Calculate majority using circular moving window on categorical data in R

Question

I'm trying to calculate the majority value in a categorical raster data set in R, for example using land cover data. This would be similar to the focal statistics tool in ArcGIS using the majority statistic. I am able to calculate the majority land cover type using a rectangular moving window and the modal function:

library(raster)

# create data
r <- raster(nrows = 120, ncol = 120, xmn=0)
r[] <- sample(3, ncell(r), replace=TRUE)

a<-focal(r, w=matrix(1,3,3), fun=modal)    # 3x3 moving window
plot(a)

However, when I apply a focal weight to define my circular moving window, the output values appear to be treated as continuous numbers and not discrete values, even if I convert the raster to a factor:

#convert to a factor factor
r.f<-as.factor(r)
#set up window
fw <- focalWeight(r.f, 4.5, type='circle')
#apply focal fxn
r.f.focal<-focal(r.f, w=fw, fun=modal, na.rm=TRUE) 

It appears zeros might be added into the analysis and that is creating the problem. Can anyone steer me in the right direction?

Answer 1

I think it will be fixed if d (which is now 4.5) is an integer. Your defined radius should be based on cell counts (1,2,3,...,n). However, still, the output will be in floating format as each xij in the kernel is a floating point with sum of 1. To achieve an integer output there is also a third option.

library(raster)

set.seed(070319)
# create data
r <- raster(nrows = 120, ncol = 120, xmn=0)
r[] <- sample(3, ncell(r), replace=TRUE)
a<-focal(r, w=matrix(1,3,3), fun=modal)    # 3x3 moving window

par(mfrow=c(1,2)) 
plot(r)
plot(a)

#set up window
fw <- focalWeight(r, 4, type='circle')
#apply focal fxn
r.f.focal<-focal(r, w=fw, fun=modal, na.rm=TRUE)

par(mfrow=c(1,2)) 
plot(r)
plot(r.f.focal)

for integer output you can also do this:

#set up window
fw <- ceiling(focalWeight(r, 4, type='circle'))#for integer output
#apply focal fxn
r.f.focal<-focal(r, w=fw, fun=modal, na.rm=TRUE)

par(mfrow=c(1,2), oma=c(0,0,0,1)) 
plot(r)
plot(r.f.focal)