Nighttime light recovery tracking as a reconstruction progress proxy
Nighttime light radiance from VIIRS Day-Night Band composites gives disaster responders and donors an objective, neighbourhood-scale measure of power restoration after earthquakes, hurricanes, and conflict. The method is fast, free, and honest about what it cannot see.
Sensors
- VIIRS Day-Night Band (Suomi-NPP / NOAA-20): Panchromatic low-light band, 750 m native resolution at nadir, daily global overpass near 01:30 local time. Monthly cloud-free composites from the Colorado School of Mines VIIRS Nightfire and NOAA EOG programmes are the standard time-series input. Nightly granules are available within hours but carry significant cloud and moonlight contamination requiring careful masking.
- DMSP-OLS archive: Predecessor sensor, ~2.7 km resolution, operational 1992-2013. Useful for establishing pre-disaster baseline radiance in historical events and for long-run trend context, though its coarser pixel and lack of on-board gain control make it incompatible with direct radiance comparison to VIIRS without inter-calibration.
- Luojia-1A: Chinese experimental nighttime-light satellite launched 2018, approximately 130 m resolution panchromatic night band. Sparse revisit (roughly monthly) and limited archive depth, but the finer pixel allows sub-neighbourhood discrimination that VIIRS cannot provide. Coverage and data access are inconsistent for non-Chinese users.
- ISS DSLR imagery (NASA Earth Observatory / Johnson Space Center): Astronaut photographs from the International Space Station at sub-100 m effective resolution, opportunistic and non-systematic. Invaluable for qualitative validation and public communication after major events (Puerto Rico post-Maria imagery is a well-documented example), but cannot be scheduled or relied upon for operational monitoring.
Why light is a proxy, not a measurement
Electrical power restoration is not the same thing as economic recovery, and radiance detected from orbit is not the same thing as power restoration. That chain of inference matters. What VIIRS actually records is upwelling photon flux at 750 m resolution. From that, analysts infer: lights are on; therefore power is available; therefore some threshold of normalcy has returned. Each step is defensible in aggregate and unreliable for individual pixels.
The proxy holds surprisingly well at the district or municipality level. Studies of post-earthquake Haiti (2010), post-hurricane Puerto Rico (2017), and post-conflict Syria published in peer-reviewed remote sensing literature consistently show that VIIRS radiance recovery curves correlate with independently measured restoration of water pumping, hospital operations, and commercial activity. The correlation weakens at the block scale and breaks entirely for individual buildings.
What a monthly composite actually shows
The NOAA Earth Observation Group produces monthly cloud-free VIIRS Day-Night Band composites by selecting the minimum-noise, cloud-masked observations across each calendar month. A pixel value represents average radiance in nanowatts per square centimetre per steradian. Comparing the pre-disaster baseline composite to each subsequent monthly composite gives a radiance-recovery ratio: values near 1.0 indicate full restoration, values below 0.2 indicate near-total outage.
Monthly composites trade temporal resolution for quality. A neighbourhood that restored power on day 20 of a month appears partially recovered in that month's composite. For the first weeks after a fast-onset disaster, nightly granules are more informative despite their noise. NOAA and NASA release VIIRS nightly data with latency typically under 24 hours through the FIRMS and STAR portals, which makes near-real-time monitoring feasible if analysts are prepared to handle cloud masking and lunar contamination manually.
The archive runs from 2012 (Suomi-NPP launch) to present, with NOAA-20 providing a second daily overpass since 2018. Twelve-plus years of monthly composites is long enough to establish seasonal baselines and to track multi-year recovery trajectories in protracted crises.
Three contamination problems every analyst must name
Saturation in dense urban cores is the most common artefact. VIIRS DNB was designed to detect faint light; bright city centres saturate the detector, producing a flat radiance ceiling that makes it impossible to detect partial dimming or partial recovery in the brightest pixels. Post-disaster dimming in a city centre may be real but undetectable if the surviving lights still saturate the sensor. Analysts should flag all pixels above the sensor's linear range (approximately 200 nW/cm²/sr for standard DNB) as unreliable for change detection.
Gas flares are a different problem. Oil and gas infrastructure produces intense, stable point sources that can dominate pixels in energy-producing regions. A pixel that appears brightly lit after a disaster may simply be a flare that was unaffected. The VIIRS Nightfire product from the Colorado School of Mines uses multi-band temperature fitting to separate flare combustion from reflected or emitted broadband light, and should be used to mask flare pixels before recovery analysis.
Generator-powered light is the most consequential ambiguity for humanitarian interpretation. A hospital running on diesel, a military checkpoint, a wealthy household with a petrol generator: all produce radiance indistinguishable from grid power at VIIRS resolution. In post-conflict settings particularly, light recovery can reflect the presence of armed actors rather than civilian normalcy. No satellite sensor currently resolves this ambiguity. Ground truth, social media signals, or auxiliary data (fuel import records, for instance) are required to interpret what the light source actually represents.
Building a recovery trajectory: the practical workflow
The standard approach takes a pre-event baseline (typically the median of 12 monthly composites before the disaster), then computes a normalised recovery index for each subsequent month: (post-event radiance minus minimum post-event radiance) divided by (baseline minus minimum). This index runs from 0 at peak outage to 1 at full baseline restoration. Plotting it by administrative unit gives a recovery curve that aid coordinators can read without remote sensing expertise.
Spatial disaggregation matters more than the index formula. Overlaying the recovery index on ward or neighbourhood boundaries, then joining to census-derived poverty or population data, identifies which communities are recovering slowest. That spatial join is where satellite data becomes actionable: not in the headline statistic but in the map that shows a specific district still at 30 percent of baseline radiance six months after an earthquake while the city average has reached 80 percent.
Satellize runs this workflow on open VIIRS composites and can add Luojia-1A tasking where finer spatial discrimination is required. The Tonga crop-estimation programme demonstrated the same underlying principle of using open-constellation time series to produce decision-relevant indices for a government client with limited in-country technical capacity.
Honest limits and what to pair with this method
VIIRS cannot see below approximately 750 m. A neighbourhood of 500 households is roughly one pixel. Recovery within that pixel is invisible. For sub-neighbourhood assessment, Luojia-1A or ISS photography are the only current satellite options, and neither is operationally reliable.
Cloud cover is a genuine constraint in the first weeks after tropical cyclones and monsoon-season floods, precisely when early recovery data is most valuable. Monthly composites mitigate this by aggregating across clear-sky nights, but in persistently cloudy regions a monthly composite may rest on only a handful of valid observations. Analysts should always report the number of valid observations per pixel alongside the radiance value.
This method should not stand alone. SAR-derived building damage assessments (covered separately in this library) tell you what was destroyed. Nighttime light tells you what has been reconnected. Pairing the two gives a clearer picture: areas with high structural damage and low light recovery are the reconstruction bottleneck; areas with low damage but low light recovery may indicate infrastructure or governance failures upstream of the neighbourhood. That combination is more useful to a reconstruction planner than either dataset alone.
Typical figures
| Native spatial resolution | 750 m (VIIRS DNB at nadir); degrades to ~1,600 m at swath edge |
| Revisit frequency | Daily (Suomi-NPP); twice daily with NOAA-20 added from 2018 |
| Monthly composite latency | Typically 1-2 weeks after month end (NOAA EOG release schedule) |
| Nightly granule latency | Under 24 hours via NOAA STAR and NASA FIRMS portals |
| Spectral band | Panchromatic, 500-900 nm (DNB); designed for 10⁻⁹ to 10² W/cm²/sr dynamic range |
| Saturation threshold | Approximately 200 nW/cm²/sr; dense urban cores frequently exceed this |
| Archive depth | VIIRS: 2012-present; DMSP-OLS: 1992-2013 (requires inter-calibration for VIIRS comparison) |
| Cloud sensitivity | Optical band; cloud-contaminated nights excluded from composites, reducing valid observations in persistently cloudy regions |
| Minimum detectable radiance change | Approximately 0.5 nW/cm²/sr in monthly composites (published noise floor estimates); single-night noise is higher |
| Delivery formats | GeoTIFF radiance rasters, CSV recovery indices by administrative unit, web map tiles |
Analytics Satellize can run
| Pre/post radiance change map | Pixel-wise ratio of post-event monthly composite to pre-event baseline median (12-month), with saturation and flare masking applied | GeoTIFF and styled web map layer, updated monthly |
| Normalised recovery index by administrative unit | Zonal statistics aggregation of DNB radiance over ward or municipality polygons, normalised to pre-event baseline; time series from disaster onset to present | CSV time series and PDF chart report, monthly cadence |
| Recovery laggard alert | Threshold detection: administrative units where recovery index remains below 0.4 at three months post-event, flagged against population-weighted exposure data | Automated alert feed (email or API) with ranked list of underperforming districts |
| Nightly outage snapshot (acute phase) | Single-night VIIRS granule differencing against baseline, with cloud and moonlight masking; intended for first 30 days when monthly composites are unavailable | Daily GeoTIFF with confidence flags per pixel, delivered within 36 hours of overpass |
| Flare-masked recovery surface | Integration of VIIRS Nightfire combustion-source classifications to exclude gas-flare pixels before recovery computation; critical for energy-producing conflict zones | GeoTIFF with flare-excluded pixels clearly coded; methodology note included |
| Damage-to-light-recovery gap analysis | Spatial join of SAR-derived building damage density (from coherence loss or amplitude change) with DNB recovery index; identifies areas where infrastructure damage explains slow light recovery versus areas where governance or supply-chain factors are implicated | GIS layer and two-page analytical brief |
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.