the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 30 Jul 2026
RABTracks: Reanalysis-Augmented Best Tracks for Tropical Cyclones
Abstract. A new dataset, Reanalysis-Augmented Best Tracks for Tropical Cyclones (RABTracks), is created for the study of Tropical Cyclones (TCs) globally. The goal of RABTracks is to bring together multiple sources of information on historical tropical cyclones in one place. To do so, we augment the International Best Track Archive for Climate Stewardship (IBTrACS) with track data from five analyses (four reanalyses and one operational analysis), using two different tracking methods, in all basins, from 1940 to present. This provides us, in particular, with extended tracks that reconstruct the early and late stages of each recorded cyclone. It is also an opportunity to gather information on cyclones’ intensity, size, and to retrieve information about the cyclones’ nature at different stages, derived from the Cyclone Phase Space (Hart, 2003), and SyCLoPS classification (Han and Ullrich, 2025). This dataset has a high potential to support robust TC-related climatological studies. It was also designed to be easy to extend with additional data, and all the code for creating the dataset, making the analyses and adding new data is provided in an open-source repository.
- Preprint
(1159 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on essd-2026-403', Anonymous Referee #1, 19 Aug 2026
-
RC2: 'Comment on essd-2026-403', Anonymous Referee #2, 17 Sep 2026
This manuscript presents RABTracks, a valuable multi-source dataset that augments IBTrACS with trajectories and storm characteristics derived from multiple reanalyses and tracking methods, thereby extending the recorded lifecycles of tropical cyclones into their precursor and post-tropical stages. The dataset has considerable potential for case studies and climatological analyses of tropical cyclone genesis and extratropical transition. However, several aspects require clarification or further validation. I recommend publication after the authors adequately address the comments below.
1. Sections 3.1.1 and 4.2
The matching criteria described in Sections 3.1.1 and 4.2 are inconsistent. Section 3.1.1 introduces “[dmax]” as the distance threshold but subsequently refers to an undefined dmin=400 km. In contrast, the following text and Section 4.2 state that only one time step within 300 km is required. The latter appears to correspond to the original procedure of Bourdin et al. (2022), whereas Section 3.1.1 appears to describe a modified procedure. Please clearly define the distance threshold using consistent notation, state whether the actual threshold is 300 or 400 km, and clarify whether one or three paired points are required.2. Section 3.1.1
When multiple detected tracks are matched to the same IBTrACS storm, they are merged into a single source-specific track. However, the criteria used in this merging step are not sufficiently described. Please clarify whether a maximum temporal gap, spatial separation, or continuity criterion is imposed between the track segments. The treatment of overlapping segments, conflicting positions, and missing periods should also be explained. These details are important because merging physically unrelated segments could produce an apparently continuous precursor or post-tropical track.3. Sections 3.2.2
Figure 5 provides a clear example in which the SyCLoPS-MERRA2 track follows to a different disturbance. However, the manuscript does not quantify how frequently such cases occur. Figure 12 evaluates the distance between each analysis track and IBTrACS, but it does not directly assess track positional disagreement among the analysis-derived tracks, particularly outside the IBTrACS period. Furthermore, Figure 13 demonstrates that TRACK and SyCLoPS frequently produce longer tracks, but track length alone does not validate the accuracy of those extensions; erroneous linkage to another disturbance would also produce an apparently longer track. To establish confidence in the tracking and matching methodology, the authors should demonstrate that erroneous linkage to an unrelated disturbance events and large discrepancies among the reanalysis-derived tracks are rare.4. lines 216–220
The authors first state that ETCs are better represented than TCs in analyses, but then state that analysis models “do not resolve better” the baroclinic processes involved in the extratropical phase than the deep-convective or mesoscale processes important during the tropical phase. Did the authors instead mean that analysis models resolve the baroclinic processes better? Please revise this sentence for clarity and correct the typo “than the the.”5. Section 4.1
The use of 2006–2022 provides a fair comparison across all sources because this is their common period. However, excluding ECMWF-OP-AN would allow the validation period to be extended to 1980–2022. Please examine the temporal variation in hit rates for each reanalysis–tracker combination over this longer period. This information is necessary to assess the temporal homogeneity of the dataset and determine whether it is suitable for analyses of climatological variability.6. Sections 4.4 and 5.2
RABTracks provides multiple source-specific trajectories for each IBTrACS storm rather than a single consolidated track. However, the manuscript provides limited practical guidance on how users should select or combine these trajectories when the sources disagree. The authors should provide guidance for single-source and consensus-based analyses, including whether two trackers applied to the same reanalysis should be regarded as independent evidence, how large positional discrepancies and missing sources should be handled, and what agreement criteria are recommended for precursor, extratropical-transition, and climatological studies. Such guidance is important because different choices can produce substantially different results from the same dataset, as illustrated by the decrease from 25 to seven French landfall cases as the required number of agreeing sources increases.Citation: https://doi.org/10.5194/essd-2026-403-RC2
Data sets
Reanalysis-Augmented Best Tracks Stella Bourdin https://zenodo.org/records/20426511
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 240 | 120 | 137 | 497 | 174 | 115 |
- HTML: 240
- PDF: 120
- XML: 137
- Total: 497
- BibTeX: 174
- EndNote: 115
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
In this manuscript, the authors present RABTracks, a global tropical cyclone dataset that augments IBTrACS best tracks with matched tracks from a set of different reanalyses using two tracking approaches (SyCLoPS and TRACK). The dataset includes added information on storm intensity, size, cyclone phase space, and lifecycle classification. The authors suggest the dataset could be useful for TC climatology, transition studies, and post-tropical impacts because it explicitly targets early and late lifecycle stages that are often inconsistently represented in best-track archives.
In general, I find the idea novel. There has been a great deal of work evaluating tracked TCs in reanalyses over the last decade or so, and the authors cite many of these papers (although Schenkel et al., 2012, 10.1175/2011JCLI4208.1, is conspicuously absent). However, the idea of standardizing and compiling all of these across multiple tracking algorithms seems unique. I was looking forward to seeing how such trajectories could have "storm-centered" fields attached to evaluate the simulated structure within the datasets and was disappointed they were not included in this submission, although the authors highlight this as a potential future exploration area!
Overall, the manuscript was a pleasant read and easy to follow, albeit a little rough around the edges, grammar-wise, and some sections are a little light on content. But I do like how the authors walk through different cases/examples that fall out of the dataset and seem to be "priming the pump" for future analyses. My main concern focuses on the inclusion of Section 4, which compares metrics in a very narrow sense that requires qualification since they lack "false alarm" context. This is described in detail below, along with some other questions that should be addressed/suggestions to consider before final publication.
--
Section 2. I assume the answer to this is "data availability," but why only three reanalyses for SyCLoPS versus 5 for TRACK? For the CPS/WCSI evaluation, why TRACK and not SyCLoPS?
Figure 3. The yellow and orange arrows seem different (between disks and tracker, and then the arrows below). Suggest homogenizing the arrows. Also, I am not sure unit conversions or download steps need to be highlighted; it is self-evident.
Section 3.1.1. Removing false alarms is acceptable here given the goal of matching IBTrACS, but it is worth noting that false alarm evaluation is extremely relevant for evaluating datasets where no known TC record exists (e.g., climate data). More below.
Section 3.1.1. The matching algorithm uses a distance-time formulation, with a maximum distance threshold of 400 km and a requirement for three matched points. Lines 170-171 are a bit inconsistent here with "Note that following this procedure, only one point is needed to match two tracks." These are probably defensible choices (although 400 km feels quite large for reanalyses), but have the authors explored how sensitive key outcomes are to these thresholds? For example, how do hit rates, false matches, track-extension lengths, and ET timing change if the distance threshold is 100 km or if the minimum matched-point criterion changes?
Section 4 feels a bit out of place in this paper. My gut reaction is that it’s a bit miscast. The authors use automated trackers, not manual tracking, but then throw out false alarms. It appears that the trackers are used in their "operational configurations" that are tuned to behave well on reanalysis but also on free-running NWP or climate data. As such, any implication of skill in this section must be considered in the context of ignoring false alarms.
Section 4.1: I'll admit that I am confused by the hit rate discussion. In Section 3.1.1, the authors state "False alarms are discarded." If that's the case, I don't see any reason why you wouldn't just use a very relaxed cyclone tracker (e.g., any tiny deficit in the sea-level pressure field or curl in the 850 mb wind), which would have a tremendous number of false alarms but would also have a very high hit rate. This section as written will inevitably lead to (perhaps informal) comparison of trackers, but that cannot be done without a more rigorous Type I/II error analysis (as many of the authors listed on this manuscript have previously done).
Section 4.2: This matching distance section is very brief; I believe the point is to show that, in general, the tracked centers in the reanalyses are well matched to IBTrACS. Do both trackers define the storm center by sea-level pressure? Are the metrics "standardized" so that they all match? I.e., if TRACK has a longer track for Storm A, do you include all TRACK's points or only the ones that SyCLoPS also has? I assume the answer is no, since if this was the case I would suspect the answers to be very, very similar between the two trackers for a given data product (i.e., it would seem rare that given the same gridded x, y data that TRACK and SyCLoPS would have matching distances that are different from one another, assuming they are both matching the track adequately).
Section 4.3: The track extension discussion dovetails with my comments from 4.1. It is very trivial to get extended tracks by reducing thresholds and permitting more false alarms. For a tracker-derived dataset that leverages reanalysis, this may be acceptable, but for operational NWP or climate applications, this is not (necessarily) a desirable trait. If the goal is to have significantly extended tracks relative to IBTrACS for operational analysis, it may be worth retuning SyCLoPS and/or TRACK for this specific application. Alternatively, one could "merge" TC/ETC tracks with other non-cyclonic feature detection (e.g., easterly wave, tropical convective cluster trackers, see 10.5194/gmd-17-6035-2024). Otherwise, I'm not sure comparing the extended tracks between the datasets here adds value in the absence of a more technical discussion of how the extended tracks are arrived at.
Section 4.4. This section also would seem to fall into the same trap of "tracker tuning." E.g., would Figure 14 look different if you used TRACK vs. SyCLoPS? Would it look different with a very permissive pressure-minimum-only tracker? I would suspect the CTO agreement is dependent on the tracker configuration.
Section 4.5. This would seem more logical to have as part of Section 5.
Section 5.1. What does "computing additional attributes" mean?
Section 5.1. Is the plan for the authors to update RABTracks on a yearly basis akin to NHC Best Tracks or IBTrACS updates? Is this trivial for a user to do with the provided code? Could it be automated in some way?
Are 10m winds actual 10m winds in the reanalyses (using their internal PBL algorithms) or winds corrected from a different level (e.g., from the free atmosphere or the lowest model level)?
Minor comments/typographical errors:
Generally, "NCEP" has historically been used colloquially to mean the "NCEP-NCAR reanalysis." See https://en.wikipedia.org/wiki/NCEP/NCAR_Reanalysis. In this manuscript, "NCEP" should be replaced with "CFSR."
Throughout: Both "extra-tropical" and "extratropical" appear. Suggest picking one.
Line 86: "what ever" should be "whatever"
Line 103: "EMCWF-OP-AN" flips M and C
Figure 1: "breaking points" should probably be "data discontinuities" or thereabouts.
Line 219: "the the"
Line 222: IBTrACS needs a capital T.
Line 224: Need to define "nature." It might be worth simplifying this to "cyclone type."
Line 251: "C-shaped?" Classic recurving Cape Verde storm?
Line 266-267: For an intense TC that is relatively axisymmetric, the RMW method described here shouldn't differ much from the methods outlined earlier (calculating the profile of azimuthal wind and finding the radius at which it is maximized). So this is almost assuredly a resolution issue. See Davis (2018, GRL).
Figure 7: "times are the" should be "times as the"
Line 276: Would make this a standalone equation to align with the thermal wind equations. Also, would number equations, even if not referred to in text.
Line 284-285: Is this a very onerous computational cost? Seems the biggest challenge is acquiring the reanalysis data.
Line 300: "the storm lost it warm core" should be "the storm lost its warm core."
Line 314: There is some additional literature describing the "pathway" of undergoing ET that may be relevant here in addition to the Bieli work (10.3402/tellusa.v67.24474, 10.1002/2016MS000775, 10.1016/j.isci.2025.112814).
Line 314: "unexact" should be "inexact"
Line 316: "traditionnally" should be "traditionally"
Line 345: "in in"