Wanting extract to produce sensible answers for NAs in R

Question

So, I'm working on extracting values from multiple different rasters in R. However, some of our data is marine, some terrestrial. Some rasters cover both, but a few are ocean only or land only. At times, this leads to extract in the raster package throwing errors in unpredictable ways.

For example, if I throw a land point into the mix when extracting sea surface temperatures, I WANT an NA return value. However, despite my best efforts, I still get the dreaded

Error in apply(x, 2, fun2) : dim(X) must have a positive length

error.

I don't get it consistently. Sometimes I get NA returns, which is what I want. But for some points...no. Here's what I've done. Does anyone have any pointers on avoiding how this extraction breaks?

To try this example, grab the data from http://www.metoffice.gov.uk/hadobs/hadisst/data/HadISST_sst.nc.gz

library(raster)

hadsst <- raster::brick("./HadISST_sst.nc")

#A lat/long throwing errors
ull <- data.frame(site_lat = 37.6, site_long = 101.3)  

#A function to make things safe
safe_mean <- function(x){
  if(sum(is.na(x))==length(x)) return(NA)
  if(sum(x == -1000, na.rm=T) == length(x)) return(NA)
  return(max(x, na.rm=T))

}

#Where it all goes wrong
hadsst_vals <- raster::extract(hadsst, 
                               cbind(ull$site_long, ull$site_lat),
                               buffer=50000, fun=safe_mean, na.rm=FALSE)

Answer 1

To work this out, I would do

hadsst_list <- raster::extract(hadsst, cbind(ull$site_long, ull$site_lat), buffer=50000)

and then

sapply(hadsst_list, safe_mean)

if that fails

for (i in 1:length(hadsst_list)) {
    safe_mean(hadsst_list[[i]])
}

hadsst_list[[i]]

to see why this happens.