In March 2025 the Joint Research Centre of the European Commission published the EFFIS Advance Report 2024. Among its record-breaking figures, one number stood out: 965,000 hectares burned in Ukraine during 2024 — more than in the entire European Union combined. Yet scale is only half the story. The nature of these fires — military ignitions, strikes on energy and chemical infrastructure, radiological scenarios, post-Kakhovka steppe events — makes the Ukraine wildfire corpus 2022-2025 a scientific resource that cannot be obtained anywhere else. This article sets out what the corpus contains, why it matters for peer review and grant applications, and how the WildFiresUA system — our complete stack for Ukraine — uses it as the foundation for validation.
Satellite observations, ground measurements, validated models — the Ukraine wartime corpus as a peer-reviewed resource. Image: Copernicus, Wikimedia Commons (CC BY-SA 3.0 IGO).
The figures that make the corpus unique
2022. Zibtsev et al. 2024 (Ukrainian Journal of Forest and Wood Science 15(1)) reported 749,500 hectares burned during 2022. The breakdown: cropland 419,100 ha, other natural vegetation 273,800 ha, coniferous forest 31,100 ha, other forest 25,500 ha. Greenhouse-gas emissions: 5.20 Mt CO₂ plus 0.28 Mt of other greenhouse gases. Spatial pattern: 68.9% of burned area within 60 km of the front line; 42.5% in occupied territory.
2023. A similar order of magnitude. Analysis of fire clusters using Sentinel-2 and MODIS data (UWEC “Flames of War”) indicated that around 80% of burned area fell within 30 km of the front, up from 66% in 2022. The concentration of fires along the military contact line intensified.
2024. JRC EFFIS: 965,000 hectares, the worst season on record. 8,753 individual fires; 317 exceeded 500 ha, 110 exceeded 1,000 ha, and the two largest exceeded 8,000 ha each. Distribution by land cover: other natural vegetation 39%, agriculture 33%, mixed forest 19%. The summer of 2024 was concentrated on the eastern and southern fronts, coincident with a pronounced Fire Weather Index anomaly.
Cumulative 2022-2024: at least 755,638 ha of forest destroyed (State Forest Agency of Ukraine) and more than 20,000 fires linked to the war. Donetsk region lost 180 km² and Luhansk 195 km² during 2022-2023 (analysis from the University of Bologna and Global Forest Watch; see also the independent assessment of Luhansk and Donetsk fires in Trees, Forests and People (2024)).
For context, the pre-war baseline for 2010-2021 from EFFIS was approximately 30,000 to 50,000 ha per year in a typical year, with episodic peaks in 2015 and 2020 (the major Chornobyl zone fires). The transition to 750,000-965,000 ha represents a 20- to 30-fold step change.
Why this is more than large numbers
For the scientific community, the figure of 965,000 ha on its own is a statistical artefact. What makes the corpus scientifically valuable is the diversity of event types it represents — types that are simply unavailable in peacetime scenarios.
Type 1 — strikes on oil infrastructure. The Kremenchuk refinery sustained at least 8 separate attacks between February 2022 and September 2023, with more than 15 distinct fires on the complex (CEOBS documentation). The characteristic FRP profile: a sharp rise to 100+ MW, steady sustainment over hours (hydrocarbon pool fires burn hotter and longer than vegetation), and a persistent 12 µm blackbody signature even through heavy soot. Lysychansk: fires during 2022. Drohobych: strikes in 2025. These events are a stress test for FRP detection algorithms — they saturate the VIIRS M13 and MODIS channels — and at the same time ideal cases for FLEXPART dispersion validation, because the source is well localised, intense and long lasting.
Type 2 — Chornobyl Exclusion Zone fires. April 2020: approximately 115,000 ha burned, the largest event since 1986. Evangeliou et al. 2016 (Scientific Reports 6:26062) and Masson et al. 2022 (Atmospheric Environment 291:119402) estimated releases of around 341 GBq of ¹³⁷Cs, 51 GBq of ⁹⁰Sr, and around 2 GBq of ²³⁸Pu during 1-22 April 2020. Nine orders of magnitude below the 1986 accident, yet still a documented resuspension of radionuclides. Kyiv recorded +5-15% above background, with PM₂.₅ at 60-150 µg/m³ sustained over several days. March-April 2022: during the Russian occupation of the zone, at least 31 fires burned about 10,000 ha, with documented disturbance of the Red Forest by heavy armour and resuspension of contaminated dust. These events are a unique natural experiment for validating CALPUFF with radionuclide source terms.
Type 3 — Kakhovka aftermath (June 2023). The breach of the dam released approximately 18 km³ of water over four days, exposed about 1,870 km² of former reservoir bed, and removed irrigation for 584,000 ha of southern agricultural land. In the summer of 2024, Kherson oblast experienced around 200,000 ha of steppe fires on the drained bed, with PM₂.₅ peaks of 80-300 µg/m³ in Kherson, 40-150 µg/m³ in Mykolaiv and 30-100 µg/m³ in Odesa. Satellite observations tracked plumes as far as Chișinău. (A running chronicle is maintained in Wildfire Today’s Ukraine archive.) This represents a new category of landscape-conversion fires that EU fire models have not previously encountered.
Type 4 — destroyed urban infrastructure. Mariupol (Azovstal), Kharkiv (Saltivka), Chernihiv, Borodianka, Irpin. Ruined residential buildings and chemical facilities (Sumykhimprom — partial damage; Sievierodonetsk Azot — total destruction). The fuel load is a mixture of wood (~18 MJ/kg), plastics (~35 MJ/kg), asbestos, lead from paints and PVC. No SB40 or FBP fuel model covers this combination. We were therefore obliged to adapt the pipeline using urban-fire emission factors from the SFPE Handbook rather than wildland values.
Type 5 — stubble burning. A traditional agricultural practice, formally banned since 2015 but still widespread. From the 2022 Zibtsev sample, 419,100 ha of cropland burned, a significant share of it stubble. No standard SB40 or FBP class exists for stubble (GR1 is the closest analogue, with a suboptimal combustion model). We had to calibrate emission factors separately, using field measurements from EcoCity and data from Ukrainian agricultural universities.
Type 6 — Polissia peat fires. Peat ignitions from surface fires smoulder underground for weeks or months at a vertical descent rate of 1-2 mm per day. One hectare with a one-metre peat layer contains around 500 t of carbon (approximately 1,800 t CO₂-equivalent). Polissia has around 500,000 ha of drained peatland from Soviet-era reclamation. Moskalchuk et al. 2025 (Sustainability 17:2223) document a warming trend of +0.60°C per decade over 1990-2021, together with a 3-5% decrease in precipitation. Peat fires constitute a distinct combustion regime, not represented in SB40, and require dedicated emission factors and duration models.
Validation track record to date
The most significant peer-reviewed validation to emerge from the Ukrainian corpus is our CJFR 2026 paper in the Canadian Journal of Forest Research. The YourAirTest team, together with Oles Honchar Dnipro National University, UHMI (the Ukrainian Hydrometeorological Institute) and the Marzieiev Institute of Hygiene and Medical Ecology, used VIIRS hotspots to localise the fire of 23 March 2022 in the Kyiv area. The full pipeline followed: FCCS fuels → CONSUME fire thermodynamics → FEPS emission factors → CALPUFF dispersion. The results, validated against the Kyiv City State Administration ground stations:
- BIAS = +2.77 µg/m³ PM₂.₅ (a modest overestimate)
- RMSE = 48 µg/m³
- Pearson r = 0.40 (moderate correlation during a peak event)
All values lie within a factor of two of the observed concentrations — an acceptable level of accuracy for operational air-quality modelling. This is the first peer-reviewed wildfire dispersion pipeline validated against Ukrainian ground stations.
A companion body of CNN work published by our team comprises three papers: ISPRS Geospatial Week 2023, DGPF 2023 in Munich, and Springer LNDECT 2022. The theme is CNN-based burned-area detection from Sentinel-2, achieving 97% overall accuracy for the classification of fire, burned area, smoke and background. Our 30-metre Anderson 13 fuel map for Ukraine (CJFR 2026) has the following oblast-level composition:
- Kyiv oblast: C-1 Spruce-Lichen 0.1%, C-3 Mature Pine 4.4%, C-5 Red/White Pine 6.2%, D-1 Leafless Aspen 14.8%, M-1/M-2 Mixedwood 4.2% (seasonal), O-1b Standing Grass 61%
- Dnipro oblast: O-1b dominant at 79% (steppe)
- Seasonal curing transitions: winter 90-95%, spring 55-85%, summer 45-85%, autumn 75-90%
Why this corpus cannot be reproduced elsewhere
Among world regions with large burned areas of scientific interest for wildfire research:
- Canada 2023: 18.5 Mha — larger in scale, but homogeneous (boreal conifers), predominantly lightning-ignited, and mostly remote, without urban or industrial damage. Byrne 2024 Nature 633:835 measured 647 TgC of emissions — important for climate science, but not for integrated stack tests.
- California: 4.2 Mac in 2020 and less in 2024, but with urban-interface concerns and PG&E-driven investment (a leading fire-spread SaaS vendor and a dedicated tower-camera AI operator). The fire aetiology, however, is lightning, electrical ignition and PSPS scenarios, not military.
- Mediterranean 2021 and 2025: Türkiye 206,000 ha, Greece 125,000 ha, Iberia 460,585 ha in 2025. Arson- and heat-dome-driven. Again, civilian ignition patterns rather than military, infrastructural or radiological ones.
- Ukraine 2022-2025: a combination found nowhere else — military ignitions (artillery, missiles, drones), strikes on energy and chemical infrastructure, radiological scenarios (the Chornobyl Exclusion Zone and contingency scenarios at Zaporizhzhia NPP), landscape-conversion events (Kakhovka), urban destruction fires (Mariupol, Kharkiv), traditional stubble burning and peat fires in Polissia.
No other region in the world presents this test matrix simultaneously. Peer reviewers recognise this. Grant applications that place the Ukrainian corpus at the foundation of their validation dataset carry a defensibility that American and Western European teams cannot reproduce.
How the corpus is structured within WildFiresUA
The operational architecture:
(1) Raw data layers: VIIRS NOAA-20/21/SNPP NRT via the NASA FIRMS API; MODIS NRT; Sentinel-2 MSI for burn-scar mapping (dNBR); Sentinel-1 SAR for cloud penetration; Copernicus Sentinel-5P TROPOMI for CO, NO₂, SO₂ and formaldehyde; ERA5 reanalysis together with a local WRF configuration for meteorology; and EFFIS pan-European burned-area polygons for cross-validation.
(2) Clustering: DBSCAN (eps = 1.5 km, haversine on 6,371 km sphere, min_samples = 2) collapses multi-pixel hotspots into fire events. Union-Find handles temporal chaining.
(3) Event annotation: significant events are marked manually — oil depots, chemical plants, the Chornobyl zone, the Kakhovka aftermath, urban destruction, agricultural burning. Metadata records the type, probable ignition source and confidence level.
(4) Ground-truth validation: reports from the State Emergency Service (SES), Kyiv City State Administration monitoring (central Kyiv), the EcoCity mesh (80+ cities), Arnika studies and dedicated campaigns with the Marzieiev Institute.
(5) Publication pipeline: each significant event becomes a case study with reproducible code and data. Target journals: Atmospheric Chemistry and Physics, Earth System Science Data, Environmental Research Letters and Fire (MDPI). One peer-reviewed publication every twelve months.
Implications for grant strategy
For Horizon Europe Cluster 5 (Climate services and adaptation), the 2026-2028 call texts explicitly mention cascading multi-hazards, wildfire in the context of climate adaptation and the Eastern European frontier. The Ukrainian corpus addresses several priorities at once: climate (Canada 2023 demonstrated that large years are now 3-6 times more probable, per Kelley 2025), military externalities, public-health attribution and cross-border smoke.
For the EIC Pathfinder, a physics-ML hybrid validated on this corpus has a foundational defensibility. Reviewers ask: “Where have you validated?” The answer — “on 20,000 Ukrainian fires with a documented variety of ignition types” — is difficult to match.
For the EU Mission on Adaptation and the Union Civil Protection Mechanism, the system is an operational tool validated and used by the SES, with transfer potential to other agencies (Portugal ICNF, the Greek Civil Protection, Romanian IGSU). The corpus is evidence that the system performs under extreme conditions.
For EU4Health and DG SANTE, smoke-exposure attribution rests on our CALPUFF and FLEXPART outputs together with KMDA validations. Sofiev 2025 in The Lancet Planetary Health opened this line of inquiry; our Ukrainian operational case is its realisation.
Limitations and honest caveats
Several points must be acknowledged for the sake of peer-review honesty:
1. Incomplete ground truth. For many fires, particularly in occupied or front-line territory, ground-based measurements are simply unavailable. The monitoring network in Kyiv, Lviv and Kharkiv is adequate; in Donetsk, Luhansk and occupied districts it is fragmentary. For validation in such cases we rely on remote proxies (Sentinel-2 post-burn, Sentinel-5P CO), but physico-chemical confirmation is limited.
2. Ignition-source uncertainty. FIRMS does not distinguish military ignition from agricultural burning or forest fire. Classification is a manual OSINT effort (cross-checked with Bellingcat-style methodology, Sentinel-2 context, CORINE land cover and infrastructural polygons). It is not automated and does not scale to real time.
3. Rapid landscape change. Ukrainian landscapes are changing quickly: demining, abandoned fields, salinisation after Kakhovka, reconstruction in built-up areas. LANDFIRE-style static fuel-map datasets cannot keep pace. Our 30 m Anderson 13 map is a 2024-2025 snapshot; updates for 2026-2027 are required.
4. Military data withholding. A share of the strikes is documented in operational channels with restricted public access. For academic publication we work only with OSINT-available events. The internal operational database used by the SES is more complete.
Frequently asked questions
Is this corpus a competitive advantage that will be lost after the war?
No. The historical dataset remains. The methodology is published in peer-reviewed journals. Continuous data generation will continue through post-war reconstruction scenarios (residual fires, peat fires, demining operations). Ukraine 2022-2030 is set to be for fire science what California 2017-2025 has been.
Is the dataset available to other researchers?
The primary components (VIIRS NRT, Sentinel-2, EFFIS polygons) are publicly available. Our annotated validation layer and clustering pipeline are partly on GitHub and partly in data papers under preparation (Earth System Science Data is the target).
Can Western companies purchase access to the validated corpus?
To a limited extent. We publish through peer review and open the data progressively. Commercial licensing remains an open question, depending on the business model and grant obligations.
Who owns the corpus?
Academic IP is shared among YourAirTest, Oles Honchar Dnipro National University, the Marzieiev Institute, ULCO/LPCA (France) and UHMI. Operational data belong to the SES and YourAirTest. Open content is released under CC BY 4.0.
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
Zibtsev S, et al. (2024) Assessment of war-induced fires in Ukraine 2022. Ukrainian Journal of Forest and Wood Science 15(1).
European Commission Joint Research Centre. (2024) EFFIS Advance Report 2024 — Ukraine record-breaking fire season. JRC News, 25 March 2025.
Our peer-review paper CJFR 2026: 30-metre Anderson 13 fuel map for Ukraine, Canadian Journal of Forest Research (cjfr-2025-0035). DOI: 10.1139/cjfr-2025-0035
Evangeliou N, et al. (2016) Wildfires in the Chernobyl exclusion zone and their radiological impact. Scientific Reports 6:26062.
Masson O, et al. (2022) Radiological impact of the 2020 Chornobyl exclusion zone fires. Atmospheric Environment 291:119402.
Byrne B, et al. (2024) Carbon emissions from the 2023 Canadian wildfires. Nature 633:835.
Kelley D, et al. (2025) State of Wildfires 2023-24: attribution of extreme seasons. Nature Communications 16.
Jones MW, et al. (2024) Global fire CO₂ emissions from forest fires driven by climate change. Science 386:eadl5889.
Moskalchuk N, et al. (2025) Climate trends in Polissia 1990-2021. Sustainability 17:2223.
Conflict and Environment Observatory (CEOBS). (2022-2024) Dossiers on the Kremenchuk and Lysychansk refineries and on Kakhovka.
Post-invasion wildfire impacts in Luhansk and Donetsk regions. Trees, Forests and People (2024) 18:100639. DOI: 10.1016/j.tfp.2024.100639
Wildfire Today — Ukraine news archive. wildfiretoday.com/tag/ukraine
European Innovation Council (EIC) — Horizon Europe framework. eic.ec.europa.eu