Module 01 · Detect · Be An Earth System Trend Detective!
One fire record from 2000 to 2025, on a single sensor scale.
Interactive and fully static: everything runs in your browser from precomputed data. Use the "Experiment briefing" button inside the tool for a guided walkthrough.
Fire managers can't compare a 2005 fire season with a 2023 one, because NASA's fire record switches sensors in 2012 and VIIRS sees about 5.6 times more fire than MODIS. Without one consistent record, no one can say if fire is really increasing.
NASA has watched fires from orbit since 2000, but the record is split. MODIS (1 km pixels, from 2000) and VIIRS S-NPP (375 m pixels, from 2012) see fire differently: over Canada since 2012, VIIRS reports about 5.6 detections for every MODIS detection.
So a fire manager can't put a 2005 season next to a 2023 season and trust the comparison. Hot Spots merges both records into one burning-activity calendar for any area, so a trend can be read across the sensor switch.
The data story · Detect
Five beats: the raw signal, what changed, how sure we are, where it gets subtle, and who acts on it.
01 · The signal
NASA FIRMS records every fire a satellite sees: MODIS since 2000, VIIRS since 2012. Over Canada since 2012, VIIRS reports about 5.6 times more fire than MODIS for the same ground, so the raw record jumps when the sensor changes.
Hot Spots joins them into one 2000–2025 calendar per area, so a 2005 season and a 2023 season sit on the same scale.
02 · What changed
MODIS MCD64A1 burned area for tile h11v03 (the boreal forest of northeastern BC, Alberta and Saskatchewan) shows the ground that actually burned each April–October. 2023 burned 28,262 km², more than double any earlier season in the record.
Across 2003–2025 the Sen's slope is +272 km² a year.
03 · How sure we are
MODIS detections over Canada rise by +3,019 a year. The 95% confidence interval runs from +81 to +5,184: wide, but every plausible slope is above zero.
p = 0.035 means that if there were no real trend, a pattern this one-directional would turn up by chance about 3 times in 100. The harmonized count (p = 0.020), fire radiative power (p = 0.011) and MCD64A1 burned area (p = 0.004) all point the same way.
Is fire activity in Canada rising?
All four agree: rising04 · The nuance
At province level the harmonized series is significant in five provinces but MODIS-only in none. The harmonized record joins a model back-cast (2003–2011) to observed VIIRS (2012 on), so a step at 2012 can pass for a trend. We call those tentative.
MCD64A1 here is one tile, not all of Canada, and the biggest season sits at the end of the record. Monthly-total error for Canada is worse than the baseline (39.4% vs 16.7%), even though per-cell error is better.
05 · Who acts on it
Fire-management agencies and forest planners compare seasons to set crews, budgets and prevention work. A record that doesn't jump when the sensor changes means a busy season reads as a busy season, not a sensor artefact.
Unusual-month flags and calibrated intervals tell an analyst which months to check first, and how much to trust each number.
Launch the dashboardWhat this means
+272 km²more burned ground each yearIn northeastern BC, Alberta and Saskatchewan, 2003–2025 (MCD64A1, p = 0.004).
One tile; 95% CI +57 to +503 km² a year.Is fire activity in Canada rising?
All four agree: risingWhat this means
+272 km²more burned ground each yearIn northeastern BC, Alberta and Saskatchewan, 2003–2025 (MCD64A1, p = 0.004).
One tile; 95% CI +57 to +503 km² a year.Who uses it
The path through the dashboard, from the situation they start in to the decision it changes.
Alberta or British Columbia
2023 was enormous, but they can't say whether their province is in a multi-year rise: VIIRS sees about 5.6× more fire than MODIS since 2012, so 2005 and 2023 aren't in the same units.
Pick the Alberta or B.C. preset, or draw a fire district.
Read the disagreement map and the held-out test card, including the monthly-total miss.
Scroll the 2000–2025 calendar: unusual months outlined, hover for count, z-score and interval; download the CSV.
On the module page, switch the trend between harmonized, MODIS-only and fire radiative power.
Brief a national rise on the conservative record (MODIS-only +3,019 detections a year, p = 0.035), but not a provincial trend where MODIS-only isn't significant.
One calendar that can sit next to a 2005 season, a CSV, a known error, and a trend statement no stronger than the conservative test.
An illustrative journey built from what the dashboard does today, not a field-tested outcome.
Launch the dashboardA town or regional district
They need to know whether this month is unusual for their area, in words a mayor can use.
Draw a box around the community.
Read the calendar for that box; a cyan outline marks a month with |z| > 3.5 against its 26-year median.
Hover for the z-score and interval; download the CSV for the briefing pack.
Either "this month is flagged unusual against the 2000–2025 median for this box" or "it isn't", with the interval attached.
A single calendar tile and a sentence that is dated, bounded and honest.
An illustrative journey built from what the dashboard does today, not a field-tested outcome.
Launch the dashboardLower error than the best baseline on Canada's held-out 2022–2025 data (498,302 cell-months).
Lower error than the best baseline in Argentina (193,472 test cell-months). Total-count bias fell from −24.4% to −1.6% after 2021 calibration.
Interval coverage on cells where VIIRS saw fire, Canada. Argentina: 90.1%.
FIRMS detections behind the model (Canada and Argentina, 2000–2025).
Unusual months flagged across Canada's 26-year calendar.
Honest miss: monthly-total error in Canada is worse than the baseline, even though per-cell error is better. Argentina too (16.6% vs 12.2%). Both are shown in the demo.
Held-out test data, compared against baselines. One seed per run, not a published benchmark.
Visuals
Earthdata upgrade · MODIS MCD64A1
We added the MODIS MCD64A1 burned-area product (500 m, monthly), downloaded with an Earthdata login: 161 monthly files for tile h11v03, which covers the boreal forest of northeastern British Columbia, Alberta and Saskatchewan, about 50–60°N, April to October, 2003–2025. Detections tell you where fire burned at the moment of an overpass; burned area tells you how much ground it took. Summed per season, burned area in this tile rose by about 272 km² a year (Sen's slope, 95% CI +57 to +503; Mann-Kendall p = 0.004, also significant after the Hamed-Rao check). 2023 was the largest season at 28,262 km².
One MODIS tile, not all of Canada. A single huge year (2023) sits at the end of the record, so the slope is sensitive to the last few seasons. Unmapped pixels average 1–3% per season.
Trend layer · Fire
23 years of NASA FIRMS active fires (2003–2025), summed per province and tested with Mann-Kendall and Sen's slope. Switch between the harmonized record, the single-sensor MODIS record and fire radiative power to see which trends hold up.
Mann-Kendall (α = 0.05) and Sen's slope with 95% CI, one value per year. Hover or tap the chart for yearly values.
Where it is changing
The approach
NASA FIRMS country archives (MODIS C6.1 and VIIRS S-NPP) for Canada and Argentina, binned into 0.25° cell-months.
An empirical-Bayes ratio per cell: how many VIIRS detections each MODIS detection is worth there, shrunk towards the regional ratio.
A LightGBM model corrects the baseline using MODIS fire power, day/night, Aqua/Terra and neighbouring months. Trained 2012–2020, tuned on 2021, tested on 2022–2025.
Conformalized quantile intervals, calibrated per MODIS-count stratum, with a 90% coverage target.
2000–2011 MODIS is converted to VIIRS scale. Unusual months are flagged with a robust z-score (|z| > 3.5 against that calendar month's 26-year median).
Pick a region or draw any box: the calendar, disagreement map and CSV export are computed client-side from precomputed tiles.
Limits we know about
| Dataset | How it's used | Access | Link |
|---|---|---|---|
| NASA FIRMS MODIS Collection 6.1 (MCD14ML) | Active fire detections, Terra + Aqua, 2000–2025 | Public, no login | source |
| NASA FIRMS VIIRS S-NPP 375 m (VNP14IMG) | Active fire detections, 2012–2025 | Public, no login | source |
| Esri / OpenStreetMap basemap via Leaflet | Map context only | Public, no login | source |
| NASA FIRMS MODIS + VIIRS (trend layer) | Annual detections and FRP per province, 2003–2025; Mann-Kendall + Sen's slope | Public, no login | source |
| MODIS MCD64A1.061 burned area (module upgrade) | Monthly 500 m burned area, tile h11v03, Apr–Oct 2003–2025; Mann-Kendall + Sen's slope | Earthdata login | source |
| Natural Earth admin-1 boundaries | Province outlines for the trend map (public domain) | Public, no login | source |
For the judges
Gives fire managers and researchers a 26-year, single-scale fire calendar for any box on the map, so they can see whether a season is truly unusual.
Learns the cross-sensor correction on top of a strong per-cell baseline, rather than one global ratio, and attaches calibrated uncertainty to every back-cast month.
Strict time split (train 2012–2020, tune 2021, test 2022–2025), four baselines and coverage checks. It reports where it loses (Canada monthly totals) as well as where it wins.
Trend Detective: tests 23 years of NASA FIRMS fire activity per province with Mann-Kendall and Sen's slope, and shows why the sensor switch has to be harmonized before you call a trend. Built entirely on NASA FIRMS MODIS and VIIRS records.
One-click presets, draw-your-own area, hover details, CSV export and a built-in briefing deck that explains each result in plain language.
First designed around the related Space Apps 2026 brief Harmonization of MODIS and VIIRS Hot Spots; here it is the detect module of one system in Be An Earth System Trend Detective!