The global wildfire detection ecosystem in 2026: satellites, sensor networks, AI, drones

4 August 2023, 06:37 local, Maui. NOAA-20 registered a 470 K thermal anomaly inside a 375 m pixel over Lahaina. NASA FIRMS published the alert 12 minutes later. A CAL FIRE dispatcher saw the push 47 minutes after that and dismissed it as a controlled burn. By visual confirmation, 17 buildings were gone. Final toll: 102 dead, 2,200 buildings destroyed. Detection without an operational loop does not save lives.

Why one technology is not enough

There is a temptation to build a detection system on a single “silver bullet” — either satellites, cameras, or drones. None of these technologies alone covers operational requirements.

Polar satellites with VIIRS and MODIS sensors provide a 375 m or 1 km pixel but pass over a given point only 4-6 times per day. Between passes there is a blind window of 3-6 hours during which a fire can grow from a point source to a 100 ha front.

Geostationary platforms GOES-R/MTG observe continuously at 1-10 minute intervals but have a 2 km pixel and lower contrast for small sources. Camera networks like AlertWildfire see fires at night better than satellites, but are limited by terrain and meteorological visibility. Drones provide detailed information but are constrained by 25-60 minute battery life and the regulatory framework of the pilot’s direct line of sight.

Justice et al. (2002) in their fundamental work on the MODIS Active Fire Product clearly justified why satellite algorithms are built on multiple spectral channels and spatio-temporal consistency rather than a single temperature threshold (Justice et al., 2002, Remote Sensing of Environment). Wooster et al. (2012) showed that even for advanced geostationary platforms such as MSG SEVIRI, the detection completeness for fires below 1 MW radiative power is around 30%, which immediately requires complementing geostationary data with higher-resolution polar sensors (Wooster et al., 2012, Remote Sensing of Environment).

The conclusion is clear: operational detection is a multi-layered architecture with data fusion, not the choice of a single instrument.

Space layer: polar systems

VIIRS (Visible Infrared Imaging Radiometer Suite) is the foundation of contemporary global detection. Mounted on Suomi NPP (launched 2011), NOAA-20 (2017), and NOAA-21 (2022), VIIRS provides a 375 m pixel in I-Band mode and 750 m in M-Band mode. The VNP14 algorithm uses channels I4 (3.74 µm) and I5 (11.45 µm) to compute the contrast index between a hot spot and the background. Schroeder et al. (2014) showed that the 375 m VIIRS product detects approximately 4 times more small fires than the 1 km MODIS at the same error rate (Schroeder et al., 2014, Remote Sensing of Environment).

MODIS (Moderate Resolution Imaging Spectroradiometer) on Terra (1999) and Aqua (2002) satellites. The Collection 6 MOD14/MYD14 algorithm has been the gold standard of fire science for two decades, but both satellites are ageing. Giglio et al. (2016) describe Collection 6 as a 50% reduction in commission errors compared to Collection 5, achieved through stricter neighbourhood criteria (Giglio et al., 2016, Remote Sensing of Environment).

Sentinel-3 SLSTR from ESA. Two satellites, Sentinel-3A (2016) and Sentinel-3B (2018), carry the Sea and Land Surface Temperature Radiometer with channels F1 (3.74 µm) and F2 (10.85 µm) designed specifically not to saturate at high temperatures: F1 saturates near 650 K instead of 470 K as on VIIRS I4. This makes SLSTR a unique instrument for assessing the intensity of large fronts. Wooster et al. (2021) validate the SLSTR FRP product against ground experiments in detail and report an RMSE of about 18% (Wooster et al., 2021, Remote Sensing).

NASA FIRMS (Fire Information for Resource Management System) is the global distributor of MODIS and VIIRS products in near real time. Access through FIRMS with a 3-hour delay for the standard stream and approximately 60 minutes for the US/Canada Land Atmosphere Near real-time Capability for EOS (LANCE-MODIS).

Space layer: geostationary systems

Geostationaries view a single hemisphere continuously. This makes them indispensable for tracking fire dynamics throughout the day.

GOES-R / GOES-18 / GOES-19 from NOAA. The ABI (Advanced Baseline Imager) instrument has 16 channels, of which channel 7 (3.9 µm) is used for hot-spot detection. Spatial resolution is 2 km at nadir. The Fire Detection and Characterization (FDC) algorithm, described by Schmidt et al. (2017), provides updates every 5 minutes for CONUS and every 1 minute for mesoscale sectors (Schmidt et al., 2017, Atmospheric Environment).

Meteosat Third Generation (MTG-I1) from EUMETSAT, launched in December 2022. The Flexible Combined Imager (FCI) has a 3.8 µm channel with improved dynamic range compared to SEVIRI. Coverage: Europe, Africa, Middle East. Mission description on the EUMETSAT site.

Himawari-9 from JMA — a geostationary satellite over the Pacific Ocean with the AHI instrument and a 3.9 µm channel. It covers Australia, Southeast Asia, and Japan.

FY-3 and FY-4 are Chinese polar and geostationary platforms. FY-3D carries MERSI-II with a 3.72 µm fire channel. FY-4A provides geostationary monitoring of East Asia. FY products are available through CMA, but integration into international fusion systems remains limited as of 2026.

Ground layer: camera networks

AlertWildfire is the largest American ground-based network, with over 1,100 cameras across California, Nevada, Oregon, Idaho, Washington, and Colorado. The system was launched by the University of Nevada, Reno in partnership with UC San Diego and the University of Oregon. High-resolution PTZ (pan-tilt-zoom) cameras are deployed mostly on mountain peaks for maximum viewshed. Public access via alertwildfire.org.

UC San Diego ALERTCalifornia studies (published by Govil, Welch, Ball, Pennypacker, 2020) show that a human operator detects smoke on average 16 minutes before an official 9-1-1 call, while an automatic AI detector reduces this time by an additional 7-10 minutes (Govil et al., 2020, Remote Sensing).

The European sector develops camera networks primarily through commercial operators. Major players in the “AI camera detector” category (including a US and a European leader) deploy combined RGB and thermal camera networks with neural-network analysis. Greece, Spain, and Portugal operate state CCTV networks for forest monitoring, integrated with regional civil protection centres.

Australia: Bushfire CRC and CSIRO have run a series of validation experiments with camera networks. Allison et al. (2016) analyse the trade-off between camera count, installation height, and territorial coverage completeness (Allison et al., 2016, Sensors).

Airborne layer: aircraft and drones

The United States operates a fleet of MAFFS (Modular Airborne FireFighting System) aircraft on C-130s and an aerial sensor fleet NIROPS (National Infrared Operations) — aircraft with thermal imagers that scan fires at night to build front maps. NIROPS is the operational accuracy benchmark for tactical map updates. An overview is presented in the USFS programme document USFS Aviation.

Canada uses CL-415/CL-515 fleets for suppression and FireWatch aircraft for reconnaissance. Australia operates AVIN (Aerial Intelligence Network) with thermal imaging sensors on BAe-146 aircraft.

UAS / RPA drones are actively tested by USFS in the USFS UAS Program. Restas (2015) provided an early review of multispectral drone capabilities for post-fire assessment (Restas, 2015, Procedia Engineering). In 2023-2025, leading agencies gradually deployed nighttime multi-drone swarms for front map updates instead of expensive crewed aircraft.

Software layer: AI fusion

The final layer — software — combines raw data from various sources into a single operational picture. It does not detect anything new but excludes what should not be detected (commission errors) and prioritises what requires immediate response.

Basic techniques: spatio-temporal clustering (DBSCAN, ST-DBSCAN), Bayesian fusion of probabilities from different sensors, random-forest contextual classifiers (land-use type, prior fire history, meteorology).

Modern 2024-2026 approaches: U-Net and Transformer models for smoke detection on CCTV (covered in detail in our article on AI fire detection), federated learning for cross-validation between regional centres, Graph Neural Networks for modelling spread with pixel-neighbourhood awareness.

Pinto et al. (2020) demonstrate CNN segmentation of burned areas using Sentinel-2 with F1 ≈ 0.89 on a Southern Europe test set (Pinto et al., 2020, Remote Sensing). Hu et al. (2018) review CNN architectures for video smoke detection, with emphasis on the problem of false alarms on fog and clouds (Hu et al., 2018, Sensors).

Regional comparison of national systems

The key observation: the presence of camera networks distinguishes mature fire agencies from those that rely solely on satellites. Camera networks are an investment of €50,000-200,000 per point plus operational costs, and so are deployed where fire risk is high and the territory is compact (California, the Mediterranean). For countries with large territories and uneven risk (Brazil, Canada, Ukraine) the foundation remains satellite-based.

Detection latency: real numbers

Time from fire ignition to a confirmed alarm is the principal operational metric.

Csiszar et al. (2014) estimated the average latency of the global VIIRS network (Suomi NPP) at about 110 minutes from observation to publication on FIRMS, with a 15-minute median lag for Direct Broadcast zones in the US and Canada (Csiszar et al., 2014, JGR Atmospheres).

Detection errors: commission and omission

No system delivers 100% accuracy. Two key errors:

Commission error (false positive): the system says “fire” where there is none. Typical causes: hot roofs of industrial buildings, metallurgical flares, sun glints from water bodies, technological hot spots (blast furnaces, oil and gas flares), and agricultural burns classified as uncontrolled fires.

Omission error (false negative): the system misses a real fire. Causes: small size (below the sensor’s sensitivity threshold), dense cloud cover, vegetation canopy concealing a ground fire, a nighttime fire in the gap between polar satellite passes.

Schroeder et al. (2014) for VIIRS 375 m estimated commission error at 1.2% and omission error at 8% on a CONUS test set. Wooster et al. (2012) for SEVIRI: 70% omission for fires under 1 MW. Roberts and Wooster (2008) analysed the MSG SEVIRI FRP product over Africa and showed that for savanna fires the typical sensitivity is 25-50 MW (Roberts and Wooster, 2008, IEEE TGRS).

A separate major issue is sensor saturation. VIIRS I4 saturates near 367 K in fire-detection mode and 470 K with reprogrammed parameters. For very intense fires (such as the Australian Black Summer 2019-2020) this means that FRP is underestimated by a factor of 2-5. Sentinel-3 SLSTR with the F1 channel (650 K) is the best available solution against saturation for civilian satellites.

Fusion and open standards

The modern fusion approach relies on open data standards. NASA FIRMS publishes points in CSV, KML, GeoJSON, and WMS formats. Copernicus EFFIS (European Forest Fire Information System) integrates VIIRS, MODIS, MSG SEVIRI, and MTG FCI into a single database through the EFFIS interface. JRC coordinates the Global Wildfire Information System (GWIS), which gathers data for 100+ countries.

Ukrainian operational services can join EFFIS and GWIS as associated users. Since 2022, DSES Ukraine has received EFFIS products in KML and Sentinel Hub WMS formats. This creates the basis for a national AI fusion system that ingests data from multiple satellite streams and produces a unified active-fire map.

Limitations and open problems in 2026

Despite two decades of rapid progress, several key limitations remain unresolved.

Subcanopy fires: in dense boreal and tropical forests, peat or litter pyrolysis can continue for weeks without a hot surface visible from space. Indonesia is a classic case: 2015 peat fires smouldered below the surface and were detected mostly by CO concentration in CAMS models rather than by thermal anomaly.

Cloud cover: even the best sensors cannot see through optically thick clouds. In the Carpathians and Polissia this means that during fog or rain, satellite detection temporarily fails.

Small fires under low contrast: sun-heated soil in deserts or farmlands reduces thermal contrast. VIIRS may miss a fire smaller than 0.5 ha at noon over dry steppe.

Data polarisation and distribution: some national providers (China FY, Russia Meteor, India IRS) do not publish their fire products in open standards. This creates gaps in global coverage, especially in regions with low Western satellite activity.

Wilson et al. (2017) in their IGARSS paper discuss how future satellite constellations (such as EUMETSAT MTG-S, NASA Landsat Next, ESA TRUTHS) may close some of these gaps by 2030 (Wilson et al., 2017, IGARSS).

Where Ukraine fits in this architecture

Ukraine has no domestic satellite infrastructure for fire monitoring. The space layer relies on foreign satellites: VIIRS (NOAA-20, NOAA-21, Suomi NPP), MODIS (Aqua, Terra), Sentinel-3 (Copernicus). A nationwide camera network does not exist. Before 2022, individual forestry operators had local CCTV cameras, but many were damaged by the war or redirected to other priorities.

In this context, WildFiresUA — the national fire service built by the YourAirTest team in partnership with Oles Honchar Dnipro National University (DNU), EcoCity, and Arnika — forms the first complete detection loop in Eastern Europe based on satellites and atmospheric modelling. The architecture: VIIRS + Sentinel-3 SLSTR + FIRMS LANCE as input streams, FLEXPART for smoke transport forecasting, AI validation for reducing commission errors (such as excluding industrial hot spots). The active fire map is publicly available at partner.yourairtest.com/map, and the smoke transport forecast at partner.yourairtest.com/forecast.

Why a satellite-based foundation rather than camera or drone: Ukraine has 60+ million hectares in relevant forest and steppe zones, which makes camera coverage economically untenable without international funding. Drones are constrained by combat operations and the regulatory framework. Satellites are the only layer that delivers national 24/7 coverage. Future deployment of camera nodes in the Carpathians and the Chornobyl exclusion zone is considered a second layer after the basic satellite architecture is reinforced.

Summary

The 2026 global wildfire detection ecosystem is a multi-layered architecture with several hundred technical and organisational components. No single component alone is sufficient. Polar satellites provide global coverage with a frequency of 4-6 times per day and 375-1000 m resolution. Geostationaries deliver minute-level updates with a 2 km pixel. Camera networks provide sub-minute response to visibility but are limited to small zones. AI fusion combines everything into a single operational picture.

For civil-protection agencies, the critical task is not to choose one technology but to build an integration loop: data from different layers flow into a single decision-support system where a human dispatcher sees a fire within 5-15 minutes rather than within an hour. Lahaina 2023 showed that even with available technologies, the price of delay is measured in lives. Ukraine, with foreign satellite infrastructure and a domestic scientific base, is building a national loop in 2026 that meets contemporary European standards.

Ukrainian startup ecosystem: follow TechUkraine and AIN.ua — the two leading outlets covering Ukrainian deep tech, climate tech, and environmental startups.

Related reading on yourairtest.com

• WildFires page of the WildFiresUA project
• YourAirTest air quality monitoring system overview
• YourAirTest blog with reviews of air science

MASK0

Related reading — other scientific reviews

• AI fire detection in satellite data and video streams 2024-2026: deep learning state of the art
• Wildfire smoke emission factors: Andreae 2019, Akagi 2011, and the global GFED database
• Atmospheric dispersion of wildfire smoke plumes: HYSPLIT, FLEXPART, SILAM, NAME, CALPUFF — Lagrangian vs Eulerian
• Ground-based camera networks for fire detection: global experience and constraints in 2026

What to do today

1. Check the YourAirTest air quality map for your city — recent PM2.5 readings.
2. If the topic resonates, share with colleagues; we monitor referrals via Google Search Console.
3. To contribute data (sensor measurements, regional models) reach out via the contact form.

References

1. Justice C.O., Giglio L., Korontzi S. et al. (2002). The MODIS fire products. Remote Sensing of Environment 83(1-2): 244-262.
2. Wooster M.J., Roberts G., Freeborn P.H. et al. (2012). LSA SAF Meteosat FRP products. Remote Sensing of Environment 122: 236-251.
3. Schroeder W., Oliva P., Giglio L., Csiszar I.A. (2014). The New VIIRS 375m active fire detection data product. Remote Sensing of Environment 143: 85-96.
4. Giglio L., Schroeder W., Justice C.O. (2016). The collection 6 MODIS active fire detection algorithm. Remote Sensing of Environment 178: 31-41.
5. Wooster M.J., Xu W., Nightingale T. (2021). Sentinel-3 SLSTR active fire and FRP product validation. Remote Sensing 13(9): 1627.
6. Schmidt C.C., Hoffman J., Prins E.M. et al. (2017). GOES-R ABI fire detection algorithm. Atmospheric Environment 150: 98-113.
7. Csiszar I., Schroeder W., Giglio L. et al. (2014). Active fires from VIIRS. Journal of Geophysical Research: Atmospheres 119(2): 803-816.
8. Roberts G., Wooster M.J. (2008). Fire detection and FRP retrieval from MSG SEVIRI. IEEE Transactions on Geoscience and Remote Sensing 46(10): 2950-2962.
9. Wilson R.C., Freeborn P.H., Wooster M.J. (2017). Future fire monitoring satellite missions. IGARSS Proceedings.
10. Govil K., Welch M.L., Ball J.T., Pennypacker C.R. (2020). Preliminary results from a wildfire detection system using deep learning on remote camera images. Remote Sensing 12(1): 166.
11. Hu Y., Lu X. (2018). Real-time video fire smoke detection by utilizing spatial-temporal ConvNet features. Sensors 18(3): 712.
12. Pinto M.M., Libonati R., Trigo R.M. et al. (2020). A deep learning approach for mapping burned areas. Remote Sensing 12(17): 2739.
13. Allison R.S., Johnston J.M., Craig G., Jennings S. (2016). Airborne optical and thermal remote sensing for wildfire detection and monitoring. Sensors 16(8): 1310.
14. Restas A. (2015). Drone applications for supporting disaster management. Procedia Engineering 100: 316-322.
15. NASA FIRMS — Fire Information for Resource Management System.
16. AlertWildfire camera network (UNR / UCSD / UO).
17. Copernicus EFFIS — European Forest Fire Information System.
18. JRC GWIS — Global Wildfire Information System.
19. EUMETSAT MTG mission documentation.
20. USFS Aviation programme overview.
21. USFS UAS Program.
22. NOAA JPSS / VIIRS mission documentation.
23. ESA Sentinel-3 mission page.
24. NOAA GOES-R series programme.
25. BoM Australia satellite services.
26. INPE TerraBrasilis — Brazilian fire data.