ERA Interim, how to handle total precipitation

Question

I am trying to compute accumulated annual precipitation for year 2000-2001 based on the ERA Interim, but I am stuck with some unresolved questions:

If I download the synoptic monthly means of "total precipitation" from ECMWF for year 2000-2001 with "Select time" set to "00:00:00" and "12:00:00" respectively, and chose "Select step" as "3", I get a *.nc -file with 48 time steps. The file also contains a "scale_factor" and "add_offset". I assume that I must use those values in (add_offset + (downloaded data * scale_factor)) to get actual precipitation in meters? And I assume that if I sum the 48 time steps of data (corrected with add_offset and scale_factor) and divide by 2 (2 years of data) then I get the average accumulated precipitation per year. Is this correct?

What then puzzles me is that the "scale_factor" is very low (3.9E-8), meaning that the values I get will be almost equal to the "add_offset" which is 0.0013. Are there any obvious mistakes I make? Or do any of you know of a guide that is more elaborate than the documentation found at ECMWF?

Thanks a lot in advance for any ideas.

Answer 1

First of all, these netCDF files follow the CF Metadata Conventions, which describe the use of scale_factor and add_offset in section 8.1 Packed data of the conventions description. In short, you're applying them correctly:

If both attributes are present, the data are scaled before the offset is added.

However, I think that you've selected the wrong fields for your purpose. The total precipitation variable is an accumulation since the start of the re-forecast, and you've selected a monthly mean of accumulations over the first 3 hours of the 00Z re-forecasts and the same thing from the 12Z re-forecasts. This means that you're missing the accumulated precipitation from hours 3 to 12 of the re-forecasts, so your annual mean estimates might only be 1/4 of any ballpark values you're expecting. I suspect that you need to choose step=12 rather than step=3, as described in the ERA FAQ.

Also note, I think the comment in that FAQ about the monthly mean from daily mean data availability being "planned" but "not yet implemented" is out of date and that may be the better route to calculate the annual means (see also section 3 of Berrisford et al, 2011).