LONG-TERM STATISTICS Of STORMS IN THE BALTIC, BARENTS AND wHITE SEAS AND THEIR fUTURE CLIMATE PROjECTIONS

The numerical model simulations of storm activity in the White, Baltic and Barents Seas were analyzed for the period from 1979 to 2015. In this paper the storm number of these seas was calculated. The connections of wind wave climate with indecies of large-scale atmospheric circulation such as NAO, AO and SCAND were estimated. Also, the future changes of wind wave climate were analysed. KEY wORDS: wind waves, storm, climate change, SWAN, reanalysis, Baltic Sea, Barents Sea, White Sea, significant wave height, trend, storminess, model, future change, NAO, AO, SCAND, large-scale atmospheric circulation CITATION: Stanislav Myslenkov, Alisa Medvedeva, Victor Arkhipkin, Margarita Markina, Galina Surkova, Aleksey Krylov, Sergey Dobrolyubov, Sergey Zilitinkevich, Peter Koltermann (2018) Long-term Statistics of Storms in the Baltic, Barents and White Seas and Their Future Climate Projections. Geography, Environment, Sustainability, Vol.11, No 1, p. 93-112 DOI-10.24057/2071-9388-2018-11-1-93-112


INTRODUCTION
Wind waves are an important component of climate system and study of their interannual variability helps to understand current climate changes and to assess wave impact in the future. The study of the storminess in the Baltic, Barents and White Seas have great importance for shipping, constructing on coasts and shelf, oil and gas field development. In this paper, a new approach estimating climate changes has been implemented. We counted a number of storms using significant wave height threshold in three neighbor seas and estimated the connection of storminess with indices of large-scale atmospheric circulation.
The wave conditions in the Baltic Sea were investigated in several studies (Kriezi and Broman, 2008;Lopatukhin et al. 2006;Medvedeva et al. 2015;Sommere, 2008;Zaitseva-Pärnaste, 2009). Different wave models and wind forcing have been used in these papers. The results of wave simulations have been estimated by comparing with buoy measurements and wave sensors (Kriezi and Broman, 2008;Medvedeva et al. 2016). There are many interesting papers about wave hindcast and climate in the Barents and White Seas (Arkhipkin et al. 2015;Korablina et al. 2016;Lopatukhin et al. 2003;Myslenkov et al. 2016;Reistad et al. 2011). Stopa et al. (2016) presented wave climate and hindcast based on altimeter data set and investigation of wave trend in the Arctic region from 1992 to 2014. The decrease of the sea ice extent in the Arctic Ocean over the last years is described in Mokhov (2013). The altimeter data and model results show that the reduction of the sea ice coverage causes a growth of wave heights instead of the increasing wind speeds. However, regional trends are influenced by largescale interannual climate oscillations like the North Atlantic Oscillation (NAO) and Pacific Decadal Oscillation (Stopa et al. 2016).
The wind and wave climate in the Arctic region based on altimeter measurements were presented by Liu et al. (2016). Trend analysis shows a clear spatial (regional) and temporal (interannual) variability in wave height and wind speed. Wave heights in the Chukchi, Beaufort (near northern Alaska) and Laptev Seas increase at 0.1-0.3 m per decade. These trends have been found statistically significant at the 90% level. The trends of wave heights in the Greenland and Barents Seas, on the contrary, are weak and not statistically significant. In the Barents and Kara Seas, winds and waves increased between 1996 and 2006. Largescale atmospheric circulation variations such as the Arctic Oscillation (AO) and the Arctic dipole anomaly have a clear impact on the variation of winds and waves in the Atlantic sector (Liu et al. 2016). Wang (2001) investigated wave heights in the Northern hemisphere and related atmospheric circulation regimes.
In order to take adequate precautions and to reduce risks and damages from the storm the information about the time of occurrence and magnitude of such events is required. This sort of research has been done for other basins of the World Ocean e.g. North Atlantic Rusu et al. 2015). There are a limited number of studies devoted to storm number, their interannual variability and their connection with global atmospheric circulation in the Baltic, Barents and White Seas, so these subject still remains challenging (Korablina et al. 2016;Medvedeva et al. 2015). Our paper focuses mainly on the storm statistics in the Baltic, Barents and White Seas and their connection with large-scale atmosphere circulation indices. The aim of this paper was to compare storm interannual variability in these three neighbor seas and to reveal common and different from each other features and also to assess the connection of storms with indices of largescale atmospheric circulation.

MATERIALS AND METHODS
In order to estimate decadal and interannual changes of the wind wave fields the SWAN, short from Simulating WAves Nearshore, (Booij et al. 1999) and WaveWatch III (Tolman 2009) numerical wave models were used. These models are state-of-the-art and are widely applied for reconstruction of wave fields and such parameters as significant wave heights (in this paper it was considered as mean of 1/3 of the highest waves), periods, lengths, swell heights and energy transport with different spatial and temporal resolutions by solving the energy balance equation (1) in spectral dimensions Reistad et al. 2011;Stopa et al. 2016).
For the Baltic and the White Seas the third-generation spectral wind wave model SWAN (version 41.01) has been implemented in order to obtain wind wave parameters. As a wind forcing we used 10-m wind from Climate Forecast System Reanalysis from the National Centers for Environmental Prediction (NCEP/CFSR) with a spatial resolution 0.3°×0.3° and a time step 1 hour (Saha et. al. 2010) for the period from 1979 to 2010. Starting from 2011, we used NCEP/CFSv2 (Saha et. al. 2014), which is the extension of NCEP/CFSR; it has a spatial resolution ~ 0.205×0.204 and a time step of 1 hour. The accuracy of the obtained wave parameters is high and it has been estimated by using measurements of the wave parameters (Medvedeva et al. 2016). Accuracy of the model was estimated for the same period, basin and data as in present paper, see details in (Medvedeva et al. 2016): in average R was 0.96, Bias 0.05, RMSE 0.29 and scatter index 0.18.
For the Barents Sea, the spectral model WaveWatch III version 4.18 has been implemented with an unstructured grid covering the North Atlantic basin and Arctic seas (Fig. 1). "ST1" parametrization has been used for energy input and dissipation. This scheme is based on the same equations that the SWAN configuration used in this study. "DIA" scheme has been implemented for nonlinear wave interactions (Hasselmann and Hasselmann, 1985). In addition, WaveWatch allows involving ice coverage. "IC0" scheme has been used for wave energy attenuation in the ice where wave energy reduces exponentially in grid points with ice concentrations between 25% and 75%, otherwise grid points are considered to be open water (<25%) or land (>75%). In WaveWatch III simulations in the Barents Sea we used wind forcing fields and ice concentration from NCEP/CFSR reanalysis from 1979 to 2010 (Saha et al. 2010). This model showed a good agreement with measurements for these regions (Stopa et al. 2016;Tolman 2009). In the White Sea storm statistics was calculated only for days without ice fields. For the Baltic Sea, we didn't exclude any data because Baltic Proper and South-eastern basin usually are not covered by ice.
Calculations have been performed using unstructured grids for the Barents and the White Seas with a spatial resolution 10−20 km in the central and open parts of the seas and 200−500 m in the coastal zones. The unstructured grid for the Barents Sea included the North Atlantic region from the Equator to the North Pole with a spatial resolution ~1° (Fig. 1). The grid for the Baltic Sea was rectangular and had a resolution 0.05°×0.05°. The computational grid was created on the basis of the General Bathymetric Chart of the Oceans (GEBCO) with a spatial resolution of one nautical mile for the deep sea. Data from high-resolution navigation maps were used for the bathymetry in coastal zones.

The model description
The SWAN model is a third-generation spectral wind wave model developed by Delft University of Technology in the Netherlands (Booij et al. 1999) which is traditionally applied for shallow water areas. WaveWatch III has been developed in many research groups, predominantly by the United States' National Oceanic and Atmospheric Administration (NOAA). Based on the same energy balance equation, WaveWatch III is more developed and has more different formulations for the wind input and the whitecapping. It incorporates state-of-the-art formulations for the deep water processes of wave generation, dissipation and the quadruplet wave-wave interactions used by the WAM model (Komen et al. 1994).
The processes included are wind input, whitecapping, bottom friction, depthinduced wave breaking, dissipation due to vegetation, mud or turbulence, obstacle transmission, nonlinear wave-wave interactions (quadruplets and triads) and wave-induced set-up (2).
All information about the sea surface is contained in the wave variance spectrum or energy density E(σ, θ), distributing wave energy over (radian) frequencies σ (as observed in a frame of reference moving with current velocity) and propagation directions θ (the direction normal to the wave crest of each spectral component). The action density is defined as N = E/σ and is conserved during propagation along its wave characteristic in the presence of ambient current, whereas energy density E is not.
The evolution of the action density N is governed by the action balance equation, which reads (Komen et al. 1994): The left-hand side is the kinematic part of this equation. The second term denotes the propagation of wave energy in twodimensional geographical ~x-space, with the group velocity ~c g = ∂σ/∂~k following from the dispersion relation σ 2 = g| ~k|tanh (|~k|d) where ~k is the wave number vector and d is the water depth. The third term represents the effect of shifting of the radian frequency due to variations in depth and mean currents. The fourth term represents depth-induced and current-induced refraction.
The quantities C σ and C θ are the propagation velocities in spectral space (σ, θ). The right-hand side contains S tot , which is the non-conservative source/ sink term that represents all physical processes, which generate, dissipate, or redistribute wave energy. In shallow water, six processes contribute to S tot : These terms denote, respectively, wave growth by the wind, nonlinear transfer of wave energy through three-wave and four-wave interactions and wave decay due to whitecapping, bottom friction, and depth-induced wave breaking.

The model output and storm count
The significant wave heights, periods and wavelengths were reconstructed with wave models for the period from 1979  All situations when Hs exceeded the chosen threshold (from 2 to 10 meters) were considered as a storm. The number of storm situations with different significant wave heights was calculated for every year from 1979 to 2010 and respectively 2015.
For example, if Hs in one grid node is higher than 4 meters then it was considered as the start of a storm with criteria Hs ≥ 4 m. An event is considered to be finished when the Hs in all nodes was less than the chosen threshold. When the Hs reached the level of 4 meters next time it was considered as another next storm event.
A storm with Hs ≥ 10 meters is taken into account for each other of the lower criteria from 2 to 10 m and it was included in all these selections. This method of the storm count has some inaccuracy, firstly, when two events happen directly one after another and, secondly, when in the sea under investigation there are two storms in different parts of the sea. Nevertheless, these cases are not very frequent.

RESULTS AND DISCUSSION Storm number and trends
In the Baltic Sea during 37 years the number of storms with Hs ≥ 2 m amounts to 2559, approximately 70 per year, with criteria 3 m -1285, 4 m -1107, 5 m -649.
These results indicate that about a half of all storm situations have a significant wave height of more than three meters (Fig. 2).
Typical periods of intensification and relaxation of wind waves are 10−12 years for the Baltic Sea, see Soomere and Räämet (2014). According to the obtained running average of storm number, there is a 10-year period of intensification/weakening of the storm activity. For various parts of the Baltic Sea, there is a discrepancy between the trends of the ten-year increase or decrease. Notably, that the rapid changes of intensification or weakening of wave activity can happen during one decade (Soomere et al. 2008;Broman et al. 2006 (Fig. 2). As for storms with Hs ≥ 3 m, they occurred only 5−6 times per year. Interannual variability of the number of storms in the White Sea is less determined than in the Barents, but for Hs ≥ 2 m there is a maximum in 1995 and a minimum in 1999. Since 1999, the storm number increases, but positive trend in storminess isn't significant. For Hs ≥ 3 m, there are two maxima in 1986 and 1994, two minima in 1985 and 1999.
In the White Sea, the following features of the wave climate are observed: the maximum of number of storms in the autumn and winter months, a significant spatial and temporal heterogeneity of the properties of wind wave fields, where each part of the sea such as Onega Bay, Basin, Gorlo, and Voronka has their own determined wave mode.
In addition, the relations of storm number between different Seas was studied. It is interesting that the highest significant correlation (0.56) was discovered between the storm number, discussed above, of the Barents storms with Hs ≥ 7 m and Baltic storms with Hs ≥ 4 m, which have statistically significant negative trend.
These events occur about 20 times per a year. For others Baltic storms with Hs ≥ 3, 5 m and Barents events with Hs ≥ 5, 6, 8 m R was also approximately 0.5. We can make a supposition that it reflects the connection of these processes and its common origin. For more severe Barents storms with upper thresholds 9 and 10 m the link R = 0.52 m is observed with the White Sea storms with Hs ≥ 3 m level (about 5 times per a year), however this connection is reflected only on scales of greater wave heights and isn't observed with the Baltic Sea. All other correlation coefficients are small. So high correlation is not observed between White and Baltic Seas at all. And for events with Hs ≥ 2 m it was not revealed too.
The relations of storm number with atmospheric indices. In order to study the connection with the global atmospheric circulation, 3 indices North-Atlantic Oscillation (NAO), Arctic Oscillation (AO), Scandinavian Index (SCAND) have been considered. The correlation coefficient (R) was calculated between these indices and the max Hs, the results are presented in table 1.
We will regard as the state of the atmosphere with a positive value of the NAO index as the positive phase of the oscillation and when the value is below zero -as negative (http://www.cpc.ncep. noaa.gov/ 2017). In the positive phase of the oscillation, the Icelandic minimum and the Azores maximum are well developed, the pressure gradients between them are increased, the zonal circulation is strengthened. In the negative phase, there is a weakening of the zonal transport and an intensification of the meridional processes.
The AO is a large-scale mode of climate variability also referred to as the Northern Hemisphere annular mode (http://www.

Fig. 2. The number of storms with the different Hs for the Barents (a), Baltic (b) and White Seas (c)
masses and increased storminess into the mid-latitudes.
The Scandinavia pattern SCAND consists of a primary circulation center over Scandinavia, with weaker centers of opposite sign over Western Europe and eastern Russia/ western Mongolia. The Scandinavia pattern has been previously referred to as the Eurasia-1 pattern by (Barnston and Livezey 1987). The positive phase of this pattern is associated with positive height anomalies, sometimes reflecting major blocking anticyclones, over Scandinavia and western Russia, while the negative phase of the pattern is associated with negative height anomalies in these regions. The positive phase of the Scandinavia pattern is associated with below-average temperatures across central Russia and also over western Europe. It is also associated with aboveaverage precipitation across central and southern Europe and below-average precipitation across Scandinavia.
Correlation analysis of the mean annual number of the storm and mean annual indices of atmospheric circulation is presented in table 1. It showed that connection of storm number in the Barents Sea R about 0.5 is only with AO, for Hs ≥ 5 m (0.45) and Hs ≥ 7 m (0.49). However, if we consider not annual but winter averaged monthly index AO (DJFM) and monthly values of storm number, then, in the Barents Sea, the connections between a number of storms and AO on decadal scales is reflected better. The highest correlation coefficient between AO index (averaged from December to March) R (0.6) was obtained for storms with Hs ≥ 7 m (Fig. 3a), and for Hs ≥ 8 m with AO it amounts to 0.57. With SCAND pattern the link isn't observed.
For the White Sea, the relations between a number of storms and AO, NAO indices are weak (Fig. 3c, d). Maximum R = 0.26 and other coefficients are less. The liaison between storm number in the Barents and White Seas and SCAND pattern is weak and it was not revealed.
For the Baltic Sea the connection of storm number is mostly pronounced with SCAND. The highest correlation is -0.59 for Hs ≥ 2 m (Table 1). Also R with AO reaches 0.45-0.49 for different thresholds. Summarizing this section, we can say that the connection with indices of global atmospheric circulation is mostly developed for the Baltic Sea with SCAND pattern, less with AO. Secondly, it is reflected in the Barents Sea with AO. For the White Sea such link was not revealed.  (a, b), White (c, d) and Baltic Seas (e, f). Indices are presented by running average with 13 months values wave heights are 1) a reduction of sea ice in northern parts of the sea and 2) an increase in the wind speed (Hünicke et al. 2015). Both these reasons should lead to a spatially inhomogeneous increase in the wave heights, first of all in seasonally ice-covered northern part of the Sea and along the eastern segments of the basin where the predominant south-westerly and north-north-westerly winds usually create the severest wave conditions. Analysis of Kudryavtseva and Soomere (2017) reveals an unexpected strong meridional pattern of changes: the wave heights have increased in the western offshore of the sea and have decreased (or exhibit no changes) along the eastern nearshore. It is, therefore, unlikely that a discernable increase in the wind speed has occurred in this region. This among other things means that a greater level of storms and swells (Bertin et al. 2013) may only characterize some parts of the North Atlantic. This is consistent with the conclusion that the basin-wide average geostrophic wind speed has not increased over the entire Baltic Sea (Soomere and Räämet, 2014).

Fig. 3. Connections between indices AO, NAO and the number of storms in the Barents
In this paper the entire Baltic Sea was divided into 5 areas (Fig. 5): the South-Eastern Baltic (I), the Gulf of Bothnia (II), the Gulf of Finland (III), the Gulf of Riga (IV) and the Baltic Proper (V) (Fig. 5) and the maximum Hs was identified for each part for every month (2200 values in all). As maximum Hs, we considered maximum value registered in this area for a selected period (for example the maximum value of Hs in one node in the Gulf of Finland in January 1990). Then for every month of this period from January to December, the For the Baltic Sea in the cold season, the field variability connected to the NAO is most pronounced. If we take into account only the period from December to March (with R ≥ 0.5), the connection of max Hs is more significant with AO, slightly less significant with NAO and negative (-0.4; -0.5) with SCAND (Table 2).
We observe a positive high correlation coefficient of ≥ 0.5 with the NAO index for five months: January, February, March and April, and for the Baltic Proper for November. For all other months, it has a low value close to the zero or negative. The highest value of R = 0.7 was observed for December in the Gulf of Finland (Fig. 6, Table 2). It is evident from the figures 3, 4, that every increase of the Hs in the Gulf of Finland corresponds to the positive phase of the NAO and AO indices. It is obvious that with positive NAO the number of deep cyclones over the North Atlantic region increases and maximum Hs increase too. Negative NAO phase in most cases coincides with the Hs decrease.
The AO index R ≥ 0.5 was identified for the winter period (plus April), the maximum value equals 0.64 for the Southeastern part of the Baltic Sea for January. In the springautumn period, the link is not obvious.
All coefficients R with the SCAND are negative. Notably, for January, February, April, and September with R ≤ -0.5, i.e. the Hs increases with the negative phase of the index. The highest values of R -0.67, -0.65 are found in the Southeastern part of the Baltic Sea. From the geographical perspective the highest R was noted in the Gulf of Finland (Fig. 6, Table 2) as with NAO, so with AO. For this area, the maximum value for NAO corresponds to December (R = 0.7) and for AO in March (0.68). Thus, with such complex configuration of the Baltic Sea, the NAO and AO indices have the most influence on the Gulf of Finland and secondly on the Baltic Proper. The lowest R coefficients were observed for the Gulf of Bothnia. It is worth noting, that in the last decade from 2005 to 2015 the storm number with Hs ≥ 2, 3, 4 m increases in the entrance of the Gulf of Finland. This conclusion is based on obtained results and on analysis of maps of maximum Hs distribution (constructed for every month for every year). In addition, it propagates deeper to the east, so it reflects the displacement of the trajectories of cyclones, which have moved 5° to the north. It changes the length of the wave fetch and promotes wave penetration into the Gulf of Finland.
As for storm number in separate parts of the sea (Table 3), the highest R is noted in the Baltic Proper with AO for Hs ≥ 3 m and R equals 0.62. For storm events with Hs ≥ 4 m, the situation is the same: the correlation for the Baltic Proper is stronger 0.54.
Summarizing, the connection between significant wave height and indices is stronger in the Baltic Proper and it was observed in more months than in any other part of the sea, th Baltic Proper is on the second place. If we consider the absolute value of correlation, in the Gulf of Finland R amounts 0.7 in the December.
As for seasonal variability of correlationthe connection is more determined from November to March with a maximum in the February.
If we take into account, storm number separately calculated for parts of the Baltic then in the Baltic Proper R is higher.

ENVIRONMENT future changes
The climate projection of weather pattern accompanying extreme winds over the Barents and Baltic Seas is carried out with the database of CMIP5 models ensemble runs (RCP8.5 scenario), see Moss et al. 2008;Taylor et al. 2012. According to this scenario, the global surface air temperature will be 3.5-4°C higher than in .
The key idea relies upon the "environment -to circulation" method (Huth et al. 2008). It is based on the assumption that extreme weather phenomena (local or mesoscale) are connected through physical mechanisms with large-scale (synoptic) events. Then it is possible to make projections indirectly, studying configuration and intensity of sea level atmospheric pressure (SLP) fields which are supposed to be the determining factor of the wind speed and thus of wind waves. In this way, there is no need to run a wave model for the future climate, but to look into the climate projection for the fields of SLP, which are associated with storm situations in the modern climate.
Due to cyclonic activity in high latitudes, strong winds and stormwind waves are frequently observed there all year round, especially, in the cold season. For the severe and specific climate of the Barents and Baltic Seas, it is complicated to adequately observe directly the atmosphere and ocean and even to use information from satellites. In this case, weather and climate models are of great value to understanding the present atmosphere-ocean interaction processes and their physical mechanisms. Earth system modeling is an important instrument for future climate projection of extreme weather events, which should help to identify and manage the risks of extreme events (Field et al. 2012).

ENVIRONMENT
To make a projection of extreme events in the future we applied the original method described in Surkova et al. (2013), Surkova and Krylov (2017). We used two approaches to get the calendar of these events for the last decades. For the Baltic Sea, a calendar of storms was derived from results of experiments of the wave model SWAN for 1948-2011. We choose such days when the modeled Hs was 4 m or higher (the government standard of general requirements for safety in emergencies specifies waves with a height of 4 m or more in the coastal zone and 6 m or more in the open sea as hazardous ones). For the Barents Sea, we considered a day as an extreme event when the wind speed was higher than the value of its 99th percentile. We found 364 events for the Baltic Sea  and 240 events for the Barents Sea .
Based on these calendars, a catalog of atmospheric SLP fields was prepared for each sea. The data used was the one of NCEP/NCAR reanalysis (Kalnay et al. 1996) for the Baltic Sea, and ERA-Interim for the Barents Sea (Dee et al. 2011). Every storm SLP field from reanalysis then was compared with everyday models SLPs of 1950-2005 period for every climate model (24 CMIP5 models for the Baltic Sea and 27 models for the Barents Sea). When for the present climate the SLP from the climate model had a coefficient of spatial correlation R more than the critical one R C and the same spatial variance as the storm field it was taken into account. It was found that when R C ≥ 0.97-98 (individually calculated for each model) the number of storm events simulated by the model is as many as in the storm calendar.
Then analogs of "storm SLP fields" and their frequency were investigated for the climate models results for an RCP8.5 CMIP5 experiment for 2006-2100. For this period the days were chosen when the correlation of modeled SLP and storm SLP from reanalysis was higher than RC.
The results show (Fig. 7)  year. The number of storms in the White Sea is four times lower: ~20 per a year with Hs ≥ 2 m. In the Barents Sea, the most severe storms have threshold Hs ≥ 10 m, which even doesn't occur in the Baltic and White Seas. For the Baltic, the considered limit was 5 m (5-7 per a year) and for the White Sea -only 4 m (2-3 per a year).
There is no significant trend of storm number in the Baltic, Barents and White Seas detected for the entire period. For the Barents and Baltic Seas the variability period about 10-12 years was identified. In the Baltic and Barents Seas, the small-scale increase of the number of storms was found in 1992−1994 years. It corresponds to the high positive NAO and AO and significant negative SCAND values. For the White Sea, the positive NAO phase in contrary corresponds to the decrease of the storm number. On average, the connection with global atmospheric circulation is stronger (R ≥ 0.5) for the Baltic Sea, then for the other two seas. The most pronounced liaison for the Baltic is with SCAND pattern, for the Barents -with AO. The connection with AO index is clearer than with NAO for all three seas. Notably, that for stronger storms with Hs ≥ 4 m in the Baltic the significant negative linear trend was identified. It is interesting that the number of these events is connected with Barents storms with Hs threshold 7 m and correlation between them is 0.56. Also correlation of number of stronger storms in the Barents Sea (Hs ≥ 8, 9 m) and storm number of the White Sea (Hs ≥ 3 m) is significant (0.52). It reflects the common origin of these events.
According to the RCP8.5 scenario, in the second part of the XXI century the number of storm events will rise in the Baltic and Barents Seas.

ACKNOwLEDGMENTS
The results of part 1 (Storm number and trends) were obtained in the framework of the state assignment of FASO Russia