What the Blazes? Using evidence-based uncertainty estimates to find what we really know about the world's changing fire regimes.

Performance and unaccounted uncertainties of fire models make it challenging to determine causes of shifts in past and future fire regimes and attribute fire event drivers. We develop a new, probabilistic style of fire modelling to optimise model performance against satellite observations of wildfires, vegetation cover and meteorological conditions. We also preserve information on the remaining error and parameter uncertainty. This allows us to calculate an evidence-based likelihood of fire regime change, given our scientific understanding and observations of fire drivers. We apply this to answer three vital and little understood questions about fire in our changing world: 1) What are the drivers of historical global fire regime change; 2) and how likely are global extreme fire events in the future; 3) what caused recent extreme fire events, using 2019 Amazonia fires as an example.

Boreal and tropical forests show the most perturbed recent fire regime and, together with peatland areas, are most likely to shift under future climate change. Historically, changes in fuel loads and moisture content drove most of the 2.3-2.6% shift in forest fire regimes since 2002. Under future climate change, burning in Boreal and Tropical Forests is likely to increase by 2100 irrespective of future emissions. Globally, there will likely be a 31-57% increase in extreme fire events (defined as a one-in-one hundred event in 2010-2020), with some boreal forest and permafrost regions seeing up to 10x more extreme events.

Perhaps most concerning is that our future analysis does not yet include landscape management and policy changes. However, we attribute that the 2019 Amazonia forest fires were very likely driven by human management, with only a 7% probability of being influenced by non-human drivers. This motivates preliminary work at incorporating management decisions into our future projections of fire regime uncertainty. Given forest's role in the global carbon cycle and potential for Earth System feedbacks, we also show preliminary work coupling this uncertainty estimation framework and fire emissions to a dynamic land surface model.