Combining modeling and observations for improved sea ice predictions

Sea ice drift velocity averaged over the years 2008-2018 in the Arctic and Antarctic. The column show: (left) the experiment without data assimilation, (middle) the experiment with data assimilation with improved drift, (right) comparison data from the Ocean and Sea Ice Satellite Application Facility (OSISAF).

Sea ice is an important component of the Earth system because it strongly influences heat exchanges between the ocean and the atmosphere. Models are used to predict the state of the sea ice over time scales from days to years.

In this study, recently published in the Journal of Advances in Modeling Earth Systems, Longjiang Mu and his colleagues from the Alfred Wegener Institute Helmholtz Center for Polar and Marine Research developed and assessed a seamless sea ice prediction system with a focus on the so-called data assimilation component. Data assimilation is the methodology that combines models with real observational data. In the study the methodology is used to generate improved model fields which are then used to initialize the computation of model predictions. Other applications of data assimilation are the assessment of model error and the optimized representation of model processes.

A particularity of the model and data assimilation system is that a so-called coupled model was used. This model, the Alfred-Wegener-Institute Climate Model (AWI-CM) simulates the ocean and sea ice as well as the atmosphere and land surface. As such it simulates interactions between these components. The data assimilation applies a particular software, the Parallel Data Assimilation Framework (PDAF), which is directly connected to AWI-CM to provide the on-line data-assimilation functionality. Directly combining the model with the data assimilation reduces the computation time dramatically. In the study the data assimilation focused on the ocean and sea ice, by assimilating observation of the sea surface temperature and sea ice properties like thickness, concentration and drift velocity.

With assimilation the sea ice and the ocean circulation become more realistic as can be shown when comparing with independent, non-assimilated, observations. In general, the data-assimilation methodology is configured in a way that each observation type can influence all model variables. These effects can be assessed by studying single data types. For example, the sea ice drift velocities help to improve the representation of the sea ice thickness and the sea surface temperature observations improve the circulation at mid depth.

The study is an important step toward a fully-featured sea-ice prediction system. It further builds the sea-ice component of the assimilation system that is currently developed in the ESM-project. Here, also other observations of the ocean and the atmosphere will be assimilated to find good initial fields for prediction simulation, but also for the assessment of the model skill.

The open-access paper by Longjiang Mu, Lars Nerger, Qi Tang, Svetlana N. Losa, Dmitry Sidorenko, Qiang Wang, Tido Semmler, Lorenzo Zampieri, Martin Losch and Helge F. Goessling is available at