Fire MemoryPatagonia Español

NASA Space Apps 2026 · Challenge proposed by CONAE · Argentine Patagonia

Fire didn't increase ninefold. The satellite changed.

If the records from the MODIS (since 2000, 1 km pixels) and VIIRS (since 2012, 375 m pixels) satellites are merged as they come, hot spots in Patagonia appear to have shot up in 2012. Here are the 24 fire seasons converted to a single unit, with the method tested on years and regions it did not use for calibration.

is how much fire “increased” between 2003–2010 and 2012–2024 if the two records are added together without correction.
With the same instrument (MODIS)
between averages; explained by two large fire seasons
Harmonized series
VIIRS expressed as “what MODIS would have seen”
Detections processed
Terra, Aqua, S-NPP, NOAA-20 and NOAA-21 · 2000–2026

Hot spots per fire season in Argentine Patagonia

Fire season: July to June. The line that shoots up is the naive merge; the other two are comparable with each other.

The tool

Pick an area and a period.

Draw a rectangle on the map, type the coordinates or use one of the preset zones. The calendar shows, month by month since November 2000, the MODIS and VIIRS hot spots converted to the same unit, so that the satellite of origin no longer matters. Click a month and the calendar compares it with the historical median for that month in that area. It works for all of Argentina.

Loading the area calendar…

MODIS-Aqua-equivalent hot spots. One row per year, one column per month. Click a month to compare it.

The area

Each square is a 0.25° cell (about 25 km) coloured according to the hot spots in the selected period. The coordinates are snapped to that grid. Where a 0.5° cell recorded at least 100 Aqua detections in 2012–2018, the day and night factors are adjusted by calendar month from that area's own history; elsewhere the two national factors apply.

In which months does it burn?

The median for each month over the period and, beside it, the year of the selected month.

Which years had the most fire?

MODIS-Aqua-equivalent hot spots per calendar year.

Download what you are looking at

The monthly series and individual detections are available for any area of Argentina. The CSV retains date, time, fire day, satellite and power; static-source candidates remain identified.

The calendar by zone, week by week

The fire calendar, 2002–2026.

Each row is a fire season (July to June) and each column a week. The unit is always the same: MODIS-Aqua-equivalent hot spots. Pick a zone, see when it burns and which seasons were outside the usual range.

View as table

Total per fire season

MODIS-Aqua-equivalent hot spots. Click a bar to see that fire season on the map.

Most unusual episodes in the zone

Weeks above the 95th percentile for that time of year. The location is approximate.

    Where

    The map, season by season.

    Base map: Instituto Geográfico Nacional (Argenmap). Daily satellite image: NASA GIBS (VIIRS NOAA-20, true colour). 0.1° cells (about 11 × 8 km).

    Near real time

    The last few days, measured against 24 years.

    New hot spots this week, sorted by escape profile

    Each row is a separate fire seen by NOAA-20. The level comes from “the first look” triage (see further down).

    First finding

    Not everything that glows is a wildfire.

    VIIRS is so sensitive that it sees oil and gas flares. In Patagonia that carries a lot of weight: the NASA product flags some of them as “static source”, but thousands of detections that burn in the same square kilometre every month of the year are left unflagged. Unless they are removed, the map of Patagonian fire is to a large extent the map of the Neuquén Basin.

    Of what VIIRS detects in Patagonia
    classified by the filter as static or industrial sources
    What the NASA product flags
    the rest was found by the persistence filter
    Last few days, near real time

    Gas flares and static sources in the Neuquén Basin, as seen by VIIRS S-NPP

    Detections per year. At the bottom, those the product already flags; on top, those it does not flag and that burn all year in the same place.

    The relationship between the two sensors

    MODIS missed three out of four fires. It saw almost all the energy.

    When the detections are grouped into “fire objects” and MODIS-Aqua is matched against VIIRS S-NPP (same overpass time), the probability that MODIS sees a fire depends on its energy, and that curve is practically the same in northern Argentina and in Patagonia. The fires that MODIS missed are many, but small.

    What MODIS-Aqua missed in Patagonia, by day

    Out of the fire objects detected by VIIRS, 2012–2025.

    That is why the 2000–2011 decade is not lost: the small fires cannot be recovered, but the record captured most of the energy.

    Probability that MODIS-Aqua detects a fire, by its energy

    By day. Total energy of the object as measured by VIIRS (FRP, in megawatts).

    Day factor
    Night factor
    Handover: NOAA-20 relative to S-NPP
    0.98
    S-NPP stopped transmitting on 2 Nov 2026; the series continues with NOAA-20 and NOAA-21

    One daytime VIIRS hot spot is equivalent to MODIS-Aqua hot spots; one at night, to . Only the afternoon orbit is used (Aqua, S-NPP, NOAA-20/21): Terra passes at a different time of day and is left out of the series.

    Validation

    Tested in another region, then block by block across the whole country.

    Four ways of converting VIIRS to “MODIS equivalent” were calibrated only with northern Argentina between 2012 and 2018, and were asked to predict what MODIS-Aqua saw in Patagonia between 2019 and 2025. The simplest one won: a day factor and a night factor. The most widely used one, the monthly ratio over cells where both sensors detect, overestimates almost twofold.

    Predicted total over actual total

    1.00 is an exact match. Out-of-sample and out-of-region test.

    Test seasons in Patagonia

    Actual MODIS-Aqua detections and those predicted from VIIRS with the two factors.

    The second test leaves no region out. Argentina was split into 3° by 3° blocks. Each block was predicted with the two factors fitted on the rest of the country between 2012 and 2016, and compared with what MODIS-Aqua saw there between 2017 and 2021. A second analysis then redid it from scratch to try to refute it, shifting the grid and the years: the ranges below are its own.

    Predicted over actual, block by block

    1.00 is an exact match. Outlined: the blocks that fall outside the 0.80 to 1.25 band. Hover over a block to see its numbers.

    Where it holds and where it does not

    2017–2021 test, with the MODIS-Aqua clock still on time.

    The underlying finding

    Two twin satellites agree on only half of the fires.

    VIIRS flies on two identical satellites, S-NPP and NOAA-20: same instrument, same orbit, minutes apart. If a satellite took a census of fires, the two would record the same thing. That does not happen, and when both see the same fire they measure very different power for it. A satellite does not count fires: it takes a snapshot of something that flickers. That opens a door. Just as the number of fish in a lake is estimated by casting the net several times and seeing how many turn up again, with three satellites that pass almost together it is possible to estimate how many fires none of them saw.

    An experiment no one designed

    Between 20 March and 4 April 2024 NOAA repositioned NOAA-20 in its orbit, and the twin satellites went from 50 to 23 minutes apart. Of the 3 to 10 MW fires that one VIIRS records, how many its twin also sees. In a lighter shade, the years when they passed 50 minutes apart.

    Of the fires that existed, how many each satellite saw

    Argentina, by day, 2018–2025. The total number of fires is estimated by capture–recapture and is a minimum, so these percentages are upper bounds.

    Detection curve without ground truth

    When one VIIRS records a fire of a given power, does its twin see it? Does MODIS-Aqua see it? Argentina, by day.

    No ground data was needed: each satellite serves as a witness for the other.

    In Patagonia, the blind spot is the steppe

    Daytime fires that none of the three satellites saw, at a minimum. In Patagonia only those far from gas flares and static sources are counted.

    Attempts to break this result

    Three tests were written before running the calculation. The other four were added afterwards, looking for boring explanations that could manufacture the same number. Three forced an adjustment to what is reported.

    The black swan

    The yardstick moved: MODIS no longer passes at the same time.

    Harmonization also requires examining observation times. In Argentina, the median local solar time of Terra's daytime detections shifted from (2012–2020 baseline) to in 2025; Aqua shifted from to . Thermal activity can vary within a day. We measure how the MODIS/S-NPP ratio changed on days with coverage from both, without treating VIIRS as a perfect measure of fire. Observation time may influence this comparison, but the data do not fully separate its effect from sensitivity, geometry and processing.

    The drift

    Difference, in minutes, in the local solar overpass time relative to the 2012–2020 average. Daytime overpass, over Argentina.

    What MODIS sees depending on the time it passes

    Hot spots seen by MODIS, measured against VIIRS, with its original clock and with the drifted clock. Aqua at 13:55 = 100. The direction of the three firm changes is the one predicted by the daily fire cycle. The horizontal axis runs from 6 to 6, like the fire day; the vertical axis is logarithmic: the same distance is the same percentage change.

    What MODIS sees by day, relative to VIIRS

    Change in the MODIS / S-NPP hot spot ratio relative to 2012–2020. All of Argentina. The change for Aqua does not exceed what was already moving with the stable clock.

    What MODIS sees at night, relative to VIIRS

    The same ratio, night-time overpass.

    How it was put to the test

    The predictions, with their sign, were written before looking at the result. Then an independent reviewer was instructed to refute it: three of the four numbers held up, the fourth did not, and the reading ended up more modest than at the start.

    A pattern

    On the steppe, fire rests on Sundays.

    Nature does not know what day of the week it is. If fire starts vary with the working calendar, someone is lighting them. Taking the first day of each fire: on the Patagonian steppe and shrubland fewer fires start on Saturdays, considerably fewer on Sundays and also fewer on public holidays, and on Mondays they come back with no rebound. In northern Argentina, half as many start on Sundays. Gas flares, which depend on no one, show no such rhythm.

    Fire starts by day of the week

    The Monday-to-Friday average is set to 100. Fires followed day by day with VIIRS S-NPP, 2012–2025.

    What can be anticipated and what cannot

    There is no crystal ball. There is a first look.

    Three ways of anticipating fire were tested, always on fire seasons the model did not see. Two did not work and one did. All three are here, with their numbers.

    Will next week be unusual?

    By zone. AUC on onsets (0.5 is chance).

    It does not improve with soil moisture or weather.

    Will the fire season be larger than normal?

    Predictor fixed in advance. AUC by zone type.

    A hint, only on the steppe.

    If it starts small, does it escape?

    With what the satellite sees on the first day. AUC.

    Yes: the first look separates the ones that escape.

    The watch list

    What share of the escapes ends up inside the list, depending on what percentage of the new hot spots is watched (those with the highest score).

    L-band (NISAR, SAOCOM): the hypothesis that remains open

    • The 50 km weather added nothing to any of the tests. The missing variable is fuel moisture at valley scale.
    • NISAR has published calibrated data since July 2026 (observations since 17 Jun 2026): it does not yet cover a full fire season, so it cannot be validated.
    • SAOCOM (CONAE) has observed in L-band since 2018; the documented soil moisture product is for the Pampas region.
    • Once there is a history, the test is the same one used here: beat the first look on unseen seasons.

    Where it started, how far it got

    Andean fire runs east.

    Each fire is followed day by day from its first detection. That shows where it began, how far it got and on what bearing, and makes it possible to ask whether that experience is useful for the next one. Pick one of the largest fires on record: the blue dot is the origin, the colour advances with the days and the line marks the farthest point it reached.

    Reach from the origin, day by day

    Kilometres to the farthest point with fire. The other two lines are the reach fan learned from the fires that escaped in that type of zone.

    Spread rose · Andean forest

    Spread rose · steppe, shrubland and coast

    Is it useful for a new fire?

    Predicting the bearing in seasons that were not used for learning. Hits within ±45°.

    History helps. The 50 km wind adds nothing.

    The synthesis

    The brain: small laws, wired together.

    A fly solves flight with a small circuit of simple parts. This is the same thing for fire: each part is a short law that got in only if it passed a judge written in advance, and together they take a raw satellite data point all the way to a decision. Click a part to see what it says, how it was judged and the result.

    Pruned synapses

    What was tested with the same rigour and did not pass. They are as useful as the others: they say where not to look.

    The brain reading this week's hot spots

    Each new hot spot goes through the five layers. It is what the circuit says with what it sees today; it is not a weather forecast.

    View all parts as table

    To review the work

    How it was made, and what it does not do.

    Method

    • Data: NASA FIRMS, annual files for Argentina (MODIS 2000–2025, VIIRS S-NPP 2012–2025, VIIRS NOAA-20 2018–2025) and the last-7-days files for South America.
    • Fire day: from 06:00 to 06:00, Argentina time, so that the afternoon overpass and that night's overpass count together.
    • Cleaning: nominal or high confidence; static sources are left out (those the product flags and those that burn in 12 or more different months, spread over 8 or more calendar months and 3 or more years), and so are the oil-field cells.
    • Unit: MODIS-Aqua-equivalent hot spots. Aqua until January 2012; after that, VIIRS times a day factor and a night factor. In the area explorer, where a cell has its own history (at least 100 Aqua detections in 2012–2018), those factors are adjusted by calendar month; the validation tests evaluate the two national factors.
    • What is unusual: the 7-day rolling sum compared with the same time of year (±15 days) in the 2002/03 to 2024/25 fire seasons.
    • Validation: retrospective evaluation with separate years and regions; the static-source filter uses the complete archive; intervals by resampling seasons.

    Limits

    • The small fires of 2000–2011 are not reconstructed: the series becomes comparable “downwards”, to what MODIS could see.
    • Validated for Argentina. It is not a global standard.
    • A hot spot is the centre of a pixel with fire, not a fire event or a burned area.
    • The zone boundaries and the border with Chile are approximate, not official.
    • January to October 2026 is not loaded, and for November only the last few days: a FIRMS API key is needed.
    • In the Northern plateau about 8% of the hot spots are right next to static sources; the near-real-time data do not carry the type flag, so a new gas flare can pass for a wildfire.
    • In the Patagonia 2019–2025 test: total difference, 2.1%; typical monthly error, a factor of 1.37, evaluated on months with at least five Aqua detections.
    • The equivalence holds for sums, not day by day: VIIRS records low activity on days when MODIS recorded zero. That is why the calendar goes by week.
    • It does not add up in every zone: in Tierra del Fuego (peat fire, few MODIS detections) and in the Northern plateau (temporary gas flares) the conversion overestimates. The calendar warns about it in each zone.
    • The fire no satellite saw is a lower bound, not a measurement: it assumes that every detection is fire and depends on how much time passes between one satellite and the other, which is why it is reported by period. It counts fires that were burning while the satellites passed; those that burned between overpasses are not included. Other, less cautious estimators give between 20% and 30% (and one, unstable, 61%).
    • MODIS changed its overpass time: since 2024 what a single Terra or Aqua overpass sees is not comparable with what it saw before. Here it does not affect the calendar, which uses Aqua only until January 2012.
    • Before July 2002 it is an estimate: those months come from Terra, with a typical monthly error of a factor of 1.6, and are shown only in the area explorer.
    • The area explorer works with 0.25° cells (about 25 km) and by month; the calendar by zone, by week.
    • At night, three out of four fires in Patagonia that pass the filter are within about 15 km of a static source: it is likely that some of them are gas flares that the filter does not remove.