During an ERS-1 acquisition on 4 April 1996, 10:34 UTC (orbit 24688, frame 2547), convective rolls occurred in the boundary layer [Hanssen et al.(1999)]. In the interferogram with an ERS-2 acquisition on 5 April 1996 (orbit 5015), see Figure 3, these rolls are clearly visible over the land areas. Since no clouds were observed at the surface stations in the area, the delay variation is a clear air effect.
Several studies have indicated that the ascending parts of the rolls transport warm, moist air at the surface upward, while the descending parts transport dry air originating witin the inversion downward, see, e.g. LeMone [1973]. Weckwerth et al. [1996] observed variations in the water vapor mixing ratio of 1.5-2.5 g/kg associated with rolls, whereas temperature variations were only 0.5 K. Mapped to water vapor pressure, these moisture variation are 2.4-4 hPa. The sensitivity of the delay to a 1 hPa change in water vapor pressure is in this situation at least 3.5 times as large as to a 1 K change in temperature. The delay variations of up to 10 mm are therefore dominantly caused by water vapor differences in the updraft and downdraft branches of the rolls. Not more than 5% of the observed delay can be explained by the temperature variations in the rolls. The rolls are approximately parallel to the wind direction, and the wind speed observed at several surface stations is 6.2-7.7 m s-1.
Over the water areas in IJssel lake, wind streaks are clearly visible, oriented in the same direction as the rolls over land areas. Although streaks caused by rolls are frequently observed in SAR images, see Alpers and Brümmer [1994], Mourad and Walter [1996], the location of the streaks with respect to the downdraft and updraft branches has been uncertain. Spectral analysis reveals a streak spacing of approximately 2 km. Over land, the spacing between the bands is also approximately 2-3 km. A radiosonde in the area, launched 90 minutes after the SAR acquisition, indicated a boundary layer depth of 800 m, which gives an aspect ratio of 2.5-2.9, typically for rolls [Atkinson and Zhang(1996)].
For wind speed estimation with a ±1-2 m s-1 accuracy, an averaging area of 25 km2 is needed [Lehner et al.(1998)]. Using square areas of 10 ×10 km, average wind speeds of 5.3-5.9 m s-1 were found. For the analysis of the wind speed differences between the streaks, square areas of this size cannot be used. Instead, we applied a normalized Radon transform to two areas of 11 × 11 km and used a profile perpendicular to the streaks to estimate the average differences in 
. This method reduces speckle effects significantly and enables direction dependent amplitude estimation. Variations over a range of 0.4-0.5 dB are observed, corresponding with wind speed variations of 1 m s-1 or wind direction variations of up to 45 degrees (i.e., the sensitivity to lateral wind velocity changes is about four times the sensitivity to transverse velocity variations).
We note the change in characteristics of the boundary layer rolls as they are advected over the cold water surface, where the fluxes of momentum and heat are expectedly quite different from those over land areas that are heated by the sun. The combination of measurements shown in this paper is ideally suited to show the transformation of air flow after advection over a land-water boundary.
The location of the streaks aligns approximately with the part of the rolls which have less delay, which correspond to the relatively dry and descending air. This is in correspondance with classical theory, which expects the maximum wind speed in a roll in the downdraft area [Alpers and Brümmer(1994), Atkinson and Zhang(1996)]. Midlatitude cases, without cloud and with important dynamical and thermal instability, such as presented here, are uniquely documented by the SAR methodology presented here.