**Multi-Model Forecasts of Very-Large Fire Occurences during the End of the 21st Century**

**Harry R. Podschwit 1,***∗***, Narasimhan K. Larkin 2, E. Ashley Steel <sup>3</sup> and Alison Cullen <sup>4</sup> and Ernesto Alvarado <sup>5</sup>**


Received: 9 November 2018; Accepted: 13 December 2018; Published: 19 December 2018

**Abstract:** Climate change is anticipated to influence future wildfire activity in complicated, and potentially unexpected ways. Specifically, the probability distribution of wildfire size may change so that incidents that were historically rare become more frequent. Given that fires in the upper tails of the size distribution are associated with serious economic, public health, and environmental impacts, it is important for decision-makers to plan for these anticipated changes. However, at least two kinds of structural uncertainties hinder reliable estimation of these quantities—those associated with the future climate and those associated with the impacts. In this paper, we incorporate these structural uncertainties into projections of very-large fire (VLF)—those in the upper 95th percentile of the regional size distribution—frequencies in the Continental United States during the last half of the 21st century by using Bayesian model averaging. Under both moderate and high carbon emission scenarios, large increases in VLF frequency are predicted, with larger increases typically observed under the highest carbon emission scenarios. We also report other changes to future wildfire characteristics such as large fire frequency, seasonality, and the conditional likelihood of very-large fire events.

**Keywords:** mega-fires; Bayesian-model averaging; model uncertainty; climate-fire models
