The coordinates of the fire locations were provided in terms of
latitude and longitude, to the nearest minute of arc. These were
converted to New Brunswick stereographic projection coordinates
(Thomson, Mephan and Steeves, 1977) which was the coordinate
system in which the map of New Brunswick --- which constitutes the
observation window for the pattern --- was obtained. The conversion
was done using a C program kindly provided by Jonathan
Beaudoin of the Department of Geodesy and Geomatics, University of
New Brunswick.Finally the data and window were rescaled since the use of the
New Brunswick stereographic projection coordinate system resulted
in having to deal with coordinates which are expressed as very
large integers with a bewildering number of digits. Amongst other
things, these huge numbers tended to create very untidy axis labels
on graphs. The width of the bounding box of the window was made
equal to 1000 (nameless) units. In addition the lower left hand
corner of this bounding box was shifted to the origin. The height
of the bounding box was changed proportionately, resulting in a
value of approximately 959.
The window for the fire patterns comprises 6 polygonal components,
consisting of mainland New Brunswick and the 5 largest islands.
Some lakes which should form holes in the mainland component are
currently missing; this problem may be remedied in future releases.
The window was formed by ‘simplifying’ the map that was originally
obtained. The simplification consisted in reducing (using
an interactive visual technique) the number of polygon edges in
each component. For instance the number of edges in the mainland
component was reduced from over 138,000 to 500.
For some purposes it is probably better to use a discretized
(mask type) window. See Examples.
Because of the coarseness of the coordinates of the original
data (1 minute of longitude is approximately 1 kilometer at the
latitude of New Brunswick), data entry errors, and the simplification
of the observation window, many of the original fire locations
appeared to be outside of the window. This problem was addressed
by shifting the location of the ‘outsider’ points slightly,
or deleting them, as seemed appropriate.
Note that the data contain duplicated points (two points at the
same location). To determine which points are duplicates,
use duplicated.ppp.
To remove the duplication, use unique.ppp.
The columns of the data frame comprising the marks of
nbfires are:
Note that due to data entry errors some of the “out dates” and
“out times” in the original data sets were actually earlier
than the corresponding “discovery dates” and “discover times”.
In such cases all corresponding entries of the marks data frame
(i.e. dis.date, dis.julian, out.date, and
out.julian) were set equal to NA. Also, some of the
dates and times were missing (equal to NA) in the original
data sets.
The ‘ignition source’ data were given as integer codes
in the original data sets. The code book that I obtained
gave interpretations for codes 1, 2, ..., 15. However
the actually also contained codes of 0, 16, 17, 18, and in
one instance 44. These may simply be data entry errors.
These uninterpretable values were assigned the level
unknown. Many of the years had most, or sometimes
all, of the ignition source codes equal to 0 (hence turning
out as unknown, and many of the years had many
missing values as well. These were also assigned the
level unknown. Of the 7108 fires in nbfires,
4354 had an unknown ignition source. This variable
is hence unlikely to be very useful.
There are also anomalies between cause and ign.src,
e.g. cause being unknown but ign.src
being cigs, burn.no.perm, mach.spark,
hot.flakes, dump.fire or ashes. Particularly
worrisome is the fact that the cause ltning (!!!) is
associate with sources of ignition cigs, burn.w.perm,
presc.burn, and wood.spark.