Every year since 2012, every month since 2018: satellite radiance translated into a regency-level map of economic activity, refreshed monthly against regional accounts that arrive once a year.
VJ146A3 · 2026-07 · 15 ARC-SECOND COMPOSITE67% of Indonesia’s night light comes from 6.6% of its land.
One island holds the archipelago’s economic gravity. Every regency’s share of that light — and how it moves month to month — is measurable from orbit. That is what this instrument does.
National sum of lights per month, 2018 to the latest composite; the ledger grows by one month each time NASA publishes. Hollow points mark monsoon months, where cloud cover leaves few clear observations — the pipeline reports what it cannot see rather than guessing. Press play on the map below to watch every regency move.
Values are lights-implied activity, not GDP. The coverage-weighted, seasonally adjusted version of this series meets BPS’s regional accounts in chapter 04.
Breadth and intensity tell different stories: industrial Bekasi and Bogor lead on total light, while Jakarta’s dense urban cores lead on brightness per pixel. (Kutai Kartanegara’s total is inflated by gas flaring — chapter 04 shows the flare mask that corrects for it.)
Do the lights actually track the economy? Here the index meets BPS’s regional accounts: real PDRB for all 514 regencies — annual and quarterly. The verdict is two-sided, and both sides are printed. Where activity is, lights explain about 70% of the cross-regency variation, every year (the location check). When it moves, the short-run link is too weak to use: the growth coefficient’s interval contains zero, and out of sample the mapping loses to simply assuming last year’s growth. 2020 is usually offered as the proof case; on our own numbers it is not one — lights fell 2.1% and GDP fell 2.1%, but satellite coverage collapsed that year too, so signal and artefact cannot be separated. The uncertainty bands below are that sentence, drawn.
Lights-implied growth = the quarterly growth calibration (a + β·Δlog lights) applied to the deseasonalised national series; the band is a 95% interval from the calibration’s parameter and residual uncertainty. It is an activity index, never GDP. The shaded zone is the nowcast — months where BPS has not yet published.
Largest standardised residuals in the latest BPS year. A regency that is bright but poor, or dark but rich, is one where the satellite and the statistical office tell different stories — a question worth asking, and the most useful thing this instrument produces. Hollow dots on the scatter are flare-corrected oil & gas regencies.
Calibration data hasn’t published yet — this chapter fills in when the model stage lands on the server.
Total light is the sum of radiance falling inside a regency, so it rewards size and breadth — big industrial regencies lead. Intensity is the mean radiance per pixel, which is independent of size — dense city cores lead. Darker means dimmer, on a log scale that is fixed across all 103 months, so colours are comparable as the player runs. One consequence is worth knowing before you read the animation: a cloudy monsoon month goes dark across the whole map because the satellite saw less of it, not because the lights went out. The coverage figure below the slider tells you which months those are.
An independent read of the same data against the published literature. It replicates Gibson et al. (2021) on Indonesian kabupaten and finds their rural verdict is a property of the sensor product, not of night lights; it explains why the growth nowcast has been withdrawn; and it corrects a claim this page used to make about 2020.
Read the review article →