A Kyiv oblast fuel-depot fire on 23 March 2022 burned above 100 MW for hours. VIIRS M13 saturates at 659 K; Sentinel-3 SLSTR’s F1 channel does not, holding dynamic range above 450 K. That is the difference between an FRP estimate you can trust and a number that drifts by a factor of three. Our Kyiv pipeline runs VIIRS NRT every 10 minutes against MODIS, SLSTR, and Sentinel-2 — validated to BIAS +2.77 ug/m³ PM2.5 against KMDA stations.
Polar-orbiting and geostationary sensors together yield 8-12 overpasses per day across Ukraine. Photo: NASA/NOAA (public domain).
What we actually use
The operational pipeline of the WildFiresUA system — our complete stack for Ukraine — rests on four sources. VIIRS on NOAA-20/21/Suomi-NPP in NRT is the primary channel, polled every 10 minutes. MODIS NRT is secondary. Sentinel-3 SLSTR provides radiometric FRP quality. Sentinel-2 MSI supports post-hoc validation of burn scars through dNBR. The geostationary channel combines Meteosat SEVIRI and MTG-I1 with 15- and 10-minute cadence. Mixing sensors without care produces false alarms or missed events.
VIIRS 375 m: the workhorse
Three NOAA platforms: Suomi-NPP (2011), NOAA-20/JPSS-1 (2017), and NOAA-21/JPSS-2 (2022). A 3040 km swath yields 6-8 overpasses over Ukraine per 24 hours.
I-band 375 m: I4 (3.55-3.93 um MWIR, saturation at 367 K), I5 (10.5-12.4 um LWIR). M-band 750 m: M13 (3.973-4.128 um, saturation at 659 K, critical for FRP on intense fires), M15 (10.263-11.263 um). The Active Fire Product (VNP14IMG/VJ114IMG) applies the Schroeder et al. (2014) contextual-threshold algorithm: candidate pixels against a moving window of 3×3 to 21×21, and an anomaly test against local standard deviation.
Minimum detectable FRP: approximately 0.5 MW at night (flaming area 0.01-0.02 ha at 800 K), approximately 5 MW by day due to solar contamination at 3.9 um. Relative to MODIS at 1 km, VIIRS 375 m detects three times more fire pixels by day and roughly 25 times more at night. In forests it recovers 65% of pixels missed by MODIS; in low-biomass terrain, 83%. NRT latency is 3-4 hours.
Sentinel-3 SLSTR: the F1 channel that does not saturate
SLSTR on Sentinel-3A (2016) and 3B (2018). Channel S7 at 3.74 um saturates at 311 K (unusable for fires). The dedicated F1 fire channel at 3.74 um has an extended dynamic range above 450 K and does not saturate on typical wildfires. MODIS and VIIRS saturate their primary detection channels over intense fires, degrading FRP. SLSTR F1 with dual-gain avoids this. For fuel-depot fires burning above 100 MW for hours, SLSTR provides FRP estimates that can be trusted.
Xu et al. (2020): aircraft FRP comparison yielded r^2 = 0.9, slope = 1.1. Globally, SLSTR detects seven times more fire pixels than MODIS Terra. The SL_2_FRP product is distributed by EUMETSAT. The overpass at approximately 10:00 local time misses the afternoon peak.
Landsat 8/9 TIRS: post-fire only
TIRS-1 (10.6-11.19 um) and TIRS-2 (11.5-12.51 um), 100 m native resampled to 30 m. A 16-day revisit per satellite (8 days combined) does not support fire dynamics. TIRS remains the gold standard for post-fire burn severity through dNBR calculated from NIR and SWIR2 pre/post. The Key et al. (2006) classification: unburned <0.1, low 0.1-0.27, moderate-low 0.27-0.44, moderate-high 0.44-0.66, high >0.66. The 30 m resolution supports perimeter accuracy at approximately 0.09 ha.
GOES and MTG: geostationary cadence
GOES-R ABI: 36,000 km geostationary, Band 7 (3.9 um), Band 14 (11.2), Band 15 (12.3). 2 km at nadir. Cadence: full-disk 10 min, CONUS 5 min. Over Ukraine the relevant platforms are Meteosat SEVIRI (3 km, 15 min, FRP-PIXEL via EUMETSAT LSA SAF) and MTG-I1 (2 km, 10 min, launched December 2022). At a zenith angle of 55 degrees, the effective resolution over Kyiv is approximately 4-5 km. The 5-minute cadence identifies blow-up behaviour within the hour, which polar-orbiting platforms cannot structurally achieve.
The small thermal-satellite class
Since 2022 a new category of thermal satellites has emerged with 3-10 m GSD and AI processing onboard: GPU boards perform fire/no-fire classification in orbit and downlink alert packets only. Latency is 3-5 minutes, and 50-100 LEO satellites yield 10-30 minute revisit. The challenges: radiometric calibration drifts; uncooled detectors saturate on high-FRP events; gas flares (1400-1700 K) produce false positives; and ML updates remain a bandwidth problem.
NASA FIRMS API
Endpoint: https://firms.modaps.eosdis.nasa.gov/api/area/csv/{MAP_KEY}/{SOURCE}/{BBOX}/{DAYS}. Sources: VIIRS_NOAA20_NRT, VIIRS_NOAA21_NRT, VIIRS_SNPP_NRT, MODIS_NRT. Rate limit 5000/10 min. CSV fields: latitude, longitude, bright_ti4, acq_date, acq_time, satellite, confidence (l/n/h), frp (MW), daynight. NRT latency is 3-4 hours.
Algorithms and false positives
Polar-orbiting fire algorithms (MOD14, VNP14, SLSTR AF) share a common contextual-threshold architecture: an absolute threshold on mid-IR, a growing window from 3×3 to 21×21, and an anomaly test against local standard deviation.
Known false positives: gas flares (1400-1700 K, filtered with a static flare mask plus temporal persistence), industrial hotspots (steel mills, thermal power plants), hot bare soil, cloud edges, sun glint, and volcanoes.
Accuracy benchmarks
- VIIRS 375 m minimum: approximately 0.5 MW at night, approximately 5 MW by day (Schroeder 2014).
- MODIS minimum: 3.7-8 MW at night, 10-15 MW by day (Freeborn 2014).
- MODIS FRP uncertainty: +/- 26.6% (1 sigma) per pixel (Freeborn 2014).
- VIIRS-MODIS bias: M13 sits in the CO2 absorption band and underestimates; after atmospheric correction, +20.8% MODIS, +65.8% VIIRS (Li 2018).
- MODIS at 1 km misses approximately one-third of the real FRP (Sperling 2020).
- Sentinel-3 SLSTR: r^2 = 0.9, slope = 1.1 (Xu 2020).
- CNN on Sentinel-2: F1 0.84-0.94 across biomes (Zhang and Ban 2023).
A reported “FRP 50 MW” on MODIS means 50 +/- 13 MW, interval 37-63.
The Ukrainian operational pipeline
Since February 2022, FIRMS has become the de facto OSINT channel for Ukrainian fire monitoring. FIRMS does not distinguish military fires from agricultural or forest events. Our workflow cross-references VIIRS NRT with Sentinel-2 burn scars (10 m), CORINE land cover, and infrastructure polygons.
- VIIRS NOAA-20/21/SNPP polled every 10 minutes (primary).
- MODIS NRT at 15 minutes (secondary).
- DBSCAN clustering, eps = 1.5 km (haversine, 6371 km), min_samples = 2 to aggregate multi-pixel detections into a single event.
- Sentinel-3 SLSTR FRP-PIXEL for intense fires where VIIRS M13 is near saturation.
- Sentinel-2 MSI pre/post for dNBR validation.
- For fires in complex terrain (the Carpathians, Dnipro ravines) we downscale WRF 1 km winds to 30-100 m using USFS WindNinja.
Fuel-depot fires have a recognisable FRP profile: rapid growth above 100 MW followed by stable burning for hours. Hydrocarbon pool fires show a persistent 12 um blackbody signature beneath a soot layer. To forecast smoke transport from such sources to neighbouring regions we run NOAA HYSPLIT smoke-forecasting.
Peer-reviewed work with Ukrainian authorship
Our CJFR 2026 paper (cjfr-2025-0035), “Mapping and development of 30-meter landscape fuel data for Ukraine”, a joint paper with UkrHMI and the Marzieiev Institute. VIIRS hotspots localised a fire on 23 March 2022 in Kyiv oblast. The validated FCCS-fuel -> CONSUME -> FEPS -> CALPUFF pipeline against KMDA stations yielded BIAS +2.77 ug/m^3 PM2.5, RMSE 48, Pearson r = 0.40, within factor-of-two.
Our CNN paper (ISPRS Geospatial Week 2023): a CNN for burned-area mapping on Sentinel-2, 97% overall accuracy.
Copernicus ESOTC 2024: parts of Ukraine experienced approximately 25 additional days with FWI > 50. Cumulative forest loss 2022-2024: 965,000 ha (JRC 2024). A regional assessment for Luhansk and Donetsk is available in Trees, Forests and People (2024).
FAQ
Why not a single sensor? Each has blind spots. VIIRS cannot see through clouds. SLSTR misses the afternoon peak. Geostationary platforms offer coarse resolution. A stack delivers overlapping coverage.
Why does 375 m matter relative to MODIS 1 km? In a 1 km pixel, a 100 m^2 fire is 0.01% of the area. In a 375 m pixel, it is 0.07%, at the detection threshold.
Can PM2.5 be estimated from satellite alone? Only through FRP-to-emission with accumulated uncertainties: +/- 26.6% FRP plus the emission factor plus the dispersion model. Confidence intervals require cross-validation against ground stations.
Monitoring system | Fire map | Partners
Ukrainian startup ecosystem: follow TechUkraine and AIN.ua — the two leading outlets covering Ukrainian deep tech, climate tech, and environmental startups.
What to do today
- Check the YourAirTest air quality map for your city — recent PM2.5 readings.
- If the topic resonates, share with colleagues; we monitor referrals via Google Search Console.
- To contribute data (sensor measurements, regional models) reach out via the contact form.
References
- Schroeder W. et al. (2014). VIIRS 375 m active fire detection. RSE 143:85-96.
- Xu W., Wooster M. J. et al. (2020). RSE 248:111947.
- Freeborn P. H. et al. (2014). MODIS FRP uncertainty. GRL 41(6):1988-1994.
- Li F. et al. (2018). VIIRS-MODIS FRP comparison. JGR Atmospheres 123(9):4545-4563.
- Sperling S., Wooster M. J., Malamud B. D. (2020). Fire 3(2):11.
- Giglio L., Schroeder W., Justice C. O. (2016). Collection 6 MODIS. RSE 178:31-41.
- Key C. H., Benson N. C. (2006). Normalized Burn Ratio. USDA RMRS-GTR-164-CD.
- Zhang P., Ban Y. (2023). Deep learning burned area. ISPRS JPRS 203.
- Our peer-review CJFR 2026 — “Mapping and development of 30-meter landscape fuel data for Ukraine”, cjfr-2025-0035.
- Our CNN paper (ISPRS Geospatial Week 2023) — CNN burned area for Ukraine.
- Copernicus / JRC (2024). ESOTC; Ukraine wildfire 2022-2024.
- NASA FIRMS (2024). firms.modaps.eosdis.nasa.gov
- NOAA Air Resources Laboratory. HYSPLIT smoke forecasting. arl.noaa.gov/hysplit/smoke-forecasting
- USDA Forest Service. WindNinja — high-resolution wind for fire behaviour. ninjastorm.firelab.org/windninja
- Post-invasion wildfire impacts in Luhansk and Donetsk regions. Trees, Forests and People 18:100639 (2024). DOI 10.1016/j.tfp.2024.100639
- Wildfire Today — Ukraine news archive. wildfiretoday.com/tag/ukraine