Ignopticon

Technical details

Experimental research output for the NASA Space Apps 2026 MODIS/VIIRS challenge, not an official forecast. Do not use it for evacuation or tactical decisions. Numbers below come from MODELLING_LOG.md in the repository (entry numbers in brackets) and the result files it cites.

What it doesDataHarmonizationForecastingVerificationWhat did not helpLimits and next

1. What the system does

Fire watch shows 12-hour spread forecasts for the 76 fires in NASA's FEDS perimeter snapshots of 2 August 2026, with the burning perimeter, a most-likely case and a worst case, hotspots and wind. History is a calendar of satellite fire activity over the lower 48 states from 2000 to 2026, with MODIS made comparable to VIIRS, plus a daily map and an overlay of NASA FIRMS hotspots. A shorter overview, with the field-map features and the Space Apps summary, is on the About page. This page goes deeper on the modelling.

2. Data

SourceUsed for
MODIS Terra MOD14A1 and Aqua MYD14A1 (daily, 1 km)Fire record 2000 on (Aqua from 2002); inputs to harmonization
VIIRS S-NPP VNP14A1, NOAA-20 VJ114A1 (LAADS DAAC), and FIRMS hotspots incl. NOAA-21Target of harmonization (S-NPP, from 2012-01-19); hotspots at fire start and in the History overlay
NASA FEDS perimetersFire start state and truth, every 12 h
NIFC WFIGS incidentsNames, containment dates (display and scoring only)
NOAA HRRR (15-minute wind)Fire-model weather
gridMETDaily weather for the activity and growth models
LANDFIRE LF2025Fuels and canopy
GHS-POP R2023A, 2025 epochPopulation density layer and people-inside-forecast counts (~1 km; European Commission JRC, CC BY 4.0)
USGS 3DEP and ETOPO elevationTerrain

Formats and the layout of the files the site serves are in docs/DATASETS.md in the repository.

3. Harmonizing MODIS with VIIRS

MODIS (2000 on) and VIIRS (2012 on) see fire differently, so a raw MODIS count is not comparable with a VIIRS count. We built a per-cell model of one question: on a given day, in a 0.1 degree cell (about 10 km), would VIIRS S-NPP have seen fire, given what MODIS Terra and Aqua saw? Inputs are MODIS fire pixels and power, the same for neighbouring cells and the previous and next day, days since fire, month and location. It is a boosted-tree model trained on 2012-2024 overlap years, using only days VIIRS looked at clearly, so cloud is never counted as "no fire" [10].

Held-out years (2013, 2016, 2019, 2022; the model was trained on the other nine) [10]:

MethodAUCBrierLog lossExpected fire cell-days (observed 183,677)
MODIS fire used as-is0.7390.0012550.01733235,121
Per-month logistic rescaling0.7830.0008090.00543180,007
Boosted-tree harmonization0.9610.0007090.00393179,844

4. Forecasting

5. Verification

Headline subset: wildfires that were not contained at issue time and match a WFIGS incident, since forecasting unsuppressed spread for fires already contained is not meaningful. Truth is the FEDS perimeter 12 hours later; the issue time is 2 Aug 2026 12:00 UTC. Growth IoU is the overlap between the predicted and the observed new area; "no growth" scores 0 by construction.

Most likely, held-out, 7 uncontained fires (6 with growth) [27]Growth IoUGrowth recallWhole IoUMedian area ratio
No growth000.9330
Raw ELMFIRE0.1040.520.7443.7
Most likely (area-constrained)0.1130.200.8970.64

On all 12 held-out fires the worst case recovered about three quarters of raw ELMFIRE's growth recall at about half the false-alarm area (11.4 vs 20.5 km2 per fire) [25]. These fires were frozen before any model was run on them.

Where does it grow? With "how much" removed (every method is given the true growth area), ELMFIRE's arrival order was compared with a plain distance-to-fire buffer on 26 uncontained, incident-matched fires [29]:

n = 26 firesAUCRecall at observed areaGrowth IoU
ELMFIRE arrival time0.8630.2770.170
Distance buffer0.8800.2410.148
Stretched downwind0.8250.2200.133

ELMFIRE minus buffer on 2 Aug: AUC -0.017 (95% CI -0.045 to +0.006), recall +0.037 (-0.018 to +0.085), within the noise. On three later issue days, held out (no choice was made on them), the edge became clear: recall at observed area 0.273 / 0.197 / 0.172 (Aug 3, n=31), 0.331 / 0.252 / 0.224 (Aug 4, n=29), 0.361 / 0.282 / 0.252 (Aug 5, n=37) for ELMFIRE / buffer / downwind, with ELMFIRE above both baselines on every day (95% CIs above zero) [33].

Against an operational system. PyreCast (Pyregence's operational ELMFIRE), from its public archive, on the same fires, truth and valid time, issues 12, 13 and 14 Aug 2026 (33 uncontained fire-days, frozen before any run): growth located at equal area 0.328 ours, 0.262 PyreCast median, 0.267 buffer; ours better on 25 of 33 fire-days, per-fire mean difference +0.086 (95% CI +0.042 to +0.131), pooled +0.066 (-0.001 to +0.143). PyreCast's runs start up to 15 h earlier from their own footprint, ours from the FEDS perimeter at issue [37, 38].

Worst case, graded on capture. A worst case is judged by how much of the growth that actually happened falls inside it, at what area. The worst case captured 90.5% of observed growth on the 7 held-out uncontained fires (pooled), against 37% for raw ELMFIRE, at about 25 times the observed growth area [35]. Capture against area marked, 97 held-out fire-days (Aug 3-5), pooled [34]:

Area marked (x observed growth)Distance bufferDownwindELMFIREELMFIRE 11-run union
1x0.240.240.310.32
2x0.400.390.450.46
3x0.510.490.540.56
5x0.670.650.680.68
10x0.870.840.870.87

ELMFIRE's shape adds capture up to about three times the observed area (union minus buffer +0.08, +0.06, +0.04, CIs above zero); by five times and beyond, a plain buffer captures as much. So at 80% capture the ordering matters little; the worst case uses ELMFIRE's ensemble order anyway, so its outline follows terrain and wind where the runs reach, and it captured slightly more on held-out fires than a pure distance envelope (90.5% vs 88.9%) [35].

Caveats. Small samples, one issue date, FEDS edges are uncertain by about 375 m (comparable to the 12 h advance of many fires), many fires are small, and the held-out results are descriptive. In-sample results (the area conversion, development fires) are marked as such in the log.

6. What we tested that did not help

These are findings, not failures to report. Each was tested on development fires first and then on the 12 frozen held-out fires.

7. Limitations and what is next

Code and notes: github.com/scholar-e/Ignopticon. Full log: MODELLING_LOG.md. Overview: About.

Built by team Babylonian Belugas for NASA Space Apps 2026.