Urban parking utilisation monitoring from very-high-resolution revisit
Sub-metre satellite imagery acquired at multiple times of day lets analysts count parked vehicles across surface car parks and on-street bays, producing occupancy curves that inform dynamic pricing and land-value assessments. The method is honest about what it cannot see: covered decks, underground structures, and steeply oblique acquisitions above 30 degrees off-nadir.
Sensors
- Planet SkySat: Delivers 50 cm pan-sharpened imagery with tasked revisit up to several times per day over a target. The constellation of around 20 satellites allows flexible scheduling across morning, midday and afternoon windows, which is essential for building occupancy curves rather than single snapshots.
- Maxar WorldView Legion: Targets 29 cm native resolution at nadir. Six planned satellites in two orbital planes are designed to achieve up to 15 revisits per day over mid-latitude cities, making it the highest-resolution frequent-revisit option currently entering commercial service. Individual vehicles are unambiguous at this scale.
- Airbus Pléiades Neo: 30 cm resolution with a daily revisit cadence from two satellites. Stereo and tri-stereo modes allow shadow-geometry analysis that helps distinguish parked vehicles from slow-moving traffic at acquisition moment, and can produce a surface model useful for identifying covered structures not visible in nadir imagery.
- Satellogic EarthView: 70 cm multispectral resolution with a growing constellation that offers competitive tasking costs for city-scale surveys. Useful for large-area parking audits where per-image cost matters more than the finest resolution, though vehicle-level detection confidence drops slightly relative to 30 cm sensors.
What a parked car actually looks like from 500 km
At 30 to 50 cm ground sample distance, a standard passenger car occupies roughly 8 to 15 pixels in the along-track dimension. That is enough to distinguish a vehicle from an empty bay, read lane-marking geometry, and separate tightly packed rows from dispersed parking. The shadow cast at low sun angles adds a second confirmation: a stationary vehicle casts a fixed shadow consistent with its geometry and the solar elevation angle at acquisition time, whereas a vehicle moving during the roughly 1-millisecond line exposure of a pushbroom sensor produces a characteristic smear or positional offset between panchromatic and multispectral channels.
Colour also helps. Vehicle rooftops span a narrower spectral range than road surfaces or painted bay markings, and the contrast between a white or silver roof and dark tarmac is detectable even in standard RGB. Multispectral sensors add a near-infrared channel that sharpens the separation between painted metal and asphalt. None of this requires exotic processing: the signal is physically strong at sub-metre resolution.
From single image to occupancy curve
A single acquisition tells you how full a car park was at one moment. That is useful for a snapshot audit but insufficient for pricing policy, which needs to know how occupancy varies across the working day and across days of the week. The tasked-revisit capability of SkySat and WorldView Legion changes the economics here. By scheduling acquisitions at, say, 08:00, 11:00, 14:00 and 17:00 local time over a target district, analysts can fit an occupancy curve to each surface facility across a multi-week campaign.
The analytic pipeline runs object detection, typically a convolutional neural network trained on labelled satellite imagery, across each acquisition. Detected vehicle centroids are mapped to individual bays using a pre-registered bay-polygon layer derived from the first high-resolution image or from cadastral data. Occupancy rate per bay polygon is then aggregated to facility level and time-stamped. The output is a time series: facility X was at 87 per cent occupancy at 11:00 on a Tuesday in March, 43 per cent at 08:00, and 91 per cent at 14:00. Repeat that across several weeks and seasonal and day-of-week patterns become visible.
Where the method fails, and why that matters for procurement
Covered multi-storey car parks are invisible to any optical satellite. The roof is all you see. Underground parking is entirely unobservable. This is not a processing limitation; it is physics. A city where a large fraction of parking supply is structured rather than surface-level will produce systematically incomplete occupancy data, and any pricing model built on satellite data alone will be miscalibrated unless the covered supply is measured by other means, such as entry-exit sensor counts or mobile data.
Oblique acquisitions above roughly 30 degrees off-nadir introduce a second failure mode. Vehicle sides become visible, rooftops shrink in apparent area, and shadows from adjacent buildings can occlude entire rows of bays. Most commercial tasking operators allow the client to specify a maximum off-nadir angle; for parking work, keeping this below 25 degrees is worth the reduced collection opportunity. Finally, cloud cover is the mundane enemy. Optical sensors of any resolution cannot penetrate cloud, and a city that experiences persistent overcast during the target campaign window will produce sparse time series. Synthetic aperture radar does not solve this: at the wavelengths used by Sentinel-1 (C-band, 5.6 cm) or ICEYE (also C-band), individual passenger cars are detectable as point scatterers but cannot be reliably counted in dense parking arrays.
Pricing policy and land valuation: the two applications that justify the cost
Dynamic parking pricing, the practice of raising kerb tariffs in high-demand zones and lowering them in underused ones, requires empirical occupancy data at the facility or block level. Ground-sensor networks provide this but cost tens of thousands of pounds per district to install and maintain. Satellite-derived occupancy curves over a multi-week campaign can calibrate a pricing model at a fraction of that cost, particularly in cities where sensor infrastructure is absent. The satellite data does not replace real-time ground sensing for operational pricing, but it provides the demand map that tells planners where to invest in sensors and what the target occupancy thresholds should be.
Land valuation is a less obvious application but arguably more durable. Surface car parks in urban centres are frequently underused relative to their land value, and planning authorities in several jurisdictions use parking utilisation evidence to assess whether a site qualifies for redevelopment pressure or a higher rateable value. Satellite-derived occupancy data, timestamped and reproducible, provides an auditable evidence base that a ground survey conducted on one day cannot match. Satellize has built occupancy analytics for agricultural contexts, including the Tonga crop-estimation programme, and the same time-series infrastructure applies directly to urban facility monitoring.
Accuracy, minimum detectable targets, and what the published literature says
Published studies using WorldView-2 and Pléiades imagery at 50 cm to 1 m resolution have reported vehicle detection precision and recall figures in the range of 85 to 95 per cent under good conditions: low sun angle, clear sky, nadir or near-nadir acquisition. Performance drops to roughly 70 to 80 per cent in complex urban canyons where shadow occlusion is significant, or when vehicles are smaller than about 3.5 metres in length, which excludes motorcycles and bicycles from reliable detection at 50 cm GSD. At 30 cm, motorcycles become detectable in open areas, though not reliably in shadow.
The minimum bay size detectable as occupied versus empty is approximately 2.5 by 5 metres at 50 cm GSD, which covers standard UK and EU parking bay dimensions. Compact bays in older city centres, sometimes as narrow as 2.1 metres, sit at the edge of reliable discrimination. Analysts should validate against ground truth for at least one acquisition in any new city before treating the occupancy figures as production-grade.
Typical figures
| Spatial resolution (best available) | 29 cm (Maxar WorldView Legion, nadir); 30 cm (Pléiades Neo); 50 cm (Planet SkySat) |
| Tasked revisit (single target) | Up to 15 times per day (WorldView Legion, mid-latitude cities); up to 12 times per day (SkySat); daily (Pléiades Neo) |
| Recommended maximum off-nadir angle | 25 degrees for parking work; detection confidence degrades materially above 30 degrees |
| Spectral bands | Panchromatic plus RGB and near-infrared (all four sensors listed); WorldView Legion adds coastal blue, yellow, red-edge and two SWIR bands |
| Minimum detectable vehicle length | Approximately 3.5 m reliably at 50 cm GSD; approximately 2.0 m at 30 cm GSD in open areas |
| Coverage per tasking pass | SkySat: up to 5,000 km² per day per satellite in strip mode; Pléiades Neo: 3,750 km² per day |
| Latency (tasked to delivery) | Typically 2 to 6 hours from acquisition to orthorectified product for priority tasking orders |
| Archive depth | SkySat: from 2016; WorldView-series archive (predecessor to Legion): from 2007; Pléiades: from 2012 |
| Failure modes | Covered and underground parking (unobservable); cloud cover (optical only); off-nadir above 30 degrees (shadow occlusion) |
| Delivery formats | GeoTIFF orthoimage, GeoJSON occupancy polygons, CSV time-series per facility, QGIS/ArcGIS compatible layers |
Analytics Satellize can run
| Bay-level occupancy map | Convolutional neural network object detection applied to orthorectified sub-metre imagery, vehicle centroids matched to pre-registered bay polygons | GeoJSON layer with occupancy status (occupied / vacant / occluded) per bay, per acquisition |
| Facility occupancy time series | Aggregation of bay-level detections across multiple tasked acquisitions; time-stamped occupancy rate per facility | CSV and interactive chart showing occupancy rate by hour and day of week across the campaign window |
| Demand-zone heatmap | Spatial aggregation of peak occupancy rates across all surface facilities within a defined district boundary | GeoTIFF and vector heatmap layer for import into planning GIS or pricing-policy dashboards |
| Underutilisation flag report | Threshold-based classification: facilities below a configurable occupancy threshold (e.g. 40 per cent) across majority of observation windows flagged as candidate redevelopment or repricing sites | PDF report with ranked facility list, occupancy statistics, and satellite image thumbnails |
| Shadow-geometry vehicle confirmation | Solar geometry calculation at acquisition time used to predict expected shadow length and direction; detections inconsistent with shadow geometry excluded as likely false positives from road markings or debris | Quality-flagged detection layer with confidence score per vehicle detection |
| Change detection: parking supply loss | Multi-date comparison of bay-polygon inventory to detect permanent removal of surface parking through construction or repurposing, using archive imagery | Change log with before/after image pairs and estimated bay-count delta |
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.