Abstract

Forecasting and monitoring polar lows are, to a large degree, based on satellite observations from passive radiometers and from scatterometer winds in addition to synoptic observations and numerical models. Synthetic aperture radar (SAR) brings higher resolution compared to other remotely sensed sources of ocean wind, such as scatterometer data and passive microwave wind products. The added information in polar low situations from SAR and the increased-resolution scatterometer wind fields are investigated. Statistically, higher variability in the MetOp Advanced Scatterometer (ASCAT) wind is clearly found during polar low situations compared to all situations. Slightly more variability is also seen in the ASCAT 12.5-km wind product compared to the operational European Centre for Medium-Range Weather Forecasts (ECMWF) model surface winds. In two analyzed polar low cases, Environmental Satellite (Envisat) Advanced SAR (ASAR) images reveal numerous interesting features, such as the sharp fronts and the location and strength of the strongest wind field in the polar low. It is likely that if SAR images are available to operational weather forecasting, that it can in some cases lead to earlier detection of polar lows. However, a reliable wind field from SAR is still needed.

1. Introduction and state of art

Since synoptic observations are sparse in the polar regions, forecasting and monitoring of polar lows (PL) are to a large degree based on satellite observations from passive radiometers, in particular the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR), and from scatterometer winds in addition to numerical models. Polar lows are relatively small (200–1000 km) short-lived (1–2 days) cyclones that form over open water in polar regions, typically during cold-air outbreaks (Rasmussen and Turner 2003). With the satellite coverage of polar areas by AVHRR and wind scatterometer, most polar lows are identified by the forecasters at an early stage and the models are able to predict their development reasonably well. However, a misplacement of the PL by some tens of kilometers when the cyclone makes landfall is of high local importance due to the intensity and small size of these cyclones. Since PLs develop over the ocean where conventional observations are scarce, the models are often more inaccurate than for phenomena that occur over land.

Scatterometer observations have been routinely available since the launch of the European Remote Sensing Satellite-1 (ERS-1) scatterometer in 1991. The conventional observing network only consists of scattered buoy, oil rig, and ship observations to give information about a weather system at sea and before it eventually makes landfall. Scatterometry allows for retrieval of the near-surface ocean wind vector and has made a drastic improvement in the availability of ocean wind information. The fact that scatterometers are available on polar-orbiting satellites gives good data coverage at high latitudes that is well suited for observing polar lows. The scatterometer wind vector is typically provided with 25-km grid spacing. More detailed level-2 products have been developed through reprocessing of the original measurements from both QuikSCAT (Tang et al. 2004) and MetOp-A Advanced Scatterometer (ASCAT). ASCAT covers the surface in two 550-km-wide swaths and the revisit time at 70°N is around 4 times per day. The validation of the ASCAT 12.5-km products shows that they have a larger number of nonrejected retrievals both offshore and near the coasts and that they have a higher information content than the standard product, according to the spectral analysis (Verhoef and Stoffelen 2013; Vogelzang et al. 2011). Portabella et al. (2012) relates some of the increased wind variability to tropical rain cells by overlaying scatterometer wind field on rain data from the Tropical Rainfall Measuring Mission Microwave Imager.

Wide coverage (~400 km) synthetic aperture radar (SAR) images have been available for more than a decade from the Environmental Satellite (Envisat) and RADARSAT-1 and RADARSAT-2 satellites. The images have not been utilized to any great degree in operational monitoring of PL, partly because the temporal coverage is still too poor (revisit time at 70°N is about every second day, provided that images are acquired), and partly, probably, because there is no wind product directly available. The SAR provides finescale observations of the sea surface roughness [150-m spatial resolution for Envisat Advanced SAR (ASAR) wide-swath mode], giving information about the surface wind speed, surface location of the PL center, fronts, and numerous details, independent of daylight or cloud cover. Such observations are not possible from AVHRR due to clouds, and SAR winds are also far more detailed than what can be obtained from the scatterometer. The drawback of SAR is that the signal is a function of a combination of wind speed and direction, so an external wind direction input is needed in order to retrieve the wind speed. Wind direction is usually taken from an atmosphere model.

SAR has got some attention in the literature in connection with tropical cyclones and typhoons (Li et al. 2013) but less on the observation of PL, even if their similarities have been acknowledged (e.g., Linders et al. 2011). Moore and Vachon (2002) studied a PL in the Labrador Sea using an image from RADARSAT ScanSAR. Du and Vachon (2003) used SAR to characterize the eyes of hurricanes, and Vachon and Wolfe (2011) developed a C-band cross-polarization relationship, enabling wind speed retrieval from SAR images in strong wind cases (such as hurricanes and polar lows) without the need for wind direction input. This has led to some interest from the forecasting community (Zhang and Perrie 2012).

The objective of the present work is to study the added information content in existing ASAR images of PL and in the high-resolution wind products from ASCAT during PL situations. This is done through a statistical analysis of the ASCAT data and through case studies of PL with ASAR and scatterometer coverage. Section 2 describes the data and methodology, and the study takes advantage of a polar lows database covering the Nordic seas that is also described in the same section. The results are given in section 3 and section 4. The discussion is found in section 5 and conclusions are in section 6.

2. Data and methodology

a. STARS-DAT

The Sea Surface Temperature and Altimeter Synergy dataset (STARS-DAT) archive is a collection of numerical model output, altimeter, AVHRR images, ASAR, and scatterometer gridded to the same grid of the Nordic seas for the PL situations registered by Noer et al. (2011). The time period is 2002–12. ASAR scenes in proximity of registered polar lows and within ±1 day of the low were included. The normalized radar backscatter, retrieved wind speed, and atmospheric wind speed and directions were resampled to a subset of the STARS-DAT grid. All data are available for download from the STARS-DAT database (at http://polarlow.met.no/stars-dat/).

b. ASAR

The Envisat ASAR scenes are calibrated and processed into wind fields with 500 m × 500 m grid cells using the C-band model 5 (CMOD5; http://www.knmi.nl/scatterometer/cmod5/) with wind directions from the Norwegian reanalysis of wind and waves [10-km Norwegian Reanalysis Archive (NORA10)] and when available from ASCAT. To get collocations between ASCAT and ASAR, we used a time collocation window of ±3 h, which is very large when considering PL situations. If ASCAT wind directions showed a better fit than the model with ASAR image signatures, then these were used in the CMOD5 wind retrieval from ASAR.

NORA10 is a downscaling of the European reanalysis ERA-40 (Uppala et al. 2005) over the Greenland, Norwegian, Barents, and the North Seas using the High Resolution Limited Area Model (HIRLAM) (Undén et al. 2002) with approximately 10 km × 10 km spatial resolution (Reistad et al. 2011; Furevik and Haakenstad 2012). HIRLAM was used in operational forecasting at the Norwegian Meteorological Institute (MET Norway) until summer 2014, in two different model setups with 8- and 12-km spatial resolution, denoted as HIRLAM8 and HIRLAM12, respectively. Surface fields of the operational models have also been used in the case studies in order to illustrate differences between forecasts and observations.

The case studies were selected from a number of archived PL cases included in the Norwegian Meteorological Institute PL climatology (Noer and Lien 2010; Noer et al. 2011).

c. ASCAT

Scatterometer data have been obtained from ASCAT on the MetOp-A satellite. Three different ASCAT datasets are available with different spatial accumulation/averaging procedures (the time period in parentheses shows the dataset used in the present study):

  • A 25-km product (2007–12).

  • A 12.5-km “standard” product (2009–12).

  • A 12.5-km “coastal” product (with different backscatter windowing from the 12.5-km standard product to allow for footprints closer to the coast; 2010–12).

Collocated near-surface winds from the ECMWF numerical weather prediction (NWP) model are used as a reference to compare the statistics of the NWP model and scatterometer. These model wind data were easily available on the scatterometer wind product files. The model winds were interpolated from short-range (around 3 h) forecasts with the main operational runs of the ECMWF model valid at the time of the observations. The actual model resolution was around 25 km before January 2010, and around 16 km afterward, but the model data used here have a coarser resolution, as it has been extracted from storage on a Gaussian grid (denoted N160, corresponding to a grid resolution of about 0.5625°) and has gone through a horizontal interpolation to observation location.

The nature of polar lows makes it interesting to use the possibilities brought by the collocation datasets with PL in the STARS archive to produce some general statistics of the wind field in polar low situations. For this we have used the 12.5-km standard ASCAT product and computed both wind statistics for polar lows as defined by the PL mask in the STARS data archive, as well as the overall statistics for ocean wind over all the available time period. The PL mask of the STARS data archive is based on manually extracted information on the position, time, and radius of the polar lows in the above-mentioned PL climatology by Noer et al. (2011). The mask interpolates position and radius between the available analysis times. This mask allows us in particular to present wind statistics restricted to the area covered by a PL.

The statistics are presented in terms of wind vector dot product autocorrelation functions, which are easy to compute for a spatially irregularly distributed dataset.

If we let and represent a wind vector pair satisfying a certain criterion (for instance, belonging to a given separation distance bin), then the dot product correlation coefficient (autocorrelation) between such wind vectors is defined as

 
formula

Here the overbar denotes averaging over all pairs conditioned by the criterion. The criterion is here defined as various separation distance bins between the observation locations, which allows us to compute the autocorrelation as a function of separation distance. For the autocorrelation data we have binned the data into separation distance bins of 40 km for all scatterometer observation pairs, irrespective of the direction of the separation vector (no discrimination between along- and across-satellite-track separation). For the autocorrelation defined in terms of wind vector dot product used here, values near one indicate that the deviations in wind tend to be correlated both in magnitude and direction. Values around zero indicate a lack of correlation in magnitude or that the wind deviations tend to be orthogonal in direction. Negative autocorrelations indicate that wind deviations tend to be in opposite-pointing directions.

The autocorrelation functions have the property that the more finescale information content in the observations and wind field, the faster the autocorrelation will fall off with increasing separation. The fact that polar lows contain small-scale wind structures and a cyclonic circulation will be reflected in differences from the average wind statistics over ocean. For numerical tractability, some thinning of the observation dataset was applied, so that only every fourth wind vector cell in both coordinate directions was used.

3. Information content in ASCAT data

In Fig. 1 we present the wind vector dot product autocorrelation for the PL situations, with the number of data points as shown in Fig. 2. Figure 3 shows the overall statistics of scatterometer ocean wind for all cases, independent of PL presence. Polar lows contain cyclonic circulations spanning certain scales that influence the statistics and make the correlation structures look different from those for overall conditions. We see that the autocorrelation falls off faster in the PL cases, indicating smaller spatial scales than in average ocean conditions, and autocorrelations become negative beyond a certain separation, which is related to the typical scale of the cyclonic circulation in the PL cases. The PL mask ensures that data are taken around a PL, and therefore there is a high probability that the winds will be of opposite direction at large distances. For the overall dataset in Fig. 3, the autocorrelation does not show the same features related to cyclone circulations, and autocorrelation values do not become negative at large separations. In the autocorrelation plots, we also include corresponding curves from collocated ECMWF model background winds at scatterometer observation positions. The autocorrelations of the model fall off slower, indicating that the model has a limited horizontal resolution and captures less small-scale detail than scatterometer data.

Fig. 1.

Wind vector dot product autocorrelation for the 12.5-km ASCAT product in PL situations (curve with circle markers). Curve with triangle markers is for the collocated ECMWF NWP winds.

Fig. 1.

Wind vector dot product autocorrelation for the 12.5-km ASCAT product in PL situations (curve with circle markers). Curve with triangle markers is for the collocated ECMWF NWP winds.

Fig. 2.

Number of samples used in the statistics in Fig. 1 as a function of separation.

Fig. 2.

Number of samples used in the statistics in Fig. 1 as a function of separation.

Fig. 3.

Wind vector dot product autocorrelation for the 12.5-km ASCAT product for all data irrespective of PL presence (curve with circle markers). Curve with triangle markers is for the collocated ECMWF NWP winds. Note that the vertical axis here has a smaller range than in Fig. 1.

Fig. 3.

Wind vector dot product autocorrelation for the 12.5-km ASCAT product for all data irrespective of PL presence (curve with circle markers). Curve with triangle markers is for the collocated ECMWF NWP winds. Note that the vertical axis here has a smaller range than in Fig. 1.

4. Information content in ASAR data

A statistical analysis of ASAR data, similar to the study with ASCAT, is possible but would require a large number of ASAR scenes. Also, the ASAR wind field is not a purely independent observational product, since an external wind direction is utilized to retrieve the wind speed. In the case of mismatch in time or space between ASAR and the external wind direction, the resulting errors in the wind field can also be significant (see, e.g., Yang et al. 2011; Furevik and Espedal 2002). We therefore choose in this study to illustrate the information content and the added value from ASAR in forecasting of polar low events through two case studies.

The first case is a possible early detection of a polar low by ASAR, while the second case illustrates the advantage of detailed surface wind information. The radiometer images used are NOAA-15, NOAA-16, NOAA-17, and NOAA-18, and MetOp-2, downloaded and interpreted at MET Norway.

a. Case 1: ASAR early detection

A series of AVHRR images (Fig. 4) over the Norwegian Sea at 0957–2028 UTC 18 November 2008 shows a fully developed polar low in the northwest and a polar low emerging around 71°N, 12°E. From the 1000 UTC analysis, the clouds in this area were analyzed as a back-bent occlusion to the rear of the passing synoptic low (Fig. 4a, circle). Not until 2 h later (Fig. 4b), a first indication of a polar low can be seen from the AVHRR; a feeble eye barely visible, but the waves in the cirrus clouds (very white clouds) to the west of this area is an early warning that something is about to happen. At 1407 UTC (Fig. 4c), a more clear signature in the AVHRR appears, the eye still only barely visible, but the cloud structures now have the distinct signature of a polar low. At 2028 UTC (Fig. 4d) the polar low is showing the typical cyclonic signature, with a cirrus shield with cloud-top temperatures at −52° to −55°C to the north of the center, and a cloud band with a gradually lesser vertical extent (more gray) leading in toward the center around the western part of the low. The AVHRR gives the visual appearance and the cloud-top temperatures are inferred from the radiance at the cloud top (NWCSAF 2014).

Fig. 4.

NOAA AVHRR channel 4 images from (a) 0957, (b) 1159, (c) 1407, and (d) 2028 UTC 18 Nov 2008. Circle indicates the coverage of the ASAR image in Fig. 5a.

Fig. 4.

NOAA AVHRR channel 4 images from (a) 0957, (b) 1159, (c) 1407, and (d) 2028 UTC 18 Nov 2008. Circle indicates the coverage of the ASAR image in Fig. 5a.

An ASAR image from 1013 UTC (Fig. 5a) and the corresponding wind field (Fig. 5b) displays a well-defined area of stronger wind (with a magnitude of 21 m s−1) to the west of a sharp shear zone at 71°N, 12°E. This wind speed agrees with NORA10, which has up to 21 m s−1 near the shear zone (not shown). The shear zone of just a few kilometers width is separating this area from a more tranquil area to the east with winds of 12–15 m s−1. The dark red area (>25 m s−1) here is probably due to about 25-km misalignment of the shear zone toward the west in NORA10. We assume that the wind speed from ASAR in this area is of the same magnitude as farther west. Farther east, a signature similar to what could be expected from the eye of a polar low is seen, with apparently calm winds of magnitude of 5 m s−1 (white arrow in Fig. 5a).

Fig. 5.

(a),(c) Envisat ASAR normalized radar backscatter images over the Norwegian Sea on 18 Nov 2008 with the PL center indicated by the white arrow in (a). Corresponding retrieved wind speeds using wind directions from (b) NORA10 for the morning pass and (d) ASCAT for the evening image. Red arrows show satellite travel direction.

Fig. 5.

(a),(c) Envisat ASAR normalized radar backscatter images over the Norwegian Sea on 18 Nov 2008 with the PL center indicated by the white arrow in (a). Corresponding retrieved wind speeds using wind directions from (b) NORA10 for the morning pass and (d) ASCAT for the evening image. Red arrows show satellite travel direction.

There is a possibility that backscatter in the center of the PL could be contaminated by rain. Synoptic ship observations indicate a 2-m temperature of +2.5°C 130 km south of this area at 1000 UTC, and −1.3°C 300 km to the north, and the point itself is embedded in a frontal zone. Thus, the actual 2-m temperature at the location is uncertain, as is thus the nature of precipitation. In the areas north and west of the center, temperatures are below zero, with precipitation in the form of snow, so the backscatter is not impaired in these areas.

Several interesting features can be seen from the images in Fig. 5. The most prominent is the sharp fronts in the northern sector of the low. The increase from a northeasterly fresh breeze at 12 m s−1 to a more than 20 m s−1 gale in this area occurs in less than 20 min as a result of the sharp front (3 km wide) and a propagation speed of 2.4 m s−1.

In the area south of the center, at around 69.5°N, 12°E, there is deep convection and well-developed open cell cumulonimbus (Fig. 5c). The HIRLAM8 model +18-h prognosis indicated 20.5 m s−1 wind here, which is representative for the eyeball average over the ASAR winds and agrees with ASCAT wind speeds. The wind directions for the evening image (2006 UTC) are taken from ASCAT almost 3 h earlier (1718 UTC), but since the low is moving relatively slow, the directions fit well with the wind signatures in the ASAR image and the resulting wind speeds are believed to be good in this case.

Experienced forecasters allow for some gusting winds associated with convective air masses, by forecasting a lower baseline wind and then typically adding two Beaufort for intermittent bursts. The nature of this variation is seldom observed directly, but the SAR wind map (Fig. 5d) illustrates this in showing the wind varying from 12 to 25 m s−1 over short bursts, probably associated with the downdraft from the convective clouds.

A commercial vessel was moving directly in toward the center of the low at less than 5 kt (2.6 m s−1) opposite of the wind, and measuring at most 30.8 m s−1 at 0000 UTC 19 November. Three hours later, the ship observed the sea level pressure at 963.6 hPa some 60 km west of the center, which is 4 hPa less than the +3-h prognosis of HIRLAM8. This model error can partly be explained by the center pressure not being deep enough, but also a model positional error of some 30–50 km. If we compare the model and observed mean sea level pressure (MSLP), the actual PL center pressure can be estimated as 960 hPa.

In the present case, the numerical model was able to predict the polar low and the analysis issued at 1000 UTC 18 November read, “Storm center 973 hPa position 70°N, 12°E is expected to be 969 hPa at 70°N, 23°E by Wednesday 0600 UTC.” Without the situation picked up well by the model and just relying on the analysis of the AVHRR, this polar low would probably first have been suspected at the 1407 UTC AVHRR image and not be identified with certainty until the 2028 UTC image, by which time it would be too late for an effective warning. If the 1013 UTC ASAR image had been readily available to the forecaster, even with some delay, it would have proven a very useful additional source of information and would most likely have had an impact on the timing and focus of the weather warnings.

b. Case 2: The Honningsvåg low

This polar low in northern Norway was misplaced by about 60 km by the models, which led to calm weather in areas where a gale was forecasted. The ASAR image illustrates these wind gradients well (Fig. 7). The AVHRR image from the early morning showed a cluster of polar lows in the Barents Sea. The southernmost of these lows hit land at 0900 UTC as seen in Fig. 6b, where the high wind area is collocated with the thick gray area. The polar low in the model forecast from 9 h earlier (mean sea level shown in blue contours in Fig. 6a) was, however, placed too far to the east, resulting in a dramatic error in the forecast east of the town Honningsvåg (forecasted 23 m s−1 as shown by filled contours in Fig. 6a; observed 7.7 m s−1 at Slettnes lighthouse shown in Fig. 6b).

Fig. 6.

AVHRR image from 0900 UTC 7 Jan 2009 over the Barents Sea and northern Norway. (a) Overlaid with HIRLAM8 mean sea level (MSLP) (blue isolines, minimum of 975 hPa) and the strongest wind field (yellow-filled contours 15–22.5 m s−1). (b) Overlaid with in situ wind observations: Honningsvåg, northwest 20.5 m s−1 (40 kt); Slettnes, northeast 7.7 m s−1 (15 kt). The box in (a) indicates the area covered in (b).

Fig. 6.

AVHRR image from 0900 UTC 7 Jan 2009 over the Barents Sea and northern Norway. (a) Overlaid with HIRLAM8 mean sea level (MSLP) (blue isolines, minimum of 975 hPa) and the strongest wind field (yellow-filled contours 15–22.5 m s−1). (b) Overlaid with in situ wind observations: Honningsvåg, northwest 20.5 m s−1 (40 kt); Slettnes, northeast 7.7 m s−1 (15 kt). The box in (a) indicates the area covered in (b).

Fig. 7.

(a) Envisat ASAR normalized radar backscatter image from 0900 UTC 7 Jan 2009 over the Barents Sea and northern Norway. Red arrows show the satellite travel direction. White arrow indicates the center of the polar low. (b) Corresponding retrieved wind speeds using wind directions from ASCAT.

Fig. 7.

(a) Envisat ASAR normalized radar backscatter image from 0900 UTC 7 Jan 2009 over the Barents Sea and northern Norway. Red arrows show the satellite travel direction. White arrow indicates the center of the polar low. (b) Corresponding retrieved wind speeds using wind directions from ASCAT.

The ASAR image (Fig. 7) shows the maximum wind area coincident with the clouds in the AVHRR image and a very sharp front to the low wind area. Indication of an “eye” is also visible very close to land (white arrow). The highest wind speed from the ASAR image is about 21 m s−1 when using ASCAT directions from 0818 UTC. This agrees well with ASCAT wind speeds (Fig. 8a) and the synoptic observations of 20–23 m s−1. The lower winds in the area east of the front are about 10 m s−1. The gradient of 11 m s−1 happens over a distance of just 1 km in the SAR image. ASCAT records this as subcell variability, as seen from the high values of the maximum likelihood estimator (MLE) in the frontal zone in Fig. 8b. The MLE information is the backscatter distance (in the level-2 product files) and is a measure of the wind inversion residual (Lin et al. 2014).

Fig. 8.

ASCAT (a) wind fields and (b) MLE distribution at 0845 and 1025 UTC 7 Jan 2009 over the PL in the southern Barents Sea.

Fig. 8.

ASCAT (a) wind fields and (b) MLE distribution at 0845 and 1025 UTC 7 Jan 2009 over the PL in the southern Barents Sea.

5. Discussion

The wind autocorrelation statistics based on scatterometer data quantify how wind field properties in the area defined by the polar lows mask differ from the overall ocean wind statistics by the presence of rotational wind structures. A more rapid falloff of the autocorrelation is seen, indicating smaller spatial scales than in the average ocean wind conditions. The autocorrelations for the 12.5-km ASCAT product also indicate that this observation dataset is less smooth and contains independent small-scale structure not present in corresponding NWP wind fields. This is in line with earlier studies by, for example, Vogelzang et al. (2011). Such variability is an indication of the usefulness and added value that can be expected in application of scatterometer data in monitoring of polar lows.

Going further with the ASAR dataset with even higher resolution, our case studies have demonstrated that small-scale structures present in polar lows are captured in these observations, and that pieces of information on a new level of detail relative to the ASCAT winds can be added. The case studies have revealed a number of wind-related features in SAR in connection to polar lows that can have importance for understanding their dynamics and aid their monitoring. SAR winds can also help operational forecasting by comparison with NWP forecasts, giving information on the model performance in these situations.

Because of the high resolution from SAR and its ability to look through clouds, as opposed to AVHRR, it is possible to see the wind impact on the sea surface. In many cases, wind signatures in the SAR image can confirm or invalidate scatterometer or NWP model wind direction and show wind variability near the coast, gustiness, or even an early development of a polar low (ref. case 1). The features of wind impact on the sea surface (gravity waves, wavy fronts, etc.) are pieces of information that might be important, but these elements have not yet been studied in connection to polar lows. When it comes to operational use (i.e., weather forecasting), so far the most valuable piece of information we can obtain from SAR and the scatterometer is the visualization of the sea surface wind impact. This may in some cases allow for earlier detection of a polar low and may further aid in accurate location of wind shear zones and the region of the strongest wind in connection with the PL.

The PL in case 1 was moving slow compared to the typical speed of propagation for polar lows, which is found to be 8–13 m s−1. Given a more typical speed of propagation and a more favorable orientation of the shear zone, this wind speed change can take place in just a few minutes. Eyewitnesses often report of such rapid changes in wind speed, stating the increase to last from almost instantaneous to 15 min. Narrow transitions in a polar low observed with SAR was earlier pointed out by Moore and Vachon (2002), who also attributed them to changes in wind speed and/or wind direction.

In the present work, external wind direction input from the scatterometer and model was used to retrieve the wind speed from SAR with the CMOD5 geophysical model function. As the PLs are highly variable in wind direction and speed, a small error in the NWP model representation of the wind field may lead to large errors in the retrieved wind speed. An offset of 25 and 60 km between the wind front in ASAR and NORA10 is experienced in the two cases (Figs. 5 and 7), respectively.

As demonstrated, using scatterometer wind directions in the ASAR retrievals can result in good SAR wind maps, but it further limits the coverage because of the need for temporal and spatial collocation of scatterometer and SAR (cf. the extent of the SAR images and the corresponding wind maps in Figs. 5c and 5d and Figs. 7a and 7b). A possible solution could be to blend the model and scatterometer wind directions, in particular to fill the gap near the coast. The two-dimensional fast Fourier transform (FFT) method used to derive wind direction from wind streaks in ERS SAR images (see, e.g., Furevik et al. 2002) is seldom applicable on wide-swath SAR images due to the coarser resolution. Other methods for wind retrieval, such as the cross-polarization (Vachon and Wolfe 2011) or the Doppler technique (Mouche et al. 2012), have recently been developed. As a consequence, data from Sentinel-1 will have the Doppler anomaly included in the level-2 product (Mouche et al. 2012). This could be a step toward a SAR wind field product that would rely on NWP fields solely for a priori information.

6. Conclusions and outlook

Overall, the statistics of the ASCAT winds in PL situations demonstrate the presence of small-scale wind structure in these cases, and the statistics also demonstrated the added representation of independent small-scale wind information in scatterometer data relative to an NWP model.

The case studies with ASAR data of two polar lows illustrate the added information on smaller scales than can be obtained with the scatterometer. It indicates that SAR will be very helpful to forecasters, given that retrieved wind field data with high quality can be provided in near–real time (within about 2 h). The problem so far has been the low coverage both in time and space and missing wind direction input.

It has further been demonstrated that using wind direction input from the scatterometer for the SAR wind retrieval can provide high-resolution-quality wind fields, although with reduced spatial coverage. The new satellites MetOp-B and Sentinel-1 are expected to improve this situation significantly through improved coverage with both the scatterometer and SAR.

Acknowledgments

The authors acknowledge the funding from the Norwegian Space Center through the Projects JOP 17.11.3 and JOP 07.13.2 and from the European Space Agency through the Support to Science Element (STSE) Project STARS (Project 22644/09). The ASAR images are obtained through Data Projects 4109 and 9968.

REFERENCES

REFERENCES
Du
,
Y.
, and
P. W.
Vachon
,
2003
:
Characterization of hurricane eyes in RADARSAT-1 images with wavelet analysis
.
Can. J. Remote Sens.
,
29
,
491
498
, doi:.
Furevik
,
B. R.
, and
H. A.
Espedal
,
2002
:
Wind energy mapping using synthetic aperture radar
.
Can. J. Remote Sens.
,
28
, 196–204, doi:.
Furevik
,
B. R.
, and
H.
Haakenstad
,
2012
:
Near-surface marine wind profiles from rawinsonde and NORA10 hindcast
.
J. Geophys. Res.
,
117
, D23106, doi:.
Furevik
,
B. R.
,
O. M.
Johannessen
, and
A. D.
Sandvik
,
2002
:
SAR-retrieved wind in polar regions—Comparison with in situ data and atmospheric model output
.
IEEE Trans. Geosci. Remote Sens.
,
40
, 1720–1732, doi:.
Li
,
X.
,
J.
Zhang
,
X.
Yang
,
W.
Pichel
,
M.
DeMaria
,
D.
Long
, and
Z.
Li
,
2013
:
Tropical cyclone morphology from spaceborne synthetic aperture radar
.
Bull. Amer. Meteor. Soc.
, 94, 215–230, doi:.
Lin
,
W.
,
M.
Portabella
,
A.
Stoffelen
,
A.
Turiel
, and
A.
Verhoef
,
2014
:
Rain identification in ASCAT winds using singularity analysis
.
IEEE Geosci. Remote Sens. Lett.
,
11
, 1519–1523, doi:.
Linders
,
T.
,
Ø.
Sætra
, and
T. J.
Bracegirdle
,
2011
: Limited polar low sensitivity to sea-surface temperature. Quart. J. Roy. Meteor. Soc.,137, 58–69, doi:.
Moore
,
G. W. K.
, and
P. W.
Vachon
,
2002
:
A polar low over the Labrador Sea: Interactions with topography and an upper-level potential vorticity anomaly, and an observation by RADARSAT-1 SAR
.
Geophys. Res. Lett.
,
29
,
1773
, doi:.
Mouche
,
A. A.
,
F.
Collard
,
B.
Chapron
,
K.
Dagestad
,
G.
Guitton
,
J. A.
Johannessen
,
V.
Kerbaol
, and
M. W.
Hansen
,
2012
:
On the use of Doppler shift for sea surface wind retrieval from SAR
.
IEEE Trans. Geosci. Remote Sens.
,
50
, 2901–2909, doi:.
Noer
,
G.
, and
T.
Lien
,
2010
: Dates and positions of polar lows over the Nordic Seas between 2000 and 2010. Norwegian Meteorological Institute Rep. 16/2010, 7 pp. [Available online at http://met.no/filestore/report16_10.pdf.]
Noer
,
G.
,
Ø.
Sætra
,
T.
Lien
, and
Y.
Gusdal
,
2011
: A climatological study of polar lows in the Nordic Seas. Quart. J. Roy. Meteor. Soc.,137, 1762–1772, doi:.
NWCSAF
,
2014
: Algorithm theoretical basis document for cloud top temperature, pressure and height of the NWC/PPS. NWC SAF Rep. NWC/CDOP2/PPS/SMHI/SCI/ATBD/3, Issue 1.0, 35 pp. [Available online at http://www.nwcsaf.org/scidocs/Documentation/NWC-CDOP2-PPS-SMHI-SCI-ATBD-3_v1_0.pdf.]
Portabella
,
M.
,
A.
Stoffelen
,
W.
Lin
,
A.
Turiel
,
A.
Verhoef
,
J.
Verspeek
, and
J.
Ballabrera-Poy
,
2012
:
Rain effects on ASCAT-retrieved winds: Toward an improved quality control
.
IEEE Trans. Geosci. Remote Sens.
, 50, 2495–2506, doi:.
Rasmussen
,
E. A.
, and
J.
Turner
, Eds.,
2003
:
Polar Lows: Mesoscale Weather Systems in the Polar Regions
. Cambridge University Press, 612 pp.
Reistad
,
M.
,
Ø.
Breivik
,
H.
Haakenstad
,
O. J.
Aarnes
,
B. R.
Furevik
, and
J.-R.
Bidlot
,
2011
:
A high-resolution hindcast of wind and waves for the North Sea, the Norwegian Sea, and the Barents Sea
.
J. Geophys. Res.
,
116
,
C05019
, doi:.
Tang
,
W.
,
W. T.
Liu
, and
B. W.
Stiles
,
2004
:
Evaluation of high-resolution ocean surface vector winds measured by QuikScat scatterometer in coastal regions
.
IEEE Trans. Geosci. Remote Sens.
,
42
, 1762–1769, doi:.
Undén
,
P.
, and Coauthors
,
2002
: HIRLAM-5 scientific documentation. Swedish Meteorological and Hydrological Institute, 144 pp. [Available online at http://www.hirlam.org/index.php/meeting-reports-and-presentations/doc_download/308-unden-et-al-2002.]
Uppala
,
S. M.
, and Coauthors
,
2005
:
The ERA-40 Re-Analysis
.
Quart. J. Roy. Meteor. Soc.
, 131, 2961–3012, doi:.
Vachon
,
P. W.
, and
J.
Wolfe
,
2011
:
C-band cross-polarization wind speed retrieval
.
IEEE Geosci. Remote Sens. Lett.
,
8
, 456–459, doi:.
Verhoef
,
A.
, and
A.
Stoffelen
,
2013
: ASCAT coastal winds validation report. Version 1.5, Royal Netherlands Meteorological Institute and Ocean and Sea Ice SAF Tech. Note SAF/OSI/CDOP/KNMI/TEC/RP/176, version 1.5, 19 pp. [Available online at http://www.knmi.nl/scatterometer/publications/pdf/ASCAT_validation_coa.pdf.]
Vogelzang
,
J.
,
A.
Stoffelen
,
A.
Verhoef
, and
J.
Figa-Saldaña
,
2011
:
On the quality of high-resolution scatterometer winds
.
J. Geophys. Res.
,
116
,
C10033
, doi:.
Yang
,
X.
,
X.
Li
,
W. G.
Pichel
, and
Z.
Li
,
2011
:
Comparison of ocean surface winds from ENVISAT ASAR, MetOp ASCAT scatterometer, buoy measurements, and NOGAPS model
.
IEEE Trans. Geosci. Remote Sens.
, 49, 4743–4750, doi:.
Zhang
,
B.
, and
W.
Perrie
,
2012
:
Cross-polarized synthetic aperture radar: A new potential measurement technique for hurricanes
.
Bull. Amer. Meteor. Soc.
, 93,
531
541
, doi:.