The Value of Sentinel-1 Ocean Wind Fields Component for the Study of Polar Lows

Abstract

Polar lows can pose serious threats to maritime operations and coastal communities in polar regions, especially due to their extreme wind speeds. The accurate and reliable representation of their wind field thus plays a crucial role in forecasting and mitigating the risks associated with this phenomenon. This study aims to evaluate the value of the SAR-based Sentinel-1 Ocean Wind Field product compared to two reanalysis products—regional CARRA and global ERA5—in studying the spatial wind speed distribution of polar lows. A visual comparison of the wind direction and wind speed fields was performed, as well as a brief quantitative analysis of wind speeds. Despite notable differences in spatial resolution, all of the data sources are able to identify the polar lows. However, the SAR-based product remains unmatched in capturing the intricate structure of the wind field. Although CARRA resolves more details than ERA5, it still deviates from the SAR image to a degree that suggests that the difference in spatial resolution is not the only source of disparity between the sources. Both CARRA and ERA5 underestimate the maximum wind speed as compared to the SAR data. Only the SAR data seems capable of providing the information necessary to study the details of the wind field of polar lows.

1. Introduction

Polar lows are intense mesoscale cyclones that appear particularly frequently over the Arctic Ocean [1]. They develop over open water near the sea ice margin or snow-covered land masses during cold air outbreaks [2]. Their lifetime ranges from 3 to 36 h, and their horizontal scale varies between 200 and 1000 km [3]. Polar lows are characterized by their rapid development, high wind speeds with surface winds near or above gale force (minimum 15 m/s), and intense precipitation (typically 1 to 10 mm per hour) [1]. All of those characteristics make them a critical threat to maritime operations, navigation, transportation, and coastal or inland communities in Arctic regions [3]. Thus, accurate forecasting and timely identification procedures are crucial to reducing the hazard potential of polar lows. Their small spatial scale, however, hampers the accurate representation of polar lows in numerical weather prediction (NWP) models as well as reanalysis products.

ERA5 is the state-of-the-art global reanalysis by the European Center for Medium-Range Weather Forecasts (ECMWF) [4]. It has a horizontal resolution of 31 km, is among the most commonly used global reanalysis product and is publicly available at [5]. It has been shown to capture more polar lows than its predecessor ERA-Interim [6] but substantially underestimates the near-surface wind speed associated with polar lows, especially close to the center of the cyclones (see Figure 8b in [7]). The Copernicus Arctic Regional Reanalysis (CARRA), which is publicly available at [8], has a much higher spatial resolution of 2.5 km and performs better than ERA5 in the representation of polar lows [7]. Although global and especially regional reanalysis products have improved significantly in representing polar lows over the past few years, near-surface wind speeds are still being underestimated [3,6]. This may be a consequence of the coarse spatial resolution of reanalysis products relative to the complex wind fields of polar lows. However, Hallerstig et al. found in their experiments that even finer grid spacing compared to the standard grid resolutions used in reanalysis does not necessarily improve wind speed accuracy [9]. Instead, changes in the model setup, such as the use of resolved vs. parameterized deep convection, have a greater effect on the accuracy of wind speed simulations.

Since the 1960s, satellite imaging has been an essential aid in understanding polar lows. Synthetic aperture radar (SAR) is among the most common technologies for wind retrieval. SAR-based wind retrieval relates the surface wind speed to the capillary waves on the ocean surface as a function of the radar backscatter and the incidence angle (see [10,11] for more details). Few studies have been presented using SAR observations to study polar lows. Hallerstig et al. employed SAR images as qualitative reference data to analyze the structure of convective cells in polar lows [9]. Grahn et al. developed a deep learning network trained on SAR scenes to detect polar lows [12]. Other than that, it seems that SAR-based wind retrieval had quickly been disregarded for the research of polar lows due to a few general concerns: One major drawback is its low temporal resolution, limited by the rate at which the satellite revisits the region of interest on its orbit—in our case (high latitudes), often only one to two times in two days. As a result, it is considered less accurate than other remote sensing wind retrieval technologies, e.g., scatterometry [9,10]. Secondly, Tollinger et al. criticized the limitations of established SAR wind retrieval methods and suggested a modified approach [13]. According to them, there are two factors negatively affecting the common methods: (1) the use of co-polarized backscatter signals alone (as opposed to cross-polarized), which saturate with high wind speeds; and (2) the dependence on a priori wind direction information from NWP models. Although these studies provide valuable insights into the limitations of SAR-based wind retrieval, there are no studies known to us that evaluate these drawbacks and compare them to other wind data sources. After all, SAR-based wind retrieval is the only spaceborne observation method capable of providing the wind field at a spatial resolution fine enough to resolve the strong wind gradients present in polar lows [10]. We, therefore, think it is worth evaluating readily available SAR-based wind products and assessing their accuracy in capturing the wind field of polar lows in comparison to other available datasets.

This study particularly focuses on the Ocean Wind Fields (OWI) component of the Sentinel-1 Level 2 Ocean (OCN) product. It provides the 10 m wind speed and direction at a spatial resolution of 1 km and is publicly available through the Copernicus Data Space Ecosystem [14]. Although Sentinel-1 provides multiple sensing modes and polarizations, this surface wind component only uses information from the co-polarized signal and relies on a priori wind direction information. It has, therefore, been assumed to be susceptible to large errors for high wind speeds, but to the authors’ best knowledge, no study finding such results has been presented. We aim to examine the value of this state-of-the-art, publicly available SAR-based wind product as compared to two commonly used reanalysis products in the representation of the wind speed of polar lows.

2. Data and Methods

We make our comparison of wind data sources by means of two cases of polar lows. Figure 1 displays the geographical area investigated in this study. This region is particularly relevant for polar low research due to its geographical and climatic conditions. It experiences a high frequency of polar lows, making it an ideal location for studying their characteristics and impacts. The Sentinel-3 optical image shows the polar low detected on 23 March 2022, near the southern part of the Svalbard. The area of the other polar low identified on 28 December 2021 also falls within this region; no optical images were available for this example due to the polar night. The centers of both polar lows are indicated by pink circles. We compare three sources of 10 m wind speed and direction: the global ERA5 reanalysis, the regional, Arctic-specific CARRA reanalysis, and the SAR-based Sentinel-1 OWI component. The following subsections briefly describe the technical details of each data source as well as our methodology.

2.1. Reanalysis Products

2.1.1. ERA5

ERA5 is the fifth-generation global reanalysis developed by ECMWF [4,15]. It has a coarse spatial resolution in comparison to other sources used in this study of approximately 31 km, making it challenging to resolve some of the mesoscale features. Nevertheless, it has a very high temporal resolution providing hourly wind information and is publicly available through the Copernicus Climate Data Store [5].

2.1.2. CARRA

CARRA is the regional reanalysis product used in this study. It has a coarser temporal resolution than ERA5, providing 3-hourly data of atmospheric and surface meteorological parameters on a 2.5 km horizontal grid mesh. It is divided into CARRA-East and CARRA-West. In this study, we are interested in the east domain, which covers Svalbard, Franz Josef Land, Novaya Zemlya, and the northern parts of Scandinavia [8]. CARRA is generated using the HARMONIE-AROME model, with the ERA5 global reanalysis providing the lateral boundary conditions [16]. However, some enhancements have been implemented in comparison to both ERA5 and the operational HARMONIE-AROME modeling systems. It utilizes an improved data assimilation system and includes extensive utilization of satellite data from the HARMONIE-AROME operational weather prediction system, significant augmentation in the surface observation datasets, and substantial improvements in the regional physiography and orography [7].

2.2. Sar-Based Wind Product

The Sentinel-1 OCN Level-2 OWI components were acquired through the publicly available Copernicus Data Space Ecosystem, the European Union’s Earth observation program. This product is fully calibrated and provided as an ocean surface wind vector, estimated from Sentinel-1 Level-1 SAR images by inversion of its associated Normalised Radar Cross Section (NRCS). It is a ground range gridded estimate of the surface wind speed and direction at 10 m above the surface with a spatial resolution of 1 km. Moreover, we extracted the corresponding Sentinel-1 Level-1 Ground Range Detected scenes in Extra-Wide Dual-Polarization mode with 40 m spatial resolution as well as a Sentinel-3 Ocean and Land Color Instrument optical scene for complementary visual comparison. We performed several preprocessing steps on the Sentinel-1 SAR scenes: thermal noise removal, speckle noise correction, calibration to sigma-nought in dB, and ellipsoid correction. Additionally, we reprojected the Sentinel-3 optical scene. All the steps were performed using the ESA Sentinel-1 Toolbox [17].

In addition to the limitations of SAR-based wind retrieval mentioned in the introduction, it should be noted that, since the measured signal does not contain information about atmospheric stratification at this point, the geophysical model function used to relate the backscatter to a wind speed assumes neutral stability at all times [18,19]. Therefore, the method cannot respond to changes in wind speed due to atmospheric stratification: the SAR-based wind speed is defined as the equivalent neutral wind [18]. This is the wind speed for which a neutral drag coefficient accurately provides the correct stress, differing from the definition of wind speed as measured by an anemometer or output from an NWP. It is important to note that equivalent neutral wind is not the same as neutral wind. That is, unstable atmospheric conditions typically result in higher equivalent neutral wind speeds [18]. In the context of polar lows, often linked with cold air outbreaks, the stability differences between an equivalent neutral wind and an anemometer wind can significantly contribute to discrepancies observed between model predictions and SAR measurements. However, since there are no offshore in situ measurements available to use as reference values for the true wind speed, there are no grounds for a discussion of those discrepancies, and we will treat the two definitions as one in this work.

2.3. Methodology

The example polar lows we chose for this study were detected in the south of Svalbard on 23 March 2022, and to the north of Jan Mayen on 28 December 2021. For this region of interest, Sentinel-1 OWI data only became available at the beginning of 2021. This limitation not only restricts the number of polar lows available for examination but also makes it more challenging to find suitable and useful overlapping scenes for our analysis. The polar low near Svalbard occurred less than 100 km from land, making it a relevant and informative case for comparing SAR-based and reanalysis wind data since reanalysis tends to perform less accurately close to the coast due to the complexity of the nearby terrain. In contrast, the polar low detected north of Jan Mayen was further offshore, approximately 350 km away, meaning that reanalysis data can be expected to perform more accurately for this case. It is worth noting that SAR-based wind retrieval also encounters potentially distorting effects in coastal areas due to factors like land–sea contrast and varying surface roughness. However, in the case of polar lows, these systems typically develop and evolve over the open ocean far from the coastline, where such distortions are minimal. As a result, we do not expect these coastal effects to significantly impact the accuracy of wind retrieval in the areas investigated here.

The SAR scenes were produced at 06:40 UTC on 23 March 2022 and 07:30 UTC on 28 December 2021. The outputs from the reanalysis that were closest in time to the SAR scene were used for analysis; 7:00 UTC for ERA5 and 6:00 UTC for CARRA. Note that the optical scene seen in Figure 1 was acquired at 11:30 UTC, which is why the polar low has moved further south relative to the corresponding SAR scene.

As is generally the case with offshore wind research, the analysis is limited by the lack of objective reference data, since reliable in situ observations are extremely sparse in the polar offshore areas. We are, thus, limited to a more qualitative comparison between the SAR image and the reanalysis products. As the SAR-based wind retrieval does not provide a time series of the polar low but only a single image, the quantitative analysis approach focuses on a simple comparison of the wind speeds found in the data sources. While this approach is somewhat limited in scope, it nonetheless provides valuable insights by enabling the extraction of several key parameters related to the polar low. The metrics compared in this work, e.g., the maximum, minimum, and mean wind speed, are essential for understanding the intensity and dynamics of polar lows. Additionally, we identify shifts in the center of the polar lows for different sources, which is vital for accurately tracking its movement and predicting its behavior. Furthermore, we provide the approximate scale of the polar low’s eye, including its length and width, which is important for assessing its overall size and potential impact. Despite the limitations inherent in this semi-quantitative approach, the extracted parameters are critical for advancing our knowledge of polar lows and improving forecasting models.

3. Results

3.1. Wind Direction Comparison

Figure 2 displays the polar low detected by the three data sources, namely Sentinel-1 OWI, CARRA, and ERA5, along with a false-color composite of the SAR backscatter for visual comparison. Green vectors represent ERA5, blue vectors CARRA, and red vectors correspond to the Sentinel-1 OWI product. Neither CARRA nor Sentinel-1 are shown in their full spatial resolution for better visualization. The colors of the wind vectors provide a loose representation of wind speed, with darker colors indicating lower and brighter colors indicating higher wind speeds, however, this aspect will be discussed in detail in the following subsection. The pink circles illustrate the approximate cyclone centers obtained from the SAR image. The yellow dashed lines on the reference images display the approximate shape of the eye of the cyclone along with the major and minor axes. The images on the first column (a–d) correspond to 23 March 2022, while the images on the second column (e–h) correspond to 28 December 2021. Note that we downscaled the wind vectors to a closer spatial resolution by taking every 5th point for CARRA and every 10th point for Sentinel-1. This approach avoids a dense and cluttered visualization with overlapping vectors, enabling clearer interpretation by highlighting key patterns and differences without excessive detail. The spatial resolution for ERA5 remains unchanged, allowing for a more precise and consistent visual comparison between data sources.

Despite the differences in spatial resolution, all data sources display wind fields that indicate a cyclonic structure, which is the counter-clockwise rotation in the wind field (in the northern hemisphere). However, the centers of rotation of all three data sources deviate from the center of the cyclone based on the reference source. This would result in a discrepancy between the location of the detected and the real polar low. Interestingly, there is no one clear direction of deviation that would indicate a common bias. Furthermore, it is important to note that no temporal interpolation was conducted between the data sources, and there is a time difference between the SAR scene and both CARRA and ERA5, which might also contribute to the observed shifts in the polar low centers. For the polar low on 23 March 2022, CARRA and Sentinel-1 both display shifts of approx. 50 km towards the south and the east, respectively. ERA5 presents the smallest deviation; here the center has been shifted approx. 30 km eastward. Even though ERA5 has a smaller shift, due to its coarse spatial resolution, it can hardly be considered an optimal source for automatic polar low identification. In our case, even though a cyclone was detected, the information provided by ERA5 is very limited, especially compared to Sentinel-1 and CARRA. For the second polar low detected on 28 December 2021, the shifts are similarly varied: CARRA shows a shift of approximately 45 km to the north and east, ERA5 a shift of 40 km to the east, and Sentinel-1 a shift of 30 km to the south and west. Despite these deviations, the general indication of cyclonic structure remains consistent across all data sources.

As the Sentinel-1 OWI component is not able to retrieve wind direction by itself but instead relies on a priori wind direction information provided by an atmospheric model in addition to the backscatter information, it was to be expected that the information content of the wind vectors of the SAR-based wind retrieval could not be independent of the reanalysis products. The much more pressing question, however, will be whether the wind speed obtained from SAR will be more useful than the reanalyses.

3.2. Wind Speed Comparison

Figure 3 displays the wind speed field of the polar lows from the three data sources: Sentinel-1 OWI component on the left, CARRA in the center, and ERA5 on the right (the first column corresponds to 23 March 2022, while the second column illustrates polar low detected on 28 December 2021). The pink circle corresponds to the approximate centers of the polar lows based on the false-color composites of the SAR images. Again, each source successfully represents the general structure of the polar low. Nevertheless, the distinct differences not only in spatial resolution but in capturing the details of the wind field structure of the polar low are evident.

The wind speed map derived from the Sentinel-1 OWI product for the polar low detected on 23 March 2022 offers the highest spatial resolution, capturing intricate details and small-scale features within the polar low system. The second image represents the wind speed map derived from regional CARRA reanalysis, clearly identifying the cyclone as such. However, although CARRA has a very high spatial resolution compared to other reanalyses, and is relatively high even compared to the Sentinel-1 OWI component (2.5 km vs. 1 km), many of the small-scale details of the wind field are not discernible. Even the mesoscale structure of the cyclone does not agree as well with the SAR image as the proximity in spatial resolution might imply. The third image displays the wind speed map derived from ERA5. ERA5 has by far the lowest spatial resolution among the three sources, resulting in a much coarser depiction of the features of the polar low. Even though ERA5 can still be considered capable of identifying the polar low, it no longer captures the structure of the polar low, let alone small-scale details. For the second instance detected on 28 December 2021, the situation is relatively similar. The Sentinel-1 OWI wind speed map reveals detailed features of the polar low, such as the eye of the polar low and the intricate wind patterns associated with the cold air outbreaks. The CARRA reanalysis image identifies the cyclone but shows an even more different shape compared to the Sentinel-1 image than observed with the first polar low. The ERA5 image, with the lowest spatial resolution, still detects the polar low but provides the coarsest representation. Despite these discrepancies, each source is capable of identifying the polar low, though neither reanalysis product matches the structural detail provided by the Sentinel-1 OWI component. Thus, while the Sentinel-1 OWI component provides a more detailed representation of polar lows compared to the ERA5 and CARRA reanalysis products, neither of the reanalysis products can be considered fully relevant for studying the wind speed structure of polar lows based on the examples analyzed.

Identifying the structure of a polar low is crucial for understanding its development and potential impact. Detailed analysis of the system’s cloud patterns, including the identification of eye-like features or banding structures, provides valuable insights into its intensity. These structural characteristics are key indicators of the polar low’s strength and organization, which are essential for accurate forecasting and risk assessment. Looking more closely at the structure of the wind speed maps for 23 March 2022, obtained from each of the data sources, we will see that the characteristics of the cyclone are not illustrated equally well. The general shape of the cyclone center differs significantly: CARRA has a much rounder shape than the rather sustained oval shape shown in the SAR image, while ERA5 does not resolve any shape at all. The eye of the cyclone, which is supposed to be the calmest part of the structure with the lowest wind speed, seems to be slightly misaligned with the rest of the structure in CARRA. ERA5 struggles to resolve the eye at all; if anything, its location could only be estimated. At the same time, the eyewall and rainbands are significantly more detailed on the Sentinel-1 wind speed map, while on the ERA5 map, these structures are almost undetectable. This again shows the differences in the amount of complementary information provided by different sources. The structure of the polar low detected on 28 December 2021, varies significantly across the three data sources. Similar to the previous example, the shape of the polar low and, in particular, the eye of the cyclone differ noticeably. The Sentinel-1 OWI product shows a very round and circular eye, providing a detailed and well-defined depiction of the polar low, including clear rainbands and an eyewall. In contrast, the CARRA reanalysis presents a much thinner and shapeless eye, with no informative details about the rainbands and eyewall compared to the 23 March 2022 event. The ERA5 reanalysis fails to resolve any concrete shape, depicting a very wide and poorly defined polar low. It should be noted that this, in particular, affected the estimation of approximate length and width since it is challenging to precisely identify the eye of the cyclone with CARRA and ERA5. This further emphasizes the superior spatial resolution and detail captured by the Sentinel-1 OWI product compared to the more generalized and less detailed representations from CARRA and ERA5.

Table 1. Key parameters for different data sources and polar low characteristics. The approximate scale parameters, namely length and width correspond to the major and minor axes of the polar low, while the shift parameter demonstrates the approximate distance between the polar low center from the wind products and the reference source. Wind speed values were extracted from the polar low region depicted in the cropped images of Figure 3.

Date	Data Source	Wind Speed [m/s]			Shift [km]	Scale [km]
		Min	Mean	Max		Length	Width
23 March 2022	Sentinel-1	0.00	10.97	26.00	50	125	50
	CARRA	0.02	8.43	20.70	50	95	80
	ERA5	0.46	8.22	14.95	30	95	80
28 December 2021	Sentinel-1	0.00	10.49	24.40	30	50	45
	CARRA	0.06	11.15	17.69	45	90	30
	ERA5	0.32	9.71	16.53	40	80	60

contains the quantitative differences among the three data sources and their corresponding characteristics for two detected polar lows. In particular, both examples display relatively similar minimum and mean wind speed values, suggesting a good agreement between the data sources. Nevertheless, ERA5 shows slightly different values compared to CARRA and Sentinel-1, particularly in the minimum wind speed, with ERA5 recording 0.46 m/s and 0.32 m/s. Furthermore, there is a substantial discrepancy in the maximum wind speed, especially for the polar low identified on 23 March, where the Sentinel-1 OWI product provides a much higher maximum wind speed of 26.00 m/s compared to ERA5’s maximum of 14.95 m/s. For the polar low detected on 28 December, CARRA also shows a lower maximum wind speed of 17.69 m/s compared to Sentinel-1 24.40 m/s, and a value very similar to ERA5’s maximum of 16.53 m/s, which is expected given their related nature and the offshore scenario, likely leading to a similar performance in wind speed estimation. This discrepancy clearly indicates that Sentinel-1 is much more capable of capturing the strong wind speeds and wind speed gradients that can be expected in polar lows. The reanalysis products, on the other hand, seem to underestimate wind speed, as was expected [20,21].

Table 1 additionally provides the approximate scale parameters, i.e., the length and width that correspond to the major and minor axes of an ellipse fitted to the eye of the polar low. The polar low signatures detected on 23 March 2022 exhibit notable differences in terms of scale when observed using Sentinel-1, CARRA, and ERA5 data sources. Sentinel-1 reveals a distinctly ellipsoid shape of the polar low eye, characterized by an approximate length of 125 km and a width of 50 km. This elongated form contrasts sharply with the observations from CARRA and ERA5, which depict the polar low eye as significantly more circular. In both CARRA and ERA5 datasets, the eyes have similar dimensions, with a length of 95 km and a width of 80 km. For the polar low detected on 28 December 2021, the scale parameters also vary across the different data sources. The Sentinel-1 OWI product shows a polar low with approximate dimensions of 50 km in length and 45 km in width, providing a detailed and well-defined structure. In comparison, the CARRA reanalysis depicts a larger polar low with dimensions of 90 km in length and 30 km in width, reflecting a more elongated and less detailed structure. The ERA5 reanalysis, with its coarser resolution, shows dimensions of approximately 80 km in length and 60 km in width, presenting a broader and less precise depiction of the polar low. These variations highlight the differences in how much of the polar low structure is captured by each data source, not only due to differences in spatial resolution. It is important to note that the scale assessment focuses solely on the eye of the polar low, as this is the most easily identifiable part of the polar low across all sources.

4. Discussion and Conclusions

In this study, we briefly evaluated the value of a common, publicly available SAR-based wind product—the Sentinel-1 Level-2 OCN OWI component—for studying the wind field of polar lows. The SAR product was compared to two state-of-the-art reanalysis products—the global ERA5 reanalysis and the regional, Arctic-specific CARRA reanalysis. Despite significant differences in spatial resolution among the sources, all are capable of identifying polar lows, but there are great differences in the amount of information they can provide about the wind field. While ERA5 does not contain much more information than the general presence of a cyclonic structure in the region of interest, CARRA is able to capture much more of the spatial structure due to its higher spatial resolution and presumably due to its improved assimilation scheme too. Nevertheless, the Sentinel-1 OWI product remains unmatched in its capability of capturing many more details, small-scale structures, and formations within the polar low. The difference in the wind field shapes observed by CARRA and Sentinel-1 suggests that simply increasing the spatial resolution of reanalysis products may not be enough to match the level of detail provided by the SAR-based product; instead, it indicates that the reanalyses may not yet have the capability to fully capture these finer details. These findings clearly emphasize the trade-off between detail and temporal coverage in the analysis of this weather phenomenon.

Moreover, a notable observation of the brief quantitative analysis is a substantial disparity in wind speed between the coarsest (ERA5) and highest (Sentinel-1) spatial resolution data sources, highlighting the limited ability of even the best reanalysis products to accurately capture the high wind speeds present in extreme weather events. This finding underscores the importance of understanding the benefits and limitations of different data sources for polar low identification and emphasizes the need for improved methodologies to enhance the reliability of extreme weather predictions in polar regions.

In light of the limitations discussed, future research should explore advanced assimilation techniques and higher-resolution models to better capture polar lows and their associated wind fields. Furthermore, the Sentinel-1 OWI wind product should be analyzed on larger samples to demonstrate its relevance for polar low detection and forecasting. Further research in this domain is crucial for advancing our understanding and ability to accurately forecast extreme weather phenomena. This study has shown the value and potential that can be added by utilizing common SAR-based wind retrieval products in the study of polar lows despite their limitations.