8. NetCDF Proxy Workflows

David Blodgett

For a better version of the stars vignettes see https://r-spatial.github.io/stars/articles/

NetCDF data sources are available via more and less granular files and/or OPeNDAP endpoints. This article demonstrates how stars enables discovery, access, and processing of NetCDF data across a wide range of such source-data organization schemes.

We’ll start with some basics using datasets included with the stars installation. A call to read_ncdf(), for a dataset smaller than the default threshold, will just read in all the data. Below we read in and display the reduced.nc NetCDF file.

library(stars)
f <- system.file("nc/reduced.nc", package = "stars")
(nc <- read_ncdf(f))
## no 'var' specified, using sst, anom, err, ice
## other available variables:
##  lon, lat, zlev, time
## 0-360 longitude crossing the international date line encountered.
## Longitude coordinates will be 0-360 in output.
## Will return stars object with 16200 cells.
## No projection information found in nc file. 
##  Coordinate variable units found to be degrees, 
##  assuming WGS84 Lat/Lon.
## stars object with 4 dimensions and 4 attributes
## attribute(s):
##                Min. 1st Qu. Median       Mean 3rd Qu.  Max.  NA's
## sst [°C]      -1.80   -0.03 13.655 12.9940841 24.8125 32.97  4448
## anom [°C]     -7.95   -0.58 -0.080 -0.1847324  0.2100  2.99  4449
## err [°C]       0.11    0.16  0.270  0.2626872  0.3200  0.84  4448
## ice [percent]  0.01    0.47  0.920  0.7178118  0.9600  1.00 13266
## dimension(s):
##      from  to offset delta         refsys         values x/y
## lon     1 180     -1     2 WGS 84 (CRS84)           NULL [x]
## lat     1  90    -90     2 WGS 84 (CRS84)           NULL [y]
## zlev    1   1     NA    NA             NA              0    
## time    1   1     NA    NA        POSIXct 1981-12-31 UTC

Let’s assume reduced.nc was 10 years of hourly data, rather than 1 time step. It would be over 10GB rather than about 130KB and we would not be able to just read it all into memory. In this case, we need a way to read the file’s metadata such that we could iterate over it in a way that meets the needs of our workflow objectives. This is where proxy = TRUE comes in. Below, we’ll lower the option that controls whether read_ncdf() defaults to proxy and use proxy = TRUE to show both ways of getting the same result.

old_options <- options("stars.n_proxy" = 100)
(nc <- read_ncdf(f, proxy = TRUE))
## no 'var' specified, using sst, anom, err, ice
## other available variables:
##  lon, lat, zlev, time
## 0-360 longitude crossing the international date line encountered.
## Longitude coordinates will be 0-360 in output.
## No projection information found in nc file. 
##  Coordinate variable units found to be degrees, 
##  assuming WGS84 Lat/Lon.
## netcdf source stars proxy object from:
## [1] "[...]/reduced.nc"
## 
## Available nc variables:
## sst
## anom
## err
## ice
## 
## dimension(s):
##      from  to offset delta         refsys         values x/y
## lon     1 180     -1     2 WGS 84 (CRS84)           NULL [x]
## lat     1  90    -90     2 WGS 84 (CRS84)           NULL [y]
## zlev    1   1     NA    NA             NA              0    
## time    1   1     NA    NA        POSIXct 1981-12-31 UTC
options(old_options)

The above shows that we have a NetCDF sourced stars proxy derived from the reduced.nc file. We see it has four variables and their units are displayed. The normal stars dimension(s) are available and a nc_request object is also available. The nc_request object contains the information needed to make requests for data according to the dimensions of the NetCDF data source. With this information, we have what we need to request a chunk of data that is what we want and not too large.

(nc <- read_ncdf(f, 
                 var = "sst", 
                 ncsub = cbind(start = c(90, 45, 1 , 1), 
                              count = c(90, 45, 1, 1))))
## 0-360 longitude crossing the international date line encountered.
## Longitude coordinates will be 0-360 in output.
## Will return stars object with 4050 cells.
## No projection information found in nc file. 
##  Coordinate variable units found to be degrees, 
##  assuming WGS84 Lat/Lon.
## stars object with 4 dimensions and 1 attribute
## attribute(s):
##          Min. 1st Qu. Median     Mean 3rd Qu.  Max. NA's
## sst [°C] -1.8   -1.04     14 12.92722   25.13 29.81  757
## dimension(s):
##      from to offset delta         refsys         values x/y
## lon     1 90    177     2 WGS 84 (CRS84)           NULL [x]
## lat     1 45     -2     2 WGS 84 (CRS84)           NULL [y]
## zlev    1  1     NA    NA             NA              0    
## time    1  1     NA    NA        POSIXct 1981-12-31 UTC

plot(nc)

The ability to view NetCDF metadata so we can make well formed requests against the data is useful, but the real power of a proxy object is that we can use it in a “lazy evaluation” coding style. That is, we can do virtual operations on the object, like subsetting with another dataset, prior to actually accessing the data volume.

Lazy operations.

There are two kinds of lazy operations possible with stars_proxy objects. Some can be applied to the stars_proxy object itself without accessing underlying data. Others must be composed as a chain of calls that will be applied when data is actually required.

Methods applied to a stars_proxy object:

Methods that add a call to the call_list.

Methods that cause a stars_proxy object to be fetched and turned into a stars object.