Defining Extreme Wildfires from Geospatial Data
The entire project team (social sciences, remote sensing scientists, and ecologists) worked together to explore whether potentially extreme wildland fire events could be determined through analyzing different geospatial metrics. This analysis has led to Lannom et al (2014, IJWF). In this study, a temporal series of Landsat imagery was acquired from 1984-2009 for all fires that occurred within the western United States. This temporal series was analyzed to identify both widespread fire years (i.e. years with significantly larger quantities of area burned) and individually extreme wildfires, using metrics based on fire size, percentage of area burned with high vegetation mortality, duration, and distance to the Wildland Urban Interface (WUI). Potentially extreme wildland fire events were identified using the cross-section of distributional statistics for each metric.
Results: An important result was recognition and agreement of the methodology used to characterize (A) “extreme fire years” and to characterize (B) “individual wildland fires as extreme”.
Product A: Method to identify “Extreme Disturbance Years”
In agreement with the work of Dillon et al (2011, Ecosphere), the project team within the Lannom et al (2014) paper used the Tukey’s 75th percentile + 1.5 interquartile–range (IQR) rule to identify statistically significant “extreme fire years”. Importantly, this concept can be readily applied to evaluating “extreme years” of any disturbance from geospatial data acquired from NASA assets, e.g., “extreme hurricane years”, or “extreme flood years”.
Product B: Method to identify “Extreme Individual Disturbance Events”
The method presented in Lannom et al (2014) collates continuous distributions of various geospatial metrics related to biophysical and social impacts of fires. On the continuous data the 90th, 95th and 99th percentiles were calculated and the fires identified. Fires that were present in these percentiles for 75% of the tests were identified as potentially extreme. This was done separately for the 90th, 95th, and 99th.
The project team has also conducted and completed a series of projects to assess the utility of various geospatial, ecology, and social science metrics as indicators of extreme fires:
Results: We demonstrate that the MTBS burned area product has high commission errors (4-16%) and if it is used to generate time series data of area burned, the burned area perimeters should be constrained by spectral index based classifications of burned and unburned surfaces. This work was also partially funded under award NNX11AF19G (Boschetti).
Results: Results indicate a significant relationship: FRE per kilogram of fuel consumed =-5.32 WC+ 3.025 (were WC = water content) and imply that not taking into account fuel moisture variations in the assumed relationship between FRE and fuel consumed can lead to systematic biases. A similar relationship has been determined for another pine species and ongoing research in summer 2014 is evaluating peat fuels.
Product C: Method to determine fuel moisture controls on fire radiative energy (FRE) retrievals
Results: They found that daily area burned is a poor predictor of burn severity as the correlation is significant but weak, even for the very largest (95th percentile) of Daily Area Burned (Birch 2013, Birch et al 2014). This suggests that other factors influence burn severity. They found that topography and vegetation were more important than climate and weather in influencing vegetation mortality (Birch 2013, Birch et al. in review).
Results: Carbon in trees killed by wildfires during 1984-2010 was 146-285 Tg, representing 2.4-4.6% of the total carbon in trees in the western US, and was highest in northern California/southern Oregon, central Idaho, and northwest Wyoming (Appendix B, Figure 8).
Product D: Methodology to construct forest-fire chronosequence from Landsat imagery.
In year 3 we made substantial progress on exploring the utility of different temporal series data to characterize trajectories. Data sets included Landsat, sub-orbital aerial photography, and agency records of burned area. Highlights in development of the fire recovery chronosequence include:
Results: It is possible to assemble a 100 year post-fire recovery chronosequence for forested systems within the project area. Ongoing research in preparation for Remote Sensing of Environment includes incorporating USDA Forest Service Forest Inventory Analysis data for these areas to link in situ recovery with spectral recovery.
Result: They concluded that too much variation exists to reliably use the method in these fires. Chronosequences in rangelands are much more difficult to identify than in forests because of the seasonal variability and differences between ecological sites. Due to these issues they are changing direction and are now looking at quantifying change based on ecological site and pre-fire vegetation cover levels.