convert a netcdf time variable to an R date object

Question

I have a netcdf file with a timeseries and the time variable has the following typical metadata:

    double time(time) ;
            time:standard_name = "time" ;
            time:bounds = "time_bnds" ;
            time:units = "days since 1979-1-1 00:00:00" ;
            time:calendar = "standard" ;
            time:axis = "T" ;

Inside R I want to convert the time into an R date object. I achieve this at the moment in a hardwired way by reading the units attribute and splitting the string and using the third entry as my origin (thus assuming the spacing is "days" and the time is 00:00 etc):

require("ncdf4")
f1<-nc_open("file.nc")
time<-ncvar_get(f1,"time")
tunits<-ncatt_get(f1,"time",attname="units")
tustr<-strsplit(tunits$value, " ")
dates<-as.Date(time,origin=unlist(tustr)[3])

This hardwired solution works for my specific example, but I was hoping that there might be a package in R that nicely handles the UNIDATA netcdf data conventions for time units and convert them safely to an R date object?

Answer 1

There is not, that I know of. I have this handy function using lubridate, which is basically identical to yours.

getNcTime <- function(nc) {
    require(lubridate)
    ncdims <- names(nc$dim) #get netcdf dimensions
    timevar <- ncdims[which(ncdims %in% c("time", "Time", "datetime", "Datetime", "date", "Date"))[1]] #find time variable
    times <- ncvar_get(nc, timevar)
    if (length(timevar)==0) stop("ERROR! Could not identify the correct time variable")
    timeatt <- ncatt_get(nc, timevar) #get attributes
    timedef <- strsplit(timeatt$units, " ")[[1]]
    timeunit <- timedef[1]
    tz <- timedef[5]
    timestart <- strsplit(timedef[4], ":")[[1]]
    if (length(timestart) != 3 || timestart[1] > 24 || timestart[2] > 60 || timestart[3] > 60 || any(timestart < 0)) {
        cat("Warning:", timestart, "not a valid start time. Assuming 00:00:00
")
        warning(paste("Warning:", timestart, "not a valid start time. Assuming 00:00:00
"))
        timedef[4] <- "00:00:00"
    }
    if (! tz %in% OlsonNames()) {
        cat("Warning:", tz, "not a valid timezone. Assuming UTC
")
        warning(paste("Warning:", timestart, "not a valid start time. Assuming 00:00:00
"))
        tz <- "UTC"
    }
    timestart <- ymd_hms(paste(timedef[3], timedef[4]), tz=tz)
    f <- switch(tolower(timeunit), #Find the correct lubridate time function based on the unit
        seconds=seconds, second=seconds, sec=seconds,
        minutes=minutes, minute=minutes, min=minutes,
        hours=hours,     hour=hours,     h=hours,
        days=days,       day=days,       d=days,
        months=months,   month=months,   m=months,
        years=years,     year=years,     yr=years,
        NA
    )
    suppressWarnings(if (is.na(f)) stop("Could not understand the time unit format"))
    timestart + f(times)
}

EDIT: One might also want to take a look at ncdf4.helpers::nc.get.time.series

EDIT2: note that the newly-proposed and currently in developement awesome stars package will handle dates automatically, see the first blog post for an example.

EDIT3: another way is to use the units package directly, which is what stars uses. One could do something like this: (still not handling the calendar correctly, I'm not sure units can)

getNcTime <- function(nc) { ##NEW VERSION, with the units package
    require(units)
    require(ncdf4)
    options(warn=1) #show warnings by default
    if (is.character(nc)) nc <- nc_open(nc)
    ncdims <- names(nc$dim) #get netcdf dimensions
    timevar <- ncdims[which(ncdims %in% c("time", "Time", "datetime", "Datetime", "date", "Date"))] #find (first) time variable
    if (length(timevar) > 1) {
        warning(paste("Found more than one time var. Using the first:", timevar[1]))
        timevar <- timevar[1]
    }
    if (length(timevar)!=1) stop("ERROR! Could not identify the correct time variable")
    times <- ncvar_get(nc, timevar) #get time data
    timeatt <- ncatt_get(nc, timevar) #get attributes
    timeunit <- timeatt$units
    units(times) <- make_unit(timeunit)
    as.POSIXct(time)
}