Nighttime light intensity as a construction activity proxy
Round-the-clock construction leaves a measurable light signature in low-light satellite imagery. VIIRS Day-Night Band and Luojia-1 data turn that signature into an independent activity index, though urban adjacency and seasonal darkness complicate the reading.
Sensors
- VIIRS Day-Night Band (Suomi NPP / NOAA-20): 750 m native pixel, daily global revisit, panchromatic low-light band sensitive from 0.5 to 0.9 µm, capable of detecting radiances down to roughly 3 × 10⁻⁹ W cm⁻² sr⁻¹ µm⁻¹. The workhorse for time-series analysis; archive runs from October 2011.
- Luojia-1 (Wuhan University): 130 m resolution nighttime panchromatic imagery, approximately 15-day revisit for any given point, covering 250 km swath. Significantly finer spatial detail than VIIRS, useful for isolating individual construction zones within dense urban areas, though the archive is limited (launched June 2018) and tasking is not routine.
- DMSP-OLS archive (NOAA): Roughly 2.7 km pixel, 6-bit radiometric depth, no on-board gain control, prone to saturation over bright sources. Useful only for historical trend context pre-2012; not suited to site-level construction monitoring.
- VIIRS Nightfire (Colorado School of Mines / NOAA): A derived product from VIIRS shortwave-infrared bands that characterises combustion sources. Occasionally relevant when construction involves gas flaring, welding or open burning, but not a direct luminance proxy for floodlighting.
What a floodlit site actually looks like from 800 kilometres up
A fast-track construction project running two or three shifts typically deploys between 50 and several hundred metal-halide or LED floodlights, each rated at 400 W to 2 kW. The aggregate upward radiance from a large site can reach tens of milliwatts per square metre at sensor altitude, well above the VIIRS Day-Night Band detection floor. At 750 m per pixel, a site of two hectares or more will register as a statistically distinct brightening against its pre-construction baseline, provided that baseline is cleanly established.
The signal is not a photograph of the site. It is a radiance value integrated across the pixel footprint. That distinction matters: a single VIIRS pixel can contain a construction site, a motorway interchange and a petrol station simultaneously. Separating the construction contribution from the rest requires careful baseline subtraction and, ideally, Luojia-1 imagery to confirm spatial attribution at 130 m.
Building the activity index: baseline, change and the limits of the method
The standard analytical approach compares monthly median radiance composites over a defined site polygon against a pre-construction reference period, typically twelve months before ground-breaking. A statistically significant increase in median radiance, sustained across multiple overpasses, is the primary signal. VIIRS provides one usable nighttime observation per 24-hour cycle under clear skies; cloud cover blocks the optical signal entirely, which is the single biggest operational constraint. In persistently cloudy tropical climates, monthly compositing may yield only four to eight cloud-free observations.
Seasonal variation in ambient darkness is a subtler problem. At latitudes above roughly 45° north or south, astronomical twilight at the time of the Suomi NPP equatorial crossing (approximately 01:30 local solar time) varies by tens of minutes across the year. Winter observations are darker, which slightly increases apparent site radiance even with no change in actual floodlighting. Analysts must normalise against surrounding dark reference pixels or apply a moon-illuminance correction using the published lunar irradiance model embedded in the VIIRS DNB processing chain.
The method produces a relative index, not an absolute measure of construction intensity. A doubling of radiance does not imply a doubling of workforce or expenditure. It implies more light output, which correlates broadly with extended shift operations but can equally reflect a change in lamp type, reflector orientation or temporary scaffold sheeting that redirects spill light upward.
Urban adjacency: the confounder that defeats naive analysis
Construction sites in or near cities sit inside a background glow that can be orders of magnitude brighter than the site signal itself. The VIIRS DNB is not saturated by most urban cores, but the spatial blurring from its point-spread function means that light from adjacent streets bleeds into the site pixel. This is not a minor correction. Published studies of VIIRS urban radiance have documented point-spread function wings extending two to three pixels beyond a bright source.
Practical mitigation involves defining a tight site polygon that excludes known permanent light sources, applying a local background subtraction using an annular reference zone, and cross-checking anomalous readings against Luojia-1 imagery when available. For greenfield sites in dark rural settings, the method is considerably cleaner. For a site inside a port, an industrial estate or a city centre, the analyst must be explicit about the uncertainty range, which can easily exceed 30 to 50 per cent of the estimated site contribution.
What the time series can and cannot tell a project financier
A consistent nighttime light time series over a construction site answers a narrow but valuable question: is work happening at night, and is that pattern changing? For project-finance purposes, sustained nocturnal luminance is consistent with the contractor meeting an accelerated programme. A sudden drop, particularly one not explained by cloud cover or a known scheduled shutdown, is a flag worth investigating.
The method cannot verify physical progress milestones, structural completion or material quantities. Those questions belong to optical progress imagery and stereo-derived volume models, covered in sibling pages. Nighttime light is best treated as a behavioural indicator: it captures effort and schedule pressure rather than output. Combined with daytime optical change detection, it adds a dimension that daytime imagery alone cannot provide, since a site can appear unchanged in a morning overpass yet have logged eight hours of night-shift activity.
Satellize runs this analysis as part of its broader infrastructure-monitoring service, applying VIIRS compositing and Luojia-1 cross-validation over client-defined polygons. The methodology shares the same open-data pipeline used in the Tonga crop-estimation programme, adapted from radiance time-series to a construction-activity context.
Archive depth and what history reveals
The VIIRS record from Suomi NPP begins in October 2011, giving over a decade of nightly global coverage. NOAA-20 adds a parallel record from January 2018 onward, providing cross-calibration and redundancy. For any construction project that broke ground after late 2011, it is possible to reconstruct a full nighttime activity history from publicly available data. This matters for retrospective due diligence: an investor reviewing a completed asset can ask whether the contractor's claimed shift patterns during a critical phase are consistent with the satellite record.
The DMSP-OLS archive extends back to the 1990s but at 2.7 km resolution with severe saturation over bright sources. It is useful for city-scale or regional trend analysis, not for individual site attribution. Treat it as context, not evidence.
Typical figures
| VIIRS DNB native spatial resolution | 750 m per pixel at nadir |
| Luojia-1 spatial resolution | 130 m per pixel |
| VIIRS DNB revisit | Daily (one nighttime pass per 24 h, Suomi NPP); dual-satellite with NOAA-20 from 2018 |
| Luojia-1 revisit | Approximately 15 days for any given point |
| VIIRS DNB spectral range | 0.5 – 0.9 µm (panchromatic low-light) |
| VIIRS DNB minimum detectable radiance | ~3 × 10⁻⁹ W cm⁻² sr⁻¹ µm⁻¹ (published sensor specification) |
| Cloud penetration | None; optical sensor blocked by cloud cover |
| VIIRS archive depth | October 2011 to present (Suomi NPP) |
| Minimum detectable site size (VIIRS) | Approximately 2 ha for a statistically distinct signal in a dark rural setting; larger in urban adjacency |
| Typical analysis latency | 24 – 72 hours after VIIRS daily product release for near-real-time composites; monthly composites within 5 days of month end |
Analytics Satellize can run
| Monthly radiance anomaly index | Median compositing of cloud-free VIIRS DNB overpasses within calendar month; pixel-level difference against pre-construction baseline period; local background subtraction using annular reference zone | Time-series chart and GIS polygon layer with radiance anomaly values, delivered as GeoJSON and PDF report |
| Shift-pattern activity flag | Individual overpass radiance extraction at site polygon centroid; statistical threshold crossing against rolling 90-day baseline to flag nights of elevated activity or unexpected shutdown | Alert feed (JSON or email) triggered within 48 hours of a threshold event |
| Luojia-1 spatial attribution layer | 130 m Luojia-1 imagery co-registered to site boundary; brightness segmentation to isolate site contribution from adjacent permanent sources | Annotated image tile with source-attribution polygons, delivered as GeoTIFF |
| Retrospective activity reconstruction | Full VIIRS archive compositing over user-defined historical period; DMSP-OLS trend context pre-2012 where relevant; anomaly scoring against regional dark-pixel reference | Historical activity report with monthly index table, suitable for due-diligence or dispute-resolution use |
| Seasonal and lunar normalisation correction | Application of published VIIRS DNB lunar irradiance model and twilight-offset correction to remove astronomical confounders from multi-year time series | Corrected radiance time series supplied alongside raw values, with methodology note |
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.