Greenfield site screening for development suitability
Satellite-derived slope, land-cover, flood exposure and access metrics let developers cut a long-list of greenfield parcels to a short-list before committing survey budgets. The approach is fast and repeatable, but it cannot see what lies beneath the soil.
Sensors
- Copernicus DEM GLO-30: 1 arc-second (approximately 30 m at the equator) digital elevation model derived from TanDEM-X radar acquisitions. Vertical accuracy is typically better than 4 m RMSE over open terrain. Used to derive slope angle, aspect and upstream catchment area for each candidate parcel.
- Sentinel-2 MSI: 13 spectral bands from 443 nm to 2190 nm; 10 m resolution in the visible and near-infrared bands used for land-cover classification. Revisit is 5 days at the equator with both Sentinel-2A and 2B operating. Cloud cover is the principal gap; persistent overcast can delay a usable acquisition by weeks in humid climates.
- Sentinel-1 SAR: C-band synthetic aperture radar at 5.405 GHz, 10 m resolution in Interferometric Wide Swath mode. Penetrates cloud and works at night, making it useful for detecting standing water and for cross-checking land-cover boundaries that optical imagery cannot resolve under canopy.
- Landsat 8/9 OLI: 30 m multispectral bands with a 16-day repeat per satellite (8-day combined). The longer archive, extending back to 1972 across the Landsat series, is the main advantage here: it lets analysts assess whether a parcel has historically been used for agriculture, waste disposal or industrial activity before current vegetation obscures the evidence.
What you are actually trying to avoid
Most greenfield screening failures are not failures of imagination. They are failures of sequence: a developer commissions ground surveys on a dozen sites, spends six figures, and then discovers that three of them sit in a 1-in-20-year floodplain, two have slopes that would require retaining walls costing more than the land, and one has a land-cover history suggesting former landfill. Satellite data cannot replace the geotechnical investigation that follows, but it can ensure that investigation is spent on sites that have already passed a rigorous first filter.
The four criteria that satellite analytics can assess objectively are slope, land-cover class and history, flood-return-period exposure and road-access distance. Each has a measurable threshold. Each can be scored, weighted and mapped across hundreds of parcels in a single processing run.
Slope and terrain: what the DEM gives away
The Copernicus DEM GLO-30 is the practical standard for this work. At roughly 30 m posting and a vertical accuracy better than 4 m RMSE over open terrain, it resolves slope angles to within about half a degree across most of the land surface. That is sufficient to flag parcels exceeding a client-defined threshold, typically 8 to 12 degrees for standard residential development, and to identify concave drainage hollows that will collect water even when the average slope looks benign.
Slope alone is not the story. Aspect matters for solar access and frost-pocket risk. Upstream contributing area, calculated from the DEM flow-direction grid, indicates where surface runoff will concentrate after rainfall. A parcel that looks flat on a planning map can sit at the foot of a 200-hectare catchment. The DEM makes that visible in minutes.
Land cover: the present picture and the inconvenient history
Sentinel-2's 10 m bands support supervised classification into standard land-cover classes: arable, improved grassland, scrub, woodland, wetland, bare ground, and built surfaces. For most greenfield screening purposes, the classification accuracy achievable with a well-trained random forest or support-vector model on Sentinel-2 imagery is in the range of 85 to 92 percent overall, depending on class complexity and the availability of training data. Wetland and scrub are the classes most likely to be confused with each other in a single-date image; a time-series approach using the full Sentinel-2 archive substantially reduces that confusion by capturing seasonal phenological differences.
The historical dimension matters as much as the current classification. Landsat's archive, which extends to 1972 in some regions, can reveal whether a parcel carried industrial buildings, waste tips or mine workings before it was returned to grassland. Vegetation that looks uniform today may be masking contamination that will become a planning constraint tomorrow. This is not a substitute for a Phase 1 environmental desk study, but it is a rapid flag that the desk study is warranted.
Flood exposure: GloFAS and what the return-period map cannot tell you
ESA's Global Flood Awareness System, GloFAS, produces probabilistic flood-extent maps at approximately 1 km resolution for return periods ranging from 1-in-2 to 1-in-100 years. For coarse screening of large land areas, this is adequate. It will correctly eliminate parcels that sit in active floodplains. It will not resolve the difference between a parcel that is marginally inside or marginally outside a 1-in-100-year boundary when the parcel is only a few hundred metres across.
Sentinel-1 SAR provides a useful complement. Backscatter change detection between a dry-period baseline and a post-rainfall acquisition can map actual inundation extents at 10 m resolution. Running this comparison across multiple historical flood events, using the Sentinel-1 archive which extends to 2014, gives a site-specific empirical flood record rather than a modelled one. The honest caveat: SAR detects open water. Dense vegetation canopy can mask shallow inundation beneath it, and urban drainage infrastructure can prevent flooding that the terrain model would predict.
Access distance: the metric that planning committees notice
Road-access distance is calculated as the shortest network distance from each candidate parcel centroid to the nearest classified road of a specified type, using a routable road network combined with satellite-derived road detection. The satellite contribution here is identifying unmapped tracks and field roads that may provide access, and confirming that mapped roads are still present and passable rather than overgrown or washed out. Sentinel-2 imagery at 10 m resolves paved roads reliably; unpaved tracks above roughly 5 m width are detectable but less consistently classified.
Satellize combines these four scored criteria into a weighted composite rank for each candidate parcel, delivered as a GIS layer with per-parcel attribute tables. The weighting is client-defined. A residential developer and a logistics operator will set very different thresholds on slope and very different weights on access distance. The Tonga crop-estimation programme demonstrated that the same multi-criteria scoring framework adapts readily to different land-use objectives when the underlying spectral and topographic data are consistent.
What the satellite cannot see, and why that matters
The fundamental limit of this entire approach is the surface. Satellite sensors observe reflected and emitted electromagnetic radiation from the top few centimetres of the land surface or the vegetation canopy above it. They have no view of bearing capacity, soil shrink-swell behaviour, groundwater depth, buried services, archaeological remains or contamination plumes. A parcel that scores well on all four satellite-derived criteria can still fail a geotechnical investigation. The satellite screen is designed to ensure that investigation budget is concentrated on the sites most likely to succeed, not to certify that they will.
Resolution also imposes a practical minimum parcel size. At 10 m, Sentinel-2 classification is unreliable for parcels smaller than roughly 0.5 hectares; the mixed-pixel problem degrades class assignments at boundaries. For very small infill plots, the sibling page on vacant land and derelict site detection covers methods better suited to sub-hectare urban parcels. For coastal sites, flood exposure assessment for property asset valuation addresses the additional complexity of tidal and storm-surge modelling that GloFAS does not capture.
Typical figures
| Topographic resolution (DEM) | ~30 m posting (Copernicus DEM GLO-30); vertical accuracy typically <4 m RMSE over open terrain |
| Land-cover resolution | 10 m (Sentinel-2 visible/NIR bands); 30 m (Landsat 8/9 OLI) |
| Sentinel-2 revisit | 5 days at equator (2A + 2B combined); cloud cover can extend effective revisit to weeks in overcast climates |
| Sentinel-1 revisit | 6 days at equator in Interferometric Wide Swath mode |
| Flood model resolution | ~1 km (GloFAS probabilistic extents); 10 m for SAR-derived empirical inundation mapping |
| Minimum scoreable parcel size | Approximately 0.5 ha for reliable Sentinel-2 land-cover classification; smaller parcels carry elevated mixed-pixel uncertainty |
| Archive depth | Sentinel-2 from 2015; Sentinel-1 from 2014; Landsat from 1972 (selected regions); Copernicus DEM is a single epoch (TanDEM-X 2011–2015 acquisitions) |
| Delivery formats | GeoPackage or GeoTIFF per criterion layer; composite rank as vector polygon layer with attribute table; optional PDF summary per candidate parcel |
| Processing latency | Typically 3–10 working days from parcel boundary submission to ranked GIS output, depending on archive cloud cover and area of interest size |
Analytics Satellize can run
| Slope and terrain suitability layer | DEM-derived slope, aspect and upstream contributing area calculation using standard D8 or D-infinity flow-routing algorithms applied to Copernicus DEM GLO-30 | GeoTIFF and vector polygon layer with per-parcel mean slope, maximum slope, aspect class and catchment area; flagged against client threshold |
| Land-cover classification (current) | Supervised random forest classification on multi-date Sentinel-2 MSI composites; minimum six spectral bands including red-edge and SWIR; validated against ESA WorldCover 10 m product as reference | 10 m raster land-cover map clipped to area of interest; per-parcel dominant class and class-area breakdown in attribute table |
| Historical land-use change flag | Landsat time-series change detection using spectral indices (NDVI, NDBI) across available archive; anomalous spectral signatures cross-referenced against known industrial and waste-disposal spectral profiles | Per-parcel binary flag (change detected / no change detected) with date range and index values; PDF summary of flagged parcels |
| Flood-return-period exposure score | Intersection of parcel boundaries with GloFAS probabilistic flood-extent polygons for 1-in-20 and 1-in-100 return periods; supplemented by Sentinel-1 SAR backscatter change detection across historical wet-season acquisitions | Per-parcel flood-exposure class (none / marginal / moderate / high) with supporting SAR-derived inundation frequency raster |
| Road-access distance metric | Network distance calculation from parcel centroid to nearest classified road using OpenStreetMap network supplemented by satellite-detected road features from Sentinel-2; road presence confirmed against most recent imagery | Per-parcel access distance in metres and road class; vector line showing optimal access route |
| Composite site-suitability rank | Weighted linear combination of normalised criterion scores; weights set by client; sensitivity analysis run across three alternative weighting schemes to test rank stability | Ranked candidate parcel list as GeoPackage with all input scores, composite score and rank; visualisation as choropleth map PDF |
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.