Night-lights mapping of urban economic activity
Calibrated night-light radiance from VIIRS and DMSP-OLS reveals electrification extent, relative wealth gradients, and economic activity patterns. The signal is real but indirect: physics, saturation, and mixed light sources all require careful handling before any GDP inference is defensible.
Sensors
- VIIRS Day/Night Band (Suomi NPP / NOAA-20): 750 m nadir resolution, daily global revisit, panchromatic 500–900 nm range, linear radiometric response across roughly seven orders of magnitude. Resolves individual lit roads and industrial sites that DMSP-OLS saturates. Monthly composites are publicly archived from 2012 onwards via NOAA/NGDC and the Earth Observation Group at Colorado School of Mines.
- DMSP-OLS archive: 6 km effective resolution, ~6-bit dynamic range that saturates over most city cores above roughly 60 nW/cm²/sr. Useful for long-run trend analysis 1992–2013. Saturation means total radiance in dense urban centres is systematically underestimated; inter-satellite calibration between the multiple DMSP platforms is also inconsistent without post-processing.
- Luojia-1 (LJ-1 01): Chinese experimental night-light satellite launched 2018, approximately 130 m resolution, 260 km swath. Demonstrates that sub-100 m night-light mapping is physically achievable. Not yet in routine operational service, but published imagery shows resolution of individual street-lamp clusters and factory yards.
- ISS DSLR archive (NASA Earth Observatory / Johnson Space Center): Opportunistic astronaut photography, variable resolution typically 10–100 m depending on focal length and altitude. No systematic revisit or radiometric calibration. Useful for qualitative illustration and for identifying specific light-source types by colour temperature; not suitable for time-series analysis.
Why upward radiance reaches a sensor at all
Artificial light at night escapes upward through three mechanisms: direct upward emission from poorly shielded luminaires, scattering off atmospheric aerosols and water vapour, and reflection off surfaces including roads, rooftops, and vegetation. A VIIRS DNB pixel records the integral of all three. That integral is a function of installed lamp power, luminaire shielding geometry, atmospheric clarity, and surface albedo. None of those variables maps cleanly onto economic output.
The spectral composition matters too. High-pressure sodium lamps, which dominated street lighting through the 2000s, emit strongly in the yellow-orange band. LED conversions, accelerating globally since roughly 2012, shift emission toward blue-white. VIIRS DNB is panchromatic and cannot distinguish them. A city that replaces sodium lamps with more energy-efficient LEDs may appear brighter in the DNB because LED spectra overlap better with the sensor's peak response, even though actual electrical consumption has fallen. Any economic inference that ignores the ongoing LED transition will overstate growth in cities that have recently upgraded their street lighting.
The saturation ceiling and what VIIRS does about it
DMSP-OLS was designed for cloud detection, not radiometry. Its on-board gain control was never calibrated for absolute radiance, and its 6-bit digitisation saturates over any large city with significant commercial or industrial lighting. The practical consequence is that Shanghai, Lagos, and a mid-sized provincial capital all return the same maximum digital number. Comparing them is meaningless without correction, and even corrected values are extrapolations rather than measurements.
VIIRS DNB was purpose-built with a three-gain-stage design that keeps the detector in its linear range from moonlit rural fields (roughly 2 × 10⁻⁹ W/cm²/sr) up to brightly lit urban cores (roughly 2 × 10⁻² W/cm²/sr). That is a genuine improvement of several orders of magnitude over DMSP-OLS. It does not eliminate saturation entirely: the very brightest industrial flares and stadium lighting can still push individual pixels into non-linearity, but for most urban analysis the DNB delivers calibrated, comparable radiance values. Monthly composites from the Earth Observation Group filter out ephemeral sources such as fires and aurora, producing a stable economic-activity proxy.
What the signal is actually measuring, and what it is not
The correlation between night-light radiance and GDP per capita is well-documented at national scale, with published studies finding R² values in the range of 0.7–0.9 depending on the sample and year. That correlation is useful. It is not causal, and it breaks down in several important ways.
Industrial lighting is the most obvious confound. An oil refinery, a fishing fleet, or a large greenhouse complex generates intense radiance with minimal resident population and limited relationship to household welfare. Gas flaring from petroleum extraction is bright enough to appear in DMSP-OLS and VIIRS composites as apparent urban centres in otherwise uninhabited desert. Conversely, dense informal settlements with high population and economic activity but minimal grid connection appear dark. Night-lights map electrification and installed lamp power; they infer economic activity only where those correlate with it, which is not everywhere.
Sub-national analysis compounds the problem. At city-district or neighbourhood scale, the mix of commercial, industrial, and residential lighting varies sharply. A warehouse district and a high-density residential area may produce similar radiance for entirely different reasons. Resolving that ambiguity requires auxiliary data: land-use classification, building footprint density, road network topology, or mobile-network activity records. Night-lights are a first-order signal, not a complete picture.
Electrification mapping in peri-urban and rural peripheries
Where night-lights are most unambiguous is at the margin: detecting whether a settlement has grid electricity at all. A village that transitions from no lighting to even minimal street lamps produces a detectable increase in VIIRS DNB radiance that can be tracked in monthly composites. The detection limit for VIIRS DNB is roughly 0.2–0.5 nW/cm²/sr above background in clear-sky conditions, which corresponds to a small cluster of street lamps or a generator-powered facility.
This makes the DNB archive genuinely useful for tracking rural electrification programmes, monitoring whether new grid connections are actually energised, and identifying peri-urban growth corridors where informal development is outpacing formal infrastructure. Cloud cover is the main operational constraint: equatorial and monsoon-affected regions may have only a handful of clear-sky nights per month, making monthly compositing essential and introducing latency of four to six weeks before a reliable composite is available.
Building a defensible economic-activity layer
A credible night-lights economic-activity product requires at minimum: inter-annual radiance normalisation to correct for sensor degradation and gain changes; masking of ephemeral sources (fires, flares, fishing vessels); separation of industrial from residential radiance using ancillary land-use data; and explicit acknowledgement of the LED transition effect in any time series crossing 2015 onwards.
Satellize applies these steps when building radiance-change layers for government clients, drawing on the same open VIIRS composites that any analyst can access. The value is in the processing chain and the interpretation, not in proprietary data. For clients with specific sub-national requirements, commercial very-high-resolution optical imagery can be used to disaggregate the VIIRS signal at block or parcel level, though that introduces its own cost and revisit constraints. The Tonga crop-estimation programme demonstrated a similar principle: open sensors provide the baseline signal; the analytical rigour determines whether the output is actionable.
Honest limits before you commission an analysis
Night-lights analysis will not tell you GDP. It will not reliably distinguish a prosperous low-density suburb from a dark informal settlement of equal or greater economic activity. It cannot resolve individual buildings at VIIRS resolution, and Luojia-1's 130 m capability is not yet in routine operational supply. Cloud cover degrades temporal resolution in tropical regions. The LED transition introduces a systematic bias in any time series that spans the mid-2010s without correction.
What it will do reliably: map electrification extent at settlement level, detect rapid economic expansion in previously unlit areas, compare relative activity levels across cities or regions using consistent radiance units, and provide a long archive (DMSP from 1992, VIIRS from 2012) that no other open sensor class matches for this purpose. Used within those limits, it is one of the more cost-effective spatial economic indicators available to a planning ministry or development lender.
Typical figures
| Spatial resolution (VIIRS DNB) | 750 m at nadir |
| Spatial resolution (DMSP-OLS) | ~6 km effective (oversampled from ~2.7 km instantaneous FOV) |
| Spatial resolution (Luojia-1) | ~130 m (experimental, not in routine service) |
| Revisit (VIIRS DNB) | Daily global; monthly cloud-free composites typically available within 4–6 weeks of month end |
| Spectral band (VIIRS DNB) | Panchromatic 500–900 nm; no spectral discrimination between lamp types |
| Radiometric dynamic range (VIIRS DNB) | ~2 × 10⁻⁹ to ~2 × 10⁻² W/cm²/sr (linear); ~7 orders of magnitude |
| Minimum detectable radiance above background | ~0.2–0.5 nW/cm²/sr under clear-sky conditions (VIIRS DNB) |
| Archive depth | DMSP-OLS 1992–2013; VIIRS DNB 2012–present |
| Cloud sensitivity | Optical sensor; cloud cover blocks signal; monthly compositing required in tropical regions |
| Delivery formats | GeoTIFF radiance grids, CSV zonal statistics, change-detection GIS layers |
Analytics Satellize can run
| Electrification extent map | Thresholding of monthly VIIRS DNB composites against background radiance; settlement boundary overlay from open building datasets | GIS polygon layer showing lit versus unlit settlements, updated monthly |
| Relative economic-activity index by district | Sum of radiance (SoL) and mean radiance per unit area extracted for administrative zones; normalised against multi-year baseline to control for sensor drift | Tabular index with confidence intervals and year-on-year change, delivered as CSV and PDF report |
| Radiance change-detection layer | Pixel-level differencing of annual VIIRS composites with outlier masking for fires and flares; LED-transition correction applied to series crossing 2015 | GeoTIFF change layer and summary statistics report for planning or investment review |
| Industrial versus residential radiance separation | Overlay of VIIRS DNB with land-use classification (OpenStreetMap or national cadastre) to attribute radiance by zone type; flagging of anomalous industrial emitters | Annotated GIS layer distinguishing commercial, industrial, and residential radiance contributions |
| Peri-urban growth corridor identification | Multi-year VIIRS time series analysis identifying pixels transitioning from dark to lit; cross-referenced with road network proximity | Priority corridor map and ranked list of emerging lit clusters for infrastructure planning |
| Long-run electrification trend report | Harmonised DMSP-OLS to VIIRS DNB intercalibration using published regression methods; consistent radiance units across 1992–present archive | 30-year trend charts by administrative unit, suitable for development-finance reporting |
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.