Aim: Wildfire activity in recent years is notable not only for an expansion of total area burned but also for large, single‐day fire spread events that pose challenges to ecological systems and human communities. Our objectives were to gain new insight into the relationships between extreme single‐day fire spread events, annual area burned, and fire season climate and to predict changes under future warming.Location: Fire‐prone regions of the western USA.Time period: 2002–2020; a future +2°C scenario.Methods: We used a satellite‐derived dataset of daily fire spread events and gridded climate data to assess relationships between extreme single‐day fire spread events, annual area burned, and fire season maximum temperature, climate moisture deficit, and vapour pressure deficit. We then developed models to predict fire activity under a 2°C warming scenario.Results: Extreme single‐day fire spread events >1,100 ha (the top 16%, >1 SD) accounted for 70% of the cumulative area burned over the period of analysis. The variation in annual area burned was closely tied to the number and mean size of spread events and distributional skewness towards more large events. For example, we identified 441 extreme events in 2020 that together burned 2.2 million ha across our study area, in contrast to an average of 168 per year that burned 0.5 million ha annually between 2002 and 2019. Fire season climate variables were correlated with the annual number of extreme events and area burned. Our models predicted that the annual number of extreme fire spread events more than double under a 2°C warming scenario, with an attendant doubling in the area burned.Conclusions: Exceptional fire seasons like 2020 will become more likely, and wildfire activity under future extremes is predicted to exceed anything yet witnessed. Safeguarding human communities and supporting resilient ecosystems will require new lines of scientific inquiry, new land management approaches and accelerated climate mitigation efforts.

The distribution of fuels is recognised as a key driver of wildland fire behaviour. However, our understanding of how fuel density heterogeneity affects fire behaviour is limited because of the challenges associated with experiments that isolate fuel heterogeneity from other factors. Advances in fire behaviour modelling and computational resources provide a means to explore fire behaviour responses to fuel heterogeneity. Using an ensemble approach to simulate fire behaviour in a coupled fire–atmosphere model, we systematically tested how fuel density fidelity and heterogeneity shape effective wind characteristics that ultimately affect fire behaviour. Results showed that with increased fuel density fidelity and heterogeneity, fire spread and area burned decreased owing to a combination of fuel discontinuities and increased fine-scale turbulent wind structures that blocked forward fire spread. However, at large characteristic length scales of spatial fuel density, the fire spread and area burned increased because local fuel discontinuity decreased, and wind entrainment into the forest canopy maintained near-surface wind speeds that drove forward fire spread. These results demonstrate the importance of incorporating high-resolution fuel fidelity and heterogeneity information to capture effective wind conditions that improve fire behaviour forecasts.

Biomass burning by wildland fires has significant ecological, social and economic impacts. Satellite remote sensing provides direct measurements of radiative energy released by the fire (i.e. fire intensity) and surrogate measures of ecological change due to the fire (i.e. fire or burn severity). Despite anecdotal observations causally linking fire intensity with severity, the nature of any relationship has not been examined over extended spatial scales. We compare fire intensities defined by Moderate Resolution Imaging Spectroradiometer Fire Radiative Power (MODIS FRP) products with Landsat-derived spectral burn severity indices for 16 fires across a vegetation structure continuum in the western United States. Per-pixel comparison of MODIS FRP data within individual fires with burn severity indices is not reliable because of known satellite temporal and spatial FRP undersampling. Across the fires, 69% of the variation in relative differenced normalized burn ratio was explained by the 90th percentile of MODIS FRP. Therefore, distributional MODIS FRP measures (median and 90th-percentile FRP) derived from multiple MODIS overpasses of the actively burning fire event may be used to predict potential long-term negative ecological effects for individual fires.