Baltic Sea modelling
Johan Söderkvist
Dear all
I have problems simulating deep water salinity in the Baltic Proper using GETM.
The main issue I want to address is that there is a build-up of a bottom layer with constant salinity.
Observations show that there is a slow decay of about 0.5 PSU during this 3 year period, which is not happening in the GETM setup. The salinity reaches a value of about 12.8 PSU for the lower 125 meters.
To illustrated this, profiles from Gotland Deep after about one month and 2 years of model simulation is attached.
There are some difference in setup between the experiments v015Ebw and v015Eed, mainly initial field. That is why v015Ebw is also show, even though v015Eed seems to be better
#
Also attached, HOTSTART-input from that particular position, from two epochs, one-month simulation and about 3 years of simulation
There seems to be no change in turbulence parameters eps and tke with time (20200116 vs. 20211201)
#
Furthermore, there is a tendency to have a weak halocline. Some input about this issue will also be appreciated.
I would really appreciate feedback!
Before, going into details:
iw_model = 2 (Large et al) have no effect on model results.
I conducted to experiments with iw_model=2, which gave exactly the same result at iw_model=0
Long story:
There was an upstart meeting with IOW-researchers and Germo Väli, where some serious problems with the setup at our office was discussed.
Thank you very much for the input!
Many of these issues have been resolved, such as:
a) Shift from General Vertical Coordinates to Adaptive Vertical Coordinate
b) Too low salinity in the deep channel of Great Belt, Arkona Basin and Bornholm Basin
Much of this was resolved deepening the channel in Great Belt and with improved filtering of the bathymetry.
Current setup uses Adaptive Vertical Coordinate system with 60 vertical layers. The horizontal resolution is 1 nm.
After more than 60 1-year experiments and some 3 -year runs also , with adjusting bathymetry, Smagorisnky and z0 2D-fields, tuning of parameters in gotm.nml, getm.inp and AVC-input file I am soon running out of options.
Furthermore, I have also changed initial field, with too low salinity in the northern and north-western part of the Baltic Proper, to “let” GETM fill up the deeper parts (experiment v015Eed in attached figures).
Still, salinity below 125 meters seems to be decoupled from the upper layers. Inflowing waters is not able to penetrate to deeper layers.
Some parameter values in best-experiment-sofar:
Molecular diffusivity of heat hand salt: avmol = 5.0e-07
advection schemes for v, w, S, and T : 2/4/4 (=half splitting with TVD-Superbee (second-order, monotone))
eps_min= 5.e-13
k_min= 1.e-9
smag_const in Baltic = 0.12
ri_st= 0.25
z0 in Baltic = 0.01
iw_model = 1; alpha=0.5; ri_cr=0.6; numshear=5e-6; numiw = 1e-4; nuhiw=1e-5
best regards
JOHAN sÖDERQVIST
Oceanographer, PhD
JOINT GEOMETOC SUPPORT CENTRE
DANISH MINISTRY OF DEFENCE ACQUISITION AND LOGISTICS ORGANISATION
Lautrupbjerg 1-5, DK-2750 Ballerup
Denmark
Phone: +45 4078 7321
E-mail: j...@fcoo.dk
This email was scanned by Bitdefender
Hans Burchard
Am 22.08.2025 um 22:11 schrieb 'Johan Söderkvist' via GETM-users <getm-...@googlegroups.com>:
--
---
You received this message because you are subscribed to the Google Groups "GETM-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to getm-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/getm-users/82143472979240dd9bc8f93a4dca6c3a%40fcoo.dk.
<by15_v015Eed_2021120100.in><by15_v015Eed_20200116.in><NS1C_BY15_202111_v015Ebw_v015Eed.png><NS1C_BY15_202001_v015Ebw_v015Eed.png>
Knut
Dear Johan,
thanks a lot for sharing.
@Hans: Last night I was wondering about the same, but the layer heights in the restart files that Johan attached look reasonable(?)
The question is whether the conditions at BY15 just show the symptom and not the cause. Maybe you miss boundary mixing elsewhere? Did you already tried AH_min? Could you also share getm.inp?
Cheers, Knut
To view this discussion visit https://groups.google.com/d/msgid/getm-users/4D9573E6-DEFA-4468-9E7C-07E5B334E568%40io-warnemuende.de.
Ilja Maljutenko
--
Johan Söderkvist
Dear all
Thank you very much for all feedback!
Attached requested image and getm.inp file, also the adaptive coordinate input file.
I am also trying with decreased deep water salinity in the initial field, with largest changes in north and north-western Baltic Proper.
Earlier experiments had too high bottom layer salinity, which could diminish the estuarine circulation with high-saline northward flow.
Not sure though how many years is required for the build-up to proper stratification, hopefully less than 50 years!
I would appreciate some insights on how you generate the initial field and how to spin-up the model.
I have tried
TVD-P2-PDM (third-order, monotone) advection scheme, but that also create a bottom layer with constant salinity.
However in this older test the initial condition in temperature have an increase from bottom towards 125 meter that might prevent high saline inflow to the bottom, if that would have been the case.
I will try MUSCL or P2PDM, thank you Ilja.
The problem with bottom layer with constant salinity appears at all monitoring stations that are deeper than 125 meters.
The locations of the SMHI monitoring stations in the Baltic Proper are separated by sills of about 120 – 130 meters.
So, I guess that it is not only the pathway from Bornholm Basin to Gotland deep basin that is the issue, that is: a local bathymetry problem.
BY15 is just one example of the model response in the experiments. I agree with Knud that it might not be at that specific location where the problem is.
The problem might be due to missing dynamics below sills, less an advection problem between basins.
What is really bothering me is that there is no slow decay in bottom layer salinity, of about 0.5 PSU over ~3 years, as shown in observations.
I have tested different values of molecular diffusivity and Galpering length scale, but without success.
Internal wave model Large et al. 1994 have been tested but gives identical results as internal wave model inactivated.
Current testing is limited to test parameter settings for the Mellor 1989 model.
This is a delicate process though, to increase deep-water mixing on longer time scales, and to keep the shape halocline layer.
There has been some testing with different AVC-settings, and will continue to do so. I agree with Knut that there seems to be a reasonable
distribution of the vertical layers at BY15 (also on other locations in the Baltic Proper).
Present AVC-configuration is attached (experiment v015Eed in attached image).
Have I misunderstood how to set hmidd? Should it be a negative or positive value? The idea with hmidd = -60 is to focus on depth layer around 60 meters.
No breakthrough yet.
Next planned test is to change to (feel free to comment on this):
chbott=0.1
d_dens = 0,4
#
Previous config used (experiment v015Ebw in attached image, does give similar results):
chbott = 0.5,
chmidd = 0.2,
hmidd = -4,
d_dens = 0.3
The distribution of the vertical layers between the previous (v015Ebw) and present setup (v015Eed) is very similar.
To view this discussion visit https://groups.google.com/d/msgid/getm-users/CALriNdPVMAoNDk%3DisePTFGbTL1AtH_kaZuU%2BUUPjR4_AktFE6Q%40mail.gmail.com.
germo...@gmail.com
smag_const in Baltic = 0.12
Karsten Bolding
--
---
You received this message because you are subscribed to the Google Groups "GETM-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to getm-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/getm-users/82143472979240dd9bc8f93a4dca6c3a%40fcoo.dk.
Johan Söderkvist
Dear Karsten
Using configuration from Ulf was not good enough here. We (me and Bjarne) was very close to that configuration some years ago.
The shape of the halocline was well simulated for about nine months, but then the halocline became more linear, with lacking salt below 60 meters. Main progress during present project are increasing depth in Great Belt and filtering method for the bathymetry. So, since the previous setup at our office was to diffusive, it is too early to conclude what changes we will end up with. Maybe more work is needed on the bathymetry. But is seems that advection scheme (TVD superbee) and Prandtl number needs to be changed. I have tested several smagorinsky-values but do not where I will end up with yet. I believe that our set up will not have Langmuir circulation but internal wave model activated, and ip_ramp = 30 (split factor 15 and dt=6)
best regards
JOHAN sÖDERQVIST
Oceanographer, PhD
JOINT GEOMETOC SUPPORT CENTRE
DANISH MINISTRY OF DEFENCE ACQUISITION AND LOGISTICS ORGANISATION
Lautrupbjerg 1-5, DK-2750 Ballerup
Denmark
Phone: +45 4078 7321
E-mail: j...@fcoo.dk
From: getm-...@googlegroups.com <getm-...@googlegroups.com> On Behalf Of Karsten Bolding
Sent: Wednesday, 27 August 2025 08.46
To: getm-...@googlegroups.com
To view this discussion visit https://groups.google.com/d/msgid/getm-users/CAFvy56Wu%3DRWmG%3DgP9OkKEYS97SbEKMw6ceb1_mGz9BW%3DsB%2BZ9Q%40mail.gmail.com.
Karsten Bolding
To view this discussion visit https://groups.google.com/d/msgid/getm-users/d3db684e31d24afa9e7fab323d68ea41%40fcoo.dk.
Johan Söderkvist
Hello again!
First things first.
Karsten: Bottom salinity is not that bad in our setup, but we have not simulated salinity during a Major Baltic Inflow event, though.
To remind you, current setup tends to get too linear halocline and a bottom layer with constant salinity, see images and BY15_20211013.png (=Gotland deep), and further south where to problem is worse (BY10_20211014.png).
Furthermore, the bottom salinity does not decay with time during stagnant periods. In fact, there is no stagnant period in this setup.
I hope that the analyse here can give you some ideas of what I am struggling with, and hopefully some suggestions of way ahead.
In our GETM 1nm setup there are oscillations in the velocity field that has a period of 14 hours, very much similar as reported by van der Lee and Umlauf (https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2011JC007072), thus near inertial waves that is initiated by inertial currents.
However, I suspect that the oscillations in our GETM setup are too frequently occurring, currents are too strong (up to 0.3m/s), the decay rate is too weak, and are too frequent occurring.
Going through the literature, it seems like currents in Baltic Proper at about 80 meter rarely is above 10 cm/s. Currents of about 2-5 cm/s are much more common below mixed layer.
I try to illustrate this with nearly two years of model data from Gotland Deep (BY15) at about 150 meters depth, see figure BY15_v015Eeq_Lev30.png.
Top panel shows original, and low pass filtered northward velocity component, together with triangles that indicates max and min values for each 14-hour period. (Unfiltered data will be discussed below)
Middle panel show the absolute value of the max and min values (the triangles in top panel) for the full time series.
Bottom panel show the number of events when the northward velocity component is within a certain interval (cm/s)
It should be noted that the northward velocity component at 150 meters depth is larger than 10 cm/s for about 25% of the total time period, which seems much too high.
This phenomena seems to be restricted to the Baltic Proper. Modelled near Inertial waves is also present in Bothnian Bay, but much weaker currents, and not always present.
#
The velocity field is low pass filtered because there is also a high frequent variability in the velocity field during summer), see BY15_v015Eeq_highfreq.png.
The frequency of this variability is higher than the model output (=10 minutes).
The amplitude of the high frequent variability can be as high as 0.05 m/s.
The high frequent variability, decay and increase several times during the summer season.
#
The strong oscillation of with 14 hour period and high frequent variability might explain the near linear-type halocline.
I also suspect that the velocity field is the reason for bottom layer with constant salinity.
I have tried to get some idea of how much mixing is generated in this GETM setup, using computed TKE (based on uu, vv, and w) and local Richardson gradient number (Ri_grad).
The u_prime in the TKE calculation here, is computed from deviation from the 2-day mean value.
The model has been run for one year, and the analyse is based on the following 10 months
The first three panels in attached figure mixing_contours.png show time, depth of Richardson gradient number, TKE and Mixing classification.
The Richardson gradient number shows that mixing most likely occur in surface and bottom layer. Black line is isoline for the Ri_critical values to 0.25 (as in current GETM setup).
Second panel (TKE) show that turbulence is very high below 100m, but surface layer has TKE values above 1e-5 on several occasions.
Third panel illustrates where and when mixing occur (see below for explanation of the regimes).
This panel show that there are long time periods with strong mixing through the water column.
This is also illustrated in the bottom right panel, here regimes with Potential mixing is excluded.
Changing parameters in the GETM setup has not yet resulted in weaker velocity field in the Baltic Proper.
For example:
Increasing z0 to 0.03, activating internal wave breaking (with high alpha values 0.8) or increasing k_min to 5e-9 (from 1e-9) are not enough to dampen the internal waves.
Current setup use TVD-Superbee advection scheme. Have tried several others but these are too diffusive. Furthermore, the upstream (first-order, monotone) also generate high frequent variability.
During the start-up phase: The model is run for 5 days with frozen coefficients, one month in with full 3D, time step = 4 and 3D time step = 40s.
The strong periodic currents is generated during this start-up phase. For the rest of the simulation, the 2D time step is 6 s, 3D time step 180 s. Horizontal resolution is 1 nm.
Thus, the CFL criteria is not exceeded.
Bathymetry is filtered and should not generate any numerical mixing in the region.
Possible spurious effects due to the initial field should (?) be dampen out with time.
Have you ran into similar problem as described above?
Any comments or thoughts regarding the struggles described above are very much appreciated.
It would be interested to see model time series of the velocity in Baltic Proper from another Baltic Sea setup, or typical amplitude, decay rate and frequency of occurrence of near inertial waves.
Links to observations of currents in Baltic Proper are also very welcomed.
Five Mixing Regimes in lower left panel in figure mixing_contours.png:
Classical Mixing (Navy Blue)
• Criteria: Ri < 0.25 AND TKE > threshold
• Physics: Shear-driven instability with sufficient turbulent energy
• Result: Active mixing (textbook scenario)
Strong TKE Mixing (Dark Red)
• Criteria: TKE > threshold AND Ri ≥ 0.25
• Physics: Energetic turbulence overcomes stable stratification
• Examples: Wind mixing, internal wave breaking, convection
Potential Mixing (Orange)
• Criteria: Ri < 0.25 AND TKE < threshold
• Physics: Favourable conditions but insufficient energy
• Result: Mixing-ready but energy-limited
Stable Region (Dark Green)
• Criteria: Ri ≥ 1.0 AND TKE < threshold
• Physics: Strong stratification + weak turbulence
• Result: No mixing, preserved layering
Intermediate (Gray)
• Criteria: 0.25 ≤ Ri < 1.0 (transitional regime)
• Result: Variable/intermittent mixing
To view this discussion visit https://groups.google.com/d/msgid/getm-users/CAFvy56Wfh5x7AOeNC01kGNeC1CuM3J0rCvoh0B5AK4x4qrZq9w%40mail.gmail.com.
Adolf Stips
To view this discussion visit https://groups.google.com/d/msgid/getm-users/916a0180e2a842829a113a6d330a83ba%40fcoo.dk.
Lars Umlauf
Hello Johann,
I agree with Adolf that the amplitudes you see in your inertial
oscallations are not unusual at all. Have a look, for example, at
Fig. 5 in https://doi.org/10.1002/2013JC009483
And please also keep in mind that topographic waves with periods
of a couple of days may reach velocities of order 0.1 m/s (see
Fig. 7 in the above paper), comparable to the NIWs.
I'd say that excessive numerical diffusion is the main reason for the smeared halocline you see in your results (but I haven't really worked with GETM in a long time).
All the best,
Lars
p.s. I don't understand your regime TKE
> threshold AND Ri ≥ 0.25 . I would expect that
outside the surface and bottom boundary layers, shear is never
large enough to trigger turbulence. So turbulence in this region
should be exclusively "generated" by the prescribe (fixed) TKE
threshold, which corresponds to a primitive IW mixing model
predicting nu_t ~ 1/N
To view this discussion visit https://groups.google.com/d/msgid/getm-users/CAEXw-aZ894NCivdhL2ux2wCU6YX9hqD5xZiBN5rGp9WdhK0vgQ%40mail.gmail.com.
-- ------------------------------------------------ Lars Umlauf Department of Physical Oceanography Leibniz-Institute for Baltic Sea Research phone : ++49 381 5197 223 fax : ++49 381 5197 114 web : www.io-warnemuende.de/lars-umlauf-en.html address: Leibniz-Institute for Baltic Sea Research Seestrasse 15 D-18119 Rostock-Warnemuende Germany -------------------------------------------------
Johan Söderkvist
Dear Lars and Adolf
Very useful information, thanks!
I see that there are this type of velocity field in the deep Baltic. I found an error in my plot-script, and the time series for velocity was higher up in the water column. At 200 meters depth the model velocity was more in the range as observed.
With that said, I am still surprised that inertial currents are so frequently occurring in the Baltic .
Regarding Lars comment on mixing for Ri >025, my reference is not that solid. Might occur on rare occasions of deep ocean, if I remember correctly. Thanks Lars, for raising this.
So, the model generates of 14-hour cycle currents similar to what is observed.
However, bottom layer mixing seems to be too large. Otherwise I can’t explain the near 50m-75m thick bottom layer with constant salinity that not decay with time.
I agree that numerical diffusion is a strong candidate for the “too-linear” salt-profile.
Have any modellers experience in decreasing the numerical diffusion or removing high frequent variability in the 3D-velocity field?
Several advection schemes have been tested for velocity and tracers, but no success yet. Still there are some tests remaining.
Next step is to only change vertical advection schemes for salt from TVD-Superbee to TVD-HSIMT (third-order, monotone), thus not for temperature or velocity.
To view this discussion visit https://groups.google.com/d/msgid/getm-users/5693ae17-3695-4374-903b-0e599cb06ea9%40io-warnemuende.de.
Knut
Dear Johan,
one fast thought in between regarding the advection schemes...
theoretically it could also be the other way around. because of using anti-dissipative superbee for momentum your velocity field contains too much shear which in turn induces too much mixing. this is true for physical and numerical mixing (https://adcroft.github.io/assets/pdf/ilicak_et_al_OM_2012.pdf).
while superbee was improving the simulations 10+ years ago, it seems that this is not true for our present configurations (higher hor. resolution, improved adaptive coordinates) anymore. so far we did not find the time to deal with this.
cheers, knut
To view this discussion visit https://groups.google.com/d/msgid/getm-users/289dd193e30347739e27cf66c21ca1e8%40fcoo.dk.
Johan Söderkvist
Dear Knut
Thanks for sharing useful aspects. It seems like delicate process.
I assume higher horizontal resolution refers to grid size smaller than our current setup of 1nm?
To view this discussion visit https://groups.google.com/d/msgid/getm-users/3df25468-a574-40ed-a784-712ce51c61f7%40gmx.net.
Johan Söderkvist
Dear all
The source for the high frequent variability during summer have been identified. It was
ip_method 2: Blumberg and Mellor (linear)
I made mistake to shift from
6: Shchepetkin and McWilliams (2003)
ip_ramp also helps but I am not sure what order of magnitude is recommended.
Anyone have experience with the parameter ip_ramp?
To view this discussion visit https://groups.google.com/d/msgid/getm-users/cb0bca700f2d4e259e2e446924561937%40fcoo.dk.
Ulf Graewe
The source for the high frequent variability during summer have been identified. It was
ip_method 2: Blumberg and Mellor (linear)
I made mistake to shift from
6: Shchepetkin and McWilliams (2003)
ip_ramp also helps but I am not sure what order of magnitude is recommended.
Anyone have experience with the parameter ip_ramp?
Another note:
If the inertial oscillation have a to high magnitude, I could be an artifact of the grid resolution. I had the discussion with Peter Holtermann. He saw a similar behaviour in his 600 m setup. The issue is likely the energy of the inertial and internal waves. Usually, you have an downward cascade, to shorter wave lengths. Since the model has a finite resolution, the energy can not be dissipated below length scales of 2*dx. Thus, the energy is trapped and reflected upwards. Thus, you have to much energy at the smaller scales and this energy is partially pumped into the inertial oscillations. Hence they are to strong.
But I have no clue on how to fix it.
Cheers,
ulf
To view this discussion visit https://groups.google.com/d/msgid/getm-users/58353a0c0be042b18a85a054ed79b358%40fcoo.dk.
Johan Söderkvist
Hi Ulf
> I was already wondering, why you used ip_method=2
My worst mistake so far.
I believe it is the high frequent variability that also generated the strong oscillations in the velocity field.
Preliminary results show weaker currents in deep Baltic
Best regards
Johan
To view this discussion visit https://groups.google.com/d/msgid/getm-users/ef1755b6-3fac-4745-ae23-018b18846369%40io-warnemuende.de.
Johan Söderkvist
Dear all
Update from the 1nm. North Sea – Baltic GETM Setup at GEOMETOC in Denmark:
Finally, some progress here with simulating the the Baltic Halocline.
After 4 years of simulation with a GETM setup v015Ehy is able to maintain the halocline.
Also, the problem with constant salinity in a thick bottom layer is no longer present.
Initial conditions fits very well with observations at several locations in the Baltic.
The problem, lately, was that either was too much vertical mixing that resulted in too linear vertical salinity profile OR too sharp halocline with much too fresh surface layer and too saline bottom water with constant vertical salinity.
Several success stories helped to reach the present best case experiment
First of all, go back to Adaptive Vertical coordinates (AVC).
Old setup (v014Gbd) uses General Vertical Coordinates (GVC)
Increasing minimum depth to 0.5 meter, solved the problem with too many vertical iterations.
Then:
1) Deepening the Great Belt and make it even more smooth than the already filtered bathymetry.
Similar change was also mad in The Sound.
(With the old bathymetry, not enough salt water entered the Arkona and Bornholm Basin)
Also, some filtering along the slopes in Baltic Proper was also made.
2) Decreasing z0 and smag_const in deep parts of Great Belt and the Sound: smag_const=0; z0=0.001
Also z0=0.003 in Arkona basin – Slupsk Furrow area. (z0 in Baltic is 0.03)
3) Focusing adaptive vertical coordinates to 60 meters.
4) Major progress was also achieved by testing several combinations of advection schemes and ip_method.
Also, testing with different advection schemes for vertical and horizontal advection.
Here, different values of k_min and eps_min were also investigated.
Some experiments generated high frequent variability in several parameter, mainly due to choice of
ip_method.
It turns out that with TVD-Superbee and ip_method = 6 now caused too strong halocline.
Updating to ip_method =7, improved the vertical structure, but still too strong halocline, and with drift.
Best experiment so far (v015Ehy) use TVD-P2-PDM advection scheme and ip_method=6.
And k_min=1.e-7,eps_min=5.e-11. Earlier in the process.
#
There is however some issues remaining.
The surface salinity becomes too low. After about 2-3 years of simulation there is a tendency towards too low surface values.
On some occasions the model agrees well, but most of the time the salinity is too low.
It is most pronounced in the southern parts (e.g. Gdansk Basin, station BCSIII-10).
But further north, up to Gulf of Finland the model surface layer is also too fresh.
Decreasing cross sectional area and depth in the Danish Straits had nearly NO effect on surface salinity, but Baltic deep water salinity became too low. Now, the effect of decreasing the fresh water outflow is investigated. Results not ready yet.
#
Another issue is that west of Gotland the model halocline is too strong, with too fresh surface layer and too saline bottom layer (e.eg. BY31). Here, one possible solution would be to fine increase the smag_const in this area. or change the bathymetry.
#
Profiles in Arkona Basin, Bornholm Basin and time series at the German buoys in Fehmarn Belt, Darss Sill and Arkona captures most of the variability. It is a large improvement compared to the v014Gbd experiment.
But, the 2 layer structure often observed in Great Belt and northern part of Sound is rarely captured.
The halocline is much too weak. I am not sure if this can be solved with a 1nm horizonal resolution model.
#
During the process of identifying areas with large numerical mixing, I found some occasions with problematic heating periods. For example, in Arkona basin during calm summer days there is a horizontal sea saw pattern in surface temperature. Heating and cooling is stronger at every 2nd grid point along a transect, compared to neighbouring grid points. The amplitude in the variation along the transect is less than 0.5 degrees. These are episodic events during summer. I did some experiments with increasing k_min and eps_min, decreasing 3d-time step to 45 sek, changing Jerlov g1_const from 0.35 to 0.45 (to allow deeper penetration of solar radiation). Nothing solved this problem.
I would like to hear your thoughts on some of the issues remaining:
1) What river forcing do you use? Have you adjusted the freshwater outflow? Have you noticed the adjustment time scale in surface salinity in experiments with changed river forcing?
2) I am interested to hear how you have solved the inflow through the Danish Straits to the Baltic Sea.
Is there a 2-layer structure observed in great Belt in your model setups?
3) How does your model setup simulate the vertical structure at other locations than the Gotland Deep (BY15)? Experiment v015Ehy is able to pretty well describe the vertical profile from Arkona Basin up to entrance to the Gulf of Finland (BY29). But, there is some variability not captured by the model, meaning that the halocline becomes occasionally too weak compared to observations. This variability is most pronounced in the southern Baltic, for example close to station BY10, east of the southern part of Swedish Island Öland. Also, best experiment so far exhibit too strong stratification west of Gotland. Have you made some regional adjustments to improve the model results here.
4) Have you noticed a sea saw pattern in surface temperature during intense heating periods?
To view this discussion visit https://groups.google.com/d/msgid/getm-users/f6e71a99987841829e157761a9ef878b%40fcoo.dk.
Johan van der Molen
|
|||||||||||||||||||||||||||||||
To: getm-...@googlegroups.com <getm-...@googlegroups.com>
Subject: SV: [getm-users: 3441] Baltic Sea modelling
Johan Söderkvist
Dear Johan
Thank you very much for your input!
I will look into the source code we use and get back to you.
50% reduction of river forcing is huge.
Before I got your email, I started two experiments with some 10%. It takes some time for adjustment, but stopped one of the two, in favour of a 50% reduction in river forcing. It will be a very exciting Monday.
I have had a look at the Rhein observational data set, from BSH Germany.
The difference between SMHIs EHYPE and the BSH data are sometimes several hundred m3/s, both too high and too low values. I will make some more thorough comparisons.
To view this discussion visit https://groups.google.com/d/msgid/getm-users/PRAP192MB150519CB39E20CCDD6D01B08E7C5A%40PRAP192MB1505.EURP192.PROD.OUTLOOK.COM.
Johan van der Molen
Hi Johan,
Looking forward to see what you’ll find.
For the Rhine, we use data compiled for ICG-EMO based on data from Rijkswaterstaat. If you like, I can extract that into a netcdf file so you can throw it in the mix?
Best wishes,
Johan
|
| |||||||||||||||||||||||
|
To view this discussion visit https://groups.google.com/d/msgid/getm-users/aa94f32af40f4ab78ea96c59594d989d%40fcoo.dk.
Ulf Graewe
true, the river discharge into the Baltic is a crucial issue.
We used a deep neural network, to predict the river discharge based on atmospheric forcing fields, to get rid of a hydrological model. As trainings data , we used the E-hype data set. By comparing the E-Hype data with observations, we also saw huge differences. Although it would be nice to get the peaks and the droughts right, in my opinion, the most important thing is to get the annual discharge right (the long-term mean of 15.000 m³/s).
If that is fine, the correct annual cycle is a second order effect.
In my case, I reduced the E-hype data by 10%.
Cheers.
Ulf
To view this discussion visit https://groups.google.com/d/msgid/getm-users/PRAP192MB150519CB39E20CCDD6D01B08E7C5A%40PRAP192MB1505.EURP192.PROD.OUTLOOK.COM.
Johan Söderkvist
Dear Ulf
Great to hear from you!
Very interesting to hear your thoughts on this, and what number you ended up with.
By chance, the first shot here was 12% reduction.
From the neural network results, where there any regions where deviation from EHYPE was very large?
I agree that it is the annual cycle are more important than specific events.
It takes several years for the model to adjust to a new state, when changing the river discharge.
best regards
JOHAN sÖDERQVIST
Oceanographer, PhD
JOINT GEOMETOC SUPPORT CENTRE
DANISH MINISTRY OF DEFENCE ACQUISITION AND LOGISTICS ORGANISATION
Lautrupbjerg 1-5, DK-2750 Ballerup
Denmark
Phone: +45 4078 7321
E-mail: j...@fcoo.dk
Fra: getm-...@googlegroups.com <getm-...@googlegroups.com> På vegne af Ulf Graewe
Sendt: Tuesday, 11 November 2025 08.51
Til: getm-...@googlegroups.com
To view this discussion visit https://groups.google.com/d/msgid/getm-users/bb685885-ff26-4222-93a3-0dde115ecae0%40io-warnemuende.de.
Johan Söderkvist
Dear Johan
A Rhine NetCDF file would be very appreciated thank you.
A 50% reduction was too much for our setup.
Had already started an experiment with 12% reduction.
This helped a lot.
Surface layer salinity improved and closer to observations.
However, preliminary results show too high saline values in the 100-150 depth interval.
Earlier in the process, the deep channel in Great Belt that connects the Baltic with Kattegat was increased and smoothened. This might be adjusted .
So, there is some fine-tuning remaining, but adjusting after freshwater outflow seems to be a good ide .
To view this discussion visit https://groups.google.com/d/msgid/getm-users/PRAP192MB1505A1C62954D18E98CE0936E7CEA%40PRAP192MB1505.EURP192.PROD.OUTLOOK.COM.