Thus synergetic use of observations and model through a data assimilation process ensures an optimal use of the available information both for monitoring and for forecasting. For operational purpose the observations need to be available in near-real-time and preferable at regular time intervals. We therefore propose to assimilate primarily satellite-based sea ice concentration, sea ice drift and SST. In addition we will explore the possibility of assimilating the salinity and temperature profiles from existing ARGO-buoys and the iAOOS glider-data from weather station Mike.  Other in-situ observations that will be obtained during IPY (e.g., from WP1) will be used as independent validation to the assimilation schemes.

The coupled ice-ocean model that will be used builds on the current semi-operational ice-ocean system for the Arctic at the Norwegian Meteorological Institute (met.no). In this way, the step from improvements in models, observations and assimilation methods, to a daily operating monitoring system is short. Today, the operational sea ice - ocean model runs with 20 km horizontal resolution covering the Arctic Ocean.(http://met.no/kyst_og_hav/northern_anim.html). Assimilation of OSISAF products are currently done with a univariate nudging technique (Albretsen and Burud, 2006). However, in a univariate method, only the assimilated field (e.g. SST) is updated in the procedure; all other parameters (e.g. velocity) are left to the model. As a result, the assimilation has large impact on the assimilated parameter but only a minor one on the others.  A multivariate assimilation scheme is therefore needed, where all the model fields are updated via cross-correlations contained in an error covariance matrix. We are currently implementing a so-called error reduced Kalman filter technique called SEIK (Singular Evolutive Interpolated Kalman filter) first introduced by Pham (1998). This method has been shown to work well still being relatively computationally efficient, and hence feasible to run in an operational context (Nerger, 2003; 2005).

In addition to the conventional multivariate assimilated scheme we also propose to implement a completely new approach to make optimal use of  SST observations. The SEIK filter uses empirical function to project the surface information down the water columns. However, in certain applications, a more dynamical projection method seems to be required, that is, one that correctly assigns a depth profile to surface features (e.g. Tang et al., 2004). We are exploring such a method (LaCasce and Mahadevan, 2005) which involves a modification of the surface quasigeostrophic (SQG) approximation, a construct used in simulating atmospheric motion in the tropopause. The method predicts not only subsurface temperature, but subsurface pressure and velocity as well. We have found it works remarkably well in tests involving model data and in situ oceanographic data (LaCasce and Mahadevan, 2005; Lapeyre and Klein, 2005). It also appears to be ideally suited for SST assimilation, although it has never been used in this way. We propose to do so, and to integrate the method with the SEIK filter.

A great advantage would of course be to directly assimilate in-situ profiles of the parameters  down the water columns. Until now, temperature and salinity profile measurements from about 30 drifting buoys connected to the ARGO program (www.argo.ucsd.edu) are available in near-real-time in our Arctic model region. The iAOOS glider observations planned in Task 1.2 will also be available in near-real-time and we will explore the effect of assimilating these profiles, together with the ones from ARGO, into the operational model system. During the first part of the period we will investigate the characteristics of these data, e.g., measurement accuracy, observation positions (density), observation times, data latency, etc., for use in the assimilation scheme. The next stage will be to build programs which extract "observations" from model results so as to mimic the real observations. Eventually, test runs will be performed to verify that the new data forms are assimilated correctly in the scheme, both separately and in combination with other data types.


Subtask 1: Multivariate assimilation of ice concentration, ice drift and eventually ice thickness.

1.    Use a multivariate assimilation technique, e.g. SEIK to assimilate ice concentration and ice velocity into the ice-ocean model.
2.    Test the system with ice thickness measurements that may be included in the monitoring system if near real-time observations are available.

Subtask 2: Test and implement the SST – dynamical projection method

1.    Examine the applicability in the Nordic Seas using fields from the operational model.
2.    Assimilate the dynamically-derived fields into the operational model.
3.    Compare the dynamical projection method with conventional (SEIK) multivariate assimilation of SST and validate against in-situ observations from WP1.

Subtask 3: Explore the effect of assimilating in-situ observations from gliders (WP1, Task 1.2) and ARGO-floats.

1.    Define the character of the data for use in the assimilation scheme.
2.    Test runs with simulated observations that mimic the real data.
3.    Assimilate the real data and study the impact by validating the results against independent in-situ observations (e.g., from WP1).