Power-outage and blackout mapping with nighttime lights
VIIRS Day/Night Band radiance composites detect city-block-scale electricity loss at 500 m resolution by comparing pre- and post-disaster light fields. Cloud cover, moonlight and fire glow all confound the retrieval, and honest analysis accounts for each.
Sensors
- VIIRS DNB (Suomi NPP): Primary operational sensor. Day/Night Band covers 0.5–0.9 µm, native pixel size 742 m at nadir, resampled to 500 m in standard products. Single nightly overpass near 01:30 local time. Detects radiances as low as roughly 2 × 10⁻¹¹ W cm⁻² sr⁻¹, making it sensitive to single streetlight clusters. Launched 2011; archive depth over a decade.
- VIIRS DNB (NOAA-20): Near-identical instrument to SNPP DNB, launched 2017. Offset orbital phasing gives a second nightly overpass roughly 50 minutes apart from SNPP, doubling the chance of a cloud-free acquisition on any given night. Used in tandem compositing to reduce latency.
- DMSP OLS (archive): Predecessor sensor, 1992–2013 archive. Coarser resolution (roughly 2.7 km), no onboard gain control, and significant blooming around bright sources. Useful only for historical baseline construction or multi-decadal trend context, not for acute disaster response.
- Planet SuperDove (supplementary): Visible-band imagery at 3–5 m resolution with daily revisit. Cannot detect nighttime light directly in standard operations, but daytime imagery of the same footprint helps confirm whether apparent light loss corresponds to physical infrastructure damage or is a DNB artefact.
What a blackout looks like from 830 km up
The VIIRS Day/Night Band measures upwelling radiance in a broad visible-to-near-infrared window. On a clear, moonless night, a functioning city produces a stable, spatially consistent light field that changes slowly over months. A major power failure removes that signal abruptly. The contrast between a pre-event composite and a post-event acquisition is, in principle, straightforward: pixels that were bright and are now dark mark the outage boundary.
In practice the signal is clean enough to resolve individual neighbourhoods. Studies following Hurricane Maria in Puerto Rico in 2017 used VIIRS DNB to track grid restoration block by block over the following months, a process that took the island's grid nearly eleven months to complete. That kind of temporal granularity is not available from any ground-based reporting system at equivalent spatial coverage.
Four things that will fool the sensor if you let them
Moonlight is the most predictable confound. The DNB is sensitive enough that a full moon can add several tenths of a nanowatt per square centimetre per steradian to every pixel, partially masking a real dimming. NOAA's monthly cloud-free composites use lunar-cycle filtering to mitigate this, but single-night acquisitions taken near full moon require explicit lunar irradiance correction before any difference image is meaningful.
Cloud cover is the blunter problem. Thick cloud blocks upwelling light entirely, producing a dark pixel that is indistinguishable from a genuine blackout in a single pass. The standard workaround, compositing multiple cloud-free acquisitions, introduces latency of days to weeks depending on local cloud climatology. In persistently overcast disaster zones, this can render the DNB operationally useless for the first week, precisely when situational awareness is most urgent.
Fire glow is the opposite failure mode. A city that has lost grid power but is burning produces anomalously bright DNB pixels, not dark ones. Separating fire-related radiance from electric-light radiance requires cross-referencing with a thermal infrared active-fire product such as VIIRS I-band fire detections from FIRMS. Without that step, a post-earthquake fire in a blacked-out district could be misread as partial grid survival.
Seasonal vegetation change affects peri-urban and rural pixels. Deciduous tree canopy that is leafed-out in summer attenuates streetlight reaching the sensor; the same streets appear brighter in winter. A naive year-on-year comparison that ignores phenological phase will attribute seasonal brightening to grid recovery or seasonal dimming to partial outage. Baselines should be drawn from the same calendar window as the event.
Compositing latency versus single-night acquisitions: the honest trade-off
NOAA publishes monthly cloud-free VIIRS DNB composites through the Earth Observation Group at Colorado School of Mines. These are the most reliable products for stable baseline construction, but a monthly composite is obviously useless for a disaster that happened three days ago. For acute response, analysts work with raw single-night Level-1 radiance swaths, accepting higher noise and the cloud-masking problem in exchange for immediacy.
With two satellites (SNPP and NOAA-20) now providing near-nightly overpasses at offset times, the probability of at least one cloud-free acquisition within a 48-hour window is meaningfully higher than with a single satellite, though it remains below 50 % in tropical and monsoon-affected regions during peak cloud season. Realistic latency for a usable post-event image in a cloudy disaster zone is two to seven days, not hours. Buyers should plan accordingly and not expect DNB to substitute for SAR-based damage assessment in the immediate aftermath.
Building a defensible baseline
The quality of a blackout map is entirely bounded by the quality of the pre-event baseline. A single pre-event night is insufficient: transient dimming from a local power cut, a public holiday, or a passing cloud that escaped the mask will produce false positives. Standard practice is to median-composite at least 30 cloud-free nights from the same seasonal window in the preceding one to three years.
For countries with highly variable electricity access, the baseline itself carries information. Areas that are already dark before the disaster cannot show further loss in the DNB signal; the sensor is measuring light, not the absence of infrastructure. This is a real detection floor. Informal settlements with minimal grid connection before an event are systematically under-represented in blackout maps derived from nighttime lights alone.
From radiance difference to a grid-restoration timeline
The most operationally useful product is not a single post-event snapshot but a time series of nightly or bi-nightly acquisitions that tracks the percentage of pre-event radiance recovered across administrative units. Each polygon, a municipality, a district, a utility service zone, gets a restoration curve. Thresholds such as 50 % and 90 % radiance recovery are reported as dates, giving humanitarian coordinators and utility operators a common metric.
Satellize runs this compositing and change-detection workflow on open VIIRS data, cross-validated against daytime optical imagery where resolution permits. The methodology is the same class of analysis applied in the Tonga crop-estimation programme: open-sensor time series, rigorous baseline construction, and explicit uncertainty reporting rather than a single confident number that papers over sensor noise.
The output is a GIS layer updated on each new cloud-free acquisition, with a confidence flag per pixel reflecting lunar phase, cloud-mask quality, and fire-contamination risk. Analysts and decision-makers see not just where the lights are off, but how certain that assessment is.
What this method cannot do
VIIRS DNB will not locate a damaged substation, a downed transmission line, or a failed transformer. It shows the spatial footprint of darkness, not its cause. Attributing the outage to a specific infrastructure failure requires ground data or high-resolution daytime optical imagery interpreted alongside the light-loss map.
The 500 m pixel also means that a single city block going dark within a lit neighbourhood is below the reliable detection threshold. The sensor is well-suited to district-scale and city-scale outages; it is not a tool for identifying which street lost power. For sub-kilometre granularity, there is currently no operational nighttime sensor that fills the gap. The International Space Station's DSLR photography programme has captured individual city blocks at night, but with no systematic revisit and no radiometric calibration suitable for change detection.
Typical figures
| Spatial resolution (VIIRS DNB) | 742 m at nadir, resampled to 500 m in standard products |
| Revisit (SNPP + NOAA-20 combined) | Two overpasses per night, offset by roughly 50 minutes |
| Spectral band (DNB) | 0.5–0.9 µm (panchromatic visible to near-infrared) |
| Minimum detectable radiance | ~2 × 10⁻¹¹ W cm⁻² sr⁻¹ (single streetlight cluster detectable under ideal conditions) |
| Latency for cloud-free single-night image | 12–24 hours after overpass under clear skies; 2–7 days in cloudy conditions |
| Latency for monthly composite (NOAA EOG) | Approximately 2–4 weeks after month end |
| Archive depth (VIIRS SNPP) | 2012 to present |
| Archive depth (DMSP OLS) | 1992–2013 (historical baseline only) |
| Global coverage | Full daily global coverage at all latitudes |
| Key confounds | Moonlight, cloud cover, fire glow, seasonal vegetation, sensor gain variation |
Analytics Satellize can run
| Pre/post radiance difference map | Pixel-wise subtraction of median-composited pre-event baseline from post-event DNB radiance, with lunar correction applied | GeoTIFF and vector polygon layer showing percentage radiance loss per 500 m pixel, with confidence flag |
| Outage extent by administrative unit | Zonal statistics of radiance-loss pixels aggregated to municipality or district boundaries | CSV and GIS layer with percentage of area blacked out, updated per cloud-free acquisition |
| Grid-restoration time series | Sequential nightly compositing tracking radiance recovery against pre-event baseline; 50 % and 90 % recovery thresholds flagged per zone | Interactive restoration-curve dashboard and tabular report by administrative unit |
| Fire-contamination mask | Cross-reference of DNB bright anomalies with VIIRS I-band active-fire detections from FIRMS; contaminated pixels excluded from outage map | Masked GeoTIFF with fire-affected pixels flagged separately |
| Cloud-gap-filled composite | Multi-night median compositing over rolling 5- and 10-night windows to reduce cloud-induced data voids while minimising temporal blur | Updated GeoTIFF delivered within 24 hours of each new qualifying acquisition |
| Pre-event baseline radiance atlas | Three-year same-season median composite from VIIRS DNB monthly cloud-free products, used as the reference state for all change detection | Baseline GeoTIFF and metadata report documenting seasonal window, cloud-mask quality, and pixel-level data density |
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.