Urban electricity grid infrastructure mapping from imagery
Sub-metre optical and SAR imagery can locate transmission towers, substations and overhead distribution lines with enough precision to supplement or challenge an ageing asset register. Underground cables remain invisible to optical sensors, and the confusion between transmission towers and mobile masts is a real operational problem.
Sensors
- Maxar WorldView Legion: 30 cm panchromatic resolution; up to 15 revisits per day over a target city. Sufficient to resolve individual tower legs, insulator strings and substation bus-bar layouts. The high revisit rate makes it practical to schedule cloud-free collection within a short campaign window.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral, four-satellite constellation. Tri-stereo collection mode generates a digital surface model alongside the orthoimage, which helps distinguish tower height from shadow length when sun angle is ambiguous.
- Planet SkySat: 50 cm resolution, tasked video or still. Useful for rapid, lower-cost surveys of distribution-line corridors where 30 cm detail is not strictly required. Less suited to resolving tower structure in dense urban canopy.
- COSMO-SkyMed (X-band SAR): Spotlight mode delivers roughly 1 m resolution. Metallic lattice towers produce strong, stable double-bounce and specular backscatter signatures independent of cloud cover or sun angle. Useful for confirming tower presence in persistently cloudy climates and for detecting structures hidden under tree canopy.
What a shadow gives away about a tower
At sub-metre resolution, a lattice transmission tower casts a shadow whose length is a direct function of tower height and solar elevation angle at acquisition time. Both quantities are known precisely. A 40-metre tower at a solar elevation of 45 degrees casts a 40-metre shadow; the geometry is unambiguous. Object-detection models trained on labelled WorldView and Pléiades imagery exploit this relationship: the shadow is often easier to detect than the tower itself, particularly where the tower base sits on a concrete pad that blends with surrounding paving.
Substation compounds are more straightforward. Their characteristic footprint, the combination of a fenced rectangular perimeter, transformer pads, bus-bar structures and access roads, is distinctive enough that even 50 cm imagery supports reliable automated detection. The principal difficulty is not detection but classification: distinguishing a 132 kV substation from a 33 kV switching station requires either ground-truth labelling or inference from the number and spacing of transformer bays.
The mast problem: when a tower is not a tower
Mobile-phone masts and transmission towers share enough visual characteristics at medium resolution to cause systematic confusion. Both are tall, metallic, lattice or tubular structures with guy wires or cross-arms. At 30 cm resolution the differences become resolvable: transmission towers carry multiple conductor attachment points and large insulator strings; masts carry antenna panels and cable runs. Even so, automated classifiers trained on one geography often fail in another because tower designs vary substantially by country and era.
The practical mitigation is to run object detection as a first pass for candidate metallic structures, then apply a secondary classification step that uses aspect ratio, cross-arm geometry and proximity to known road or power corridors. Proximity to an existing partial asset register, even an incomplete one, dramatically improves precision. A detected structure 50 metres from a mapped 33 kV line is almost certainly a distribution pole; the same structure 2 km from any mapped infrastructure warrants field verification.
Wind turbines introduce a third class of confusion in peri-urban areas. Their shadow pattern is distinctive, but in a first-pass detection run they appear as tall metallic objects with a regular spacing. Filtering by shadow shape and the circular cleared-ground footprint resolves most cases.
Overhead lines: detectable but demanding
Transmission conductors at 30 cm resolution appear as thin, nearly one-pixel-wide lines with a characteristic catenary sag. Detection relies on line-segment extraction algorithms, typically variants of the Hough transform or deep-learning edge detectors, followed by catenary-curve fitting to confirm the sag profile. High-voltage lines at 220 kV or above are typically easier to detect because the conductor bundle is wider and the tower spacing is longer, giving a cleaner segment.
Distribution lines at 11 kV or 33 kV, strung on wooden poles at 50 to 80 metre intervals, are at the limit of reliable automated extraction in sub-metre imagery. Urban tree canopy interrupts the line segment; parked vehicles and building overhangs create false gaps. Extraction completeness in dense urban areas rarely exceeds 70 to 80 percent without manual editing. That figure is worth stating plainly to any client expecting a complete distribution-line inventory from imagery alone.
Underground cables: the honest limit of optical remote sensing
Optical imagery cannot see underground cable routes. Full stop. Ground-penetrating radar, cable-locator surveys and utility records remain the only reliable methods for underground infrastructure. SAR interferometry can detect millimetre-scale surface deformation caused by trench settlement over buried cables, but the signal is subtle, slow to develop and easily masked by other deformation sources in active urban environments.
The practical implication for asset-register projects is that imagery-derived maps will systematically under-represent underground networks, which are disproportionately common in city centres and newer residential developments. A grid-mapping programme that relies solely on imagery will produce an accurate picture of the overhead network and a significant gap where underground cabling dominates. Clients should plan for ground-truth campaigns in those zones rather than assume imagery will fill them.
SAR as a cloud-independent cross-check
COSMO-SkyMed in Spotlight mode, and to a lesser extent Sentinel-1 in Interferometric Wide Swath mode at 10 m resolution, provide a complementary detection layer. Metallic lattice structures produce a strong and repeatable backscatter signature in X-band and C-band SAR because the corner-reflector geometry of the lattice amplifies the radar return. This makes SAR particularly useful in tropical cities where persistent cloud cover limits optical collection to a handful of usable scenes per year.
Sentinel-1's 10 m resolution is too coarse to resolve individual distribution poles, but it can confirm the presence of large transmission towers and substation compounds. Change detection between two SAR acquisitions, separated by weeks or months, can flag new structures that were not present in the baseline, which is a practical way to monitor grid-expansion construction without tasking expensive commercial sensors on every revisit.
From detected assets to a usable register
The output of an imagery-based grid survey is a candidate asset list, not a certified asset register. Every detected tower carries a confidence score, a location in decimal degrees, an estimated height derived from shadow geometry or stereo DSM, and a structure-type classification. That list needs to be reconciled against whatever utility records already exist. In practice, the most valuable finding is often not the towers the imagery confirms but the ones it finds that are absent from the register, or the registered assets for which no imagery evidence can be found.
Satellize has applied similar object-detection and change-detection pipelines in agricultural contexts, including the Tonga crop-estimation programme, and the underlying workflow transfers directly to linear infrastructure. The grid-mapping analytics stack produces a GeoPackage or Esri File Geodatabase of detected assets, a confidence-graded line layer for overhead conductors, and a change report flagging discrepancies against a client-supplied baseline register. Those outputs feed directly into outage-risk models and grid-expansion planning tools without requiring intermediate manual digitising.
Typical figures
| Best optical spatial resolution | 30 cm (WorldView Legion, Pléiades Neo panchromatic) |
| SAR resolution (metallic-structure detection) | ~1 m (COSMO-SkyMed Spotlight); 10 m (Sentinel-1 IW) |
| Revisit for tasked optical | Up to 15 times/day (WorldView Legion over target city); 1–2 days (Pléiades Neo) |
| Sentinel-1 SAR revisit | 6 days at equator; 1–3 days at mid-latitudes with both satellites |
| Minimum detectable tower height (shadow method) | ~10 m at 30 cm resolution and solar elevation above 20 degrees |
| Distribution-line extraction completeness (dense urban) | 70–80% without manual editing; canopy and clutter are the limiting factors |
| Underground cable detection | Not possible with optical imagery; SAR deformation proxy is experimental only |
| Archive depth (commercial optical) | WorldView series from 2007; Pléiades from 2011 |
| Delivery format | GeoPackage, Esri File Geodatabase, GeoJSON; confidence scores per feature |
| Positional accuracy (orthorectified sub-metre imagery) | Typically 1–3 m CE90 with ground control; 3–5 m without |
Analytics Satellize can run
| Transmission tower inventory | Shadow-geometry analysis combined with convolutional object-detection model trained on labelled sub-metre imagery | Point GIS layer with tower location, estimated height, structure-type classification and detection confidence score |
| Substation compound mapping | Footprint segmentation using high-resolution optical imagery; compound boundary, transformer-pad count and access-road extraction | Polygon layer of substation compounds with attribute table of internal feature counts |
| Overhead line corridor extraction | Line-segment detection (Hough transform or deep-learning edge detector) with catenary-curve fitting for conductor confirmation | Polyline layer of detected conductors with confidence grade and voltage-class inference where tower spacing allows |
| Asset-register discrepancy report | Spatial join of imagery-derived detections against client-supplied register; flagging of unmatched detections and registered assets with no imagery evidence | Tabular report and map of discrepancies, prioritised by asset class and confidence |
| New construction change detection | Bi-temporal SAR backscatter change detection (Sentinel-1 or COSMO-SkyMed) to flag new metallic structures between baseline and current acquisition | Change-alert GIS layer with acquisition dates and backscatter-change magnitude |
| Mast versus tower classification | Secondary classifier using aspect ratio, cross-arm geometry and proximity-to-network features applied to candidate metallic structures from first-pass detection | Reclassified point layer with structure-type probability distribution per candidate |
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.