Data Disclosure

The Tail Doppler Radar is mounted on the tail of NOAA's hurricane hunter aircraft. These are the WP-3D Orions N42RF and N43RF and the Gulfstream IV-SP N49RF. The dataset has the raw radar data recorded on the aircraft. It also has the wind and reflectivity analyses made from it for each flight. The data comes from NOAA's yearly Hurricane Field Program. The program works with the NOAA Office of Marine and Aviation Operations (OMAO) Aircraft Operations Center.

• Provider: NOAA Atlantic Oceanographic and Meteorological Laboratory, Hurricane Research Division (AOML/HRD)
• Principal scientists (Level 1 & 2 data): Dr. Paul Reasor and Dr. John Gamache, NOAA/AOML/HRD

Data levels shown on this site

• Level 1b (real-time). Made on the aircraft during the flight with automated quality control. It's sent to a ground server and archived as it is. It has little or no quality checking. Use it for situational awareness only. Retrieved from the OMAO real-time archive. This site indexes, processes and serves it.
• Level 2 (post-processed). Made again on the ground after each hurricane season. Each analysis has been inspected and meets basic standards for research. Only analyses from 2020 on are formally part of the Level 2 database. Retrieved from the AOML/HRD Level 2 archive. This site indexes and serves it.
• Level 1 + Experimental QC. An extra automated quality check that this site runs on the Level 1b grid. It is not an official NOAA, NHC or HRD product. They haven't reviewed or endorsed it, and it hasn't been checked against Level 2.

NOAA strongly advises users of Level 1 and Level 2 products to do their own quality control. HRD flags some Level 2 analyses as unsuitable for use. See the reprocessing notes in the TDR dataset README.

Interpreting the data

• The 3D analyses use a 2 km grid across and a 0.5 km grid up. They reach 250 km from the grid's center. That is usually the storm's center at flight level. G-IV (N49RF) flights may use other spacing and sizes.
• The lowest level (z = 0) is not the surface.
• Winds are Earth-relative. Products marked storm-relative (_rel) have a constant storm motion taken out.
• Vertical profiles use data up to 10 km from the flight track. They aren't always true lines out from the storm's center.
• For flights that cross 00 UTC, NOAA writes analysis times on a 48-hour clock. For example, 2530 is 01:30 UTC the next day.

Data use and attribution

Unless stated otherwise, AOML/HRD data belongs to the U.S. Department of Commerce and is fully open. tropicalrecon.com is not affiliated with or endorsed by NOAA. The visualizations on this site are for information only. To use TDR data in research or publications, get it directly from NOAA. Follow the AOML/HRD Data Policy and User Agreement.

Method references

• 3D synthesis Gamache, J. F., 1997: Evaluation of a fully three-dimensional variational Doppler analysis technique. Preprints, 28th Conf. on Radar Meteorology, Austin, TX, Amer. Meteor. Soc., 422–423.
• 3D synthesis Reasor, P. D., M. Eastin, and J. F. Gamache, 2009: Rapidly intensifying Hurricane Guillermo (1997). Part I: Low-wavenumber structure and evolution. Mon. Wea. Rev., 137, 603–631.
• Profile synthesis Zhang, J. A., Rogers, F., Reasor, P. D., and Gamache, J., 2022: The mean kinematic structure of the tropical cyclone boundary layer and its relationship to intensity change. Mon. Wea. Rev., 151, 63–84.
• Quality control Gamache, J. F., 2005: Real-time dissemination of hurricane wind fields determined from airborne Doppler radar data. National Hurricane Center, 38 pp. Available online.

Source documentation

• TDR dataset README (NOAA/AOML/HRD)
• Guide to AOML/HRD Data Accessibility (data policy and user agreement)
• HRD real-time Doppler radar

Dropsondes are small instrument packages dropped from hurricane hunter aircraft. NOAA flies the WP-3D Orions N42RF and N43RF and the Gulfstream IV-SP N49RF. The U.S. Air Force Reserve's 53rd Weather Reconnaissance Squadron flies the WC-130J. Each sonde falls by parachute from flight level to the ocean. It measures pressure, temperature and humidity on the way down. It tracks its own position by GPS. The wind comes from that. The aircraft's Airborne Vertical Atmospheric Profiling System (AVAPS) receives the data during the fall.

• Provider: NOAA Office of Marine and Aviation Operations (OMAO), Aircraft Operations Center
• Retrieved from: the OMAO real-time data server and OMAO's post-flight AVAPS archive. The aircraft post what they send to the real-time server. NOAA's TEMP DROP messages also come from NHC (latest, NWS and archive, NHC). OMAO's folders often miss them. This site indexes, decodes and serves the data.
• Air Force Reserve sondes: the coded TEMP DROP messages NHC sends out (latest, NWS and archive, NHC). They're only published in this form.

Data shown on this site

• Post-flight soundings (ASPEN). Each drop's full sounding after ASPEN quality control. It has a level every quarter second. It's posted to the AVAPS archive a week or more after the flight. It includes drops that were never sent, such as every G-IV drop. Once posted, it replaces every other version of the same drop. Some TEMP DROP messages have no real observation number. The G-IV's are all "OB 99". Those sondes are numbered in the order they were dropped.
• Soundings (BUFR). The detailed message each aircraft sends after a drop. It has a level every second of the fall, or about every 10 m. The sonde operator on board runs it through NCAR's ASPEN quality-control software before it's sent.
• TEMP DROP messages. The coded message sent with it. This site uses it for the mission, the storm and the operator's remarks. The remarks say where in the storm the sonde was dropped, such as "Eyewall 045" or "Center". They also give the splash position and the mean winds over the lowest 150 m (WL150), the lowest 500 m (MBL) and a deep layer (DLM).
• Coded soundings. Some drops have no detailed sounding, including every drop before 2020. For those, the sounding is decoded from the TEMP DROP message. These are quality-controlled but coarse. They have only standard and significant levels, usually 20 to 40. This site works out their heights from the surface pressure and temperatures. They're marked "TEMPDROP" (or "T") in the list.
• Preliminary soundings. The raw data each sonde sends back, as the aircraft receives it. It shows up a few minutes after launch. It stays until the quality-controlled sounding is sent. Drops that never get one keep it. These have not been through the operator's quality control. This site runs them through its own copy of ASPEN's quality-control steps instead (see "ASPEN QC on this site" below). These are automatic checks. They are not an official NOAA product and not a substitute for the operator's ASPEN run. They're clearly marked "Prelim" (or "P"). Don't rely on them. When a preliminary sonde's TEMP DROP message has also arrived, it's marked "TEMPDROP" too.
• One sonde, every version. Each drop is listed once and opens on its best version. A switch shows the others. These are the preliminary sounding, the TEMP DROP sounding and the quality-controlled BUFR (or ASPEN) sounding. A TEMP DROP is matched to its preliminary sonde by release time. When two drops were launched too close together to tell apart, it's listed on its own.

Until the post-flight sounding is posted, this is real-time data. It has only had the quality control done on board. None of it has had the post-season processing used for research datasets. Past seasons go back to 2016. That is as far back as the OMAO server keeps them. Some flights have no archived soundings and keep what was sent.

Air Force Reserve dropsondes

The Air Force Reserve's detailed soundings aren't published. Their drops are shown from the coded TEMP DROP message alone. It has only standard and significant levels, usually 20 to 40. They're marked "TEMPDROP" in the list. Each is labeled with the bulletin it came in. That is UZNT13 KNHC in the Atlantic, or UZPN13 or UZPA13 KNHC in the Pacific. NHC's archive files them as REPNT3, REPPN3 and REPPA3. They're the lowest-quality version on this page, below even preliminary soundings. They're also the only version of these drops there is.

• Flights are grouped by aircraft and mission code, as in the aircraft's HDOBs. An example is AF308, mission 0602C. Each is named by the date of its first drop. Training and weather reconnaissance flights use codes like WXWXA. They're included with no storm, as "Other". A storm follows its system from invest to named storm, by NHC's genesis number. If a flight's sondes name no storm, it takes the storm the Plan of the Day tasked it for.
• Launch times come from the release time in the operator's remarks. Without one, the launch time is the splash time less the fall time. The fall time assumes a typical 13 m/s from the release height. Failing that, it's the time the message was sent, less 10 minutes. Launch positions come from the remarks. Without those, they come from the message's coded position, to a tenth of a degree.
• Heights are worked out up from the surface, as for every coded sounding.
• The drift map shows the aircraft's track from its HDOBs.

ASPEN QC on this site

Preliminary soundings are put through the steps of NCAR's ASPEN, in ASPEN's order. The settings are the "mini-dropsonde" set NOAA AOC runs ASPEN with. Those settings come from the attributes of AOC's own ASPEN output files.

• The steps. Frames with checksum errors are dropped. Data after the end of the drop are removed. Data are removed while the sensors settle after launch. That time comes from ASPEN's sensor time constants. Then come the limit, GPS satellite, wind error, buddy, outlier and filter checks. Pressure, temperature, humidity and winds are smoothed with ASPEN's B-spline filter. The filter keeps more detail in the last 4 seconds near the surface. Temperature and winds are corrected for sensor lag and the sonde's fall. Winds are checked against the GPS fall speed. A surface point is added for the short fall after the last report. Heights are integrated up from the surface, or down from the aircraft when the sonde didn't reach it.
• What the operators also do. NOAA's operators edit nearly every sonde by hand. They end the drop just before the last sample, which is the splash. On the G-IV they remove temperatures above 190 hPa. That is where the RD41's sensor reads warm after a high launch. This site does both too, as their own marked steps.
• Warnings. ASPEN warns of post-splash data, where the sonde kept sending after it hit the water. It also warns of a fast fall, where the parachute likely failed. This site shows the same warnings as tags and a pop-up. It also flags heights that disagree with the aircraft's altitude, and sondes that stopped before the surface.
• QC flags. Under each sounding, a dropdown lists what each check removed. It can show the raw data on the Skew-T, with removed samples crossed out. It also lists the raw samples, with the check that removed each one. This is shown for any drop whose raw data is in, even after NOAA's version arrives.
• How close it comes. It was checked against NOAA's post-flight ASPEN soundings of the same 70 G-IV drops from September and October 2026, sample by sample. Pressure came within 0.01 hPa RMS, temperature within 0.07 °C, humidity within 0.02% and wind speed within 0.1 m/s. Heights came within 5 m RMS. The two kept and removed the same samples over 99.6% of the time. The rest are mostly where operators made one-off judgment calls by hand.
• Where this differs from ASPEN's docs. Two values were fitted to ASPEN's actual output. The temperature sensor's time constant is used without the docs' division by 4 for RS41 sensors. The settling time after launch uses 2·ln(ΔT) − 0.75 time constants, with a 6-second minimum.

Interpreting the data

• Heights are geopotential heights reported by the sonde. The surface is the level the sonde flags as the surface. In post-flight soundings, it's the last pressure within 30 m of the surface.
• Winds come from GPS and are Earth-relative. They're shown in knots. Wind barbs follow the usual convention. A pennant is 50 kt, a full barb 10 kt and a half barb 5 kt.
• The sonde drifts with the wind as it falls. So a sounding isn't a straight column above the launch point. The launch and splash positions are both listed. The sonde map shows each drift path. It also shows the aircraft's track from its HDOB reports when available.
• The notes below on preliminary heights, winds and fall rate apply only to the few preliminary soundings that couldn't go through ASPEN QC.
• Heights in preliminary soundings are worked out by this site. It integrates the hypsometric equation with virtual temperature. It starts at the surface if the sonde reached it. A sonde that stopped short is tied to its GPS altitudes instead. It uses their median offset. That way a few bad GPS fixes don't move it. Raw GPS altitude can drop out by hundreds of meters or even kilometers.
• Winds in preliminary soundings are a 3-second centered mean of the raw 4 Hz GPS winds.
• Fall rate. A preliminary sonde is flagged as doubtful if its typical fall speed is under 6 m/s or over 25 m/s. It's also flagged if it isn't descending for more than a fifth of its fall. Fall speed comes from its heights, over 10-second steps.
• How close preliminary soundings come. They were checked against the quality-controlled versions of the same 225 drops from 2026, at the same levels. Temperature came within 0.2 °C RMS and dewpoint within 0.3 °C RMS. Heights came within 25 m RMS, or 18 m when worked out up from the surface. Raw GPS altitude was off by 114 m. Wind speed came within 0.24 m/s on average. The 3-second wind mean was the best of the windows tried, from 0 to 10 s. It gains little over raw winds. Fall speeds ran 8 to 18 m/s. None were flagged.
• Relative humidity, potential temperature (θ), equivalent potential temperature (θe), mixing ratio (w) and the lifting condensation level (LCL) are worked out by this site. They come from the reported temperature, dewpoint and pressure. They aren't part of the data the aircraft sends.
• WL150 is the mean wind over the lowest 150 m of the sounding. It's taken from the operator's TEMP DROP message when there is one. For every other sounding, it's worked out by this site the same way. That includes preliminary ones. It starts at the lowest wind report, which must be within 50 m of the surface. The speed is the height-weighted mean up to 150 m above that. The direction is that of the mean wind over the layer. It's only worked out when winds cover at least 100 m of the layer. For 94 sondes with both values, the site's speed came within 1 to 2 kt of the operator's on average. MBL and DLM only come from the message.
• The Skew-T background is also worked out by this site. That's the dry and moist adiabats and the saturation mixing ratio lines. The moist adiabats are pseudoadiabats, integrated numerically.

Data use and attribution

Dropsonde data collected by NOAA aircraft belongs to the U.S. Department of Commerce and is openly available. tropicalrecon.com is not affiliated with or endorsed by NOAA. The visualizations on this site are for information only. To use dropsonde data in research or publications, get the quality-controlled dataset directly from NOAA.

Method references

• QC NCAR Earth Observing Laboratory: ASPEN documentation, quality control algorithms and formulas.
• Instrument Hock, T. F., and J. L. Franklin, 1999: The NCAR GPS dropwindsonde. Bull. Amer. Meteor. Soc., 80, 407–420.
• θe, LCL, saturation Bolton, D., 1980: The computation of equivalent potential temperature. Mon. Wea. Rev., 108, 1046–1053.

The Aircraft Data page shows hurricane hunter flight tracks, live and archived. It shows the flight-level wind, surface wind estimates and other measurements along each track. These appear on a map and in time plots. It draws on three sources.

• HDOBs (high-density observations). Every reconnaissance aircraft sends these bulletins during a flight. That's the U.S. Air Force Reserve WC-130J and NOAA's WP-3D and G-IV. Each has an observation every 30 seconds. An observation gives position, pressure, geopotential height, extrapolated surface pressure or D-value, temperature, dewpoint and flight-level wind. It also gives the peak 10-second wind, the SFMR surface wind estimate, rain rate and quality-control flags. They're retrieved from the NWS data feed as they're sent. Gaps and past flights come from the NHC recon archive. This site stores, decodes and serves them.
• IWG1 (NOAA aircraft, live). The 1-second data stream NOAA's aircraft send during a flight. It gives position and altitudes (geopotential, pressure and radar). It gives flight-level pressure, extrapolated surface pressure and D-value. It gives temperature, dewpoint, relative humidity and equivalent potential temperature. It gives horizontal and vertical wind and radiometric surface temperatures. It also gives air speed, ground speed, vertical speed, pitch and roll. It's retrieved from the OMAO real-time data server while the aircraft flies. It usually arrives 15 to 60 seconds after it's sent. The aircraft's own data system works out values like relative humidity and θe. This site does not.
• NOAA flight-level (MET) archive. NOAA's own archive of its aircraft's flight-level data, recorded every second. It comes from the NOAA Aircraft Operations Center. The same IWG1 fields are read from its text files. It's usually posted a day or more after a flight. Until then, a flight followed live keeps its live IWG1 copy. Some older flights only have a thinned-out copy. For those, this site uses its companion service, noaa-recon-api. That gives position, flight-level wind, SFMR and altitude every 5 seconds.

NOAA flights show the 1-second data in place of their HDOBs. A live flight uses IWG1 first, then HDOBs. An archived flight uses the MET archive first, then the live IWG1 copy, then the thinned-out archive, then HDOBs. The flight's HDOBs still add what the 1-second data lacks. That's the SFMR surface wind, the peak 10-second wind and the rain rate. Plots mark these series "HDOB". Air Force WC-130J flights are shown from HDOBs only.

• Provider: U.S. Air Force Reserve 53rd Weather Reconnaissance Squadron and NOAA Office of Marine and Aviation Operations (OMAO), via the National Hurricane Center
• Coverage: live flights as they report. Archived HDOB flights from 2026. NOAA flight-level data from 2011 for the 1-second archive and 2013 for the thinned-out copy.

Interpreting the data

• Flight-level winds are not surface winds. Wind barbs show the wind at the aircraft's altitude. They're colored by the Saffir-Simpson category that speed would be at the surface. The scale runs from below hurricane force through Category 5. The colors are only a quick visual guide. Surface winds are usually well below flight-level winds. The SFMR value is the aircraft's own surface wind estimate.
• SFMR is the Stepped-Frequency Microwave Radiometer. It estimates surface wind speed from the microwaves the sea surface gives off. It can be unreliable in heavy rain, in shallow water and at low wind speeds.
• Extrapolated surface pressure is worked out by the aircraft from flight level down to the surface. It's reported below about 550 hPa. Higher up, the aircraft reports a D-value instead. That's the difference between its actual height and its height in the standard atmosphere.
• 1-second tracks are trimmed to the flight itself. They run from 10 minutes before takeoff to 10 minutes after landing. A ground speed above 60 kt counts as flying. The data files also cover time on the ground. Flight extremes like peak wind and lowest pressure only count airborne data.
• Telemetry plots show time in UTC. Zooming only changes the time axis. Each plot's value axis fits whatever is in view. Wind direction is plotted as points because it wraps from 360° to 0°.
• Radar altitude is height above the surface below the aircraft, sea or land. Pressure altitude is the standard-atmosphere altitude for the measured pressure. The altitude shown elsewhere is geopotential. Vertical wind is the air's own up and down motion. Updrafts are positive. It's not the aircraft's motion.
• The radiometric sea surface temperature and downward-looking infrared temperature measure whatever is below the aircraft. In clear air that's the sea surface. With cloud underneath, it's cloud tops or rain.
• Storms follow each system from invest to named storm. They use NHC's genesis number from its ATCF best-track files. That number stays the same when an invest is named. Flights into Invest 92L list as "Invest 92L" until it's named. After that they list with the storm, for example "Imelda (AL09, formerly Invest 94L)". Some flights' own data names no storm. This is common for G-IV surveillance and early invest flights. Those take the storm the Plan of the Day tasked them for. Flights with no storm are under Training / Research. These include training, ferry and research flights.
• Obviously bad positions are removed before tracks are drawn. An example is 0°N 0°E, sent before the GPS has a fix. Jumps no aircraft could fly are removed too.
• A nonzero QC flag means the aircraft's own checks marked some of that observation's values as suspect.
• HDOBs and IWG1 are real-time data with no post-flight quality control. The flight-level archive is NOAA's recorded data. The map draws 1-second tracks at one point every 5 seconds. Plots and readouts use every second.

Data use and attribution

The U.S. Air Force Reserve and NOAA collect reconnaissance data. The National Hurricane Center sends it out. It's openly available. tropicalrecon.com is not affiliated with or endorsed by NOAA, NHC or the U.S. Air Force. For research or publication, get the data directly from NOAA and NHC.

The UAS Data page shows flights of the Black Swift Technologies S0. It's a small uncrewed aircraft system (UAS). It's launched in flight through a tube on NOAA's WP-3D Orion aircraft, N42RF and N43RF. It flies in the hurricane boundary layer where crewed aircraft can't safely go. Flights often last one to two hours. It reports pressure, temperature, humidity and wind. NOAA runs the program with Black Swift Technologies. 2026 is the first season its data goes into NOAA's hurricane models.

• Provider: Black Swift Technologies, with NOAA's Atlantic Oceanographic and Meteorological Laboratory (AOML) and Office of Marine and Aviation Operations (OMAO)
• Retrieved from: the National Weather Service's recon bulletin feed. Observations arrive minutes after they're sent. Also from Black Swift Technologies' data server, which has every flight since 2023. Data there is usually posted later. A summary follows once each flight is over. This site indexes, decodes and serves the data.
• Coverage: flights from October 2023 onward, starting with Hurricane Tammy. Training and test flights are included.

Data shown on this site

• Observations. Sent in the same format as aircraft HDOBs. There's an observation every 30 seconds. Each gives position, pressure, geopotential height, extrapolated surface pressure, temperature, dewpoint and wind. It also gives the peak 10-second wind and quality-control flags. They come in bulletins about every 10 minutes. So a UAS in flight moves on the map in 10-minute steps. The site checks for new bulletins every minute. While the UAS page is open during a flight, it checks every 15 seconds. During a flight the site shows the National Weather Service's copies, as sent live. Black Swift later posts a reprocessed copy of each bulletin. That copy replaces the live one. So archived flights can differ slightly from what was shown live.
• Flight summary. For each flight, it gives the vehicle's serial number, such as S00070. It gives the P-3 that launched it and that P-3's flight. It also gives the storm, the flight pattern and the flight time. The bulletins name the ground station that received them, such as UA212. They don't name the vehicle. So this site matches each flight to its summary by launch time and position. The summary is usually posted after the recon mission. Until then, the UAS flight is shown under its ground station ID. Its launching aircraft is not shown.

Interpreting the data

• The S0 flies low, so its winds are close to the surface. Unlike crewed aircraft at flight level, it often measures within a few hundred meters of the sea. Wind barbs are colored by the Saffir-Simpson category the speed would be at the surface. That's only a quick visual guide. It isn't an official surface wind.
• The aircraft's motion can affect winds and pressure altitude. This happens most in strong updrafts and downdrafts, right after launch and when it's losing altitude near the end of a flight. Some eyewall flights report very high winds. These are shown as sent. Treat them with caution.
• Cleaning by this site. Observations with no position are dropped. Dewpoints below −60 °C or above the temperature are removed as sensor errors. Positions that would need impossible speeds are dropped from tracks. When the link drops, the ground station repeats the last fix. Long runs of these repeats are left out of tracks. Observation times are tied to each bulletin's issue time. This is because the bulletins keep the launch date after 00 UTC. Nothing else is adjusted.
• Quality-control flags are the aircraft's own. A nonzero flag means some values in that observation were marked as suspect.
• This is experimental, real-time data with no post-flight quality control.

Data use and attribution

NOAA and Black Swift Technologies collect S0 data. tropicalrecon.com is not affiliated with or endorsed by NOAA or Black Swift Technologies. The visualizations on this site are for information only. For research or publication, get the data directly from NOAA or Black Swift Technologies.

The ROARS page shows ocean surface winds from ROARS, the Rain, Ocean and Atmosphere Radar System. It's an experimental X- and Ku-band radar on NOAA's WP-3D Orion hurricane hunters, N42RF and N43RF. Its beam can be steered from straight down to 60 degrees off vertical, all the way around the aircraft. It's being tested during NOAA's 2026 Hurricane Field Program (APHEX).

It estimates the wind at the ocean surface below and around the aircraft. It uses how strongly the sea reflects the radar. That depends on how rough the wind has made the sea. It also measures the wind in the air below the aircraft. It uses the Doppler shift of what the radar sees there. The page gathers every file posted for a mission into one view. It groups them by mission ID, such as 20260929H1.

• Provider: NOAA's National Environmental Satellite, Data, and Information Service (NESDIS) Center for Satellite Applications and Research (STAR), with NOAA's Atlantic Oceanographic and Meteorological Laboratory (AOML) Hurricane Research Division and Office of Marine and Aviation Operations (OMAO)
• Retrieved from: NESDIS/STAR's ROARS retrieval files and the data in their interactive wind curtain pages. The files are NetCDF. Some are split into one file per flight leg. They're checked every 10 minutes. This site decodes, stores and serves them. It redraws the curtains from their data. Each mission's page links to its original files. Images NESDIS/STAR posts without data aren't shown.
• Coverage: whatever has been posted. As of September 2026, that's a surface wind retrieval for one flight, winds at a few heights for another, and wind curtains for a third. The format may change as the instrument develops. The page shows any new values a file adds, labeled the way the file describes them.

Data shown on this site

• Surface wind speed and direction. Retrieved about every 1.3 seconds along the flight. Eight beams scan a cone 30 degrees off vertical. Their backscatter is compared against a geophysical model function. That is a fitted link between wind and backscatter. On the map, the aircraft's track is colored by the Saffir-Simpson category of the surface wind. The color bins have hard cutoffs.
• Rain attenuation (PIA, path-integrated attenuation). How much rain between the aircraft and the sea weakened the signal. It's estimated from the straight-down beam, in decibels one-way and two-way. The retrieval corrects for it. It also classes each point as clear, light rain or rain. Rain is shaded grey on the plots.
• Beyond the fitted range. The model function was fitted to winds of 5 to 70 m/s, about 10 to 136 kt. But the retrieval searches up to 100 m/s. Faster retrievals are flagged. The plots show the 70 m/s limit as a dashed line.
• Wind curtains (Ku-band). Shown first where they're posted. Each leg's wind speed from the sea surface up to about 4 km, scan by scan, about every 1.2 seconds. The solid color bins are below 34 kt, 34 kt, 50 kt, then the Saffir-Simpson categories. The black line is the aircraft's altitude. Hovering a scan shows its speed and direction against height. The aircraft's own flight-level wind is marked. "30-s mean" is each height's wind averaged over 30 seconds, as the file gives it. "Each scan" is the unaveraged retrieval at finer heights. The map and plots read the curtains at a few fixed heights.
• Winds at heights (Ku-band). Where posted, the 30-second mean wind speed and direction at a few heights, from 25 m to 1 km, along each leg. The legs are joined into one timeline. The map colors the track by the wind at the height picked beside the replay controls. The default is 300 m.
• The file's description. Each flight's page lists the file's own settings, such as its method, thresholds and averaging windows. It also lists the file's description of each value.

Interpreting the data

• This is experimental research data. It hasn't been quality-controlled. The National Hurricane Center doesn't use it operationally. It isn't an official wind measurement. Compare it with the SFMR and flight-level winds on the Aircraft Data page. That page links to and from the same flight.
• Winds beyond the fitted range are extrapolated. Treat flagged retrievals, and anything above the dashed line, as very uncertain.
• Rain reduces confidence. Heavy rain weakens the signal. The correction for it is rough. Treat shaded stretches with more caution.
• Heights below about 300 m are less certain. Near the surface, the sea's own echo spoils the Doppler wind. The files say so. The page starts at 300 m.
• Winds aloft aren't surface winds. The curtains and height series show the wind at that height. That is usually stronger than at the surface.
• Directions can flip by 180 degrees. Upwind and downwind look alike to the radar. So single points can point the opposite way.
• Nothing is adjusted by this site. Values are shown as the file has them. They're converted to knots where the site uses knots. Points without a time or position are left out.

Data use and attribution

NOAA collects and processes ROARS data. tropicalrecon.com is not affiliated with or endorsed by NOAA. The visualizations on this site are for information only. For research or publication, get the data directly from NOAA.

Data by Storm is under the Data menu. It lists every recon flight into each storm of a season. It goes back to 2016. It has no data of its own. Each part comes from the source in its own section of this page. The notes there apply here too.

How flights are matched

• Flights. These are the Aircraft Data page's flights. NOAA flights come from NOAA's flight archive or the live feed. Air Force flights come from their HDOBs. Training, ferry and research flights with no storm are left out.
• Storms. A flight goes with the storm in its own data. If that's missing, the Plan of the Day's tasking or the storm's name decides. Flights into an invest that later became a storm are listed with the storm.
• Active storms. A storm is active while NHC is still writing advisories on it. A flight is flying now while it is still sending data.
• Data links. A link is greyed out when that page has nothing for the flight. Dropsondes are matched by the flight's AOC folder. Air Force folders are matched by aircraft, mission code and takeoff date. Tail Doppler Radar, UAS and ROARS are matched by the NOAA flight ID.

The map

• Best track. NHC's best track for the storm, from its ATCF b-deck. Dots mark each fix. They are hollow below 34 kt, filled for tropical storms and larger for hurricanes. Each 00Z fix is labeled with its day. While a storm is active, the newest fixes can lag by a few hours.
• Flight tracks. Each flight's HDOB positions, one point every 2 minutes. A flight still flying ends in the aircraft arrow.

Data use and attribution

NOAA, the U.S. Air Force Reserve and NHC collect and process the data. Each source's section gives the details. tropicalrecon.com is not affiliated with or endorsed by NOAA, the U.S. Air Force or NHC.

Mission Review is under the Tools menu. It replays one flight with everything it collected, in step. That's the aircraft's track and flight-level winds, its dropsondes, the Black Swift S0 UAS it launched, its Tail Doppler Radar analyses and its ROARS winds. It has no data of its own. Each part comes from the source in its own section of this page. The notes there apply here too.

How the parts are put together

• Which data goes with a flight. Everything is tied to the flight by its NOAA flight ID, such as 20260924H1. A flight rebuilt from HDOBs gets its ID from the same aircraft's dropsondes, live feed or ROARS files during the flight. Air Force flights have no dropsondes, radar, UAS or ROARS on this site. So they replay the aircraft's track only.
• Aircraft. The 1-second track for NOAA aircraft when there is one. Otherwise it's the HDOBs, every 30 seconds. The map draws a point every 5 seconds. SFMR and extrapolated surface pressure come from the flight's HDOBs, within a minute of the time shown.
• Dropsondes. Each sonde appears at its launch time. It follows its reported positions down to the surface, drawn every 5 seconds. Coded soundings come from TEMP DROP only. They have no positions aloft. They're drawn straight from launch to splash, timed by a typical fall speed. Preliminary sondes have dashed outlines.
• UAS. Each flight as its bulletins report it, every 30 seconds. It runs from its first to its last observation.
• ROARS. The surface wind retrievals along the track. They're colored by Saffir-Simpson category at and above hurricane force. Some flights have ROARS winds at heights or wind curtains instead. Then the track is colored by the 30-second mean wind at the height picked under the layer. The default is 300 m, the same as on the ROARS page. The timeline shows that wind and each ROARS leg. The readout gives the wind at that height and the strongest of the heights. It links to the ROARS page at that moment. Winds aloft are usually stronger than at the surface.

Tail Doppler Radar in the replay

The radar part of the replay is a reconstruction. The analyses are real. The order their cells appear in is estimated.

• Legs. Each analysis comes from one radar leg. Its on and off times come from the name of the leg's analysis bundle. A leg with no bundle is assumed to run from the previous analysis to this one. The timeline marks those legs with a dashed outline.
• Scan lines. The P-3's tail radar looks about 20 degrees ahead of and 20 degrees behind a line straight out from the fuselage. It does this on both sides. While a leg runs, the map draws those four beams out to 60 km, or about 32 nm. The real range varies with rain and altitude.
• Revealing the analysis. As the aircraft flies a leg, each cell appears faintly once the forward beam has passed over it. It appears fully once the aft beam has. Winds can only be worked out where both beams have looked from two angles. The times come from the aircraft's track and the beam angles. They don't come from the radar data. When the leg ends, the whole analysis is shown, including cells the estimate missed.
• The values are the finished analysis. A cell that appears mid-leg shows its value in the analysis made after the leg ended. It's not a reading at that moment. The analysis is centered on the storm at its analysis time. So it may be a few kilometers from where the storm was when a cell was first scanned.
• Earlier legs. The latest finished analysis stays on the map until the next one ends. While a new leg runs, the earlier analysis is shown faintly underneath.
• Storm centers. Each analysis's center appears when its leg ends. The latest is drawn solid and earlier ones faint.

Data use and attribution

NOAA and Black Swift Technologies collect and process the data. Each source's section gives the details. tropicalrecon.com is not affiliated with or endorsed by NOAA or Black Swift Technologies.

The Mission Planner is under the Tools menu. It lays a standard recon or research flight pattern over a storm's forecast position. It lists the pattern's waypoints. It's a planning aid. It is not an official product. Check the waypoints against the tasking, the Plan of the Day and the aircraft's own navigation before they're flown.

Where the center comes from

• NHC forecast. The official forecast (OFCL) from NHC's public ATCF a-deck files. It covers forecast hours 0, 3, 12, 24, 36, 48, 60, 72, 96 and 120. The a-deck is checked every 2 minutes and fetched again when it changes. The 12Z forecast comes out with the 15Z advisory. So a forecast appears a few minutes after its advisory. The last 10 forecasts are kept so an older one can be picked.
• Invests have no official forecast. The planner uses the TVCN model consensus track instead and labels it that way. It gives position only. It has no intensity or wind radii. With no consensus either, it uses the latest best-track fix.
• Between forecast hours, the center is placed along a great circle between the forecast positions either side. The intensity, pressure and wind radii aren't blended. They come from the nearest forecast hour. That hour is shown with them. NHC gives the pressure at only some hours. So it may come from a different hour than the winds.
• Storm motion is NHC's forecast heading and speed at the nearest forecast hour. For the consensus track, it's worked out from the track itself.
• Manual center. A position typed in by hand replaces the forecast position. The forecast's intensity and wind radii for that hour still show around it.
• Best track. The past three days of NHC's best track (b-deck) are drawn for context. Its wind radii are drawn too.

How the pattern is worked out

• Every pattern is placed around the center on a sphere. Turn points sit at a set bearing and distance from the center. Legs are great circles. A leg's course is its starting great-circle course. Over a long leg, it can differ by a degree or so from the line it follows out from the center.
• Figure-4. In through the center on the starting inbound course and out to the leg length. Then a downwind leg a quarter of the way round, left or right. Then through the center again at right angles to the first pass.
• Rotated Figure-4. A second Figure-4 turned 45 degrees. The four passes are then 45 degrees apart.
• Butterfly. Three passes 60 degrees apart, joined by 60-degree turns.
• Box. A square with its corners at the leg length from the center.
• Lawnmower. Parallel lines across a square twice the leg length on a side.
• Circumnavigation. A circle at the leg length with a turn point every 45 degrees. A second circle is optional.
• Downwind legs can be straight chords or arcs at a steady distance from the center.
• Times are distance divided by the chosen true airspeed. They ignore wind, turns, climbs and descents. They're only a rough guide to the pattern's length.
• The storm is held still. The whole pattern is placed around the center at one time. A slow pattern on a fast storm won't cross the center where the map shows.
• Vortex Alignment Module (APHEX 2026, P-3 Pattern #1). It's flown after the pattern or instead of it. It follows the 2026 VAM Flight Pattern Description. The low-level center is the planner's center. The mid-level center is that plus the tilt. It's marked "M" on the map.
  • Box (Option A). V1 is 25 nm uptilt of the low-level center. V2 is 25 nm downtilt of the mid-level center. Both sit half the leg spacing to one side of the tilt axis. A turn crosses to V3, 32 nm (60 km) away. The turn is left by default, as in the document's figure. Then it flies back parallel to V4.
  • Triangle. V1, then 25 nm downtilt of the mid-level center on the axis, then V4's position.
  • L. The box's first two legs.
  • Radial (Option B). From 25 nm uptilt of the low-level center, through both centers, to 25 nm downtilt of the mid-level center, and back.
  • The legs run along the tilt unless a leg course is entered. The distance past each center and the leg spacing can be changed. The document allows 40 to 80 km between legs.
  • The document's wording puts V1 on the other side from V2. It uses the words "downwind" and "upwind". But its figure shows a rectangle. The rectangle is what's drawn.
  • The module's own legs are dashed on the map. The leg into V1 and every other leg are solid.
• The tilt for the module comes from the storm's latest Tail Doppler Radar analysis in the last 24 hours. It uses the 2 km and 6 km centers, as on the Vortex Alignment page. A typed-in tilt replaces it. With no analysis, the planner guesses 30 nm along the storm's motion and says so. The document recommends TDR centers first, then a flight-level or recon fix, then IR imagery or the HAFS forecast tilt.
• Satellite (a map layer). A GOES Band 13 infrared image around the center. It's the same as on the Tail Doppler Radar page. It's the latest scan at or before the center's valid time. A forecast center in the future gets the latest scan. It's for orientation. It isn't moved with the storm.
• TDR swath. The band 50 km (27 nm) either side of each leg. That's the usual edge of a Tail Doppler Radar analysis. The real coverage depends on rain and the aircraft's altitude.

Route and point names

• The route is one line of text, like the flight plan strip of an electronic flight bag. It starts as the pattern's own points in order. It follows the pattern as it changes. Points typed in before, between or after them are flown where they sit. Leaving a pattern point out skips it.
• Point names are numbered by pass. IP is the initial point, CP the center crossing and EP the end point. MP is a midpoint. A marks the inbound half of a pass and B the outbound half. The box and circumnavigation don't cross the center. They go IP1, EP1, EP2 and so on. MPn is midway along leg n. The lawnmower names each line IPn to EPn. Midpoints are in the route when midpoint drops are on. Otherwise they can be typed in by name.
• Moving a pattern point. Write a pattern point's name with a position, such as EP2=2530N07512W, to fly it there instead. Any of the forms below works. Write plain EP2 again to put it back. A midpoint beside a moved point moves with it. It stays halfway along its leg, flown straight. That doesn't happen if the midpoint itself was moved.
• Moving a center crossing. A moved CP becomes the center for its own pass only. That pass's IP, EP and midpoints are laid around it. They keep the same bearings and ranges as before. The other passes stay on the storm's center.
• Changing the pattern asks first when the route has points of your own or moved points. So does changing the module's shape. It deletes them if you go ahead.
• Points of your own can be a bearing and range from the center (045/80, or CTR045080). They can be an offset north or south and east or west of the center (CTR30N20W). Or they can be a latitude and longitude (2530N07512W, 25.5N75.2W). A name can be given, as in SW1=045/80. A star marks a dropsonde, as in CP1*. Distances are nautical miles.
• Points given relative to the center move with it. A latitude and longitude stays put. An offset is laid off as a bearing and range along a great circle. That matches north and east closely over a few hundred nautical miles.
• The pattern's point names can change without the pattern changing. This can happen with more lines, or with midpoints turned on or off. Then your points stay after the pattern point they followed. Points at the start or end stay there. V1, V2 and so on are the Vortex Alignment Module's points.

TDR Operator Logsheet

• The TDR Operator Logsheet was revised 8/28/26. It's in the TDR folder of Documents, under Tools. Documents needs a password. A new logsheet can have any committed plan's route drawn on it. The plot is storm-relative and north up, with true bearings. Its rings are 20 nm apart. The bold ring is at 105 nm. If the route reaches beyond 140 nm, a larger ring step is used. The step is printed beside the plot.
• Beside the plot is the plan. It gives the pattern, legs, course, airspeed, center and motion. It also lists each point's bearing and range from the center in nautical miles. D marks a dropsonde. Latitudes and longitudes are on the planner's waypoint table.
• Filling it in. The logsheet opens in the site's PDF editor, in a new tab. What's typed is kept in that browser on that device until it's downloaded, shared or saved to the server. Save to server keeps it in Documents. Anyone with the Documents password can reopen it there. It isn't live editing. A copy open elsewhere doesn't update. Text that doesn't fit its box is printed in full on extra "continued" pages at the end.
• Share links. Sharing uploads the filled PDF to this site. It gives a link that anyone with it can open. The file is deleted 48 hours after it's shared.
• Email. The filled PDF can be emailed as an attachment from [email protected]. It's sent through Gmail. The site doesn't keep the file or the addresses.

Committed plans

• A verified crew can commit a plan under a NOAA mission ID, like 20260929H1. That's the date, the aircraft's letter and the flight that day. Anyone can open committed plans. They're kept until deleted.
• On the Aircraft Data page and the home page, a flight with a committed plan shows its newest version where it was planned. It's drawn around the center it was committed with, usually a forecast position. It doesn't move as the storm moves. When the crew update the pattern, they commit a new version. The map shows that one. The planned center can be tens of miles from the real one. So the plan on the map shows where the aircraft means to go. It doesn't show where the center is.

Data use and attribution

NOAA's National Hurricane Center makes the forecasts and best tracks. It publishes them in the ATCF format at ftp.nhc.noaa.gov. tropicalrecon.com is not affiliated with or endorsed by NOAA.

The Vortex Alignment page is under the Research menu. It follows how far the storm's mid-level and high-level vortex centers sit from its low-level center. It also follows their direction. It covers every Tail Doppler Radar analysis of a storm from both P-3s, across all flights, in time order. The G-IV flies its legs far from the center, so its centers aren't reliable enough to use. The page is built for the 2026 Hurricane Field Program's Vortex Alignment Module. That module studies how weak, sheared storms align.

At the top, a summary row gives the latest tilt and its direction, the trend by flight, the best-track intensity and the number of analyses. Below it, three time panels share one time axis. Each flight is shaded and named along the top. They show intensity, tilt and a time-height section. A polar trace sits beside them. The page has no data of its own. The analyses and centers come from the Tail Doppler Radar archive in its own section of this page. The notes there apply here too.

Which analyses

• Every horizontal (xy) analysis of each flight the TDR archive lists for the storm. That includes ones HRD didn't mark acceptable for compositing. The CSV download carries that flag.
• Each flight's Level 2 analyses when it has any, or else its Level 1b ones. That's the default, for research use. Level 2 is reprocessed and quality-controlled. It comes out weeks to months after a flight. The Radar data option can also add Level 1b analyses where a flight's Level 2 has gaps. Level 2 reprocessing re-times some analyses. For example, Level 1b's 0129Z can become Level 2's 2529. That means 0129Z the next day. Level 1b's 2233Z can become 2235Z. So a gap-filling Level 1b analysis is only added when no Level 2 analysis is within 8 minutes of it. Otherwise one analysis would count twice. Level 1b analyses are real-time and not quality-controlled. The subtitle says how many of each are shown.
• Jobfiles with no grid are analyses the archive lists without a horizontal grid. They get a tick so you can see when they were run. They have no centers or tilt.
• Analysis ticks. A tick along the bottom of the tilt panel marks every analysis. It's dark when the analysis had a low-to-mid tilt and light when it didn't. It's dashed for a jobfile with no grid. Hover one for its time.
• Flights. Each flight is shaded from its first analysis to its last, plus 20 minutes either side. It's named by its mission ID. Where space is short, it's named by its day and suffix, such as 21H1.
• Times are the analysis times from the TDR archive, on the flight's date. Times past 2400 are the next day. So are early-morning times on a flight with evening analyses.

Vortex centers

• Method. The centers come from noaa-recon-api's objective center finder. It works at each height, every 0.5 km of the analysis grid. The center is the point within 50 km of the analysis's origin with the strongest average tangential wind around it. That average is taken over radii of 2 to 50 km and must be at least 3 m/s. It comes from the analysis's horizontal wind. The finder starts at an anchor height. It keeps each height's center within 12 km of the one below. That way the column follows the vortex's tilt. Centers stuck against the edge of the search area are dropped. Heights with no clear circulation have no center. Weak or disorganized vortices often have no center at some heights, or at all.
• Heights. 2 km for the low level and 6 km for the mid level by default. These follow Fischer et al. (2024). There's also a high level. It's 10 km by default and can be set to none. Any of them can be changed.
• Jobfile center. Each analysis's jobfile has the center the radar operator gave it. That is the grid's origin. By default, it stands in when the center finder has no low-level center. That way more analyses get a tilt. That's true at the chosen height and at a stand-in height. It's the operator's center. It's not an objective one. Across 2026's analyses, it sat a median 3.6 km from the objective 2 km center. But one time in ten it was more than 25 km away. That is as large as many tilts. Those analyses are drawn hollow with a dot and marked * in the table. In the CSV they have low_source=jobfile. Turn the option off to use objective centers only. Every analysis's jobfile center is also in the readout and the CSV.
• Stand-in heights. By default, when the chosen height has no center, the nearest height within 0.5 km that has one is used. If two are equally near, the lower one is used. Those analyses are drawn with hollow markers. The readout gives the height used. Turn the option off to use the exact heights only.
• Tangential wind and RMW in the readout are the average tangential wind at that center and the radius where it peaks. Both come from the same center finder.

Tilt

• Size. The straight-line distance from the low-level center to the mid-level center, in km on the analysis grid. Both centers are on the same grid, so no map projection is involved. A jobfile center is placed on the grid from its latitude and longitude relative to the grid's origin.
• Tilt panel. One arrow per analysis. It sits at the height of its tilt and points in its direction, north up. The arrows are joined within each flight.
• Flight boxes. Behind the arrows, each flight gets a box and whiskers across its analyses. The box covers the middle half of its tilts, the 25th to 75th percentile. The whiskers reach its smallest and largest tilts. A bar marks the median. It uses every tilt drawn, including those with a stand-in height or the jobfile center. A dashed line joins the flight medians in time order to show how the tilt changes.
• Analysis Summary (the Show option) draws only the boxes. Each has an arrow at its median, pointing in the flight's average tilt direction. The low-to-high series becomes its flight medians. The time-height section shows each height's median offset per flight. The polar trace shows one mark per flight, at the average of its analyses' east and north offsets.
• Trend by flight. The summary compares the last flight's median low-to-mid tilt with the flight before. It says "Aligning" when the median fell and "Tilting more" when it rose. It says "Little change" when it moved less than 5 km or 20%. It's a quick read. It's not a statistical test.
• Tilt now. The latest analysis with a low-to-mid tilt. Its direction is given as a compass point.
• Time-height section. Every height from 0.5 km up where the center finder found a center. Each analysis is a column. Each cell is colored by how far that height's center is from the low-level center, in steps. In km, the steps are under 5, 5–10, 10–20, 20–30, 30–50, and 50 and more. Nautical miles and miles use similar round steps. Blank cells are heights with no center. The dashed lines mark the chosen low, mid and high heights. It uses the same low-level center as the tilt, including the jobfile center when that stood in.
• Low to high. The same, from the low-level center to the high-level one. It's a second series of small grey dots on the tilt panel and a dotted trace on the polar plot. The readout and CSV also give mid to high.
• Direction. The direction from the low-level center toward the mid-level center, in degrees clockwise from north. A tilt toward 90° means the mid-level center is east of the low-level one. The arrows, the polar trace and the readout all show it.
• Polar tilt trace. The mid-level center plotted relative to the low-level center, north up. There's one mark per analysis, joined in time order. Older ones are fainter. When the mid-level center circles the low-level one, the marks walk around the middle. This is called precession. When the vortex aligns, they close in on it. Rings are round distances in the chosen unit.
• Shear-relative views will come once the site has an environmental shear source. Shear will point up in them. Until then everything is relative to north.

Intensity

The best-track maximum sustained wind from noaa-recon-api's storm archive. That's NHC's HURDAT2 once it's published. Before that, it's the operational ATCF best track, updated through the season. Invest and depression stages that were never named may be missing. The readout interpolates between the 6-hourly fixes.

Method references

• Fischer, M. S., R. F. Rogers, P. D. Reasor, and J. P. Dunion, 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. The objective TDR vortex centers and tilt the Vortex Alignment Module uses.
• NOAA AOML, 2026: 2026 Hurricane Field Program Early Stage Science: Vortex Alignment Module. https://www.aoml.noaa.gov/wp-content/uploads/2026/05/2026HFP_EarlyStage_Science_VAM-1.pdf

The Aircraft Performance page is under the Research menu. It looks at how the G-IV (NOAA 49) flew where a hurricane's warm core reached its cruise altitude. A jet engine's thrust falls as the outside air warms compared to the International Standard Atmosphere (ISA) for its altitude. So near its ceiling, the G-IV can't hold its speed and altitude in air several degrees warmer than ISA. The page sets the aircraft's behavior beside the storm's warm core. The warm core comes from dropsondes and satellite soundings. It also shows the vortex tilt from the P-3s' Tail Doppler Radar. It's experimental research. It's not an operating limit or a safety tool.

Flights

• G-IV storm flights of each year from NOAA's flight-level archive. Only storms with an ATCF ID are included. It's the same list as on Aircraft Data. A P-3 flight of the same storm is listed as paired if its time in the air overlaps the G-IV's or comes within 3 hours of it.
• Flight data. NOAA's 1 Hz IWG1 data, averaged to 10 seconds. Only airborne data is used. It's the MET archive once that's posted. Before that, it's the live copy.

Performance

• No engine data is public. There's no N1, EGT or fuel flow. So performance is read from what the aircraft did.
• ISA deviation is the static air temperature minus the ISA temperature at the pressure altitude. ISA cools 6.5 °C per km up to 11 km and stays at −56.5 °C above. Tropical air is far warmer than ISA below about 36,000 ft and colder above. So away from storms, the deviation at the G-IV's cruise levels (FL410 to FL450) is usually −5 to −12 °C.
• Sensors. Temperature (static and total), Mach, true and calibrated airspeed and pressure altitude come from the aircraft's own air data computer. That's what the crew sees and the engines are flown on. They come from the flight's MET netCDF. Static pressure is the ISA pressure at that pressure altitude. Position, attitude, vertical speed and vertical wind come from the 1 Hz IWG1 data. Some flights are not yet in NOAA's MET archive because they're in the air or just landed. These show IWG1's air data until the archive is posted. The page says so.
• Mach comes from true airspeed and static temperature.
• Energy rate is the climb rate plus the speed term (V/g, dV/dt), less the measured vertical wind. It's taken over 60 seconds, in ft/min. In steady flight it's the aircraft's thrust minus drag. Values that stay negative mean it's losing energy. The panel is cut at ±4,000 ft/min.
• Cruise in the summary means within 3,000 ft of the flight's top altitudes. Those are its 95th percentile.
• Performance events are the aircraft losing height from level cruise. It must be at or above 35,000 ft and level for 5 minutes. Level means altitudes within 400 ft of each other. Then the pressure altitude must drop at least 500 ft below that level for at least 30 seconds. Anything shorter is a bump of turbulence. The event runs from leaving the level until the aircraft has been back within 300 ft for a minute, or for 15 minutes. Events less than 2 minutes apart are merged. A drop of more than 5,000 ft is a planned descent and isn't kept. A step down to another flight level without warming is a planned level change and isn't kept either. That means level for its last 3 minutes within 150 ft of a round thousand feet. Mach and airspeed are reported for each event. They don't start one. That changed on 5 Oct 2026. Hurricane Sally's 14 Sep 2020 drop had been missed. It dropped 2,300 ft into air 11 to 15 °C warmer than ISA. But it kept its speed.
• Kinds of event.
  • Warm air needs four things. First, the ISA deviation rose at least 3 °C above its median of the 5 minutes before. That uses the warmest 30 seconds of the event. About 2 °C is taken off per 1,000 ft descended above FL361. This is because ISA stops cooling there but tropical air doesn't. Second, it reached air at least as warm as ISA. Colder air leaves the engines thrust to spare. Third, the aircraft lost at least 300 ft. Fourth, the warm air came first. At least 1.5 °C of that warming had been met by the time the aircraft was 500 ft down. A descent finds its warmer air only on the way down. So it counts as other, marked as a descent. It also must be within 300 nm of the storm's center. Further out it counts as other, marked far from the core.
  • In a turn means the aircraft was banked more than 15° for most of it.
  • Other is everything else. It's often turbulence or a downdraft. Check the vertical wind.
  • The thresholds were tightened on 5 Oct 2026. Before that, events like Hurricane Karl's on 24 Sep 2016 counted as warm air. That event was 2.7 °C warmer but still 6 °C below ISA. It lost no height. They were tightened again on 8 Oct 2026. Before that, descents counted. One was a 4,000 ft descent on 31 Aug 2026 far from any storm. So did an event 900 nm from the center. They'll change as more cases are studied.
• Potential warm air is the same warm air met at cruise with no performance event. That's at least 3 °C warmer than the 5 minutes of level cruise before. The air is at least as warm as ISA. The G-IV held its level or only descended gradually. It had no sudden drop of 500 ft or more. These aren't performance events. They're listed apart. They're in the overview's Show menu, within 300 nm of a center. They show how often such air is met without trouble. Any forecast threshold has to allow for that. ISA readings of +20 °C or more are marked as likely an iced total-temperature probe. They likely don't reflect the air.
• Event summary. Each event is compared with the 5 minutes before it. It gives the reference and lowest Mach. It gives the peak ISA deviation at any height during the event. It gives the warming, corrected for the descent. It gives the altitude lost and gained. It gives the course deviation. That's the largest change of ground track from the average before. Ground track comes from successive positions. It also gives the maximum bank and the true and calibrated airspeed lost and gained. Calibrated airspeed comes from Mach and static pressure. Finally it gives the vertical speed and vertical wind ranges and the lowest energy rate.
• Overview. Every G-IV storm flight since 2016 is worked out once and kept. New flights and their satellite passes are worked out while the page is open. A progress bar shows until it's done. The overview shows the event of any kind that lost the most altitude. It counts warm-air events within 300 nm of a storm's center. The rest show under Show all events. It lists them and plots each one's starting altitude against its peak ISA deviation. It also places each event by its bearing relative to that flight's shear. Downshear is within 90° of the way the shear points. Left and right are looking downshear.
• Storm center. The best track from noaa-recon-api's storm archive, interpolated in time. Range and bearing are from that center. So they carry the best track's error, often 10 to 20 km.

Data available

Under the summary, a strip shows what the flight has. That's flight data, dropsondes (with a count), shear (or no data), model times, ATMS passes and P-3 radar analyses. Parts of the page that need something the flight doesn't have are hidden. They're not shown empty.

Warm core

• Dropsondes. The flight's sondes at standard pressures. The anomaly is against the average of the flight's outermost third of sondes, at least 100 nm out. The radius used is in the data. It can be set against the satellite's ring average instead. That's described below.
• Model replay. The warm core as a model had it during the flight. It's the backup source. It's shown first only when no ATMS pass saw the storm. A model smooths a sheared, tilted warm core and puts it where its own storm is. The satellite sees it as it was. The model is there every 3 hours. Satellite passes come when they come.
  • Every 3 hours from 3 hours before takeoff to 3 hours after landing, it uses the newest run that had started by then. It uses that run's shortest lead. That's 0 or 3 hours, as near an analysis as these files get.
  • It's HAFS-A when NOAA ran it for the storm. HAFS-A is NOAA's hurricane model, with a 0.02° storm domain, from mid-2023. Otherwise it's GFS at 0.25°, from 2021. Both come from NOAA's open data on AWS.
  • Each is placed on the same storm-relative grid as the ATMS passes. It's measured the same way, as the anomaly against the 550 to 800 km ring. It's centered on the model's own storm, as for planned flights below. ATMS passes are offered beside it. Flights before 2021 have no model replay.
• ATMS. NOAA's MiRS temperature retrievals from the ATMS sounders on NOAA-20, NOAA-21 and S-NPP. They come from NOAA's open data on AWS. It covers every pass over the storm from 3 hours before takeoff to 3 hours after landing. Footprints that MiRS flags as bad are left out. Each level's anomaly is the temperature minus its average in a ring 550 to 800 km from the center. That's the usual AMSU/ATMS warm-core measure. It's gridded every 25 km out to 600 km with a Gaussian weight (22 km). ATMS is the preferred source. The page opens on the pass nearest the worst event. Footprints are large at the swath's edge. So a pass with the center there counts as 2 hours further off. The model only opens first when there's no pass.
• ATMS before 2026. NOAA's open archive keeps MiRS retrievals for the current year only. So flights from November 2022 through 2025 use the ATMS brightness temperatures instead. These are the calibrated SDR, in the same buckets.
  • It uses channel 7 (54.94 GHz), which senses a layer centered near 250 hPa, about FL340. Channel 8 (55.50 GHz) is near 175 hPa, about FL410. Channel 9 (57.29 GHz) is near 100 hPa.
  • Each footprint's anomaly is against the ring's footprints at the same scan position. That removes the cooling toward the swath's edges, called limb darkening.
  • A channel's anomaly is a layer average. So it's smaller than the temperature anomaly at the level, typically 1 to 3 K in a strong hurricane. It shows where the warm core sits and how it leans. It's less good at showing how warm it is.
  • Channels 5 and 6 (700 to 400 hPa) are left out. Rain and ice in the eyewall scatter them colder. There are no isotherms. Sondes can't be compared with the ring. Flights before November 2022 have no ATMS.
• Archive. NOAA's open buckets keep MiRS soundings for the current year only. As of October 2026, that's 2026. So earlier flights have no satellite warm core. The page hides those views and marks ATMS as not available.
• Limits. ATMS footprints are 15 to 75 km across. So the retrieved warm core is smoothed and weaker than the real one. Below about 650 hPa, rain spoils the retrieval. The warm core's axis is tracked from 600 to 150 hPa only. It starts at the lowest level's warmest point within 150 km of the center. Then it finds each level's warmest point within 80 km of the one below. That's refined to the anomaly-weighted center around it.
• Shear. The 850 to 200 hPa wind difference from the flight's dropsondes launched 200 to 800 km from the center. Each quadrant, relative to north, is averaged first. Then the quadrants are averaged. That way the storm's own circulation mostly cancels out. It needs sondes in at least 3 quadrants. Without them the page says No shear data. That means not enough sondes far out. That's never the same as no shear. A measured shear under 10 kt is labeled weak. The arrow points the way the shear vector points. That's toward where the upper winds push the top of the vortex. The summary gives the direction it comes from. That's the usual weather convention.
• Flight levels. The ATMS levels nearest the G-IV's usual flight levels in the standard atmosphere. Those are FL300 (300 hPa), FL340 (250), FL390 (200), FL410 (175) and FL450 (150). Each map shows the anomaly and isotherms of the retrieved temperature, every 1 °C or 2 °F. It also shows the G-IV's track within 6% in pressure of the level, and the sondes there. All levels overlays each level's half-max outline. That's where the anomaly is half the level's warmest.
• Stretch. For each level, it looks at the region at least half as warm as the level's core and within 200 km of it. Each point is weighted by how much warmer than that it is. The ratio is its spread along its long axis to its short axis. So 1 : 1 is round. The long axis is given as a bearing. The angle between the long axis and the shear is also given. 0° is along the shear and 90° is across. Shear stretches a warm core along the shear and leans it downshear with height.
• P-3 radar tilt. From the Vortex Alignment page's data. See its section above. It's the median 2 km and 6 km centers of the P-3 analyses from 3 hours before the G-IV's takeoff to 3 hours after its landing. They're placed relative to the same best-track center.

Planned flights

• Plans. The G-IV's flight plans committed in the Mission Planner, using the newest version of each mission ID. They're listed from when they're committed until 12 hours after the route ends. A plan is stored relative to the storm's center. So the route is flown relative to the storm wherever it goes. Plans have no times. So the route is timed at the plan's airspeed. That's its TAS line, or 450 kt. The plan's valid time is put at its middle. That's the forecast hour it was laid around. Planned drops are placed where the route passes them.
• Forecast. HAFS-A when NOAA ran it for the storm. HAFS-A is NOAA's hurricane model. Its 0.02° storm domain comes from NOAA's open data on AWS. Otherwise it's GFS at 0.25°, cut to the storm. GFS comes from NOAA's NOMADS server or from NOAA's open data on AWS.
  • The newest run that covers the whole route is used. A newer one replaces it when it's posted. The site checks every 15 minutes, whether or not anyone has the page open.
  • Each planned flight has a tag. Current means it's on the newest run. Updating means a newer run is out and being read. Out of date means it hasn't been checked lately. Hover for the run and the times.
  • Every 3 hours through the flight, the model's temperature at 13 pressure levels goes on the same storm-relative grid as the ATMS passes. That's every 15 km out to 600 km. It's the anomaly against the 550 to 800 km ring.
  • It's centered on the model's own storm. That's the model's lowest sea-level pressure within 250 km of NHC's forecast position. If that low isn't at least 2 hPa below its surroundings, as with a weak invest, NHC's position is used.
  • GFS has no 125, 175 or 225 hPa level. Those are interpolated in log pressure.
  • The shear is the model's 850 to 200 hPa wind difference, averaged around a ring 200 to 800 km out.
• Along the route. Every 30 seconds at FL350 to FL450, the temperature is interpolated between the frames before and after in time. It's also interpolated between pressure levels, in log pressure. Then it's compared with ISA and with the level's ring average. Forecast warm air follows the rule past events met. That's air at least as warm as ISA that's at least 3 °C warmer than the level's ring. The ring stands for the air the aircraft would have cruised in before reaching it.
• Compared with past events. The warmest forecast ISA deviation on the route at each flight level is plotted beside every past event. Each past event shows its warmest ISA deviation and starting altitude.
• Satellite passes. ATMS passes (MiRS) over the storm from the last 12 hours are offered beside the model's frames. Older passes say little about the storm by the time it's flown. Where the model run reaches back to a pass, the same run is shown at the nearest 3-hour step. That way the model's warm core can be checked against the satellite's at cruise level. MiRS footprints are tens of kilometers wide. So the satellite's warm core reads weaker than a model's even when both are right.
• Limits. A forecast's warm core is only as good as the model's storm. Invests and weak storms are poorly forecast. The timing is an assumption. The warm-air rule has no aircraft performance data behind it. No engine data is public. The comparison with past events is a guide. It's not a limit.

Hazardous Winds is under the Research menu. It checks how one recon flight's flight-level wind reduces to the surface. It uses the flight's dropsondes and Tail Doppler Radar. It shows the profile from flight level down and the pattern by ring and quadrant around the storm. It also fits reduction formulas to the flight and compares them with the standard one. It's built for the 2026 Hurricane Field Program's Distribution of Hazardous Winds experiment (NHC). It has no data of its own. The notes in the Aircraft, Dropsondes and TDR sections of this page apply here too. This is a research aid. It's a formula fit to one flight's sondes. It's not an operational reduction.

Which flights

• Flights with dropsondes that are tied to a storm by an ATCF storm number. That way everything can be placed relative to its center. In practice these are the NOAA WP-3D (P-3) flights. Training and test flights aren't listed.
• All flights of a storm can be pooled into one fit. Each sonde keeps its own flight's center, RMW and radar.

Each sonde

• Surface wind. The average of every wind report in the lowest 20 m of the sounding. A sonde whose lowest wind is above 20 m isn't used.
• Flight-level wind and pressure. The aircraft's 30-second HDOB nearest the sonde's launch, within 30 s. This is the wind the reduction starts from.
• Reduction factor. Surface wind ÷ flight-level wind.
• Place. Where the sonde splashed. That's its lowest report with a position. It's placed in the storm-relative frame below.
• Profile. The sounding's winds averaged into height bins. They're 10 m up to 200 m, 25 m up to 1 km and 50 m above that.
• Radar column. The Tail Doppler Radar wind and reflectivity at the sonde's levels, every 100 m. They come from the nearest analysis in time, within 30 minutes. These are the TDR vs. In-Situ page's samples. See that section. A sonde counts as in convection when the radar's strongest reflectivity in its lowest 2 km reaches the chosen threshold. That's 35 dBZ by default.
• By default, a sonde isn't used when its flight-level wind is under 34 kt. It also isn't used when it splashed in the eye, inside half the RMW. Both can be changed.
• Some preliminary messages duplicate a QC'd sonde launched within 30 seconds of them. These are dropped.

The storm-relative frame

• Center. NHC's best track, shifted onto the aircraft's own center fixes. It's the same method and code as the Wind Radii Check.
• RMW. One per flight. It's the median RMW of the legs the sondes were dropped on, from 3 to 150 km. Without those, it's the radar's RMW at 0.5 km. That's the radius of its strongest average wind around the center. Failing that, it's the best track's.
• Rings can be set three ways. By the RMW, they're the eyewall (0.5 to 1.5), the inner core (1.5 to 3) and the outer vortex (beyond 3 × the RMW). By NHC's best-track 64, 50 and 34 kt radii in the sonde's quadrant, interpolated to the flight's middle. Or by distance, at 30, 60 and 120 km.
• Quadrants can be by compass. These are NE, SE, SW and NW, as NHC gives its radii. Or they can be relative to the storm's motion. These are front right, rear right, rear left and front left.

The radar's estimate

• For every TDR analysis of the flight, two winds are read from its horizontal slices. One is at 0.5 km and one is at the flight's height. The flight's height is the sondes' median launch height, to the nearest 0.5 km. They're summed into cells 4 km by 22.5°. The cells sit around that analysis's own vortex center at 0.5 km, or else the grid's origin. They reach out to 200 km. Cells where the flight-level wind is under 10 m/s aren't counted.
• The radar's 0.5 km ÷ flight-level ratio is then multiplied by the sondes' surface ÷ 500 m ratio in that ring. That uses the median of the sondes in the ring when there are 2 or more. Otherwise it uses all of them. The result is the radar's estimate of the same reduction factor where no sonde fell. The radar can't see the lowest 0.5 km. So that last step always rests on the sondes.

The formulas

• Standard reduction (Franklin et al. 2003). It's 0.90 of the 700 mb wind, 0.80 at 850 mb and 0.75 at 925 mb. The flight's median pressure picks which one.
• Franklin ranges. The middle of Franklin et al.'s 700 mb ranges. That's 0.95 in the eyewall (0.9 to 1.0). In the outer vortex, it's 0.825 in convection (0.75 to 0.9) and 0.775 outside it (0.7 to 0.85). Where the radar didn't say, it's 0.80. This is for 700 mb flights only.
• Fit to the flight by least squares. The formulas are one ratio, a line, a quadratic, a power law and a ratio per ring. There's also a ratio that changes with the log of r/RMW. Another adds a wavenumber-1 term around the storm to that.
• Scoring. Each formula gets its RMS error and bias on the sondes. It also gets its leave-one-out error. Each sonde is predicted by the formula fit to all the others. That's the error to expect for a sonde the fit hasn't seen.
• Best guess. The lowest leave-one-out error. Only formulas with at least 3 more sondes than terms count. A simpler formula within 3% of it is preferred. A formula in wind speed alone must stay physical across the sondes' winds. It must rise with the flight-level wind and never go negative. No best guess is named with fewer than 5 sondes.

Limits

• A sonde drifts as it falls. In the eyewall it drifts a few kilometers. So its surface wind and the aircraft's flight-level wind aren't measured in quite the same column. Where the eyewall slopes outward, a sonde launched at the flight-level wind maximum can splash inside the surface maximum. These sondes have low factors and weigh heavily on any fit.
• One flight rarely has more than a few sondes in each ring and quadrant. Treat the table's single-sonde cells as single measurements.

Sources: Franklin, J. L., M. L. Black, and K. Valde, 2003: GPS dropwindsonde wind profiles in hurricanes and their operational implications. Wea. Forecasting, 18, 32–44. NOAA AOML, 2026: 2026 Hurricane Field Program Mature Stage Science: Distribution of Hazardous Winds. https://www.aoml.noaa.gov/wp-content/uploads/2026/06/2026HFP_MatureStage_Science_DistHazWinds.pdf

The Wind Radii Check is under the Research menu. It compares NHC's wind radii with what recon measured around the same time. The radii say how far from the center 34, 50 and 64 kt winds reach in each quadrant. It's built for the 2026 Hurricane Field Program's Distribution of Hazardous Winds experiment. It has no data of its own. The notes in the Aircraft, Dropsondes and ROARS sections of this page apply here too. This is a research aid. It is not a check of NHC's analyses.

Which flights

• NOAA WP-3D (P-3) flights can carry SFMR, ROARS and dropsondes. Air Force WC-130J flights carry SFMR. Both are listed. The G-IV flies too high for a surface-to-flight-level ratio to mean anything. So it isn't listed.
• Only flights tied to a storm by an ATCF storm number are listed. This is because everything is placed relative to the storm's center. Training and test flights aren't listed.

Where each value comes from

• SFMR surface wind and rain rate. From the aircraft's 30-second high-density observations (HDOBs), as on the Aircraft Data page. An SFMR wind of 0 kt is how the HDOBs say "no retrieval". So it's treated as missing. HDOB quality flags that mark the SFMR or the flight-level wind as suspect are kept. Those points are drawn faded. The NOAA P-3s' HDOBs have had no SFMR since 2025. The SFMR fields in the 2026 NOAA flight-level archive files are empty too. So on NOAA flights from 2025 and 2026, only ROARS and the dropsondes are available.
• Flight-level wind and pressure. From the same HDOBs, as 30-second averages.
• ROARS. The retrieval file for the P-3 flight, at about one sample a second. See the ROARS section. Its rain class is kept and shown. The class is clear, light rain or rainy. It comes from the one-way path-integrated attenuation. Its "beyond the fitted range" flag is kept and shown too.
• Dropsondes. Every sonde from the flight. That's the QC'd sounding, or the preliminary one until the QC'd version arrives. There are two surface wind measures.
  • Lowest 20 m (the default) is the average wind speed of every report in the lowest 20 m of the sounding. Sondes whose lowest wind report is above 20 m have no value.
  • WL150 is the average wind in the lowest 150 m. It's from the TEMP DROP message when it gives one. Otherwise it's worked out from the sounding. See the Dropsondes section. No factor is applied to bring WL150 down to 10 m.
  • The sonde's place and time are those of its lowest report with a position. That's close to where it splashed. Coded-only soundings have no positions aloft. So they use the splash from the message and a typical fall time.
  • The flight-level wind for a sonde is the aircraft's HDOB nearest its launch, within 30 s.

The storm-relative frame

• Center. NHC's best track, the ATCF b-deck. It's interpolated along a great circle between its 6-hourly fixes. The latest fix can lag by several hours. When a flight runs past it, the newest NHC forecast's positions are added for an active storm.
• The aircraft's own center fixes. On each pass within 60 km of the best-track center, the HDOB with the lowest extrapolated surface pressure is taken as a center fix. It must fall inside the pass. It can't be at either end. It must also have a flight-level wind under half the pass's peak. The best track is then shifted onto these fixes. The offset is interpolated in time between fixes. It's held steady before the first fix and after the last. Fixes more than 60 km from the best track are ignored. These are not vortex data messages (VDMs). A pass that misses the true center is off by its miss distance. HDOB positions are rounded to about 1.8 km. On the flights checked, sondes labeled CENTER splashed about 1.4 to 2.5 km from the shifted center.
• Radius and azimuth are measured from the center at each observation's own time. Storm motion is the best track's direction over 12 hours, 6 hours either side.
• Radius of maximum wind (RMW). The flight is split into radial legs where its distance from the center turns by more than 4 km. Each leg's RMW is the radius of its peak flight-level wind. That only counts if the peak is within 200 km and not at the leg's outer end. A point on a leg without an RMW takes the nearest leg's RMW within 90 minutes. Otherwise it takes the best track's RMW.

What's compared

• NHC's radii. From the best track, the ATCF b-deck, at the chosen 6-hourly time. Or from the advisory for that time. That's the official forecast at hour 0, from the a-deck. If there isn't one for that time, the latest one before it is used. Radii are given for the NE, SE, SW and NW quadrants. They're in nautical miles.
• Recon around it. Every P-3 and WC-130J flight with data within the chosen window, from ± 3 to 12 hours. It's placed relative to the storm's center at each observation's own time.
• Observed winds. SFMR, ROARS and dropsonde surface winds (lowest 20 m, or WL150). Also flight-level wind reduced to the surface by a single factor, after Franklin et al. (2003). This can be turned off. The factor is 0.8 by default and can be set from 0.75 to 0.9. Flight-level winds are only used between 600 and 850 hPa. That's around the 700 mb level the factors are for. By default, some data is left out. That's HDOB-flagged SFMR, ROARS in its "rainy" class or beyond its fitted range, and preliminary dropsondes.

The observed radius

• In each quadrant and for each threshold, it's the farthest observation at or above the threshold.
• Edge found means the sampling went on at least 5 nm past it, with at least two more observations, all below the threshold. Otherwise the radius is "at least" that far, shown "≥". That means the sampling ended while winds were still above the threshold.
• None means the quadrant was sampled. Nothing reached the threshold.
• The verdict. It agrees when the observed and NHC's radii are within 5 nm. NHC gives radii in 5 to 10 nm steps. Otherwise it's larger or smaller. An "at least" radius can only be called larger. It must already be more than 5 nm past NHC's. A quadrant with no threshold winds only counts against NHC's radius if it was sampled inside it.

Figure 1 ellipse

The experiment's Figure 1 turns four quadrant radii into an ellipse. The radii are placed on the quadrant diagonals. Each opposite pair makes a circle whose diameter joins them. The larger circle's radius is the semi-major axis. The smaller one's is the semi-minor axis. The shifts of the two circles give the ellipse's center, or its skew. It's drawn dashed for NHC's radii. When every quadrant's edge was found, it's also drawn solid for the observed ones.

Radial passes

A pass is the part of one flight leg inside one quadrant. Legs are split where the distance from the center turns. A pass needs at least four observations over at least 10 km of radius. Each is listed with its peak surface wind and how far out it found 34, 50 and 64 kt.

Method reference

• NOAA AOML, 2026: 2026 Hurricane Field Program Mature Stage Science: Distribution of Hazardous Winds. https://www.aoml.noaa.gov/wp-content/uploads/2026/06/2026HFP_MatureStage_Science_DistHazWinds.pdf
• Franklin, J. L., M. L. Black, and K. Valde, 2003: GPS dropwindsonde wind profiles in hurricanes and their operational implications. Wea. Forecasting, 18, 32–44.

Ocean Winds is under the Research menu. It checks ROARS's experimental winds against other instruments. It uses every instrument that measured the same air at nearly the same time and place. Then it scores how well they agree. It's built for the 2026 Hurricane Field Program's Ocean Winds experiment. ROARS's own data and its notes are in the ROARS section. This is a research aid. A high score means ROARS agreed with the other instruments on that flight. It doesn't mean any of them is right.

What's compared

Which checks run depends on what ROARS produced on the flight.

• Surface wind is checked against dropsonde surface winds, ASCAT and SFMR. The dropsonde surface wind is the average of the lowest 20 m. ROARS is averaged over 10-second blocks.
• Wind curtain and wind levels are checked against dropsonde winds at the same height, thinned to 50 m steps. They're also checked against Tail Doppler Radar analyses at the same point and height. Curtains are compared in 100 m steps. Where a curtain's top valid height is within 300 m below the aircraft, it's also checked against the flight-level wind.

The references:

• Dropsondes. Within 10 km and 15 minutes of a ROARS sample by default. The range can be set from 5 to 25 km and 5 to 30 minutes. It uses each observation's own position and time.
• TDR. The analysis nearest in time, sampled at the ROARS sample's position and height. It comes from the noaa-recon-api volumes.
• ASCAT. Metop-C's 4-hourly gridded winds from NOAA CoastWatch ERDDAP. They're in 1/3° cells, for ascending and descending passes.
  • The grids only give a 4-hour bin. So each pass's time is estimated from Metop-C's sun-synchronous orbit. It crosses the equator at 09:30 local solar time going south and 21:30 going north. That's good to about 20 minutes.
  • ASCAT is compared one cell at a time. Each 10-second ROARS block is paired with the one cell whose center is within 10 km of it. That can be set from 5 to 25 km. The cell is first moved with the storm from the pass's time to the block's. The storm's motion comes from NHC's best track. Pairs only count within 3 hours of the pass by default. That can be set from 1 to 6 hours.
  • ROARS winds are never averaged across a cell. Near a hurricane, the wind turns within one 37 km cell. ROARS's surface directions can also flip by 180° from one block to the next. This is the scatterometer's direction ambiguity. Pairs more than 120° apart are counted as flipped.
  • ASCAT is a 10 m neutral-stability wind over a 25 km footprint. So expect some scatter even when both are right.
• SFMR. From the aircraft's HDOBs, within 30 seconds and 5 km of a ROARS sample. Some flights have no SFMR in their HDOBs.

When a reference that could check the product has no pairs, the page says why.

Agreement, 1 to 5

Each pair is scored by its wind speed difference. The allowance is 2 m/s or 10% of the reference's speed, whichever is larger. Within the allowance scores 5 (very high). Within twice it scores 4 (high), three times scores 3 (moderate) and four times scores 2 (low). Anything more scores 1 (very low). The allowance roughly matches the instruments' own stated accuracy.

• Each comparison is one ROARS product against one reference. Its score is the rounded average of its pairs' scores. It also gives the bias with its 95% interval, the RMS difference and the correlation. The bias is ROARS minus the reference.
• Overall is the average of the comparisons' scores. Each is weighted by the square root of its number of pairs. That way one large comparison can't drown out the rest.

Direction agreement

Wind direction is scored the same way, but separately. The allowance is 10°. The difference is ROARS's direction minus the reference's. It's taken the short way round, from −180° to +180°. Within 10° scores 5, within 20° scores 4, within 30° scores 3 and within 40° scores 2. Anything more scores 1. A pair only gets a direction score when both winds are at least 5 m/s, about 10 kt. Direction means little in light winds. Both instruments must also report a direction. SFMR doesn't. Each comparison gives the direction bias with its 95% interval, the RMS difference and the share of pairs within 10°. The overall direction score is weighted like the speed score. The "Agreement in" switch picks speed or direction. That sets how the views are colored and the scatter drawn.

Confidence

• A comparison is high with at least 30 pairs and a bias interval within ±1.5 m/s. For direction the interval must be within ±5°. It's moderate with at least 10 pairs. Otherwise it's low.
• Overall is high with at least 100 pairs from at least two references. It's moderate with at least 20 pairs. Otherwise it's low.

Pairs along one leg aren't independent. Neighboring samples see the same air. So the intervals are narrower than they'd be for independent samples. Read them as a guide.

Method reference

• NOAA AOML, 2026: 2026 Hurricane Field Program Mature Stage Science: Ocean Winds. Linked from https://www.aoml.noaa.gov/2026-hurricane-field-program/
• ASCAT winds: NOAA CoastWatch ERDDAP, datasets erdQCwindproducts4hrAscending and erdQCwindproducts4hrDecending.

TDR vs. In-Situ is under the Research menu. It checks a flight's Tail Doppler Radar wind analyses against what the aircraft and its dropsondes measured in place. It follows the 2026 Hurricane Field Program's TDR Analysis Evaluation module. That module asks how the radar's winds compare with in-situ observations. It also asks how the errors grow away from the flight track and at the lowest levels. The page has no data of its own. The analyses come from the Tail Doppler Radar archive. The flight-level data comes from the aircraft's IWG1 stream. The soundings come from the dropsonde archive. Each has its own section on this page. The notes there apply here too.

Which analyses

• Every horizontal (xy) analysis of the flight. That includes ones HRD didn't mark acceptable for compositing.
• Level 2 when the flight has it. Otherwise Level 1b. They're never mixed. Level 1b analyses are real-time and not quality-controlled. The page says so when it shows them.
• Analysis time is the center time in the analysis's jobfile, to the second.
• The xy product's winds are Earth-relative. So are the aircraft's and the sondes' winds. So they're compared as they are. Taking the storm's motion out of both would leave every difference the same.

Time window

• Radar leg (default). Each analysis is compared with the observations made while the radar collected its data. That's the start and stop times of the analysis bundle that holds it. They're usually 60 to 100 minutes apart. They're not centered on the analysis time. The center crossing is often early in the leg. When the archive lists no bundle for an analysis, a window of 25 minutes either side of the analysis time is used instead. That's marked "est." in the table.
• ±10 to ±60 min. A fixed window either side of the analysis time instead. Wider windows can overlap. Then an observation is counted once for each analysis whose window holds it. That includes the whole-flight statistics.
• Time offsets. Every point carries its time minus the analysis time. It's shown in the readout, in the CSV and on the "Time from analysis" error plot. An analysis is a single picture of the storm built from data collected over the whole leg. So observations far from the analysis time are compared with a storm that has moved and changed in between.

How the analysis is sampled

• Position. Each observation's latitude and longitude are turned into km east and north of the analysis grid's origin. That's the storm center the analysis was built around. It uses the same flat-earth projection the TDR archive uses for its centers. The point is then moved back to the analysis time by the storm motion in the analysis's jobfile. The radar data is placed the same way when the analysis is built. On Gabrielle 20250921H1, this brings the flight-level wind speed RMS error from 1.6 m/s down to 1.2 m/s.
• Height. The aircraft's GPS (geopotential) altitude and the sonde's geopotential height. These are matched to the analysis levels every 0.5 km above sea level.
• Interpolation. Trilinear, between the 8 grid points around the observation. They're 2 km apart across and 0.5 km apart up. A point is left out unless all 8 have data. So there are no comparisons at the edges of the radar's coverage. There are also none below its lowest level with data. That level is usually 0.5 to 1 km. Earlier evaluations went the other way and interpolated the flight-level data to the grid points. Sampling the grid at the observations keeps each observation's own time and position.
• Fields. u, v and w. Reflectivity is added for context. Wind speed is worked out from u and v on both sides.
• Flight level. Every 1 Hz IWG1 record. It's the MET archive's copy once that's posted. Before that, it's the live feed. The page names which.
• Dropsondes. Every sounding with positions, thinned to one level per 100 m of height. Each level uses its own time and position. This is because the sonde drifts. Preliminary soundings are used until their QC'd sounding replaces them. Then the analysis is sampled again.

Components

• Center. The vortex center at the observation's height. It comes from the TDR archive's objective center finder, described under Vortex Alignment. It's interpolated between the two nearest levels with a center when they're no more than 1 km apart. Otherwise it's the nearest level with a center within 1 km. With no center there, the grid origin stands in. The readout then says "grid origin".
• Tangential wind is positive cyclonic. That's counterclockwise north of the equator and clockwise south of it. Radial wind is positive outward. Neither is given within 1 km of the center. The direction is undefined there.
• u and v are the east and north components.

Statistics

• Error is TDR minus in situ. Bias is its average. RMS is the root mean square of the errors. r is the Pearson correlation between the TDR and in-situ values. N is the number of pairs.
• Points aren't independent. 1 Hz flight-level data is far denser than the 2 km grid. The aircraft crosses a grid cell in about 15 s. So N overstates how many independent comparisons there are.
• Flight level is near zero range by design. The aircraft is on its own flight track. There the radar sees the air from close by, and its fore and aft looks are nearly at the same time. So flight-level comparisons show the analysis at its best. The radar's errors should grow with distance from the track. That's because the time between the fore and aft looks grows. So does the beam's volume. The dropsonde comparison against distance from the flight track is there to show this.
• Distance from the flight track is measured on the analysis grid, at the analysis time. It runs from the sonde level to the nearest point of the aircraft's track during that analysis's radar leg, sampled every 5 s. Sondes are dropped from the aircraft and drift. So most are within a few km of the track. Comparisons far from it are rare.
• Vertical velocity is only compared as distributions. It's never compared point by point. That means histograms, the average and the standard deviation. The analysis's mass-continuity constraint works on scales much larger than single updrafts and downdrafts. The distributions use the flight-level points where both have a value. Dropsondes don't measure vertical velocity.
• The binned lines on the error plot are the average and RMS error of the points in each bin. Bins are 2 km, 250 m or 5 min. Each needs at least 3 points.

What to expect

Past evaluations along the track found tangential-wind average errors under 1.5 m/s. RMS errors were 3 to 4 m/s (Reasor et al. 2009, Reasor and Eastin 2012). Correlations were 0.98 to 0.99 for tangential wind and 0.81 to 0.85 for radial wind. Simulated-radar tests found smaller errors (Lorsolo et al. 2013). But those didn't model real beam volumes or the loss of near-surface data.

Method references

• NOAA AOML, 2026: 2026 Hurricane Field Program Mature Stage Science: TDR Analysis Evaluation Module. https://www.aoml.noaa.gov/wp-content/uploads/2026/04/2026HFP_MatureStage_Science_TDR_Evaluation.pdf
• Reasor, P. D., M. D. Eastin, and J. F. Gamache, 2009: Rapidly intensifying Hurricane Guillermo (1997). Part I: Low-wavenumber structure and evolution. Mon. Wea. Rev., 137, 603–631.
• Reasor, P. D., and M. D. Eastin, 2012: Rapidly intensifying Hurricane Guillermo (1997). Part II: Resilience in shear. Mon. Wea. Rev., 140, 425–444.
• Lorsolo, S., J. Gamache, and A. Aksoy, 2013: Evaluation of the Hurricane Research Division Doppler radar analysis software using synthetic data. J. Atmos. Oceanic Technol., 30, 1055–1071.
• Gamache, J. F., F. D. Marks Jr., 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.

The ADECK Viewer is under the Tools menu. It shows NHC's ATCF a-deck files as published. They hold every forecast aid for every cycle of a storm, invest or genesis area. Aids include models, consensus aids and the official forecast. There's one line per aid, forecast hour and wind radii threshold.

Source

• This season. The a-decks in NHC's public aid directory (ftp.nhc.noaa.gov/atcf/aid_public). The directory listing is read again when it's more than 2 minutes old. A file is fetched again when its listing time changes. While a storm is open, the page checks for a newer file every 3 minutes. It reloads it in place and keeps your filters.
• Earlier seasons. NHC's ATCF archive (ftp.nhc.noaa.gov/atcf/archive), back to 1990.
• Names. From ATCF's storm list (atcf/index/storm_list.txt). Invests are numbers 90 to 99. They're listed as "Invest". Genesis areas are 70 to 89. They're listed as "Genesis area". Each shows its ATCF ID.
• Active means NHC updated the file in the last 12 hours.

Official forecast tab

• The official forecast is NHC's OFCL aid in the a-deck. It has one line per forecast hour and wind threshold. The latest is shown in full. Each earlier one is folded to one line below it. Invests and genesis areas have no official forecast.
• Initial and Advisory. Hour 0 is the synoptic time the forecast starts from. Hour 3 is the advisory time. The summary strip describes the storm then. NHC writes the hour 0 lines without a pressure. It gives the pressure on the hour 3 lines.
• Type is ATCF's storm type in words, such as HU Hurricane, TS Tropical storm, PT Post-tropical cyclone and LO Low.
• Category is Saffir-Simpson from the maximum sustained wind alone. Category 1 starts at 64 kt, 2 at 83, 3 at 96, 4 at 113 and 5 at 137. Below 64 kt, it's TS from 34 kt and TD below that. For a system that isn't a tropical cyclone, it's shown greyed. That's a post-tropical or extratropical system or a low. There it describes the winds. It's not a hurricane category.
• Motion is the heading and speed NHC gives on each line. The heading is the direction it's moving toward. Where a line gives none, it's worked out from the forecast track to the next point. It's then shown in italics. The last point uses the one before.
• Wind radii are the 34, 50 and 64 kt radii in the NE, SE, SW and NW quadrants, as NHC gives them. A 0 means that quadrant doesn't reach that wind. A threshold the storm doesn't reach isn't listed.
• Units follow the site settings for wind, pressure and distance. The radii thresholds stay in knots. That's how they're defined.
• The forecast hour valid nearest now is highlighted, as in the table.

What's shown in the table

• The file's own values. Lines and fields are shown as the file has them. They use its units of knots, hPa and nautical miles. They aren't converted to the units in the site settings. Latitude and longitude stay in the file's tenths of a degree with a hemisphere letter. Filters and sorting on them use signed degrees. South and west are negative.
• Column names are ATCF's. Hover a column header for its meaning. Columns empty on every line are left out. Fields past the standard ones are ATCF's user-defined name and data pairs.
• VALID is added by the viewer. It's the cycle time plus the forecast hour. It isn't in the file.
• Order. Newest cycle first. Within each cycle, lines keep the file's own order. This holds until a column is sorted.
• Current cycle is the newest cycle's lines in the file. Early in a cycle, it may only hold the first aids. Those are CARQ and early models. The official forecast arrives with the advisory.
• Valid now. In the current cycle, each aid's lines at the forecast hour valid nearest the present time. An aid gets none when the present is more than 3 hours outside its forecast hours. So an old storm's file has no "valid now" lines.
• Download is in the share button's box. It saves the lines shown in the original a-deck format. It keeps the filters and order shown.

NOAA's National Hurricane Center produces the a-decks. The forecast centers and modeling groups whose aids they carry also help produce them. Model aids are guidance. They are not forecasts. NHC's official forecast is the OFCL aid. tropicalrecon.com is not affiliated with or endorsed by NOAA.

The VDM Viewer is under the Tools menu. It shows vortex data messages (VDMs). These are the recon aircraft's center fixes. Each one gives the center's position and time, the lowest pressure and the eye's shape. It also gives the strongest winds found on the way in and out.

• Provider: U.S. Air Force Reserve 53rd Weather Reconnaissance Squadron and NOAA Office of Marine and Aviation Operations (OMAO), via the National Hurricane Center
• Retrieved from: the NWS data feed as they're sent, under URNT12 and URPN12. Gaps and past seasons come from the NHC recon archive. This site stores and decodes them.
• Coverage: from 2016 on.

Data shown on this site

• Decoded. Each message's fields in plain words, beside the message as sent. Messages from 2018 on use the current A to U layout. Older ones use the earlier A to P layout. Both are decoded into the same fields.
• Raw. Every message as sent, newest fix first.
• Corrections. A corrected message (CCA, CCB) replaces the one it corrects. The original is still shown, struck through.
• Comm checks and training messages aren't stored. That includes messages with no position or a training mission code.

Interpreting the data

• VDMs are real-time messages with no post-flight quality control.
• Minimum pressure may be measured by a dropsonde in the center. Or it may be extrapolated from flight level. The message says which.
• Flight-level winds are not surface winds. The estimated surface winds in a VDM are the crew's estimates.
• Heights, pressures and winds are shown in the units chosen in the site settings. Accuracy figures are shown as sent.

Data use and attribution

Air Force Reserve and NOAA crews write VDMs. The National Hurricane Center sends them out. They're openly available. tropicalrecon.com is not affiliated with or endorsed by NOAA, NHC or the U.S. Air Force.

The Static Images page has ready-made pictures of the newest recon data. Anyone may put them on their own website. Each picture is made from this site's own pages. It's made the same way their share box makes one. It carries the same title, legend, details and credit.

• Made from: the homepage map, the Dropsondes page, the Tail Doppler Radar viewer and the UAS Data page. The sections above describe where their data comes from.
• How often: the site checks for new data every 5 minutes. A picture is made again only when its data has changed. The recon map is made again every 10 minutes while aircraft are reporting.
• Fixed links: each picture's link always gives the newest one. Browsers may keep a copy for up to a minute.
• By storm: most pictures are also made for each storm. A storm's picture shows its newest data, from Data by Storm's list of its flights. They're made while the storm is active and for two days after its newest data. Then they're removed.

Data shown on this site

• Active recon. The homepage map of storms and recon aircraft. Aircraft show if they reported in the last 30 minutes.
• Latest dropsonde. The newest sonde launched, from any aircraft. A raw sounding is replaced by its quality-controlled one when that comes in.
• Latest TDR reflectivity. Reflectivity at 2 km from the newest real-time (Level 1b) analysis.
• TDR winds. Wind speed and wind barbs at 0.5 km and at 3 km from the same analysis.
• TDR reflectivity by height. An animation of the same analysis. It steps every 0.5 km from the lowest level with data to the highest.
• TDR winds by height. The same kind of animation for wind speed with wind barbs.
• Latest storm map. The newest storm with recon flights, from Data by Storm. Its best track and every flight.
• Latest ROARS. The wind curtain from the newest ROARS mission. Its surface winds when it has no curtain. ROARS data is posted after the flight, so the past 6 hours version is often empty.
• Latest UAS flight. The newest Black Swift S0 flight's track and readings.

Interpreting the data

• Two groups. Each picture comes in two versions. "Past 6 hours" shows a picture that says so when nothing is that new. "Latest data" always shows the newest data, however old. Check the times in the picture.
• Like the pages they come from, these pictures show real-time data without post-flight quality control. A picture will say so if it differs.

The Embeds page has live boxes of recon data. Anyone may put them on their own website. Each box is a small page of this site, shown in a frame.

• Recon flights. The homepage's table of tasked flights. It comes from the NHC Plan of the Day and the recon reports, as described under Aircraft above.
• One storm's flights. The recon flights box can show a single storm. Its flights are Data by Storm's, with the flights still scheduled.
• Active storms. Every active storm and invest that NHC lists in the Atlantic and Pacific. It's the same list as the homepage map.
• How often. The boxes check for new data every minute.

The homepage map has a Satellite view. It shows GOES GeoColor imagery under the storms and aircraft.

• GeoColor. True color in daylight. At night it shows clouds over a background of city lights. It was developed by CIRA and NOAA.
• Satellites. GOES-East covers most of the map. GOES-West covers the area west of 106°W.
• Where it comes from. NASA's Global Imagery Browse Services (GIBS) at gibs.earthdata.nasa.gov. This site keeps a copy of each picture for 10 minutes.
• How new. GIBS posts each image about 1 to 2 hours after the scan. The map shows each satellite's time. Images come every 10 minutes.
• Provider: NOAA GOES satellites, through NASA GIBS