Re: [GRASS-user] v.kernel - Points weighted with attribute ?

[Micha Silver]

Hi Taïs Thanks for testing. As Moritz mentioned, the important parameter in spatial kernel density is the bandwidth. Here attached is an improved script that uses the spatialEco R package (instead of density.ppp from spatstat). To run this, you'll need to install spatialEco in your R environment. This package allows to specify both the bandwidth and the target raster extents and resolution. In this script I take the extents and resolution from the GRASS region. Bandwidth is specified on the command line. So to try this, in your running GRASS session: Rscript kernel_density.R <GRASS point vector> <weight column> <bandwidth> I also attached two images in a test I did, with bandwidths of 10,000 and 20,000. Hope this helps, Micha On 10/25/19 12:17 PM, Taïs wrote: Hi Micha, I successfully run your R script. However to output is weird and I don't know how to fix it. In v.kernel, you can setup the "raduis" parameter to control what I assume to be the size of the kernel (of the moving window). I made one test with radius=300 and another with value 3000. The result of the first is more what I would expect in terms of spatial variability. When I try your script, the output raster has a size of 128x128 which did not correspond at all to my computational region, and thus the resolution is not the same as the one defined in the computational region. The other thing is that I'm wondering if it is possible to control the size of the moving window with the "density" function in R. I already tried the 'n' parameter but it doesn't change anything.. I also tried with real weights (the number of inhabitants) and a unity weighting (value 1 for all points) to see it there is a change and there is which proof the weights affect the output :) Le 22/10/19 à 13:41, Micha Silver a écrit : Here is the script. Let me know how it goes. (I might try to take time to make it into a GRASS addon) On 10/22/19 2:33 PM, Taïs wrote: Hi Micha, thanks for your repply. For my own need, I think your solution is enough and I would like to try if you agree to send the R script. Le 22/10/19 à 12:53, Micha Silver a écrit : On 10/21/19 1:33 PM, Tais wrote: Hi all, I would like to compute a raster density map ("heat map") from vector points using v.kernel. The points represent house addresses. However, I would need to weight the points with a value stored in the attribute table (number of inhabitants). I saw on the module manual page that this functionality is a "known issue". I imagined a solution to by-passe the limitation: duplicate each point as many time as needed (according to the value stored in the attribute table) and use this new layer directly in v.kernel. However, I think it will definitely not be the most efficient way to do the trick. Do you have an alternative in mind ? Of course the best solution would be that someone in the development community would have the time and the kindness to implement this function directly in the module (I'm not skilled in C and thus cannot try myself). Maybe not the answer you are looking for, but you can create a weighted kernel density in R. I can send an R script that is intended to be run from within a GRASS session. It accepts a point vector and attrib column, and creates a kernel density raster, weighted by the attribute values (using the R defaults for bandwidth and Gaussian kernel) Requires, of course, that R is installed, with the packages: spatstat, raster, rgrass7. NB: Sorry if I should have posted this on the developer mailing list. I hesitated but decided to post here finally. Best, -- Micha Silver Ben Gurion Univ. Sde Boker, Remote Sensing Lab cell: +972-523-665918 -- Micha Silver Ben Gurion Univ. Sde Boker, Remote Sensing Lab cell: +972-523-665918 -- Micha Silver Ben Gurion Univ. Sde Boker, Remote Sensing Lab cell: +972-523-665918

# Weighted Kernel Density in R
# Using the spatialEco package
# execute from a GRASS session
# Command line arguments: GRASS point vector, column of weights
#####################################################################
suppressMessages(library(rgrass7, quietly=TRUE))
suppressMessages(library(spatialEco, quietly = TRUE))
suppressMessages(library(sp, quietly = TRUE))
suppressMessages(library(raster, quietly = TRUE))
# Test case, in GRASS location nc_basic_spm, using the schools vector 
# This vector has attribut CORECAPACI, used as weight in this case

args = commandArgs(trailingOnly=TRUE)
if (length(args) < 3) {
    	cat("Syntax:\nRscript kernel_density.R <GRASS point vector> <weight attribute> <bandwidth>")
    	print("Exiting...")
    	quit(save="no")
} else {
	input_vect = args[1]
       	output_rast = paste0(input_vect, "_density")
	weights_col = args[2]
	bandwidth = as.integer(args[3])
}

use_sp()
pts = readVECT(input_vect)
# Check geometry type and weight attribute column
if (class(pts)!= "SpatialPointsDataFrame") {
	print("Wrong geometry type. Must be POINT")
	quit(save="no")
}
weights_idx = which(names(pts) == weights_col)
if (length(weights_idx) == 0) { 
	print(paste("No attribute column:", weights_col))
       	quit(save="no")	
}

# Remove NA
pts = pts[!is.na(pts[[weights_idx]]),]
wts = pts[[weights_idx]]
print(paste("Using:", length(wts), "points"))

# Prepare newgrid object (as R raster)
execGRASS(cmd = "r.mapcalc", parameters=list(expression = 'tmp_one = 1'))
newgrid = raster(readRAST("tmp_one"))
execGRASS(cmd = 'g.remove', parameters=list(type = 'rast', name = 'tmp_one'), flags = 'f')

# Density (default Gaussian kernel)
pts_dens = sp.kde(x = pts, y = wts,
                  bw = bandwidth, newdata = newgrid)
png(paste("density_plot", as.character(bandwidth), ".png"))
plot(pts_dens, main = paste(input_vect, "Density"),
     sub = paste("bw:", as.character(bandwidth)))
dev.off()
# Convert to SGDF to save back to GRASS
pts_sgdf = as(pts_dens, "SpatialGridDataFrame")
writeRAST(x = pts_sgdf, vname = output_rast, overwrite = TRUE)

_______________________________________________
grass-user mailing list
grass-user@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-user