Creating a new raster layer based on mean of other raster layers, ignoring NA values in R

Question

I am trying to create a new raster layer that is made up of the mean values from 7 other raster layers. I get results using overlay function mean:

mean <- overlay(2000, 2001, 2002, 2003, 2004, 2005, fun=mean)

The results I get however are skewed by a large amount 0 values in many of the cells. As the data within the rasters changes each year some cells could have 0 values one year and a value of 56 the following, for example any particular cell may contain the following:

+----------+---------+
|    Year  |  Value  |
+----------+---------+
| 2000     | 0       |
| 2001     | 54      |
| 2002     | 78      |
| 2003     | 102     |
| 2004     | 0       |
| 2005     | 0       |
+----------+---------+

So while the mean is:39, I am trying to achieve the mean ignoring the 0 values so the result should be: 78.

I have tried changing the 0 values to NA:

2000[2000 == 0]<-NA
2001[2001 == 0]<-NA
2002[2002 == 0]<-NA
2003[2003 == 0]<-NA
2004[2004 == 0]<-NA
2005[2005 == 0]<-NA

Which works fine, however the bit that I can't get to work is the calculating of the mean as before but with ignoring these NA values. I have tried

mean <- overlay(2000, 2001, 2002, 2003, 2004, 2005, fun=mean,  na.rm=TRUE)

but the results this returns are incorrect. So I am just looking for a way to find the mean whilst ignoring zero values or NA values.

Answer 1

I wanted to point out that you can rewrite the mean function, mean, which you can write yourself to do anything you want, including ditch the 0 values and calculate the mean. For example, if you want to ignore 0s:

meanIgnoringZeroes <- function(x) {
  mean(x[x!=0],na.rm=T)
}

Then you can pass the function, meanIgnoringZeroes to overlay:

mean <- overlay(r2000, r2001, r2002, r2003, r2004, r2005, fun=meanIgnoringZeroes)

You can make the function anonymous as well:

mean <- overlay(r2000, r2001, r2002, r2003, r2004, r2005, fun=function(x){ mean(x[x!=0],na.rm=T)})

Answer 2

You cannot have R objects called "2000", so presumably these are fake names? Your example actually should work, so you may want to double check why you think that the results are incorrect.

@aaryno's approach should work. I would do this:

library(raster)
s <- stack(r2000, r2001, r2002, r2003, r2004, r2005)
x <- reclassify(s, cbind(0, NA))
r <- mean(x, na.rm=TRUE)