QuikSCAT Level 1C Dataset Known Issues and Updates

QuikSCAT Level 1C Dataset Known Issues and Updates

Postby dmoroni » Tue Aug 23, 2016 3:25 pm

The QuikSCAT Science Data Systems team has recently delivered an updated batch of approximately two more months of data for the Level 1C dataset, bringing the dataset time series up to 12 August 2016 (calendar day 255), as made available here: https://podaac.jpl.nasa.gov/dataset/QSC ... 0_WINDS_V1

Now disclosed for public awareness is a subtle difference in the wind vector retrieval between spinning scatterometers (e.g. RapidSCAT, QuikSCAT before 2009) and QuikSCAT L1C (QuikSCAT nonspinning configuration). This difference resulted in a change to the Geophysical Model Function (GMF) that went into effect for the QuikSCAT L1C dataset starting with orbit 87727. Earlier orbits have less consistent wind speeds with systematic errors of 0.3 to 0.4 m/s.

The new QuikSCAT GMF (hereafter, QSNS2016a) was developed from nonspinning QuikSCAT data. This new GMF differs from the NSCAT2014 GMF used to retrieve the publicly available RapidScat and spinning QuikSCAT winds in two primary ways. First, there is a constant Normalized Radar Cross Section (NRCS) bias of -0.25 dB (-5.9% linear scale) with QSNS2016a producing lower NRCS values. This difference is due to a calibration adjustment applied to QuikSCAT NRCS that was consistently used in the Remote Sensing Systems (RSS) wind retrieval, and GMF development but was missing from JPL wind retrievals until recently. The second difference is that the azimuthal modulation in QSNS2016a is decreased by several tenths of dB (varying with wind speed). This reduction is consistent with differences between how winds are retrieved for nonspinning and spinning data.

For this unique method of retrieving a wind vector from a fixed-beam scatterometer, one needs a GMF that models the conditional expected value of NRCS given true speed and direction. One first estimates the expected NRCS value given speeds and directions from some sensor or NWP (e.g., WindSAT and ECMWF), and then one increases the directional modulation to account for a reduction in modulation due to errors in the NWP directions. However, for nonspinning QuikSCAT, we cannot independently estimate wind direction. Instead, we retrieve speeds using the ECMWF directions as inputs along with the QuikSCAT NRCS data. In this case, no correction for azimuth modulation needs to be applied, because it is precisely the conditional expectation of NRCS on ECMWF direction that is needed to consistently retrieve wind speed.

appendix_c_figure1.png
Figure 1: Random error in triple collocations QuikSCAT, RapidScat, and ECMWF wind speeds.
appendix_c_figure1.png (182.9 KiB) Viewed 1299 times

Figure 1 (See above) compares the RMS wind speed error between WindSAT and QuikSCAT/RapidSCAT/ECMWF over time for the two GMFs. Figure 2 (See below) compares wind speed bias and RMS errors strictly between QuikSCAT and WindSAT over specific years for the two GMFs. The biases and scaling factors estimated by the WindSAT/QuikSCAT/ECMWF triple colocation analysis was significantly dependent on the GMF version. For colocations within 10 minutes, when it was presumed that WindSAT was unbiased and unscaled, QuikSCAT speeds retrieved using NSCAT2014 (i.e., GMF labeled "V1") had a bias and scaling factor of -0.3467 m/s and 1.0166, respectively for HH and -0.2848 m/s and 1.0106 for VV. When the QSNS2016a GMF (i.e., GMF labeled "V2") was applied, biases were reduced and the scaling factor was brought closer to 1 (with bias and scaling factor of -0.0155 m/s and 0.9950 for HH and -0.0032 m/s and 0.9965 for VV). The simultaneously-determined ECMWF offset and scaling factors were also made more consistent between HH and VV cases when the newer GMF was applied. The QNS2016a GMF substantially improves the agreement between WindSAT and QuikSCAT wind speeds.

qscat_l1c_speed_bias_rms_windsat.jpeg
Figure 2: QuikSCAT L1C wind speed bias and RMS error relative to WindSAT for version 1 and version 2 of the GMF. The left panel represents the HH-polarized data while the right panel represents the VV-polarized data.
qscat_l1c_speed_bias_rms_windsat.jpeg (47.38 KiB) Viewed 1299 times
dmoroni
 
Posts: 33
Joined: Wed Sep 03, 2014 1:24 pm

Return to Ocean Wind