Nuclear facility operational-status change detection from open imagery
Multispectral and thermal open imagery can reveal operational changes at nuclear facilities, from cooling-tower plume behaviour to vehicle surges, without access to classified data. Surface signatures are real; subsurface enrichment is not.
Sensors
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral, 3.7 m SWIR (8 bands). SWIR bands distinguish water vapour from dry exhaust and can characterise rooftop material changes. Revisit roughly 1 to 4.5 days depending on latitude and tasking priority. Commercial tasking required; not freely archived.
- Landsat 8 and 9 TIRS: Thermal Infrared Sensor captures two bands centred at 10.9 µm and 12.0 µm at 100 m resolution (resampled to 30 m in products). Detects elevated surface and water temperatures from reactor cooling-water discharge. 16-day repeat per satellite; combined 8-day revisit. Free archive back to 1972 (Landsat 1 MSS) for historical baselining.
- Sentinel-2 MSI: 10 m visible and near-infrared, 20 m red-edge and SWIR. No thermal band, but SWIR band 11 (1.61 µm) and band 12 (2.19 µm) are sensitive to surface temperature extremes and can detect hot effluent in water bodies under favourable conditions. 5-day revisit at mid-latitudes with two satellites. Free and open.
- Planet SuperDove: 3 m resolution, 8-band multispectral (coastal blue through NIR). No thermal capability, but daily or near-daily revisit makes it the primary sensor for vehicle counting, construction-activity sequencing and plume-shadow geometry analysis. Commercial licence required.
- VIIRS (Suomi NPP / NOAA-20): 375 m day-night band and thermal channels. Useful for detecting gross thermal anomalies and night-light changes at facility level, particularly for facilities in regions with poor commercial tasking coverage. Free and near-real-time via NASA FIRMS.
What the surface gives away
A nuclear reactor cannot hide its waste heat. Every thermal megawatt rejected through a cooling tower or a river-water discharge circuit leaves a signature that passive infrared sensors can read from orbit. The physics is straightforward: light-water reactors typically reject roughly two-thirds of their thermal output to the environment. At a 1,000 MWe plant, that is around 2,000 MWt going somewhere visible. Landsat TIRS has detected river-temperature anomalies of 1 to 3 degrees Celsius above ambient at distances of several kilometres downstream from operational plants in published studies.
Cooling towers add a second observable: the condensation plume. Plume presence, height and direction are functions of reactor power level and ambient meteorology. A facility running at reduced power or in cold shutdown produces a smaller or absent plume. Analysts at 38 North and the James Martin Center for Nonproliferation Studies have used precisely this signature, cross-referenced against wind and temperature data, to infer operational status at facilities in North Korea and Iran where no inspector has access.
Vehicle traffic and construction as leading indicators
Thermal signatures confirm operation. Vehicle and construction signatures often precede it. A surge in heavy-goods vehicles at a declared facility can indicate fresh fuel delivery or spent-fuel cask movement. A pattern of light vehicles arriving in shifts, consistent with rotating work crews, suggests sustained operations rather than a maintenance outage.
Planet SuperDove's near-daily revisit at 3 m resolution makes vehicle counting tractable. Analysts can build a baseline distribution of vehicle counts per day-of-week and flag statistically significant departures. The method is the same one applied to port activity and logistics depots, adapted to the specific access-road geometry of nuclear sites. Construction activity, new earthworks, or modifications to the security perimeter are detectable at WorldView-3 resolution and can be sequenced using Planet's daily cadence to establish when work began and at what pace it is proceeding.
Thermal infrared through cloud: what is possible and what is not
Thermal infrared penetrates thin cloud and haze better than visible bands, but it does not pass through thick convective cloud or precipitation. Landsat TIRS data is flagged for cloud contamination using the CFMask algorithm; analysts must composite multiple passes to build a reliable thermal baseline. At an 8-day combined revisit for Landsat 8 and 9, a facility in a persistently cloudy region, northern North Korea in winter, for example, may yield only a handful of clean thermal observations per quarter.
VIIRS at 375 m provides more frequent thermal observations but cannot resolve individual buildings or discharge channels at most facilities. The practical workflow is to use VIIRS for rapid anomaly flagging and Landsat TIRS for quantitative temperature retrieval, then cue commercial optical tasking for surface-detail confirmation. No open sensor currently provides high-resolution thermal imagery with daily revisit; that remains a gap.
Honest limits of the open-source method
The methodology is explicitly a surface-signature discipline. Centrifuge halls are underground or inside windowless buildings. Isotopic enrichment levels, fuel fabrication status, and weapons-component machining leave no reliable optical or thermal signature detectable from orbit with open sensors. Claims to the contrary should be treated with scepticism.
Ambiguity is real. A cold reactor may be in planned maintenance rather than shut down under pressure. A vehicle surge may be a safety drill or a fuel-rod inspection rather than a covert transfer. Analysts following the 38 North and CNS methodology explicitly caveat their assessments with alternative explanations and assign confidence levels accordingly. The analytic product is a change indicator and a set of hypotheses, not a verdict. Decision-makers who expect binary confirmation from imagery will be disappointed; those who want a structured, evidence-based update on observable activity will find genuine value.
Building a monitoring programme: archive, baseline, alert
Effective facility monitoring rests on three layers. First, a deep historical archive: Landsat's record back to the 1970s and Sentinel-2's archive from 2015 allow analysts to characterise what normal looks like across seasons, weather conditions and known operational cycles. Second, a statistical baseline: vehicle counts, thermal anomaly frequency distributions, plume-day ratios. Third, an alert layer triggered when current observations depart from the baseline by a defined threshold.
The alert layer is where commercial tasking earns its cost. When a Sentinel-2 or VIIRS pass flags an anomaly, a same-day or next-day WorldView-3 task can confirm or dismiss it at sub-metre resolution before an analyst writes a report. Satellize structures this workflow for government clients, combining open-constellation monitoring with commercial tasking on client licence.
Governments running their own treaty-verification programmes, or think-tanks conducting open-source nuclear risk assessments, should plan for a minimum six-month baseline-building period before alert thresholds are meaningful. Rushing to alerts on a thin baseline is how false positives accumulate and credibility erodes.
Spectral band selection: why SWIR matters beyond the visible
Visible-band imagery tells you a building exists and whether vehicles are present. SWIR adds material discrimination. WorldView-3's eight SWIR bands (1.195 to 2.365 µm) can distinguish concrete from metal cladding, fresh earthwork from stabilised ground, and water from dry surfaces. At a facility where construction is ongoing, SWIR time-series can reveal whether a new structure has a metallic roof consistent with a turbine hall or a concrete roof consistent with a containment structure.
Sentinel-2 SWIR bands 11 and 12 provide a free, lower-resolution version of this capability. They are sufficient for tracking large-scale construction phases and for detecting the thermal signature of warm water in discharge channels under low-cloud conditions. The combination of free Sentinel-2 SWIR with commercial WorldView-3 tasking for confirmation represents the cost-effective standard workflow for sustained facility monitoring.
Typical figures
| Best optical spatial resolution | 31 cm panchromatic (WorldView-3); 3 m multispectral (Planet SuperDove) |
| Thermal spatial resolution | 100 m native / 30 m resampled (Landsat 8/9 TIRS); 375 m (VIIRS) |
| Thermal sensitivity | Landsat TIRS noise-equivalent delta temperature approximately 0.4 K; river-discharge anomalies of 1 to 3 °C detectable in clear conditions |
| Revisit (optical) | Daily to near-daily (Planet SuperDove, VIIRS); 5 days at mid-latitudes (Sentinel-2 two-satellite); 8 days combined (Landsat 8+9) |
| Revisit (thermal) | 8 days combined (Landsat 8+9 TIRS); 12-hour (VIIRS, coarse resolution) |
| Cloud penetration | Thermal IR penetrates thin cloud and haze; fails under thick cloud or precipitation. No open optical or TIR sensor provides all-weather imaging at facility scale. |
| Spectral bands used | Visible (RGB), NIR, SWIR (1.2 to 2.4 µm), TIR (10.9 µm and 12.0 µm) |
| Archive depth | Landsat MSS from 1972; Landsat TM/ETM+/OLI-TIRS from 1982; Sentinel-2 from 2015; Planet from approximately 2016 (commercial) |
| Minimum detectable construction feature | Approximately 1 m linear features at WorldView-3 resolution; approximately 10 m at Sentinel-2 10 m band |
| Delivery formats | Georeferenced GeoTIFF change layers, annotated PDF site reports, GIS-compatible vector overlays, periodic written assessments |
Analytics Satellize can run
| Cooling-tower plume-activity index | Automated plume detection on Sentinel-2 and Planet imagery using spectral water-vapour contrast; cross-referenced against ERA5 reanalysis meteorology to separate meteorological suppression from operational shutdown | Monthly time-series chart with plume-day ratio and confidence band; alert if ratio drops below site-specific threshold |
| Thermal anomaly map for cooling-water discharge | Landsat TIRS land-surface temperature retrieval using split-window algorithm; comparison against multi-year seasonal baseline per pixel | Georeferenced temperature-anomaly raster per Landsat pass; quarterly summary report with trend line |
| Vehicle-count time-series | Manual and semi-automated vehicle detection on Planet SuperDove daily imagery; statistical baseline modelling with control-chart alert thresholds | Weekly vehicle-count dataset per access road; alert notification when count exceeds two standard deviations from baseline |
| Construction-phase sequencing | Multi-date change detection on WorldView-3 and Sentinel-2 SWIR; earthwork extent mapped using normalised difference built-up index variants | Annotated image series with construction-stage classification; GIS polygon layer of new structures with estimated footprint and first-detection date |
| Security-perimeter modification alert | Object-based image analysis comparing current WorldView-3 scene against reference scene; fence-line and berm geometry extracted and differenced | Alert report with annotated before-and-after image chips; vector overlay of changed perimeter segments |
| Integrated operational-status assessment | Structured analytic technique combining thermal, optical and construction indicators following CNS and 38 North published methodology; explicit alternative hypotheses and confidence levels | Written assessment report, updated on defined cadence, with evidence annex and imagery citations |
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.