JET PROPULSION LABORATORY INTEROFFICE MEMORANDUM 3348-99-008 June 16, 1999 To: From: CC: Subject: Philip S. Callahan Young-Joon Kim SAPIENT, SVT Validation of the NOAA Processor through a comparison with the standard SeaWinds on QuikSCAT Processor 1. Introduction In SeaWinds data processing, normalized microwave radar backscatter (aka sigma0) can be grouped differently depending on whether and how to weight the slices using the X-factor [1]. The SeaWinds on QuikSCAT processor uses either so-called egg method or composite (or in short comp1 ) method. The NOAA processor uses so-called composite of composite or composite squared (or in short comp2 ) method. This study compares the winds retrieved using sigma0 s grouped by using these three methods. The three sigma0 grouping methods do not always have data at the same points. For a fair comparison, the NOAA MGDR (Merged Geophysical Data Record) data are re-mapped into the Level 2B data points for collocation and only those points where all three methods have wind data are included in the comparison. 1 x 1 NCEP surface wind analyses gathered from observations within 3 hours apart are used to simulate the test data and also to nudge the retrieved winds. These true wind fields are again used for calculation of the wind error and the ambiguity removal skill. The ambiguity removal skill is defined as 100% when the closest wind vector is the selected vector or the selected vector is within 20 range of the closest vector. For this study, we use a recent version of the QuikSCAT test data [2]; Five orbits of simulated Level 2B data were processed by S. Craig using the SeaPAC operational software, and the NOAA MGDR data were processed by R. S. Dunbar. The main purpose of this study is to validate the NOAA processor with the focus on the performance of winds. The validation of sigma0 is reported elsewhere [3]. The data used in this study have a higher level of noise than those used in many other studies. Therefore, the results may be somewhat pessimistic and should only be used for comparison among the three grouping methods. 2. Results Figure 1 shows the mean and RMS differences of selected wind vectors in wind speed (from 3 to 30 m/s) among the three cases and the mean and RMS errors from the true winds, as a 1
function of cross-track distance. Figure 2 is as Fig. 1 except for the wind direction. The ambiguity removal skill shown in Fig. 3 is calculated for four wind speed subranges; 3-5, 5-7, 7-12, 12-30 m/s and the whole range of 3-30 m/s. A region of selected wind vectors from an orbit is compared in Fig. 4 for visual check of the vectors. Figure 5 shows the errors of the selected wind vectors compared with the true wind vectors. The difference in wind speed is larger between Comp1 and other cases with the typical mean magnitude of less than 1 m/s (Fig. 1). The difference between Egg and Comp2 is smaller than that of other combinations. Mean error from the truth is apparently the smallest for Comp1, but it is a result of sign cancellation as seen in the RMS. The RMS errors reveal that Comp2 has the best performance. The mean speed differences among the methods and from the truth are all less than 1 m/s. The RMS differences among the methods are well below 2 m/s except in a few bins in the far swath. The somewhat greater RMS from the truth is probably caused by the extra noise mentioned in the introduction. On the other hand, the difference in wind direction shows that Comp2 is also the best for most of the cross track except near the edges where it is among the worst (Fig. 2). The RMS direction errors show the typical SeaWinds shape across the swath and are less than 20 in the sweet zone (200-700 km). The ambiguity removal skill is generally higher for higher speed subrange (Fig. 3). Comp2 performs generally better than other methods except near the edges of some speed subranges and in average, which seems to be due to that the number of sigma0 s are limited to only two in far swath where inner beam drops out whereas the number is four in other regions. Visual comparison of the selected wind vectors reveals that the three grouping methods produce overall similar winds (Fig. 4). The errors from the truth (Fig. 5) suggest that the performance of the wind retrieval degrades near the edge of the swath and the cyclone as anticipated. 3. Summary and discussion The differences among the three sigma0 grouping methods are compared in terms of the retrieved winds. Five orbits of QuikSCAT test data are used for the comparison. The result of this study is in line with the sigma0 study by S. V. Hsiao [3] that showed the difference in sigma0 between Egg and Comp2 is less than that between Comp1 and Comp2. Our results that the errors of the retrieved winds are in general the smallest for Comp2 are in agreement with other independent studies performed by K. S. Pak and R. S. Dunbar, and thus are regarded highly convincing. It should be noted here that the cases Egg and Comp1 may have been affected by a minor error in the SeaPAC processor regarding the Kpr noise handling. It is believed, however, that the error is not serious enough to change the conclusion drawn in this study. In conclusion, Comp2 2
method, i.e., the NOAA Processor is highly favorably validated against the standard QuikSCAT / SeaWinds processor. 4. References [1] NASA/JPL Scatterometry Processing Algorithm and Analysis Group, Science Algorithm Specification for SeaWinds on QuikSCAT, JPL, May 1999. [2] Y. -J. Kim, R. S. Dunbar, K. S. Pak, S. V. Hsiao and P. S. Callahan, 1998: "Simulation of test data for QuikSCAT Level 1A and Level 0 processing". JPL IOM 334YJK-98-002, July 24, 1998 (available from <http://sapient.jpl.nasa.gov/paper/iom_334yjk-98-002.ps>). [3] S. V. Hsiao, 1999: QuikSCAT/SeaWinds Test Data Sigma0 Comparisons. JPL IOM 3348-99-007, June 15, 1999. 3
Fig. 1. The mean and RMS differences in wind speed (from 3 to 30 m/s) of selected wind vectors among the three sigma0 grouping methods averaged for five orbits of the test data, and the mean and RMS errors from the true winds, as a function of the crosstrack distance. 4
Fig. 2. As Fig. 1, but for the wind direction. 5
Fig. 3. The ambiguity removal skill, averaged for five orbits of the test data, of the three sigma0 grouping methods calculated for four wind speed subranges; 3-5, 5-7, 7-12, 12-30 m/s and the whole range of 3-30 m/s, as a function of the cross-track distance. 6
Fig. 4. A region of selected wind vectors of the three sigma0 grouping methods from orbit number 71. The longitudinal boundaries are those of the swath s. 7
Fig. 5. As in Fig. 4, but for the errors of the selected wind vectors from the true wind vectors. 8