Palaeodata-informed modelling of large carbon losses from recent burning of boreal forests

Wildfires play a key role in the boreal forest carbon cycle(1,2), and models suggest that accelerated burning will increase boreal C emissions in the coming century (3). However, these predictions may be compromised because brief observational records provide limited constraints to model initial conditions (4). We confronted this limitation by using palaeoenvironmental data to drive simulations of long-term C dynamics in the Alaskan bo- real forest. Results show that fire was the dominant control on C cycling over the past millennium, with changes in fire frequency accounting for 84% of C stock variability. A recent rise in fire frequency inferred from the palaeorecord5 led to simulated C losses of 1.4 kg C m?2(12% of ecosystem C stocks) from 1950 to 2006. In stark contrast, a small net C sink of 0.3 kg C m?2 occurred if the past fire regime was assumed to be similar to the modern regime, as is common in models of C dynamics. Although boreal fire regimes are heterogeneous, recent trends6 and future projections (7) point to increasing fire activity in response to climate warming throughout the biome. Thus, predictions (8) that terrestrial C sinks of northern high latitudes will mitigate rising atmospheric CO2 may be over-optimistic.