Wind-farm wake mapping from SAR wind-speed retrievals
Synthetic aperture radar retrieves 10-metre wind-speed fields precise enough to resolve velocity deficits and turbulence signatures in the lee of operating offshore wind farms, giving developers and operators quantitative wake data without deploying a single met mast.
Sensors
- Sentinel-1A/B SAR-C (IW mode): 10 x 10 m spatial resolution in Interferometric Wide-swath mode, 250 km swath, C-band (5.405 GHz). Revisit 6 days per satellite, 3 days with both active. The workhorse for offshore wake studies; free archive back to 2014.
- Sentinel-1 EW mode: 400 km swath at 40 x 40 m resolution. Useful for regional wake climatologies across multiple farm clusters; spatial detail is coarser, but synoptic coverage compensates.
- RADARSAT-2 Fine-Beam Single (FBS): 8 x 8 m resolution, 50 km swath, C-band. Tasked commercially; useful when Sentinel-1 geometry or timing is unfavourable and higher spatial detail is needed for near-rotor flow.
- TerraSAR-X Stripmap: 3 x 3 m resolution, X-band (9.65 GHz), 30 km swath. Highest geometric detail of the three; X-band NRCS-to-wind conversion is less mature than C-band GMFs, introducing additional retrieval uncertainty of roughly 0.5 m/s.
What a SAR image actually measures, and what it does not
SAR does not measure wind speed directly. It measures the normalised radar cross-section (NRCS) of the sea surface, which is governed by centimetre-scale capillary and gravity-capillary waves. Those waves are set in motion by surface wind stress. Geophysical model functions (GMFs), principally CMOD5.N for C-band, convert NRCS to a 10-metre equivalent neutral wind speed. The retrieval requires wind direction as an auxiliary input, typically from ECMWF or NCEP reanalysis, because NRCS is anisotropic with respect to the radar look angle.
The practical consequence is that SAR wind speed carries an inherent retrieval uncertainty of roughly 1 to 1.5 m/s under typical open-ocean conditions. Inside a wind-farm wake, where the turbulence spectrum differs from the open-sea assumption baked into GMFs, that uncertainty may be somewhat larger. Velocity deficits of 2 m/s or more are therefore resolved with reasonable confidence; deficits below about 1 m/s sit near the noise floor and should be treated as indicative rather than definitive.
Published studies have resolved wakes 20 km downwind. Here is what that means in practice.
Multiple peer-reviewed studies using Sentinel-1 IW scenes over the North Sea have tracked wake signatures extending 20 km or more behind large offshore arrays such as Horns Rev and Anholt. The wake appears in the SAR image as a spatially coherent reduction in NRCS relative to the undisturbed upwind fetch, which maps to a velocity deficit when the GMF is applied. Cross-wake width narrows with distance as ambient turbulence re-energises the boundary layer, and the centreline deficit decays approximately as a power law with downwind distance, consistent with Jensen-model predictions.
For a developer planning an adjacent farm, the key quantities are the centreline deficit at the proposed rotor positions and the added turbulence intensity, which drives fatigue loading. SAR retrieves the former directly. Turbulence intensity is inferred indirectly from scene-to-scene variance in the wind-speed field across many passes, not from a single image. That distinction matters: a single SAR pass is a snapshot of one atmospheric stability regime, one wind speed, one wind direction. Building a statistically meaningful wake climatology requires aggregating dozens of scenes spanning neutral, stable, and unstable stratification, which at Sentinel-1 revisit rates means months of acquisition.
Atmospheric stability is not a nuisance variable. It is the variable.
Wake length and width are strongly modulated by atmospheric stability. Under stable stratification, vertical mixing is suppressed, wakes persist further downwind, and the velocity deficit at 10 km can be two to three times larger than under convective conditions at the same hub-height wind speed. SAR captures this variability naturally because it images whatever state the atmosphere is in at overpass time. The challenge is sorting the archive by stability class after the fact.
A practical approach pairs each SAR scene with a collocated stability proxy: sea-surface temperature minus air temperature from ERA5, or the Monin-Obukhov length estimated from reanalysis. Scenes are then binned by stability class before computing mean deficit profiles. This is not exotic; it is the method used in published North Sea wake climatology studies. It does require a reasonably large scene archive, typically 50 or more usable passes per wind-direction sector, to populate each stability bin with statistical confidence.
Where SAR wake data fits in the modelling chain
Jensen (top-hat) and Gaussian-plume wake models are the workhorses of wind-farm energy-yield assessment. Both carry free parameters: the wake-decay coefficient in Jensen, and the width-growth parameters in Gaussian variants. These are typically calibrated against met-mast or LiDAR measurements, which are expensive and sparse. SAR-derived wake fields offer a spatially dense, independent dataset for calibration and validation across a range of atmospheric and wind conditions.
The comparison is not trivial. SAR retrieves a 10-metre wind speed; models compute hub-height (typically 80 to 150 m) velocity deficits. A vertical profile assumption is needed to bridge the two, introducing additional uncertainty. Despite this, published comparisons show that SAR-calibrated decay coefficients reduce energy-yield prediction errors relative to default literature values, particularly for large arrays where inter-row wake accumulation is significant. SAR data does not replace high-quality in-situ measurements for final bankable yield assessments, but it provides a cost-effective first-order constraint on model parameters across a wide range of conditions.
Honest limits: cloud, ice, rain, and the single-pass problem
SAR is unaffected by cloud cover, which is its principal advantage over optical sensors in the North Sea and similarly overcast regions. Rain is a different matter. Heavy precipitation contamination of C-band NRCS is well documented; rain rates above roughly 2 mm/hr can produce NRCS signatures that mimic or mask wind-speed features. Scenes acquired during precipitation events should be flagged and excluded, using collocated weather-radar composites or reanalysis precipitation fields as a quality screen.
Sea ice is a hard stop. C-band NRCS over ice is dominated by volume scattering and surface roughness unrelated to wind, making GMF-based wind retrieval meaningless. For sites in seasonally ice-affected waters, the usable archive is restricted to ice-free months. Finally, the single-pass temporal limitation means that SAR cannot resolve wake dynamics on sub-hourly timescales. It captures the time-integrated surface signature at the moment of overpass. Rapid changes in wind direction or speed between scenes are invisible unless they leave a persistent imprint on the statistical ensemble.
From archive to deliverable
A typical SAR wake-mapping engagement begins with archive screening: identifying all Sentinel-1 IW scenes over the target farm with wind direction within a specified sector and wind speed above the cut-in threshold, then applying quality filters for rain and ice. Scene counts for a North Sea site with a multi-year Sentinel-1 archive typically run into the hundreds, of which 60 to 80 per cent pass quality screening. Wind-speed fields are retrieved using CMOD5.N, co-registered to a common grid, and composited by stability class and wind-direction bin.
The output is a set of mean velocity-deficit maps and centreline deficit profiles, delivered as georeferenced GeoTIFF layers and tabular profiles ready for import into wake-model calibration workflows. Satellize runs this processing pipeline on the open Sentinel-1 archive, with optional commercial tasking via RADARSAT-2 or TerraSAR-X for higher resolution or targeted acquisition windows. The analytics follow the same published GMF and compositing methods used in peer-reviewed North Sea wake studies, with full processing provenance documented so that results can be independently audited.
Typical figures
| Spatial resolution (Sentinel-1 IW) | 10 x 10 m (ground range detected, multi-looked) |
| Spatial resolution (TerraSAR-X Stripmap) | ~3 x 3 m; X-band wind retrieval less mature than C-band |
| Revisit (Sentinel-1, both satellites) | ~3 days at mid-latitudes; single satellite ~6 days |
| Radar frequency | C-band 5.405 GHz (Sentinel-1, RADARSAT-2); X-band 9.65 GHz (TerraSAR-X) |
| Wind-speed retrieval uncertainty | ~1.0 to 1.5 m/s under open-ocean conditions; potentially higher inside wake |
| Minimum detectable velocity deficit | ~1.5 to 2 m/s (below noise floor of GMF retrieval uncertainty) |
| Archive depth (Sentinel-1) | April 2014 to present; EW mode coverage from launch |
| Swath width | 250 km (IW), 400 km (EW), 50 km (RADARSAT-2 FBS), 30 km (TSX Stripmap) |
| Delivery formats | GeoTIFF wind-speed grids, CSV centreline deficit profiles, GPKG wake-polygon layers |
| Auxiliary inputs required | Wind direction (ERA5 or NCEP), SST for stability classification, precipitation mask |
Analytics Satellize can run
| Mean velocity-deficit maps by wind-direction sector | CMOD5.N GMF applied to Sentinel-1 NRCS; scene compositing by wind-direction bin (typically 30° sectors) | GeoTIFF layers, one per sector, georeferenced to farm grid |
| Centreline wake deficit profiles | Spatial averaging across cross-wake transects at successive downwind distances; power-law decay fitting | CSV tables of deficit vs. downwind distance, with confidence intervals from scene ensemble |
| Stability-stratified wake climatology | ERA5 Monin-Obukhov length or SST-air temperature proxy used to bin scenes into stable, neutral, unstable classes before compositing | Three-class deficit map set plus summary statistics report (PDF) |
| Jensen / Gaussian wake-model calibration dataset | SAR-derived deficit profiles compared to model output; least-squares fit of wake-decay coefficient or width-growth parameters | Calibrated parameter set with uncertainty bounds, formatted for WindPRO or OpenWake import |
| Wake-affected area polygon layer | Threshold masking of composite wind-speed deficit field at 5%, 10%, 15% deficit contours | GeoPackage polygon layer for GIS-based turbine siting constraint analysis |
| Scene-quality-screened archive inventory | Automated rain-flag (ERA5 precipitation) and ice-flag (OSI-SAF sea-ice concentration) applied to full Sentinel-1 archive over target area | Annotated scene inventory spreadsheet with pass/fail flags and usable-scene count per sector |
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.