Continuing with Ole's description, maybe it is worth reminding a potential trap in the HARMONIE script for users at operational services:
if the vertical levelling in your model domain is different from "65" or "MF_60", you need to find adequate numbers for the following ZNEN1 and ZNEN2 as specified in Forecast/Dfi scripts:

=============================================

# Domain dependent settings for transition zones in AROME spectral nudging
# horizontally it is approximately 50 and 100 km in wavelength
# vertically a height correspnding to around 50 and 100 hpa

gs=$( echo $GSIZE | cut -f 1 -d . )
ZNEK0=$( expr \( $NLON \* $gs \) / 100000 || echo "" )
ZNEK1=$( expr \( $NLON \* $gs \) / 50000 || echo "" )

if [ "$VLEV" = MF_60 ]; then
ZNEN1=6
ZNEN2=9
else
ZNEN1=3
ZNEN2=6
fi

Regarding LSMIXBC, I also think, out of experiences with similar schemes used in HIRLAM, it is better not to activate LSMIXBC in most of the DA researches, especially in impact studies about observations. Further, it is probably worth some study with a much increased cut-off distance in LSMIXBC to establish that one achieves equally good, if not better, results.

These said, I think from the MetCoop study quoted by Ole in his post, and the recent validation about trunk/38h1.beta3 with Danish DKA (800x600) domain and the reference DKCOEXP domain (600x600), there still seems to be evidence that LSMIXBC improves to some extent the MSLP score and upper air parameters such as T, H and W. One may argue though that the benefit here in HARMONIE may not be as significant as what we used to see in HIRLAM system, and since the scheme does not appear to benifit much about surface parameters, its usefulness may not be so clear. On the other hand, nor do I see the option damaging.

I have looked at the verification scores presented by Xiaohua for three
periods with and without LSMIX in HARMONIE. From this I can only conclude
that LSMIX seems to help a little bit to repair problems that HARMONIE
has with vertical mixing in physics (yes, data assimilation experts use to blame physics because it is very difficult to do data assimilation with biased models - experience from HIRLAM). In all three cases HARMONIE is too dry and too warm at 850 hPa, too cold at 925 hPa, too cold compared to SYNOP stations over land and consistently there is a positive surface pressure bias in the SYNOP stations over land.
So in my view the problem is not related to small scale features (let us
say weather disturbances) that the HARMONIE model cannot handle in the data
assimilation, but related to systematic errors in the physics that make data assimilation difficult or impossible.
Somebody must look at the "Whole NWP problem with HARMONIE"; Maybe simply
there are some default coefficient values that are set wrong - This happended quite often in HIRLAM, particularly in the beginning.
The are the views of a brave senior mathematician

Alena Trojakova replied on the 24th of October about ALADIN experience

Ensemble based technique is mostly used to generate background error
statistics and all the resolved scales are influenced by analysis.
Regarding tuning only REDNMC is considered with Desroziers et al (2005)
method used for its estimation and the most recently we focus on
grid-point sigma_b maps. The idea of a scale-selective background errors
is quite interesting, but to be honest I have no idea how to do it.

Regarding technique to account for host model information the Digital
Filter (DF) blending technique (Brozkova et al 2001) is used currently
in Czech Republic and Slovakia, tuning is detailed in the documentation
by Dominique Giard
www.cnrm.meteo.fr/gmapdoc//spip.php?article97&lang=en. I found
your studies on spectra presented on HIRLAM Forum quite interesting and
I'll try to contribute with our results with DF blending.

In longer term I would like to revitalize work on Jk within ALADIN and
any experience is welcome.

Additional comments from Mariska Derkova on the 24th of October

Just a complement to Alena`s ınformatıon on blending by DFI technıque:
- the technıque is used ın LAEF as well, wıth dıfferent tunıng of DFI and blendıng truncatıon parameters (ECMWF EPS analysıs/forecast resolutıon). More detaıls could be found ın LAEF artıcles/reports (Yong Wang or Martın Bellus).
- we have seen ın standard VARBLEND/BLENDVAR/BLENDING experıments that the dıgıtal fılter ınıtıalısatıon applıed on analyses reduces the small scale features brought from the fırst guess. So there ıs no ınıtıalızatıon used ın assımılatıon, and ıncremental dıgıtal fılter used ın productıon (or no ınıtıalızatıon at all) wıth DF blendıng.
Several technıcal reports exısts on thıs topıc as well (Prague, Budapest?)

Contribution from Mate Mille on the 25th of October

For the time being I can give similar answers for your questions as Alena already did. Therefore this is tipical of LACE DA practice to apply background error statistics which describe all range of the resolved scales. Additionally to tuning, the investigation of spatially varying sigma B maps are restarted this year in several LACE countries in order to bring flow-dependency approach to ALADIN and AROME DA. From the first study (Strajnar 2008 MF report) it was found that the spatial variability of the AEARP sigma B maps are not enough over Central Europe, therefore we would like to focus on the sigma B maps from LAM ensemble system for next.

On the blending side, Czech and Slovakian Colleagues are the most experienced on this and they have it operationally.

I tend to agree with Nils analysis that is supported with new results, summarized in the attached powerpoint.

I considered a number of alternative ways of mixing, including vertical mixing only and extreme mixing by using ECMWF as first-guess rather than some mixture with Harmonie.

The main conclusions are
1. Whatever mixing with ECMWF helps to improve Harmonie forecasts
2. The cold start experiment (using ECMWF as first-guess) gives similar results as the other mixing options

Shocking I would say since it means that either the information content from previous observations is not captured by the system or the (non-verifying) small-scale structures added by Harmonie overwhelm the analysis increment already after 3 hours. In both cases DA will not help to improve Harmonie forecasts.

Why do the small-scale Harmonie structures not verify? Are Nils assumptions correct?

I think the uplaod of the powerpoint ent wrong. Another try:

Hi all,

no scientific input here but if you wish to receive e-mail notifications of new postings on this thread click on "SUBSCRIBE" at the top.

Eoin

Hello,

Using an ECMWF forecast as FG is an issue that has arisen in the course of this discussion. It seems to me that it makes sense. I have been doing this in assimilation experiments with radar data and the results are encouraging. This however can at the moment only work for wind observations because there are no hydrometeors in the ECMWF files interpolated to HARMONIE grids, and therefore no reflectivity guess. I want to test with a diabatic initialization to produce these hydrometeors. But here comes the problem. The following error is raised:

YQ_NL%LGP true only works with LSPRT true

changing LSPRT to true (default is false) seems to work during the DFI process but it fails when fields are to be dumped to a file becasue SC2RDG detects an empty buffer.

After this finding, I have tried with a standard assimilation (from a HARMONIE first guess) but it gives back the same error.

It seems that this is a mere technical issue related to the representation of some fields in GP or SP forms.

I wonder why it is not possible to run a DFI initialization with HARMONIE (arome+surfex) at this moment! Can this apparently minor problem be overcome in a simple way? (I use 37h1.2)

Here are some trials to start to understand how reasonable the background error statistics at short time and space scales used in HARMONIE data assimilation are. The input comes from Nils and Ljubljana University. Kinetic energy spectra are often used to study the behaviour of a model. However, in a regional domain kinetic energy spectra are associated with aliasing effects due to the bi-periodization (through the use of an extension zone for example). Therefore so called “structure functions” in gridpoint space are instead used. For the U-wind component we may for example look at

Mean value over x and y of ( u(x+dx) – u(x) )**2 as a function of grid separation dx.

Such structure functions have the same characteristics as kinetic energy spectra. In order not to confuse you by mixing up with assimilation structure functions (seen by single observation experiments for example), we will simply refer to kinetic energy spectra in the following.



On all plots you can see the theoretical kinetic energy spectra (red solid curve), the observed kinetic energy spectra (green curve, based on the aircraft measurements) and the kinetic energy spectra calculated from the HARMONIE forecasts for different forecast lengths. One can see a clear spin-up of the kinetic energy spectra. HARMONIE Model fields are obtained through downscaling of the global ensemble fields from the ECMWF (non-cycled mode). This information is used to calculate structure functions for full model fields as well as for ensemble perturbations.
Notice that most of the plots contain the spectra of the kinetic energy for the field itself (mean value + perturbation). Mean value is what we are able predict. Perturbations are the differences between the ensemble members (a proxy of uncertainty/forecast error).

At the larger scales the kinetic energy spectra is consistent with the theoretical estimate, at smaller scales the lack energy is obvious for small scales at the initial times. The energy at small scales is quickly generated most due to adjustment processes going from coarse resolution to high resolution. At 300hPa the spin-up of energy goes faster during winter (strfun_winter) than during summer (strfun_summer) (most likely due to turbulence associated with stronger jet-streams). In any case it seems that it takes more than 3 hours to spin-up the structures, up to 8-10 hours in summer case. This might have a serious consequences for the data assimilation. These "adjustment" or “spin-up” structures are probably not so suitable to constrain the analysis. If so, this might be one of the reasons why the HARMONIE data assimilation is not able to improve the HARMONIE forecast. It might be better to use a more mature structures if they are based on the dowscaled perturbations. The situation can be quite different if one uses perturbations from cycled experiments (type EDA/ETKF) for which the the model and the data assimilation have a real chanse to insert mesoscale structures.

On the other hand, one should should also remember that the spectra for vorticity and divergence that we often look at may have a lot of variance at the smallest scales, but since vorticity and divergence are spatial derivatives of the wind components, one needs to normalize with 1/(k**2+l**2) in order to see the effects on physical variables such as kinetic energy. Doing such a re-normalization, the spectra of vorticity and divergence will look more like spectra of temerature and humidity, with maximum in the energy spectra around 500-1000 km scales.

It is interesting to notice that in the boundary layer (900hPa) the spin-up of spectra goes quite quickly during winter (strfun_winter_900) (vertical stability?) and much slower during summer (strfun_summer_900) (more convection turbulence/more mixing?).

In summary, the forecast error kinetic spectra have the most energy on quite large (strfun_summer_diff) (synoptic) scales and that is what we have seen from Gert-Jan single observation experiments and is in line with the message Åke tries to bring up.

It'll be interesting if a similar study can be made using ensemble HARMONIE forecasts through purturbation of observations, assuming one can create sufficiently large spread.

Dear Jelena,

Coming back to one of your earlier posts, further information for scatterometer NWP data assimilation is provided for example in research.metoffice.gov.uk/research/inter...similation_workshop/

Please do not treat all scatterometer data with the same characteristic length scale. We have analysed spatial error properties of several scatterometer instruments in detail, and generally conclude that error correlation depends on satellite and product. Please contact me or the NWP SAF Helpdesk ( This email address is being protected from spambots. You need JavaScript enabled to view it. ) for further guidance.

Cheers,

Ad

Dear Jelena,

From your report:

"In any case it seems that it takes more than 3 hours to spin-up the structures, up to 8-10 hours in summer case. This might have a serious consequences for the data assimilation. These "adjustment" or “spin-up” structures are probably not so suitable to constrain the analysis. If so, this might be one of the reasons why the HARMONIE data assimilation is not able to improve the HARMONIE forecast. It might be better to use a more mature structures if they are based on the dowscaled perturbations"

Are you suggesting that it may perhaps not so fruitful to base the climatological B on too short forecast ranges (1 hour), but instead to use an overall parameter (like REDNMC) to modify the B-amplitude (based on 6h forecast differences) for each analysis. I tend to agree with this. From what I have seen thus far there are quite some similarities between B's based on differences say 3h and 6h forecasts. Please let me know if you have seen other results.
Adding some form of flow dependency to B will really improve the DA performace,

Best regards,
Jan

Happy to knowing more about Harmonie experiments, with different observation settings. Really good analysis you done on it and providing here its result. By profession I am an assignment writer so I don’t have more knowledge about it but I love music so like to read such kind of interesting issues that can explain the so far results from Harmonie. I am also enjoying and listening music at my workplace Quality Assignment UK where we deal with those students who are facing so many problems with their writing assignment and getting an excellent & professional academic service from educational experts.