Force 'float' precision for the coordinate variable of an unlimited dimension variable (time) in netcdf using R(ncdf4)

Question

The command ncdim_def from package ncdf4 is built to encourage the automatic creation of the coordinate variable that goes with the dimension, a very good practice.

However, it only allows to create "double" or "integer" precision for this coordinate variable. For external reasons, I need write the coordinate variable "time" as a float.

To do so, I use the following structure which consists in creating the coordinate variable separately from the dimension definition ( ie. using the option create_dimvar = FALSE described in the doc of ncdim_def)

timevalue= seq(0.5,10.5)
VAR1value= seq(10.2,20.2)

# define time dim, but without the time var
timedim <- ncdim_def( name   = 'time'  ,
                  units  = '', 
                  vals   = seq(length(timevalue)),
                  unlim  = TRUE,
                  create_dimvar = FALSE)

# define time coordinate variable  
timevar <- ncvar_def(name  = 'time',
                 units = 'days since 1950-01-01 00:00:00',
                 dim = list(timedim), 
                 longname = 'time', 
                 prec = "float")

# define another variable
var1var<- ncvar_def(name = 'VAR1',
                units    = 'unit1',
                dim      = list(timedim),
                missval  = -9999.0, 
                longname = 'VAR1 long name')

defVar<-list(timevar,var1var)

# creating ncfile (removing any previous one for repeated attempt)
ncfname='test.nc'
if (file.exists(ncfname)) file.remove(ncfname)
ncout   <- nc_create(ncfname,defVar,force_v4=T, verbose = T)

# writing the values
ncvar_put(ncout,timevar,timevalue)
ncvar_put(ncout,var1var,VAR1value)

nc_close(ncout)

However, this returns the error :

"ncvar_put: warning: you asked to write 0 values, but the passed data array has 11 entries!"

Indeed, the resulting netcdf shows (ncdump) :

dimensions:
    time = UNLIMITED ; // (0 currently)
variables:
    float time(time) ;
        time:units = "days since 1950-01-01 00:00:00" ;
    float VAR1(time) ;
        VAR1:units = "unit1" ;
        VAR1:_FillValue = -9999.f ;
        VAR1:long_name = "VAR1 long name" ;

I guess I need to force the dimension of the unlimited 'time' dimension upon creation, but i don't understand how to this in the frame of ncdf4.

Roman Roman · Accepted Answer · 2020-02-11T10:00:35

I ran into the same issue (needing the time dimension as float instead of double) and was able to solve it by defining the time variable as a regular non-unlimited dimension with the length of my hours vector

hours <- c("595377")
t <- ncdim_def("time", "", 1:length(hours), create_dimvar = FALSE)
timevar <- ncvar_def(name = "time",
                     units = 'hours since 1950-01-01 00:00:00',
                     dim = list(t),
                     longname = "time",
                     prec = "float")
variables <- list(timevar)
ncnew <- nc_create("TEST.nc", variables )
ncvar_put(ncnew, timevar, hours, start=c(1), count=c(length(hours)))
nc_close(ncnew)

Not sure though if this only works if the time dimension only has one value...

Force 'float' precision for the coordinate variable of an unlimited dimension variable (time) in netcdf using R(ncdf4)

1 Answers