QSCAT-KU2000 Curl

Global Annual Average Wind Stress Curl and Divergence from QSCAT: Comparisons between QSCAT, NCEP, and a Blended Product

Jan Morzel and Ralph Milliff (Colorado Research Associates/NWRA), Michael Freilich and Barry Vanhoff (Oregon State University)
Based on presentation at the NASA Oceanographic Scientific Conference, April 2001, Miami Beach, FL

Abstract

The global distributions of annual-average wind stress curl and divergence are compared for the calendar year 2000 (CY2000) using surface wind data from: a) QSCAT wind retrievals; and b) the standard NCEP analyses. This work extends the analyses based on NSCAT data in a recent paper by Milliff and Morzel (2000). Annual average wind stress curl estimates from QSCAT are shown to be biased smaller than NCEP in sub-polar regions of the Northern Hemisphere. Annual average wind stress divergence estimates from NCEP are plagued with Gibbs artifacts that are known to stem from the NWP spectral model. A hybrid wind product is presented based on enhancements of a blending method first developed by Chin et al. (1998). Here we blend QSCAT-Ku2000 and the NCEP surface wind analyses. The blended dataset consists of 6-hourly global maps of surface vector wind 0.5° x 0.5° resolution. The blending method ameliorates the wind stress curl bias in QSCAT data and the Gibbs artifacts in NCEP wind stress divergence.

Wind Retrieval Quality

The QSCAT mission provides global wind direction and speed observations over the ocean surface, beginning in July 1999. At a resolution of 25km, the measurement swath cross-section covers at least 1800km, or 72 wind vector cells (WVC), and occasionally 1900km (76 WVC). This represents almost 60% more data than was available during the NSCAT mission (September 1996 - June 1997), which scanned 600km on either side of the satellite with a nadir gap of 400km (48 WVC). Two QSCAT vector wind datasets are available: a) the QSCAT-1 product provided by JPL; and b) the QSCAT-Ku2000 product from Remote Sensing Systems. In addition to higher coverage, the QSCAT data also provide several new quality flags which allow the user to estimate the data reliability/quality: e.g. by knowing how well the retrieved radar signal matches the model function (from "SOSAL" in KU2000), or whether rain was detected in the WVC (from the rain flags in both products).

When extracting highest quality data, additional factors need to be considered. Due to the SeaWinds instrument design yields radar beam geometries that are not uniformly optimal across the wind vector cell swath. There are principally three areas in the QSCAT swath where the measurements of wind speed and direction are less reliable: the two outer edges of the swath and the nadir region. In general, the instrument receives four different radar reflections from the same patch of the ocean's surface. As the antenna spins at a rate of 18rpm, two radar pencil-beams scan the ocean with horizontal and vertical polarization (H-pol and V-pol beams). The inner H-pol beam scans with a slightly smaller radius than the outer V-pol beam. Each beam results in two signals from the same surface patch as the receiver looks ahead and back. The retrieved wind is most reliable when both polarization signals are available and when the fore and aft look angles are different from each other (preferably 90° apart). There are several regions in the satellite swath, however, where the reliability suffers: in the nadir region, where the two look angles are nearly 180° apart; in the outer regions, which are only scanned by the V-pol beam; and along the extreme edges of the swath, where the fore and aft looks are nearly in the same direction.

Below are two samples of typical wind vector swath data from QSCAT-Ku2000 and QSCAT-1. In both cases, about the same regions and numbers of WVC are associated with rain (light blue arrows). Also, clearly visible are regions of higher WVC-to-WVC variability: along the edges of the swaths and near nadir. In this and other examples we have examined, QSCAT-1 exhibits higher variability than QSCAT-Ku2000 everywhere in the swath. The high-variability nadir region is broader in QSCAT-1 than in QSCAT-Ku2000. It is not clear whether the higher WVC-to-WVC variability of QSCAT-1 with respect to QSCAT-Ku2000 is due to more noise in the former, or due to over-smoothing in the latter product.






The figure below demonstrates the same effects in an average sense. The CY2000 annual average RMS differences, NCEP minus QSCAT-1, across the swath are computed, where NCEP has been sampled to mimic QSCAT-1. As in the case of the swath snapshots above, the regions of highest RMS variability are in the far-swath and near-nadir. The figure demonstrates that the RMS differences in these regions are probably not idiosyncrasies of the surface wind distribution since there is no similar signal in the NCEP data (i.e. the differences are largest there). The higher RMS variability in the far-swath regions of the QSCAT data result from insufficient azimuth angle variability, and the fact that only the outer, vertical polarization, radar beam illuminates this region. Higher RMS variability in the nadir region of the swath is again due to sub-optimal azimuth angle diversity.



Blending QSCAT and NCEP

QSCAT provides high-wavenumber, but temporally intermittent data: each revolution takes 101min and covers a 1800km wide swath at 25km resolution. The NCEP fields are ubiquitous, but low-wavenumber: each global field is available every 6 hours on a T62 Gaussian grid (ca. 1.8° grid), but the true spatial resolution is coarser than T62. The blending scheme (adapted from Chin et al., 1998) creates 6-hourly global fields by retaining QSCAT wind retrievals in swath regions, and in the unsampled regions by augmenting the low-wavenumber NCEP fields with a high-wavenumber component that is based on monthly regional QSCAT statistics. These statistics are derived from 4° x 4° bin averages and preserve the observed power-law relation between power-spectral density (PSD) and wavenumber k: PSD ~ k p for each of the vector wind components u,v. The exponent p takes values between -2 at high latitudes and -5/3 at the equator.

Nearly uniform global coverage from QSCAT is achieved in 12hr composites. So each 6-hourly analysis field from NCEP is blended with a QSCAT 12hr composite centered on the analysis time. The two figures below depict typical 12hr QSCAT coverage (left) and the result of our quality control that limits the QSCAT data entering the blended wind product (right). For highest data quality, the following data were excluded from the blending scheme:

These quality-flagged WVC are colored sequentially in the left figure (i.e. later colors overwrite earlier colors). They are not included in the blended product, as are WVC from the nadir region identified above, are removed to yield the QSCAT-Ku2000 data distribution remaining in the righthand panel.



The following panels depict stages of the blending method for the analysis day 23.75 (January 24, 2000 at 1800 UTC). The left panel is the wind stress curl from NCEP. QSCAT wind stress curl is computed within each swath, from the highest quality data in the 12hr window centered on the analysis time. These wind stress curls overlay the NCEP wind stress curl in the middle panel. Finally, the blending method adapted from Chin et al. (1998) and described above is applied to create the hybrid field in the right panel. Color contour intervals are 20 x 10 -8 N m-3. The zero wind stress curl amplitude separates red from blue.



Annual Average Wind Stress Curl and Divergence

Annual average wind stress curl and divergence fields were computed from: NCEP; QSCAT-Ku2000; and the blended product. In the QSCAT-Ku2000 case, wind stress is accumulated in 0.5 ° bins, and the derivative fields are computed, orbit by orbit (see Milliff and Morzel, 2000). These fields are depicted here for the calendar year 2000.



The differences in annual average wind stress curl fields (lefthand column), between QSCAT-Ku2000 (top) and NCEP (middle), include results consistent with the analysis by Milliff and Morzel (2000) using NSCAT, as well as a surprising new distinction most evident at high northern latitudes. Qualitative comparisons with similar maps based on QSCAT-1 demonstrate results very similar to those obtained here with QSCAT-Ku2000. Consistent with the former work using NSCAT data, the spatial scales are finer and "patchy" in the QSCAT case. The QSCAT annual average resolves realistic wind stress curl features associated with: a) narrow cross-shore and extensive alongshore ocean basin eastern boundary extrema; b) narrow meridional and broad zonal features in the equatorial Pacific and Atlantic; and c) wind stress curl features associated with topography (e.g. islands, mountain gaps). These features are either missing or contaminated by model artifacts in the NCEP annual average. The implications for possible ocean general circulation model response has been investigated by Milliff et al. (1999), based on NSCAT winds.

The surprising difference between the annual average wind-stress curls from NCEP and QSCAT is the apparent bias toward negative wind stress curls in the sub-polar gyre regions of the Northern Hemisphere (an analogue bias toward positive wind stress curls in the high latitude Southern Hemisphere is also evident).

The implications of this bias for differences in implied Sverdrup transport of the sub-polar gyres remain to be explored. One possible explanation for the bias is that the rain-flagged WVC occur most often with atmospheric cyclone and associated frontal systems. These systems are sources of large positive and negative wind stress curls as well as rain. It has become apparent that the two, relatively steep incidence angles for QSCAT are more sensitive to contamination due to rain than was the broad range of incidence angles in the NSCAT system.

The annual average wind stress divergence maps (righthand column above) from QSCAT (top) and NCEP (middle) reiterate a result documented in Milliff and Morzel (2000) as well. Gibbs artifacts from the spectral model that underpins the NCEP analyses contaminate the global fields of wind stress divergence. Related features appear as oversized eastern and western boundary wind stress curl artifacts as well (lefthand middle). Wind stress divergence features in the equatorial Pacific QSCAT average exhibit large amplitudes, narrow meridional and very long zonal scales. Nothing of the kind is apparent in the NCEP case. The implications for equatorial upwelling in the ocean remain to be explored.

The annual average wind stress curl and divergence from the blended QSCAT/NCEP product are depicted in the bottom two maps. The blended fields exhibit high-wavenumber structures in the wind stress curl field; e.g. including realistic features such as narrow zonal bands in the equatorial Pacific and Atlantic. Gibbs artifacts from the NCEP wind stress curl are still evident, but reduced in offshore extent. Similarly, the spectral ringing signature in the NCEP wind stress divergence is largely replaced by physical signals from the QSCAT influence in the blended product. The blended wind product provides a useful compromise between ubiquitous but coarse and contaminated NCEP, and intermittent by high-resolution QSCAT surface wind products.

The blending method in its current implementation is particularly sensitive to high-wavenumber variability, causing us to be conservative in discarding scatterometer data prior to blending. This sensitivity might be reduced through modifications in the means by which regional spectral estimates used in the blending are computed. Such a change will allow us to supply more of the QSCAT winds, particularly from the nadir region, to the blending algorithm.

References

last modified on May 31, 2001
Go Back