Point pattern to grid

On Fri, 18 Nov 2005, Leaf Sun wrote:

Hi Roger,

Thanks again for your kind help.

Yes, I still use the 200K points data applying this program but the good
thing is I found it finished in no time.

I'm moving my reply to R-sig-geo, a more suitable list.

The questions again here are:

1)   >try0 <- lapply(split(as(df1, "data.frame"), res), mean)  

When I tried to replace mean to sum, error looks like this:

Error in x at data[, i, drop = FALSE] : undefined columns selected 

Please take a small subset of your data (say 25-50 points), and fake the 
grid (say just 4 cells 250x250) to try to reach a function that seems 
suitable. Then write your own function to give the output you need - look 
carefully at ?split and ?lapply. Your function needs to return a single 
vector (which can be a one-row data frame). The nafill values should suit 
the ones you want to return. 

2) If I just need to know the number of points in each cells, how can I
modify the codes. The codes still a bit beyond me.

It might be length() or nrow() in your own function. If you try first with 
a small example, and look at the results of each step in turn - so break:

try0 <- lapply(split(as(df1, "data.frame"), res), mean)

into

try00 <- split(as(df1, "data.frame"), res)

sgrd1 <- SpatialGridDataFrame(slot(sgrd, "grid"), try1)

image(sgrd1, "x")
image(sgrd1, "y")
image(sgrd1, "z")

look what is inside a chosen member of the list try00, and try to work it 
out from there. Once your function does what you want, run it on the full 
data set, but still be careful in case impossible values have happened 
(things like dividing by zero). Use summary() on the input point data to 
get a good feel for the spread of the data. You're very close, but you 
need to make these decisions in relation to your research problem, and how 
you've operationalised your observed variables at the points (maybe you 
need to store mean, sd, fivenum, and count for a numerical variable, but 
only you can know).

Roger

Thanks!

Leaf

======= At 2005-11-18, 01:39:05 you wrote: =======

On Thu, 17 Nov 2005, Leaf Sun wrote:

Dear all,

I'd like to change a point pattern to a grid of cells and use one of the
variables as the output.

e.g.  The point pattern is of a window of (500*500) and several features
such as pH, SoilType etc.  I like to divide it into a grid with cell
size 5*5, and use the mean of the point values falling inside the cell
as the output.

Is there any package in R working with this? Thanks in advance!

This might have been better posted on R-sig-geo. Try this:

library(sp)
df1 <- data.frame(x=runif(10000,0,500), y=runif(10000,0,500),
 z=rnorm(10000))
coordinates(df1) <- c("x", "y")
summary(df1) # SpatialPointsDataFrame
grd <- GridTopology(c(2.5,2.5), c(5,5), c(100,100))
sgrd <- SpatialGrid(grd) #SpatialGrid
bbox(sgrd)
res <- overlay(sgrd, df1)
# find which grid cells the points are in
str(res)
try0 <- lapply(split(as(df1, "data.frame"), res), mean)
# take means by grid cell - assumes all numeric columns in df1
# (soil type??) - maybe write a custom function to handle non-numeric 
# columns sensibly
try01 <- vector(mode="list", length=prod(slot(slot(sgrd, "grid"),
 "cells.dim")))
nafill <- rep(as.numeric(NA), ncol(as(df1, "data.frame")))
try01 <- lapply(try01, function(x) nafill)
# make a container to put the means in with the right number of columns
try01[as.integer(names(try0))] <- try0
# insert means into correct list elements
try1 <- data.frame(t(data.frame(try01)))
# transpose
summary(try1)
sgrd1 <- SpatialGridDataFrame(slot(sgrd, "grid"), try1)
image(sgrd1, "x")
image(sgrd1, "y")
image(sgrd1, "z")

It goes a bit further than the short description of the sp package in the 
latest R-News, and will most likely be a new method for overlay in sp. If 
these are your 200K points, it may take a little longer ...

Cheers,

Leaf

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no

= = = = = = = = = = = = = = = = = = = =

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no