Inferring extreme fire theory from land surface models: from imperfect proxies to predictive power.

Wildfire cause, effect and severity are driven by interactions between an array of climatic, biotic, and anthropogenic factors at multiple spatio-temporal scales.  While a broad theory of fire causation has been unveiled by a vast body of in vivo, in vitro and satellite studies, this complexity and wildfire’s destructive nature have precluded large-scale experimentation of remaining unresolved drivers and mechanics. This hampers theoretical advances for fire prediction at scale, acutely so where global climate and anthropogenic change amplifies hitherto minor or only-hypothesised processes.  Here, we show that where process representation is task-sufficient and appropriate, global land surface models can step in to infer and resolve these theoretical gaps.  This is possible precisely because these models currently fail to reproduce observed burned area and/or fire intensity patterns in a substantive number of space-time and biome-level configurations, despite reasonable performance at global and annual scales.  These in turn provide clues towards the primary theoretical deficiencies in contemporary fire ecology, as well as a platform for resolving them.

 

We present two studies that achieve this, which suggest that appropriate construction of model protocols enables hypothesis testing that can reject the null where simulation outcomes simultaneously meet both alternative hypothesis criteria and expected simulation improvements with respect to observed patterns, paving the way for improved theoretical understanding and predictive capacity.

 

The first study constructs a simplified yet powerful proxy for anthropogenic land fragmentation’s effects on fire activity at global scale.  Including this complex interaction of increased human ignition potential, fire size constriction, wind infiltration and land surface desiccation drives fire decreases in temperate and cold areas of moderate to high population density, while causing substantial increases in tropical areas subject to high levels of fragmentation.  In aggregate, including fragmentation effects decreased simulated global burned area by -6% and increased it by +5% (-1% net), while 7% of grid cells’ fire activity was affected by >25%.  These results were consistent with both global and regional (e.g. Brazil, Indonesia) -scale statistical and fire-fragmentation relationships.  

The second study provides a solution for representing the critical bifurcation of fire phenomena and severity between boreal Eurasia and North America, previously unachievable in global land surface models. Our solution results in wide-ranging improvements to the simulated space-time patterns of boreal burned area, fire intensity and their divergence.  The initial theoretical gap was addressed by hypothesizing that a previously described (Rogers et al., 2015) vegetation -and hence fire -ecology split between the two continents could be fundamentally defined by a top-down (climatic) signal, rather than the bottom-up (vegetation) driver identified by that study, which cascaded into ground/crown fire probability, fire spread and combustion dynamics. 

Process-based theoretical inference, in combination with high resolution machine learning techniques, may pave the way for future advances in global-scale fire ecology.