Wildfires play a central role in the Earth's carbon cycle and are a major driver of regional air quality. As the climate warms, regional megafire complexes are becoming increasingly common. Canada has recently emerged as a focal region of this shift. In 2023, record-breaking fires in Canada burned an estimated 12. 7–18. 6 million hectares (Mha; Pelletier et al. 2024; Canadian Forest Service 2026), releasing substantial amounts of carbon into the atmosphere (Byrne et al. 2024) and causing widespread impacts on air quality (Field et al. 2025; Zhang et al. 2025) and vegetation loss (MacCarthy et al. 2024). Initially regarded as an anomalous year, the 2023 fire season was followed by two additional years of extreme activity, forming an unprecedented three-year period of elevated fire activity from 2023 to 2025 (Tymstra and Flannigan 2025). Satellite observations indicate that annual active fire counts during this period (1. 71, 0. 70, 0. 90 million in 2023–2025, respectively) consistently exceeded the 2002–2022 mean (0. 23 Mha yr. −1) by approximately six standard deviations, averaging 1. 10 million detections per year (Figure 1a). The Canadian National Fire Database (CNFDB) and Global Fire Emissions Database version 5 (GFED5) (van der Werf et al. 2025; Chen et al. 2026) report mean non-crop burned areas of 10. 6 and 12. 5 Mha yr. −1 in Canada during 2023–2025, respectively—more than seven standard deviations above the 21st century mean (2. 4 ± 1. 2 Mha yr. −1) and 5 standard deviations above the longer historical baseline (2. 1 ± 1. 7 Mha yr. −1 for 1959–2022; Figure 1b). Beyond the extent of burned area, this three-year fire anomaly was characterized by exceptionally high gaseous and particulate emissions. Using VIIRS active fire detections, we extended the GFED5 emissions record through recent years (Chen et al. 2026). Total carbon emissions in Canada during 2023–2025 were 574, 243, and 308 Tg C yr. −1, respectively, marking the highest values in the past two decades. Together, these three years constitute consecutive extremes (Figure 1c), and account for about 10% of global fire emissions. These values correspond to Z-scores of ~5–15 relative to the historical mean (67 ± 33 Tg C yr. −1), placing them far outside the range of expected variability. These sustained emissions led to widespread degradation of air quality across Canada (Figure 1d). Summer mean aerosol optical depth (AOD) was strongly influenced by wildfire activity during this period. The frequency of smoke–haze reports at weather stations declined steadily between 1970 and 2000, reflecting improvements in air quality following regulatory measures such as the US Clean Air Act and comparable environmental legislation in Canada. However, this trend has reversed over the past 15 years due to escalating wildfire-driven smoke (Field et al. 2025). During 2023–2025, smoke–haze frequency in Canada exceeded 3% (corresponding to ~109 smoke-hours per year), compared with an average of 0. 3% (~10 smoke-hours per year) during the 2000s, when Canadian air quality was at its best. As noted by Tymstra and Flannigan (2025), wildfires in North America have entered uncharted territory. Here we show that the 2023–2025 period represents not only the most extreme multi-year fire episode in Canada this century, but also the most severe since records began in the late 1950s. Past clusters of high-fire years, including 1979–1981, 1994–1995, and 2013–2015 (Amiro et al. 2001; Hanes et al. 2025), were significant but fall well short of the recent period in terms of burned area, carbon emissions, and smoke frequency. The persistence of these extreme fire years has also generated escalating public health and economic impacts. Fires spreading into the populated areas forced tens of thousands of Canadians to evacuate, with disproportionate impacts on First Nations communities. Multiple interacting factors influence extreme wildfire activity in Canada, including climate variability, ecosystem conditions, and human influences (Jain et al. 2024). The sequence of extreme fire years from 2023 to 2025 therefore raises an important question: whether this period represents a rare natural anomaly or the emergence of a new baseline under ongoing human-caused climate change (Curasi et al. 2024). If consecutive extreme fire seasons become more common, pollutant emissions and carbon losses from Canada's forests and other ecosystems could rival or exceed those from human activities, potentially undermining air-quality gains and climate-mitigation progress achieved through reductions in transportation, energy, and industrial emissions. Y. C. , R. D. F. , R. C. S. , and J. T. R. conceived of the analysis and methodology. Y. C. and R. D. F. conducted the data curation, formal analysis and wrote the original draft. Y. C. conducted visualization. All authors contributed to reviewing and editing the manuscript. This work was supported by the NASA Earth Information System—Fire project (Grant No. 80NSSC20K0590), the NASA FireSense Technology Program (Grant No. 80NSSC24K1823), and the NASA FireSense Implementation Program (Grant No. 80NSSC24K1317). Additional support was provided by the U. S. National Science Foundation through the Collaborations in Artificial Intelligence and Geosciences (CAIG) program (Grant No. RISE-2425932) and by the Dutch Research Council (NWO) through a Rubicon Grant (Grant No. 019. 241EN. 018). This work was supported by the NASA (Grants 80NSSC20K0590, 80NSSC24K1823, 80NSSC24K1317), NSF (Grant RISE-2425932) and the Dutch Research Council (NWO) (Grant 019. 241EN. 018). The authors declare no conflicts of interest. All datasets used in this study are publicly available from the repositories and sources listed below. MODIS (MCD14ML, Collection 6. 1, https: //doi. org/10. 5067/FIRMS/MODIS/MCD14DL. NRT. 0061) and VIIRS (VNP14IMGML, Collection 2, https: //doi. org/10. 5067/VIIRS/VNP14. 002) active fire location data were obtained from the University of Maryland FTP server (fuoco. geog. umd. edu). For MCD14ML, only detections from the Terra satellite were used to calculate active fire counts, and the resulting counts were scaled to match the VIIRS data during the overlap period (2013–2022). The Global Fire Emissions Database version 5 (GFED5, https: //doi. org/10. 5281/zenodo. 16794692) and its near-real-time extension (GFED5NRT, https: //doi. org/10. 5281/zenodo. 18702700) were obtained from https: //globalfiredata. org. GFED5 standard data were used for the period 2002–2022, and GFED5NRT data were used for 2023–2025 for burned-area and fire-emissions analyzes. Burned-area data from the Canadian National Fire Database (CNFDB, https: //doi. org/10. 18739/A24F1MM0S) were accessed through the Canadian Wildland Fire Information System (CWFIS) portal (https: //cwfis. cfs. nrcan. gc. ca/ha/nfdb). Burned-area estimates for 2025 rely on provisional provincial agency reports and near-real-time polygon data and may be subject to retrospective adjustments. CNFDB emissions estimates are described in Amiro et al. (2001). CNFDB burned-area and emissions data were scaled to match GFED5 over their respective overlap periods (2002–2022 for burned area and 1997–1999 for emissions). Top-down inversion estimates for 2023 emissions were derived from Byrne et al. (2024) (https: //doi. org/10. 48577/jpl. V5GR9F). Aerosol optical depth (AOD) data were obtained from the MODIS/Terra Aerosol Cloud Water Vapor Ozone Daily L3 Global 1° CMG product (MOD08D3, https: //doi. org/10. 5067/MODIS/MOD08D3. 061) and the VIIRS/SNPP Deep Blue Level-3 daily aerosol product on a 1° × 1° grid (AERDBD3VIIRSSNPP, https: //doi. org/10. 5067/VIIRS/AERDBD3VIIRSSNPP. 002), both accessed through NASA's Level-1 and Atmosphere Archive https: //ladsweb. modaps. eosdis. nasa. gov/). Observational smoke and haze frequency records from weather stations were derived from historical climate observations provided by Environment and Climate Change Canada (ECCC) (https: //dd. weather. gc. ca).
Chen et al. (Wed,) studied this question.