This dataset is a record of forest fires in the Castilla-La Mancha region of Spain between 1998 and 2007. This region is approximately 400 by 400 kilometres. The coordinates are recorded in kilometres.
The dataset clmfires is a point pattern (object of class
"ppp") containing the spatial coordinates of each fire,
with marks containing information about each fire. There are 4
columns of marks:
cause | cause of fire (see below) |
burnt.area | total area burned, in hectares |
date | the date of fire, as a value of class Date |
julian.date | number of days elapsed since 1 January 1998 |
The cause of the fire is a factor with the levels
lightning, accident (for accidents or negligence),
intentional (for intentionally started fires) and
other (for other causes including unknown cause).
The format of date is “Year-month-day”, e.g.
“2005-07-14” means 14 July, 2005.
The accompanying dataset clmfires.extra is a list
of two items clmcov100 and clmcov200 containing covariate
information for the entire Castilla-La Mancha region. Each
of these two elements is a list of four images (objects of
class "im") named elevation, orientation,
slope and landuse. The landuse image is
factor-valued with the factor having levels urban,
farm (for farms or orchards), meadow,
denseforest (for dense forest), conifer (for conifer
forest or plantation), mixedforest, grassland,
bush, scrub and artifgreen for artificial
greens such as golf courses.
These images (effectively) provide values for the four
covariates at every location in the study area. The images in
clmcov100 are 100 by 100 pixels in size, while those in
clmcov200 are 200 by 200 pixels. For easy handling,
clmcov100 and clmcov200 also belong to the
class "listof" so that they can be plotted and printed
immediately.
data(clmfires)clmfires is a marked point pattern (object of class "ppp").
See ppp.object.
clmfires.extra is a list with two components, named
clmcov100 and clmcov200, which are lists of pixel images
(objects of class "im").
The precision with which the coordinates of the locations of the fires changed between 2003 and 2004. From 1998 to 2003 many of the locations were recorded as the centroid of the corresponding “district unit”; the rest were recorded as exact UTM coordinates of the centroids of the fires. In 2004 the system changed and the exact UTM coordinates of the centroids of the fires were used for all fires. There is thus a strongly apparent “gridlike” quality to the fire locations for the years 1998 to 2003.
There is however no actual duplication of points in the 1998 to 2003
patterns due to “jittering” having been applied in order to
avoid such duplication. It is not clear just how the fire
locations were jittered. It seems unlikely that the jittering was
done using the jitter() function from R or the
spatstat function rjitter.
Of course there are many sets of points which are virtually identical, being separated by distances induced by the jittering. Typically these distances are of the order of 40 metres which is unlikely to be meaningful on the scale at which forest fires are observed.
Caution should therefore be exercised in any analyses of the patterns for the years 1998 to 2003.
if(require(spatstat.geom)) {
plot(clmfires, which.marks="cause", cols=2:5, cex=0.25)
plot(clmfires.extra$clmcov100)
# Split the clmfires pattern by year and plot the first and last years:
yr <- factor(format(marks(clmfires)$date,format="%Y"))
X <- split(clmfires,f=yr)
fAl <- c("1998","2007")
plot(X[fAl],use.marks=FALSE,main.panel=fAl,main="")
}
Run the code above in your browser using DataLab