I have fire occurrence data with x,y coordinates and area burned. Been using adaptive kernel density estimation to derive fire occurrence maps with just the x,y coordinates (evaluating variable bandwith selection based on kth nearest neighbor as described by Amatulli et al 2007- ecological modelling 200:321–333).
Does anyone have experience including fire size as an 'intensity' variable in the kernel interpolation. Having a hard time finding any fire-relevant literature on this. But seems potentially useful to identify high risk areas for larger fires.
Hi,
We did it, but have not published the data. It works just fine. Nikos Koutsias may have published some work on this topic, specifically applied to fire mapping. The literature on marked point processes ought to provide the theoretical background.
Such a map will not necessarily represent fire risk, it depends on the length of your dataset and the time scale you're analysing. If there has been a cluster of large fores in a given area, it may be depleted of fuel for a few years, therefore less prone to burning. At the time scale of decades, yes, such a cluster probably indicates a region of high fire incidence. But a few large fires does not necessarily mean high fire risk, since the return interval may be very long.Very interesting topic!
Thanks José,
Excellent point regarding large fire risk. In this case, the fire occurrence data are from Hawaii, where most of our large fires occur in nonnative grasslands - and where they occur in forests, fires will typically convert that forest to grassland. Because these grasslands recover fuel loads so quickly, I'm pretty confident that these fire occurrence maps with size as the intensity variable convey a picture of large fire risk - even if our data set is only on the time scale of a decade.
Curious about the units you are using for the fire size KDEs? For the simpler analysis of just point data, I've been using relative density (fires/sq. km).
Hi Clay,
We interpolated the centroids of fire perimeters, marked with the area of the corresponding fire. The result is dimensionless (i.e. km2/km2) and may exceed 1, in areas with high fire incidence in our 40 yr. fire atlas.
OK - that makes sense - also clear when you look back at where 'intensity' fits into the KDE equation. Thanks to both of you for your input.
I am not sure if it will benefit the KDE equation- but MODIS team in UMD has burned area product:
http://modis-fire.umd.edu/Burned_Area_Products.html