Doubts about the ERA-Interim

[backkom]

Hi all,

I have some doubts about ERA-Interim datasets after going through some posts in this forum, and wish someone can give me some guidance and comments.

The ERA-Interim data used comes from http://rda.ucar.edu/datasets/ds627.2/, 6 hourly, thus for the forecast variables, such as precipitation and heat flux, data is provided on hour 06, 12, 18, 24.

From the official website of ERA-Interim, http://www.ecmwf.int/products/data/arch ... erminology, it can be found:

14. Many fields have seconds in their units e.g. precipitation and radiation fields. How can instantaneous values be calculated?

Some forecast fields (fluxes) are accumulated from the start of the forecast, so to get averaged values you need to divide by the length of the forecast (the forecast step). In this way the average between the start of the forecast and the forecast step is obtained. To obtain the average between two forecast steps (e.g. STEP1 and STEP2) it is necessary to retrieve the fields for the two steps
( FLD ( STEP1) and FLD( STEP2))
then calculate the difference and divide by the time difference in seconds:
( FLD( STEP2)- FLD( STEP1))/(( STEP2-STEP1)*3600).

34. What are the 'steps' for the surface daily fields in ERA-Interim?

The ERA-Interim data server surface archive has a mixture of analysis fields, forecast fields and fields available from both the analysis and forecast. The other daily archives have only analysis data.

If step 0 is chosen, then only analysed fields, which are produced for 0000, 0600, 1200 and 1800 UTC, are available.

If step 3, 6, 9 or 12 is selected then only forecast fields which are produced from forecasts beginning at 0000 and 1200 UTC, are available.

Thus, it can be concluded that:

(1) the forecast varialbles at 06, 12 hr are intialized and integrated from the reanalysis data at 00 hr, and similarly those at 18, 24hr are from reanalysis at 12 hr.

(2) the accumulated variables, e.g. precipitation and radiation fields, representes the time integral from the start of the forecast, so e.g. for solar surface radiation(ssrd), ssrd at 06hr(ssrd06 hereafter) reprensents its temporal sum from 00hr ~ 06hr, ssrd at 12hr(ssrd12 hereafter) for the period 00hr ~12hr, ssrd at 18hr(ssrd18 hereafter) for period 12hr ~ 18hr, and ssrd at 24hr(ssrd24 hereafter) for period 12hr ~ 24hr.

If that is correct, the temoral average ssrd, with its ROMS units W/m*m, should be caculated as ssrd06/(3600*6hr) with its central temporal point at 03hr, and (ssrd12-ssrd06)/(3600*6hr) at 09hr, ssrd18/(3600*6hr) at 15hr and (ssrd24-ssrd18)/(3600*6hr) at 21hr.

Instead of the simpily divided by 3hr or 6hr, just like the folowing posts:
viewtopic.php?f=23&t=2293&hilit=era+interim

W/m^2/s --> 1/(60*60*dt) W/m^2

http://stommel.tamu.edu/~baum/ecmwf.html

2. The "Surface solar radiation" in "W m**-2 s" needs to be converted into a "shortwave radiation flux" in "W m**-2".

shortwave radiation flux (W/m^2) = surface solar radiation (W/m^2/s) * 3600*3 s

viewtopic.php?f=14&t=2948&hilit=era+interim

swrad (W m-2) ssrd / (3*3600); 3-hour step
lwrad_down (W m-2) strd / (3*3600); 3-hour step
latent (W m-2) slhf / (3*3600); 3-hour step
sensible (W m-2) sshf / (3*3600): 3-hour step
rain (kg m-2 s-1) tp * Rho_w / (3*3600)
evaporation (kg m-2 s-1) e * Rho_w / (3*3600)

Best,

[lolhsson]

...good eye! Thanks! I think you're right; when I look at the forcing files generated with the conventional wisdom of these forums, my swrad (generated from ERA's ssr) and swflux (generated from ERA's e and tp) are clearly cumulatively additive through each set of four (00,03,06,09;12,15,18,21) 3-hour timesteps. No wonder my ocean was still getting too hot. (My shflux seems fine, because I used the ERA's instantaneous sensible heat flux, ishf, and although it is a forecast term, it seems to be calculated non-cumulatively by the ERA folks. Which makes sense; cumulative instantaneous flux would be particularly bizarre.)

I think your estimates for flux average-centered at 03, 09, 15 and 21 make sense, and I'm writing some scripts to convert my forcing files accordingly. I'm not positive if we can get away with using 00 and 12 directly; they are ... instantaneous? perhaps integrated only over an hour? I feel like following your logic, ssrd00 = temporal sum from 00hr ~ 01 hr(?), so in W/m^2 that's just a simple ssrd00/3600, and the same for ssrd12, but that's pure intuition, nothing more.

[backkom]

Hi lolhsson,

It is really glad to get some comments and reply~

I think your estimates for flux average-centered at 03, 09, 15 and 21 make sense, and I'm writing some scripts to convert my forcing files accordingly. I'm not positive if we can get away with using 00 and 12 directly; they are ... instantaneous? perhaps integrated only over an hour? I feel like following your logic, ssrd00 = temporal sum from 00hr ~ 01 hr(?), so in W/m^2 that's just a simple ssrd00/3600, and the same for ssrd12, but that's pure intuition, nothing more.

I downloaded my 6hr ERA-Interim from http://rda.ucar.edu/datasets/ds627.2/, and for my case, there is no "instantaneous" ssrd00, instead the ssrd at 00hour comes from a 12-hr integration staring from the noon(12hr) in the previous day.

Besides, I am not sure if you calculated the relative humidity according to Hernan's equation in this post: viewtopic.php?f=14&t=2948, it seems that those equation is kind of different with equation 7.4 and 7.5 in http://www.ecmwf.int/research/ifsdocs/C ... SPart4.pdf.

Best,

[lolhsson]

I thought about this problem a little bit more thoroughly, and it occurs to me that what is essentially a discontinuity every four steps (at least in my fields; it's very clear to me that 03-12 are cumulatively additive, as are 15-24) introduces a major problem when you try to do simple centered averages such as, say, 18-12/6; unfortunately, my estimates at 09 and 21 get rather screwed up from this, with (expected) unreasonable values of net radiation flux. That being said, I'm using ssr, not ssrd; I haven't examined the ssrd fields in detail, if they're calculated differently...

A simpler and possibly safer compromise would be to only use the 4th and 8th steps (12 and 24) integrated over 12-hour periods, but that isn't particularly good temporal resolution...

I don't use bulk fluxes (they've caused some major problems for me, and in fact getting away from them is what led me to ERA), so relative humidity isn't something I work with directly from ERA. The evap/precip fluxes, though, have the same problem as radiation.

[backkom]

I just checked the official website of ERA-Interim, and found that only four step (03, 06,09,12) are available with for the forecast varaibles such as precipitation and radiation flux, as listed in FAQ:

34. What are the 'steps' for the surface daily fields in ERA-Interim?

The ERA-Interim data server surface archive has a mixture of analysis fields, forecast fields and fields available from both the analysis and forecast. The other daily archives have only analysis data.

If step 0 is chosen, then only analysed fields, which are produced for 0000, 0600, 1200 and 1800 UTC, are available.

If step 3, 6, 9 or 12 is selected then only forecast fields which are produced from forecasts beginning at 0000 and 1200 UTC, are available.

So from my understanding, there should be no such istantaneous staus for the 00hr or 12hr , instead another four step 03,06,09,12 will be provided as the accumulated variables starting either from 00hr or 12hr based on the reanalysis results. It can also be confirmed in the pds (Product Definition Section, http://rda.ucar.edu/#gribdoc) when reading WMO GRIB file, which indicates the P1 varing from 03,06,09,12 for each reanalysis time (only 00hr or 12hr for each day).

when you try to do simple centered averages such as, say, 18-12/6;

I do think you can just implement the centered averages like (ssr18 - ssr12)/6, which make no sense. Since ssr18 is a 6-hour accumulated surface solar radiation starting from 12hr, while ssr12 a 12-hour accumulated one from 00hr.