Offshore and coastal wind resource prospecting from SAR wind-speed climatology
C-band SAR backscatter converted via CMOD5.n resolves coastal wind features that ERA5 and mesoscale models miss, giving developers a decade-long wind-speed record at 500 m resolution before a met mast goes in.
Sensors
- Sentinel-1 A/B (C-band SAR, IW and EW modes): Interferometric Wide swath delivers 250 km swath at 5 × 20 m range-azimuth resolution; Extra Wide swath covers 400 km at 25 × 100 m. Repeat cycle 6–12 days per satellite depending on acquisition plan. The global archive runs from April 2014, giving more than a decade of acquisitions at many European and North Atlantic sites. VV polarisation is used for CMOD5.n inversion.
- Radarsat-2 (C-band SAR): Canadian Space Agency and MDA commercial system, operational since 2007. ScanSAR Wide mode provides 500 km swath at roughly 100 m resolution, extending the climatological record back several years before Sentinel-1. Useful for sites where Sentinel-1 revisit is sparse or where a longer baseline is commercially licensed.
- ASCAT scatterometer (MetOp-A/B/C, C-band): Provides 25 km and 12.5 km coastal wind products at roughly daily global revisit. Spatial resolution is too coarse to resolve mesoscale coastal features, but the long, consistent time series makes it an independent validation reference for SAR-derived climatologies and catches systematic biases in the CMOD5.n retrieval.
- ERA5 reanalysis (ECMWF): 31 km horizontal grid, hourly output, available from 1940. Used as a bias-correction reference and to fill temporal gaps in the SAR archive, not as the primary wind-speed source. ERA5 systematically underestimates wind speed in coastal jets and topographically channelled flows by amounts that vary by site and season.
What a decade of radar passes actually tells you
Every time Sentinel-1 crosses a candidate site in VV polarisation, the normalised radar cross-section of the sea surface encodes a wind speed. The geophysical model function CMOD5.n translates that backscatter value into a 10 m neutral wind speed, provided the incidence angle and wind direction are known. Repeat that retrieval across several hundred passes and you have a wind-speed frequency distribution at 500 m spatial sampling that no met mast installed today can replicate for the past.
The archive depth matters enormously. A single year of SAR passes captures seasonal patterns but misses inter-annual variability driven by large-scale circulation anomalies. At European Atlantic sites, Sentinel-1 alone provides roughly 100 to 200 usable passes per year in IW mode, depending on the acquisition plan in force at the time. Combined with Radarsat-2 going back to 2007, a developer can construct a 15-plus-year climatology at some locations. That is comparable in length to the reference periods used in bankable wind resource assessments, and it costs a fraction of a floating LiDAR deployment.
The features that reanalysis products smooth away
ERA5 runs on a 31 km grid. Coastal jets, topographic channelling through fjords or headland gaps, and low-level jets associated with stable marine boundary layers operate at scales of 2 to 20 km. A grid cell that straddles a headland averages the accelerated flow and the sheltered lee into a single number that describes neither. SAR wind fields at 500 m resolve these features directly.
The practical consequence is site-specific. A candidate zone 5 km downwind of a prominent headland may show a persistent wind speed enhancement of 10 to 20 per cent relative to the surrounding field, a feature invisible in ERA5 but consistent across dozens of SAR passes. Conversely, a site in a sheltered bay may look attractive in reanalysis but reveal persistent speed deficits in the SAR climatology. Getting this wrong at the prospecting stage is expensive: a 1 m/s error in mean wind speed translates to roughly a 10 per cent error in annual energy production for a typical power curve, which can flip a project's internal rate of return.
Where the method breaks down
Candour is worth more here than optimism. The 500 m spatial resolution is a floor, not a ceiling. In EW mode, the effective resolution is closer to 100 m in range but the noise floor is higher, and wind retrievals become unreliable below about 3 m/s and above roughly 25 m/s where the CMOD5.n function saturates. Rain cells corrupt the backscatter signal: precipitation attenuates the radar pulse and creates surface roughness that mimics high wind speed, so passes acquired during rain events must be flagged and excluded. At high latitudes in winter, this can remove a significant fraction of acquisitions.
Wind direction is not retrieved from a single-polarisation SAR image directly. CMOD5.n requires an external wind direction input, typically from ERA5 or ECMWF analysis fields. Errors in the assumed direction propagate into the speed retrieval, with sensitivity of roughly 0.5 m/s per 20-degree direction error at moderate wind speeds. Near-shore, land contamination in the SAR signal extends 1 to 3 km from the coast depending on mode and processing, which limits the method's utility for sites very close inshore. Finally, the SAR archive is not uniformly distributed in time: acquisitions cluster around the satellite's fixed repeat orbit, so certain times of day are systematically over-represented and diurnal wind cycles must be corrected using ancillary data.
Building a bankable climatology: the processing chain
Raw Sentinel-1 Level-1 GRD products are radiometrically calibrated to sigma-nought, terrain-corrected using a digital elevation model to remove geometric distortions, and then passed through a noise-removal step to reduce thermal noise in the near-range. CMOD5.n is applied pixel by pixel using co-located ERA5 wind direction at the SAR acquisition time. The resulting wind-speed map is quality-screened for rain contamination using ancillary precipitation analyses or internal SAR texture metrics.
Individual wind maps are then co-registered to a common grid and stacked. From the stack, the analyst extracts Weibull shape and scale parameters at each grid cell, mean wind speed at 10 m, the 90th-percentile wind speed, and directional frequency distributions. These are extrapolated to hub height using a log-law or power-law profile with stability corrections derived from sea-surface temperature data. The output is a GIS-ready wind atlas covering the candidate zone, with uncertainty bands that reflect the sample size, the rain-exclusion rate, and the wind-direction input error.
Fitting SAR climatology into a broader prospecting workflow
SAR wind climatology is most valuable in the early screening and site selection phases, before a developer commits to floating LiDAR or met mast deployment. It narrows a large candidate area to a shortlist of zones with the most favourable wind distributions and the least exposure to coastal sheltering, and it flags sites where the reanalysis-based resource estimate is likely to be materially wrong. It does not replace in-situ measurement for bankable energy yield assessments: lenders and independent engineers will still require a period of calibrated anemometry or LiDAR data, and the SAR climatology feeds in as a long-term reference rather than a primary measurement.
Satellize runs this processing chain on open Sentinel-1 acquisitions and adds commercial Radarsat-2 tasking where archive depth or coverage requires it. The Tonga crop-estimation programme demonstrated that the same archive-mining approach, applied to a different sensor type, can produce decision-grade outputs from open data at a fraction of the cost of field campaigns. The principle transfers directly to wind prospecting. A developer scoping a 500 km² coastal zone can receive a preliminary SAR wind atlas within weeks of instructing the analysis, giving the project team quantitative grounds to prioritise or deprioritise candidate sites before any capital is committed to in-situ equipment.
Typical figures
| Spatial resolution (wind field) | 500 m typical (Sentinel-1 IW, after smoothing for CMOD5.n stability); 1–2 km in EW mode |
| Temporal revisit per site | 6–12 days per Sentinel-1 satellite in IW mode; combined A+B gives 6-day repeat where both are tasked |
| Archive depth | Sentinel-1: April 2014 to present; Radarsat-2: 2007 to present (licensed access) |
| Radar frequency and polarisation | C-band, 5.405 GHz; VV polarisation used for CMOD5.n wind retrieval |
| Wind speed retrieval range | Approximately 3–25 m/s; CMOD5.n saturates above ~25 m/s; retrievals below ~3 m/s unreliable |
| Wind direction input | External (ERA5 or ECMWF operational analysis at SAR acquisition time); not retrieved from SAR directly |
| Land-contamination exclusion zone | 1–3 km from coastline depending on mode and processing; sites closer inshore require caution |
| Rain-contamination exclusion rate | Varies by region and season; can exceed 20–30% of passes in high-latitude winter; flagged and excluded |
| Climatology output height | 10 m neutral wind speed (CMOD5.n native); extrapolated to hub height via log-law or power-law with stability correction |
| Delivery formats | GeoTIFF wind atlas layers, NetCDF climatology stacks, Weibull parameter grids, PDF summary report |
Analytics Satellize can run
| Multi-year SAR wind-speed climatology | CMOD5.n geophysical model function applied to calibrated Sentinel-1 sigma-nought; Weibull fitting to per-pixel wind-speed distributions | GeoTIFF and NetCDF wind atlas covering candidate zone; Weibull shape/scale grids; mean and 90th-percentile wind-speed maps |
| Coastal jet and topographic channelling identification | Spatial gradient analysis of SAR wind fields; composite maps of persistent speed anomalies relative to ERA5 background | GIS polygon layer marking zones of consistent wind enhancement or deficit; annotated PDF site report |
| ERA5 bias characterisation at candidate sites | Pass-by-pass comparison of SAR-derived wind speed against co-located ERA5 10 m wind; regression and seasonal stratification | Bias and RMSE tables by season and wind-direction sector; correction factors for reanalysis-based energy yield estimates |
| Rain-contamination screening and data-quality log | SAR texture metrics and ancillary precipitation analysis (ERA5 total precipitation, IMERG) used to flag and exclude corrupted passes | Per-pass quality flag file; summary of effective sample size and exclusion rate by month |
| Hub-height wind-speed extrapolation | Log-law and power-law vertical profile with stability correction using sea-surface temperature from Copernicus Marine Service | Wind-speed and Weibull parameter grids at specified hub heights (e.g. 100 m, 120 m, 150 m); uncertainty bounds |
| Directional wind-rose climatology | Binning of SAR wind-speed retrievals by ERA5 wind direction sector; frequency-weighted speed distributions per sector | Per-grid-cell wind-rose data in CSV and GIS format; site-averaged rose plots for candidate turbine positions |
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.