Sports and recreation facility audit from satellite imagery
Spectral classification and shape-template matching across Planet, Sentinel-2 and Pléiades Neo imagery can inventory formal and informal sports facilities city-wide, exposing gaps in provision that ground surveys routinely miss.
Sensors
- Planet Dove (PlanetScope): 3 m native resolution, daily revisit at most latitudes, four multispectral bands (blue, green, red, near-infrared). Sufficient to distinguish a full-size football pitch from surrounding hard surfaces and to detect informal five-a-side cages. Cannot reliably resolve a single badminton court line.
- ESA Sentinel-2 MSI: 10 m resolution in visible and NIR bands, 20 m in red-edge and SWIR, 5-day revisit at the equator with both satellites. Ideal for city-scale spectral classification of grass, artificial turf and hard courts, and for time-series analysis of seasonal grass condition. Too coarse to detect facilities smaller than roughly 400 m².
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral, tasked on demand. Resolves court markings, goal-post shadows and surface texture well enough to distinguish natural grass from artificial turf by texture alone. Revisit over a single city is typically 1 to 2 days when both satellites are tasked. Cost per km² is substantially higher than open-data options, so it is best reserved for validation strips or high-priority districts.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral, plus eight SWIR bands at 3.7 m. The SWIR bands are particularly useful for separating artificial turf polymers from natural grass, which look similar in standard RGB. Archive depth extends to 2014 for many cities.
What a sports surface looks like from 500 km up
Natural grass on a well-maintained pitch has a strong near-infrared reflectance signature driven by chlorophyll and leaf-cell structure. The normalised difference vegetation index (NDVI) for healthy turf typically sits between 0.5 and 0.8, well above the 0.1 to 0.3 range of sparse urban grass or stressed lawns. Artificial turf made from polyethylene fibres reads differently: its NDVI is lower and more stable across seasons, and its SWIR reflectance is elevated relative to living vegetation. WorldView-3's SWIR bands exploit this directly. Hard courts, whether asphalt, concrete or acrylic-painted surfaces, have characteristically flat spectral curves with high visible reflectance and negligible NIR response.
Shape matters as much as spectra. A football pitch is a rectangle of roughly 68 by 105 metres. A running track is an annulus with a standard inner radius near 36.8 metres. A basketball court is 28 by 15 metres. Template-matching algorithms applied to pan-sharpened Pléiades Neo or WorldView-3 imagery can search for these geometries systematically across a city. At 30 cm resolution, the white line markings themselves are detectable, which removes most ambiguity. At Planet's 3 m, lines are not resolved, so classification relies on the spectral signature and the gross shape of the cleared area.
The lawn problem: why spectral data alone is not enough
The most persistent confusion in urban sports-facility mapping is between public recreation grass and private residential lawns. Both have similar NDVI values, similar geometry at moderate resolution, and similar seasonal behaviour. A large back garden in a low-density suburb can be spectrally indistinguishable from a small public park at Sentinel-2's 10 m pixel size. Even at Pléiades Neo's 30 cm, a well-tended private lawn and a public pitch share texture and colour.
Resolving this requires auxiliary data. Land-registry parcels, cadastral boundaries and planning-use designations are the primary tools. Where a parcel is registered as residential, a grass patch inside it is almost certainly private. Where no parcel boundary encloses a green area, or where the parcel is recorded as public open space, the probability of a recreation facility rises sharply. The catch is that cadastral data quality varies enormously across cities and is frequently out of date, particularly in rapidly urbanising areas or informal settlements. In those contexts, field-survey sampling is the honest fallback for ground-truth.
Informal facilities are where the equity gap actually lives
Formal facilities, those with fences, floodlights and registered ownership, are generally known to municipal authorities already. The analytical value of satellite imagery is greatest for informal provision: the concrete slab in a housing estate car park repurposed as a basketball court, the vacant lot with improvised goalposts, the rooftop play area. These are systematically under-recorded in planning databases and systematically over-represented in low-income districts.
Detecting them requires sub-2 m resolution and a classification approach that does not presuppose standard dimensions. Object-based image analysis (OBIA), which segments imagery by spectral homogeneity and shape compactness before applying rules, performs better here than pixel-based classifiers trained on formal facilities. Published research using WorldView imagery in cities across sub-Saharan Africa and South Asia has demonstrated detection of informal pitches down to roughly 200 m² when imagery quality is sufficient. Below that threshold, confusion with other hard surfaces becomes difficult to resolve without field confirmation.
Linking detected informal facilities to population density grids, such as those derived from WorldPop or Facebook's High Resolution Settlement Layer, allows planners to calculate a recreation-space-per-capita figure at sub-district level. That calculation is the quantitative input that equitable-access policy actually needs.
Building the city-wide inventory: a practical workflow
A sensible approach starts with Sentinel-2 for city-wide spectral classification, producing a coarse map of candidate recreation surfaces at 10 m. This is cheap, fast and covers even large metropolitan areas in a single scene. Planet Dove adds daily temporal depth, useful for confirming that a green patch is actively maintained rather than seasonally dormant. Pléiades Neo or WorldView-3 tasking is then concentrated on the ambiguous polygons: areas that the coarse classification flags as probable facilities but that cannot be confirmed without higher resolution.
Accuracy assessment is non-negotiable. A stratified random sample of classified polygons, verified against street-level imagery or field visits, is the standard method. Published studies on urban green-space mapping with Sentinel-2 report overall accuracies of 80 to 90 per cent for broad land-cover classes, but sports-facility-specific precision is typically lower, around 70 to 85 per cent, because the category is spectrally heterogeneous. Users should budget for a validation campaign before treating the inventory as authoritative for policy purposes.
Satellize has applied analogous spectral classification pipelines in agricultural contexts, including the Kingdom of Tonga crop-estimation programme, and the same multi-sensor fusion logic transfers directly to urban surface classification.
Change detection: tracking provision over time
A single-epoch inventory answers the question of where facilities exist today. A time-series answers the more politically useful question of whether provision is improving, degrading or simply being redistributed as the city grows. Sentinel-2's archive runs from 2015, Planet's usable archive from roughly 2016. WorldView-3 archive coverage for many cities extends to 2014. That gives a decade of comparable imagery for most urban areas.
Practical applications include monitoring whether a facility that appears in planning approvals has actually been built, detecting conversion of public recreation land to other uses, and tracking the condition of grass surfaces through seasonal NDVI trajectories. A pitch that shows declining NDVI over successive summers is likely underfunded for maintenance, a signal that is invisible to any audit method that relies solely on administrative records.
Typical figures
| Spatial resolution (city-wide classification) | 10 m (Sentinel-2 MSI); 3 m (Planet Dove) |
| Spatial resolution (facility confirmation) | 30 cm pan / 1.2 m multispectral (Pléiades Neo); 31 cm pan / 1.24 m multispectral (WorldView-3) |
| Revisit cadence | Daily (Planet Dove); 5 days at equator (Sentinel-2, both satellites); 1–2 days tasked (Pléiades Neo) |
| Minimum detectable facility area | ~400 m² at Sentinel-2; ~200 m² at WorldView-3 sub-metre (informal surfaces, with OBIA) |
| Key spectral bands | Visible + NIR (NDVI for grass/turf); SWIR (WorldView-3 for polymer vs. vegetation separation); Red-edge (Sentinel-2 for turf stress) |
| Archive depth | 2015 to present (Sentinel-2); 2016 to present (Planet); 2014 to present (WorldView-3 for most cities) |
| Typical classification accuracy | 70–85% precision for sports-facility class specifically; 80–90% for broad urban green-space (published Sentinel-2 studies) |
| Auxiliary data required | Cadastral / land-registry parcels to resolve public vs. private ownership; population density grid for per-capita access calculation |
| Delivery formats | GeoPackage / Shapefile polygon inventory; GeoTIFF classified raster; CSV attribute table with facility type, area and confidence score |
Analytics Satellize can run
| City-wide sports-facility polygon inventory | Multi-sensor spectral classification (NDVI, SWIR indices) combined with OBIA shape-template matching on pan-sharpened imagery | GeoPackage layer with facility type (natural grass, artificial turf, hard court, playground), area, centroid coordinates and classification confidence |
| Public vs. private green-space disambiguation | Spatial intersection of classified polygons with cadastral parcel boundaries and planning-use designations | Attributed GIS layer with ownership-class field and flagged ambiguous parcels requiring field verification |
| Recreation-space-per-capita access map | Zonal statistics of confirmed public facility area against population density grid (WorldPop or equivalent) at sub-district level | Choropleth GIS layer and tabular report by administrative unit, suitable for equity-planning input |
| Informal facility detection in under-served districts | OBIA segmentation on Pléiades Neo or WorldView-3 imagery with relaxed geometry constraints; spectrally homogeneous hard or grass segments flagged for analyst review | Point layer of candidate informal facilities with thumbnail image chips for rapid analyst confirmation |
| Facility condition time-series (grass surfaces) | NDVI trajectory extraction from Sentinel-2 and Planet Dove archive over detected grass pitches; anomaly detection against seasonal baseline | Per-facility NDVI trend chart and annual condition-change GIS layer |
| Planning-compliance check (approved vs. built facilities) | Change detection between planning-approval date imagery and current epoch; presence/absence classification of expected facility footprint | Tabular report flagging approved facilities not yet detectable in imagery, with before/after image pairs |
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.