Estimating monthly climatology on NetCDF data using R

问题

I am working on NOAA AVHRR daily Sea Surface Temperature (SST) data of 31 years. The data is in NetCDF format with dimensions as 28 (lon) x 40 (lat) x 11686(days). I am supposed to compute monthly climatological mean (e.g. mean of all Januaries of 31 years and so on). Using ncdf4 and chron libraries I was able to get it in array form.

ncin <- nc_open('sstfile.nc')
sst_array <- ncvar_get(ncin, 'sst')

Since time variable is separate from SST data, I had to use it loop over the array.

is.leapyear <- function(year){
return(((year %% 4 == 0) & (year %% 100 != 0)) | (year %% 400 == 0))
}

dateseq <- seq(as.Date("1987-01-01"), as.Date("2018-12-31"), by=1)

Using raster library I am converting to rasters and then doing computation.

for ( i in seq(11686)) {
dtft <- strsplit(as.character(as.Date(dateseq[i])), split = '-')
y <-  as.integer(dtft[[1]][1])
m <-  as.integer(dtft[[1]][2])
d <-  as.integer(dtft[[1]][3])
while (m == 1){
assign(paste0('r',y,'.',d), raster(matrix(sst_array[1:27, 1:38, i], 
nrow = 27, ncol = 38)))
m = m + 1
}
if (is.leapyear(y) == TRUE) (i = i + 366)
else (i = i + 365)
}

Problem is it's creating far too many rasters and first computing monthly mean and then yearly.

r87jan <- stack(mget(paste0('r1987.',1:31)))
r87janmean <- calc(r87jan, mean)

Is there any function/method which can compute over this time duration without making so many rasters and computation can remain as array or a matrix? Or can the above code can be improved to compute the monthly mean for all years at once?

回答1:

You do not provide your data, but I think you can do something like this:

library(raster)
nc <- brick('sstfile.nc')

dates <- getZ(nc)
months <-  as.integer(format(dates, "%m"))

s <- stackApply(nc, months, fun=mean)

回答2:

If I can offer an answer not in R, if you have cdo (climate data operators installed) you can simply do this on the linux command line:

cdo ymonmean sstfile.nc sst_climate.nc

the file sst_climate.nc will contain 12 timesteps, with the average of all the January, Feb and so on...

You can install cdo easily in, say, ubuntu/mint with

sudo apt-get install cdo

and you can also install ubuntu easily within Windows 10 these days to access these useful tools easily. Documentation is available here https://code.mpimet.mpg.de/projects/cdo/

来源：https://stackoverflow.com/questions/54821567/estimating-monthly-climatology-on-netcdf-data-using-r

标签

netcdf

cdo-climate