Where and why do conifer forests persist in refugia through multiple fire events?

Changing wildfire regimes are causing rapid shifts in forests worldwide. In particular, forested landscapes that burn repeatedly in relatively quick succession may be at risk of conversion when pre-fire vegetation cannot recover between fires. Fire refugia (areas that burn less frequently or severely than the surrounding landscape) support post-fire ecosystem recovery and the persistence of vulnerable species in fire-prone landscapes. Observed and projected fire-induced forest losses highlight the need to understand where and why forests persist in refugia through multiple fires. This research need is particularly acute in the Klamath-Siskiyou ecoregion of southwest Oregon and northwest California, USA, where expected increases in fire activity and climate warming may result in the loss of up to one-third of the region's conifer forests, which are the most diverse in western North America. Here, we leverage recent advances in fire progression mapping and weather interpolation, in conjunction with a novel application of satellite smoke imagery, to model the key controls on fire refugia occurrence and persistence through one, two, and three fire events over a 32-year period. Hotter-than- average fire weather was associated with lower refugia probability and higher fire severity. Refugia that persisted through three fire events appeared to be partially entrained by landscape features that offered protection from fire, suggesting that topographic variability may be an important stabilizing factor as forests pass through successive fire filters. In addition, smoke density strongly influenced fire effects, with fire refugia more likely to occur when smoke was moderate or dense in the morning, a relationship attributable to reduced incoming solar radiation resulting from smoke shading. Results from this study could inform management strategies designed to protect fire-resistant portions of biologically and topographically diverse landscapes.