Critical transitions in
fire spread in
savanna landscapes
Carla Staver, EEB, Princeton University & Sally Archibald, CSIR, South Africa
Greg Asner & Ty Kennedy-Bowdoin, Carnegie Institution for Science, Stanford University
Emmanuel Schertzer & Simon Levin, EEB, Princeton University
Navashni Govender & Isak Smit, SANParks, South Africa
The distribution of fire in the earth system
(Bowman et al. 2009)
Effects of fire on the distribution of biomes
At intermediate
rainfall, fire is a
primary
determinant of
tree cover.
(Staver et al. 2011)
Fire in global biome models
…matters, but how to deal with it?
Models operate at huge scales, but fire spread interacts with vegetation at much smaller scales.
How sufficient are current mean field models?
Tree cover (and landscape fragmentation) stop the spread of fires.
How much does their configuration in landscapes matter?
(Bond et al. 2005)
Buzzwords in fire science
Self-organized criticality (SOC) = in the absence of an externally changing parameter, the system
evolves towards a critical point
Highly optimized tolerance (HOT) = introduces a global optimization principle
Both have been proposed as
explanations for power laws in the
distribution of fire sizes in boreal forest.
Fire size distributions in savanna systems
(1) The pattern is clear (?). How well
supported is the process?
(2) What is the role of memory in a
savanna, where the fuel layer in
savanna recovers quickly from?
(3) …
(Pueyo et al. 2010)
(3) Tree
cover as a barrier to fire spread
(i.e. Criticality in response to an externally changing parameter).
No framework for understanding the threshold response of fire spread
to barriers, like tree cover…
(Archibald et al. 2009)
(Staver et al. 2011)
Fire spread as percolation
=1.0
Fire spread as percolation
…and, more generally, as infection.
=0.6
Fire spread as percolation
…and, more generally, as infection.
=0.5
Fire spread as percolation
…and, more generally, as infection.
(Archibald et al. in press)
Whence data sufficiently fine-scaled and spatial?
The Carnegie Airborne
Observatory
LIDAR (LIght Detection and Radar) =
1.12m resolution vegetation height data,
good above 0.5-1m height
Imaging spectroscopy =
vegetation presence/absence
The Experimental Burn Plots
in Kruger National Park,
South Africa
Mopani EBPs
Satara EBPs
Skukuza EBPs
Pretoriuskop EBPs
The EBPs in KNP
Landscapes
N
Fire spread through an EBP (Wet Season 2yr)
Fire spread through
the EBPs
Spread appears to be relatively insensitive
to tree cover or its configuration in the
landscape, at least within limits.
Perhaps explaining the 40% threshold in
the context of heterogeneous landscapes is
not so hard after all?
• Does spatial variability in the arrangement of barriers to
spread matter less than variability in probability of spread?
• E.g. grass biomass, season of fire, etc?
• Types of spatial variability in the landscape? E.g. the
arrangement of barriers to spread at the catchment scale?
• Perhaps explaining the global 40% threshold in the context of
heterogeneous landscapes is not so hard after all?
Making probability of
spread meaningful
…is doable.
(Archibald, Staver, Levin 2012, PNAS)
(Govender et al. 2006)
(Scholes et al. 2004)
Complications translating data
into flammability landscapes…
r2=0.721*
(Hoffmann et al. 2005)
ground LAI = 2.47 + 0.95*(tree LAI)
• Comparisons of modeled results with real burns from KNP
• Fine-scale in EBPs
• Broader-scale with coarser resolution fire data
Many thanks!
The Andrew W. Mellon Foundation & DARPA
The LevLab, past and present
William Bond