Southern Ocean sea ice reanalysis
(published: 8th November 2013, by F. Massonnet)
Sea ice in context: large seasonal variability and interannual fluctuations
Sea ice is a product of atmosphere and ocean interactions. It forms at high latitudes during the cold seasons as a result from the large thermodynamic imbalance between the ocean and the atmosphere. In spring and summer, the sea ice cover melts partially. This strong seasonality is illustrated here for the year 2007 with model results:
Figure: Arctic and Antarctic sea ice development in the year 2007 simulated by the Louvain-
la-Neuve sea ice model v3, LIM3. Arctic sea ice shrinks to its minimum extent in September
(February) in the Arctic (Antarctic), and expands to its maximum in March (September).
Note that because the seasons are reversed between the Northern and Southern Hemispheres, the Arctic and Antarctic sea ice developments are out of phase with each other by about 6 months.
Sea ice also exhibits fluctuations from year to year, and from decade to decade. Passive-microwave satellite observations available since 1979 revealed over the past 3 decades
- a significant decrease in Arctic sea ice extent ( ~ - 4.5% / decade in the annual mean)
- a significant increase in Antarctic sea ice extent ( ~ 1.5% / decade in the annual mean), but the positive trend is the result from both positive and negative regional contributions.
Observations of sea ice thickness are sparse in the polar regions. In the Arctic, several submarine cruises during the 80s and the 90s revealed a global thinning of the ice cover compatible with the observed decline in areal coverage. In the Antarctic, sea ice thickness was measured at the occasion of multiple ship campaigns. However, these valuable measurements lack sufficient sampling in time and space. Therefore, it is impossible to reconstruct the changes in Antarctic sea ice thickness from observations only. Such a reconstruction could be a valuable tool to get a better grasp of the complex Antarctic sea ice variability.
Figure: Antarctic sea ice is characterized, in a first simple view, by its area and its thickness (credit: NASA)
Why models can help
Climate models are numerical tools based on basic conservation laws. They solve the spatial and temporal evolution of prognostic variables on discretized grids. Such models are of course approximations of the actual world, but they have the advantage to output many variables, including sea ice thickness, with high spatial and temporal sampling. In Louvain-la-Neuve, at TECLIM, we maintain the development of the LIM (Louvain-la-Neuve sea ice model, more info here) which is the sea ice model used in several climate models. At TECLIM, LIM is also used for retrospective understanding of the recent sea ice changes.
Reconciling model and observations
Models are valuable tools but arguably not entirely free of biases due e.g. to their limited resolution and to our understanding and implementation of physical processes. On the other hand, observations are also valuable but incomplete: as mentioned earlier, the areal coverage of sea ice is continously monitored since the late 70s, but sea ice thickness data are much more limited in time and space. The challenge is to make a wise use of those two sources of information and their uncertainties in order to estimate the state of the system to be studied with best accuracy.
We implemented at UCL a sequential data assimilation technique, the ensemble Kalman filter (EnKF) into the ocean-sea ice model NEMO-LIM. The scheme works in two steps:
1. The model is run forward a large number (typically 25 to 100) of times by perturbing either its initial conditions, its uncertain parameters or the boundary conditions. With this approach, the goal is to sample the uncertainty of the model and to construct, based on the many model forecasts, an approximation of the true model error covariance matrix.
2. The forecasts are "analyzed", or updated with the use of observations. That is, the observations and their uncertainties are used as a constraint on the forecasts. Each of the prognostic model variables is updated: the correction is larger when the mismatch between the model and the observations is large.
Figure: A schematic description of ensemble methods for data assimilation
This two-step cycle continues: new model forecasts are generated from the previous analyses, etc. The ensemble of analyses constitutes a "reanalysis" for the system.
Reanalysis of Southern Ocean sea ice thickness
The above method has been implemented and tested in the ocean-sea ice model NEMO-LIM, with observations of sea ice concentration. We were able to show that the reanalysis has a lower bias in sea ice thickness than the model run without data assimilation. This motivated us to construct a map of sea ice thickness trends in the Southern Ocean:
Figure: linear trend (1980-2008) in Southern Ocean sea ice thickness.
The grey contours enclose significant trends at p=0.05. From Massonnet et al., 2013.
The map reports an overal increase in sea ice thickness but as a result of positive and negative trends. The changes are not negligible, since Antarctic sea ice thickness ranges between 0 and 1.5 m on average. The trends are compatible with observed trends in sea ice concentration. Both thermodynamical and dynamical processes are suspected to play a role in the spatial distribution of trends.
Other contributors: Martin Vancoppenolle, Pierre Mathiot, Thomas Lavergne, Chris König Beatty
F. Massonnet,P. Mathiot, T. Fichefet, H. Goosse, C. König Beatty, M. Vancoppenolle, T. Lavergne, 2013, A model reconstruction of the Antarctic sea ice thickness and volume changes over 1980-2008 using data assimilation, Ocean Modelling, 64 67-75, doi:10.1016/j.ocemod.2013.01.003.
P. Mathiot, C. König Beatty, T. Fichefet, H. Goosse, F. Massonnet, M. Vancoppenolle, 2012, Better constraints on the sea-ice state using global sea-ice data assimilation, Geosci. Model Dev., 5 1501-1515.