Progress: 45%

WP4.1 covers several model development, validation and different types of sensitivities by different models. In some cases, we even explore the extent of comparability of similar sensitivity experiments with very much different model types. Sensitivity studies and model development always benefit from each other and thus are presented together.

SMHI: (50%)

SMHI's coupled model validation

During project months 13-24, SMHI has carried out further validation and sensitivity runs with the coupled ocean-ice-atmosphere model RCAO and its standalone components.

After first tests in the 1980's and 1990's, RCAO is now able to run through the complete ERA-40 period (1960-2000). The sea ice extent of the coupled model is given in Fig. 1 together with observations. The first long run (red in Fig. 1) has been initialized in 1959 with actual ERA-40 data in the atmosphere and with T/S climatology in the ocean. Sea ice has initially been set to 2.30 m thickness and 0.95 concentration where the SST is less or equal the local freezing point. The development of sea ice extent shows clearly a spin-up phase of about 20 years. After the late 1970's the simulated sea ice extent is close the interannual average of observations. Contrary to ocean-ice standalone models, no agreement in the year-to-year variations can be expected due to internally generated variability in the Arctic coupled system. This internal variability is substantial and is at least 50% of the externally generated variability. This is indicated by first results from predictability runs by SMHI in WP4.2.

The second coupled run (blue in Fig. 1) starts from an ocean snapshot from year 2000. After the first 40 years of integration the coupled ocean and ice can be expected in quasi-equilibrium, i.e. adjusted to the model and in agreement with the advective regime. This gives the second run a more realistic start as indicated by the much improved ice extent during the first few years. Still the second ocean initial fields are not adjusted to the conditions of the year 1959, which leads to an inability to cover multiyear variability during the spinup period as indicated by the winter ice thickness disagreement around 1970. The oceanic start field can be further improved by iteration through the ERA-40 period several times. This has been demonstrated by INM-RAS (this task WP4.1). However, the genuine problem with the initial multi-year variability can only be addressed by an earlier start time. This is currently not possible for the coupled setup RCAO. SMHI plans to run the ocean component RCO through the complete 20^th century, forced by reconstructed AOMIP atmospheric forcing. Results from this ocean run will then be used to start the coupled runs in 1959. For now, we focus on the period after the spinup, i.e. after 1978.

Figure 1: Arctic sea ice extent for two RCAO coupled runs (red, blue) and observations (black)

Between 1978 and 2000 the sea ice anomaly is depicted in Fig. 2. The two simulations start in 1959 and differ only by a slight difference in the initial field. Differences are due to internally generated non-linear interaction. The simulations are compared with the anomalies of satellite observations and with ERA-40 data. All observations and model runs are supplemented by the associated trends during this period. Trends of observations, reanalysis and model simulations are similar. The spread in trend between the two simulations is only little larger than the difference between the direct observation and the reanalysis. Thus, the coupled model is well capable to resemble the decreasing sea ice extent during the satellite age. Many global climate models (GCMs) such as used for the IPCC report, often underestimate this trend. It remains to be seen if the better Arctic performance of the regional model (RCM) is due to better process descriptions, or if it is due to the ERA-40 forcing at the lateral atmospheric boundaries. SMHI plans to utilize climate control forcing from a GCM at the lateral atmospheric boundaries (WP5.3).

Another interesting feature of Fig.2 is the distinct agreement between simulations and observations during the years 1990 and 1995. During these specific years the NAO index is extremely positive. This indicates a control of Arctic internal variability by the large scale circulation.

Figure 2: Arctic summer sea ice anomaly (with respect to 1978-2000) for RCAO runs (red) and observations (black)

 

SMHI's experiments on parameterization of mixing in regional circulation models for the Arctic Ocean

Figure 3: RCO (blue) and RCA (red) model domain

In this study we have investigated the role of mixing parameterizations in three-dimensional modelling of the Arctic Ocean and its impact on the ice cover, temperature and salinity fields, and on the circulation using the regional coupled ice-ocean model RCO, the Rossby Centre Ocean model, developed at SMHI. RCO is also used as ocean component in the coupled atmosphere-ice-ocean model RCAO.

A special focus in this study will be on the ice cover and on the Atlantic Water circulation. A result of the Arctic Ocean Model Intercomparison Project (AOMIP) was that even regional models with higher horizontal grid resolution than RCO can have problems to reproduce the circulation of the Atlantic Water realistically (Proshutinsky et al, 2005). Approximately half of the models investigated within AOMIP showed cyclonic circulation of Atlantic Water (as observed) and the other half showed the opposite circulation. Below results from two sensitivity experiments using our Arctic Ocean model are presented.

Model and Experiments

RCO is a regional version of the global OCCAM model coupled with a Hibler-type two-level (open water and ice) dynamic-thermo-dynamic sea-ice model using viscous-plastic rheology. The model domain of RCO covers the central Arctic Ocean, the Nordic Seas, and the North Atlantic roughly to 50°N (see Figure 3). Open boundary conditions are implemented at 50°N. The horizontal resolution is rather coarse and amounts to 0.5° or approximately 50 km in a rotated coordinate system centered over the North Pole. Some of the model's characteristics and parameters are presented in Table 1. For a more detailed model description see e.g. Meier et al (2003).

 

Table 1. Model characteristics

In each sensitivity experiment the model was integrated over the period 1958 to 2005. The model was forced by ERA-40 data for the period 1958-2001 and by weather forecast data from ECMWF for the period after 2001. As the initial state the PHC (Polar Hydrographic Climatology) was used. To test the sensitivity of mixing several experiments with different mixing schemes and different values of background horizontal eddy viscosity and diffusivity were carried out. In this report we only present two of them, the Richards (Ri) number dependent empirical model and the two equationk – ε model.

Data and Analysis

To evaluate the impact of mixing on salinity and temperature distributions in our experiments we have compared the model simulations with observational data from the Submarine Science Expedition (SCICEX) in fall 2000 and with climatology (PHC).

To investigate the influence of mixing on the sea ice we compare sea ice thickness and concentration with satellite derived products. For the sea ice thickness we compare seasonal (October-March) data averaged for the period 1993 to 2001 with results reported by Laxon et al (2003). For sea ice concentration the mean state and variability is studied by using EOF (Empirical Orthogonal Functions; see e.g. Von Storch and Zwiers, 1999) analysis on monthly averaged winter (November-April) and summer (May-October, not shown in this report) data for the period 1978 to 2005. The same analysis is carried out for sea ice concentration derived from satellite-borne passive microwave sensors available at NSIDC (http://nsidc.org/). The model results of the two experiments are compared with these observations.

Results

One of the key processes in the Arctic Ocean (and also one of the major goals to study in AOMIP) is the inflow and circulation of warm North Atlantic water (AW) at depths from approximately 200 and down to 800 metres. This AW layer is especially evident in the Eurasian basin for the SCICEX observations and the climatology (Figure 4a and 4b), although the climatology is somewhat colder. For the Ri model experiment (Figure 4d) the AW layer is also present but with a somewhat disconnected AW core. The k – εmodel experiment on the other hand does not have the observed AW core along the transect (Figure 4c). In the Amerasian basin the latter model experiment show a much too warm AW layer whereas the Ri model has a reasonable, although somewhat too cold temperature distribution. Its vertical distribution is also very variable along the transect in this region.

Figure 4. Temperature distribution along the SCICEX transect for a) SCICEX datab) PHC c) k – εmodel experiment and d) Ri model. For the model simulations it is the mean of 2000

a)

b)

Figure 5. October 2000 mean potential temperature integrated over 3 model levels (approximately 350- 530 metres depth) for thek – ε model experiment (a) and Ri model (b).

C)

Figure 6. Mean seasonal (October-March) sea ice thickness for the period 1993-2001 for a)k – ε model experiment, b) Ri model and c) Laxon et al (2003) redrawn from Nature. Note that panel c) has a different range for the colorbar.

From the horizontal temperature distribution (Figure 5a and 5b) in the AW layer it is evident that the Ri model has a larger inflow of AW through the Fram Strait. What is also notable is that the Barents Sea branch of the AW (entering the Arctic Ocean through St. Anna Trough ) is too cold. The circulation in the Eurasian basin is cyclonic for both experiments. In the Amerasian basin the circulation is weakly anticyclonic for thek – ε model experiment sustaining the warm core from the initialization. For the Ri model the circulation is cyclonic.

The average ice thickness distributions (Figure 6) show that both model experiments have a similar but thinner ice cover compared to the satellite derived distribution in Figure 6c. It is also visible that the Ri model experiment (Figure 6b) has a region between the Canadian Archipelago and the North Pole with significant thicker ice cover, compared to the k – εmodel experiment.

From the analysis of the average winter sea ice concentrations one can see that the modelled and the satellite derived long term mean fields are very similar on the basin scale, however small regional differences are seen along the ice edge, especially along the Siberian coast where the model has too low ice concentration and outside the eastern coast of Greenland where the famous “Is Odden” (the blob shaped region) is visible in the satellite data (Figure 7c) and partly visible in the Ri model experiment (Figure 7b) but not in the k – εmodel experiment (Figure 7a). The first EOF mode is shown in Figure 8a – 8c where it is evident that most of the variability is located in the Barents Sea, Labrador Sea, Bering Sea and along the eastern coast of Greenland (only for the satellite data). The correlation between the first principal component (PC) of each model experiment and the satellite data is fairly good, 0.83 for the k – εand 0.78 for the Ri model experiment. In Figure 6d and 6e the modelled and satellite derived PCs are compared to the Arctic Oscillation index. They follow the low frequency (decadal) oscillations quite well, whereas the high frequency (interannual) component is less accurate represented by the models.

a)

b)c)

Figure 7. Mean winter (November-April) sea ice concentration for the period 1978-2005 for a)k – ε model experiment, b) Ri model and c) NSIDC satellite data.

a)

b) c)

d)

e)

Figure 8. The first EOF-mode for winter (November-April) sea ice concentration from the period 1978-2005 for

Laxon, S. N. Peacock and D. Smith, 2003. High interannual variability of sea ice thickness in the Arctic region, Nature 425, pp. 947-950, doi:10.1038/nature02050.

Meier, H.E.M., R. Döscher and T. Faxén, 2003: A multiprocessor coupled ice-ocean model for the Baltic Sea: application to salt inow J. Geophys. Res., 108(C8), 3273, doi:10.1029/2000JC000521

Proshutinsky, A., Yang, J., Gerdes, R., Karcher, M., Kauker, F., Häkkinen, S., Hibler, W., Holland, D., Maqueda, M., Holloway, G., Hunke, E., Maslowski, W., Steele, M., and

Zhang, J., 2005: Arctic Ocean Study - Synthesis of Model Results and Observations. EOS, Transactions, American Geophysical Union, 86(40), 368-371

Von Storch, H. and F. W. Zwiers, 1999. Statistical Analysis in Climate Research, Cambridge University Press, New York.

Zhang, J., and M. Steele (2007), Effect of vertical mixing on the Atlantic Water layer circulation in the Arctic Ocean, J. Geophys. Res., 112, C04S04, doi:10.1029/2006JC003732.

UG in collaboration with SMHI (35%):

Introduction

A coupled atmosphere-ice-ocean column model (Figure 9) is used to calculate the Arctic ice cover and its sensitivity to changes in the forcing parameters. The ice cover is described by a thickness distribution where each ice category is allowed to independently evolve thermodynamically. The model comprises formulations for ridging and ice export, two processes that are important for the ice thickness distribution. The study will focus on how the ice cover reacts to changes in the cloud distribution and to varying albedo. The effect of changes in the atmospheric heat flux will also be considered.

The first issue has been to look at the effects of a varying albedo and some results from this investigation are presented here.

Figure 9. Schematic sketch of the coupled model.

Effects of a varying albedo

In this first study two different models for the atmospheric part have been used. The first one, denoted I, is rather simple and based on Thorndike (1992) whereas the second one is a version of the NCAR Community Climate Model (CCM2), denoted II. In both models the albedo, a, is described according to Maykut (1982): a = G1 a_snow + (1-G1) a_ice ; G1 = 0 if h = 0 and G1 = 1 if h > 0, where h is the snow thickness. The ice albedo is a function of the ice thickness H, a_ice = min (0.44 H^0.28 + 0.08, 0.64), and the snow albedo is given as monthly values shown in Table 2.

Table 2. Monthly values of the snow albedo according to Maykut (1982).

In order to test the sensitivity of the modelled ice thickness to changes in the albedo, the maximum value of the ice albedo was varied between 0.6 and 0.7 for three different cases of the snow albedo. The standard case of this latter quantity was the one shown in Table 2 (ASN=1) while the two others were obtained by changing the values with ± 5 % (ASN=1.05; 0.95). The results are shown in Figure 10 where the annual mean of the ice thickness is plotted for the various albedo combinations. A general outcome is that model II, with the more advanced description of the atmosphere, gives a 0.5- 1 m thinner ice cover than model I. The ice thickness is very sensitive to changes in the snow albedo, a 5 % decrease causes a thinning of H ranging between 1.5- 2.25 m (model II) and 2- 2.7 m (model I). Since the ice cover becomes very thin most ice categories do not reach the thickness for maximum albedo. In model II the ice will for this case totally disappear during summer. For the examined span of the maximum ice albedo the ice thickness change is about 0.75 m for ASN=1 and ASN=1.05, respectively, regardless of the model used.

Figure 10. The figure shows the annual mean ice thickness for different values of the snow albedo (ASN) and the maximum ice albedo.

The next tests to be carried out will be for the albedo formulations used in the 3-dimensional RCAO model.

FIMR: Progress (30 %)

FIMR has prepared NSIDC upward looking sonar draft data for model validation purposes. The dataset consists of 37 submarine cruises between 1975-2000. Monthly statistics of the ice thickness have been calculated for specific regions. A first comparison has been carried out with the MPI-climate model (i.e ECHAM5/MPI-OM1, sea-ice in that model is described as a classical two-level Hibler model). Results are:

• Modelled maximum ice thickness realistic, minimum ice thickness overestimated
• Modelled variability of ice thickness rather close to submarine observations
• Change in the ice thickness nonlinear in time, during 2000-2030 decrease of ice thickness is 0.7 m/30 a but during 2030-2050 decrease is 1.5 m/30 a
• Variability of ice thickness increases when the ice gets

In addition, FIMR have examined how sea-ice thickness distribution depends on the synoptic scale variability of the wind field. For this purpose, the HELMI multi-category sea-ice model was used. The OMIP atmospheric forcing data provided by the MPI were used in these simulations. We have made two simulations, a 20 years reference simulation which takes account of synoptic scale variability of the wind field and an other 20 years simulation where the effect of the atmospheric cyclones was removed from the forcing data. The steady state results of those simulations are shown in the figure 11. In general, the sea-ice thickness fields are similar in both simulations, but the simulation with the effect of cyclones removed produces less ridged ice and resulting thinner mean sea-ice thickness. We can conclude that variability of the Arctic sea ice conditions is mainly driven by the large scale atmospheric circulation, but regionally synoptic scale variability of the atmosphere is important for generating anomalies into the ice thickness field.

Fig. 11 Simulated mean sea-ice thickness and deformed ice fraction from the HELMI simulations for the reference simulation (left panels) and from the case where the synoptic scale variability of the wind is neglected.

Met.no (40%)

Several important improvements have been implemented in ORCM during the first phase of DAMOCLES and these are described in deliverable report D4.1- 02. In addition, early sensitivity results for the model will be described in report D4.1-04. However, due to problems with porting the numerical code to the new super computer facility available for this project in Norway, the progress has been delayed the last 9 months. The latest status now is that the computer facility was upgraded October 1 2007, such that we are again allowed to run the ORCM on the available super computer. Also, an alternative computer should be available for us during the start of 2008. During the transition to the new computers, the atmosphere component of ORCM is upgraded to met.no-version II of HirHam (Haugen & Haakenstad, 2006). The main change in the model is a transition to a semi-Lagrangian, semi-implicit time integration scheme. The new atmosphere model runs more effectively on the computers and shows reduced pressure, precipitation and temperature errors compared to the old model. In addition, the ice-ocean component has been prepared for salinity restoring or salinity flux correction. This is necessary due to the unbalanced hydrological cycle in the regional model systems.

Reference:

Haugen, J.E. and H. Haakenstad (2006): The development of HIRHAM version 2 with 50km and 25km resolution. In RegClim General Technical. Report 9. Norwegian Meteorological Institute. http://regclim.met.no/results/gtr9.pdf

INM-RAS: (25%)

The main goal for the INM RAS (Nikolay Yakovlev) within DAMOCLES is the implementation of the climate model with the explicit tides to investigate their role in the formation of the sea ice patterns (thickness and compactness). During the first 6 months of the DAMOCLES Extension the work was subdivided into several subtasks:

1. Forcing data preparation for the 1/6 deg. model for the area 40-90N in the North Atlantic-Arctic + Bering Sea. Forcing data consists first of all of bottom topography in the rotated spherical coordinates, air temperature and pressure, wind velocity, river runoff, etc. The planned experiments with the 1/6 deg. model will be coordinated with the 100-year runs in the frame of the Arctic Ocean Model Intercomparison Project (AOMIP).
2. The development of the parallel code for the INM RAS cluster.
3. The development of the code with the free upper surface of the ocean, investigation of the a priori requirements for the model with tides. Some theoretical work to investigate the possibility to use Z-star vertical coordinate system.
4. Methodological experiments with the low-resolution model (100km, 16 vertical levels, North Atlantic north from 65N) to investigate methods of introducing tidal forcing, and ways of ocean-ice coupling.

State of realization:

1.–2. The work is in progress, bottom topography and ocean temperature-salinity data are interpolated on the model grid. There are some problems with the atmosphere forcing data, for there were problems with the AOMIP data server. Parallel code development also was slowed due to INM RAS mainframe reconfiguration. The first results are expected in month 24 (after 12 months of the work start in the frame of the DAMOCLES-TTC). This work requires additional training of young scientists (graduate students), involved in the project.

3. The use of the Z-star coordinate system appeared is very attractive, especially to describe the under-ice boundary layer correctly, but appeared to be extremely difficult for the realization in the frame of the finite-element formalism in short time. Nevertheless the code for the unstructured grids was developed and tested. The first version of the model with the free upper surface in the z-coordinate system will be developed till the end of month 21.

4. A set of methodological experiments were carried out to test the model capability to work with tides. The latter were specified as an incident M2 wave (normal to the boundary barotropic velocity) on the liquid parts of the boundary (in the GIN Sea, in the Denmark Passage and in the straits of the Canadian Archipelago). The data on the tidal velocities are by Kowalik and Proshutinsky, 1994 (data are available on the site http://www.ims.uaf.edu/tide/m2tide.html). Simulated sea level elevation amphidromic map by the model in the “no wind, baroclinicity, ice and rivers” mode presented on Fig. 12 (the phase of the tide is arbitrary). These results, compared with the data by Kowalik and Proshutinsky (Fig. 5), show that the model may be used for further investigations of the role of tides in Arctic Ocean climate formation. The only significant differences are in the vicinity of the White Sea, which is not represented in the low-resolution model. Note, that the large scale model was run in the regular mode, with the time step 1 hour.

Fig. 12: Simulated M2 tide.

Fig. 13: M2 tide by Kowalik and Proshutinsky, 1994.

Preliminary tests with the full coupled model showed that

• There is no significant change in sea ice extent and ice mass, although tides tends to stabilize ice mass decreasing (Fig.14);
• There is some increase of ice mass and change in sea ice thickness distribution with the increase of more thick gradations;
• Tide mixed vertical column and cooled Atlantic Water layer significantly. This is clear in the Fram Strait temperature section (Fig. 15). This problem should be investigated.
• The parameterization of the ice-ocean dynamical interaction appeared to be of principal importance for the sea ice thickness modeling under the high frequency forcing. The problem was already indicated in the number of papers. In the series of papers by W. Hibler with coauthors there was proposed a solution, although this approach is not commonly adopted due to obvious lack of the momentum conservation. The main problem is to take into account not the skin-layer drag only, as it is common in large-scale models, but the shape drag as well. The latter is important for the ridged ice with large keels. Simple theoretical estimates show that for the realistic Arctic keel statistics shape drag may be 10 times more important, than skin-layer drag. Modeling of the ice-ocean fluxes by Large-Eddy-Simulation model shows that the process is very sophisticated due to salinity stratification, but in general shape drag do very important for large keel. Low drag leads to very weak effect of inertial and tidal oscillations on the sea ice, it may be the explanation of the weak effect of tides on ice, discussed above. The number of tests to estimate the appropriate model output by virtue of increased drag coefficient are in progress. This problem may be addressed to other WPs of the DAMOCLES as well.

 

Fig. 14. Ice mass series for two versions of the low-res. model.

Fig. 15. Fram Strait temperature sections – no tide and with tide. Note the intensive cooling in the Atlantic Water layer.

NERSC: (30 % progress in WP4.1)

TOPAZ Model system:

The TOPAZ model system has lately been updated, and the third version was set in real time operation in June 2007. It has a double horizontal resolution by comparison to the previous version (11 km in the Arctic until 16 km in the Equator, against 20 - 36 km in the previous version). The model is a North Atlantic and Arctic configuration of the HYCOM model in its recent version 2.1. See an illustration Figure with Google Earth.

Assimilation in TOPAZ

The assimilation system used in the real-time TOPAZ system is the Ensemble Kalman Filter (see the book Evensen 2006). The capability for assimilating sea-ice drift data from Ifremer has been added recently. The ice drift products are based on scatterometer and AMSR-E data.

Data delivery from TOPAZ

The numerical forecast from TOPAZ and the past analyses are available online via the THREDDS server, following the standard OPeNDAP protocol. This means that the fields can be retrieved directly from Matlab or any GIS software. See an example on http://topaz.nersc.no

In particular, the Fram Strait section is one of the selected output sections of the Arctic thematic portal and the DAMOCLES hydrographic data in the Fram Strait will be a major input data for the validation of operational systems in the Arctic.

Special forecast for the Tara expedition:

The drift of the Tara expedition is followed and forecast every week by the TOPAZ system. The forecasts are updated and presented in Google Earth http://google.nersc.no/TOPAZ_Tara_drift.kml, (see Figure 17)

Figure 16. Model output from TOPAZ as presented in Google Earth.

Figure 17. TOPAZ forecast for the Tara drift expedition

Read part 2 of this article here.

Home : Research : Work packages : WP4: Integration and data assimiliation in large-scale modelling and forecasting : Task 4.1 Model sensitivity and studies combining models with observation

Search:

Nov 10, 2008



• Home
• About the Arctic
• About DAMOCLES
• Research
• Publications
• DAMOCLES DATABASE
• SEARCH for DAMOCLES (S4D)
• Education
• Arctic news
• Contacts
• Press room