NASA Space Apps Challenge 2026 · Puerto Madryn
Harmonization of MODIS and VIIRS Hot Spots · Proposed by CONAE

Fire Memory
Patagonia

Fire did not multiply ninefold.
The satellite changed.

Fire Memory corrects the record and shows where you can trust it.

Team Marx
Matías Ceballos

Landsat 9 image of smoke from the Valle Magdalena fire in Parque Nacional Lanín, Patagonia, on 9 February 2025
Valle Magdalena · Neuquén · 9 February 2025
NASA Earth Observatory / Michala Garrison · Landsat / USGS

The problem

Fire did not increase ninefold.
The satellite changed.

MODIS records 1 km pixels; VIIRS records 375 m pixels. Adding their detections directly creates a jump when the instrument changes.

×9.5
apparent increase from merging the records directly
×1.8
activity ratio after harmonization into one comparable unit

The same Patagonian record, processed two ways. These are ratios of recorded thermal activity, not counts of unique fires or burned area. Harmonization removes a large instrument-driven discontinuity; it does not show that fire activity stayed constant.

The proof

Calibrated in the north,
tested in Patagonia: 2.1% off.

CALIBRATENorthern Argentina2012–2018 · day and night factors TESTPatagonia · MODIS-Aqua reference2019–2025 · different years and region 2.1%difference in total

The 2.1% is the difference in the whole test total. Typical monthly error was ×1.37 on months with at least five Aqua detections. The app also maps where regional conversion fails.

NASA FIRMSMODIS + VIIRS · five satellites Filter static sourcesflags · persistence · industrial cells One common unitAqua-equivalent hot spots · day / night Calendar · maps · downloads6.35 million detections · 2000–2025

Retrospective evaluation: static-source preprocessing uses the full archive. Production factors were refitted after testing. The aggregate result is not a guarantee for each month or selected area.

NOAA-20 and NOAA-21 extend the record after S-NPP data stopped arriving in our FIRMS feed in early November 2026. Spacecraft status is not inferred from this feed gap.

Why it matters here

Patagonia’s fire record includes industrial heat

≈37%

of Patagonian VIIRS detections are classified as static or industrial by the filter.

Around Vaca Muerta, persistent heat can be confused with vegetation fire. Product flags, recurrence and industrial cells help separate the signals.

36.6% before filtering; 15.3% from product flags alone. Classification is not a confirmed inventory of gas flares. The filter can also exclude vegetation fire in industrial cells.

Real Fire Memory atlas showing the observed thermal detections of the 2015 Cholila fire in Chubut on white 3D terrain
Cholila · Chubut · Observations replayed on 3D terrain
Smooth heat shows detection density, not a burned perimeter or a spread forecast.
USER EXPERIENCE Draw an area Read the calendar Month vs historical median Download CSV or PNG

The discovery in the record

The yardstick moved

In 2025, Terra’s recorded daytime median was 69 minutes earlier and Aqua’s 68 minutes later than in 2012–2020.

The detection ratios against S-NPP changed too:

−35%
Terra · day
+28%
Terra · night
−31%
Aqua · night

A change in the observation system can look like a change on the ground. The orbital drift is known; this project measured its timing and sensor-specific differences in the Argentine record.

Real Fire Memory satellite-clock viewer with Terra, S-NPP and Aqua at their recorded local solar observation times
The satellite clock · recorded local solar time, year by year
NASA technical spacecraft renders; positions are explanatory.

Observed ratios are not changes in fire activity and do not isolate the causal effect of timing. Sensitivity, geometry and processing also matter. The project retained these three changes and rejected daytime Aqua as a robust effect.

Why trust the method

13 laws survived.
22 did not.

13

findings passed the project’s validation criteria

22

candidate laws tested and discarded

  1. Write the claim and what would refute it before running the test.
  2. Test on withheld years or regions, with shuffled-data controls where applicable.
  3. Keep the failures visible. The app shows rejected ideas and regional limits.

The findings register retains 25 entries: 13 validated, 4 fragile, 2 hints, 2 operating rules and 4 descriptive results. “Validated” refers to the project’s stated tests; it is not a claim of universal laws.

For a brigadista comparing this week with 23 fire seasons

“Is this week unusual here?”

Fire Memory puts the satellite record in one unit, compares recent activity with history and shows where the conversion can be trusted.

Show the map of where it fails. Regional limits and missing coverage remain visible.

Weekly regional context: 2002/03–2024/25. The area calendar compares complete months with their historical median. Intended monitoring support, not demonstrated improvement in firefighting outcomes.

Real Fire Memory calendar and selected Patagonian area with a complete month compared against its historical median
Draw a Patagonian area → calendar → complete month vs median → download
Then open the regional reliability map.

Fire did not multiply ninefold. The satellite changed.
Fire Memory corrects the record and shows where you can trust it.

Argentina · Retrospective evaluation · Small fires before 2012 cannot be reconstructed · January–October 2026 is not loaded · Observed heat is not a spread forecast.