Update on Antenna Elevation Pattern Estimation from Rain Forest Data Manfred Zink ENVISAT Programme, ESA-ESTEC Keplerlaan 1, 2200 AG, Noordwijk The Netherlands Tel: +31 71565 3038, Fax: +31 71565 3191 Manfred.Zink@esa.int 1 ABSTRACT Antenna elevation pattern estimation is based on acquisitions over a large homogeneous rain forest area in the Amazon basin. The estimation procedure accepts IMP, APP and WSM products as input and generates a combined estimate from as many input products as are available for a certain beam. For IS3-IS6 IM and WS products can be combined as well. Beam patterns for vertical polarization have been presented at the ENVISAT Calibration Review (ECR) [1]. Here we present some fine-tuning of the WS VV beam patterns at the swath boundaries and the patterns in HH, HV and VH polarizations. Comparison with the pre-launch patterns versus time in the receive window allows verifying the compensation of the receiver gain droop. Pattern corrected range profiles have been analysed with respect to beam-tobeam radiometric normalization. 2 ANTENNA PATTERN ESTIMATION APPROACH Pattern estimation from several products requires a three-step approach (for details see [1]): Range power profiles are calculated applying a simple threshold based masking of non-homogeneous areas like rivers and clear-cuts. The antenna pattern correction is then reversed using the pattern applied during processing and the range power profile is multiplied with the tangent of the incidence angle. The resulting gamma profile is only modulated by the elevation antenna pattern, which in turn can be estimated under the assumption that gamma profiles are flat for rain forest. Individual profiles are combined to one cloud of estimates versus elevation angle. The final step is a least squares polynomial fit to the estimates. Outside the imaged swath we use the pre-launch pattern to extrapolate the patterns to ±5 deg. Proc. of Envisat Validation Workshop, Frascati, Italy, 9 13 December 2002 (ESA SP-531, August 2003)
3 RAIN FOREST AREA AND ACQUISITIONS Fig. 1. Rain forest area used for antenna pattern estimation, the red marked area was used for ERS. The rain forest we are using for antenna pattern estimation covers a latitude/longitude range of 4-11 deg South and 65-71 deg West respectively. We selected this rather large area based on information from RADARSAT and on the Rain Forest Mosaic produced from JERS-1 data. Fig. 1 shows the area in an ESOV plot indicating one of the IS2 acquisitions in orbit 973. The area used for ERS antenna pattern determination is shown as a red rectangle. Compared to ERS we had to increase the extent of the area because on ASAR we have to determine 32 beam patterns. In one 35-day repeatcycle there are in total 30 acquisition opportunities. Fig. 2 shows some typical IMP, APP and WSM products. The area is mostly flat and quite homogeneous with a few rivers and some clear-cuts. IMP&WSM products, HH - polarization
APP products, VH - polarization APP products, HV - polarization Fig. 2. Examples of IMP and WSM products in HH and APP products in VH and HV polarization over the selected area in the Amazon basin
4 FINE-TUNING OF WS VV PATTERNS AT THE SWATH BOUNDARIES In ECR we reported problems at the sub-swath boundaries in the case of WSM products. We used only data samples well away from the beam-merging region (about 0.5deg on either side). On the other hand the beam patterns at the transition from one sub-swath to the other are very critical for WS beam merging. Consequently all WS beams need further fine-tuning at the beam boundaries (see for example the far range edge of the SS1 beam in fig. 3). Fig. 3. Rain forest estimates (green), in-flight antenna pattern as polynomial fit (red) and pre-launch pattern (blue) for SS1-VV; vertical axis: two-way gain in [db], horizontal axis: off-reference angle [deg]; left: first estimate not using the data close to the beam merging region, right: improved estimate using all available data per subswath; red circles indicate the change at the far range edge of SS1. Fig. 4. Range power profiles for WSM VV products; left/right: before/after fine-tuning of the beam patterns.
The consequence of not using the data close to the beam merging region is a too steep drop of the beam patterns in near and/or far range and as a result an over-compensation of the beam pattern especially for the far range edge of SS1 and the near range edge of SS2 and SS3 (fig. 4, left). After applying the refined patterns the range power profiles show consistent slopes in the beam merging regions (see fig. 4, right). Please note the change in vertical scale for the plots in fig. 4. Furthermore between the processing of the WSM products used in the left and right plots of fig. 4 the processing gain from beam to beam has been adjusted resulting in an improved normalization across the whole WS image. The absolute levels are in good agreement with the calibration factors from the transponders and the expected σ o for the rain forest. 5 ESTIMATED ELEVATION BEAM PATTERNS FOR HH, VH AND HV POLARIZATIONS Figs. 5 and 6 show some of the estimated elevation beam patterns for HH, VH and HV polarization. The numbers in the title of the plots indicate the orbit numbers of the products used. If the same orbit number appears more than once several non-overlapping IMP or APP products processed from one long acquisition have been used. For HH polarized beams IS3 to IS6 only WSM products have been used in the refined approach. The green cloud represents the combined range power profiles from all products. The red curve is the final antenna pattern obtained by piecewise least squares fitting of 3 rd order polynomials to the green cloud of estimates. The polynomial pieces have been merged to obtain a smooth beam pattern over the whole angular range. Pre-launch patterns are show as blue curves. Range power profiles from the different products used agree very well with each other and the standard deviation of the green estimates is below 0.1dB as for the VV patterns [1]. Estimated patterns (red curves) deviate by up to 0.5dB from the pre-launch measurements (blue curves). Differences are most likely due to pre-launch measurement artifacts. Fig. 5. Rain forest estimates (green), in-flight antenna pattern as polynomial fit (red) and pre-launch pattern (blue) for HH polarized beams IS1, IS3, IS5, SS1; vertical axis: two-way gain in [db], horizontal axis: off-reference angle [deg]
Fig. 6. Rain forest estimates (green), in-flight antenna pattern as polynomial fit (red) and pre-launch pattern (blue) for VH polarized beams IS2, IS4, IS5 and HV polarized beams IS1, IS2, IS7; vertical axis: two-way gain in [db], horizontal axis: off-reference angle [deg]
The good agreement of range power profiles from different acquisitions is a further indicator of the stability of the antenna and the low standard deviation of the estimates confirms the high performance of the pattern estimation from rain forest data. Furthermore we can conclude that one single product can provide a reasonable estimate for the beam pattern. This was important for the cross-pol patterns, where in some cases only one product was available (see fig. 6). For HV polarization and beams IS3 and IS6 we have not received any product yet. Patterns synthesized from Module Stepping Mode measurements are currently being used to correct these beams in the processor. 6 COMPARISON WITH PRE-LAUNCH PATTERNS Fig. 7 shows the difference between pre-launch and in-flight patterns versus time in the receive window for VV polarization. Any common trend would indicate inaccurate knowledge of the receiver gain droop function. No such trend is visible and we conclude that the receiver gain droop has been well characterized pre-launch. It is also a verification of the gain droop compensation in PF-ASAR. Any residual receiver gain droop effects do not affect our products. Fig. 7. Difference between pre-launch and in-flight patterns (VV polarization) versus time in the receive window To further understand the difference between pre-launch and in-flight antenna beam patterns we plotted this difference for IS1, IS4 and IS7 in fig. 8. These plots include all 4 polarizations and we see similar behavior for different polarizations within one beam but quite different shapes from beam to beam. The main reasons for these differences are: The accuracy of pre-launch measurement does not match highly accurate RF estimates. The limited control/knowledge of module gain/phase settings during pre-launch characterization The limited accuracy of the pre-launch characterisation measurements also impacts on the patterns synthesis from Module Stepping and internal calibration data (see below).
IS1 IS7 IS4 Fig. 8. Difference in db between pre-launch and in-flight antenna elevation beam patterns as a function of offreference angle for beams IS1, IS4 and IS7: HH (solid ), VV (dotted), VH (dashed), HV (dash-dot) 7 PATTERN CORRECTED RANGE POWER PROFILES Calibration pulse measurements are used to derive the elevation gain for product normalization. After correction of the in-flight antenna pattern we would expect flat range power profiles within each swath and, if the internal calibration is able to track the gain difference from swath to swath, even across the whole incidence angle range from IS1 to IS7. VV HH Fig. 9. Range power profiles after correction of the in-flight antenna pattern for IMP products in beams IS1-IS7
Fig. 9 shows the pattern corrected range power profiles for all IMP products analyzed in all 7 beams and for both VV and HH polarizations. Within one swath the profiles are quite flat and the different measurements are within 0.5dB. From swath to swath we see jumps of up to 1dB. This means that the internal calibration does not completely track the gain change from beam to beam and we believe it is due to the limited accuracy of the pre-launch characterization data (embedded row patterns were only measured on an incomplete antenna array - 14 out of 20 tiles). The variation across all swath is different for different polarization (slightly higher in HH) due to the polarization dependence in the antenna characteristics. Absolute levels are consistent with calibration factors from transponder measurements and the expected σ o of the rain forest. Profiles for IS1 are above all the others in both polarizations. In [1] we concluded, that this might be due to terrain effects but we now believe that the flatness of the γ-profiles across the whole incidence angle range has to be confirmed. Fig. 10. Range power profiles after correction of the in-flight antenna pattern for HH polarized WSM products Fig. 10 shows similar plots for the WSM HH polarized products. The range power profiles are flat across the whole incidence angle range and the slopes are consistent at the beam boundaries. These products have been processed using the processing gains optimized for VV polarization. As explained above, the gains are polarization dependent and we therefore see some increase in the variation across the whole images. Finally fig. 11 contains all the pattern corrected range power profiles from APP products. These products where generated in a very hectic period of processor adjustments, which we could not fully recover. We therefore discarded the profiles, which where completely off from the rest. As for IMP and WSM products the individual profiles are flat and we see similar variations in gain from beam to beam. The absolute levels are close to the IMP levels in fig. 9 above despite the different processing algorithm (SPECAN versus Range-Doppler). VV profiles are approximately 2 db lower than HH profiles. Cross-polar returns are about 7 db below co-polar measurements.
Fig. 11. Range power profiles after correction of the in-flight antenna pattern for APP products 8 CONCLUSIONS Our antenna elevation pattern estimation from rain forest acquisitions is based on the combination of IMP, APP and WSM products. Range profiles from different IMP, APP and WSM products are very consistent with a standard deviation of better than 0.1dB. Because of this excellent agreement we could obtain reasonable patterns from single products in AP mode. HV IS3&IS6 patterns are still missing as we have not received any products in these modes yet. Fine-tuning of the WS beam patterns at the swath boundaries has been performed successfully resulting in consistent slopes of the range power profiles in the beam merging regions. Differences to the pre-launch patterns reach up to 0.5dB and show no systematic trend versus time in the receive window. This confirms the pre-launch receiver gain droop characterization and the compensation in the processor. Differences are due to limitations in the pre-launch characterization measurements. Pattern corrected range power profiles are within 0.5dB for one beam, but beam-to-beam variations can reach up to 1dB. The internal calibration does not fully track this gain differences. The most likely reason is the limited accuracy of the pre-launch characterized embedded row patterns, which was performed on an incomplete antenna array (only 14 from the 20 tiles). 9 REFERENCES [1] M. Zink & B. Rosich, Antenna Elevation Pattern Estimation from Rain Forest Acquisitions, Proceedings ENVISAT Calibration Review, ESTEC, 9.-13. September 2002 [2] R. Torres, ASAR On-board Instrument, Proceedings ENVISAT Validation Workshop, ESRIN, 9.-13. December 2002