Hello,
I would like to know what values P-E land is initialized with. The fields in question are PFWRO1 + PFWROD for surface layer runoff and runoff at depth. It seems to me that the initial values are too large. The values drop by a factor 10 within the first 24 hours of the simulation and never reach initial values again. They stabilize only after several weeks. I first encountered this behavior in the OIFS-FESOM2 coupled model where the high initial runoff values crashed the ocean model. I have since been able to reproduce it in standalone OIFS runs. The plot below shows P-E land for 4 random points in the Amazon (red), Kongo (white), Mackenzie(blue) and Ob (green) river basins. Its the same story everywhere. The global sum of river runoff is initially ~ 20 times higher than in equilibrium.
P.s. I had a look at what IFS cy 36r4 sends to NEMO in EC-Earth it does not have unusually high initial values for PFWRO1 + PFWROD.
Best regards,
Jan
12 Comments
Unknown User (nagc)
Hi Jan, can you remind me what experiment id this is from? I can check the source of the initial data.
Glenn
Unknown User (jstreffi)
Hi Glenn,
this is the experiment from the EC-Earth OIFS branch. The experiment id is ECE3. The data can be found here:
I don't know what the original experiment id was. I assume Uwe changed it to ECE3.
Cheers
Jan
Unknown User (nagc)
Ah ok, I thought this was from one of the experiments I gave you. This might be because ERA-Interim was used to create the initial conditions.
Land surface scheme changes between ERA-Interim and OpenIFS
OpenIFS is based on a newer version of IFS than used for ERA-Interim (CY31R2) and this introduces a discrepancy in the land-surface modelling.
ERA-Interim used the TESSEL land-surface model. This was upgraded to H-TESSEL in later IFS cycles where a new soil hydrology scheme was introduced. This created a difference in the soil moisture dynamical range and a rescaling is required to use ERA-Interim for H-TESSEL in OpenIFS. An approach in which the evaporative-fraction is conserved was adopted. The changes to the soil hydrology scheme and more details of the scaling are described here: Change to soil hydrology scheme (PDF). This describes the rescaling using MARS client but the formulae could be written in say python using the python interface to eccodes or fortran and eccodes.
ERA-Interim/Land
Limitations in the rescaling approach led to the development of an offline land-surface reanalysis used to produce an ERA-Interim/land surface update. This improved the compatibility of current land surface scheme with past reanalysis. The updated land state should be used in conjunction with ERA-Interim for the atmosphere with OpenIFS.
The ERA-Interim/Land reanalysis is further described in this free-access journal paper by Balsalmo et al (2015).
Unfortunately ERA-Interim/Land is only available between 1979 and 2010. For dates beyond 2010 the rescaling approach still has to be used.
Rescaling the evaporative index is also used when interpolating across resolutions, e.g. T255 to higher initial resolutions for OpenIFS. Spatially interpolating surface properties and prognostic fields is a complex problem as the surface is inherently heterogeneous.
It is important to be aware of this if the focus of study is the surface. The OpenIFS team can provide initial conditions based on operational analysis which will be more appropriate.
I think this might be the issue but if not let me know.
Glenn
Unknown User (jstreffi)
Hello Glenn,
that looks like a good candidate for the reason for this issue. Luckily it doesn't affect me greatly since I want to do long runs. I turn runoff off in the ocean for the first 12 hours and after a few weeks the values return to normal. Impact on ocean salinity and sea surface hight should be minuscule. As I said I also found this in the standalone OIFS runs. I do those at T95L91 resolution for some quick budget runs. I looked up the initial files and they have exp id gvfc. I'm not sure if the initial files you gave me for T255L91 (exp_id: gvfb) are better. Since they are 2016 I guess they have the same problem? If not I could check see what happens when I replace the ones from Uwe with those.
Cheer Jan
Unknown User (nagc)
I checked and gvfb and gvfc were created started from ERA-Interim for atmosphere and ERA-Interim/Land for surface fields. I would hope the behaviour is better with these files.
But the initial dates for those experiments are 19900101?
We always use the rescaling in creating the files and Interim/Land if available on the required dates.
Regards, Glenn
Unknown User (jstreffi)
Ah yes, they are 1990. So rescaling is available and you used it. Does that mean there was a problem with the rescaling method here? Here is the global mean of runoff in the first three days in a standalone run using gvfc intial data.
Cheers Jan
Unknown User (nagc)
Hi Jan,
I didn't explain very well. For the gvfc experiment, because ERA-Interim/Land was available for the start date, I took the surface fields from that reanalysis and rescaling the fields was not necessary. The rescaling approach is only used for dates when interim/Land is not available or on request.
I will ask someone in the surface team about this and get back to you.
Glenn
Unknown User (nagc)
Hi Jan,
There is a mention about river runoff in the publication comparing ERA-Interim and ERA-Interim/Land:
https://www.ecmwf.int/sites/default/files/elibrary/2012/7922-era-interimland-global-land-surface-reanalysis-based-era-interim-meteorological-forcing.pdf
look at page 14 and fig 11.
Cheers, Glenn
Unknown User (jstreffi)
I see so there is the source of our initial value discrepancy. Good to have a confirmation on that.
Cheers, Jan
Unknown User (nagc)
Jan,
I have found some more information which I think applies in your case. This is taken from an old ecmwf web page that is no longer available on the new ECMWF website (I kept a copy). Even with the scaling of the soil moisture there can still be significant differences. We would not expect to see this with a more recent reanalysis like ERA-5.
The implication of the change in the soil hydrology is the that effective soil moisture range varies geographically as the permanent wilting point (PWP) and field capacity (FC) vary. For instance, sandy soils will be dryer while the fine texture soils (e.g. clay) will have a higher soil moisture than before, in line with observations.
This model change may have consequences for applications where soil moisture (SM) is used as an initial condition for other models or for applications where soil moisture is used as a dryness indicator. For those applications it might be necessary to scale to the so−called soil moisture index (SMI) defined as (SM−PWP)/(FC−PWP). This index is equal to zero at the permanent wilting point and one at field capacity but it can be larger than one after rain. It is also an indicator for the "wetness" of the soil and can be used as an intermediate step to scale soil moisture from one model to another in cases where these models use different soil characteristics.
Note that, even after scaling to the soil moisture index, the soil moisture fields from IFS cycle 32r3 can differ substantially from that given by the previous cycle. This is due to the major changes in the land surface scheme (revised runoff, new hydrology and new soil types).
(Acknowledgement: Gianpaolo Balsamo)
Regards, Glenn
Unknown User (jstreffi)
Hello Glenn,
thanks for digging deep and finding the reason for the behavior I am seeing. Just out of interest: Are there any plans to switch OIFS inital value generation over to ERA5?
Cheers,
Jan
Unknown User (nagc)
Hi Jan,
We can create initial conditions from either ERA-Interim, ERA-5 (if the dates are available) or operational data. I normally ask what's preferred. For your experiments I think ERA-5 wasn't available at the time.
Glenn