ERA5 netcdf encoding since the new CDSAPI release

Feedback and requirements
, ,
1

Hi,
I noticed that the result of requests in netCDF format for ERA5 product (reanalysis-era5-single-levels) are now declared in double precision. Before, the results were in short, with add_offset and scale_factor to convert values (done smoothy by a lot of softs, ferret ncview, xarray etc)
In my mind, that change have 2 inconvenients :

  • increasing the downloading time,
  • increasing disk usage on both side, ECMWF, and on my side
    I understand that the new cdsapi extract data from a pool in grib format. But why to drop out the previous encoding in netCDF format ?
    Thank you
2

At first glance, there are now a lot of extra insignificant digits in the data, which are essentially noise, and do not compress well.

3

Another issue is the poor reading performance that results from the chunk sizes chosen.

I point to these two posts raising this issue: Specifying the chunk size for netCDF files and Huge performance drop when reading new ERA5 netcdf4 format