Evaluation DART-TCO319 simulation without Ross Sea salinity problem

[tsemmler05]

CVDP_AWICM3.pdf

Especially trends in the Southern Ocean need to be closely monitored. General caveat: longer spin-up period because this simulation had to be cold started to avoid Ross Sea salinity issue. One year of coupled spin-up simulation with FESOM time step of 60s has been performed before the long simulation with 240s FESOM time step.

[tsemmler05]

These are some selected results of the 1950 spinup simulation:

performance index over years 20-29:

SST averaged over years 20-29:

SSS averaged over years 20-29:

AMOC averaged over years 20-29:

Ice extent as time series for years 0-29:

Tropical ocean temperature Hovmoeller diagram years 0-29:

Global ocean temperature Hovmoeller diagram years 0-29:

Atlantic ocean temperature Hovmoeller diagram years 0-29:

[tsemmler05]

Southern Ocean trends are not that great. In the old DARTY simulation we also had a dip in Southern Ocean sea ice and afterwards a recovery - so hope is there that this could happen for the present simulation, too. The DARTY simulation was warm started from a short 5 year uncoupled spin-up - so like the present ctl1950d simulation still far away from a spun-up state. The sp1950d simulation with the Ross Sea salinity issue was actually restarted from DARTY simulation (that also has Ross Sea salinity issue) - it means it had been started in quite a spun-up state compared to the new ctl1950d.

[christian-stepanek]

Hi Tido,

very promising results. Except for sea ice, std. dev. tos and mlotst all seems in order according to Jan's performance indices. Quite interesting that the 10 m thetao, and as well the tas, both look fine nearly everywhere, but the std deviation of tos is quite a bit off. So there is still a problem with variability in the model?

The deep ocean looks particularly great. Is this an effect of the spinup? Did you initialize from the ocean state against which you grade the simulation, so that the exceptional agreement of the simulated deep ocean may just be a side effect of incomplete model equilibration?

[tsemmler05]

I think the great deep ocean is an effect of the spinup since we are doing a cold start with the EN4 climatology that we are comparing against. So it might still drift away over time. But really good news: The Drake passage transport in model years 51-56 (nominal years 1900-1905) is 157 Sv. In the simulation with Ross Sea salinity problem it was about 90 Sv.

[tsemmler05]

These are some results of the first 50 years of simulation.

Sea ice is recovering after the dip half way through the 50 years. Drake passage transport is realistic. The only thing is: std. tos in El Nino region is not great while std. zos is.

Performance index matrix years 26-50:

Sea ice extent:

AMOC:

Hovmoeller diagram Global Ocean temperature bias (still some drift there particularly in around 500 m depth):

SST bias:

[tsemmler05]

In years 51-70 of the ctl1950d simulation the Southern Ocean warm bias is practically gone while the Barents Sea - Kara Sea bias is enhanced: Here the SST bias:

This is also reflected in the sea ice - Southern Hemisphere sea ice is increasing, Northern Hemisphere sea ice decreasing. Here sea ice extent:

The performance index is slightly improved in years 51-70 compared to years 26-50. Especially note the El Nino SST variability.

[tsemmler05]

This is the evaluation of Sun-Seon with the CVDP package.

CVDP_AWICM3.pdf

[JanStreffing]

I just checked in on this again after a while. The run seems to be heading in a good direction.

From Sun Seons analysis, my main takeaway for the eventual tuning towards v3.2 is that the NH sea ice fraction is strangely low throughout the year, but especially in Summer (85% vs 99% in central Arctic). I think 10 years chunks are a bit short for the SST variability metric in the nino34 region.

[tsemmler05]

I think 10 years chunks are a bit short for the SST variability metric in the nino34 region.

20 to 25 years are used for the PI metrics (for example years 26-50 or years 51-70).

[JanStreffing]

Ah right, it's 20-25 years. So something does change there. I guess you can also see this in the ENSO evolution. variability decreases starting from ~year 50.

[tsemmler05]

Performance index of years 76-100:

[tsemmler05]

This is the performance index of the first 25 years of the 1950 control simulation (nominal years 1950-1974; follow-on from the 100 years of 1950 spin-up simulation).

[tsemmler05]

Compared to the last 25 years of the 1950 spin-up simulation (see previous post: years 76-100), the Arctic sea ice has been improved. Probably because the summer sea ice extent has increased in the first few years of the 1950 control simulation. The two graphics show the sea ice extent in the 100 years of 1950 spin-up simulation and in the first 25 years of 1950 control simulation:

[tsemmler05]

Energy balance of the first 25 years of 1950 control simulation:

TOA: 0.15 W/m2 SEB: 0.07 W/m2

[tsemmler05]

Annually averaged sea ice thickness (first 25 years of 1950 control):

[tsemmler05]

SST bias (first 25 years of 1950 control):

[tsemmler05]

Hovmoeller diagrams temperature and salinity (25 first years of 1950 control) - while trends are small, there is the issue of too fresh and too cold water at the surface and too saline and too warm water underneath (please note that the unit in the second plot, the salinity plot, is, of course, psu):

[chrisdane]

Why is the spatial distribution of the SST trends in the two pdfs completely different? (Tidos initial post from March 2 versus Tidos post from March 20; both page 4)

[tsemmler05]

Because of the spin-up effect. The original post shows data from the first ~20 years of the simulation, the second post from years 33 to 82.

[tsemmler05]

Surface energy balance and TOA energy balance (W/m2) in the first 50 years of 1950-control and 1950-historical simulations:

                      SEB   TOA

1950-control 1950-1974 0.07 0.15 1950-control 1975-1999 0.07 0.16 1950-historical 1950-1974 0.19 0.25 1950-historical 1975-1999 0.51 0.58

[tsemmler05]

Performance index 1975-1999 of 1950-control and 1950-historical simulations:

[tsemmler05]

Not sure why El Nino region standard deviation is that different between 1950-control and 1950-historical (quality of tos and zos standard deviations reverse)

[tsemmler05]

Time series of ice extent in 1950-control and 1950-historical:

In the historical run, especially the Northern Hemisphere summer sea ice extent is declining towards the end of the 20th century.

[tsemmler05]

AMOC averaged over 1975-1999 in 1950-control and 1950-historical runs:

In the historical run, overturning is slightly weaker in the north and slightly stronger in the south compared to the control run.

[tsemmler05]

AMOC averaged over 2000-2014 in 1950-control and 1950-historical runs:

Maximum slightly weaker in historical run compared to 1950-control run.

[tsemmler05]

Global ocean temperature profile bias (K) (first 65 years of 1950-control run):

[tsemmler05]

Global mean 2 m temperature in 1950-control and 1950-historical simulations. Seems like climate sensitivity a little bit on the high side, though probably not as extreme as for coarse resolution run.

[tsemmler05]

Sorry, y axis should of course read degree C, not K.

[tsemmler05]

Global mean 2 m temperature in C until 2055:

around 3 degree temperature increase from mid- to mid-century.

[tsemmler05]

AMOC streamfunction 2046-2055

[tsemmler05]

2 m temperature response 2050-2057:

[tsemmler05]

Strong Arctic Amplification, Weddell Sea warming; North Atlantic Warming hole quite extensive.

[tsemmler05]

Sea ice extent:

Arctic ice free in summer too early (after 2025 September Arctic sea ice extent below 1 Million squared meters in every year); starting Southern sea ice decline in the 2020's; strong Southern sea ice decline in the 2030's.

[chrisdane]

Quite a trend in piControl tas?

[tsemmler05]

About 0.1 K from 1950 to 2050 in 1950-control.

[tsemmler05]

Towards the end of the 2070s, we are already in a 4.5 degree warming world:

[tsemmler05]

AMOC streamfunction 2050-2074:

[tsemmler05]

AMOC only slightly different compared to 2046-2055

[tsemmler05]

Sea ice extent really declining. Northern Hemisphere wintertime sea ice extent a bit more stable. If I would have had a second shot, I might have increased sea ice albedo values by 1 or 2% to avoid the too low Arctic summer sea ice extent. In the 1950 control simulation, Arctic summer sea ice extent is reasonable but concentration too low.

[tsemmler05]

This is the 1970s difference between 1950-hist and 1950-control:

Cooling over Europe and parts of East Asia, Australia, South America, Western USA, Tropical and Southern Oceans. Some of it could be due to aerosols, but not too sure about this given natural variability.

[tsemmler05]

1980s difference between 1950-hist and 1950-control:

Cooling over India, parts of Europe, and over Southern Ocean. While for the northern aerosol areas not so sure how much is just natural variability, it seems like that the Southern Ocean cooling is quite robust between 1970s and 1980s.

[tsemmler05]

And this is the 2000s difference:

Quite a change in the Southern Ocean! Now warming kicks in, maybe 10 or 20 years too early compared to observations. Cooling or only very slight warming over South-East Asia related to aerosols?

[JanStreffing]

Sea ice extent really declining. Northern Hemisphere wintertime sea ice extent a bit more stable. If I would have had a second shot, I might have increased sea ice albedo values by 1 or 2% to avoid the too low Arctic summer sea ice extent. In the 1950 control simulation, Arctic summer sea ice extent is reasonable but concentration too low.

Probably a minor issue but these plots have more data points than they should. For ever 10 years there are 20 points. The strange jumps seen are the result of this.

Regarding SO warming: I had previously seen a similar effect for TCo95L91-CORE2 where we have early onset Antarctic ice melting. At the time I had hypothesized, that this is due to ~+1.5-2K warm bias & ~-0.6m sea ice thickness low bias in TCo95L91-CORE2 PI state:

I had think that the warmer PI primes the model to react in the SO to GHG inducing global warming sooner than is the case in reality. Your TCo319L137-DART PI state has smaller warm SST biases, still ~+1K too warm. By the 2000s you are nearly at +2K in the SO with the PI bias + global warming. At +2K from present day observational estimates SO sea ice begins to melt.

[tsemmler05]

Regarding the plot problems: I was in touch with Patrick Scholz but he couldn't fix the issue with tripyview on aleph yet. Therefore, I am still using fdiag, and there are for every year 12 monthly data points put too close together - I thought I'd not invest time right now in this and rather wait for tripyview. Interesting about Southern Ocean in coarse resolution setup!

[tsemmler05]

This is now the temperature response 1960-1979 hist1950-ctl1950:

And this 2000-2019 hist1950/ssp585-ctl1950:

[tsemmler05]

Europe and the United States are heated up in 2000-2019 while limited warming or cooling is simulated for 1960-1979 compared to 1950ctl simulation. In contrast, South-East Asia still shows large areas with very limited warming. One can say this is consistent with the expected impact of aerosols.

[tsemmler05]

Milestone reached! The ssp585d simulation has arrived in 2100. We are now in a six degree warming world:

[ibs-ccp]

Wow, 6 C for SSP585 is definitely scary. Does anyone know if the low-resolution model has a similar sensitivity?

[JanStreffing]

We don't have the transient runs for SSP585. @fernandadialzira recently ran 4x CO2 simulations and found the ESC for doubling, which was 4K (8K for 4x). This is right in the middle of the CMIP6 ensemble. I might be interesting to investigate mixed layer depths and ocean heat uptake for the TCo319-DART SSP85 simulation.