Near-surface wind-resource mapping from SAR
C-band SAR backscatter, inverted through geophysical model functions, yields 10 m wind-speed maps at 100–500 m spatial detail across open water. A single overpass is a snapshot, not a climatology, so statistical fusion with ERA5 or MERRA-2 reanalysis is essential before any investment decision.
Sensors
- Sentinel-1A/B SAR-C (ESA/Copernicus): 5.4 GHz C-band, IW mode ground-range resolution approximately 5 × 20 m (multi-looked to ~100 m for wind retrieval), 250 km swath. Nominal 6-day repeat at the equator with both satellites; 12-day with one. Free and open archive from 2014. Primary workhorse for offshore wind mapping in Europe and globally.
- RADARSAT-2 (MDA/CSA): C-band, ScanSAR Wide modes give 100 m resolution over 500 km swaths, enabling large-area wind-field mapping in a single pass. Commercial tasking allows targeted acquisition windows. Operational since 2007, giving a useful multi-year archive for statistical analysis.
- ALOS-2 PALSAR-2 (JAXA): L-band (1.27 GHz) rather than C-band. GMF-based wind retrieval from L-band is less mature and less validated than CMOD5.N at C-band, but PALSAR-2 ScanSAR mode covers 350 km at 60 m resolution and is the only freely accessible archive for parts of the Asia-Pacific region not well served by Sentinel-1.
- ERA5 reanalysis (ECMWF): Not a SAR sensor, but an essential fusion partner. ERA5 provides hourly 10 m wind at ~31 km resolution globally from 1940 onward. Its temporal density compensates directly for SAR's sparse overpass rate; the standard workflow conditions SAR retrievals on ERA5 wind direction, which CMOD5.N requires as an input.
What the backscatter is actually measuring
C-band radar energy scattered back from the sea surface is dominated by Bragg resonance with centimetric capillary-gravity waves. Those small waves are in near-equilibrium with the local wind stress at the surface. The geophysical model function CMOD5.N, developed and validated at ECMWF against scatterometer data, relates normalised radar cross-section to 10 m equivalent neutral wind speed and the incidence angle and relative wind direction of the radar look. Invert the function and you get a wind-speed estimate at every resolution cell across the swath.
The retrieval is not a simple lookup. CMOD5.N is a forward model, so inversion requires an assumed wind direction, typically supplied by a collocated NWP field such as ERA5 or ECMWF operational analysis. Errors in that direction input propagate into the speed estimate. Published validation studies over open ocean report root-mean-square errors of roughly 1.5–2.0 m/s against buoy measurements under moderate wind conditions, degrading at very low winds (below about 3 m/s, where the sea surface becomes specular) and at very high winds (above roughly 20 m/s, where wave breaking saturates the signal).
Why offshore resource mapping is the primary use case
Open water is the ideal target for two reasons. First, the GMF was calibrated over the ocean; its physical assumptions hold when the surface is homogeneous and the roughness elements are wind-driven capillary waves. Second, offshore wind sites are often in data-sparse areas where mast deployment is expensive and logistically difficult. A single Sentinel-1 IW acquisition covers 250 km of coastline at roughly 100 m spatial detail, resolving fetch gradients, coastal acceleration zones, and the influence of islands or headlands that a sparse buoy network would miss entirely.
The spatial resolution advantage is decisive for layout planning. Turbine spacing in a large offshore array is typically 700–1000 m; a 100–500 m wind map resolves the intra-array variability that determines annual energy production differences between layout options. Mast data at a single point cannot do that, and mesoscale NWP at 10–30 km resolution smooths over it.
The snapshot problem: one overpass is not a wind climate
This is the most important limitation to state plainly. Sentinel-1 revisits a given offshore location every 6 to 12 days, depending on satellite availability and acquisition scheduling. Each pass captures a single instantaneous wind state. Wind energy yield depends on the full probability distribution of wind speed across all hours of the year. No realistic number of SAR acquisitions, even over a multi-year archive, samples that distribution densely enough to stand alone as a bankable wind-resource estimate.
The accepted solution is statistical fusion. SAR wind maps are used to correct the spatial structure of a long-term reanalysis climatology rather than to replace it. The method, sometimes called SAR-mesoscale wind-field downscaling, fits a transfer function between collocated SAR retrievals and ERA5 values at each grid point, then applies that spatial correction to the full ERA5 time series. The result inherits ERA5's temporal completeness and SAR's spatial resolution. Published work from the Technical University of Denmark (DTU Wind Energy) has demonstrated this approach over the North Sea and Baltic. The method is sound, but it compounds the uncertainties of both inputs, and the correction is only as good as the number of collocated SAR scenes available, typically 50–200 for a two- to three-year Sentinel-1 archive over a given site.
Onshore terrain: where the method degrades
Over land, the GMF breaks down. CMOD5.N was not designed for vegetated or built surfaces, and the backscatter from a forest canopy or a hillside has no physical relationship to the local wind speed. Attempts to apply SAR wind retrieval over complex terrain produce artefacts driven by topographic radar shadow, layover, and surface-type changes rather than by wind.
Some research groups have explored using SAR to map surface roughness length over flat terrain, which feeds into wind-shear models, but that is a separate and less mature method. For onshore sites with complex topography, SAR contributes to micrositing through terrain and land-cover characterisation rather than through direct wind-speed retrieval. Buyers assessing onshore sites should not expect a wind map from SAR; they should expect a terrain-roughness and land-cover input to a CFD or linearised flow model.
From raw scenes to a bankable wind-resource layer
A standard processing chain runs as follows. Sentinel-1 Level-1 Ground Range Detected products are ingested, radiometrically calibrated to sigma-nought, and terrain-corrected. Incidence-angle variation across the swath is accounted for. Wind direction from ERA5 at the overpass time is interpolated to the SAR grid. CMOD5.N is then inverted cell by cell to produce a 10 m equivalent neutral wind-speed raster at 100–500 m posting. Quality flags mask areas affected by rain cells (detectable as anomalously bright patches), ship wakes, and coastal land contamination within roughly 2–3 km of the shoreline.
Multiple scenes are then co-registered and the transfer-function correction to ERA5 is computed. The output is a long-term mean wind-speed map and, critically, a Weibull parameter map (shape k and scale A) that characterises the speed distribution at each pixel. Those Weibull parameters feed directly into standard wind-energy yield models. Satellize runs this chain on open Sentinel-1 archives and can incorporate RADARSAT-2 acquisitions under client licence, delivering outputs as GeoTIFF rasters and accompanying uncertainty layers.
Uncertainty quantification is not optional. A responsible deliverable states the number of scenes contributing to each pixel's statistics, the RMSE of the SAR-ERA5 transfer function, and the sensitivity of the Weibull fit to the scene count. Fewer than about 30 collocated scenes produces Weibull estimates too uncertain for bankable resource assessment; the honest answer at that point is to extend the acquisition window before drawing conclusions.
Typical figures
| Spatial resolution (wind product) | 100–500 m posting, depending on multi-looking applied to Sentinel-1 IW or RADARSAT-2 ScanSAR |
| Swath width | 250 km (Sentinel-1 IW); up to 500 km (RADARSAT-2 ScanSAR Wide) |
| Revisit (Sentinel-1, both satellites) | 6 days at mid-latitudes when both satellites operational; 12 days with one satellite |
| Radar frequency | 5.4 GHz C-band (Sentinel-1, RADARSAT-2); 1.27 GHz L-band (ALOS-2 PALSAR-2) |
| Wind-speed retrieval range | Approximately 3–20 m/s with RMSE ~1.5–2.0 m/s over open ocean; outside this range accuracy degrades |
| Wind direction input required | External NWP (ERA5 or similar); SAR alone does not retrieve direction |
| Archive depth (Sentinel-1) | From April 2014 (Sentinel-1A); Sentinel-1B data available 2016–2021 |
| Minimum scene count for Weibull fit | ~30 collocated scenes recommended; fewer than 30 produces unreliable distribution parameters |
| Coastal exclusion zone | Typically 2–3 km from shoreline masked due to land contamination in radar sidelobes |
| Delivery formats | GeoTIFF (wind speed, Weibull k and A, scene-count, uncertainty); CSV summary statistics per zone |
Analytics Satellize can run
| Single-overpass 10 m wind-speed raster | CMOD5.N GMF inversion applied to calibrated Sentinel-1 sigma-nought with ERA5 wind-direction input | GeoTIFF at 100–500 m resolution, quality-flagged, with per-pixel incidence-angle metadata |
| Multi-year mean wind-speed map | Statistical compositing of all available SAR scenes over archive period, outlier-filtered | GeoTIFF mean wind speed with scene-count layer; PDF summary report with spatial statistics |
| Weibull parameter map (k and A) | Maximum-likelihood Weibull fitting to per-pixel SAR wind-speed sample, conditioned on ERA5 transfer-function correction | Paired GeoTIFFs (shape and scale) plus uncertainty bounds derived from bootstrap resampling |
| SAR-corrected ERA5 wind climatology | Spatial transfer-function downscaling: SAR-ERA5 bias correction applied to full hourly ERA5 time series at each pixel | Hourly wind-speed time series at SAR spatial resolution, delivered as NetCDF; annual energy production estimate per layout point |
| Coastal acceleration and orographic channelling map | Spatial anomaly analysis of mean SAR wind field relative to ERA5 background, identifying persistent high- and low-speed zones | GIS polygon layer of candidate high-resource zones with mean speed and Weibull A value per zone |
| Acquisition planning and scene-count adequacy report | Monte Carlo simulation of Weibull-fit uncertainty as a function of scene count, given site latitude and Sentinel-1 coverage geometry | PDF report recommending minimum archive window or supplementary commercial tasking (RADARSAT-2) to reach target uncertainty threshold |
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.