将 kernel 密度估计提取到 R 中的新样本点

Question

I have spatial referenced data (shapefiles) of 400 polygons and points that are spread over them. I wish to create a raster of kernel density estimate for each polygon based on the points it contains. After that, I want to create a sample of 100 random points for each polygon, where each point has a spatial reference (x and y coordinates) and a kde value. All I managed to do until now was to create a kde image in the shape of the polygons using the 'ppp.density' function, but the output comes out different than the one described in the document and I don't know in what format it is and how to obtain the kde values to a sample of new points.

Any help would be much appreciated.

buffer <- st_read(".../.././buffers.shp")
pbb<- st_read(".../.../.././pbb.shp")

 library(spatstat)
 
 for (p in 1:400) {
  if(p %in% pbb$value) {
    poly123<- pbb[pbb$value == p,]
    
    C <- as.owin(buffer$geometry[p])
    point<- ppp(poly123$X,poly123$Y, window = C)
    d <- density(point, kernel = "gaussian")
    plot(d)

kde plot

The 'density' output is this:

Answer 1

You are using the function density.ppp (not "ppp.density") from the spatstat package.

If p is a point pattern (class ppp ) then D <- density(p) computes the unnormalised kernel density estimate. See help(density.ppp) for explanation and examples.

The result D is a pixel image (class im ) which can be printed, plotted and subsetted, etc. (In your question you have just printed the internal structure of the im object.)

After computing the pixel image D , if you want to know the values of D at some spatial locations, then you could

• make the query locations into a point pattern X , then extract the numerical values at those locations by values <- D[X] ; or
• convert the pixel image to a function by f <- as.function(D) then evaluate the function at desired coordinates x,y by values <- f(x,y) .

If you want to know the density values at the original data locations p , then it is faster and more accurate to use the at argument to density.ppp :

values <- density(p, at="points")

Warning: In your example (and the examples above), the smoothing bandwidth sigma was not specified. The default value of sigma may be unsatisfactory for your data. See the help file and examples for density.ppp .