FORECASTING OF COMPRESSIVE ICE CONDITIONS

Mikko Lensu1, Jari Haapala1, Jonni Lehtiranta1, Patrick Eriksson1, Pentti Kujala2 Mikko Suominen2, Anders Mård2, Leena Vedenpää3, Tarmo Kõuts4, Madis-Jaak Lilover4

1Finnish Meteorological Institute, Helsinki, Finland2Aalto University, Espoo, Finland

3ILS Oy, Helsinki, Finland4Tallinn University of Technology, Tallinn, Estonia

ABSTRACT Compressive conditions increase the difficulty of ice navigation. Channels are closed, ship speed is reduced, the ship may get stuck, and in severe case ice buildup and hull damage may occur. Inclusion of compression into ice forecasts would greatly improve the efficiency of winter navigation as difficult areas could be avoided and need of icebreaker assistance assessed better. It is well known to ice navigators that zones of compression may generate, relieve and change location in hours. The physical background is found in ice stress propagation which can experience a complete change of pattern after a slight alteration in wind forcing or ice conditions. The SAFEWIN project approach to compression forecasting is described. It is based on an extension of the Finnish Meteorological Institute operative ice model HELMI. Two day forecasts are provided with 3 hour interval and the compression is expressed as a simple numeral from 0 to 4. The forecasts exhibit similar sensitivity for forcing and ice conditions as the observed compression. This makes however the forecasts sensitive to errors in the model parameters, initial conditions and forecasted forcing fields. A natural solution is to include expected variation for all of these and derive an ensemble of predictions. The forecasts are then presented in terms of probability of occurrence and provided in several formats added to the routine ice forecasts. The forecasts are delivered trough a web site and via the icebreaker network IBNet. For selected ferries and other ships on regular routes experimental ship specific forecasts are possible. The forecasts and presentations are validated and enhanced further using icegoing data and feedback from the end users.

INTRODUCTIONCompressive conditions increase the difficulty of navigation from what is expected from ice thickness, ridging and concentration only. A compressive event is characterised by local convergence driven by ice cover stresses which in the Baltic are mainly generated by wind forcing. The usual indication of compression is a closing channel. Ridging and other ice deformation may occur close by. The ship may also feel added resistance in a narrowing channel. Increasing resistance can stop the ship's progress and ice pileup against the hull may follow. In a severe case the hull can be damaged and the ship may heel. Compressive events may occur also in fast ice zone where the ship channel itself triggers convergence in a previously immobile ice cover.

To be forewarned of compression is high on the ice navigator's wish list. However, the phenomenon is notoriously elusive. A compressive condition can generate in a short time and relieve soon, but it can persist days and even weeks. It can be localised so that other ships around have no difficulties but also aggravate the conditions of a whole sea area. It is not

POAC’13Espoo, Finland

Proceedings of the 22nd International Conference onPort and Ocean Engineering under Arctic Conditions

June 9-13, 2013Espoo, Finland

simply related to wind speed, ice drift, closing ice cover or ongoing ridging. It is a general observation that the geometry of the ice cover, which includes also ship channels, has an important role in the generation of compression.

Systematic studies of compression and involving the whole chain from ice cover dynamics to the ship experiences have been lacking. Apart from the Baltic Sea the phenomenon has attracted persistent attention in Russia, where a qualitative coding of the compression severity has been used for a long (Heideman, 1996, Juurmaa et al. 1998), and the St. Lawrence Estuary (Senneville et al. 2007). Relationship between ice model parameters and ship performance and damage in the Baltic Sea is addressed in Pärn et al. (2007) and Kõuts et al. (2007). Compression is discussed also in Leisti et al. (2009). The ship-ice interaction phenomena during compressive situation are described in Riska et al. (1996) and modeled by Kubat et al. (2010). The summary report by Eriksson et al. (2009) deals with all relevant aspects.

The principal objective of SAFEWIN project 2000-2013 is to include compression to operative ice forecasts that are daily used by icebreakers and ice navigating commercial fleet. To attain this forecast model development is supported by extensive field campaigns addressing both ice cover state and ship performance. An important part is the collecting of ship reports from compressive situations, and over 1000 reports were obtained during winter 2010-2011 only (Leisti et al. 2011). Research on ship-ice interaction under compressive conditions has been conducted to connect the forecast model scale to the ship scale (Suominen and Kujala 2012). Here the development and validation work on Finnish Meteorological Institute ice forecast model HELMI is described.

THE FORECAST MODEL HELMIIce forecast models Dynamic-thermodynamic ice models are developed to describe dynamic movements of the ice cover and the changes in ice thickness and concentration. The models describe the ice cover in a continuum idealisation which applies in scales large enough in comparison to floe size. Dynamics originates from atmospheric and oceanic driving forces. Thickness and concentration are changed by thermodynamics and by also deformation processed driven by the dynamics, especially ridging.

Mathematically the model includes two types of equations. Momentum balance equation describes momentum fluxes between ice cover, water and atmosphere, and within the ice cover due to ice-ice interactions. The change in thickness and concentration are described by continuity equations. The thickness description may use average thickness, or several thickness and/or ice type categories. See Feltham (2008) for ice models in general.

The effect of ice-ice interactions enters the model equations in terms of sea ice rheology. The interactions are assumed to generate a stress state σ and the additional term σ⋅∇=intF , often called internal ice friction, is included to the momentum balance equation. Rheology is relationship ( )ψεσσ ,= between the stress state, the kinematic state in terms strain rate tensor, and variables ψ related to thickness and concentration. For low concentrations internal friction term is negligible and a free drift solution of model equations holds. As concentration increases floes start bumping and grinding with each other and after a certain threshold rafting and ridging processes set on. In a closed ice cover further convergence is possible only trough simultaneous rafting and ridging.

In free drift the energy driving the ice movement comes principally from wind forcing and is dissipated by water drag. The ice-ice interactions generate new energy dissipation mechanisms and the drift slows down. If total rate of dissipation from ice-ice interactions is

calculated as the sum E over all local deformation events it is possible to impose the condition

εσ :=E . (1)

The HELMI model The FMI operative ice forecast model is an implementation of HELMI, Helsinki Multicategory Ice model. The model is outlined here without the new features related to the compression variables which are described in the next chapter.

The ice thickness description in HELMI includes five level ice categories, and categories for rafted and ridged ice. The associated continuity equations include terms describing ice mass flux between these categories. The model rheology is viscous-plastic rheology with an elliptic yield curve of Hibler (1979). See Haapala et al. (2005) for the model equations.

Although other rheological models have been suggested, viscous-plastic rheology is most common one in operative models. The elliptic yield curve is constructed to seize the observation that the ice cover can resist high compressive forces in convergence but diverges for low tensile forces, while the situation is in between when shear strains are present. The onset of drift and the drift characteristics depend on a strength parameter P of the rheology. As the viscous-plastic rheology is in effect only for a moving ice cover, inert ice cover is described as being in a state of very slow viscous creep the effect of which in the modelled results is negligible.

There are two approaches to the strength parameter P. Hibler (1979) defined P as an increasing function of thickness and concentration. The HELMI model uses another approach suggested by Rothrock (1975). This applies the dissipation equation (1) to fix the value of P. Hibler's function has a virtue of simplicity but lacks physical foundation present in the dissipation approach. On the other hand, the left hand side of (1) depends on the understanding of dissipation in individual deformation events, most importantly in ridge formation events. However, in practice the strength parameter is a tuning parameter for which best value is chosen by comparing modelled ice drift to observations.Operative practiceIn operative use HELMI model is solved numerically in grid with the resolution of one minute in latitude and two minutes of longitude. This is roughly 1x1 nautical miles at 60° N. The lower bound of the grid is on 56.74° N. During the ice season 48-hour ice forecasts with 3 hour interval are updated every 6 hours.

The model is run on FMI supercomputer and takes atmospheric forcing from FMI weather forecast model HIRLAM runs made every 6 hours as well. HIRLAM border conditions are from ECMWF (European Centre of Medium-Range Weather Forecasts) global model. The atmospheric data is interpolated as interpolation as the resolution of the ice model is much higher. The model is forced in addition by sea surface temperature (SST) and ice edge location which are taken from daily hand-made ice charts. These combine satellite data and multiple observations into a best guess. The forecasts can be viewed on a web page. They also delivered to icebreakers via the IBnet terminal, which is a communication and information system combining environmental and ship data. The standard validation and tuning of the modelIce models are constructed to forecast ice drift, concentration and thickness. The thermodynamic part of HELMI is based on submodels which can be validated independently, and the main validation and tuning effort concerns the modelled ice drift fields. All thicknesses exceeding thermodynamically attainable values and the typical complex variation of sea ice thickness are due to processes related to ice cover strains. Thus if the model describes the ice drift well there are hopes that the thickness description is reliable.

The HELMI model has been developed and compared to observations in many projects during ten years and. The in daily operative practice and the comparisons with ice charts and imagery have shown that the model serves it purposes. To add to this the SAFEWIN project undertook validation exercises related to the tuning of the strength parameter. These used data from drift buoys deployed to the Bay of Bothnia, Gulf of Finland and Gulf of Riga. Both main aspects of drift, or the onset of the drift and the actual drift velocities of moving ice cover, were targeted.

The results for seven buoys tracked for three months February-April 2011 in the Bay of Bothnia are shown in Figure 1. The usual value of P of operative model runs was observed to be too small as the onset of drift occurred too frequently and also the drift speeds were too high. This value was based mainly on comparisons made in the Arctic Ocean. After some exercises the new Baltic value of P was fixed to be 2.5 times the previous value. This produced good agreement both with the onset and the average drift speed and has been adopted as a new operative value. There is some evidence that the strength parameter is sea area dependent and that the action of intense ship traffic makes the Gulf of Finland ice cover more mobile.

Figure 1. The comparison of modelled and observed drift speed for strength parameter 2.5 times the previous operative value. Inset shows the tracks of one drifter group in the southern Bay of Bothnia.

INCLUDING COMPRESSION TO THE FORECASTSThe physical background of compression phenomenonCompressive situations are generated by local stresses and by the strains induced by the stresses. The key to understanding of the phenomenon is the distribution of stress and stress concentrations within the ice cover. The stresses originate mainly from wind forcing but can propagate and accumulate over large distances. This phenomenon is little understood and no practicable methods to observe ice cover stresses except locally exist.

Stress distributions of granular materials have been studied by numerous laboratory experiments and simulations. These have revealed that stresses propagate along a branching network of stress chains with low-stress closed cells in between (for example, Majmudar and Behringer 2005). In a deforming medium the stress chains change locations and are created

and extinguished constantly. Sea ice is a granular medium as well and although the geometry is vastly more complicated than in the stress chain experiments the phenomenon can be assumed to present. This fits qualitatively well together with the observed localisation, intermittency and geometry dependency of the compression.

The characteristic scale of low-stress cells in granular materials experiments is about ten times the grain size. Stress field averaged over scales larger than this approximates continuity while in shorter scales the stress field is patently discontinuous. These observations can be used as a thumb rule for continuum ice models. They are thus in principle not capable of forecasting the stresses deterministically in scales shorter than ten characteristic floe diameters. The compression variableThe objective of an ice forecast model is to predict ice drift realistically. Rheology is a device to describe how ice-ice interaction affects drift and it does not seek to be a characterisation of the actual physical stress state of the ice cover. The validation of the model targets the internal friction force σ⋅∇=intF and not the stress σ which is in practice unobservable.

Moreover, the viscous-plastic rheology does not include elastic stresses which are central in the buildup and propagation of stresses in a static ice cover. Therefore the safest option is to use Fint to describe compression which is also adopted in HELMI. The success in the modelling of drift and of the amount of ridged ice indicates that the internal friction term has a robust correlation with the local dynamic intensity. In the granular materials analogy Fint

would be the average force exerted on a grain by its neighbours.The value of Fint lies mostly between 0 and 5 kN/m2. In winter 2010-2011 model data

there was an exponential drop in the probability of occurrence so that values larger than 5 are attained in about 5% of the cases and values larger than 10 in 0.5% of the cases. The unit pertains to the physics of idealised continuum model and should be linked to what is experienced by ships. Two alternatives are conceived here, the estimation of line loads and an empirical interpretation as navigational compression numeral. The HELMI operative forecast uses a compression numeral ranging from 0 to 4. If the numeral is defined as 0.4Fint, truncating too high values, numeral 4 is attained in 0.5% of the cases which together with the exponentiality is appealing. However, the final assessment of this must rely on the end users' experiences that are extensively collected in the SAFEWIN project. Downscaling problematicsThe numeral is sufficient for operative forecasting purposes but a line load argument is required if calculations on the actual forces exerted on ship hull are at focus. The argument is similar to that used to estimate aggregate strength of ice cover. The grid cell is thought of as single square shaped floe with 1 NM side length. If the internal friction force integrated over the floe area is assumed to become exerted against neighbouring cell. Then the 1 N/m2 in the model physical unit would correspond to 1858 N/m line load along the cell boundary and the stress value would be obtained by dividing with ice thickness. The highest model physical values, typically of the order of 10 N/m2, correspond then to stresses of the order of 40 kPa for 0.5 m thick ice. Similar values are observed by local stress measurements described in the next section.

It is understood that this argument depends on the scale, which was here chosen to be the model scale 1 NM without further physical justification. It can be hypothesised however that there is a scale L, related to the floe size, over which the argument is sound. The granular materials analogy suggests this is of the order of ten characteristic floe diameters, or the scale where continuum approximation begins to apply. Assume a ship traversing in homogenous ice conditions with compression loads attacking perpendicularly from the sides. If the ship is shorter than L the average line load against the hull shows traits of the random variation from stress chain geometry. A downscaling approach is required to estimate the total load against

hull or the probability to get stuck. On the other hand, ships longer than L would feel about the same average line load whatever the length and downscaling is relevant for along hull load variation.

This indicates that downscaling should proceed from line loads calculated by integrating modelled Fint over an area L×L. Little is known beyond this and the observational means to include L to operative forecasting are lacking. The ice navigation community's experience that certain ice cover characteristics and certain local features are prone to generate compressive situations indicates that the downscaling is an essential part of the problematics. The modelled compression values are insensitive to this aspect as they do not take into account floe size and other horisontal ice cover structure, and no such model enhancements that could be reliable enough for operative purposes are in sight either. Validation with dataThere were two approaches to the validation of the compression forecasts. Physical validation compared the occurrence and magnitude of the forecasted compression to the observed stress state of the ice cover. This is considered in this section. The other approach compared the experience of the ice going ships on the severity of compression, quantified in terms of simple numerals, with the modelled values and their physical units. This is addressed below in the section 'Ships observations of compression'.

The physical stress in ice cover can be locally measured by stressmeter arrays. A standard instrument is a biaxial stressmeter that resolves both principal components of the stress and allows the separation of thermally and dynamically induced stresses. Several stressmeters are required to quantify the spatial distribution of stress within a floe. However, there are presently no means to observe the stress propagation and stress distributions in floe systems with reasonable effort, that is, without hundreds of stressmeters.

Figure 2 shows a two-week Bay of Bothnia stress record from a biaxial stressmeter with typical stress peaks separated by stress-free periods. For more detailed analysis see the companion POAC'13 paper (Lensu et al, 2013a). This intermittency can be understood as a time domain counterpart of the assumed presence of stress chain s in the ice cover. Same features have been observed in longer Arctic records (Lensu et al, 2013b). The stress peaks are related to three basic cases: stresses during ice drift, stresses in static ice cover, and stress buildup as drift slows down in convergence (on 9th March). Also periods with drift but very low stress level are observed. As the wind speed during the low stress phases has been about the same as the average for the whole period, 10 m/s, these indicate that the stress carrying chains circumvent the instrumented floe for this period.

Figure 2. Two-week stress record from the Bay of Bothnia in 2011.

The modelled compression variation at some grid cell is very similar to the observed stress records with spikes and stress free periods, Figure 3a. The values refer to the average state of the 1NM cell while it is expected that within the cell there is further stress variation due to stress chains, manifesting as stress concentrations and stress free lacunas. Therefore no close correspondence between the point stress record and modelled compression is expected. Rather, the correspondence can be at times quite good like in Figure 3b both for magnitude and frequency of the spikes, while at other times no correspondence whatsoever is found.

Figure 3. a) A 40-day modelled stress record (line load) from the Bay of Bothnia in 2010b) A one day comparison (06/03) of observed and modelled stress (∗) for the data of Figure 2.

Sensitivity studies and ensemble approach to compression forecastingThe HELMI forecasts are the snapshots of modelled fields taken every three hours. For most model variables this is acceptable as the fields change rather smoothly and slowly. However, the internal friction term used to describe compression shows similarities with the stress distributions observed in granular material experiments and simulations. A mass of granular materials may deform as an approximation of continuous media but the stress chain locations change over short intervals. The same is observed for the modelled compression fields that change considerably already during the 6 minute timestep of the model runs.

More importantly, the compression is also sensitive to small changes in forcing and parameters. In Figure 4 reanalysis results from winter 2011-2012 are shown. The reanalysis runs use best possible atmospheric forcing, or the HIRLAM forecasts rectified using assimilated observations. The ice edge is not forced as in forecasts and the ice thickness fields rely entirely on the physics of the model. Nevertheless the reanalyses created a very faithful replica of the ice season. The reanalysis has been made using the new standard ice strength parameter value and two values that differ by ± 8% from this. This variation shows insignificantly in all standard fields, for example in ridged ice concentration although this is a cumulative result of all simulated ridging and ice drift phases from the beginning of the ice season. From the point of view of practicable validation efforts targeting ice thickness and drift all three cases are equivalent. However, the modelled compression fields are clearly of different disposition. What appears to be a compression-free east-west corridor on the left hand side is a high compression area on the right hand side while in the middle neither is observed. Similar results follow from slight variation of other parameters as well. The forecasted wind fields have also uncertainty that is much larger than is needed to change the compression fields significantly.

The conclusion is that deterministic forecasting of compression is not feasible in the time and length scales of the forecasts. A natural solution is to adopt an ensemble approach and provide probabilistic forecasts instead. This means that the model is run several times using different parameters values and/or forcing fields and the forecast is based on the statistics of the generated ensemble. The main products in terms of the compression numeral are then 1) the fields of average of expected compression ⟨n⟩ and, 2) the fields p(n) for the probability percentage that compression numeral exceeds value n, where n=1,2,3,4. Other choices exists, for example the average of ⟨n(p)⟩ for p % of the highest values.

The analogy with granular materials experiments suggests that the changes of the compression fields following from small variations of initial conditions are predominantly of random origin. Thus any variation that does not significantly change the thickness fields can be used to create an ensemble for the compression forecasts. On the other hand, the varying of the atmospheric forcing fields, or using ensemble forecasted input for these, is not needed unless motivated from the viewpoint of the whole model, that is, also thickness fields are affected. This reduces the need of supercomputing core hours that is often the bottleneck of ensemble simulations. The compression ensemble for the three hour time interval of the operative forecast contains 1) Maximally 12 parallel core runs varying the strength parameter within ±10%, and 2) for each parallell run the 30 compression fields stored at each 6 minute time step of the model. There are thus maximally 360 members in the ensemble and results refer to the expected conditions and occurrence probabilities during a three hour period.

The operative ensemble forecasting commenced in March 2013. Before this only deterministic forecasts and reanalyses were made and compared with data.

Figure 4. Upper panel shows snapshots of modelled compression fields (kN/m2) for 25 February 2011, using standard strength parameter value (middle), and 0.92 and 1.08 times the standard value (left and right). In the lower panel ridged ice concentration for the same day.

SHIP OBSERVATIONS OF COMPRESSIONThe principal objective of the SAFEWIN development effort is to provide compression information to ships. To be useful the forecasts should have bearing on the ice navigators' interpretation on the severity of the compressive conditions. This is often coded by simple numeral ranging from 0 to 4, sometimes from 0 to 3. The experience of severity depends not only on the magnitude of local stresses but also to the ice movements and the spatial and temporal persistence of the situation. Ship particulars, ice capability, navigational status and the local geometry of ice cover and ship channels effect as well. Thus the establishing of connection between forecasted compression values and the compression numerals is not entirely unproblematic. To approach this problem an extensive set of compression reports made by ice going vessel was collected. This was complemented with AIS-retrieved navigation data.

Most compression reports are from the winter 2010-11 when in total 1058 reports from both icebreakers and commercial ships were received. The reports include time, ship identification data, location and compression numeral n with range 1-4, and in most cases wind speed and direction. A minority of the reports include also ice information like ice drift speed and direction, ice thickness, concentration, and ridging numeral. The performance of selected ships was extended from the location of report with AIS data. The comparisons for 2010-11 used deterministically modelled fields as the ensemble simulations were not yet available.

Reports were obtained almost every day during a period from mid January to late April. The share of compression values {1,2,3,4} from the reports was {49,38,11,2} percent respectively. There is general agreement between the severity of period as measured by number of reports and by average modelled compression, Figure 5. The relationship between the modelled values and the numerals was less clear however. Contrary to expectations, the reported numeral did not correlate with wind speed as the overall mean wind speed in the

reports was 9.3 m/s while for the compression numerals 3 and 4 it was 9.6 m/s. The usual approach to the model/report comparison was to determine the modelled compression in some model grid cell block centred on the reporting ship, for example a block of 3x3 or 5x5 cells. The modelled values for certain reported numeral n then typically had a distribution with rather large variance. For the four numerals {1,2,3,4} the associated model value distributions were similar but had mean values increasing with n. This indicates that the correspondence between model physical units and the numeral used in the forecasts should be statistical rather than one-to-one.

The comparison of modelled compression with the reports and progress of individual ships proved inconclusive. In the light of the above results, and of the sensitivity revealed in later studies and exemplified in Figure 4, this is expected. There was overall agreement with the location of the reports and the daily average presence of modelled compression in 10 NM scale but the correspondence of the magnitudes was poor. Some agreement with ship speed reduction and the variation of along route modelled compression could be found in some cases, Figure 6, while for other cases the agreement was lacking completely. Presently the results corroborate from the ship point of view the irreducibly statistical nature of the compression phenomenon and need of ensemble approach to forecasting.

Figure 5. The daily data on compression cases reported by ships compared with compression from the forecast model. The period is ice season 2010-2011. Compression intensity is daily number of reported cases multiplied by average reported numeral. The model value is the daily average for the whole Baltic Sea ice cover.

Figure 6. Speed (SOG) and course (COG) of MT Tempera with forecasted ice compression (pressure) on 6th of March 2010.

DISCUSSIONThe results strongly support the view that the observations on stress propagation made in

granular materials research have bearing on sea ice dynamics and especially on the compression problem. Sea ice geometry is however much more complicated. To the vast size variation of floes adds thickness variation and ridge geometry and compressibility following from ice deformation. The floe geometry exhibits often fractal features with nested geometry and no well defined mean floe size. In midwinter the ice cover does not partition to clear floes and the granular state can be said to exist only during deformation phases, the grains corresponding to coherently moving ice areas. Thus the importing of granular materials research results otherwise than as qualitative analogies is not straightforward.

The internal friction term of the model shows sensitivity and randomness that is somehow analogical to the granular materials stress propagation as well. It remains to be studied whether these two phenomena have any real connection. The model does not generate any stress chain resembling features. The approximations inherent in the numerical solving of model equations must be kept in mind as a source of randomness. However, in providing ensemble averaged mean compression fields the model appears reliable. This is corroborated by the fact that the model predicts quite well the thickness and drift fields that do not change significantly within the ensemble generated for operative compression forecasting. The success of the ensemble approach in comparison with deterministic forecasts is still to be validated. For this the season 2010-2011 comparisons with ship data will be repeated.

The possibilities to improve ice models on fundamental physics level much beyond the present performance of HELMI appear limited as the continuum assumption cannot be bypassed. Although resolution could be increased the results might prove void if the added detail vanishes in the ensemble averaging. The adding of floe size and other geometric parameters, and thereby also some downscaling scheme, is mathematically possible but the links with dynamics are not similarly constrained by dynamics as for thickness. However, information on floe size or other geometric variation quantified from satellite imagery could be used as an external module for making downscaling inferences. It also appears that the ice cover strength depends on the size of floes or whatsoever entities that are interpreted as interacting grains during deformation. Thus what is likely to be next step of development for HELMI is further validation for different seasons and the use of context dependent strength parameter instead of a universal one. This could be accompanied by season and sea area dependent downscaling approach.

CONCLUSIONSThe compression forecasting approach of SAFEWIN project has been described. The model performance has been compared with geophysical and ship observations which, together with granular materials research analogies, have resulted to the adopting of ensemble operative forecasting approach. This opens up new vistas of systematic validation and enhancing activities. On fundamental level the forecast model HELMI appears to perform as well as it is possible for a continuum ice model. It remains to be seen whether possibilities to incorporate downscaling and thereby provide more accurate forecasts on ship performance emerge by importing results from granular materials research of otherwise.

ACKNOWLEDGEMENTSThis research was conducted within the EU-funded projects ’Safety of winter navigation in dynamic ice’ (contract SCP8-GA-2009-233884 - SAFEWIN). The partners in this project are Aalto University, Arctic and Antarctic Research Institute, Finnish Meteorological Institute, Finnish Transport Agency, ILS Oy, Stena Rederi AB, Swedish Maritime Administration, Swedish Meteorological and Hydrological Institute, Tallinn University of Technology and AS Tallink Group.

REFERENCESEriksson P., Haapala J., Heiler I., Leisti H., Riska K. and Vainio J., 2009: Ships in Compress-

ive Ice, Winter Navigation Research Board, Research Report No. 59, 42 p.Feltham, D.L., 2008. Sea Ice Rheology. Annual Review of Fluid Mechanics,Vol. 40, pp. 91-

112. Haapala, J., Lönnroth, N. and Stössel, A. 2005. A numerical study of open water formation in

sea ice, J. Geophys. Res., 110. C09011.Hibler, W.D., III, 1979. A dynamic thermodynamic sea ice model. J. Phys. Oceanogr. 9, pp.

815-846. Heideman, T., 1996. Influence of ice compression on feasible navigation on the Northern Sea

Route. INSROP working paper N:o 39-1996. Juurmaa K., Heideman T., Uuskallio A., Patey M. and Kosomaa L., 1998. Influence of Ice

Compression on Feasible Navigation on the Northern Sea Route Part II, INSROP Work-ing Paper, No. 110-1998.

Kõuts, T., Wang, K. and Leppäranta, M. (2007). On connection between mesoscale stress of geophysical sea ice models and local ship load. In: Proceedings of the 10th International Symposium on Practical Design of Ships and other Floating Structures.

Kubat I., Sayed M. and Collins A., 2010. Modelling of Pressured Ice Interaction with Ships, ICETECH 2010, Paper No. ICETECH10-138-R0, 6p.

Leisti, H., Kaups, K., Lehtiranta, J., Lindfors, M., Suominen, M., Lensu, M., Haapala, J., Riska, K. and Kõuts, T., 2011. Observations of ships in compressive ice, the 21st International Conference on Port and Ocean Engineering under Arctic Conditions, Montreal, Canada, July 2011.

Leisti, H., Riska, K., Heiler, I., Eriksson, P. and Haapala J., 2009. A method for observing compression in sea ice fields using IceCam. Cold Regions Science and Technology, Volume 59, Issue 1, October 2009, Pages 65-77.

Lensu, M., Suominen, M., Haapala, J., Külaots, R. and Elder, B., 2013. Measurements of pack ice stresses in the Baltic. Proceedings of POAC'13 (this volume).

Lensu, M. , Elder, B., Richter-Menger, J. and Haapala, J., 2013. Comparison of ice stress records in terms of extremal value analysis. Accepted for publication in Annals of Glaciology 54(62).

Majmudar, T.S. and Behringer, R.P., 2005. Contact force measurements and stress-induced anisotropy in granular materials. Nature 435, 1079-1082.

Pärn O., Haapala J., Kõuts T., Elken J. and Riska K., 2007. On the Relationship Between Sea Ice Deformation and Ship Damages in the Gulf of Finland in Winter 2003, Proc. Estonian Acad. Sci. Eng., 13, 3, 1-1, pp. 1-14.

Riska, K., Kujala, P., Goldstein, P., Danilenko, V. and Osipenko, N., 1996, Application of results from the research project “A ship in compressive ice” to ship operability. Proceedings of the 13th International Conference on Port and Ocean Engineering under Arctic Conditions.

Rothrock, D.A., 1975. The energetics of the plastic deformations of pack ice by ridging. J. Geophys. Res., 80, 4514-4519.

Senneville, S., Saucier, F., Wright, B. and Tseng, J., 2007. Numerical study of Ice Conditions, Ice Dynamics and Ice Pressure Events near the Gros Cacouna Site in the St. Lawrence Estuary (winters1996/97 through 2002/03). CGU HS Committee on River Ice Processes and the Environment, 14th Workshop on the Hydraulics of Ice Covered Rivers, Quebec City, June 19 - 22, 2007

Suominen, M. and Kujala, P.,2012. Ice Model Tests in Compressive Ice, the 21st IAHR International Symposium on Ice, Dalian, China, June 2012.