Where our data comes from and how we process it

SOURCES

• NASA FIRMS, VIIRS NOAA-20 and NOAA-21 products
• SiPongi (Ministry of Forestry), for comparison

WHAT WE COLLECT

Each heat detection in the box 94°–142° E and 12° S–7° N: coordinates, satellite pass time, sensor confidence, and fire radiative power (FRP).

HOW WE PROCESS IT

• Only one satellite, NOAA-20, is counted. Two satellites can see the same fire, so adding them would double count it.
• Counts cover the latest 24 hours ending at the most recent detection, including both day and night passes.
• Points outside Indonesian city and regency boundaries are excluded. Each point belongs to exactly one area, so the area totals equal the national total.
• Confidence follows the sensor's low, medium, and high classes. FRP is grouped as below 5 MW, 5 to below 25 MW, and 25 MW or above.
• The same detection is not stored twice; its identity combines satellite, time, and coordinates.

ACCURACY AND LIMITS

• One point represents a pixel about 375 metres wide, not an exact fire location. One large fire may produce many points.
• A hotspot does not always mean a fire; chimneys, field burning, or hot roofs may also be detected.
• Clouds and thick smoke can hide heat from the satellite. No hotspot does not mean no fire.
• The satellite passes about every 12 hours, day and night. NASA provides data roughly three hours after a pass.
• Compared with NOAA-20 rows in SiPongi, our national count differs by less than 1%. Different administrative boundaries can explain the difference.

SOURCES

• SiPongi (Ministry of Forestry): indicative burned forest and land area per year
• NASA FIRMS, VIIRS S-NPP standard quality: hotspots per year since 2015
• SiPongi: Manggala Agni firefighting reports
• NASA MODIS MCD64A1: monthly burned area (Microsoft Planetary Computer copy)
• Gaveau, Descals & Salim (2026): burned area from Sentinel-2, 2019–2024 (Zenodo, CC BY 4.0)
• CHIRPS v2.0 (Climate Hazards Center, UC Santa Barbara): monthly rainfall
• NOAA Climate Prediction Center: El Niño index (ONI)
• CAMS via Open-Meteo (air) and Open-Meteo (daily temperature and rain): the 30-day comparison

WHAT WE COLLECT

For each year since 2015: burned forest and land area per province and per city/regency from SiPongi, and the number of hotspots per city/regency from the S-NPP satellite. Every day: Manggala Agni firefighting reports, with date, place, area burned, and area put out. Monthly since 2015: NASA's burned area and CHIRPS rainfall per city/regency. Once, from the researchers' file: national monthly Sentinel-2 burned area, 2019–2024.

HOW WE PROCESS IT

• Burned area is stored as SiPongi publishes it, per province and per city/regency. A year without a figure is shown as no data, not zero. A broken or incomplete SiPongi answer is refused and the previous figures stay.
• Yearly hotspots use one satellite, S-NPP, because it is the only one with data since 2015. Only detections NASA marks as vegetation fires are counted; volcanoes, fixed land heat sources such as factories, and offshore flares are excluded. These counts are never added to the 24-hour NOAA-20 counts on the hotspot page.
• Yearly hotspots are counted from FIRMS standard-quality data, which NASA reprocesses and which usually lags by a few months. The current year is counted up to the latest available day, and that date is stated.
• Each hotspot belongs to exactly one city or regency, by the same rule as the hotspot page. A province's figure is the sum of its cities and regencies.
• SiPongi only publishes the total for the year so far. Since October 2026 we keep that total each time SiPongi adds a month; a month's addition is the difference from the month before. Months before we started recording cannot be filled in.
• Firefighting reports are read again in full every 6 hours. Reports dated after today, exact repeats, and areas over 5,000 ha for one report are not used, as they are typing errors in the source.
• How NASA maps burn scars (MODIS MCD64A1): the Terra and Aqua satellites image the ground daily, and NASA's method looks for 500 m cells that suddenly turn darker in infrared light than on the days before, matched with nearby active fires. Each burned cell gets the day it burned.
• How we count it: each month we read the 17 MCD64A1 map sheets covering Indonesia. Every cell with a burn date counts as its own area, 21.47 ha, and goes to the city or regency its centre lies in, using BIG's boundaries. A month counts only when all 17 sheets are present; a year with months missing is still shown from the months present, with the missing months named.
• Sentinel-2 area is the researchers' national figure per month, used unchanged. A year is added up only when all 12 months are present.
• CHIRPS rainfall (cells of about 5 km) is averaged for each city/regency from the cells whose centre lies inside its boundary. A city smaller than one cell takes the cell at its centre. National and fire-zone means are weighted by cell count, so larger areas weigh more.
• The fire zone is the cities/regencies with the most NASA burned area that together hold 80% of NASA's burned area in its complete years. Dry-season rain is the zone's total rain over July–October.
• The peak month is the month with the most NASA burned area in a year, only for years with all 12 months. A season ends in the first month after the peak whose area is under 10% of the peak.
• Yearly hotspots of other kinds (static land heat sources, volcanoes) are counted separately from NASA's kind tag and shown on their own, never added to vegetation fires.
• The 30-day comparison uses the 30 full days before today: NOAA-20 hotspots per city/regency, ISPU from the CAMS model's 30-day mean PM2.5, and the mean daily maximum temperature and total rain from Open-Meteo. Missing figures are not counted as zero; an area's rain is used only when every day is present.

ACCURACY AND LIMITS

• Burned area is an indicative figure from the Ministry of Forestry, calculated from satellite imagery. The current year's figure is still growing, and SiPongi's city and regency figures do not always add up to its province figures; we show both as published.
• Hotspots and hectares measure different things: a hotspot is a satellite heat detection, hectares are burn scars. One cannot be converted into the other.
• Small fires, or fires under cloud and smoke, can be missed by satellites, so they count in neither the area nor the hotspots.
• Firefighting reports come only from the working areas of Manggala Agni crews, mainly in Kalimantan and Sumatra. They record the crews' work, not every fire, and their areas are not added to SiPongi's burned area.
• NASA's cells are 500 m, so small fires are often missed. The copy we read lacks some months (for example September 2023 and much of 2025); those years come only from the months present, so they are lower than they should be, and the missing months are named.
• Sentinel-2 area exists only for 2019–2024 and only as national figures.
• CHIRPS rainfall is an estimate from satellites and stations, not a gauge in every place, and is published a few weeks after each month ends.
• 24-hour hotspots have no kind yet, so industrial heat sources are still counted in the 24-hour figures and the 30-day comparison.
• Air and temperature are model output. We have only started storing their history per city, so years cannot be compared yet.

SOURCES

• BMKG open data: M5+ and felt earthquakes
• USGS: major earthquake catalogue since 1900

WHAT WE COLLECT

Time, coordinates, magnitude, depth, area, felt reports, and tsunami-potential status from BMKG open data JSON files (gempaterkini.json and gempadirasakan.json). For longer history: M6.5+ earthquakes from the USGS catalogue. We do not take data from www.bmkg.go.id pages.

HOW WE PROCESS IT

• Times from BMKG files (UTC) are displayed in WIB.
• The same earthquake across both BMKG files is merged. BMKG-reviewed calculations replace preliminary ones.
• Events centred outside the Indonesia coverage box are excluded because BMKG files can also list overseas earthquakes.
• The historical list prefers BMKG. USGS fills events absent from BMKG records. Records match when they are within 120 seconds and 300 km.
• Stored earthquakes are not deleted when they disappear from a BMKG list.

ACCURACY AND LIMITS

• Magnitude and depth in the first minutes are preliminary and often revised by BMKG.
• If BMKG corrects an event time, one earthquake may appear twice because the source files provide no event ID.
• We only take M5+ and felt earthquakes from BMKG open data, every 5 minutes. Smaller earthquakes that are not on those lists, and the full lists on www.bmkg.go.id, are not collected. Check BMKG for new or sub-M5 events.
• BMKG open files hold about 15 recent events per list. Earthquakes displaced between collections may be missed.
• A tsunami-potential marker records warning history on the M5+ file, not an active warning.
• Different agencies may report different magnitudes for the same earthquake.

SOURCES

• MAGMA Indonesia (PVMBG): activity levels, daily reports, eruption information
• Smithsonian Global Volcanism Program: eruption history and VEI
• GDACS: eruption alerts

WHAT WE COLLECT

The Geological Agency's list of 127 volcanoes, today's activity level, daily reports (visual observations, seismic activity, recommendations), today's eruption notices, and eruption history with VEI.

HOW WE PROCESS IT

• Status follows PVMBG's activity levels: Normal, Waspada, Siaga, Awas. We do not invent levels.
• We label an eruption when MAGMA records one that day or its daily report mentions one.
• Volcano names from each source are matched to the official list. Unmatched names are recorded and not forced into it.
• If the activity-level list is empty or a report date does not match, collection is cancelled so older status is not overwritten.
• Eruption history is counted by volcano from the Smithsonian catalogue. Highest VEI uses all records, not only those since 1900.

ACCURACY AND LIMITS

• One report is published per volcano each day and may arrive in the afternoon or evening.
• The Smithsonian catalogue records years, often without exact dates. Small and older eruptions may be missing.
• VEI measures eruption size, not casualties or damage.
• Official exclusion distances appear only in PVMBG reports. The recommendation we display copies the first recommendation item.

SOURCES

• CAMS (Copernicus) global model through Open-Meteo
• OpenAQ ground sensor network
• Environment and Forestry Ministry Regulation 14/2020: ISPU calculation

WHAT WE COLLECT

Hourly PM2.5 at the representative point of each city and regency from the CAMS model over the past 30 days. From OpenAQ: hourly PM2.5 from sensors active in the past 48 hours.

HOW WE PROCESS IT

• Today's value is the average PM2.5 over the past 24 hours. If today's estimate has not arrived, such as just after midnight, we show the latest estimate with its time, no older than 48 hours. Older data is marked unavailable.
• The 7 and 30 day values average available hours across those periods and require at least 75% coverage.
• PM2.5 is converted to ISPU using the regulation's breakpoints, then grouped as Good, Moderate, Unhealthy, Very unhealthy, or Hazardous.
• A sensor is used only if it has at least 18 valid hours out of 24. Hours flagged as faulty by the provider are excluded.
• Model and sensor readings are stored separately and always labelled; they are never averaged together.

ACCURACY AND LIMITS

• One model grid cell is about 45 km wide. Neighbouring cities can have the same value, and traffic on a particular road is not resolved.
• The model can differ from ground conditions, especially when wildfire smoke arrives suddenly.
• Our ISPU uses PM2.5 alone. Official government ISPU uses station instruments and several pollutants, so the values may differ.
• Ground sensors remain sparse and unevenly distributed, mostly in major cities.

SOURCES

• Open-Meteo forecasts
• Open-Meteo historical ERA5 weather
• CHIRPS v2.0 (Climate Hazards Center, UC Santa Barbara): monthly rainfall since 2015

WHAT WE COLLECT

At one representative point in each city and regency: weather condition; mean, minimum, and maximum temperature; rainfall total; maximum wind; and hourly temperature and weather. From 30 days ago through 6 days ahead.

HOW WE PROCESS IT

• A day follows WIB throughout Indonesia so national dates are consistent.
• The daily condition is the most severe condition that day. Rain does not mean it rained all day.
• Each row is labelled as forecast, a model result for a recent past day, or ERA5 reanalysis for an older day.
• Rain uses BMKG categories: light 0.5–20 mm, moderate 20–50 mm, heavy 50–100 mm, and very heavy at least 100 mm per day.
• Missing values remain missing; they do not become 0°C or clear skies.
• CHIRPS monthly rainfall (cells of about 5 km) is averaged for each city/regency from the cells inside its boundary; a city smaller than one cell takes the cell at its centre. National means are weighted by area. Each month is compared with the mean of the same month over the ten years before.

ACCURACY AND LIMITS

• This is model output, not a weather station measurement. One point represents a whole city or regency.
• Model grid cells span several to tens of kilometres. Local rain and mountain temperature differences may be missed.
• Forecast uncertainty grows with lead time and forecasts can change at each update.
• This is not an official forecast or early warning. Follow BMKG for those.

SOURCES

• National administrative boundaries, BIG
• OpenStreetMap basemap

WHAT WE COLLECT

Names, Ministry of Home Affairs codes, parent areas, area, and city, regency, district, village, and urban ward boundaries.

HOW WE PROCESS IT

• All data is stored in one process with its source address and edition date.
• BIG does not publish province outlines. We assemble them from the cities and regencies in each province, including parts BIG has not assigned to a city or regency.
• To name areas for earthquakes, volcanoes, air sensors, and your location, and to draw boundaries on the map, we simplify BIG's boundaries by about 1–3 metres for villages, districts, cities, and regencies, and about 11 metres for provinces. This remains within the source map's precision.
• We also retain full city, regency, district, village, and urban ward outlines for downloads: every point from BIG, with coordinates to six decimal places (about 11 cm).
• Areas BIG marks as 'Undefined Area' are not stored as districts or villages. Points there receive only their city or regency name.
• For hotspot counts by city and regency, boundaries are simplified further by about 100 metres for speed. A point within 3 km of a boundary, such as along a coast, is assigned to the nearest area.
• For weather and modelled air quality, each city and regency is represented by one point inside its largest contiguous part.
• When BIG publishes a new edition, other datasets are realigned. Area names for earthquakes, volcanoes, and air sensors are looked up again; the past 30 days of weather and modelled air are refetched at new representative points. Weather collected outside its area is deleted, not moved to another area.
• GeoJSON and Shapefile downloads are available by area or for all Indonesia. A province file includes that province and its cities and regencies, exactly as supplied by BIG; the largest is about 400 MB. The full Indonesia files include all provinces, cities, and regencies: about 2.2 GB (GeoJSON) and 0.8 GB (Shapefile). District and village files also preserve BIG's boundaries. Excel and CSV list provinces, cities, and regencies with their representative points.

ACCURACY AND LIMITS

• This is the June 2026 edition with 38 provinces. Most village boundaries are still indicative, meaning they have not been formally set.
• City and regency boundaries are drawn at 1:50,000 scale; district and village boundaries at 1:10,000. They may differ by several metres near borders.
• These boundaries serve data analysis, not legal boundary determination. Refer to the Ministry of Home Affairs and BIG for official boundaries.
• This dataset is not on an automatic schedule. We update it when the source publishes a new version.

SOURCES

• OpenStreetMap via Geofabrik: Indonesia on 1 January, 2015–2026

WHAT WE COLLECT

Road lines and their type from OpenStreetMap's copy of Indonesia on 1 January of each year since 2015.

HOW WE PROCESS IT

• Road types are grouped: main roads (toll, trunk, primary, secondary), local and village roads (tertiary, village links, residential, access roads), dirt tracks, and footpaths. Sidewalks, cycleways and steps are not counted.
• All vehicle roads are main roads, local and village roads, and dirt tracks. Footpaths are counted separately.
• Each stretch of road is measured on the earth's surface and counted in the village where its midpoint lies. A road crossing a boundary is split close to it. City and regency figures are the sum of their villages.
• Road per km² is road length divided by the area's land area according to BIG.
• A year whose copy we have not counted yet is shown as no data, not zero.
• GeoJSON and shapefile downloads are the OpenStreetMap lines themselves (one LineString per stretch) for one year and one province or city/regency, not city-centre points.
• The live map draws those lines, not a colour per city. From far away only main roads; zoom in for city and village roads.

ACCURACY AND LIMITS

• These are roads that have been mapped, not all roads that exist. A road enters OpenStreetMap when someone draws it, not when it is built.
• Indonesia's total rose from about 240,000 km in 2015 to 1.27 million km in 2020, mostly because existing roads were mapped. Since 2021 it has grown by under 5% a year, so recent change is more likely new road, but can still be mapping.
• Change per city or regency can be negative. It is a difference in mapped length, not a count of demolished roads: a road may be retagged, deleted from the map, or its midpoint may move into another area.
• Every road type comes from the highway tag mappers set in OpenStreetMap, not from government or Google data. Main roads are motorway, trunk, primary and secondary; local and village roads are tertiary, unclassified, residential and service; dirt tracks are track; footpaths are path.
• That type follows how mappers see the road being used, not the official national, provincial, regency/city or village status set by the government. That official status is not available as an open map.
• Road condition, such as paved or damaged, is not counted.

SOURCES

• OpenStreetMap via Geofabrik: Indonesia on 1 January, 2015–2026

WHAT WE COLLECT

Building outlines and their type from OpenStreetMap's copy of Indonesia on 1 January of each year since 2015.

HOW WE PROCESS IT

• Building types are grouped: homes, businesses (shops, offices, factories, warehouses, hotels), facilities (schools, hospitals, government offices, stations, places of worship), and other. Other includes buildings tagged only “yes”, plus garages and small sheds.
• All buildings are those four groups. A building drawn only as a point, tagged “no”, or tagged as an entrance is not counted.
• Footprint is measured on the earth's surface from the outline, with holes taken out. One building is counted whole in the village that contains its centre. City and regency figures are the sum of their villages. An outline larger than 1 km² is not counted: the 2022 copy contains one building=yes ring of about 127,000 km² covering Java, and that is not a building.
• Buildings per km² is the count divided by the area's land area according to BIG. Hectares are the ground area drawn on the map, not the floor area of upper storeys.
• A year whose copy we have not counted yet is shown as no data, not zero.
• GeoJSON and shapefile downloads are each building's outline for one year and one city, district, or village. A province and the whole country are not offered.
• The live map draws each building's outline wherever the map is open, from zoom 10. Further out, buildings become squares shaded by how many buildings are there.

ACCURACY AND LIMITS

• These are buildings that have been mapped, not all buildings that exist. A building enters OpenStreetMap when someone draws it, not when it is built.
• The national count rose from 1.9 million buildings in 2015 (about 39,000 ha of footprint) to 42.5 million in 2026 (about 469,000 ha). Until 2019 most of the rise came from existing buildings being mapped. Since 2024 it has grown by under 3% a year, so recent change is closer to newly drawn buildings, but can still be mapping.
• Change per city or regency can be negative. It is a difference in the mapped count, not a count of demolished buildings: a building may be deleted, retagged, or its centre may move into another area.
• Every building type comes from the building tag mappers set in OpenStreetMap, not from government or Google data. For example house and apartments become homes; commercial, retail and office become businesses; school, hospital, mosque and government become facilities.
• Most buildings in Indonesia are tagged only “yes”, with no type: on 1 January 2026, 97% of 42.5 million buildings fall under other. So the home, business and facility groups are far smaller than what is on the ground, and they are not suited to comparing facilities between areas.
• A building whose outline is also part of another building can be counted twice. Height and the number of storeys are not counted.