Merged Analysis and Forecasting
The amount of available and reliable Earth Observations have significantly increased in the last decades thanks to continuous progress in instrumentation and data acquisition for satellite and in-situ measurements. However, they remain still too low in number and in the range of observables to determine the Earth state completely. This is especially the case for sparse observed areas as the high latitudes (Arctic and Antarctic), which are the areas of interests of our Sea Ice Section.
Modeling techniques has been developed to complement and assist the interpretation of scientific samples and field observations. Here, we combine observations with modeling techniques to identify the leading processes determining the spatial and temporal variability of sea ice quantities, such as thickness, extent or concentration.
The modeling techniques we use and develop, range from Lagrangian tracking of individual sea ice floes backward and forward in time (Box 1: IceTrack), 1D modeling along observed (e.g. buoys) and reconstructed (e.g. IceTrack) Lagrangian tracks (Box 2: Icepack) to advanced data assimilation techniques for forecasting sea ice conditions (Box 3: Sea Ice Forecasting).
Our Sea Ice Physics section at AWI operates a Lagrangian sea ice tracking tool called IceTrack in order to examine inter annual variability and trends in sea ice pathways. The methodology was used in various studies and is capable of tracing sea ice backward or forward in time, It is based on multiple satellite-derived sea ice motion products. Along sea ice pathways, we extract information from various other data products like sea ice thickness, sea ice concentration, temperature, wind, pressure or water depth. Moreover, we calculate sea ice growth and decay from different thermodynamic models (see e.g. Box2: IcePack). For a detailed method description we refer to Krumpen et al. (2019)
The approach is used to
- predict potential pathways of sea ice, e.g. for scientific and logistic expedition planning like for MOSAiC
- identify source areas of sea ice
- assist the interpretation of scientific samples and field observations
- investigate the temporal variability of processes acting on the ice cover
Example application of how IceTrack can support campaign planning: Here IceTrack was used to investigate drift scenarios of the MOSAiC floe for the potential starting point 86°N / 130°E. The figure shows the drift trajectories for the years 2005 to 2017. The starting date is always 1st of October of the respective year. The colours used for the trajectories symbolise the month for the respective position.
Reference:
Krumpen, T., Belter, H.J., Boetius, A. et al. Arctic warming interrupts the Transpolar Drift and affects long-range transport of sea ice and ice-rafted matter. Sci Rep 9, 5459 (2019). https://doi.org/10.1038/s41598-019-41456-y
A novel sea ice model framework, based on the single-column sea ice model Icepack (CICE Consortium) is currently under development and validation at AWI in order to better conciliate observations and modeling. Typical local sea ice observations are usually available from drifting buoys, airborne tracks, or moored instruments (e.g. Upward Looking Sonar). Neither the temporal nor the spatial scales of these observations are currently resolved by 3D sea ice models which makes the data-model comparison inconsistent. The new model framework has been adapted to suit these requirements and to exploit the full strength of the extensive and multi-source observations like those currently collected during the ongoing MOSAiC campaign. The approach has been implemented along buoys where the location is tracked by GPS or combined with IceTrack when information about the floe trajectories are not available. Icepack eventually provides seasonal to inter annual along-track sea ice quantities coherently with instrumental measurements.
Left Panel: Geographical map of the South Central Laptev Sea showing the moored ULS location (black dot). The cloud of grey points corresponds to the ensemble of trajectories that were back-propagated using IceTrack. Each individual track corresponds to a floe that passes the ULS on a specific day during the studied period from October 2013 to September 2014. As an example the trajectory of the floe that reaches the ULS on 12/02/2014 is superimposed in a blue shade color scale.
Right top Panel: Daily median observed draft (orange dots) at the ULS location. Simulated draft along the trajectory of the floe that reaches the ULS on 12/02/2014 (blue shade color scale) and the draft at the end point of the trajectory on 12/02/2014 (blue triangle) are superimposed. The Eulerian simulation at the ULS location for the same period is also represented (black line).
Right bottom panel: Comparison between observed (orange dots) and simulated ULS draft (blue triangles). The simulated draft is reconstructed following the example of the right top panel for each individual day using each individual trajectory. Contrary to the Eulerian simulation the Lagrangian approach allows us to catch the drop in the draft mid March 2014 which is caused by a shift of the floe pathways advecting sea ice from a different source region to the ULS location.
The state of the Arctic climate system is rapidly changing. These changes are impacting ecosystems, coastal communities, and economic activities. High-quality predictions of the sea-ice conditions are of outstanding interest. In the Sea-Ice Outlook, for example, various international research groups are applying different approaches to predict the Arctic summer minimum sea ice extent in September from the beginning of the melting season in May/June on.
While the strongest greenhouse gas induced changes are currently observed in the Arctic, it is also the area of the largest natural variability leading to a very low theoretically predictability. However, nobody knows at the moment exactly how large or low the predictability is. The methods applied for sea ice predictions depend on the time-scales of the forecasts. For the very long time-scales (multi-decadal to century long) Earth System Models (coupled atmosphere-sea ice-ocean-land models) forced by greenhouse gas emissions are used. For shorter time-scales the quality of predictions depends much stronger on the initial state as for instance in weather prediction where a large network of atmospheric and land surface observations of different kinds is combined with an atmospheric model. With respect to the sea ice many of the data streams necessary for high-quality predictions are still in an infant state.
AWI is participating in the SIO since 2008. We are using a sea ice-ocean model (NAOSIM) forced by atmospheric surface forcing of the past to build up an ensemble of possible developments. The ensemble mean is used as the most probable evolution. However, this approach neglects all feedbacks of the sea ice-ocean system on the atmosphere. Since 2015 the initial state of the sea ice-ocean model is constrained by sea ice observations (data assimilation). The data streams used consist of AWI’s CryoSat-2 ice thickness, ice concentration, snow depth and sea-surface temperature from other sources. In 2019 the method has been refined using a version of the sea ice-ocean model with optimized model parameters (Sumata et al., 2019a and 2019b). This increased the quality of the forecasts considerably. The system has been recently applied as well to forecast the sea ice conditions at the start of the MOSAiC expedition.
The figure shows a forecast of the likelihood of encountering an ice concentration of 15 % or higher in the September mean for 2018 (left) and 2019 (right). The forecast is initialized in early July, i.e. the lead time is about 3 months. The predicted sea ice extent in mill. km2 together with the ensemble spread and the observed mean September sea ice extent from NSIDC (in brackets) are shown at the top of each panel. The observed isoline with a concentration of 15 % (OSI SAF) is marked in magenta. For 2018 and 2019 the forecasted sea ice extent is consistent to the observed sea ice extent (difference within one standard deviation) and also the spatial pattern of the area within the 15 % concentration isoline of the ensemble prediction and the observation is very similar.
References:
Sumata, H., Kauker, F., Karcher, M. und Gerdes, R., 2019a: Simultaneous Parameter Optimization of an Arctic Sea Ice–Ocean Model by a Genetic Algorithm, doi.org/10.1175/MWR-D-18-0360.1
Sumata, H., Kauker, F., Karcher, M. und Gerdes, R., 2019b: Covariance of Optimal Parameters of an Arctic Sea Ice–Ocean Model, doi.org/10.1175/MWR-D-18-0375.1