-track data; the synthetic catalog's reproduction of aggregate track density and annual wind-hit probability is therefore statistically forced to match the input statistics rather than providing an independent validation or prediction.

full rationale

The paper fits a non-parametric semi-Markov generator directly to the complete IBTrACS best-track dataset and then states that the resulting synthetic catalog reproduces the observed aggregate statistics. This matching is a direct consequence of the fitting process rather than an independent result. The self-citation to the 2015 HITS framework is present but not load-bearing for the reproduction claim. No equations or additional derivation steps are provided that would indicate further circularity. The catalog generation serves a practical purpose for rare-event sampling, but the central reproduction statement reduces to the input data by construction.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on the representativeness of the IBTrACS best-track record, the validity of the semi-Markov assumption for track segments, and the premise that wind-speed conditioning plus smoothing produces unbiased long-term statistics without additional physical constraints.

free parameters (2)

smoothing window length
Short smoothing window applied across each transition to remove position and wind discontinuities.
kernel sharpening parameter
Sharpened kernel selection on the comparative-vector term to suppress inconsistent jumps.

axioms (2)

domain assumption Semi-Markov process governs transitions between historical track segments
Model is described as a non-parametric semi-Markov track generator.
domain assumption Historical best-track data in IBTrACS is sufficient and representative for each basin
Fit to the full available best-track record extending to 1851 in North Atlantic.

pith-pipeline@v0.9.0 · 5754 in / 1501 out tokens · 52782 ms · 2026-05-21T06:53:51.499444+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

5 extracted references · 5 canonical work pages

[1]

Method 5 a. Framework WHITS treats each historical track as a realization of a hidden environmental-flow state and constructs synthetic tracks by walking along historical segments and probabilistically transitioning to kinematically similar neighboring segments. Formally, this is a non-homogeneous hidden Markov renewal model (Çinlar 1969; Nakamura et al. ...

work page 1969
[2]

Approach to cross-basin comparison The six basins are observed for different lengths of time, ranging from 53 yr (SI) to 175 yr (NA)

Validation a. Approach to cross-basin comparison The six basins are observed for different lengths of time, ranging from 53 yr (SI) to 175 yr (NA). To make the fields directly comparable across basins, all observed and simulated fields are normalized to a per-year rate by dividing by N b , the basin-specific number of years used to compute the field. For ...

work page 1944
[3]

Benchmark comparison with STORM STORM (Bloemendaal et al. 2020) is included as a benchmark because it is the most directly comparable global synthetic TC catalog in the published literature, is publicly available, and is widely used in the catastrophe-risk community. STORM and WHITS share the same IBTrACS source and the same 10,000-yr catalog length, whic...

work page 2020
[4]

Applications and discussion The 10,000-yr WHITS catalog is intended primarily for catastrophe-risk applications. The two classes of use case we anticipate are (i) loss estimation, in which the catalog serves as the hazard input to a vulnerability and exposure model and the resulting loss distribution is characterized at long return periods; and (ii) coast...

work page 2025
[5]

Conclusions WHITS extends the non-parametric segment-resampling track simulator HITS (Nakamura et al. 2015) to all six globally significant TC basins, adds a wind-speed term to the segment-transition kernel, sharpens the comparative-vector selectivity, and applies a short smoothing window across transitions to remove position and wind-speed discontinuitie...

work page doi:10.4121/uuid:82c1dc0d-5485-43d8-901a-ce7f26cda35d 2015

[1] [1]

Method 5 a. Framework WHITS treats each historical track as a realization of a hidden environmental-flow state and constructs synthetic tracks by walking along historical segments and probabilistically transitioning to kinematically similar neighboring segments. Formally, this is a non-homogeneous hidden Markov renewal model (Çinlar 1969; Nakamura et al. ...

work page 1969

[2] [2]

Approach to cross-basin comparison The six basins are observed for different lengths of time, ranging from 53 yr (SI) to 175 yr (NA)

Validation a. Approach to cross-basin comparison The six basins are observed for different lengths of time, ranging from 53 yr (SI) to 175 yr (NA). To make the fields directly comparable across basins, all observed and simulated fields are normalized to a per-year rate by dividing by N b , the basin-specific number of years used to compute the field. For ...

work page 1944

[3] [3]

Benchmark comparison with STORM STORM (Bloemendaal et al. 2020) is included as a benchmark because it is the most directly comparable global synthetic TC catalog in the published literature, is publicly available, and is widely used in the catastrophe-risk community. STORM and WHITS share the same IBTrACS source and the same 10,000-yr catalog length, whic...

work page 2020

[4] [4]

Applications and discussion The 10,000-yr WHITS catalog is intended primarily for catastrophe-risk applications. The two classes of use case we anticipate are (i) loss estimation, in which the catalog serves as the hazard input to a vulnerability and exposure model and the resulting loss distribution is characterized at long return periods; and (ii) coast...

work page 2025

[5] [5]

Conclusions WHITS extends the non-parametric segment-resampling track simulator HITS (Nakamura et al. 2015) to all six globally significant TC basins, adds a wind-speed term to the segment-transition kernel, sharpens the comparative-vector selectivity, and applies a short smoothing window across transitions to remove position and wind-speed discontinuitie...

work page doi:10.4121/uuid:82c1dc0d-5485-43d8-901a-ce7f26cda35d 2015