1a) 1-km resolution NOAA-11 AVHRR IR image acquired at 13:03 UTC
1b) 100-m resolution ERS-1 SAR image acquired at 12:31 UTC.
The temporal and spatial scales which characterise marine boundary layer processes are from seconds to days and meters to hundreds of kilometres, implying that satisfactory observations by conventional in-situ techniques are both difficult and expensive. However, satellite remote sensing observations of sea-surface quantities and processes over a wide range of spatial and temporal scales are capable of improving the situation. Under cloud-free conditions, the sea-surface temperature (SST) field with a resolution of about 1 km may be retrieved from the IR channels of the advanced very high resolution radiometer (AVHRR) via the sensed emitted brightness temperature (Steward, 1985). Under cloudy conditions the SST signal is lost.
Instead, the IR provides a measure of cloud-top temperatures which can aid in assessing weather system movement and physical processes in the atmospheric boundary layer.
Unlike the AVHRR, active microwave sensors such as the synthetic aperture radar (SAR) on ESA s remote sensing satellite ERS-1, launched in July 1991, are not affected by cloud cover and light conditions as it supplies images of the ocean- surface roughness distribution with a resolution of 25 m. The SAR directly measures the surface roughness at the scale of the radar wavelength through Bragg resonant scattering (Wright, 1978). The challenge is to interpret uniquely the SAR images in terms of physical processes in the upper ocean and the atmospheric boundary layer which modulates the surface roughness (ESA SP-359, 1993 and ESA SP-361, 1994). Coincident information from other spaceborne sensors are valuable in solving this problem. In this paper, we examine two such cases which demonstrate the capability of SAR to provide information on atmospheric boundary layer and mesoscale upper ocean processes.
A good qualitative relationship between expressions of boundary layer cloud structure and surface roughness pattern is demonstrated in Figure 1a-b. The NOAA AVHRR IR image depicts the cloud structure evolving downwind from the ice edge in the Greenland Sea. The cloud structures are associated with horizontal roll vortices in the atmospheric boundary layer (LeMone, 1973). Rolls are frequently observed in IR and visible images downwind from the ice edge, being formed in response to the convective mixing of heat and moisture in the unstable boundary layer due to cold air flowing over warmer water. The clouds are formed where the upward rising air condenses. The mean roll spacing is 1a 5 km. This spacing is reported to typically range from 2 to 4 times the planetary boundary layer height (LeMone, 1973; Brown & Liu, 1982). The boundary layer height should therefore range from 1.25 to 2.5 km. Atmospheric soundings from Jan Mayen Island, on the other hand, indicate a boundary layer height which is less than 1.0 km. Although this station is within the same weather regime (Fig. 1c), it is more than 400 km away. This could allow for the discrepancy.
The SAR image (Fig. 1b), acquired 30 minutes prior to the AVHRR image, reveals the corresponding surface roughness in the open water off the ice edge. The sea-surface roughness pattern is streak-like with an orientation aligned in the direction of the roll vortices seen in the IR image. The mean streak spacing is 5 km, in agreement with the estimate from the IR image. The short surface waves that produce this pattern are formed in response to variations in the wind stress. From the relationship between horizontal and vertical winds in convective roll vortices (LeMone, 1973) we expect the dark (cloudy) regions in the IR image to be in phase with the dark (lower wind speed) regions in the SAR image. In turn atmospheric phenomena that induce a varying, sub-mesoscale sea-surface wind field, and hence wind stress, are detectable by the SAR. Several regions of reduced radar backscatter are also observed. We interpret these to be areas of grease ice that damp the short gravity waves. Similar spaceborne SAR observations of wind streaks associated with horizontal roll vortices have been reported previously (Thompson et al., 1983; Alpers & Brummer, 1994), but this is the first time a radar image is complemented with a coincident IR image.
High-resolution wind velocity estimates near the ice edge and in coastal waters are difficult to obtain. For example, the ERS-1 wind scatterometer has a resolution of 50 km. Wind vectors derived from calibrated, 100 m resolution, SAR images can significantly improve this, in particular if wind direction retrievals are reliable. In this case the wind direction is estimated from the orientation of the wind streaks using spectral analysis. The wind speed is then estimated (Fig. 2, bottom) from the radar cross-section using CMOD4 wind retrieval model (Stoffelen & Anderson, 1993). The mean wind speed is found to be 8.0 ñ2 m/s which compares well with the measured wind speed at Jan Mayen of 7.5 m/s. In this case, the wind speed variation induced by the convective motion has an amplitude of 1 m/s.
The corresponding weather map is shown in Figure 1c. This is a classical winter situation in which cold Arctic air masses are transported southward towards Europe. The streak pattern orientation is seen to be aligned along the isobars in the geostrophic wind direction. The air temperature at Jan Mayen is reported to be 16 C, while the average cloud top temperature is about 17 C (Fig. 2, top). In the vicinity of the ice edge where the sea-surface temperature is expected to be 1.5 to 1.75 C, the unstable stratification is therefore about 15 C, leading to a bulk heat flux from the ocean to the atmosphere (Liu, 1990)of 160 W/mý. These results demonstrate that SAR images, in principle, carry qualitative and quantitative information about the surface wind field and fluxes in the atmospheric boundary layer.
Figure 1. Images obtained on 16 January 1992 off the ice edge in the Greenland Sea.
1a) 1-km resolution NOAA-11 AVHRR IR image acquired at 13:03 UTC
1b) 100-m resolution ERS-1 SAR image acquired at 12:31 UTC.
1c) the corresponding surface analysis (provided by the Norwegian Meteorological Institute) from 12:00 UTC with the coverage of the IR and SAR images indicated. The 100 200 km region of SAR coverage is
indicated in the NOAA image. In the IR image, white corresponds to warm temperatures while
dark corresponds to cold (cloud-top) temperature. In the SAR image, white corresponds to
rougher regions associated with increased wind speeds while dark corresponds to smoother
regions of lesser wind speeds. In both images the roll spacing, measured using spectral analysis,
was found to be 5-km near the ice edge.
Examples of IR and SAR manifestations of sea-surface temperature and roughness fields obtained off the west coast of Norway, are compared in Figure 3a-b. In the IR image, the surface temperature decreases from nearly 14 C (white) in the coastal water to 12 C (dark blue) in the Atlantic water offshore. The weak temperature contrast is typical for autumn. The maximum temperature gradient is about 0.6 C/km. Previous observations across such temperature fronts have established the presence of corresponding salinity and density fronts, which combine to maintain a baroclinic current boundary. The structure of the sea-surface temperature field with the curvilinear temperature fronts, represents mesoscale current variability of 10 to 50 km scale, characteristic of the unstable Norwegian Coastal Current (NCC) (Johannessen et al., 1989).
The ERS-1 SAR image (Fig. 3b), acquired 7 hours later, contains frontal features at a scale, configuration and orientation in good qualitative agreement with those seen in the IR image. It clearly suggests that the SAR can image current boundaries, including meanders and eddies (Johannessen et al., 1994a). Based upon the surface weather analysis, winds were northerly at 5 m/s with air temperatures from 12 to 14 C reported along the coast. Furthermore, a northward near surface current of 0.30 m/s, a constant temperature of about 13.5 C in the upper 20 m of the water column, and a significant wave height of 1 m were reported at the satellite overpass time from a moored buoy which was deployed about 20 km offshore (Fig. 3b).
The air/sea temperature difference indicates near neutral stability. We might thus expect that backscatter variations could be modulated by wind stress variations induced by changes in boundary layer stratification (Brown, 1990) in the vicinity of the frontal boundary. An increase in surface-wind stress due to a drop in the stratification will cause an increase in surface roughness and in turn in radar cross section. This is examined in Figure 3c by relating the observed radar cross-section to the wind speed using CMOD4 which assumes neutral stratification (Stoffelen & Andersen, 1993). A mean wind speed of about 6.2 m/s is obtained, in good agreement with the weather reports. By subtracting this mean wind speed and corresponding radar cross-section we can relate the remaining radar cross-section fluctuations to the atmospheric stratification (Wu, 1991)as shown in Figure 3d. Assuming constant air temperature, we consider the sea-surface temperature variations to represent the fluctuations in stratification. The results suggests that the bright (1 dB increase), narrow peak at 25 km ground range, would require almost 6 C unstable stratification to be totally attributed to a wind stress change. Moreover, the second bright peak at 80 km ground range would require about 12 C unstable stratification. This appears to be too much according to the weather analyses. We therefore interpret the SAR image expressions to be primarily a manifestation of wave-current interaction for short gravity waves along the current fronts (Johannessen et al., 1994a). As the short gravity waves, i.e. from centimeter waves to meter waves, propagate across the current front, they change steepness and propagation direction. The combined effect, in turn, leads to the increase in radar cross section. The width of the peaks are several kilometers, and in particular near the first peak the transition from stable (+1 C) to neutral stratification is seen to occur. However, this gentle variation in the stratification does not seem to increase the wind stress. The lack of detailed information of the air temperature and wind speed limits further examination of this. The level of agreement between the edges found in the two images can be calculated from the cross-correlation function between the two images using edge enhancement. Maximum cross correlation values between the edge enhanced (wavelet filtered) SAR and AVHRR image was about 70% for a subimage extracted from the filtered SAR image with a 18-km dimension (Johannessen et al., 1994b). This high value indicates that the same basic oceanic process, or feature, is being imaged by both sensors supporting the interpretation that the two images have the surface current boundaries in common.
In summary, these images suggest that in spacebome monitoring of mesoscale coastal ocean circulation, the SAR-derived current boundaries can replace the IR-derived sea-surface temperature fronts in the event of total cloud cover. Under conditions with partial cloud cover, the SAR can be used to connect sea-surface temperature fronts masked by the clouds. In general, the local wind speed will restrict the ability of SAR to detect surface-current related features. Winds between 3 and 10 m/s are favourable for surface current feature detection by SAR (Johannessen et al., 1994a). Therefore, combined use of near coincident SAR and IR images, both improve and optimise spaceborne observations of mesoscale ocean circulation features.
Figure 2. Top: A profile of temperature extracted from
the AVHRR image perpendicular to the roll orientation. The temperatures range from 3 C
between the dense clouds (bright) to 17 C at the cloud tops (dark). Bottom: A profile of radar
cross section extracted from the SAR image perpendicular to the streak orientation. The radar
cross section based upon CMOD4, the ERS-1 scatterometer wind retrieval model has been
plotted on the same coordinates for several wind speeds by assuming that the wind direction is
off ice and oriented along the streaks. According to CMOD4, the mean wind speed is about 8.0
m/s, while the wind speed fluctuations induced by the convective rolls are ñ1 m/s. The radar
cross-section error bar, shown in the upper right corner of the plot, is rather small due to the
spatial averaging which was carried out along the streaks to produce the profile.
There is a growing need to monitor and manage the coastal environment. Since in-situ measurements of marine boundary layer processes at sufficient spatial and temporal scales are not practical, it is recognised that multisensor remote sensing data will be an important element. One of the candidate sensor is SAR, despite that we do not fully understand the SAR imaging mechanisms over the ocean.
The modulation of short gravity waves under a mixture of surface conditions related to wave-wave interactions, surface currents, near surface wind field and conditions in the atmospheric boundary layer is complex and so is the interactions with electromagnetic waves which are used to form the images. A way to advance the qualitative and quantitative interpretation of SAR images, in particular when in-situ data are lacking, is by utilisation of other independent, coincident remote sensing data. For example, the synoptic IR observations of sub-mesoscale cloud pattern and/or sea-surface temperature field at spatial resolution of about 0.5 km importantly contribute to the SAR interpretations of the mesoscale near surface wind field and surface current pattern. This is clearly demonstrated in these comparisons between coincident SAR and IR images.
We expect that improved SAR image interpretations, in particular for process studies related to oceanic frontal dynamics and air-sea interactions along seasonal ice edge zones and coastal zones, will provide a vital contribution to advance the development of marine environmental monitoring and prediction systems. By gaining better understanding of air-sea interaction and atmospheric boundary layer processes at spatial scales of a few kilometers the characterisation of heat and moisture fluxes as well as gas exchange may moreover advance. The improved observation capabilities offered by the 500 km swath width of Radarsat and Envisat are therefore promising for such process studies including its role in climate studies.
We thank P.M.Haugan, E. Jansen, I. Robinson and B.A. Farrelly for helpful discussions, and G. Digranes and K. Kloster for conducting the image processing. Part of this work was carried out while J.A. Johannessen was at Nansen Environmental and Remote Sensing Centre (NERSC) and P.W. Vachon was visiting the NERSC with financial support from the Norwegian Research Council Visiting Research Fellowship and Canada Centre for Remote Sensing. The work was also supported by Norwegian Space Centre, European Space Agency, US Oceanographer of the Navy/Naval Research Laboratory-South.
Figure 3. Images obtained on 3 October 1992 off the west coast of Norway between 59 and 62 N.
3a) 1-km resolution NOAA-11 AVHRR IR image acquired at 14:20 UTC.
3b) 100 mresolution ERS-1 SAR image acquired at 21:35 UTC. Both
images cover the same 100 300 km region (the Norwegian coast is visible along the right-hand
edge of each image). In the IR image, white corresponds to a water temperature of nearly 14 C,
while dark blue corresponds to colder temperature of about 12 C. The temperature accuracy is
expected to be about 0.5 C. The sea-surface temperature of 13.5 C reported from the Seawatch
Europe buoy off Sotra Island (60 13.2 N, 4 44.4 E, location marked +) was used to retrieve the
absolute surface temperature in the IR image based on the assumption of a linear relationship
between sea-surface temperature and image brightness temperature. The land has been masked
in green and clouds in black.
3c) Transect of radar cross-section (extracted from the SAR image
along the line indicated) plotted using coordinates based upon CMOD4 for several wind speeds
by assuming that the radar look direction is cross wind. The mean wind speed is about 6.2 m/s.
The error bar for the cross-section measurement is indicated in the upper right of the plot.
3d) Changes in relative cross-section interpreted in terms of atmospheric stability by assuming a
constant wind speed across the swath of 6.2 m/s (10 m height neutral stability) and a constant air
temperature of 14 C. The sea-surface temperature profile from the AVHRR is shown relative to
14 C (or neutral stratification).
Just Published: ESA SP-365
Proceedings of the 1st ERS-1
Pilot Projects Workshop
(Toledo, Spain, 22-24 June 1994)
ESA SP-365 (ISBN 92-9092-302-4)
October 1994, 525 p. (+70 papers)
100 Dfl, 300 FF, 50US$
ESA EOQ Nr. 46.