Main content area

Mapping prescribed fire severity in south-east Australian eucalypt forests using modelling and satellite imagery: a case study

Loschiavo, John, Cirulis, Brett, Zuo, Yingxin, Hradsky, Bronwyn A., Di Stefano, Julian
The International journal of wildland fire 2017 v.26 no.6 pp. 491-497
Eucalyptus, case studies, confidence interval, fire severity, fire spread, forest management, forests, landscapes, normalized difference vegetation index, prescribed burning, remote sensing, satellites, simulation models, wildfires
Accurate fire severity maps are fundamental to the management of flammable landscapes. Severity mapping methods have been developed and tested for wildfire, but need further refinement for prescribed fire. We evaluated the accuracy of two severity mapping methods for a low-intensity, patchy prescribed fire in a south-eastern Australian eucalypt forest: (1) the Normalised Difference Vegetation Index (NDVI) derived from RapidEye satellite imagery, and (2) PHOENIX RapidFire, a fire-spread simulation model. We used each method to generate a fire severity map (four-category: unburnt, low, moderate and severe), and then validated the maps against field-based data. We used error matrices and the Kappa statistic to assess mapping accuracy. Overall, the satellite-based map was more accurate (75%; Kappa±95% confidence interval 0.54±0.06) than the modelled map (67%; Kappa 0.40±0.06). Both methods overestimated the area of unburnt forest; however, the satellite-based map better represented moderately burnt areas. Satellite- and model-based methods both provide viable approaches for mapping prescribed fire severity, but refinements could further improve map accuracy. Appropriate severity mapping methods are essential given the increasing use of prescribed fire as a forest management tool.