Disinformation rebuttal using dated satellite scenes
Timestamped satellite imagery, combined with shadow-angle geometry, crop phenology and building-footprint comparison, gives analysts a physics-grounded method for refuting false claims about where and when events occurred.
Sensors
- Maxar WorldView-3: 30 cm panchromatic resolution, 1.24 m multispectral. Shadow geometry is legible at this resolution to within a few minutes of solar time. Archive extends to 2014, giving years of pre-event baseline. Revisit approximately 1 day at mid-latitudes with off-nadir tasking.
- Airbus Pléiades Neo: 30 cm native resolution, four-satellite constellation achieving same-day revisit over most targets. Stereo and tri-stereo modes allow 3-D building-height extraction, which strengthens footprint comparison against claimed before/after states.
- Planet SkySat: 50 cm resolution, video-capable. Useful for confirming the presence or absence of specific vehicles or structures at a precise moment. Archive from 2016 onward. Revisit is on-demand via tasking, typically hours within a priority window.
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator (more frequent at higher latitudes with two satellites). Too coarse for individual-building analysis but well-suited to crop-state seasonal anchoring across wide agricultural areas, and freely available with a consistent metadata chain.
Why physics makes a better witness than a press release
A photograph or video can be fabricated, misdated or mislocated. A satellite scene cannot easily be forged without leaving detectable inconsistencies in its metadata, acquisition geometry and spectral signature. The sun moves on a predictable arc governed by date, time and latitude. If a claimed event occurred on a specific date, the shadows in any genuine image taken that day must fall at the angle dictated by the solar elevation and azimuth for that place and moment. Disagreement between the claimed date and the measured shadow angle is not ambiguous: it is a falsification.
This is the foundation of open-source satellite rebuttal work. Bellingcat's analysts applied shadow-angle calculations to geolocate and date imagery from the MH17 shoot-down investigation and subsequent Buk missile convoy tracking in Ukraine. The New York Times Visual Investigations team used the same principle, combined with building-footprint comparison, to document the destruction of Mariupol. The method is not new, but the availability of sub-metre commercial archive imagery has made it far more precise and far faster to apply.
Shadow-angle dating: what the calculation actually involves
Solar azimuth and elevation at any point on Earth at any moment are deterministic functions of latitude, longitude, date and time. Given a satellite image with known acquisition coordinates, an analyst measures the length and direction of shadows cast by objects of known or estimable height. The solar elevation angle is the arctangent of the object height divided by shadow length. The azimuth is read directly from the shadow direction. Both values are then compared against ephemeris tables or standard solar-position algorithms (the NREL SPA algorithm is publicly documented and widely used) for the claimed date and time.
Precision depends on resolution. At 30 cm, a shadow cast by a two-storey building can be measured to within a metre or two, yielding a solar elevation estimate accurate to roughly one degree, which corresponds to a time uncertainty of a few minutes at most solar angles. At Sentinel-2's 10 m, the same calculation carries an uncertainty of tens of minutes, which is still sufficient to rule out a claimed date that is weeks or seasons away from the true one. The honest limit is near the solstices, when solar elevation changes slowly day-to-day; a winter image taken anywhere within a two-week window around the solstice may be indistinguishable by shadow angle alone, requiring corroborating evidence.
Crop state and phenology as a seasonal anchor
Agricultural fields cycle through recognisable spectral states: bare soil after harvest, green emergence, canopy closure, ripening, senescence, stubble. Sentinel-2's near-infrared band makes the normalised difference vegetation index (NDVI) straightforward to compute, and NDVI time-series for most agricultural regions are available in the public archive back to 2015. If a claimed event is said to have occurred in, say, late October, but the surrounding wheat fields in the imagery show the NDVI signature of mid-June heading, the date claim fails.
This technique was applied publicly during the Syria conflict to anchor imagery to harvest seasons when other metadata was stripped or disputed. Its limitation is specificity: crop phenology varies by cultivar, irrigation and local climate, so a broad seasonal anchor (spring versus autumn) is reliable while narrowing to a specific week requires local ground-truth or comparison against multi-year Sentinel-2 time-series for the exact field. Satellize's crop-estimation work in Tonga, which tracks coconut and root-crop phenology using Sentinel-2 time-series, uses the same underlying NDVI methodology, which transfers directly to this forensic context.
Building-footprint comparison and the limits of change detection
Comparing a building's footprint, roofline and shadow profile across two dates establishes whether a structure existed, was damaged, or was demolished within that interval. At Pléiades Neo or WorldView-3 resolution, analysts can distinguish a standing building from a collapsed one, identify rubble scatter patterns, and detect new construction. This is the spatial equivalent of a before-and-after photograph, except the dates are cryptographically recorded in the image metadata by the operator.
The honest caveat is that commercial imagery is not continuous. A gap of several days between available scenes leaves a window of ambiguity about exactly when a change occurred. Cloud cover compounds this: optical sensors see nothing through thick cloud, and in high-latitude winters or tropical wet seasons, usable acquisitions may be separated by weeks. SAR sensors (Sentinel-1, for instance) penetrate cloud but operate at 5 m resolution in interferometric wide-swath mode, which is sufficient for large-structure change but not for detailed footprint comparison. A credible rebuttal therefore states its temporal uncertainty explicitly rather than claiming precision it cannot demonstrate.
Chain of custody: where the argument can be attacked
The entire evidentiary weight of satellite-based rebuttal rests on the integrity of the image metadata. Commercial operators record acquisition time, sensor ID, orbit parameters and ground-processing provenance in standardised metadata files. Maxar, Airbus and Planet all maintain these records and can, under appropriate legal or governmental arrangements, provide certified copies. Sentinel-2 data is distributed by ESA through the Copernicus Data Space Ecosystem with checksums and processing-level provenance that are independently verifiable.
An adversary seeking to undermine a rebuttal will attack the chain of custody first: claiming the image was sourced from an unverified third party, that metadata was altered, or that the analyst misidentified the location. The methodological response is to cross-reference at least two independent sensors covering the same event window, to publish the raw measurement steps (shadow length in pixels, scale derived from known reference objects, solar-position calculation inputs), and to use only imagery retrieved directly from operator archives rather than social-media screenshots. Reproducibility is the rebuttal's defence. Any analyst who cannot show their working is not doing forensic analysis; they are doing advocacy.
Practical constraints a buyer should understand before commissioning this work
Tasking a commercial satellite over a contested area during active conflict is not always possible. Operators assess safety, legal jurisdiction and export-control obligations before accepting a tasking order. Archive searches are usually unrestricted, but the archive may simply not contain a useful acquisition if the area was not previously of commercial interest. Coverage of rural Syria in 2012 and 2013 was sparse in the commercial archive; analysts often had to work with whatever Landsat or early Sentinel-2 scenes existed, accepting the resolution constraints that came with them.
Turnaround time matters enormously in a disinformation context. A false narrative that circulates for 72 hours before being rebutted has already shaped opinion. Rapid tasking, archive triage and automated shadow-measurement tools can compress the analytical cycle to under 24 hours for straightforward cases, but complex multi-scene, multi-date comparisons involving crop phenology and 3-D footprint reconstruction may take several days of skilled analyst time. Budget accordingly, and treat the first 48 hours after a disputed event as the critical window.
Typical figures
| Best available spatial resolution (optical) | 30 cm (Maxar WorldView-3, Airbus Pléiades Neo) |
| Shadow-angle time uncertainty at 30 cm | Approximately ±5 minutes of solar time under good conditions |
| Shadow-angle time uncertainty at 10 m (Sentinel-2) | Approximately ±30–60 minutes; sufficient to rule out seasonal misdating |
| Revisit (commercial tasking, on-demand) | Sub-daily to 1 day (WorldView-3, Pléiades Neo, SkySat) at most latitudes |
| Revisit (Sentinel-2, free archive) | 5 days at equator; 2–3 days at 45° latitude with both satellites |
| Archive depth (commercial) | WorldView-3 from 2014; SkySat from 2016; Pléiades from 2011 |
| Archive depth (Sentinel-2) | From mid-2015 (Sentinel-2A); 2017 onward with both satellites |
| Cloud penetration | None for optical sensors; Sentinel-1 SAR penetrates cloud at 5–20 m resolution |
| Metadata provenance | ISO 19115-compliant; ESA Copernicus data carries checksums; commercial operators provide certified acquisition records on request |
| Minimum detectable structural change | Individual building collapse detectable at 30 cm; large-structure change at 5 m SAR |
Analytics Satellize can run
| Shadow-angle date and time estimate | Solar position algorithm (NREL SPA or equivalent) applied to measured shadow vectors in georeferenced imagery | PDF report with annotated image, measurement table, solar-ephemeris comparison, and stated uncertainty bounds |
| Crop-phenology seasonal anchor | NDVI time-series from Sentinel-2 archive compared against multi-year phenological baseline for the specific field location | Time-series chart with annotated acquisition dates, seasonal classification, and confidence interval for the event window |
| Building-footprint before/after comparison | Object-based change detection on co-registered sub-metre optical pairs; manual QA by trained analyst | GIS layer (GeoPackage or shapefile) of changed footprints with acquisition dates, plus annotated image pair |
| Multi-sensor corroboration package | Cross-referencing optical and SAR acquisitions over the same target window to bound the change interval and reduce single-source dependency | Consolidated evidence report with independent acquisition metadata from at least two sensor systems |
| Geolocation verification | Feature matching against open-source base imagery (Google Earth historical, Sentinel-2 mosaic) using identifiable permanent structures as control points | Annotated map confirming or disputing the claimed event location, with pixel-level correspondence markers |
| Chain-of-custody documentation | Provenance logging of all imagery sources, download timestamps, checksums and processing steps in a reproducible audit trail | Signed provenance log suitable for submission to legal, governmental or journalistic review processes |
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.