Progress: 42%

The work is grouped in the following 4 activities each of which will provide analysed fields of Arctic key variables combining information from observations with a different modelling system.

MI-IM/MIPOM SEIK

(Met.no) (60% progress):

An implementation of the SEIK assimilation filter in MIPOM has been tested in a twin experiment where SST and vertical profiles of temperature and salinity from an independent model run (from the ROMS model) is assimilated into MIPOM utilizing SEIK. The results show a positive impact of the assimilation with SEIK. Assimilation of SST has the largest impact on the results when looking at upper ocean quantities, but quite many vertical profiles are needed to give an additional improvement in this part of the water mass. Increasing the number of ensembles used in the filter has a clear positive impact on the results. A problem with the present implementation is that the filter, after several sequences of assimilation, underestimates the model errors due to a too small ensemble spread. This causes the effect of the assimilation to diminish with time, giving less weight to the observations, but more confidence to a possibly wrong model state. This is a possible, known, problem with all methods that estimate the background error covariances from a model ensemble. We are considering possible methods to increase the ensemble spread to improve this situation. This weakness of the method is an important point to improve when considering assimilation of ice concentration and other possible ice related variables.

Further work on the MIPOM-SEIK assimilation system has been directed to improve the earlier reported problems where the SEIK assimilation degenerates after several sequences of assimilation. To reduce the problems a version of local SEIK is being implemented. In contrast to the first implementation that takes all model points into consideration when calculating the error covariances and model updates, local SEIK uses and updates only model variables in a localized region around the observation. This reduces problems with possible unphysically long distance correlations between model points, and it is expected to give better updates (both physically and numerically) of the model state.

Assimilation of acoustic tomography measurements

(NERSC) (20% progress):

In the past 12 month there has been no action on this item. An Acoustic tomography experiment was planned to be initiated during a field operation from 18-23 October 2007. However, the deployment of the tomographic moorings had to be terminated due to heavy waves and strong winds.

HIRLAM/HIROMB OI-3/4DVAR

(SMHI) (70 % progress):

The International Polar Year 2007/2008 (IPY) provides new and unique opportunities to observe the atmosphere, hydrosphere and cryosphere at high latitudes in the two hemispheres. Ambitious surface- and space-based programs such as iAOOS will result in high quality observational databases that will help us to improve our understanding of geophysical processes at these latitudes. However, to reap the full benefit of often disparate randomly distributed observations of varying quality they have to be assimilated into comprehensive analyses of the state of the atmosphere and ocean. Numerical models are used to fill spatial and temporal gaps in the observations through a process called data assimilation. By employing variational techniques the observations can be used in their original form and their error characteristics can be determined and corrected for.

Northern Europe and particularly the Nordic countries are often affected by high impact weather systems originating in the Arctic. Furthermore the Baltic Sea and its ice conditions are crucial factors affecting the weather forecasts in these regions. For detailed fast-access operational weather forecasts on the national level, available global weather forecasts, primarily from ECMWF, have to be supplemented by frequent higher resolution limited area forecasts. In several Nordic countries the HIRLAM (High-Resolution Limited-Area Model) NWP (Numerical Weather Prediction) system is used for this purpose.

The HIROMB (High-Resolution Operational Model for the Baltic) ocean forecasting system is used operationally at SMHI (Swedish Meteorological and Hydrologiocal Institute) for analysis and forecasts of Sea Surface Temperatures (SST), sea ice conditions, sea levels and currents for the Baltic Sea and surrounding seas (Kattegat, Skagerrak).

Anticipating the IPY, the two numerical analysis/forecast systems HIRLAM and HIROMB, are combined into a coupled system. The aim is to prepare high resolution analyses and short-range forecasts of the weather and ice conditions in the Arctic. At the same time the coupled system will be a very valuable addition to SMHIs toolbox for regional forecasts in Scandinavia and the Baltic region.

Uncoupled experiments: the atmosphere component: HIRLAM

 

A period beginning in September 2005 has been selected for a ‘Damocles' data assimilation and forecast experiment over an area covering the Arctic Ocean. Lateral boundary conditions are extracted from the operational ECMWF (European Centre for Medium-Range Weather Forecasts) forecasts, and the observations are also extracted from the ECMWF archives. Only ‘conventional' observations (i.e. surface observations from land and ships/buoys, radiosonde ascents and aircraft data) were used in the experiments reported here. The atmospheric data assimilation was done with version 7.1 of the HIRLAM/HIRVDA 3-D variational data assimilation system. Analyses and 6-hour forecasts were produced every 6 hours from September 1st to September 30th 2005 (Expt. DAD). In addition, a two-month uncoupled experiment (Expt. DAC) was done, in which SST and ice conditions were prescribed from HIRLAM climatology. See Table 1 below.

Uncoupled experiments: the ocean component: HIROMB

 

Due to the longer time scales in the ocean, a spin-up period of many years would be required. However, by assimilating observed profiles of salinity and temperature from the World Ocean Atlas 2005, daily sea surface temperature (SST) maps from NCEP analyses, and daily sea ice concentration (SIC) maps (AMSR-E, the AQUA satellite) from University of Bremen, the spin-up period has been reduced to just one week, from 2005-08-25 to 2005-08-31; see "Expt O-SU" in table 1.

A one-month simulation was made for the whole of September 2005, forced by HIRLAM (Expt. DAD), but uncoupled (no feedback from HIROMB to HIRLAM); see "Expt. O-01" in table 1. This test was made to check the physics of the model without data assimilation. The main results from that study is that the model gives realistic results in terms of currents, SIC and SST.

The second uncoupled ocean-only simulation ("Expt. O-02") was also one month long, but included data assimilation of SIC and SST. Comparison with experiment O-01 reveals that the model physics has a cold bias, as SST in expt. O-01 often is much cooler than in expt. O-02. The reason for this is unknown. A possible future test is to try forcing HIROMB directly with the surface fluxes of heat and momentum from HIRLAM. So far we have only forced HIROMB with fluxes calculated by HIROMB itself using standard bulk formulations and large-scale atmospheric variables.

Table 1. Description of uncoupled and coupled experiments.

Initially, the coupling between the atmospheric and oceanic models is done as simple as possible. The atmospheric assimilation by HIRLAM is done in six-hourly cycles, one analysis every 6 hours followed by a six-hour forecast, which in turn serves as a ‘background' (first guess) for a subsequent analysis. Then HIROMB makes a corresponding six-hour forecast for the same time period, using HIRLAM as forcing. HIROMB then concludes by making an analysis of SST and SIC (as well as 3-D fields of salinity and temperature) which is used by HIRLAM in the surface analysis in the next cycle; see Figure 24.

Figure 24. Execution of the coupled model system.

Coupled experiments

 

The first coupled experiment has now been made. A two-month reanalysis was made for the period 2005-09-01 to 2005-10-31. Figure 25 shows the ice extent (defined as the area with at least 15% ice concentration) in the coupled experiment as compared with the two uncoupled ocean-only experiments as well as observations according to AMSR-E data. The ice extent is overestimated slightly, compared to observations. The reason is unknown, but it is possible that we loose some accuracy in the interpolation from high-resolution AMSR-E data (6 km resolution) to so-called super observations (22 km resolution) done before the data assimilation. This will be investigated in the future.

Figure 25. Ice extent according to satellite data (black), the two uncoupled ocean-only experiments O-01 and O-02 without and with data assimilation (red and gren, respectively), and the two-month coupled experiment DAC with data assimilation (blue).

The interaction between the HIRLAM atmospheric analyses/forecasts and the HIROMB Ocean/Ice analyses/forecasts are most pronounced near the surface. Preliminary comparisons of e.g. monthly mean analysis increments at 500 hPa show little impact of the coupling, but this is expected to change in longer forecasts. At the surface the net energy exchanges are notably different between the coupled assimilation and the uncoupled control assimilation. In the uncoupled assimilation HIRLAM default sea ice concentrations and sea surface temperatures were used. They are based on a (doubtful old) climatology and ECMWF operational data. The two maps in Figure 26 show the monthly mean net energy exchange (i.e. the sum of the solar radiation, the thermal radiation, the turbulent latent heat flux and the turbulent sensible heat flux). The direction of the fluxes follows the coordinate system of the atmospheric model: downward fluxes are positive (red) and upward fluxes negative (blue). In addition to the different sea ice covers in the coupled and uncoupled assimilations, there are also fairly large differences in the sea surface temperatures, as discussed below. We can note [Figure 27 (a) and (b)] that the ‘coupled' (DAC) sea ice field is not only smaller in extent but also includes larger areas (for instance near the pole itself) with ice concentrations less than 99%. This allows us to speculate that the increased sensible and latent heat fluxes seen in the coupled assimilation [in figure 7 (b)] is – at least partly – due to open water (lets call them ‘leads') within the ice fields. At the time of writing no further detailed studies of these flux differences have been carried out. Suffice it to say that the air-sea exchanges are very sensitive to the sea surface temperature and the sea ice concentration. This is obviously a trivial conclusion, but we find it very encouraging that our coupled system does respond in the expected way to a much more consistent coupled analysis of the atmosphere and ocean.

Figure 28 (a) shows the SIC bias, calculated as the monthly mean of the analysis increament at 12 Z for September 2005. The bias is up to about 10 percent units near the ice edge and of different signs in different areas. Figure 5 (b) shows corresponding RMS errors for SIC, of a magnitude of up to 10-20 percent units.

Figures 29 (a) and (b) show the corresponding bias and RMS errors for SST. It is again clear that SST has a cold bias. The reason for this is unknown at present.

 

 

 

 

 

 

 

 

 

 

 

 

Figure 26 Monthly mean net energy exchanges between air and ocean/ice in the (left) uncoupled and (right) coupled assimilation.

 

 

 

 

 

 

 

 

 

 

 

 

Figure 27: Monthly mean sea ice cover in the (left) uncoupled and (right) coupled assimilations for September 2005.

 

 

 

 

 

 

 

 

 

 

 

 

Figure 28. (left) Bias and (right) RMS error for sea ice concentration for September 2005, calculated from analyses at 12Z and the corresponding 6-hour forecasts valid at 12Z. The result is for the first coupled experiment (DAC).

 

 

 

 

 

 

 

 

 

 

 

 

Figure 29. (left) Bias and (right) RMS error for sea surface temperature for September 2005, calculated from analyses at 12Z and the corresponding 6-hour forecasts valid at 12Z. The result is for the first coupled experiment (DAC).

 

Remaining work

 

Although the coupled system seems to work, there is a great deal of work remaining before the planned, final longer reanalysis is started. Here is a short list of planned improvements:

• The ocean model requires fluxes of heat and momentum at the boundary. Today these fluxes are calculated from large-scale atmospheric variables using bulk formulae. An interesting alternative is to use the fluxes from HIRLAM directly. This will be tested in the future. It is however questionable whether those fluxes can be used without so-called flux corrections. Resources allowing, research into this problem may follow later.
• The data assimilation method used so far for the ocean analysis is the Method of Successive Corrections (SCM method). It is still our goal to replace this method by the Optimal Interpolation method instead (O-I method). However, as the correlation length scales are unknown, the O-I method will be less than optimal anyway. Further, when assimilating high-resolution 2-D fields like SST and SIC the result is not very dependent on the assimilation method being used. For more sparse observations the difference is probably larger.
• Apart from 2-D fields of SST and SIC, only very few observed profiles of salinity and temperature have been used so far. Collecting more observations is helpful for the longer simulations planned, to help the ocean model keep its stratification.
• The atmospheric assimilation of AMSU data over snow and ice needs to be improved.
• The bathymetry of the ocean needs improvements.
• Apart from several month-long test runs, we will finish off by making a longer reanalysis of a year or so. For this reanalysis, the best configuration of assimilation methods and data will be used. Time and resources allowing, it would be interesting to include also the year 2007.

NAOSIMDAS

(OASYS, FASTOPT) (30% progress):

The present task is constructing the variational data assimilation system NAOSIMDAS around the coupled ocean sea-ice model NAOSIM (Kauker et al, 2003) and analysed fields for two assimilation periods. Period I from 79 to 81 is characterised by high ice cover and period II (2006 to 2008) is part of the DAMOCLES period and most likely characterised by low ice cover.

NAOSIMDAS will rely on adjoint code of two components, the dynamical model and the cost function, which compares values simulated by the model to various observational data streams. The implementation of a preliminary cost function (including observational operators) with data from period I is described in delivery report 4.3-01 of month 12. The adjoint code of this cost function has been generated and verified (see deliverable report 4.3-04 of month 18). The adjoint for most of the dynamical model was generated, verified and first operated for an adjoint sensitivity study (see delivery report of 4.1-03).

Figure 30. Sea-ice data and uncertainties.

Sea-ice concentration uncertainties to be used in the cost function have been updated from a preliminary state to more sophisticated formulation. The new formulation is based on literature study and personal communication with remote sensing experts. The uncertainties of ice concentration data products are larger in summer than winter, larger at low concentrations than at high concentrations. The NASA Team (NT) algorithm used so far has a negative bias (underestimate of ice cover). We assume a linear dependence of absolute error from concentration (see figure 30).

The cost function has been extended by inclusion of continuous (i.e. very high spatial resolution) profiling data (e.g. CTD). The extended version also allows for inclusion of profiling float data, which we expect to be a major new data type for period II. We are currently preparing software for handling of formats of ARGOS float data and are awaiting observations for period II from DAMOCLES partners in WP1 and WP3. Such data are e.g. from ARGO floats, Sea-Gliders, Ice-tethered platforms (ITPs) or fixed profiling moorings (yoyo moorings). From the CORIOLIS data center ARGOS and CTD data have been gathered and included for the year 2006 as part of the assimilation period II (2006-2008). New data will be included as they are uploaded on the DAMOCLES external databases. We are e.g. awaiting observations for period II from DAMOCLES partners in WP1 and WP3.

 

 

 

 

 

 

 

 

Figure 31. ARGO data included into the cost function for 2006 form the 'Global' (left hand side) and 'Arctic' (right hand side) data sets.

For period I (1979-1981) ice draft data from two central Arctic submarine cruises for 1979 and 1981 have been received which were recently released at the NSIDC data center. They stem from analog recordings and are the only central Arctic thickness data for period I so far. We are currently working on the coding to include such ice draft data from ULS and submarine data into the cost function recalculating them into ice thickness estimates.

The data assimilation system NAOSIMDAS was set up and is being tested with the TLM of NAOSIM. First successful identical twin experiments have been conducted.

We are well on track towards our next deliverable: D4.3-5: First NAOSIMDAS assimilation experiment in low resolution completed and evaluated (month 28)

References

• Kauker, F, Gerdes, R, Karcher, M, Köberle, C, and Lieser, JL., Variability of Arctic and North Atlantic sea ice: A combined analysis of model results and observations from 1978 to 2001. J. Geophys. Res. 108(C6), 2003.
• Giering, R., Kaminski, T., "Recipes for Adjoint Code Construction", ACM Trans. Math. Software, Vol. 24, pp. 437-474, 1998.

Home : Research : Work packages : WP4: Integration and data assimiliation in large-scale modelling and forecasting : Task 4.3 Enhancement of observational data sets by data assimilation and analysis

Search:

Nov 10, 2008



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