QSCAT and NCEP Blended Winds

(Version 5.0, released in February 2008)


Summary

Global 6-hourly maps of ocean surface winds are derived from a space and time blend of QSCAT-DIRTH scatterometer observations and NCEP analyses. This blending method creates global fields by retaining QSCAT wind retrievals in swath regions, and in the unsampled regions (between swaths and in data gaps) augmenting the low-wavenumber NCEP fields with a high-wavenumber component that is derived from monthly regional QSCAT statistics.

6-hourly maps of 10m zonal and meridional wind-components (U and V), as well as wind stress curl, are available at a resolution of 0.5° x 0.5° resolution. The global coverage datasets begin in July 1999 and are updated periodically as long as the QSCAT mission continues. Data files are available from the National Center for Atmospheric Research (NCAR) Data Supprt Section (DSS): DS744.4 - QSCAT/NCEP Blended Ocean Winds. Fortran and IDL codes to read the binary data files are available from the same webpage.

A more detailed description of a blended product similar to this one (12 months of NSCAT/ERS and NCEP, August 1996 to July 1997) can be found in the appendix of Milliff et al. (1999), and the details of the blending methodology are the subject of Chin et al. (1998); see Referencs listed below. Here is a quick summary of the product and method.

Blended Data

The blended data set includes files for 10m surface wind-components U and V, as well as for windstress curl. There are three different products available: The output grid has a resolution of 0.5° x 0.5°, and spans from 88S to 88N. Land points are not set to some missing values, but instead the technique is applied everywhere: the NCEP analyses include wind values over land and the blending adds statistical high-wavenumber variability to this background field wherever there are no satellite observations. This way, the dataset can be used to force any ocean model, regardles of its particular configuration of the land/ocean mask. Caution is to be used, however, when using near-coastal values, as these grid points may be contaminated by land-values from QSCAT and NCEP.

Caution
This blended wind product was developed for general circulation scale analyses. It should not be used when mesoscale or ultra-high resolution is required. Moreover, each 6-hourly surface wind field is derived from the latest 12-hours of QSCAT observations (centered in time on the analysis time). This means that alternating halves of the globe retain the same QSCAT obs in each blended field output. Therefore, the blended winds are not suited for point-by-point, temporal comparisons (i.e. with buoy data), or when true 6-hourly resolution is needed.

Due to seasonal ice coverage, QSCAT data poleward of 60 degree is sparse. The NCEP re-analysis are of much lower quality in polar regions as well. QSCAT winds can be contaminated near ice edges resulting in very high wind speeds that are not always removed by quality control. Therefore, blended winds poleward of 60 degree latitude are of reduced quality, and gradient wind products, such as wind stress curl and divergence, cannot be trusted.

Input Data for Blending

QSCAT Input Data

The scatterometer data is the DIRTH product from NASA's SeaWinds on QuikSCAT mission. At a resolution of 25km, the QSCAT measurement swath cross-section covers at least 1800km, or 72 wind vector cells (WVC), and occasionally 1900km (76 WVC). 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 Dirth product was developed to improve the accuracy of retrieved wind directions in two portions of the swath, the far swath and nadir regions. For reference see Special Wind Vector Data Product: Direction Interval with Threshold Nudging (DIRTH) by Bryan W. Stiles, September 13, 1999.

"At far swath, ambiguity removal skill is degraded due to the absence of inner beam measurements, limited azimuth diversity, and boundary effects. Near nadir, due to nonoptimal measurement geometry, (fore and aft looking measurement azimuths approximately 180° apart) there is a marked decrease in directional accuracy even when ambiguity removal works correctly. Direction interval retrieval (DIR) addresses the nadir performance issue, and threshold nudging (TN) improves ambiguity removal at far swath" (Stiles). In normal processing, there are up to four ambiguities available at each wind vector cell (WVC). Median filtering selects one of the ambiguities based on the relative likelihoods. In areas where the likelihood values are relatively similar for a large range of direction, the DIR median filtering does not select just between four direction values but considers ranges of direction around the likelihood maxima.

For the purpose of blending, the outermost three WVC's along the outer edges of the satellite swath are excluded. Also excluded are all rain-flagged QSCAT data.

NCEP Input Data

The National Centers for Environmental Prediction (NCEP) analysis fields are the products of the NCEP Climate Data Assimilation System (CDAS), which was the operational system developed for the NCEP-NCAR reanalysis (Kalnay et al. 1996). The CDAS surface winds are available 4 times each day (at 0000, 0600, 1200, and 1800 UTC) on a Gaussian grid consistent with T62 resolution (i.e., triangular truncation, admitting 62 zonal wavenumbers). The grid is 1.875° lon x ca. 1.9° lat, but the true spatial resolution is coarser than T62. NCEP reanalysis data are available from NOAA's NOAA's Climate Diagnostics Center .

High-wavenumber Power-Law Constraint for Blending

The blending creates global fields of surface winds by retaining QSCAT wind retrievals in swath regions, and in the unsampled regions (between swaths and in data gaps) augmenting the low-wavenumber NCEP fields with a high-wavenumber component that obeys observed power law relations:

PSD ~ k p,


where PSD is the power-spectral density, k the wavenumber, and the exponent p characterizes the power-law behavior. QSCAT data show that p takes values between -2 at high latitudes and -5/3 at the equator.

Global maps of the exponent p get computed from QSCAT data as monthly averages on a 8° x 8° grid. An entire month of scatterometer data is partitioned into continuous along track segments of at least 30° length. Wavelet coefficients are computed from the projections of segmented data onto nested wavelet basis functions for 8°, 4°, 2°, and 1° resolution intervals. The coefficients are accumulated in the latitude, longitude bin that contains the mid-point of the scatterometer data segment. Wavelet coefficient means and standard deviations are computed for each bin, and for each resolution interval. The means are essentially zero in all cases. The global maps of wavelet coefficient standard deviations for all resolution intervals are smoothed once with a 5-point spatial filter. Those coefficients are used to synthesize, locally in space and time, the high-wavenumber variability of the surface wind.

Blending Method

Scatterometer wind speed and direction estimates arise from co-located radar backscatter measurements as processed by the geophysical model function specific to the QSCAT radar system. Determining the accuracies of these wind estimates is an area of active current research, however it is not an issue here. For the purpose of blending, it is assumed that the scatterometer estimates for surface wind speed and direction, that survive the above stated quality controls, are correct.

Formally, this blending method is an efficient manipulation of 2-dimensional cubic B-splines for the joint interpolation of scatterometer and NCEP analyzed surface wind fields, and for wavelet synthesis of high-wavenumber variability wherever there are gaps in the scatterometer observations. The B-spline operation on a discrete wind component field, say û, is denoted as [ û ] SS , where SS is the spline scale in kilometers and ^ implies a discrete field. The synthesis procedure is a B-spline implementation of orthonormal wavelet-based, multiresolution analysis (Wornell, 1993), that is forced to be statistically consistent with the scatterometer data. The overall procedure is applied independently for each velocity component at each new analysis time.

Construction of a blended, global field of a surface wind component, say u(x,y), first requires an analytic low-pass filtered NCEP field, u LP(x,y), at each analysis time. It is given by

u LP (x,y) = [ û NCEP ] 220 ,

where the B-spline operation is denoted by the square brackets, whose subscript, SS = 220, is the spline scale in kilometers. This low-pass field is used to detrend all the scatterometer data in a ± 6-hour temporal window, wherever QSCAT and NCEP overlap in space. A small scale spline fit to the high-pass filtered scatterometer data gives

u HP (x,y) = [û QSCAT - û LP (x,y) ] 55 .

An analytic NCEP/QSCAT wind component field is then given by

u SUB (x,y) = u LP (x,y) + u HP (x,y).

Within a swath, u SUB (x,y) exactly reflects the scatterometer data, because the filtered NCEP field is first removed then added back. The high-pass spline provides a smooth transition over 3 spline scales (165 km) between the scatterometer winds in a swath and and the NCEP wind in the gaps between swaths.

Next global, analytic, realistic, high-wavenumber variability, u SYN (x,y), is synthesized from statistically generated wavelet coefficients. For a nested sequence of finer resolutions (8°, 4°, 2°, 1°), wavelet coefficients are obtained by randomly sampling a logarithmic-spike distribution (see Chin et al., 1998) of zero mean. The standard deviation of the distribution varies to match the regional standard deviation of the wavelet coefficients computed from QSCAT data for each resolution as 8° x 8° monthly averages. Therefore, statistically, u SYN (x,y) obeys the power-law constraint imposed by the QSCAT data. Finally, the enhanced analytic representation of the the wind component field is computed as

u(x,y) = u SUB (x,y) within swaths, and

u(x,y) = u SUB (x,y) + u SYN (x,y) within gaps.

This field can be sampled at any desired resolution. It contains realistic variability to 50 km (0.5°), given that both the Nyquist interval of the scatterometer observations, and the highest wavelet resolution correspond to about 100 km.

Wind Stress and Curl calculations

Wind stress is computed as
&tau = &rhoa CD |u| u,

where u = (u,v) is the wind vetor, |u| = (u2 + v2)1/2 is the wind speed, &rhoa is a typical air density (1.2 kg/m3), and CD is the neutral 10-m drag coefficient based on Large, McWilliams, and Doney (1994): "Oceanic vertical mixing: A review and a model with a nonlocal boundary layer parameterization", Rev.Geophys., 32, 363-403, and given by

CD = 0.0027 / |u| + 0.000142 + 0.0000764 × |u|.

Wind stress curlis calculated as the vertical component of the curl of &tau. And can be computed from the gradient of t&tau:

&nablaz × &tau = &partx&tauy - &party&taux .

For this procedure, maps of blended winds (on 0.5° × 0.5° grid) are used to compute wind stress, and the above equation is discretized in a centered finite-difference form to approximate wind stress curl. For details, see Milliff and Morzel (2001).

In the blended wind product, spline approximations for u and v allow analytic computation of the spatial derivatives as follows for the computation of curl:

&nablaz × &tau = &rhoa CD { &partx (|u| v) - &party (|u| u) },

&nablaz × &tau = &rhoa CD { |u| (&partxv - &partyu ) + |u|-1× [ v2&partxv - u2&partyu + u v (&partxu - &partyv) ] }.



References


last modified on February 20, 2008
Go Back