Onshore wind development constraint mapping from satellite land-cover classification

Satellite land-cover classification and digital elevation data can pre-screen onshore wind sites against setback, habitat and visual-impact constraints before a single site visit, cutting desktop feasibility time significantly. This page explains the sensors, methods, honest limits and analytic outputs involved.

Sensors

Sentinel-2 MSI: 13 spectral bands, 10 m resolution in visible and near-infrared, 20 m in red-edge and shortwave-infrared. Five-day revisit at the equator (two to three days at mid-latitudes with both satellites). Time-series stacking over a full seasonal cycle is the standard approach for separating land-cover classes that look similar in a single image.
ESA WorldCover 10 m: Global land-cover map at 10 m, produced from Sentinel-1 SAR and Sentinel-2 optical data. Eleven classes including built-up, tree cover, shrubland, grassland, cropland, wetland and bare ground. Published overall accuracy of approximately 75 percent globally; accuracy is higher in temperate Europe and lower in spectrally ambiguous tropical transitions.
Copernicus DEM GLO-30: Global digital elevation model at 30 m posting, derived from TanDEM-X radar interferometry. Absolute vertical accuracy better than 4 m RMSE over most terrain. Used for viewshed computation, slope and aspect layers, and terrain-following setback buffer generation.
CORINE Land Cover: Pan-European land-cover inventory at 100 m minimum mapping unit, updated at roughly six-year intervals. Forty-four classes with legal standing in several EU member-state planning frameworks. Coarser than WorldCover but directly referenced in some national statutory guidance, making it useful for cross-checking classification outputs.
Sentinel-1 SAR C-band: 6 m by 22 m ground range resolution in Interferometric Wide Swath mode, six-day repeat. Penetrates cloud, enabling land-cover discrimination in persistently overcast regions where optical time-series have seasonal gaps. Backscatter texture distinguishes forest from grassland and detects standing water in potential peatland areas.

What the planning envelope actually looks like from orbit

Onshore wind consent in most jurisdictions rests on a set of spatial rules: minimum separation distances from dwellings (typically 500 m to 2 km depending on national guidance), exclusion of designated habitats such as Sites of Special Scientific Interest or Natura 2000 areas, aviation safeguarding zones around aerodromes and low-flying corridors, and visual-impact assessments tied to receptor locations. Each of these constraints has a spatial footprint. Most of that footprint can be approximated, though not formally determined, from satellite data before any ground investigation begins.

The practical value is in rapid elimination. A 50 km² search area might contain a dozen candidate parcels after removing statutory designations from a GIS. Satellite land-cover classification can narrow that to three or four by identifying terrain that is already forested, actively cultivated under a sensitive crop regime, or topographically screened from wind. That is not a planning determination; it is a prioritised shortlist that makes the subsequent ground campaign cheaper and faster.

How a Sentinel-2 time series separates the classes that matter

A single Sentinel-2 scene is a poor classifier for land cover. Grassland and winter cereals are spectrally similar in February. Deciduous woodland and dense scrub overlap in July. The standard solution is to stack images across a full phenological cycle and compute per-pixel statistics: median reflectance, amplitude of the normalised difference vegetation index seasonal curve, and red-edge chlorophyll indices. These multi-temporal feature vectors separate classes that a single-date image cannot.

For wind-siting purposes the critical distinctions are: semi-natural grassland versus improved pasture (relevant to habitat designation risk), blanket bog and upland heath (peat depth is a statutory concern in Scotland and Ireland, for example), dense woodland requiring felling consent, and built-up footprints including isolated dwellings that trigger setback rules. Sentinel-2 at 10 m can resolve individual farm buildings reliably. It cannot reliably detect a single isolated dwelling smaller than roughly 20 m across, so settlement buffers derived from satellite data alone should be treated as indicative and checked against OS or cadastral data.

Cloud cover is the principal operational constraint. In persistently cloudy regions such as the Scottish Highlands or Norwegian coast, a full-year Sentinel-2 stack may contain fewer than ten usable scenes at any given pixel. Sentinel-1 SAR backscatter composites fill the gap for broad structural classes but cannot replicate the spectral discrimination of optical bands.

Viewshed geometry and what a DEM can and cannot tell a planner

Visual impact is a material consideration in most onshore wind planning systems. A viewshed computation asks: from which ground locations is a turbine hub or rotor tip theoretically visible, given the intervening terrain? The Copernicus DEM GLO-30 at 30 m posting is adequate for regional screening over open upland terrain. It is less reliable in narrow valleys, where 30 m cells average across ridge and floor, and it does not represent vegetation canopy height. A turbine that is technically visible over bare terrain may be screened by a mature forest block that the DEM ignores.

The standard workflow intersects the DEM viewshed with the land-cover classification to identify residential receptors within the theoretical zone of visual influence, typically defined as a 15 km to 35 km radius depending on turbine height. This produces a receptor count and a rough Zone of Theoretical Visibility, which planners use as a trigger for more detailed photomontage work. The satellite-derived output is a screening input, not a substitute for the photomontage and landscape character assessment that statutory process requires.

Stacking the constraint layers: where the candidate parcels survive

The analytic product is a constraint surface: a raster or vector layer in which each cell carries a binary or graded score reflecting how many exclusion criteria it fails. A cell classified as blanket bog within 1 km of a Natura 2000 boundary, inside an aviation safeguarding zone, and within 800 m of a dwelling fails four criteria simultaneously and is effectively eliminated. A cell on improved grassland on a south-westerly slope, 1.5 km from the nearest settlement and outside all statutory designations, survives all filters.

The honest limit here is classification error propagation. If the land-cover map misclassifies a semi-natural grassland as improved pasture at a rate consistent with WorldCover's published accuracy figures, some ecologically sensitive parcels will appear to clear the habitat filter. Ground-truthing and statutory ecological survey remain mandatory before any consent application. Satellite screening reduces the area requiring survey; it does not replace the survey.

Archive depth matters for this use case. Sentinel-2 data runs from 2015, giving roughly a decade of seasonal cycles. That is enough to detect land-cover change, flag areas that have been afforested or rewetted since a baseline cadastral survey, and identify parcels with stable open-ground character across multiple years. Stability is itself a positive signal for wind-site suitability.

Where Satellize fits in the feasibility workflow

Satellize runs land-cover classification and constraint-layer derivation on open Sentinel and Copernicus datasets, with commercial very-high-resolution tasking added where a client needs sub-metre confirmation of a specific parcel boundary or structure. The workflow produces GIS-ready constraint surfaces and candidate-parcel shapefiles that slot directly into a developer's planning GIS. Satellize's analytics team applies the same classification pipeline it uses for the Tonga crop-estimation programme to temperate land-cover problems, adapting the training classes and phenological feature set to the target region.

A typical desktop screening engagement covers a defined search area, delivers a constraint surface at 10 m resolution, a ranked candidate-parcel list with per-parcel constraint scores, and a viewshed summary for the top-ranked sites. It does not include ecological survey, noise modelling, shadow-flicker assessment or any statutory submission document. Those remain the responsibility of the developer's specialist consultants. The satellite layer is the starting point, not the finish line.

Typical figures

Land-cover classification resolution	10 m (Sentinel-2 / WorldCover); 100 m minimum mapping unit for CORINE cross-check
DEM posting for viewshed	30 m (Copernicus DEM GLO-30); vertical accuracy better than 4 m RMSE over most terrain
Sentinel-2 revisit	5 days at equator; 2 to 3 days at mid-latitudes with both Sentinel-2A and 2B operational
Sentinel-1 SAR revisit	6 days per satellite in Interferometric Wide Swath mode
Spectral bands used for classification	Sentinel-2 bands B2 to B8A (visible, NIR, red-edge) and B11/B12 (SWIR); NDVI, EVI, red-edge chlorophyll index derived
Minimum resolvable settlement feature	Approximately 20 m across at 10 m pixel; isolated small structures may be missed; cadastral cross-check recommended
Satellite archive depth	Sentinel-2 from 2015 (approximately 10 years); Landsat back to 1972 for historical land-cover change context
WorldCover published accuracy	Approximately 75% overall globally; higher in temperate Europe; lower in spectrally ambiguous or persistently cloudy regions
Viewshed radius typically modelled	15 to 35 km from proposed turbine location, depending on hub height and national guidance
Delivery formats	GeoTIFF constraint rasters, GeoPackage or Shapefile candidate parcels, PDF screening report

Analytics Satellize can run

Multi-temporal land-cover classification	Supervised random-forest or gradient-boosting classifier trained on Sentinel-2 phenological feature stack (median, amplitude, NDVI seasonality); validated against WorldCover and CORINE reference layers	10 m GeoTIFF land-cover map with per-class confidence scores; GeoPackage of class polygons
Statutory constraint surface	Spatial intersection of classified land cover with user-supplied designation boundaries (Natura 2000, SSSI, aviation zones, cadastral settlement points) and configurable setback buffers	Raster and vector constraint layer with per-cell exclusion-flag count; PDF summary table
Candidate parcel identification and ranking	Boolean and weighted overlay of constraint surface; parcels ranked by number of constraints failed, slope suitability from DEM, and land-cover stability across archive years	Ranked shapefile of candidate parcels with attribute table of per-parcel scores
Viewshed and Zone of Theoretical Visibility	Line-of-sight computation on Copernicus DEM GLO-30 for specified hub height and rotor tip; receptor count from settlement layer within defined radius	Viewshed GeoTIFF, receptor count table, per-site ZTV polygon for top-ranked candidates
Land-cover change detection	Bi-temporal or multi-temporal change detection across Sentinel-2 archive from 2015; flags afforestation, rewetting, built-up expansion within search area	Change-detection GeoTIFF with change-type labels; alert report for parcels showing recent sensitive-class encroachment
Peatland and wetland risk flagging	Combination of Sentinel-1 backscatter moisture proxies, NDVI seasonality, and DEM-derived topographic wetness index to identify probable peat or wetland ground within candidate parcels	Risk-flag overlay on candidate parcels; written caveat layer recommending ground-truth survey priority

Sources

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.