What this gives you: for any one of Indonesia's 514 regencies, how much light it sends up into the sky per unit of ground, whether that is rising or falling and how fast, how much it swings during Ramadan, and how much of it is waste gas being burned off rather than anything anyone uses.
What it does not give you: anything about who has electricity, and anything about what the sky looks like from your own back garden. This measures light leaving the ground, seen from above. It is not a reading of what you would see looking up, and a dark regency here is not a regency without power. Both of those are in the limits below, and the second one has ended careers.
Do lights actually track money? Someone checked, in Indonesia
A 2021 study did exactly this test on Indonesian data, using satellite images from 2015-16, and its answer split in two. In cities, lights tracked official economic output closely — explaining about 68% of the differences between them. In rural regencies they explained 1% — which is nothing at all. The authors concluded that night lights are "not a suitable source of data to proxy for GDP in non-urban areas in developing countries like Indonesia".
This case runs the same test — same country, same administrative level, same formula — on a different satellite product and a later decade. The city half comes out almost identical: 77% against their 68%. The rural half does not: 67% against their 1%. One of those two results is a replication and the other is a contradiction, and the case's own reading is that the sensor changed, not the country.
So "is it good?" has no single answer here. Against the country-level rule of thumb — a one per cent change in light accompanying roughly 0.28 per cent more output — this data behaves as published work expects in cities and behaves like nobody's published expectation in the countryside. That disagreement is why the rest of this page is careful about which question the lights are being asked.
The one question this data answers without standing in for something else
This case has been asked two questions before this one and it got both wrong, for the same reason: both needed the lights to be a stand-in. Asked to estimate economic output, the lights lost to a rival that simply repeats last year's growth. Asked about electricity access, there was nothing in the record to check an answer against at all.
Light pollution is not like that. The instrument measures light travelling upwards at night, and light travelling upwards at night is light pollution. There is no step where a number has to be trusted to represent something it cannot see. That is why this page can say things plainly, and it is the only question in this whole case where that is true.
Indonesia's light is not spread like its land
Java holds about 7.5% of the ground this map covers and sends up about 61% of the country's night light — roughly 8.15 times its share. Put the map on the first view and the shape of that is immediate: one island burning, and an archipelago that is mostly matte.
Which places are worst depends on a choice most reports make silently, and it changes the answer completely. Rank regencies by their total light and you get the big peri-urban districts around Jakarta and Surabaya. Rank them by light per unit of ground — which is what light pollution means, because a sky does not care how wide the district under it is — and 8 of 10 names change.
| Ranked by light per unit of ground | Ranked by total light |
|---|---|
| Kota Jakarta Pusat | Bekasi |
| Kota Jakarta Barat | Bogor |
| Kota Jakarta Utara | Tangerang |
| Kota Jakarta Selatan | Banyuwangi |
| Kota Yogyakarta | Kota Surabaya |
| Kota Jakarta Timur | Karawang |
All 5 of Jakarta's city districts are in the per-unit-of-ground top ten. Not one of them is in the total-light top ten, because each is small. The two lists share only Kota Bekasi and Kota Surabaya. The places emitting the most light in total are not the places with the brightest skies — and a programme aimed at the wrong one of those two lists would spend its money in the wrong districts.
And yes — it is getting brighter
This is the question a reader actually wants answered, and unusually for this case it can be. Over the 103 months from 2018-01 to 2026-07 the country's night light rose about +6.19% a year, ending 1.89 times where it started. Of 514 regencies, 430 are brightening and 84 are not; the middle one is gaining +6.43% a year. Show me that on the map.
Why this trend is allowed and a longer one is not. The satellite behind these monthly numbers changed in 2018. Every month on this page comes from after that change, so the whole series is one instrument comparing itself with itself. A brightening measured across the changeover would be partly a fact about satellites, and this page does not draw one.
Why you cannot simply add up what the satellite saw
Cloud and moonlight decide how much of Indonesia is usable in any given month, and the monsoon takes most of it. Add the light up as it arrives and you get the thin line below: it swings by a factor of thousands and it is mostly a weather record, tracking the cloud-free share beneath it at 0.734 on a scale where 1 would be lockstep.
Two checks matter here, because "the satellite is just seeing more of the country" would sink everything above. It is not: the cloud-free share does not climb across the record, it drifts -0.52% a year, from 25% of the grid in the first twelve months to 27% in the last. And the per-regency trends are not an artefact of filling in the missing months: refit on observed months only and the two sets of answers agree at 0.86 across 507 regencies.
Ramadan is visible from orbit, and the national figure hides it
Human habits change the night sky measurably. Averaged over the country, Ramadan shifts Indonesia's light by +0.33% — near enough to nothing, and that number is worse than useless, because it is the average of two crowds moving in opposite directions.
255 regencies go darker during Ramadan and 259 go brighter, which is very nearly a coin toss and is exactly why the national figure cancels to nothing. 259 of them — half the country — move by a tenth or more in one direction or the other. Pick a regency on the map and the card gives you its own swing.
Light nobody asked for
Some of this light has no human purpose at all. At oil and gas sites, gas that cannot be captured is burned off, and the flame is bright enough to be counted from orbit as though it were a town. Against a published survey of flare sites, 182 of them fall inside Indonesia's regencies, and about 5.3% of the country's night light comes from within 3 km of one.
Nationally that is a footnote. Locally it is the whole story:
| Regency | Share of its light from flares | Flare sites |
|---|---|---|
| Teluk Bintuni | 92% | 1 |
| Sorong | 71% | 3 |
| Kota Lhokseumawe | 68% | 2 |
| Tanjung Jabung Timur | 67% | 6 |
| Banggai | 54% | 2 |
| Kota Dumai | 52% | 1 |
In Teluk Bintuni 92% of all the light the satellite sees comes from 1 flare site. Any reading of that regency's lights as a sign of activity, prosperity or population is reading a flame. Show me the flares. This is also the one figure on this page that is already taken out of everything else: flare light is subtracted before the brightness and trend numbers are computed, which is why those two views can be trusted in the 54 regencies where it would otherwise dominate them.
What this cannot do, including the part that cuts against it
Four things, and the first is the most important on the page.
- This satellite cannot see blue light. Its night-time sensor responds to roughly the yellow-to-near-infrared part of the spectrum and is close to blind below about the blue end. The world is replacing orange sodium street lamps with blue-rich white LEDs. So a city can switch its lighting, put more glow into the sky — blue light scatters harder in air, which is what sky glow is made of — and get dimmer in this data. That means the 57 regencies shown in blue on the second view cannot be read as improving. Some of them may be; for others this is an explanation that cannot be ruled out, and nothing in this case's records addresses the sensor's colour response at all, so no number here is corrected for it.
- Light measured at the satellite is not the brightness of your sky. What reaches orbit is the light that escaped straight upwards. What you see looking up is that light scattered back down by air, dust and water vapour, and it depends on the weather, the aerosol load and the angle of every lamp. Nothing in this case measures sky brightness at the ground, so no figure here should be presented as what someone would see.
- The method is weakest where official statistics are weakest. In the most recent year checked, 29 regencies sit far outside the pattern the rest of the country follows, and most of the published examples carry a "low annual coverage" tag — Tambrauw and Pegunungan Arfak among them. Across the whole record, 44% of regency-months have no usable image at all, and 111 regencies have fewer than 24 well-observed months out of 103. These are the same places where the government's own figures are thinnest. Their trends on the second view are the least trustworthy on the map, and the card says so when you pick one.
- The thing this case was built to do, it does not do. It set out to nowcast economic output from lights. Tested on a year held back from it, its average error was 41% larger than a rival that simply assumed this year's growth equals last year's, and it beat that rival in only 27% of regencies. Worse, 99.5% of its published growth figure was the fixed part of the model rather than anything the lights contributed, and its uncertainty band was 65 times wider than the lights' contribution to it. The check written down in advance for this failed and is published as failed. That failure is why this page asks a different question, not a softer version of the same one.
And the electricity question, since everyone asks it. Lights look like they should show who is connected to the grid, and this case cannot tell you that: it holds no electrification, utility or household-survey figure to check an answer against. It can rule out one tempting story, though. The reason these lights track rural economies far better than the older literature found is not a decade of electrification — a test written down in advance checked exactly that, and the agreement was already high in 2015 and 2016, the years the earlier study used, at 0.8249 and 0.7835 against that study's 0.01 on a scale where 1 is perfect agreement and 0 is none. It barely moves over the whole record, 0.7684 in 2012 to 0.8333 in 2025. The improvement is in the data product, not in the wiring of the country.
What you must not conclude from this map. A dark regency is not a regency without electricity — a village with power and no street lighting reads dark from orbit, and Indonesia's official connection rate is near universal. A dark regency is not undeveloped, poor or empty; forest, mountain and sea are dark because nothing there emits light, which is a fact about land cover, not about people. And a bright regency is not prosperous: this case's own attempt to turn brightness into an economic figure failed its test, and the brightest thing in some regencies is a gas flare with nobody living under it. Use this to talk about light in the sky. Do not use it to rank places, allocate budgets by need, or judge anyone's standard of living.
What a government or an NGO can actually do with this
- Regulate per unit of ground, not per district. The two rankings above disagree on 8 of ten names. A lighting ordinance written against total district output would exempt exactly the densest, brightest skies in the country.
- Act on the fast tier while it is small. 33 regencies are gaining more than 15% a year. Lighting installed now sets the fixtures, the angles and the colour temperature for decades, and the cheapest moment to require full cut-off shielding and warmer lamps is before they go in.
- Flaring is the one source with no constituency. 5.3% of the national total, concentrated in 45 regencies where it is more than a trace, comes from burning waste gas. No one is lit by it, no one asked for it, and reducing it costs nobody their lighting.
- Do not measure your own progress with this satellite. Point 1 of the limits is the reason: switch a city to white LEDs and this data will report improvement whether or not the sky over it got darker. Progress on sky glow has to be measured from the ground.
Could this run in daily operations?
FEASIBILITY ONLY — none
Every pipeline here is a batch backfill: it fetches a season or an archive, not a live feed. Anything operational means rebuilding the ingest for near-real-time arrival, whatever the verdict below says.
What stops it. The out-of-sample check failed: the mapping beats simply assuming last year's growth in fewer provinces than the bar set in advance required. It was also scored on a single year rather than the rolling backtest the plan called for.
What it would need. A mapping that beats the naive baseline. Nothing about the ingest is the constraint here.
If you want the detail
The methods, the checks set in advance and how each came out, the failed economic nowcast in full, and everything still uncertain are on the technical article, and the technical dashboard carries the whole monthly record and the statistics behind it. Every figure here is read from the same files as those, so they cannot drift apart.
Words on this page, in plain language 6 terms
- Black Marble
- NASA's processed night-lights product: how much light each patch of ground emits at night, cleaned of moonlight and cloud. Used as an indirect signal of economic activity.
- a look
- One pass of the satellite over a field. More looks means the crop's cycle is sampled more often; too few and a short stage — flooding, heading — can happen entirely between two looks and never be seen.
- nowcast
- An estimate of what a statistic is doing right now, before the official figure is published. It is a forecast of the present, not the future.
- baseline
- The dumbest sensible answer, used as the thing to beat — for example "this year looks like last year". A model that cannot beat its baseline has added nothing, however good its score looks alone. Careful: in the radar sections the same word means something else entirely — a reference level a signal is measured against, as in "3 dB below the cell's own baseline".
- orbit (ascending/descending)
- Whether the satellite passed heading north or south. It looks from a different angle each way, so measurements from the two are never mixed.
- regency
- The English name for a kabupaten: the administrative level below a province, and the level Indonesia publishes most official statistics at. A regency is not a city — cities are counted separately here, and on several of these cases they behave differently enough to be reported on their own.
Night light from NASA's Black Marble daily product, 2018-01 to 2026-07. Flare sites from Elvidge & Zhizhin (2021), Global Gas Flare Survey by Infrared Imaging, VIIRS Nightfire 2012–2019, ORNL DAAC. Data vintage 2026-07. Outlines simplified from 62,980 to 12,518 points for the web from the same boundaries the analysis used; Kepulauan Seribu and Kota Sibolga are smaller than one pixel at this width and are drawn as a minimum-size mark rather than left off. Written to be read without a background in remote sensing; nothing was rounded to make a point.
Found something wrong on this page? Report a correction — it opens a pre-filled issue — or email taufik.adi@openstudy.id. Corrections are credited by name in the errata, and one that changes a finding says so on the page.