Dataset documentation

TC-RADAR v4.0

Tropical Cyclone Radar Archive of Doppler Analyses with Recentering: NOAA P-3 tail Doppler radar analyses of tropical cyclones on storm-centered grids, with best-track and environmental metadata.

Michael Fischer (University of Miami, CIMAS) · Paul Reasor, Robert Rogers, and John Gamache (NOAA/AOML Hurricane Research Division)
First released 3 April 2020 · this version November 2026
1,640
swath analyses
474
merged analyses (one per flight)
1997–2025
Atlantic, eastern and central Pacific
2 × 0.5 km
horizontal × vertical grid spacing

Overview

NOAA's WP-3D (P-3) hurricane hunters carry an X-band tail Doppler radar (TDR) that scans fore and aft of the aircraft, so a variational analysis can retrieve the three-dimensional wind field (Gamache et al. 1995). Each pass through the storm center yields one swath analysis. Observations reach roughly 50 km from the aircraft, so a single swath covers only the region near the flight track. A flight usually makes three or four center passes along different headings; averaging every swath from one flight gives a merged analysis with far fuller azimuthal coverage.

TC-RADAR gathers the post-processed swath and merged analyses into two files on common storm-centered grids, with best-track and environmental metadata attached to every analysis, so storms can be compared directly and composited. Analyses span the whole life cycle, from pre-genesis disturbances to mature hurricanes and storms nearing extratropical transition.

Swath analyses

One analysis per center pass, about an hour of observations.

  • Convective-scale features, which merging smooths out
  • Pass-to-pass evolution within a flight
  • Begin and end times and per-level coverage for each pass

Merged analyses

All swaths from one flight averaged about their 2-km centers.

  • Vortex-scale structure: azimuthal means, wavenumber asymmetries, tilt
  • Core-averaged wind and shear profiles (merged file only)
  • Number of swaths that went into each analysis
Swath analyses per year

Download

Two netCDF-4 files, one per analysis type, each holding every case from 1997 to 2025. Downloads can be resumed, so an interrupted transfer picks up where it stopped.

Merged analyses

7.1 GB
474 flights · recommended starting point
tc_radar_v4.0_xy_rel_merge.nc
SHA-256 f93ea152b77f1c071a49ccc7f4f0215aeebc2a390b0198d9751b0fe42a00619a

Swath analyses

10.6 GB
1,640 center passes
tc_radar_v4.0_xy_rel_swath.nc
SHA-256 5b0378fd40aaa3f430ded4bcb63f227aa8176564daa7da6425dbd00f93ca2b59

From the command line

# -c resumes an interrupted download
wget -c https://cdn.tcatlas.org/tc-radar/v4.0/tc_radar_v4.0_xy_rel_merge.nc
wget -c https://cdn.tcatlas.org/tc-radar/v4.0/tc_radar_v4.0_xy_rel_swath.nc

# check both files against the published checksums
curl -O https://cdn.tcatlas.org/tc-radar/v4.0/SHA256SUMS.txt
shasum -a 256 -c SHA256SUMS.txt      # Linux: sha256sum -c SHA256SUMS.txt

Downloading from a script? Use wget, curl, or Python's requests. The site's bot protection rejects Python's built-in urllib unless you set a User-Agent header, for example urllib.request.Request(url, headers={"User-Agent": "tcradar-download"}).

Usage policy

Users are free to download TC-RADAR. We ask that you tell Dr. Michael Fischer ([email protected]) and NOAA's Hurricane Research Division radar team lead, Dr. Paul Reasor ([email protected]), about your research project so the data are used appropriately. Examine the analyses carefully and apply any additional quality control your work needs; see Known issues.

Earlier releases, such as v3m (1997–2024), are available on request from Michael Fischer ([email protected]).

Quick start

Each analysis is stored as its own compressed chunk, so reading one case reads only that case from disk (about 0.4 MB per three-dimensional field), not the whole file.

import numpy as np
import xarray as xr

# Opening is lazy: only the cases you index are read from disk.
ds = xr.open_dataset("tc_radar_v4.0_xy_rel_merge.nc")

# Pick a flight by its mission ID (Hurricane Dorian, 1 September 2019)
case = int(np.flatnonzero(ds["mission_ID"].values == "190901H1")[0])
vt = ds["recentered_tangential_wind"].isel(num_cases=case)   # m/s, (y, x, height)

# Azimuthal mean about the 2-km TC center, in 2-km radial bins
radius = np.hypot(ds["eastward_distance"], ds["northward_distance"]).rename("radius")
edges = np.arange(0, 152, 2)
vt_mean = vt.groupby_bins(radius, edges, labels=edges[:-1] + 1).mean()

vt_mean.plot(x="radius_bins", y="height")   # radius-height cross section
  • Gridded fields are stored as 16-bit integers with a scale_factor. xarray, netCDF4-python, and MATLAB's ncread all unpack them on read. xarray and ncread return missing values as NaN; netCDF4-python returns a masked array.
  • Find analyses by mission_ID and time, not by index. Case indices change between versions as analyses are added.
  • Identify storms by tcid_ships (the best-track ID, such as AL162019). Some storm_name values are placeholders from the original analysis files, such as TDR, CYCLONE, or INVEST. Flights into disturbances before they became tropical cyclones have tcid_ships set to NAN; see Known issues.

Grids and centers

Every radar field comes on three grids that differ only in which center they follow. All three are storm-motion-relative unless the name says earth_relative. Choose a grid to see how it lines up with a tilted vortex.

Side view of a tilted vortex with the analysis grid at each level

Variables

Every variable in both files, read from the released netCDF headers. Search by name or description, filter by group, and click a name to copy it.

VariableDescriptionUnitsDimensions
Loading variables…

Environment and best track

Every analysis carries 17 snapshots of track and environmental data at 6-h intervals, from 48 h before to 48 h after the best-track time nearest the analysis (ships_lag_times; index 8 is the 0-h snapshot).

*_ships variables

Best-track position, intensity, and motion, plus SHIPS-style shear, humidity, potential intensity, SST, ocean heat content, and precipitable water. The names are kept for continuity with v3, but in v4.0 the track values come from IBTrACS and the environmental values are computed from ERA5 through TC-PRIMED, not taken from the SHIPS developmental data. Ocean heat content is the tropical cyclone heat potential from GLORYS12V1, averaged within 50 km of the best-track position.

era5_* variables

ERA5 diagnostics from TC-PRIMED: profiles on 23 pressure levels (1000–50 hPa), area averages over four regions (0–300, 0–500, 0–800, and 200–800 km), sector means in 15° azimuth sectors and 200-km radial bands, layer shear and helicity, cyclone phase space parameters, empirical and theoretical potential intensity, eddy flux convergence, and precipitable water in radial bins.

Because the sources changed, v4.0 environmental values differ from v3m. For the same flights, mpi_ships is 8.6 kt higher on average and shdc_ships 0.8 kt higher, while best-track intensity is unchanged. Don't mix the two versions in one analysis.

Units and angle conventions

VariableConvention
lon_shipsDegrees west, positive. Every other longitude in the files is in degrees east (western hemisphere negative).
tc_tilt_direction, core_shear_directionRadians, mathematical convention: the direction the vector points, 0 = east, increasing counterclockwise. Equal to atan2(northward, eastward) of the matching components.
sddc_ships, era5_shear_directionDegrees, compass heading toward which the shear vector points: westerly shear = 90°.
era5_azimuthDegrees, mathematical convention: 0 = due east of the TC center, increasing counterclockwise.
vmax_ships, shdc_ships, mpi_shipsKnots. Gridded radar winds are in m s−1.
tc_tilt_magnitude, tc_eastward_tilt, tc_northward_tiltKilometers, relative to the 2-km TC center.
mpi_shipsThe empirical, SHIPS-style potential intensity: identical to era5_potential_intensity_empirical, although its description in the file says theoretical. The Emanuel value is era5_potential_intensity_theoretical.
ohc_shipskJ cm−2, relative to the 26 °C isotherm; 0 where the surface is cooler than 26 °C.

Methods

Grids

Analyses use Cartesian grids with 2-km horizontal and 0.5-km vertical spacing from 0 to 18 km height (37 levels). The recentered grids span ±200 km (201 × 201 points) about the TC center. The 0-km level is kept for grid continuity only; it is missing everywhere because the Doppler analyses contain no valid data at the surface.

Finding the center

The TC center is found with a weighted circulation maximization (WCM) method (Fischer et al. 2024). For a series of trial centers, the storm-relative wind direction at each grid point is compared with purely tangential flow about that trial center. The best center is the one with the smallest mean difference. Each point's mismatch is weighted by wind speed and by distance from the trial center, so strong winds near the center count most. The search runs for every swath and merged analysis, at every height, which gives the center profile (tc_center_latitudes, tc_center_longitudes), the radius of maximum wind profile (tc_rmw), and the vortex tilt relative to the 2-km center (tc_tilt_magnitude, tc_tilt_direction, tc_eastward_tilt, tc_northward_tilt). Where coverage is too sparse to fix a center, these values are missing.

Building merged analyses

Each swath from a flight is first recentered on its own 2-km center, and the recentered swaths are averaged into the merged analysis. A revised center is then estimated from the merged tangential wind field; its greater azimuthal coverage should give a more accurate center than any single swath.

Core-averaged wind and shear

The merged file holds vertical profiles of the wind averaged over the TC core and of its vertical shear (core_*; the v3 README called this local shear). Averaging radii run from 25 to 100 km in 5-km steps (core_radii). A value is computed only where at least half of the points in every geographic quadrant (northwest, southwest, southeast, northeast) hold valid data, so small radii have values more often than large ones. The mean and minimum quadrant coverage are stored alongside (core_mean_quadrant_coverage, core_minimum_quadrant_coverage). Use these profiles with great caution.

Quality control

Quality control of the raw radar data (Level 1a) balances keeping as much meteorological signal as possible against removing non-meteorological artifacts; removing all artifacts would also remove real data. Because coverage matters for most research uses, the QC behind the Level 2 analyses in TC-RADAR does not aim to remove every artifact. The production of the Level 2 analyses since 2020 is described in the HRD radar data README.

Known issues

Vertical velocity before 2010
Analyses before the 2010 season were made with an error in the synthesis software that made vertical velocities too deep and too strong in some cases. Do not use the TDR vertical velocity quantitatively before 2010. Analyses from 2010 on have the error corrected. Details are in Fischer et al. (2022).
Reflectivity is not calibrated
Reflectivity can differ by up to 7–10 dB between the radars on the two P-3s (N42 and N43) in some seasons, and between seasons as the radar systems changed. The X-band signal is also attenuated in heavy precipitation, especially along swath edges, and TC-RADAR does not correct for it. Until a calibration is done, use reflectivity quantitatively only within individual case studies.
Analyses are lightly quality-controlled
Some non-meteorological artifacts remain by design (see Quality control). Examine the wind and reflectivity fields and apply further QC suited to your objectives.
Placeholder storm names
Some storm_name values come straight from the original analysis files and are not names, for example TDR, CYCLONE, INVEST, or PTC09. Use tcid_ships to identify the storm.
Pre-genesis flights have no track or environment data
Flights into disturbances before they became tropical cyclones (15 merged and 48 swath analyses, for example the precursor of Fay on 14 August 2008) have tcid_ships set to NAN and no best-track or environmental values at any lag.

Version history

v4.0 · Nov 2026
  • Adds the 2025 season (37 flights) and one 2019 flight, Tropical Storm Nestor (191018I1): 1,640 swath and 474 merged analyses (v3m had 1,510 and 436). No analyses were removed.
  • One file per analysis type; v3m split each type into two year ranges.
  • Track values now come from IBTrACS and environmental values from ERA5 via TC-PRIMED, replacing the SHIPS developmental data. The *_ships names are kept.
  • Adds 44 ERA5 environmental variables (era5_*), a CF time variable, and land masks (land_mask, recentered_land_mask, land_fraction).
  • mpi_ships is now the ERA5 empirical potential intensity, 8.6 kt higher on average than in v3m. Ocean heat content now comes from GLORYS12V1, in kJ cm−2.
  • The 0-km level is set to missing; the analyses contain no valid surface data.
  • Case indices match v3m through 2019 (the first 710 swath and 215 merged analyses) and shift after the added 2019 flight. Match analyses by mission ID and time.
v3m · Aug 2025
Added 2024 cases and the nearest best-track time to the SHIPS variables, for easier time referencing. The center-finding search was streamlined so a crude initial search is no longer needed; this may slightly improve swath tilt estimates in a handful of cases. The weights and cost function were unchanged.
v3l · Oct 2024
Added 2023 cases. Variable names revised to follow netCDF CF conventions, with long names added. Added four SHIPS precipitable water variables.
v3k · Jun 2023
Added 2022 cases. A larger core-radius parameter made center finding more consistent on swath analyses. The files were split into two swath and two merged files by year range to limit memory use, and some coordinate names were made easier to use.
Earlier versions (v3a–v3j)
v3j · May 2022
Added 2021 cases. Revised center finding to estimate centers more accurately from swath data, especially at mid-levels in weak storms, with a stricter coverage constraint. Fixed the divergence calculation.
v3i
Version used in the TC-RADAR paper (Fischer et al. 2022). Fixed the routine that computes each grid point's bearing, which affected center finding.
v3h
Corrected the latitudes and longitudes of each analysis for Earth's curvature and for the grid origin falling between grid points. Some centers and tilts changed.
v3g
Added swath begin and end times. Fixed date stamps for flights that crossed a month boundary, which had kept a few merged analyses from pairing with best-track and SHIPS data.
v3f
Added 2020 cases. Revised center finding so it no longer depends on swath RMW, which agreed better with subjective analyses.
v3e
Vortex tilt computed from swath data. Stricter coverage criteria for recentering; cases that fail are set to NaN.
v3d
Fixed best-track pairing for analyses between 21 and 00 UTC. Replaced the Earl (2010) analyses with an updated QC version.
v3c
Tuned the center-finding parameters and improved code efficiency.
v3b
Fixed center-finding bugs.
v3a
Beta version of the new center-finding method.

How to cite

Publications and presentations that use TC-RADAR should cite the database paper. Work that uses the center finding or vortex tilt should also cite the tilt paper.

TC-RADAR database
Fischer, M. S., P. D. Reasor, R. F. Rogers, and J. F. Gamache, 2022: An analysis of tropical cyclone vortex and convective characteristics in relation to storm intensity using a novel airborne Doppler radar database. Mon. Wea. Rev., 150, 2255–2278, https://doi.org/10.1175/MWR-D-21-0223.1.
Center finding and vortex tilt
Fischer, M. S., P. D. Reasor, J. P. Dunion, and R. F. Rogers, 2024: An observational analysis of the relationship between tropical cyclone vortex tilt, precipitation structure, and intensity change. Mon. Wea. Rev., 152, 203–225, https://doi.org/10.1175/MWR-D-23-0089.1.

Data sources in v4.0

  • TC-PRIMED (ERA5 environmental diagnostics): Razin, M. N., C. J. Slocum, J. A. Knaff, and P. J. Brown, 2023: Tropical Cyclone PRecipitation, Infrared, Microwave, and Environmental Dataset (TC PRIMED). V1.0. NOAA NCEI, doi:10.25921/dmy1-0595.
  • ERA5: Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, doi:10.1002/qj.3803.
  • IBTrACS: Knapp, K. R., M. C. Kruk, D. H. Levinson, H. J. Diamond, and C. J. Neumann, 2010: The International Best Track Archive for Climate Stewardship (IBTrACS). Bull. Amer. Meteor. Soc., 91, 363–376, doi:10.1175/2009BAMS2755.1.
  • GLORYS12V1 ocean reanalysis: E.U. Copernicus Marine Service, Global Ocean Physics Reanalysis, doi:10.48670/moi-00021.
  • TDR analysis method: Gamache, J. F., F. D. Marks, and F. Roux, 1995: Comparison of three airborne Doppler sampling techniques with airborne in situ wind observations in Hurricane Gustav (1990). J. Atmos. Oceanic Technol., 12, 171–181, doi:10.1175/1520-0426(1995)012<0171:COTADS>2.0.CO;2.

Questions about the data: Michael Fischer ([email protected]) or the HRD radar team.