Construction crane census and activity indexing

Tower crane counts derived from sub-metre satellite imagery give economists and investors a near-real-time proxy for active construction investment, months before official statistics catch up. The method is specific, repeatable and openly auditable, though it misses underground work and flatters stalled sites.

Sensors

Maxar WorldView-3: 31 cm panchromatic resolution at nadir, 8-band multispectral at 1.24 m. The panchromatic channel resolves crane jibs, counterweights and mast sections clearly enough for automated object detection. Revisit over a given city is roughly 1 to 4.5 days depending on latitude and tasking priority.
Airbus Pléiades Neo: 30 cm native panchromatic resolution, 4 or 6 spectral bands at 1.2 m. Tri-stereo collection mode allows building-height estimation alongside crane detection, which helps distinguish tower cranes from lattice masts and antenna structures. Revisit is up to twice daily at mid-latitudes.
Planet SkySat: 50 cm panchromatic, 1 m multispectral. Slightly coarser than WorldView-3 or Pléiades Neo but the constellation supports near-daily tasking of specific cities, which is valuable for tracking crane arrival and removal events. Video mode captures jib rotation, a direct indicator of operational status.
BlackSky: Approximately 1 m resolution optical, with a scheduling architecture designed for high-revisit tasking of fixed points. Useful for monitoring a defined list of sites at high cadence rather than wide-area census sweeps.

Why a crane is a better leading indicator than a building permit

A building permit is a statement of intent. A tower crane is a capital commitment. Erecting a Potain MDT 389 or a Liebherr 630 EC-H costs money, requires a foundation, and signals that a developer has drawn down finance and mobilised a contractor. Permits can sit unexercised for years. Cranes cannot.

Official construction output statistics in most jurisdictions are published with a lag of two to six months and are subject to substantial revision. Satellite-derived crane indices can be updated as frequently as weekly for cities with reliable clear-sky windows, giving investors and planning authorities a signal that precedes the official data by a full quarter or more. The analogy to shipping AIS data is instructive: both substitute a physical observable for a reported one.

What the imagery actually shows, and how detectors find it

At 30 to 50 cm resolution, a tower crane's mast, jib and counterjib are geometrically distinctive. The mast is a vertical linear feature, typically 1.5 to 3 m wide in cross-section, and the horizontal jib extends 30 to 80 m. In panchromatic imagery the shadow cast by the jib is often more reliably detected than the structure itself, particularly when the crane is surrounded by scaffolding or cladding. Shadow geometry also encodes approximate jib length and height, which can be used to classify crane size.

Automated detection typically applies convolutional neural network classifiers trained on labelled sub-metre imagery, with post-processing steps to suppress false positives from antenna masts, construction hoists and lattice towers. Published research on object detection in VHR optical imagery, including work using WorldView and Pléiades data, reports precision and recall figures in the 85 to 95 per cent range for tower cranes in open urban settings. Performance drops in dense historic city centres where cranes are partially occluded by adjacent buildings, and in cities where flat-roofed construction means cranes are surrounded by parapets that obscure the base.

Distinguishing an operational crane from a dormant one is harder. Jib orientation changes between collection dates are the most reliable indicator: if a crane's jib has rotated between two images taken days apart, the site is active. SkySat video mode, which captures roughly 90 seconds of footage at 30 frames per second, can observe jib movement directly. A crane with an unchanged jib bearing across multiple revisits is a candidate stalled site, though wind loading can rotate an unloaded jib without any human activity.

Building an index: from point detections to economic signal

Raw crane counts per image are noisy. Cloud cover creates gaps; off-nadir angles change apparent crane density; seasonal haze affects some cities for months at a time. A credible index smooths across these artefacts by combining multiple collections per period, flagging cloud-contaminated pixels, and normalising for the observable footprint rather than the nominal city boundary.

The resulting time series can be aggregated at several scales: individual development zones, administrative districts, metropolitan areas or entire countries. Cross-city comparison requires care because crane density per square kilometre of urban area is a function of construction typology. A city building primarily mid-rise reinforced concrete frames will show different crane density than one building steel-frame high-rises, even at identical levels of investment. Analysts should publish the normalisation methodology alongside the index values, not just the headline number.

Investors typically use crane indices as one input alongside planning application data, cement and steel import figures, and satellite-derived nighttime light intensity. No single proxy is sufficient. The crane index is strongest at detecting the onset and cessation of major construction phases; it is weaker at estimating the monetary value of work in progress.

The honest limits of counting cranes from orbit

Tower cranes are used primarily for above-ground structural work. Underground construction, including basement excavation, tunnelling, piling and foundation work, generates no crane signal at all. A city investing heavily in metro expansion or deep basement car parks will be systematically undercounted. Conversely, a stalled development where the developer has defaulted but the crane remains erected, because dismantling it costs money, will be counted as active until the crane is eventually removed. Both errors are real and persistent.

Interior fit-out, mechanical and electrical installation, and facade work after the structure is topped out are also invisible to crane counting. A building that is 80 per cent complete by value may have already lost its crane. This means crane counts track the structural phase of construction, not the full investment cycle. For sectors such as data centre construction, where fit-out is a large fraction of total cost, the undercount is particularly significant.

Cloud cover is a practical constraint in tropical cities. Singapore, Lagos and Jakarta can experience cloud cover rates above 70 per cent on any given day. Achieving a clean monthly composite requires either a large number of collection attempts or acceptance of temporal gaps. SAR imagery is cloud-independent but current SAR resolution, even from Capella Space or ICEYE at roughly 25 to 50 cm spotlight mode, makes crane detection substantially harder than in optical data.

Who uses crane indices and for what decisions

Macro investors and real estate funds use city-level crane counts to track construction cycles in markets where official data is slow or unreliable. The approach has been applied publicly to cities including London, Sydney, Toronto and several Chinese tier-one cities, where permit data is opaque. Planning authorities use site-level detections to cross-check whether permitted developments are proceeding on schedule and to identify unpermitted structures.

Central banks and statistical offices in several countries have explored crane indices as experimental high-frequency indicators for GDP nowcasting, following the broader literature on satellite-derived economic proxies. The method is most useful in economies where construction represents a large share of fixed capital formation and where official statistics have known quality issues.

Satellize can run crane detection and activity indexing across defined city portfolios on a recurring basis, drawing on tasked commercial imagery and delivering structured outputs to client systems. The analytics pipeline follows the same object-detection and time-series methodology described here.

Typical figures

Typical spatial resolution (panchromatic)	30 cm (WorldView-3, Pléiades Neo) to 50 cm (SkySat); minimum for reliable crane detection is approximately 50 cm
Revisit cadence	1 to 4.5 days (WorldView-3, depending on latitude); up to twice daily (Pléiades Neo); near-daily for fixed targets (SkySat, BlackSky)
Minimum detectable target	Tower crane mast cross-section approximately 1.5 m; jib reliably resolved at 30 to 50 cm imagery; small hoists and mobile cranes frequently missed
Cloud sensitivity	Optical only; cloud cover above 30 per cent per scene significantly degrades detection; tropical cities may require 10 or more collection attempts per monthly composite
Spectral bands used	Panchromatic primary; multispectral (red, NIR) used to suppress false positives from vegetation and water features
Archive depth	WorldView-3 archive from 2014; Pléiades from 2012; SkySat from approximately 2016; allows retrospective index construction for most major cities
Detection accuracy (published range)	Precision and recall 85 to 95 per cent for tower cranes in open urban settings; lower in dense historic centres or where cranes are partially occluded
Typical update latency	24 to 72 hours from image acquisition to detection output, depending on processing pipeline and cloud screening
Delivery formats	GeoJSON point layer per crane detection; CSV time-series index per zone; GeoTIFF annotated imagery; API feed for integration with financial data platforms

Analytics Satellize can run

City-wide crane census map	Convolutional neural network object detection applied to pan-sharpened VHR optical imagery, with shadow-geometry post-processing to reduce false positives	GeoJSON point layer with crane location, estimated jib length class and detection confidence score; updated per new clear-sky collection
Active versus dormant classification	Jib-bearing change detection between sequential collections; sites with no angular change across three or more revisits flagged as potentially stalled	Attributed point layer with activity status field; monthly summary table of active, dormant and newly erected or removed cranes per zone
Construction activity index time series	Normalised crane count per observable urban area per period, smoothed across cloud-affected gaps using linear interpolation with uncertainty bounds	CSV or JSON time-series feed per city or district, with methodology notes and cloud-cover quality flags per observation
Crane arrival and removal event alerts	Change detection between consecutive collections identifying new crane presences or absences at previously surveyed coordinates	Alert feed (email or API) triggered within 48 hours of a confirmed new erection or removal event at monitored sites
Cross-city comparative index	Standardised crane density per square kilometre of urban fabric, normalised by construction typology class derived from building footprint data	Quarterly comparative report covering client-specified city portfolio, with ranked index table and annotated maps
Stalled-site identification report	Persistent crane presence combined with absence of visible site activity (no material movement, unchanged scaffolding) across a defined observation window, typically 60 to 90 days	Flagged site list with supporting imagery chips and observation timeline, formatted for planning authority or credit-risk review workflows

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.