Utility-scale solar farm installed-capacity verification for green-bond audits
Satellite imagery at 30–50 cm resolution can map the physical extent of photovoltaic arrays and convert that area into an independent capacity estimate, giving green-bond auditors a check that requires no site visit and no cooperation from the asset owner.
Sensors
- Pleiades Neo: 30 cm panchromatic, 50 cm multispectral (blue, green, red, red-edge, NIR), daily revisit over most latitudes. The resolution is sufficient to distinguish individual panel rows and resolve tracker gaps, which is the critical measurement for area calculation.
- WorldView-3: 31 cm panchromatic, 1.24 m multispectral, plus 8 SWIR bands at 3.7 m. The SWIR bands are particularly useful: silicon PV panels show a strong absorption feature between 1.55 and 2.19 µm that separates them from most spectrally similar surfaces such as dark roofing or standing water.
- Planet SuperDove: 3 m resolution, 8 spectral bands including red-edge and NIR, near-daily global revisit. Resolution is insufficient to map individual panels but adequate to delineate array boundaries and detect inter-acquisition changes in installed area at large facilities.
- Sentinel-2 MSI: 10 m in visible and NIR bands, 20 m in SWIR, free archive from 2015. Useful for detecting new construction, confirming facility footprint and cross-checking area estimates from commercial imagery. Cannot resolve panel rows; minimum detectable array is roughly 1–2 hectares.
What the spectral signature of silicon gives away
Crystalline silicon photovoltaic panels have a distinctive optical fingerprint. In the visible spectrum they absorb heavily, appearing very dark. In the near-infrared, where vegetation reflects strongly, panels remain low-reflectance. In the shortwave infrared, particularly the 1.55–1.75 µm and 2.08–2.35 µm windows captured by WorldView-3 SWIR bands and Sentinel-2 bands 11 and 12, silicon shows a characteristic absorption that most competing land-cover types do not replicate. This combination of low NIR and low SWIR reflectance is the basis for automated panel mapping.
Thin-film technologies (cadmium telluride, CIGS) have slightly different spectral profiles from crystalline silicon, which matters for classification confidence. A well-designed classifier trained on labelled imagery from known facilities can separate panel types with reasonable accuracy at 30–50 cm resolution, though mixed-technology parks introduce ambiguity that should be disclosed in any audit report.
From panel pixels to megawatts: the conversion chain
Mapping panel area is the satellite's contribution. Converting area to capacity requires published engineering priors, and that is where honest uncertainty accounting becomes essential.
The standard conversion runs as follows. Mapped panel area is divided by the area per panel for the technology type (typically 1.6–2.0 m² for a 60- or 72-cell crystalline module). Panel count multiplied by rated module efficiency, expressed as watts per square metre of panel area, gives DC nameplate capacity. Published module efficiencies for commercial crystalline silicon range from roughly 17% to 23% as of recent product generations. A facility-level DC-to-AC ratio (the inverter loading ratio) of 1.2–1.4 is typical for utility-scale plants, though it varies by design. The result is an estimated AC capacity with an uncertainty range that honestly reflects the range of plausible inputs.
For a green-bond audit, the satellite estimate does not need to match the claimed figure exactly. It needs to be close enough to confirm the asset exists at the stated scale, or wide enough in its confidence interval to flag a discrepancy worth investigating on the ground. A 10–15% uncertainty band on capacity is achievable from imagery alone; tighter bounds require ground-truth data on the specific modules installed.
What orbit cannot see
Panel area is observable. Generating output is not. Soiling, shading from adjacent rows, inverter clipping, degradation over time and grid curtailment all reduce actual generation below what the panel area implies, and none of these are detectable from a single optical acquisition.
Single-axis trackers complicate area mapping directly. A tracker array imaged at solar noon looks different from the same array imaged at 09:00, because panel tilt changes the projected area visible to the satellite. This is not a fatal problem: acquisition geometry is known precisely, and tilt can be corrected if the tracker angle at acquisition time is available. But if it is not, the analyst must flag the ambiguity. Bifacial panels on trackers add a further complication because the rear surface contributes to generation in ways that are entirely invisible from above.
Cloud is the operational constraint for optical methods. Major solar facilities tend to be sited in high-irradiance, low-cloud regions, which helps. But a single cloudy acquisition over a facility under construction can delay verification by days or weeks unless archive imagery is available. SAR can penetrate cloud but does not provide the spectral discrimination needed for panel mapping; it is useful for detecting construction activity and ground disturbance, not for capacity estimation.
How auditors should use the output
A satellite-derived capacity estimate is an independent physical check, not a replacement for engineering documentation. The appropriate use in a green-bond audit is triangulation: does the satellite-mapped area support the claimed DC nameplate capacity within a declared confidence interval? If the mapped area implies a capacity 30% below the claimed figure, that is a material discrepancy requiring explanation. If it implies a capacity within 12% of the claim, that is broadly consistent and the residual uncertainty is attributable to module-efficiency and DC/AC ratio assumptions.
Archive depth matters for construction-stage verification. Sentinel-2 imagery is available from 2015, and commercial archives from WorldView and Pleiades extend back roughly a decade. This means an auditor can reconstruct the construction timeline of a facility, confirm that panels were installed before the bond issuance date, and detect any subsequent decommissioning or partial removal. That temporal record is often more valuable than a single snapshot.
Satellize structures solar-capacity verification as a repeatable annual check, suitable for ongoing sustainability-linked loan covenants, rather than a one-off audit product. The methodology is documented in the deliverable so that a client's own auditors can interrogate the assumptions.
Minimum detectable scale and practical thresholds
At Sentinel-2's 10 m resolution, the practical minimum for reliable array delineation is roughly 1–2 hectares of continuous panel area, corresponding to approximately 1–2 MW of installed capacity depending on technology. Below that, adjacent land cover begins to contaminate the spectral signal.
At 50 cm resolution from Pleiades Neo or WorldView-3, individual panel rows are resolvable and the minimum detectable array shrinks to a few hundred square metres. For utility-scale facilities, which typically exceed 10 MW and cover tens of hectares, the resolution question is largely moot: the facility is easily delineated even at Sentinel-2 resolution. The commercial sensors earn their cost by providing the panel-row geometry needed for precise area calculation and technology classification, not simply by detecting the facility's existence.
Typical figures
| Spatial resolution (panel mapping) | 30–50 cm (Pleiades Neo, WorldView-3); 3 m (Planet SuperDove); 10 m (Sentinel-2 visible/NIR) |
| Spectral bands used | Visible (RGB), NIR, red-edge, SWIR 1 (1.55–1.75 µm), SWIR 2 (2.08–2.35 µm) |
| Revisit (commercial tasking) | Daily or near-daily for Pleiades Neo and Planet; WorldView-3 typically 1–4 days depending on latitude and tasking priority |
| Archive depth | Sentinel-2 from 2015 (free); commercial archives from approximately 2008–2014 depending on sensor |
| Minimum detectable array (Sentinel-2) | Approximately 1–2 hectares (roughly 1–2 MW) |
| Minimum detectable array (50 cm optical) | A few hundred square metres; individual panel rows resolved |
| Capacity estimation uncertainty | Typically ±10–15% from imagery alone, driven by module efficiency and DC/AC ratio assumptions |
| Cloud limitation | Optical methods blocked by cloud; archive tasking mitigates but cannot eliminate delay |
| Delivery formats | GeoTIFF panel-extent masks, GeoJSON facility polygons, PDF audit report with methodology annex |
| Latency (tasked acquisition to delivery) | 3–7 business days for standard verification; expedited available |
Analytics Satellize can run
| Panel-extent map | Supervised spectral classification using NIR and SWIR band ratios, trained on labelled PV imagery; morphological filtering to remove sub-pixel noise | GeoTIFF raster and GeoJSON polygon layer of mapped panel area with per-class confidence scores |
| DC nameplate capacity estimate | Panel area multiplied by published module-efficiency priors (17–23% range for crystalline silicon) with Monte Carlo uncertainty propagation across efficiency and DC/AC ratio inputs | Capacity estimate with 90% confidence interval, tabulated in PDF audit annex |
| Construction timeline reconstruction | Change detection across Sentinel-2 and commercial archive imagery; NDVI suppression and bare-ground expansion used as construction-onset indicators | Annotated time-series chart showing installation progress by quarter, suitable for bond-issuance date verification |
| Technology-type classification | SWIR spectral unmixing to distinguish crystalline silicon from thin-film signatures where WorldView-3 SWIR bands are available | Technology-class attribution table with confidence rating; flagged where ambiguity exceeds threshold |
| Annual re-verification check | Repeat panel-extent mapping against baseline; area-change detection flags decommissioning, partial removal or expansion | Annual comparison report with delta map, suitable for sustainability-linked loan covenant monitoring |
| Tracker-tilt correction | Solar geometry calculation at acquisition timestamp applied to known tracker-axis orientation; projected area corrected to horizontal equivalent | Corrected area figure with tilt-correction methodology note; flagged as uncertain where tracker angle is unconfirmed |
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.