Greenhouse and polytunnel coverage mapping for horticultural land valuation
Greenhouse and polytunnel coverage is a direct productivity signal for horticultural land, yet planning registers routinely lag physical expansion by years. Multispectral and thermal satellite data can map covered-structure extent, density and change at scales useful for valuation, lending and investment decisions.
Sensors
- Sentinel-2 MSI: 10 m multispectral at 13 bands including NIR and SWIR; 5-day revisit at mid-latitudes. Resolves greenhouse complexes larger than roughly 20–30 m across reliably; individual narrow polytunnel rows at 3–5 m width are at or below the detection floor without sub-pixel unmixing.
- Planet SuperDove: 3 m multispectral (8 bands including red-edge and NIR); daily revisit over most land areas. Captures polytunnel rows down to approximately 6–9 m total cluster width and provides change detection at weekly cadence, useful for tracking seasonal deployment and removal of temporary structures.
- Maxar WorldView-3: 1.24 m multispectral and 0.31 m panchromatic; tasked on demand. At this resolution individual polytunnel bays and glass-house modules are directly mappable. Shortwave-infrared bands (8 SWIR bands at 3.7 m) add material discrimination between glass, polyethylene film and bare soil.
- ECOSTRESS (ISS-mounted thermal): 70 m thermal infrared at approximately 1.5–5 day revisit depending on ISS orbit precession. Greenhouses produce a measurable thermal anomaly relative to open soil, particularly in early morning acquisitions before ambient warming. Useful for confirming active heated glasshouses but too coarse for individual structure mapping.
What a greenhouse roof gives away spectrally
Glass and polyethylene film share one inconvenient property for anyone trying to hide them: both are highly reflective in the visible and near-infrared. Clean glass reflects 8–12 % of incident NIR; new polyethylene film can reflect 15–25 % depending on formulation and dust loading. Open bare soil and actively growing crops reflect in entirely different patterns. The result is a pronounced spectral contrast that multispectral sensors exploit without any special processing.
In Sentinel-2 band 8 (NIR, 842 nm), a dense greenhouse complex typically appears two to three times brighter than surrounding soil and far brighter than healthy vegetation, which absorbs NIR strongly. SWIR bands (1610 nm and 2190 nm on Sentinel-2) add a secondary discriminant: polyethylene has characteristic absorption features in SWIR that distinguish it from glass, metal roofing and concrete. WorldView-3's eight dedicated SWIR bands make this material separation considerably sharper.
Why planning records are the wrong data source for valuation
In England, polytunnels below a certain scale are frequently erected under permitted development rights and never appear in the planning register at all. Larger glasshouse complexes may have planning consent on record but the consent date tells you nothing about whether the structure was actually built, extended or demolished since. In Spain's Almería province, the world's largest concentration of greenhouse agriculture, the covered area grew from roughly 20,000 hectares in 2000 to over 40,000 hectares by the early 2020s, much of it on informal or retrospectively regularised land. Satellite archives capture that growth year by year; planning databases do not.
For a lender assessing a horticultural estate, or an investor pricing a sale-and-leaseback, the question is not what was consented but what is physically present and operating. Coverage density (percentage of a parcel under cover) correlates directly with productive capacity and therefore rental value. Satellite-derived coverage maps answer that question with a datestamp.
Resolution limits and what they mean in practice
Sentinel-2 at 10 m is adequate for mapping greenhouse complexes as a whole and tracking their expansion or contraction between seasons. It is not adequate for resolving individual polytunnel rows, which in standard strawberry or raspberry production may be 1.5–3 m wide with similar-width gaps. At 10 m, a row-gap pattern is sub-pixel; the pixel value is a mixture of film, soil and possibly crop, and the structure is detectable only as an aggregate signal over a cluster of rows.
Planet SuperDove at 3 m improves this considerably. A cluster of four or five polytunnel rows becomes resolvable as a distinct object. Individual rows still require sub-metre data. WorldView-3 at 1.24 m multispectral (or 0.31 m pan-sharpened) resolves individual bays, gutters and ridge lines, making it the appropriate sensor for precise area measurement, insurance surveys or legal boundary disputes. The trade-off is cost and revisit: WorldView-3 requires tasking, delivers a single acquisition rather than a time series, and covers a relatively small area per pass compared with Sentinel-2's 290 km swath.
Thermal confirmation and its limits
Heated glasshouses maintain interior temperatures well above ambient in winter, producing a surface thermal signature detectable by ECOSTRESS and, in some conditions, by Landsat 8/9 thermal bands at 100 m resolution. This is useful for distinguishing actively heated production facilities from cold polytunnels or derelict structures. An abandoned glasshouse with broken glazing will have a thermal signature close to ambient; an operating tomato or cucumber house in February will not.
ECOSTRESS's 70 m pixel is too coarse for individual structure mapping and the ISS orbit means revisit is irregular, sometimes daily, sometimes a gap of several days over a given location. Landsat thermal at 100 m is coarser still. Thermal data therefore serves as a confirmation layer rather than a primary mapping input. It answers whether structures are in active heated use, not where their boundaries lie.
Building a coverage map that holds up in a valuation report
A defensible horticultural coverage map combines at minimum two data sources: a medium-resolution time series (Sentinel-2 or Planet) to establish the current extent and recent change history, and a high-resolution optical acquisition (WorldView-3 or equivalent) to measure precise area within the identified clusters. Classification uses a supervised approach trained on spectral signatures from known greenhouse and polytunnel sites in the same region, accounting for local soil colour, crop type and seasonal phenology, all of which shift the background against which structures are detected.
Change analysis over a three-to-five year archive reveals whether coverage has been expanding, contracting or stable. Rapid expansion in the two years before a transaction may indicate speculative development that has not yet reached full productivity. Contraction may indicate disease, water stress or a decision to exit a crop. Neither signal appears in a planning search. Satellize's crop-estimation work for the Kingdom of Tonga demonstrated that spectral time-series methods developed for open-field agriculture translate directly to covered-structure contexts with appropriate retraining of the classifier.
Delivered outputs typically take the form of a georeferenced polygon layer (GeoJSON or Shapefile) with per-parcel coverage percentage, a change-detection summary table and, where WorldView-3 data is included, a precise area figure in hectares accurate to within the pixel resolution of the sensor.
Typical figures
| Spatial resolution (primary mapping) | 10 m (Sentinel-2); 3 m (Planet SuperDove); 1.24 m multispectral / 0.31 m pan (WorldView-3) |
| Minimum detectable cluster width | ~20–30 m at 10 m resolution; ~6–9 m at 3 m resolution; individual rows (~1.5 m) at sub-metre resolution |
| Revisit cadence | 5 days (Sentinel-2, mid-latitudes); daily (Planet); on-demand tasking (WorldView-3); irregular 1–5 days (ECOSTRESS) |
| Spectral bands used | Visible (Red, Green, Blue), NIR (~842 nm), Red-edge (~705 nm, ~740 nm), SWIR (1610 nm, 2190 nm); thermal 8–12 µm for confirmation |
| Thermal confirmation resolution | 70 m (ECOSTRESS); 100 m (Landsat 8/9 TIRS) |
| Archive depth | From 2015 (Sentinel-2); from 1972 (Landsat, coarser); from 2009 (Planet, limited early archive) |
| Cloud limitation | Optical sensors require cloud-free acquisitions; persistent cloud cover in northern Europe can limit usable imagery to 4–8 clear scenes per year in winter months |
| Typical area coverage per pass | 290 km swath (Sentinel-2); ~24 km swath (WorldView-3 tasked strip) |
| Delivery formats | GeoJSON, Shapefile, GeoTIFF, CSV summary table; compatible with QGIS, ArcGIS, ESRI REST API |
| Area measurement accuracy | ±1 pixel boundary uncertainty; at 1.24 m this equates to roughly ±1.5 % on a 1 ha parcel |
Analytics Satellize can run
| Covered-structure extent map | Supervised spectral classification using NIR/SWIR band ratios and texture metrics; trained on labelled greenhouse and polytunnel polygons | Georeferenced polygon layer (GeoJSON/Shapefile) with per-parcel coverage area in hectares and coverage percentage |
| Multi-year change detection | Bitemporal or time-series differencing on NIR reflectance composites; change confirmed against Planet daily archive | Change-detection table showing expansion, contraction or stability per parcel across a user-specified date range, with annotated imagery |
| Material classification (glass vs. polyethylene) | SWIR band analysis using WorldView-3 8-band SWIR; polyethylene absorption features at ~1730 nm and ~2310 nm distinguish film from glass and metal | Per-structure material type attribute appended to polygon layer; relevant for insurance and replacement-cost valuation |
| Active-use confirmation | ECOSTRESS or Landsat TIRS thermal anomaly detection; heated structures flagged where surface temperature exceeds ambient by a statistically significant margin in winter acquisitions | Binary active/inactive attribute per structure cluster, with acquisition date and temperature delta |
| Coverage density score per land parcel | Zonal statistics of classified coverage layer intersected with cadastral or valuation parcel boundaries | Tabular report with coverage density (%) per parcel, suitable for direct input to rental or capital value models |
| Seasonal deployment tracking | Monthly Planet SuperDove composites classified and compared to identify temporary polytunnel erection and removal cycles | Time-series chart of covered area by month, indicating whether structures are permanent or seasonal |
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.