This website uses cookies, which are small text files that are used to make websites work more effectively. In order to continue using this website, you will need to accept the use of cookies.

In order to study the movement of sea-ice, modellers usually use so-called elastic-viscous-plastic (EVP) algorithm. However, this algorithm is too slow to obtain realistic sea ice distribution, that resembles satellite images with leads and cracks, in acceptable time. In their recent study published in JAMES Nikolay Koldunov and colleagues from the Alfred Wegener Institute show that there is a way to reduce the time needed to calculate realistic sea ice distribution in high resolution model using previously developed modification of EVP algorithm.

The modelling of sea ice is quite different from modeling of the liquid ocean and the gaseous atmosphere. Often sea ice is considered as non-Newtonian fluid whose viscosity is non linearly depend on applied stress or force. If, for example, the ice is pushed together by winds, its thickness will change only after some threshold value and press ice ridges will be formed. The emergence and development of such structures is very difficult to represent in sea-ice models. One technical challenge is that a lot of short extra time steps are needed in the model in order to get the very rapid development of such structures reasonably and presentably done.

Most of the climate models use VP rheology (description of dynamical sea ice behaviour) that is solved with elastic-viscous-plastic (EVP) method. It is criticized in scientific literature for using many assumptions that to not find observational evidence. But climate models have to make a balance between the speed of the simulation (how much computer resources it requires) and realism of results. The VP rheology solved with EVP method still provide the best balance between model speed and realism in simulations with relatively low spatial resolution (100-25 km). For high resolution models (10-1 km) the classical EVP also become too slow if one would like to reproduce, say, leads in the sea ice. The reason is the higher the model resolution, the more sub-cycles of EVP solver is required to get to realistic solution.

Koldunov and colleagues for the first time applied two recently developed EVP versions (mEVP and aEVP) to high resolution Arctic Ocean model (4.5 km). Both methods can run with much smaller number of sub-cycles, therefore using less computational resources and making the time to model solution shorter. However, the results stay almost the same as if the model would run with classical EVP having a lot of sub-cycles. In aEVP method the information about local sea ice dynamics is analysed and calculations are further adjusted to increase the quality of the solution.

“Even though a distinction between both model approaches (mEVP and aEVP) cannot be made since advantages for just one are not obvious” explains Nikolay Koldunov who led the research “we found a 6-fold increase in the speed of the sea ice model calculation without affecting the quality of the simulated sea ice fields. This is very relevant for climate simulations because less time-consuming and more accurate predictions can lead to climate simulations with more realistic sea ice dynamics.”

Koldunov and colleagues also discuss that further improvements in relation to the possible optimization of the sea ice in FESOM2 should be explored in order to use different mesh partitioning for sea ice and ocean. Furthermore, effects on sea ice thermodynamics and changes in simulated temperature and salinity fields were left out in this study, but could be analysed in future studies.

The open –access research paper by Nikolay V. Koldunov, Sergey Danilov, Dmitry Sidorenko, Nils Hutter, Martin Losch, Helge Goessling, Natalja Rakowsky, Patrick Scholz, Dmitry Sein, Qiang Wang, and Thomas Jung is available at:

EVP diagram