Evaluation of wildfire propagation susceptibility in grasslands using burned areas and multivariate logistic regression

Xin Caoa, Xihong Cuia, Miao Yueb, Jin Chena, Hiroki Tanikawac, Yu Yed
a State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, Beijing 100875, China;
b Computer School of China West Normal University, Nanchong 637002, China;
c Graduate School of Environmental Studies, Nagoya University, Nagoya 464-8601, Japan;
d Aerospace Surveying and Mapping Remote Sensing Information Processing Centre, Beijing 102102, China.
Abstract: This research simulated wildfire propagation susceptibility based on multivariate logistic regression. Moderate Resolution Imaging Spectrometer (MODIS)-derived fuel indicators and topographic factors were the independent variables, and burnt areas served as the dependent variable. MODIS data were collected daily during the wildfire seasons of April to May and September to October from 2001 to 2007 to acquire information about live and dead fuel in the Mongolia–China grasslands. The inputs for the independent parameters for wildfire propagation susceptibility modelling were the normalized difference vegetation index (NDVI), optimized soil-adjusted vegetation index (OSAVI), moisture stress index (MSI), global vegetation moisture index (GVMI), dead fuel INDEX (DFI), elevation, slope, and aspect. Multivariate logistic regression ranking indicates that DFI, MSI, DEM, and OSAVI are the top four factors, with an overall accuracy of 80%. ‘Leave one out’ cross-validation demonstrated that the overall accuracy of the propagation susceptibility modelling ranged from 65% to 87%. Finally, the model was used to produce 10 day average wildfire propagation susceptibility maps during the wildfire seasons of 2001–2007 and to predict the location of burned areas. This research will be useful for understanding the propagation susceptibility of wildfires in grassland areas and for creating policies for preventing wildfire spread.
Published in International Journal of Remote Sensing. 2013, DOI: 10.1080/01431161.2013.805280.