Wednesday, December 17, 2014 08:00 AM – 12:20 PM
Moscone West Poster Hall
Fire regimes provide a sensitive indicator of changes in climate and human use as the concept includes fire extent, season, frequency, and intensity. Fires that occur outside the distribution of one or more aspects of a fire regime may affect ecosystem resilience. However, global scale data related to these varied aspects of fire regimes are highly inconsistent due to incomplete or inconsistent reporting. In this study, we derive a globally applicable approach to characterizing similar fire regimes using long geophysical time series, namely MODIS hotspots since 2000. K-means non-hierarchical clustering was used to generate empirically based groups that minimized within-cluster variability. Satellite-based fire detections are known to have shortcomings, including under-detection from obscuring smoke, clouds or dense canopy cover and rapid spread rates, as often occurs with flashy fuels or during extreme weather. Such regions are free from preconceptions, and the empirical, data-mining approach used on this relatively uniform data source allows the region structures to emerge from the data themselves. Comparing such an empirical classification to expectations from climate, phenology, land use or development-based models can help us interpret the similarities and differences among places and how they provide different indicators of changes of concern. Classifications can help identify where large infrequent mega-fires are likely to occur ahead of time such as in the boreal forest and portions of the Interior US West, and where fire reports are incomplete such as in less industrial countries.