Tropical cyclone surface wind field retrieval from SAR
C-band SAR retrieves ocean surface wind speed and direction inside tropical cyclone wind fields where conventional observations are absent, using geophysical model functions to map the asymmetric structure that drives storm surge and inland damage.
Sensors
- Sentinel-1 A/B (C-band SAR, ESA/Copernicus): IW mode at 5 x 20 m resolution, 250 km swath; EW mode at 20 x 40 m, 400 km swath. Repeat cycle nominally 6 days per satellite, 3 days with both; open data with near-real-time dissemination. Most cyclone wind retrievals use EW mode for swath coverage.
- RADARSAT-2 (C-band SAR, MDA): ScanSAR Wide mode covers 500 km at roughly 100 m resolution, well-suited to full storm coverage in a single overpass. Commercial tasking allows priority scheduling over forecast storm positions.
- RADARSAT Constellation Mission (C-band SAR, CSA): Three satellites, each with a 12-day repeat, collectively offering revisit as frequent as 4 days at mid-latitudes and daily access at high latitudes. Sub-daily revisit over a given storm track remains unlikely but improves on predecessors.
- ALOS-2 PALSAR-2 (L-band SAR, JAXA): L-band (1.27 GHz) responds differently to ocean surface roughness than C-band; CMOD functions do not apply directly. L-band retrievals require adapted GMFs and are less validated at extreme winds, but the longer wavelength penetrates rain more effectively, which is relevant inside eyewalls.
What the ocean surface tells a radar
Ocean surface wind speed is not measured directly by SAR. What the sensor records is microwave backscatter from centimetre-scale capillary and gravity-capillary waves, which are generated and maintained by local wind stress. The relationship between that backscatter and the 10-metre neutral wind vector is encoded in geophysical model functions. For C-band, CMOD5.N is the most widely validated: it relates the normalised radar cross-section to wind speed and the angle between wind direction and the radar look direction, given the incidence angle.
Wind direction must be supplied independently or inferred from image features. In practice, retrieval pipelines use one of three sources: numerical weather prediction background fields, the orientation of wind streaks visible in the SAR image itself, or Doppler centroid anomaly analysis. Each carries its own uncertainty. NWP backgrounds are smoother than reality; streak detection fails in areas of heavy precipitation; Doppler analysis requires precise knowledge of satellite velocity and is not always available in standard products.
Why asymmetry matters more than peak speed
Emergency managers often focus on maximum sustained wind speed, but the damage and surge potential of a landfalling cyclone depends heavily on the spatial distribution of the wind field. The right-front quadrant of a Northern Hemisphere storm (relative to its track) typically carries the highest winds because storm motion adds to rotational wind speed. It also generates the longest fetch for wave growth and the largest storm surge. SAR retrieves this full two-dimensional field in a single overpass, something that a sparse network of buoys, ships or aircraft reconnaissance passes cannot replicate.
Published peer-reviewed retrievals using Sentinel-1 and RADARSAT-2 have resolved eyewall wind structure in Atlantic and Pacific hurricanes, including asymmetric features of 20 to 40 km scale that are operationally relevant to surge forecasting. The spatial detail is genuinely useful: a uniform parametric wind model applied to the same storm may underestimate surge in one coastal sector and overestimate it in another.
The saturation problem at extreme winds
CMOD5.N was developed and validated primarily at wind speeds below about 25 m/s. Above roughly 30 to 35 m/s, C-band backscatter begins to saturate: the ocean surface becomes so disturbed, with foam, spray and breaking waves suppressing the capillary wave field, that additional wind speed produces diminishing increments in backscatter. Retrievals in this regime carry substantially larger uncertainty, potentially 5 to 10 m/s or more, which is exactly the range that matters most for Category 3 to 5 intensity assessment.
Research groups have proposed corrections using cross-polarisation (VH or HV) channels, which do not saturate as readily as co-polarisation (VV or HH) at extreme winds. Sentinel-1's EW mode acquires dual polarisation, and RADARSAT-2 offers quad-pol modes. Cross-pol GMFs are less mature than CMOD5.N but are an active area of development. Buyers should treat SAR-derived winds above 40 m/s as indicative rather than precise, and should combine them with aircraft reconnaissance or stepped-frequency microwave radiometer data where available.
The single-overpass constraint and what to do about it
A SAR acquisition captures the wind field at one moment. A mature tropical cyclone evolves on timescales of hours: eyewall replacement cycles can cause 15 to 20 m/s intensity changes in under 12 hours. A single overpass may catch the storm at a phase that is unrepresentative of its peak or its landfall state. This is not a flaw unique to SAR; it applies to any snapshot observation. But it is worth stating plainly because the spatial richness of a SAR wind field can create a false sense of completeness.
The practical mitigation is data fusion. SAR-derived wind fields are most valuable when assimilated into, or validated against, operational hurricane models rather than used in isolation. Several national meteorological agencies, including NOAA's National Hurricane Center and ECMWF, have explored SAR wind assimilation. The RCM's improved revisit over RADARSAT-2 reduces the timing gap somewhat, but sub-6-hour revisit over a single storm location remains out of reach for any current SAR constellation.
From backscatter to an operational product
A standard processing chain begins with calibrated sigma-nought imagery, applies incidence-angle correction, ingests a background wind direction field from ECMWF or GFS, and inverts CMOD5.N iteratively to produce a gridded wind speed map at the native SAR resolution downsampled to roughly 500 m to 1 km to suppress speckle noise. Rain contamination is flagged using co-pol to cross-pol ratio thresholds or auxiliary rain-rate estimates from microwave imagers such as AMSR2 or GPM. The output is typically a NetCDF or GeoTIFF wind speed and direction grid, georeferenced and timestamped.
Satellize processes Sentinel-1 acquisitions over active storm basins and can incorporate commercial RADARSAT-2 or RCM tasking on client licence. The Tonga crop-estimation programme demonstrated the organisation's pipeline for rapid, operationally timed processing in a Pacific island context, a region where cyclone wind field data is especially sparse. Wind field grids can be delivered to national meteorological services or civil protection agencies within a few hours of overpass, depending on data downlink and processing queue.
Honest limits, briefly stated
SAR does not observe wind speed directly, cannot overpass the same storm more than once per day with any current constellation, saturates in the most dangerous wind regimes, and requires careful rain flagging to avoid artefacts. The method works over ocean and coastal water; it provides no information over land, where the surface roughness relationship to wind breaks down entirely. It is also a night-and-day, all-weather sensor, which is a genuine advantage over optical and infrared alternatives for a phenomenon that is frequently cloud-covered.
Used honestly, within its validated range and fused with model output and other observations, SAR wind retrieval adds real information where real information is otherwise absent. That is the case for making.
Typical figures
| Spatial resolution (wind product) | 500 m to 1 km (after speckle averaging); native SAR resolution 5–100 m depending on mode |
| Swath width | 250 km (Sentinel-1 IW), 400 km (Sentinel-1 EW), 500 km (RADARSAT-2 ScanSAR Wide) |
| Revisit (single satellite) | 6 days (Sentinel-1 per satellite); 12 days (RADARSAT-2, RCM per satellite); RCM 3-satellite constellation improves to 4-day median at mid-latitudes |
| Radar frequency | C-band 5.405 GHz (Sentinel-1, RADARSAT-2, RCM); L-band 1.27 GHz (ALOS-2 PALSAR-2) |
| Polarisation | VV, VH dual-pol standard for Sentinel-1 EW; HH, HV, VV, VH quad-pol available on RADARSAT-2 |
| Wind speed retrieval range (validated) | Approximately 2 to 30 m/s with CMOD5.N; increasing uncertainty above 30–35 m/s; cross-pol extensions to ~50 m/s under active research |
| Wind speed retrieval uncertainty (moderate winds) | Approximately 1.5 to 2 m/s RMS against buoy reference in peer-reviewed studies (below saturation threshold) |
| Latency (open Sentinel-1 data) | Near-real-time Level-1 products typically available within 1–3 hours of acquisition via Copernicus Data Space |
| Archive depth (Sentinel-1) | From April 2014 (Sentinel-1A launch) |
| Delivery formats | NetCDF, GeoTIFF wind speed and direction grids; vector wind barb shapefiles; JSON metadata |
Analytics Satellize can run
| Gridded surface wind speed and direction field | CMOD5.N GMF inversion on calibrated C-band sigma-nought, with ECMWF ERA5 or GFS background wind direction | GeoTIFF and NetCDF wind field grid, timestamped to overpass, delivered within 3 hours of data availability |
| Wind field asymmetry and quadrant wind radii | Azimuthal binning of retrieved wind speeds into NE/SE/SW/NW quadrants; threshold exceedance at 34, 50 and 64 kt radii following NHC convention | Structured JSON report with quadrant radii for ingestion into storm surge models or operational advisories |
| Rain contamination mask | Co-pol to cross-pol ratio thresholding, supplemented by GPM IMERG or AMSR2 rain-rate colocation | Binary mask layer flagging pixels where wind retrieval is unreliable due to rain; included in all wind field products |
| Cross-polarisation wind speed estimate (high-wind regime) | Empirical cross-pol GMF applied to VH channel for wind speed above 30 m/s, with uncertainty bounds | Supplementary wind speed layer for eyewall region, clearly flagged as experimental with wider confidence interval |
| Multi-overpass wind field time series | Sequential CMOD5.N retrievals across multiple satellite passes during storm lifecycle, co-registered to storm-relative coordinates | Animated GIF and NetCDF time series showing intensification or weakening across available overpasses |
| Wind field ingestion-ready file for storm surge models | Regridding and formatting of SAR wind product to ADCIRC or SLOSH model input specifications | Formatted wind forcing file compatible with specified surge model, with processing notes |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.