Evaluation of the ICON-Ru Model’s Sensitivity to Sea Ice and Sea Surface Temperature Changes in Polar Low Forecasts for the Cold Seasons of 2022–2024

Abstract

Polar mesocyclones are often the cause of sudden worsening of weather conditions, including strong winds, snowfall with low visibility, and storms. The short lifetime, rapid development, high movement speeds, and small sizes, combined with a lack of meteorological observations over the Arctic seas, create difficulties in forecasting associated weather phenomena. High-resolution numerical modeling can help address this issue. The emergence and development of polar lows (PLs) significantly depend on the properties of the underlying surface, which largely determine the dynamic properties of the atmosphere in the boundary layer. This article is dedicated to assessing the sensitivity of the configuration ICON-Ru of the model ICON with a 2.0 km grid spacing to changes in the sea ice boundary and sea surface temperature (SST) when forecasting the formation and development of PLs. The results showed that the presence of artificial ice in the model almost completely suppresses the development of PLs in cases where the vortex does not have a strong connection with the jet stream. Heating the SST to 278.15 K while simultaneously shifting the ice boundary northward leads to increased thermal instability, rising sensible and latent heat fluxes, and higher CAPE, which enhances PLs, with the degree of enhancement depending on the nature of the vortex formation itself.

1. Introduction

The development and movement of PLs across the Arctic seas can have a serious impact on the stable functioning of the Northern Sea Route. Forecasting even the near-term formation of these systems is complicated due to the limited number of ground observations in the Arctic region, while non-hydrostatic numerical modeling holds significant potential for improving the predictability of these synoptic objects. Numerous studies over recent decades show that polar lows are highly sensitive to changes in the sea ice boundary and sea surface temperature (SST) [1,2,3,4]. Depending on the nature of PLs’ formation, different mechanisms of its development prevail—baroclinic instability, convective instability, and barotropic instability [5,6]. Furthermore, even during the development of a single PL, the mechanisms for maintaining energy within it can change, depending on its stage of existence. An additional factor in the development of PLs may be the presence of a low-pressure zone at 500 hPa, cyclonic vorticity in the large-scale flow, and the presence of the jet stream over the forming PL [7]. The influence of various factors on PLs’ formation can be traced using numerical modeling. The authors of the study [8], which studied a case of PL formation over the Barents Sea, conducted three numerical experiments—a control experiment, one with the ice removed in the Svalbard area, and one with an increase in sea surface temperature of 5 K (from 68° N to 84° N, from 18° E to 50° E). Analyzing the results of these experiments led to the conclusion that the formation of this PL was primarily due to baroclinic instability, while convective instability played an important role during its development.

A similar study was conducted for Hudson Bay [1]: the ocean surface temperature was increased by 8 K, which led to an intensification of wind speeds in the PL to hurricane levels. The role of barotropic instability in the development of PLs has also been the subject of several studies in the latter half of the 20th century [9,10]. In recent years, research on the influence of orography and the distribution of sea ice on PLs’ formation in the Norwegian and Barents Seas has continued [11]. The authors investigated two typical PLs that formed near Svalbard during northern intrusions of cold air masses. It was found that Svalbard serves as an additional source of cyclonic vorticity that contributes to the formation and intensification of PLs. Numerical experiments showed that a reduction in ice cover west of Svalbard leads to a moderate intensification of PLs, whereas an increase in ice cover significantly hinders their development. These experiments demonstrated that PLs in the northeastern part of the Atlantic Ocean can withstand large changes in the land–sea mask (such as the “removal” of Svalbard). Thus, according to the results of the work in [11], the orography of Svalbard and the surrounding sea ice affect PLs in the Norwegian Sea, but they modulate them rather than being the dominant factor in their formation.

Previously, studies using the COSMO-Ru model have been conducted to investigate the influence of spatiotemporal variations in SST and sea ice parameters on the evolution of PLs [12,13]. It was also shown that the mesoscale weather prediction systems COSMO-Ru (with grid spacing of 6.5 km and 13 km) and the configuration ICON-Ru of the model ICON (referred to in the text as ICON-Ru) with a grid spacing of 6.5 km are capable of predicting the emergence and evolution of polar cyclones, their stages of development, and trajectories and can successfully reproduce hazardous weather phenomena such as heavy precipitation and gusty winds on time scales of up to two days [14]. The results of these studies indicated that high-resolution forecasting systems COSMO-Ru [15] and ICON-Ru [16] can serve both as tools for investigating the properties and mechanisms of PLs formation and as tools for their prediction. The COSMO consortium transitioned to the ICON model for operational numerical forecasting in 2015. The first comparison of the capabilities of the two models (COSMO-Ru and ICON-Ru) in reproducing polar lows was presented in [14]. It was shown that ICON allows for more detailed reproduction of both the position of the cyclone itself and the meteorological parameters within it. The aim of this research is to assess the capabilities of the ICON-Ru with a grid spacing of 2.0 km for forecasting and studying PLs, particularly assessing ICON’s sensitivity to changes in sea ice and SST fields when forecasting PLs.

2. Materials and Methods

2.1. Studied Area

Among the many PLs that arise worldwide, those forming in the North Atlantic, in the Norwegian and Barents Seas, have the greatest impact on Northern Sea Route. Cases of PLs forming over the Kara Sea are also known, but they are significantly rarer, as it is covered with ice throughout the entire cold period. Moreover, the North Atlantic is the most studied region [17], with the maximum number of occurrences in the Norwegian and Barents Seas [18]. There are a number of PL archives, but all of them have a significant time delay (from several years to decades). To select new cases of PL formation, constant monitoring of satellite images is required (we used the Antarctic Meteorological Research Center archive—https://amrc.ssec.wisc.edu/data/ftp/archive/ (accessed on 16 October 2025), wind forecasts over the northern seas, and warnings that are posted on the website (https://www.barentswatch.no/polarelavtrykk/ (accessed on 16 October 2025). Based on this data, maps of the genesis and movement of the main PLs for the cold seasons of 2022–2023 (Figure 1a) and 2023–2024 (Figure 1b) were created. It is clear (Figure 1) that the main regions of PLs’ genesis are coastal areas south of Svalbard and areas east of Greenland.

2.2. Model and Domain

For modeling PLs, we choose integrated area spans from 63° N to 83° N and from 27° W to 87° E (Figure 2). The grid spacing is 2.0 km, with a time step of 20 s. This configuration covers the waters of the Greenland Sea, Barents Sea, and partially the Norwegian and Kara Seas, allowing for the reproduction of the overwhelming majority of PLs observed in this area. The initial and boundary conditions for this domain were based on the results of calculations from the global model ICON-Ru (grid spacing of 13 km), with the nested domain N29.5 (grid spacing of 6.5 km, covering all territories north of 29.5° N).

The sensitivity experiments were designed to evaluate the impact of sea ice and sea surface temperature (SST) modifications on the modeling. A series of four experiments was conducted:

• Control experiment: No changes were made to the sea temperature or ice characteristics, serving as the baseline for comparison.
• IceCut experiment: In the specified region, sea ice was completely removed. Ice thickness and the ice fraction in the grid cell were set to 0, while the ice temperature was fixed at 273.15 K.
• IceCutTemp5 experiment: This aimed to further enhance heat fluxes from the underlying surface. In addition to zeroing out ice thickness (as in IceCut), the SST was increased. In areas where the SST already exceeded 278.15 K, it remained unchanged; in other regions, it was raised to 278.15 K.
• IceBuild experiment: This implemented the opposite approach by expanding sea ice coverage. In the selected region, ice thickness was increased to 1.4 m, the ice fraction to 1, and the ice surface temperature to 263.15 K.

Details of the experimental parameters are provided in

Table 1. Parameters of numerical experiments with varying sea ice boundaries and SST.

SST and Ice Boundary Characteristics	Control	IceCut	IceCutTemp5	IceBuild
SST	Has not changed	Minimum 273.15 K	Minimum 278.15 K	263.15 K
Ice thickness	Has not changed	0	0	1.4 m
Ice content in a cell	Has not changed	0	0	1
The area of ice parameter changes (Figure 2)	No	Between 68° N and 80.5° N and between 15° W and 70° E	Between 68° N and 80.5° N and between 15° W and 70° E	Between 68° N and 80.5° N and between 15° W and 70° E

.

To evaluate the ICON-Ru model’s performance in simulating polar lows, four experiments were conducted, as detailed below:

2.3. Selection of Cases for Study

Four cases of PLs differing in their genesis and season of occurrence were examined to assess ICON-Ru’s sensitivity to changes in the sea ice and SST on the formation and development of PLs.

• 16–18 March 2022 (Figure 3a). This is the longest-lived PL, which formed off the western coast of Greenland and traveled approximately 3000 km east to the northern tip of Novaya Zemlya. This case is interesting for several reasons. First, the PL’s speed and the distances it managed to cover indicate a connection between it and the jet stream (whose axis was located at 300 hPa); second, about half of the vortex’s path (from Svalbard to Novaya Zemlya) ran over a surface covered with sea ice, so it can be assumed that changes in the ice boundary may have significantly influenced the trajectory of the PL.
• 6–8 January 2024 (Figure 3b). This period is notable for the appearance of three PLs occurring over the waters of the Norwegian and Barents Seas. The first PL (PL-1) formed south of Svalbard in the rear of a large synoptic cyclone centered over Novaya Zemlya. It moved eastward and reached Kolguev Island on the morning of 7 January, then impacted Vaigach Island and the south of Novaya Zemlya. PL-2 and -3 originated southwest of Svalbard, where there were significant horizontal contrasts between the warm sea and the icy surface of the archipelago.
• 18–19 January 2022 (Figure 3c). An intense PL formed off the eastern coast of Greenland on the night of 17–18 January. It appeared behind the large synoptic cyclone which passed over the Greenland and Norwegian Seas. The PL shifted southeastward and passed in close proximity to Jan Mayen Island on the night of January 19.
• 18–19 September 2023 (Figure 3d). The polar low that emerged on 18 September 2023 is a rare case of PL formation during the warm period. The water temperature in the area of its emergence, according to satellite maps, was 8–9 °C, and the sea ice boundary at that time was significantly displaced to the north, passing approximately at 81–82° N. A well-formed “comma” in the cloud field is visible in satellite images from 18 September at 14 UTC, with the surface temperature and 500 hPa temperature difference reaching 42 °C.

3. Results

One of the criteria for the formation of polar lows is the contrast between the sea surface temperature (SST) and the temperature at the 500 hPa level. Figure 4a shows the temperature difference map, which indicates that east of Greenland, near the sea ice edge during the formation of the studied polar low, the [SST–T500] parameter reaches 45 K, exceeding the 43 K threshold given in [19]. It can be seen that the satellite image (Figure 3a) shows two “comma-shaped” clouds—the first one is slightly weaker than the second. The ICON-Ru model reproduced both of these polar lows in the wind field. In the first polar low, wind speeds reach 25 m/s, and in the second, they reach up to 30 m/s. Both polar lows approached Svalbard on the evening of 17 March. The passage of the second, stronger polar low resulted in strong winds after 19:00. When approaching Svalbard and the sea ice edge, the first of the two abovementioned polar lows, which was less active, weakened and eventually merged with the second vortex to form a single polar low.

To verify wind speed over the sea, where meteorological stations are absent, satellite data can be used. The ASCAT scatterometer database from the Metop-B satellite (25 km grid spacing) meets the required criteria. At the time when the polar low crossed the southern regions of Svalbard and Bear Island, ASCAT data showed that wind speeds increased to 35 knots; closer to 75° N, they were locally up to 40 knots, which is equivalent to 17.5–20 m/s. This agrees well with the model data (south of Edge Island, wind gust speeds reached 15–20 m/s).

Figure 5a shows the vertical wind speed profile along the 22.74° E meridian (approximately crossing the center of Edge Island in southern Svalbard). At 77.5–78° N, it is clear that at 23:00, when the center of the polar low was located over the island, wind speeds in the 900–950 hPa layer reached 32.5 m/s. At altitudes of 250–300 hPa, a jet stream is clearly visible, with wind speeds reaching 60 m/s (Figure 5a,b). The trajectory of the polar low lies exactly along the path of the maximum wind speeds in the jet stream throughout its movement across the Barents Sea (Figure 5b).

To analyze the sensitivity of this polar low to sea ice boundary changes, a series of experiments described in Table 1 was conducted. Figure 6 shows the sea ice boundary: the blue line indicates the control experiment (Figure 6a) and the experiment with modified ice—IceCut (Figure 6b). It can be seen that the extensive area between Svalbard and Novaya Zemlya, covered by ice in the control experiment, is ice-free in IceCut—the ice along the polar low’s path was “removed”. It was assumed that this would affect the vortex’s trajectory since, in the second case, it would pass over open water. However, no significant changes in the vortex’s path occurred, nor in wind speeds and pressure fields (Figure 6). In the control experiment, at 12 UTC, wind gust speeds were slightly higher in the area where the polar low moved from ice onto open water (Figure 6a). The absence of any significant changes during the polar low’s passage across the Barents Sea is likely due to the temperature of the ice present in the control experiment—it was relatively warm, so there were no substantial gradients between the ice and the water temperatures (Figure 6b). The sea surface temperature in the area between Svalbard and Novaya Zemlya ranged from −2 °C to −4 °C, while the ice thickness in that area was over half a meter. At the same time, the “model” ice temperature near Greenland and west of Svalbard was significantly lower: from −6 °C to −12 °C. This assumption is confirmed by the fact that the polar low intensifies noticeably when the SST is increased to 5 °C in the IceCutTemp5 experiment (Figure 6c). Wind gusts in the polar low in IceCutTemp5 exceed those in the control experiment by an average of 5 m/s. Notably, the greatest impact on the trajectory and development of the polar low is exerted by the added ice in the IceBuild experiment (Figure 6d): wind speeds decrease significantly, pressure increases, and when reaching Novaya Zemlya, wind gusts do not exceed 20 m/s (in the control experiment, they reach 30 m/s or more). Upon reaching northern Novaya Zemlya, the polar low began to decay and almost immediately disappeared from the satellite images.

Figure 7 shows the wind speed’s evolution during the passage of the polar low and after it moved away, at the Hornsund and Edge Island stations (for the forecast data near the station, the average of the five nearest model grid points was used). As can be seen in Figure 7a, starting at around 18:00 on 17 March 2022, wind speed at Hornsund station began to increase, with mean values reaching 13 m/s, and the influence of the polar low lasted until about 02:00 on 18 March 2022. The ICON-Ru model reproduced this increase in wind speed in all experiments, with the smallest increase in the IceBuild experiment (with added ice) and the highest wind speeds—up to 19 m/s—in the IceCutTemp5 experiment. The polar low reached Edge Island about two hours later than Hornsund station, where the mean wind speed reached 17 m/s. According to the model data, it was noticeably lower, about 13 m/s (however, in one of the model grid points near the station, the model gave 14 m/s). In the IceCutTemp5 experiment, where the polar low intensified compared with the control experiment, wind speed reached 19 m/s. The arrival time of the polar low at Edge Island coincides with reality in all experiments except IceCutTemp5. As can be seen in Figure 7a, the arrival of the polar low on the island was marked by a significant drop in atmospheric pressure—it fell by 10 hPa. In the model, the pressure drop was less pronounced in all experiments, ranging from 4–6 hPa, depending on the experiment (the largest drop in IceCutTemp5) and was almost absent in the IceBuild experiment. Overall, the vortex reached Edge Island station in a more active stage than Hornsund station, which can also be seen from the forecast wind gust map shown in Figure 7a.

It is interesting to examine the reasons for the intensification and weakening of the polar low in the control experiment and in the experiments with modified ice. For this, we can refer to CAPE (Convective Available Potential Energy) maps—an indicator of atmospheric instability that corresponds to the amount of energy an air parcel would gain when rising from the level of free convection to the level of neutral buoyancy. Although [20] showed that CAPE values are comparable with the ocean surface heat flux over a limited time and thus do not represent a significant component of the polar low’s energy budget, CAPE still provides insight into changes in atmospheric conditions.

Figure 8 shows CAPE difference maps for three experiments—IceBuild, IceCut, IceCutTemp5—relative to the control experiment (the control values are subtracted). The data are for 13 UTC 17 March 2022, when the young vortex formed near the ice edge east of Greenland was gaining strength and rapidly moving northeast under the influence of the jet stream. In the IceBuild experiment (Figure 8a), there is a clear decrease in CAPE compared with the control, with differences of up to −100 J/kg in the area of the two polar lows (a reduction in the absolute values by half). In IceCut, the changes range from −20 to +20 J/kg; the difference field has a patchy structure (areas of decreased CAPE alternate with areas of increased CAPE), so probably no significant change occurs (Figure 8b). This is because the polar lows are over open sea in both the control and IceCut experiments. Significant CAPE increases are seen when SST is artificially raised to 5 °C (Figure 8c). Over most of the area, CAPE increases by 100–150 J/kg, and especially strongly where the polar lows pass and northwest of Svalbard, where a synoptic-scale cyclone is located.

The sensible heat fluxes increase noticeably in IceCutTemp5 (Figure 9a), which is obvious, since increasing the SST naturally boosts atmospheric instability and the heat flux from the surface to the air. Similarly, when ice is added in IceBuild, the sensible heat flux drops (Figure 9b). The biggest flux changes appear in areas with synoptic cyclones and where the surface condition changes sharply (where there was no ice in the control, but it was artificially added). Identical results were found for latent heat fluxes.

Figure 10 shows the difference maps of mean sea-level pressure. It can be seen that in the IceBuild experiment, the pressure increase of 1–2 hPa is observed everywhere (Figure 10a), and in the center of the polar low, the pressure increase reaches up to 4 hPa. Thus, it is evident that adding ice in the model suppresses the development of the polar low. In the IceCutTemp5 experiment, pressure decreases by 0.5–2 hPa over most of the area except in the polar low region, where, as we see (Figure 10b), pressure increases. This is due to a northward shift in the trajectory; therefore, north of the growth zone, there is an area with a pressure drop of up to 3 hPa. The conclusion that the largest changes in the studied parameter occur in the polar low zone is confirmed by the wind gust change map (Figure 10c). Increasing the SST to +5 °C leads to an increase in wind speeds everywhere by 1–5 m/s; however, in the area of the polar low, wind gusts in IceCutTemp5 strengthen by 10 m/s or more.

Table 2. Changes in meteorological parameters in the polar low region over 16–18 March 2022 in the numerical experiments IceBuild, IceCut, and IceCutTemp5 compared with the control experiment.

Meteorological Parameter	IceBuild	IceCut	IceCutTemp5
Wind gusts (Russian Hydrometeorological Centre, Moscow, Russia)	−4 to −10 m/s; in the polar low area, gusts decrease significantly	+3–5 m/s; gusts increase in the ice-free experiment	Substantial increase over the entire area by 3–10 m/s. Maximum increase of 12 m/s or more localized in the polar low area
Mean sea-level pressure	Widespread pressure increase of 0.5–2 hPa; in the polar low’s center, it increases by 3–4 hPa	Minimal changes	Widespread pressure decrease by 0.2–2 hPa; in the polar low’s center, it decreases by 3 hPa; the trajectory shifts north
SST–T500	−2 to −6 °C in the area where there was ice in the control; −10 to −15 °C over the ice-free zone	Differences increase with forecast time; over the ice-covered area; differences are initially around zero, then increase to 4 °C at 28 h, 6 °C at 36 h, and 10 °C at 56 h	In the area where there was ice, increases of 20 °C (east of Greenland) and of 8–10 °C east of Svalbard; −2 to +6 °C over the ice-free area
Latent heat fluxes	20–30 W/m2; latent heat flux in IceBuild is zero; in the control, fluxes are −20 to −30 W/m2	Mostly minimal changes. Locally, −10 to −20 W/m2; the latent heat flux in IceCut is 10–20 W/m2 higher (negative, surface emission)	Negative changes over most of the area; −50 W/m2 east of Greenland, −20 to −30 W/m2 elsewhere (negative, surface emission)
CAPE	Minimal changes	−10 to +50 J/kg; slightly higher without ice	+100 to +150 J/kg; CAPE significantly higher without ice and with SST = 5 °C

presents the main conclusions on the changes in the meteorological parameters in the polar low region over 16–18 March 2022, depending on the sea ice boundary modifications. The results obtained show that the absence of sea ice alone does not strongly affect the polar low: this is probably due to the existing connection of the vortex with the jet stream. The most significant differences appear in the experiment with added ice and with the SST increased to 5 °C. In particular, CAPE increases slightly over the ice-free area but grows significantly more in IceCutTemp5. Latent heat fluxes also increase in IceCutTemp5; wind gusts in the polar low are higher in the IceCut experiment by 3–5 m/s and in IceCutTemp5 by up to 12 m/s. In the IceBuild experiment, wind gusts decrease, latent heat fluxes are low, and mean sea-level pressure is higher (the polar low is shallower than in the control experiment).

3.1. Polar Lows of 5–7 January 2024

The satellite image (Figure 3b) shows all three polar lows that formed on 5–7 January over the Norwegian and Barents Seas. All these vortices were reproduced by the ICON-Ru model in the control experiment (Figure 11a–c), and the forecast wind speeds match the ASCAT scatterometer data. Comparing the wind speeds predicted by the model over the sea surface with ASCAT scatterometer data showed good agreement. However, when comparing the forecast wind speeds with station data, it turned out that in the control experiment, the polar low lagged by several hours compared with the real mesoscale cyclone, and the wind speeds predicted by the model were slightly lower than those observed at coastal stations. It should be noted that the polar low made landfall almost three days after the forecast began, so the comparison of meteorological data was carried out with forecasts with a significant lead time (about 72 h). Analysis of upper-level wind speed maps and vertical profiles showed that one of the three polar lows during this period had some connection with the jet stream.

To consider the impact of the modified SST on the polar low, we turn to the CAPE maps. Figure 12 shows the CAPE difference maps for three experiments—IceBuild, IceCut, and IceCutTemp5—relative to the control experiment (CAPE values from the control experiment were subtracted from the others). It can be seen that in the experiment with added ice, CAPE values decrease by 20–100 J/kg in the polar low area. Overall, the CAPE reduction is less pronounced compared with the 17 March 2022 case (Figure 8). In the IceCut experiment, the changes range from −20 to +20 J/kg; the difference field has a very patchy structure (areas with decreased CAPE alternate with areas with increased CAPE), and probably no significant changes occur (Figure 12b). This is because the polar low is located over open water in both the control experiment and IceCut. As in the March case (Figure 8c), significant changes in CAPE are observed when the SST is artificially increased to 5 °C (Figure 12c). In the areas of the polar lows and in the east (where a synoptic-scale cyclone is located), CAPE increases by 50–100 J/kg. Overall, these changes are smaller than those observed in the IceCutTemp5 experiment in March, which is likely due to the absence of ice in the control experiment in the areas where the polar lows originated. Thus, the only difference between the control experiment and IceCutTemp5 is the surface temperature.

The latent heat fluxes become practically zero when the sea ice is increased, change little relative to the control experiment in IceCut, and increase noticeably in the IceCutTemp5 experiment. Overall, these results are logical and consistent with the data obtained for 17–18 March 2022; therefore, separate maps for them are not provided.

Table 3. Changes in the meteorological parameters in the polar low area of 5–7 January 2024 in the numerical experiments IceBuild, IceCut, and IceCutTemp5 compared with the control experiment.

Meteorological Parameter	IceBuild	IceCut	IceCutTemp5
Wind gusts	−2 to −5 m/s over the entire area; in the polar low area, −5 to −10 m/s	In the polar low PL-2, gusts increase by 2–4 m/s; in PL-1 and PL-3, a heterogeneous structure indicating minor changes	Substantial increase over the entire area of 5–10 m/s in the synoptic cyclone and in the area above the sea ice. Maximum increase of 10 m/s or more localized in the PL-2 area
Mean sea-level pressure	Widespread pressure increases by 0.5–2 hPa; in the polar low’s center, increases of up to 10 hPa	Slight increase in the PL-1 area of 0.6–0.8 hPa, decreases elsewhere of 1–2 hPa	Widespread pressure decrease of 0.5–3 hPa; in the polar low’s center, decreases of 2–3 hPa; propagation speed decreases
Latent heat fluxes	In PL-2 and PL-3, 30–50 W/m2; in PL-1, 50–100 W/m2. Flux in IceBuild is zero; in the control experiment, fluxes are −30 to −50 W/m2 and higher	Minimal changes over most of the area	Negative changes in the area covered by ice in the control experiment (fluxes are negative, surface emission). No significant changes in the polar lows themselves
CAPE	In PL-2 and PL-3: −10 to −20 J/kg; in PL-1: −20 to −50 J/kg	Minimal changes over most of the area	In places in PL-1: +50 to +100 J/kg; in PL-2 and PL-3: +20 to +30 J/kg; in the synoptic cyclone in the east, locally up to +150 J/kg

presents the main conclusions on the changes in the meteorological parameters in the area of the formation and movement of polar lows, depending on the modification of the ice boundary. The trajectories of the studied polar lows lay entirely over areas free from sea ice, so in the IceCut experiment, only minor changes were obtained. The most noticeable differences appeared in the experiment with added ice and the SST increased to 5 °C. In IceBuild, the creation of artificial ice completely suppressed the development of the polar low, unlike the same experiment for 17–18 March 2022, when the connection of the vortex with the jet stream supported its further existence. In particular, CAPE slightly increased over the ice-free area but grew significantly more in IceCutTemp5. Latent heat fluxes also increased, and wind gusts in the polar low were higher in the IceCut experiment by 3–5 m/s and in IceCutTemp5 by up to 12 m/s.

3.2. Polar Low of 18–19 January 2022

The formation of the polar low was associated with an intrusion of cold air over a warm underlying surface—the temperature differences between the ice-free and ice-covered sea surfaces reached 25 °C. The ICON-Ru model reproduces the polar low’s trajectory quite accurately. Maximum gusts in the IceCut experiment at the initial stage of the polar low’s formation (24 h forecast) already reached 40 m/s. As the cyclone moved southward, away from Greenland’s coast, gusts in the control experiment became slightly higher; the trough in the IceCut and IceCutTemp5 experiments flattened earlier and the polar low decayed sooner before reaching the shore. In both experiments with reduced ice, the polar low started to slightly lag behind the control. With added ice, the polar low developed weakly but moved southward at about the same speed as in the control experiment. After leaving the ice for the open sea surface, it re-intensified and gusts increased to 32.5 m/s, but it was still weaker than the polar low in the other experiments and dissipated earlier. The earlier and more intense intensification of the polar low without ice is probably due to a significant increase in both the SST–H500 temperature difference (in the control experiment at the time of formation, the difference did not exceed 30 °C; without ice, the temperature difference exceeds the 40 °C threshold by the 20 h forecast) and its proximity to Greenland’s cold ice surface. In both experiments, the polar low area is characterized by an SST–H500 temperature difference of more than 47 °C, but these values are reached only several (6–10) hours after the vortex forms.

Let us examine how CAPE changed in the experiments when the sea ice boundary was modified. Figure 13 shows the differences in CAPE values relative to the control experiment at the early stage of the polar low’s development, when it had not yet left the ice surface. Notably, during this period, vorticity in the baric trough increases in the surface to the 850 hPa layer (Figure 13d), while the vortex is still weak in the wind speeds. In the IceBuild experiment, there are no significant changes in the CAPE field in the polar low’s formation area; this is because the control experiment also had ice in this area. In the IceCut experiment, CAPE increases by 20–50 J/kg, which is significantly more than in the two previous cases, where formation occurred over an ice-free surface. In the IceCutTemp5 experiment, CAPE increases by 50–100 J/kg.

After a day of existence, the vortex moved onto the ice-free surface, intensified, and reached Jan Mayen Island. During this period, in the experiment with added ice, the vortex’s development practically ceased, CAPE according to the model dropped to zero (in the control experiment, its values reached 150), and only a small trough remained in the isobars. Figure 14 shows the CAPE maps for three types of experiments—control, IceCut, IceCutTemp5—according to the ICON-Ru data at 05 UTC 19 January 2022. The black oval marks the location of Jan Mayen Island. It can be seen how the vortex’s trajectory changes depending on the experiment. In the control experiment, by 05 UTC, the polar low had already passed Jan Mayen; in IceCut, it was west of the island; and in IceCutTemp5, it lagged behind and was located northwest, closer to Greenland.

In general, the numerical experiments showed different responses of the polar low to the absence of ice, depending on the stage of development. In particular, at the initial stage of development, thanks to increased atmospheric instability and enhanced horizontal temperature gradients, the vortex rapidly intensifies—wind speed and gusts are higher. However, in the mature stage, the opposite pattern is observed, and the polar low decays faster.

It should be noted that the period of 18–19 January 2022 was characterized by large vertical temperature gradients, with differences between the surface temperature and the 500 hPa level temperature reaching 47–49 °C. The latent and sensible heat fluxes were high throughout the North Atlantic area that was free of ice. In the control experiment, the latent heat flux in the polar low reached 100–120 W/m² (Figure 15a), which is almost twice as high as in the previous case (60–80 W/m²), with the maximum values predicted in the center of the baric trough. It is evident that in the IceBuild experiment, the fluxes almost completely drop to zero in the area covered by artificially added ice; however, at the edge of this ice and the warm sea, they immediately sharply increase up to 100–120 W/m² (Figure 15b). During this period, the vortex re-intensified and gusts increased to 32.5 m/s, but it was still weaker than the polar low in the other experiments and dissipated earlier. In the IceCut experiment (Figure 15c), latent heat fluxes were lower than in the control, ranging from 80–100 W/m², and the polar low itself in IceCut was located slightly further west, passing closer to Iceland. Interesting results were obtained in IceCutTemp5, where, in addition to shifting the ice boundary, the SST was increased to +5 °C. It is evident that the baric trough in the vortex area is less pronounced than in the control experiment and IceCut, and the latent heat flux values, although quite high (80–100 W/m²), are not localized near the vortex’s center and have a “smeared” structure.

Table 4. Changes in meteorological parameters in the area of polar low’s formation and development in the numerical experiments IceBuild, IceCut, and IceCutTemp5 compared with the control for the polar low on 18–19 January 2022.

Meteorological Parameter	IceBuild	IceCut	IceCutTemp5
Wind gusts	Initial stage: −10 to −12.5 m/s	Initial stage: +3–5 m/s; mature stage: −3 m/s	Initial stage: gusts increase by 5–7 m/s, polar low decays earlier
Mean sea-level pressure	Increase relative to control	Initially minor changes, then pressure rises faster than in the control	Initially minor changes, then pressure rises faster than in the control
Latent heat fluxes	In the area of artificial ice, widespread reduction to nearly zero	Minor changes	Overall increase in fluxes across the North Atlantic, insignificant changes in the polar low
CAPE	No changes at the initial stage, then a decrease as vortex development is completely suppressed	Initial stage: increase of 20–50 J/kg	Initial stage: increase of 50–100 J/kg

summarizes the main findings on changes in the meteorological parameters in the area of the polar low’s formation and development. Special attention should be paid to the fact that the response of the meteorological parameters to the sea ice boundary modification strongly depended on the vortex’s development stage. Short-term intensification of the polar low with SST warming and a sea ice boundary shift was followed by its earlier decay.

3.3. Polar Low of 18–19 September 2023

This case drew particular attention because it formed under rather unusual conditions, as polar lows are predominantly a winter phenomenon. According to [18], in September, polar lows in the region formed only twice over 14 years from 1999 to 2013, and the average occurrence of such cases is only 0.07. The sea surface temperature of the Barents Sea in the vortex formation area—south of Svalbard—was 6–8 °C, and the sea ice boundary lay significantly farther north of the island. Figure 3d clearly shows the cloudiness of the formed vortex. The polar low slowly moved southward and by noon on 19 September (UTC), it reached the Kola Peninsula’s coast.

Figure 16 shows the forecast data at 37 h—the stage of maximum mesoscale cyclone development—from the ICON-Ru model from noon UTC on 17 September. Wind gusts (Figure 16a) reach maximum values of 30–33 m/s; as the vortex moves southward, they gradually weaken; but by the time it reaches landfall (forecast hours 45–47), they remain high: 25–27 m/s. The maximum CAPE in the mesoscale cyclone (Figure 16b) is observed west of the center and reaches 80–100 J/kg. The vortex formation area was located in the rear part of an old filled cyclone with a minimum pressure of 1010 hPa and a small pressure gradient, but to the north of Svalbard, an extensive low-pressure area with a cold core (minimum geopotential height at 500 hPa—512 dam) persisted for a long time. Its presence played a major role in the cyclone’s formation under the weak influence of other factors. Indeed, as seen in Figure 16c, the mesoscale cyclone’s position with minimum pressure aligns well with the 500 hPa geopotential contours—the polar low matches the upper-level cold air source. The maximum wind speeds (Figure 16d) are also observed in the western part, occupying a wide band and reaching 27–32 m/s. After the vortex moved ashore, wind speeds at 500 hPa increased up to 47 m/s, and the maximum wind area continued to shift eastward.

As with the previously described cases, a full series of experiments (IceCut, IceCutTemp5, IceBuild) was conducted; however, the prescribed changes in the first two experiments had virtually no impact on the life cycle and dynamics of the polar low. This is probably due to a significantly stronger factor dominating the conditions for the mesoscale cyclone’s development. In the IceCutTemp5 experiment, the SST was only increased to 5 °C in areas where lower temperatures were observed, but in this atypical case, the SST was already sufficiently high, so the changes in the experiment only affected remote areas along the sea ice boundary in the northern part of the domain and did not influence the mesoscale cyclone itself. Only in the IceBuild experiment were the changes sufficient to weaken the polar low—differences began to appear from the very first forecast hour. The polar low still formed but was characterized by weaker winds and lower surface pressure compared with the control run. The pressure difference ranged from 2 to 5 hPa, and the wind difference reached up to 7–10 m/s (Figure 17a,b). This “lag” relative to the control experiment remained throughout the forecast period. At 500 hPa, the differences were much smaller but still noticeable—up to 4–5 m/s (Figure 17c,d). Thus, under certain conditions in the lower and middle troposphere, even sea ice cover is not sufficient to completely suppress the development of a polar low.

Table 5. Changes in the meteorological parameters in the polar low region on 18 September 2023 in the IceBuild experiment compared with the control experiment.

Meteorological Parameter	IceBuild
Wind gusts	Decrease by 8–12 m/s in the mature stage
Mean sea-level pressure	Minimum pressure increases by up to 5 hPa in the mature stage
Latent heat fluxes	Sign change from slightly negative to slightly positive (up to 50 W/m2)
CAPE	Decrease to less than 20 J/kg

summarizes the main findings on changes in the meteorological parameters in the polar low region on 18 September for IceBuild, as only in this experiment were significant differences observed compared with the control.

4. Discussion

Several cases of polar low formation were analyzed. The polar low that formed east of Greenland on 17–18 March 2022 had a strong connection with the jet stream; therefore, even in the IceBuild experiment, it continued to exist (despite partially crossing ice) and completed its entire 3000 km path to Novaya Zemlya, only slightly changing its trajectory northward and noticeably weakening. The absence of sea ice alone (IceCut) did not have a strong effect on the polar low, probably due to the significant connection of the vortex with the jet stream. With artificial warming of the SST to 5 °C, wind gusts increased by 12 m/s. In the IceCut experiment, wind gusts in the polar low were 3–5 m/s higher than in the control. Artificial SST warming to 5 °C resulted in increased latent and sensible heat fluxes and a significant increase in CAPE. In IceBuild, CAPE was reduced by half compared with the control, and latent and sensible heat fluxes were low while mean sea-level pressure was higher. Sensible heat fluxes increased noticeably in IceCutTemp5.

The polar lows that formed on 6–7 January 2024 in the Norwegian and Barents Seas also showed low dependence on sea ice in the IceCut experiment. These vortices formed and passed over ice-free surfaces and had no significant connection to the jet stream (except for PL-2). PL-1 formed near the sea ice boundary south of Svalbard in the rear part of a synoptic cyclone; the main formation mechanism was likely baroclinic instability, with an additional development factor being the presence of a slightly lowered pressure zone. Therefore, in the IceCut experiment, no significant changes occurred in the vortex, but in IceCutTemp5, clear changes were noted: latent heat fluxes increased, and wind gusts in the polar low increased by 12 m/s. Thus, SST warming increased the polar low’s intensity during the period when baroclinic instability no longer affected it (the vortex moved south of Svalbard). In IceBuild, the creation of artificial ice completely suppressed the polar low’s development, unlike the same experiment for 17–18 March 2022, when the jet stream connection supported its further existence. CAPE slightly increased over the ice-free area but grew significantly more in IceCutTemp5. Latent heat fluxes also increased; wind gusts in the polar low were higher by 3–5 m/s in IceCut and by 12 m/s in IceCutTemp5.

In the third case, 18–19 January 2022, the vortex’s response to the absence of ice differed depending on its development stage. At the initial stage, in IceBuild, there were no significant CAPE changes in the polar low’s formation area because the control experiment also had ice in this area. In IceCut, CAPE increased by 20–50 J/kg, which is significantly more than in the two previous cases, where formation occurred over an ice-free surface. In IceCutTemp5, CAPE increased by 50–100 J/kg. In the mature stage, development was suppressed in IceBuild, along with reduced sensible and latent heat fluxes, and CAPE dropped to zero. Overall, in IceCut and IceCutTemp5, the polar low experienced short-term intensification but then decayed earlier and dissipated before reaching the Norwegian coast.

Analysis of the atypical September polar low case showed that its development was likely influenced by a large low-pressure area with a cold core at the 500 hPa level. Since the vortex developed under relatively warm conditions and far from the sea ice boundary, the changes applied in IceBuild and IceCut did not affect the mesoscale cyclone. However, adding ice weakened the polar low: pressure increased by 2–5 hPa and wind speeds decreased by 7–10 m/s. Overall, this polar low showed greater sensitivity to the presence of a continuous ice cover: CAPE was significantly reduced (by 20 J/kg) throughout the mesoscale cyclone’s life cycle. Latent heat fluxes relative to the control experiment also decreased significantly (±50 W/m²), which greatly affected the cyclone’s thermodynamic development and atmosphere–surface interaction.

Overall, the sea ice boundary experiments showed that adding artificial ice in the model (IceBuild) has a much stronger effect than its removal (IceCut), while warming the SST to 5 °C combined with shifting the ice boundary northward (IceCutTemp5) leads to intensification of the polar low, and the degree of this intensification depends on the vortex’s development mechanisms.

5. Conclusions

This study focused on evaluating the capabilities of the ICON-Ru model, with a horizontal grid spacing of 2.0 km, for forecasting and analyzing polar lows (PLs), with a particular emphasis on assessing the model’s sensitivity to changes in the sea ice and sea surface temperature (SST) fields. Several PL cases with different origins were examined to estimate the model’s response to variations in SST and sea ice. The results varied depending on the formation mechanism of each polar low. A common finding across all cases was that artificial changes in sea ice had a stronger impact on the forecasts than increases in SST. It is also worth noting that the ICON-Ru forecast data used for predicting the landfall of polar lows and subsequent verification had a lead time exceeding two days, which is a promising outcome for forecasting this phenomenon.

Our study has a number of limitations which should be underlined.

-

The lack of in situ measurements in the Arctic introduces uncertainty in precisely interpreting the model’s performance, as observational data along the entire PL trajectory are not available for verification.

-

Since there are several mechanisms that can generate polar cyclones, more advanced and comprehensive methods are needed to isolate and evaluate the influence of each process.

-

A small number of PL cases were analyzed, meaning that the conclusions cannot be generalized to all PLs. Nevertheless, examining a few PLs from diverse origins has enabled a detailed exploration of their sensitivity to the underlying surface conditions. Future research directions could include expanding the scope of polar low studies, increasing the number of cases for greater result reliability, and obtaining statistically significant assessments of polar mesocyclone forecasts’ quality