| Sumario: | We use LANDSTAT-TM imagery to estimate the impact of fire on a Mediterranean forest and shrubland. In our previous report, we described our segmentation-based approach to delimit the fire scar and presented a metric for fire impact, which was based on a distance in the plane defined by the first two KauthThomas components. In this report, we further deal with the assessment of fire impact by the joint analysis of pre- and post-fire images, taking the response of different landcover categories and the eventual phenologic change into account. After an atmospheric standardization, we conducted an stratified exploratory analysis of the change between both images, for which we first produced a segmentation-based classification of the pre-fire image. After studying the trajectories of burnt and not burnt regions, we defined an index of fire impact as the difference between the actual post-fire image and a model of a not-burnt image at the post-fire date. We used a simple median model and a regression tree model, with the median model producing a better discrimination of field-assessed degrees of fire impact. Finally, we produced three estimated images of fire impact by applying (i) a simple minimum absolute distance allocation rule, (ii) a regression tree, and (iii) an exponential model fitted to the data. Error, which was estimated by a jackknife procedure, was lowest for the first method.
|
|---|
