HR: 0800h
AN: B11D-0779 [Abstracts]
TI: Applying Spatial Statistics to Isolate the Effects of Fuels, Topography, and Weather on Burn Severity
AU: * Wimberly, M C
EM: michael.wimberly@sdstate.edu
AF: Geographic Information Science Center of Excellence, Wecota Hall 506B
South Dakota State University, Brookings, SD 57007-3510, United States
AU: Cochrane, M A
EM: mark.cochrane@sdstate.edu
AF: Geographic Information Science Center of Excellence, Wecota Hall 506B
South Dakota State University, Brookings, SD 57007-3510, United States
AU: Baer, A D
EM: adam.baer@sdstate.edu
AF: Geographic Information Science Center of Excellence, Wecota Hall 506B
South Dakota State University, Brookings, SD 57007-3510, United States
AU: Zhu, Z
EM: zhu@usgs.gov
AF: USGS Center for Earth Resources Observation and Science (EROS), 47914 252nd St
Sioux Falls, SD, Sioux Falls, SD 57198-0001, United States
AB:
Fire severity datasets derived from satellite remote sensing data are now being used extensively in wildfire
research and land management. Maps of burn severity based on the differenced normalized burn ratio (dNBR)
are being produced and disseminated by the Monitoring Trends in Burn Severity (MTBS) project for all major
wildfires in the United States from 1984 to present. This abundance of data presents unprecedented new
opportunities for understanding how weather, terrain, and fuels interact to determine fire severity patterns, and for
testing the effectiveness of fuel-reduction strategies for mitigating wildfire impacts. However, these datasets
present challenges for statistical analysis because of their large sizes and the non-independence of spatially
autocorrelated pixels.
To explore the importance of spatial autocorrelation, we analyzed the spatial patterns of burn severity in two recent
wildfires - the 2004 School Fire in the Blue Mountains of southeastern Washington and the
2005 Warm Fire on the Kaibab Plateau in northern Arizona. Conditional autoregressive (CAR) models were fitted
with dNBR as the dependent variable and topography, fuels, and locations of recent fuel treatments as the
independent variables. In both fires, elevation, slope, and aspect had strong effects on burn severity. Fuels had
stronger effects on burn severity for the School fire than for the Warm Fire. In both fires, fuel treatments that
combined thinning and prescribed burning resulted in statistically significant reductions in fire severity. The CAR
models were then decomposed to isolate the spatial signal, which reflected spatially structured variability in
dNBR that was not related to the independent variables. The spatial signals were correlated with the burn
progression maps, reflecting spatial and temporal variability in weather and fire behavior (e.g. wind versus plume
driven) over the course of the fire. These results suggest that spatial autocorrelation in the analysis of remotely-
sensed burn severity datasets is not simply a nuisance, but in fact captures substantive and interpretable effects
of weather and fire behavior on burn severity.
DE: 0434 Data sets
DE: 0466 Modeling
DE: 0480 Remote sensing
SC: Biogeosciences [B]
MN: 2007 Fall Meeting