Kernel Density Mapping?

[Ari Jolma]

Bill Binko kirjoitti:

> Brent,
>
> Thanks for the info.  I will certainly look at the R tools: I was 
> unaware of an S replacement (it's been years since I've seen S).
>
> It does seem (from what little research I've done) that these R tools 
> will generate images, and not shapefiles.
>
> I am primarily like to take Point datasets, and generate polygon 
> datasets.  I suppose I could take R, generate rasters, and then use 
> gdal_contour to generate the vectors?
>
> However, I am willing (and probably capable) of writing a tool that 
> can process the points, run the kernel function and generate the 
> resulting contours directly.  Does anyone know where I can find a good 
> description of the algorithm and/or some sample code for it?  Perhaps 
> one of the professors out there?


I have no experience with this (despite my status :-) but I believe the 
idea at its simplest is that the effect of the points (e.g., location of 
an incident, which has no area) are "spread out" using the kernel 
function (which has a value over some area), and thus you get a 
continuous function of incidents/unit area. The parameters of the kernel 
function are of course subjective, so visualization is used to select 
"the best" values. Google gave me this URL on doing spatial kernel 
density estimation with R:

http://www.math.uni-klu.ac.at/~agebhard/ERSA-98-D8-427/node6.html

but I'd go to some handbooks to learn more. I think the use of rasters 
for computing the density map on is a practical thing. It is easy to 
compute the density at any given point but it is much more difficult to 
find the points where the density gets a specific value (which you would 
need to do if you want to compute the contours directly).

Ari

>
> Of course, if I did so, I'd be happy to push it back to the community.
>
> Bill
>
> Brent Wood wrote:
>
>> --- Bill Binko <bill at BINKO.NET> wrote:
>>
>>  
>>
>>> Hi everyone,
>>>
>>> [I've been sidelined recently battling some painful bugs in GDAL, 
>>> Mapserver, and several raster libs, so I've been putting off some 
>>> more general questions I'd love to pose to the list.  This is my 
>>> first, but I'll probably send several tonight (sorry to flood).]
>>>
>>> I'm interested in creating "kernel method" density maps using vector 
>>> point data.
>>>   
>>
>>
>> Look at R for this sort of thing, I know several fisheries modellers 
>> using R
>> for kernel density analyses. R/PostGIS/GRASS/QGIS/mapserver (with 
>> Proj.4, GDAL,
>> GEOS, etc...) all work pretty well together to support such work as 
>> an Open
>> Source suite.
>>
>> See:
>> http://www.stat.math.ethz.ch/R-manual/R-devel/library/stats/html/density.html 
>>
>>
>> For a more general list of R functions, 
>> http://www.oulu.fi/atkk/tkpalv/unix/R/library/stats/html/00Index.html
>>
>> and for general R spatial packages/info,
>> http://r-spatial.sourceforge.net/
>>
>>
>>
>> Hope this helps....
>>
>>  Brent Wood
>>
>>
>>  
>>


-- 
Prof. Ari Jolma
Kartografia ja Geoinformatiikka / Cartography and Geoinformatics
Teknillinen Korkeakoulu / Helsinki University of Technology
POBox 1200, 02015 TKK, Finland
Email: ari.jolma at tkk.fi URL: http://www.tkk.fi/~jolma