2008 report
Progress: 45%
Predictability studies by SMHI (50%)
SMHI has started work on Arctic predictability. The goal is to distinguish between the internal and external Arctic variability. The tool to address this topic is SMHI's coupled ocean-ice-atmosphere model RCAO, which is run over the ERA40 period 1959-2000. The internal variability is that part of the total variability which is generated inside the Arctic model domain due to nonlinear interaction between ocean, ice and atmosphere components of the Arctic system. RCAO has been set up for an ensemble experiment with several runs, only differing by slightly varying initial conditions (the sea ice concentration at the pole point is initially reduced by 10 percent). The coupled model has currently been running for two times. This is of course not the final number of runs, however first results are interesting to discuss.
Fig. 22 shows the Arctic sea ice extent and thickness anomalies. The initial small differences grow larger. The sea ice extent shows differences on the 1-5 year timescale, while the longer scale variability is common to both coupled runs. The same statement can be applied to the mean sea ice thickness, which in addition shows slight biases on the 10-year timescale. The latter point needs to be further examined in terms of possibly varying ice export from the Arctic
Fig. 22: Arctic sea ice extent and mean sea ice thickness in two ensemble runs of the coupled ocean-ice-atmosphere model RCAO.
The two-dimensional view on internal and external variance (Fig. 23) gives additional information. The internal variance is assessed by the mean internal variability among the ensemble members (Fig.23a), which is the considered noise, not predictable by lateral boundary conditions. The external variance is given by the time-variance among the ensemble-averaged anomalies (Fig. 23b), which is interpreted as the variance due to external lateral forcing at the outer boundaries of the regional model. Generally both external and internal variability show strongest amplitudes at the Arctic coast. The internal part has a clear stronghold at the Siberian coasts. Mostly in coastal regions, the external amplitudes are up to two times stronger than the internal ones. Fig. 23c shows the ratio between external and internal variability. Away from the coast, external part is often larger by a factor of 1-1.4.
These preliminary results need to be substantiated by further runs. Later on, these results will be compared to met.no simulations. It is hypothesised that this kind of variability and predictability analysis will be even more interesting for transient climate runs, were thinning sea ice possibly allows for stronger internal amplitudes.
Fig. 23: (a, upper left) Summer sea ice mean internal variability, (b, upper right) summer sea ice mean external variability, (c, lower left) quotient between external and internal variability.
Predictability is defined here as the extent to which variability of Arctic variables can potentially be controlled by external forcing. Therefore the technical term “potential predictability” is identical with our external variance. The potential predictability is low when the internal (=internally generated) variability high. Predictability studies are useful to assess the feasibility of prediction systems. High sensitivity of internal processes to small disturbances is always limiting prediction efforts. In this study, we intend to approach the limits of Arctic potential predictability.
NWP reanalysis studies at SMHI (70%)
The SMHI NWP group contribution in this task is on preparing and setting up a system for coupled Arctic atmosphere-ice-ocean reanalysis. This work partly also takes place in Task 4.3, and more details are given in the progress report for Task 4.3.
Predictability studies at Met.no (20%)
Workpackage 4.2: (20%)
The planned long runs with the upgraded ORCM has been delayed due to the computer problems we have experienced this year. Earlier long simulations for the period 1970-1993 shows realistic simulations of MSLP and sea ice extend for the first 15 years, but with a small cold temperature bias over the Nordic Seas. In addition there where a strong drift in ocean salinity. This drift caused a reduced northward heat transport in the Nordic Seas in the later part of the simulations, and this gave an gradual cooling of the climate in contrast to observations. It is expected that fixes introduced to decrease the salinity drift, and to improve the ocean heat transport will improve the simulations. Also long (40 years) uncoupled atmosphere and ice-ocean runs have been made as a reference for evaluating the coupled runs that are planned.
Atmospheric assimilation studies at Met.no (20%)
The met.no work in WP4.2 on improving Arctic NWP by improving assimilation of satellite observations over sea ice has consisted of:
- started data collection of collocated satellite (AMSU-A) microwave radiance observations and NWP model output.
- started implementation of improved use of sea-ice sensitive lower atmospheric AMSU-A channels in the HIRLAM 3D-Var assimilation system
More specifically on the second point, one improvement based on work undertaken (partly) under WP1 by University of Bremen is under implementation. This approach takes into account the effect of the temperature profile within the ice on the efficient emitting temperature in each microwave channel. Furthermore, we work on the implementation of sea ice emissivity in control variable in the HIRLAM variational assimilation scheme.
As the implementation is not yet completed, no results are available yet. The first statistics to demonstrate improvements in the interpretation of the satellite observations are expected late 2007. Later on parallel impact studies quantifying the effect of the methods on weather forecasts will be performed.
Assimilation schemes NERSC (12%)
Month 13-18
During spring/summer 2007 high priority was been given to the planning and preparation of the acoustic tomography experiment within WP 3.1/8.2. Therefore, there has been low activity related to assimilation of tomography within 4.2, from month 13 to month 18. However, based on previous results of the activity in WP 4.1 we have come a step further on deciding which model system we should go for in further work in WP 4.2.
The previous high-resolution model grid covering the area in Figure 16 with a resolution of 2 km had 480x500 grid-points. With the 22 vertical layers the model became very computationally demanding, and therefore not feasible for EnKF assimilation (See D4.1-01), which requires 100 ensemble members to do assimilation. The first Fram Strait model system therefore has to go through significant changes to be used in the assimilation. Three options are under consideration
1. A three level model system consisting of: a clone of the new TOPAZ – 3 system (11 km), an intermediate model (6 km)) and an inner high resolution model (2-2.5 km). Multi member simulation.
2. A three level model system consisting of: a clone of the new TOPAZ – 3 system (11 km), an intermediate model (6 km)) and an inner high resolution model (2-2.5 km). Single member simulation.
3. A two level model system where the inner high resolution model is nested directly to the TOPAZ-3 clone. Multi-member simulation.
4. A one level model system. Perform direct assimilation into the 10 km TOPAZ-3 system. Multi-member simulation.
Comparing the three different model system options above, the option 1 is the most “accurate” approach. However, each model level has to be run with 100 ensemble members in order to do EnKF assimilation, and this makes this approach less attractive due to computation times.
Option 2 is to leave the EnKF assimilation approach, and run the three level model system as a single member simulation, and use the EnOI for assimilation of acoustic data. First of all this requires that we perform 2-3 years of simulations to establish statistics for the error covariance matrix which contain the seasonal variability. This is an approach that will not be used by NERSC within DAMOCLES.
Option 3 is to keep to the two level model system, as was used in the first Fram Strait model system. The difference from the first approach is that the model area has been reduced and that eventual effects due to the boundary conditions might influence a larger part of the model area. Accordingly, it will be essential that the 10 km TOPAZ-3 represents the Fram Strait current system in a good way and provide good boundary conditions for the inner model. Therefore, our decision here will rely on the results of the ongoing and planned validation of the TOPAZ 3 in the Fram Strait, within MERSEA. Validation depends on the data from the AWI/NPI mooring in the Fram Strait.
Depending on how efficient and good behaving the Fram Strait model is compared to TOPAZ 3, we will make the decision on if we want to assimilate data directly into TOPAZ-3 or into the Fram Strait model. Meanwhile, the assimilation scheme will be developed using the operational TOPAZ-3 simulations.
Month 19-24
The success of assimilation and model oriented inversions depend on a appropriately modelled observation matrix. Ingredients in the matrix are the oceanographic fields from the ocean model, data extraction routines, interpolation scheme (horizontal and vertical), and the acoustic model. In the last 6 months the main activity has been to test different interpolation schemes and to link the acoustic model to the interpolated modelled oceanographic fields. Oceanographic (temperature and salinity) sections along 78 50 N in the Fram Strait was produced from the TOPAZ-3 simulations. The original horizontal resolution is 11 km and each vertical profile has 25 layers. The depth of each layer and its thickness is varying with space and time and between ensemble members. The interpolation schemes were developed to produce fields with a horizontal resolution of 500 m and vertical resolution of 2 m from 0m to 1000 m and 20 m below 1000m. The vertical interpolation was then followed by a linear interpolation in the horizontal.
Several interpolation schemes were implemented and tested. Based on the numerical study we decided to go for linear interpolation in the horizontal and a special second order spline in the vertical. The study shows that the acoustic travel time calculations depend strongly on which scheme is selected. So, the result of assimilation will also rely on how good the interpolation scheme is. A, final decision will be made after comparing the modelled acoustic travel times with measurements.
After choosing the interpolation scheme the 100 ensemble members from TOPAZ-3 (model result after assimilation 25-07-2007) where run through the interpolation and used as input to the ray-tracing model to produce travel times. The purpose is to see how the statistical variability of the modelled and interpolated sound speed fields are propagated through the acoustic model. Analysis of these results has started, and a preliminary report has been produced which will be further refined and submitted as report D4.2-03 in month 30.
A 3 month internship for a French female master student, Camille Marini, from Centre de Géosciences Ecole des Mines de Paris, France was defined and carried out. The goal of the internship was to will establish the major components of the measurements matrix, required to refine the EnKF to incorporate direct acoustic measurements.
