About two months ago I posed a question to the listserver, to help in our
spatial analysis of roadkills along the Trans-Canada Highway in Banff National
Park. Here is a long overdue summary of the responses I received.
Hi I am working on a research project that studies roadkills on the Trans-Canada
Highway, a linear vector spanning over 141 miles. I am looking at hotspots, or
clusters of roadkills along the highway, from small to large mammals. I am
starting with a simple analysis that looks if the kills are related to the
access points, (breaks in the wildlife fence). Firstly, I was wondering if
someone knows a script that I can use that puts random points on my linear
vector, so I can look at a Chi-square test or Hines test for randomness. (Do I
need random points along the line vector or would a ploygon suffice?) Also once
I have tested for clusters, or aggregation, how do I link these clusters to the
access points. I am looking at a simple two step analysis, for testing for
clusters, and are these clusters related to the access points. It would be
great if someone could lead me in the right direction..............
You might want to look at "local statistics" and K functions. The journal
has a number of articles on the subject.
1. If you will test against differences between two distributions, then both
your distributions must be comparable. Therefore you need random points on your
2. Is the highway absolutely linear like a line between two points? In this
case you can use a simple worksheet function for random number generation like
in Excel. For testing you should only compare the coordinates of one axis
(west-east or north-south), its useful to choose the axis with the greatest
3. If the highway is not absolutely linear you can make a workaround. First
you should make a buffer around your highway with a very small distance/radius.
After buffering use a random number generator for polygons. For ArcView exists
a simple script, which you can find on the ESRI-Website: http://www.esri.com.
Because the lateral extension of the buffer is very small I think you can
neglect it in all cases where you will find very great differences between both
the distributions. For testing you should only compare the coordiantes of one
axis (west-east or south-north), its useful to choose the axis with the greatest
4. To link the clusters with your access points you can go in different ways.
For instance you can make buffers with a defined radius around your access
points and do an identity operation. You will find some tools in Atlas*GIS
(access points) or ArcView (XTools-Extension, Animal Movement Extension; both
you can find on ESRI's site). I think the Animal Movement Extension is very
useful for your aims.
5. I think the best way to compare your distributions is to perform a
nearest-neighborhood analysis between your access points and the road kills
compared with the nearest neighborhood distances between your access points and
the random points.
hope this helps
Dr. Reinhard Klenke
Society for Nature Conservation and Landscape Ecology (GNL)
phone 49-39822-20 474
fax: 49-39822-20 474
I'm not really an expert in spatial statitics but it seem to me that you might
apply the Ripley's K statisitcs to your data. I recomment you to take a look
athe following article:
O'Driscoll, R.L. 1998. Description of spatial pattern in seabird distributions
along line transects using neighbour K statistics. Mar. Ecol. Prog. Ser.
Also Boots, B.R. and A. Getis. 1988. Point Pattern Analysis. SAGE publications.
give some examples of the application of measures of dispersion in one
dimension. As for the software, well the only one I know is available is the
S-plus and you have to add the libraries Splancs and Spatial. THere is a
student version of S-plus for a little less than $100 dollars and libraries can
be downloaded from the internet. However, I am not sure if you can use the
software for one dimentsion directly, most likely you might need to make some
modifications and obviously you will need to study the S-plus language for that.
Anyway, hope this informations is of some help to you. Good luck,
School of Forestry, Wildlife and Fisheries
Louisiana State University
Baton Rouge, La.
First of all numbers from a uniform (0,1) random number generator can be mapped
onto any stretch of highway. With as many points from the random generator as
there are road kills you can then create two empirical distributions and test
for their being the same with the Kolmogorov-Smirnov statistic. This will
simply tell you that the road kills are (not) from a uniform distribution along
You actually do not have to generate the points. The cumulative distribution
along the highway has a slope of 1/L, where L is the length of the highway
segment. i.e. F=x/L, 0<=x<=L. Now make an empirical distribution of the
roadkills and use the K-S one sample test statistic.
You may also look at gap sizes and their distribution to obtain a 'best' Pearson
type distribution for waitng times, except your waiting time is simply the
distance between kills.
You may also divide the road into a set of segments and fill in the counts for a
one dimensional contingency table.
Lastly, you can use the scan statistic found in CLUSTER, CLUSTER can be
downloaded from CDC/ATSDR.
Send me your data, and I will find an unwilling student to have a look at them.
I wrote CLUSTER along with three others.
Professor of Biostatistics
USC School of Public Health, 205
Columbia, SC 29208
Which software are you using for the analysis? I can give you some hints on
Avenue (ArcVIew script language) and S+, although I have no premade script.
Shouldn't be too difficult, though.
You can just transform all your observation in distances from the starting point
in the hwy, and then you just have to draw a random number.
BTW, where on the Hwy are you working? Is it the Banff stretch?
Dept. of Forest Sciences - UBC
*To post a message to the list, send it to ai-geostats@...
*As a general service to list users, please remember to post a summary
of any useful responses to your questions.
*To unsubscribe, send email to majordomo@...
with no subject and
"unsubscribe ai-geostats" in the message body.
DO NOT SEND Subscribe/Unsubscribe requests to the list!