WMO/CAS/WWW

Fig. 1.b.1: Bragg Scattering. Adapted from Stoffelen (1998).

of being azimuthally dependent on the orientation of the wave relative to the emitting source, as well as being proportionally related to the amplitude of the waves. These capillary waves adapt nearly instantaneously and are oriented to the surrounding wind field. The reflective wave is also a function of both the angle of incidence and polarization of the emitting energy (see Fig 1.b.1).

b) Geometry

The SeaWinds conical scanner onboard the QuikSCAT satellite receives a multi-view observation of the backscattered microwave energy, _o, via two differently polarized emitters that have slightly different incident angles (46 and 54 deg for the horizontally polarized (h-pol) and vertically polarized (v-pol) emitters, respectively). The SeaWinds antenna footprint on the earth’s surface at zenith angle is an ellipse approximately 25 km in azimuth by 37 km in the look(or range) direction. Signal processing also provides commandable variable range resolution ‘slices’ of approximately 2 to 10 km (Perry 2000). At 18 rpm and a pulse rate of 187 Hz, each processed slice is a product of up to 12 samples. Because of the positioning on the antenna, the v-pol view is slightly wider than the h-pol view limiting the swath edge to only two independent views. The geometry of the sensors and ideal locations to obtaining up to four _o retrievals is shown in Fig 1.b.2.

Fig. 1.b.2: SeaWinds geometry and measurements. Adapted from Freilich (2000).
c) Wind Retrieval

Retrieval of the wind data from the normalized backscatter radar cross-section is an extensive process and is described in various references (Perry 2000; Wentz and Smith 1998; Dunbar and Freilich 1993). The process requires the inversion of a multi-view wind retrieval algorithm and the determination of a ‘most likely solution’ as determined by an empirically derived geophysical model function, currently called QSCAT1 (see Fig 1.b.3 for a simplified schematic of this process). Because more than one combination of wind speed and direction is possible, the process ‘ranks’ the most likely solutions. A separate wind direction ambiguity function is then used to ‘pick’ a single wind speed and direction for each wind vector cell (wvc). This selection is determined by the top-ranked solutions, location of the wvc along the cross track of the swath, comparison of neighboring wind vector cells, and a comparison (called ‘nudging’) with a numerical weather prediction (NWP) model analysis (the short-term forecast fields of the Aviation (AVN) model are used for the so-called Near-Real Time (NRT) product while using only the top two-ranked ambiguities).

Fig. 1.b.3: Wind retrieval and determination. (Adapted from Freilich (2000).)

d) Rain Effects

Although less vulnerable than the passive microwave sensors, the Ku-Band scatterometer onboard QuikSCAT is more susceptible to rain contamination than the C-Band scatterometer sensor on the earlier Earth Remote Sensing (ERS) satellites. Unfortunately, there is no rain or water vapor measurement sensor onboard QuikSCAT (there will be a microwave imager on ADEOS-II, the next NASA scatterometer mission). Therefore, any evaluation of the effect of rain on the wind retrieval process must be indirectly inferred from the behavior of the backscattered sigma-0 (__o) measurements, themselves, and from comparative data in the proximity to QuikSCAT measurements. The current method of ‘flagging’ __o values follows a sophisticated method called a Multi-dimensional Histogram Rain Flag technique (MUDH) where the normalized beam is evaluated for an expected difference as compared to a reference database developed during the calibration/validation period with nearby SSM/I data. This method is shown to be successful over large data set comparisons, but is subject to errors over small areas and time periods as it is also sensitive to increased variance between measurements that is common in high wind regions. Huddleston and Stiles (2002) show (Table 1.b.1) that the false alarm rate goes up with the increase in the wind speed (an unfortunate consequence for evaluating data around a TC). Fortunately, the recent study by Stiles and Yeuh (2001) (shown in IWTC Workshop Paper 0.2.b) provides support for a methodology (shown in a latter section) to distinguish between likely unusable data and acceptable data for those regions where the expected wind speeds are above 15 m/s (although with a higher expected variance in speed and especially direction).

Estimated Rain Flag Performance (MUDH) (Statistics for two-beams) Wind Speed % Flagged False Alarm Percentage Missed Rain 3 – 5 m/s 4.0

1.7 21 5 – 7 m/s 3.5 1.4 23 3 – 7 m/s 3.7 1.5 22 7 – 15 m/s 5.0

2.1 30 15 – 20 m/s 24.0 17.0 24 15 – 30 m/s 26.0 18.0 23

(From Huddleston and Stiles (2000))

Table 1.b.1 Rain flag performance as a function of wind speed.

1.b.3 QuikSCAT Observations and Data Availability

a) Data Availability

The QuikSCAT satellite is a polar-orbiting, sun-synchronous satellite with an equator-ascending time of approximately 0600 local (+/- 30 minutes)(Fig 1.b.4 a,b). Wind vector solutions and up to four ambiguity solutions are stored in 25 km X 25 km wind vector cells (wvc) over a 1800 km wide swath. In addition, the averaged sigma-0 for each wvc and up to eight (for each view) individual slice sigma-0 values that contributed to the average are also available. Besides including the position of the center of each cell, each wvc also contains various flagging indicators that include possible contamination with rain, land, and ice. Near-real time (NRT) wind data are available from the first three sites while the Science Data Set (more data checks and use of the AVN analysis fields with all four ambiguities) from these other sites:

FNMOC (globe) - http://152.80.49.204/CGI/scat.cgi/plot=scat/

NOAA/NESDIS (globe) - http://manati.wwb.noaa.gov/quikscat/

NOAA/NESDIS (TC page) - http://manati.wwb.noaa.gov/cgi-bin/qscat_storm.pl

JPL (various) - http://haifung.jpl.nasa.gov/index.html

BYU/MERS (various) - http://www.scp.byu.edu/data.html

FSU/COAPS (various) - http://www.coaps.fsu.edu/scatterometry/

QuikSCAT data over tropical cyclones have been archived since August 1999. This includes the individual sigma data (Level 2A) and processed wind solutions (Level 2B) from JPL (PO.DACC) as well as global wind data (Science Data set - Level 3). A partial list of agencies that have created TC-specific archives includes: NOAA/NESDIS, NRL/Monterey, Remote Sensing Systems (RSS), FSU/COAPS, JPL/PO.DACC, and the BYU MERS Laboratory.

a) b)
Fig 1.b.4: Examples of daily QuikSCAT-processed winds. Level 3 Science Data Set (a) are available from JPL/PO.DACC; while Near-Real Time (NRT) views (b) are available within 2 _ hours of data time from the NOAA/NESDIS site.

b) Coverage over TCs

It is estimated that over 800 passes each year at least partially view a TC. Since its inception (June 1999), in excess of 300 TCs have been sampled, which at an average of 5 days per cyclone and two passes per day would account for over 2500 QuikSCAT passes that contain TC data (see Figure 1.b.5). In Fig 1.b.5, one can visualize the type of coverage to be expected for varying direct hits and near-misses. Although the figure shows only 30% hitting the ‘sweet zone’, this researcher has found that center positions can be obtained almost to the very edge of the swath. However, ambiguity solutions and lower skill level
in wind estimation is to be expected here.
Fig 1.b.5: QuikSCAT coverage over TCs. (Adapted from D. Smith (2001).)

1.b.4 Working With QuikSCAT Data and TC Analysis

Before performing any critical TC analysis with QuikSCAT data the analyst must fully understand how the data are obtained and its weaknesses and strengths. The previous sections of this paper briefly described the basic principles of scatterometry and gave a summary of the data and the wind retrieval process. This next sub-section will summarize problem areas inherit with these data primarily related to what are considered ‘low skill’ areas and gives suggested solutions, especially as related to working with TCs. Then, the following sub-section will give a suggested step by step approach to a manual analysis of a TC using the Near-Real Time (NRT) QuikSCAT data set as was developed at the Joint Typhoon Warning Center (JTWC), Hawaii. Although there is no discussion of any automated process (which may be desired by many users), it is hoped that the descriptions described in these steps may give an idea how these processes can be automated. Case studies will be presented in the workshop.

a) QuikSCAT Problem Areas and Possible Solutions

Table 1.b.2 summarizes some of the more typical problem areas, typically found in the ‘low skill’ areas when working with the NRT QuikSCAT data set.
Examples of some of these situations will be shown during the Workshop.

Table 1.b.2 QuikSCAT Problem Areas/Solutions

1) Edge of swath (~7 wind vector cells, wvc) and along sub-track (3-4 wvc).
Characteristic:
a) Different sensor view angles creates an underlap of the h-pol sensor by 193 km.
b) Outer regions have less than 4 independent views (gives higher ranking to along
track solutions).
c) Area along sub-track lacks substantial view angle difference between the
upstream and downstream view for each sensor.
Solution:
d) Reanalyze edge data where wind directions are drawn parallel to the swath
direction or wind speeds are inconsistent with surrounding data.
e) Use either ambiguity displays (to see other possible solutions) or NRCS image
of forward-looking V-pol view.

2) Sensitivity to heavy rain.
Characteristic:

1. Enhances surface roughness creating a signal to noise problem and isotropic conditions, especially in low wind areas.
   
2. Rain scattering of up- and down-signal creates a loss of signal.
   
3. Less ability to see wave orientation and make direction determination.
   
4. H-pol sensor more susceptible than V-pol sensor.
   
5. Signal orientation weakens and solutions tend towards cross-track direction
   
Solution:
f) An isotach analysis of flow surrounding rain flag region will indicate areas
that are susceptible to over estimation of winds in low (< 10 m/s) regions
versus areas that are more likely to under estimate winds (> 15 m/s).
g) Reevaluate areas with excessive cross-track solutions using either ambiguities
or NRCS imagery.
h) Most vulnerable wvcs will show only two cross-track solutions-use with sat
imagery,(in a 25 km² area there is a good likelihood some pulses will not be
adversely affected by the rain, especially in convective systems).
i) Rain flag false alarm rate highest in winds >15 m/s, median filter selection
process can extend rain-affected regions to surrounding wvc (rain block),
especially in low skill areas (current wind vector selection process exaggerates
rain effect).

3) Wind selection sensitivity to errors in NWP model in low skill locations.
Characteristic:
a) Missed (or misplaced) synoptic-scale features will nudge (select) an incorrect ambiguity solution, especially if TC structure is small or in data poor region, especially vulnerable to circulations forming within trough axis.
b) Median filter (smoothing) can exasperate an incorrect choice into the neighboring wind vector cells.
c) Near real time solutions may have inconsistent solutions between data cut-off
(download) regions and between updates in the comparison NWP.
Solution:
d) Compare between different ambiguity selection processes (FNMOC AVN vs NOGAPS).
e) Recognize typical patterns that selection process has difficulties.
f) Reanalyze area using ambiguities and NRCS images.






Table 1.b.2 QuikSCAT Problem Areas/Solutions, continued


4) “Practical” wind regime between 5 and 30 m/s (especially in TC core).
Characteristic:
a) Problems in both Light and very Heavy winds.
b) Light winds have low power signal, poor wave orientation indication, and
susceptible to large areas of moderate to heavy rain.
c) Heavy winds lack proper validation due to lack of comparison data, and have
saturation and behavior issues (h-pol and v-pol have different properties)
Solutions:
d) Try to validate any analysis with independent data or imagery
e) Areas with winds outside of this range have greater uncertainty in wind
direction (sometimes indicated by 4 equal selections in the ambiguity display).
- Review possibility that circulation center may be ‘off’ by one wvc (equal
probability) even when surrounding wind field looks consistent.
- Circulation center usually within lightest wind
- High resolution NRCS image may help solve positioning along swath in cross-
track direction (east-west position).
f) Understand general characteristics of TC structure and that wind speed reflects
a 25 km² area.

5) Resolution of 25 km by 37 km footprint.
Characteristic:
a) Limits wind retrieval in tight gradient regions, especially in core region of TC
b) Represents an average wind speed and direction from up to 48 overlapping pulses
Solutions:
c) Coarse resolution more likely to obtain a solution in very high and very low
wind speed regions, and in extensive rain areas (noisier data pulses not used in
wvc calculation)
d) Use higher resolution data set (such as the NRCS images) but be aware of items
listed above.

6) Ambiguity selection process and how rain flags are used.
Characteristic:
a) The NRT QuikSCAT product uses short term AVN forecast field and performs nudging
with only top two-ranked ambiguity solutions.
b) Generally, the top-ranked solutions is expected 80% of the time (outside of
rain region and swath edge) and the 2^nd-ranked solution 15% of the time (Voorrips
1999); requirements for the 3^rd and 4^th ranked solution increases to 15% for the
swath edge (Stiles 1999).
c) Process will tend to select highest ranked solutions over NWP field except in
low skill regions (or in areas with more than one equally likely solution).
d) Median filter has characteristic of creating abrupt boundaries which can cause
an unwanted discontinuity in a circulation field (but, nice in frontal systems).
e) No direct measurement of rain or moisture on QuikSCAT.
f) Large rain blocks will eliminate or miss-position a circulation center.
g) Current models are not tuned to the tight curvature and gradients
h) Model gives higher weight to cross-track solutions in isotropic (rain)
conditions (winds are artificially interpreted as being cross-track, or
perpendicular to swath); the median filter then expands these regions
into neighboring wind vector cells (called rain blocks).
Solutions:
i) Use Science-level data (considerable delay) or perform methods suggested below.
j) Look for discontinuities in wind field and evaluate correct flow from more
confident wind field away from the TC center.
k) Use raw ambiguity solutions versus selected wind vectors.
l) Evaluate ambiguity solutions for higher confidence regions.
- Two-way solutions (usually 180^o of each other) outside of rain region are
highest confident solutions (be careful not to confuse with default rain-
affected solutions that are perpendicular to the swath orientation).
- Three-way solutions can often help eliminate a false choice,
especially when using with two-way solutions and expected flow in area.
m) Perform analysis with isotachs (look for light wind trough axis identify).
n) Use above methods in conjunction with a NRCS analysis.
b) Scatterometer Analysis Procedures for TCs

The following procedures (Table 1.b.3) were developed as a first draft for TC analysis with QuikSCAT data at the Joint Typhoon Warning Center (JTWC). Although in simple form, it requires extensive training sessions with various case studies. The goal is to be able to obtain a reconnaissance-grade position fixes (as stated in Workshop Paper 0.2.b) and to obtain with some degree of confidence a minimum (at least) maximum wind intensity and a description of the outer wind radii structure. Several cases studies will be presented during the IWTC Workshop.




Table 1.b.3 Scatterometer Analysis Procedures for TCs (Developed for forecasters at JTWC) 1) Examine satellite imagery and synoptic data for first guess position (include TC best track history). 2) Determine synoptic conditions to ensure understanding of

typical wind and center relationships (e.g., wave, monsoon trough, shear, TUTT, subtropical system). 3) Examine scatterometer wind solutions with reference to understanding of procedures 1 and 2, above. a) Use highest resolution available (currently FNMOC

display on 1 deg grid)and obtain available wind, ambiguity and NRCS images. b) Draw (or estimate) a first guess streamline analysis using wind solutions to find the center. i) Identify possible problem areas (see Table 1.b.2). ii) Is cente

r region affected by problem areas? iii) Do the different (NWP) nudging methods give different solutions? 4) Perform isotach examination. a) Perform an isotach analysis while subjectively adjusting for any rain- enhanced wind speeds. i) Draw t

he 30kt (15 m/s) isotach first while trying to visualize a smooth gradient of increasing winds towards the systems strongest area, draw with knowledge of closest in time satellite imagery. ii) Winds in excess of 30kts are le

ss likely to be affected by heavy rain; and as the winds increase, are more likely to underestimate the wind speeds versus overestimate them. iii) Rain has greatest adverse effect in light wind region (less than 5-

10 m/s) and can often be identified by comparing with up- and down- stream wind values (compare with convection seen in imagery) iv) When determining a ‘maximum’ wind for this analysis, analyst must be aware of limi

tations of the sensor and the processing of these data -- In the current QSCAT1 algorithm, winds are rarely in excess of 50kt (25 m/s) and should be considered widespread in the wvc -- Wind plots greater than 50kt near the

TC core can only represent a minimum, or ‘at least’ value. True representation of the TC maximum wind will be dependent upon the inner core gradient near this measurement, degree of extinction from

widespread rain in the wvc, and actual value of maximum wind over the nominal working range of 60-80 kt for QuikSCAT data. v) Determining the location of the 35, 50 and 64 kt wind radii -- As the system develops over each of these valu

es, a smooth increasing gradient will give confidence to the analysis -- Biggest restriction will be limited to the size of the region a) Center is often on axis of lightest winds and along trough axis. b) Center position with respect to

highest winds depending upon known characteristics and development stage (early, mature, shear, etc.). Table 1.b.3 Scatterometer Analysis Procedures for TCs, continued c) Center located near lowest wind value within highest wind isotach



(especially for TC eye cases). 5) Examine ambiguity solutions and NRCS (see guideline for NRCS, below). a) Draw streamlines to meet ambiguities (Fig. 1.b.6). i) Work inward from TC environment towards vortex center ii) Give priority to wvc’s

with 2 or 3-way ambiguities. a) Adjust center in order to avoid drawing streamlines that are not possible solutions. b) Adjust to fit NRCS image i) When given a choice, select closest direction to the cross-track or along track solution that best

fits either a dark or bright band region ii) Uses finer adjustments to fit trough axis, col region, and lee-wind effects off of higher terrain. iii) Confidence of solution can be determined by clarity of light/dark re

gions near suspected center 6) Compare and adjust position with trough axis and isotach analysis from Step 4 and satellite imagery from Step 1. 7) Examine NRCS for wind signature (Figure 1.b.7). a) In weaker systems, use only after initial guess to focus

search. b) Very precise when good signature exists (average 6km resolution). c) Try to use with dark band to refine cross-track (east-west) direction. Fig 1.b.6: Ambiguity analysis Fig 1.b.7: NRCS Solutions (7km Streamlines must

not violate resolution(Sigma-0)) available solutions (on rare cases 2-way ambiguities exist, perpendicular to swath indicating no solution).


1.b.5: Summary

Although the wind retrieval process has difficulties in both light and high winds, the data can be expected to provide the TC forecaster with knowledge of potential tropical cyclone genesis, position, movement, outer wind structure, and provides a 'minimum’ (at least) estimate of maximum sustained winds. Although rigorous procedures need to be established throughout the TC warning community as suggested in the procedures stated in this paper, the application process can be summarized by the phrase: “If it makes sense, use it”.





Acknowledgments. This research was supported by the Naval Research Laboratory, Monterey through its programs at the Office of Naval Research (PE-060243N), and by the Oceanographer of the Navy through the program office at the Space and Naval Warfare Systems Command PMW-155 (PE-0603207N).




1.b.6 Bibliography





Atlas, R., R.N. Hoffman, S.M.Leidner, J. Sienkiewicz, T.-W. Yu, S. C. Bloom, E. Brin, J. Ardizzone, J.Terry, D. Bungato, and J. C. Jusem, 2001: The effects of marine winds from scatterometer data on weather analysis and forecasting. Bull. Amer. Meteor. Soc.,82, 1965-1990.

Early, D.S and D.G. Long, 2001: Image reconstruction and enhanced resolution imaging from irregular samples. IEEE Trans. Geosci. Remote Sens.,39, 291-302.

Edson, R.T., 1997: Operational use of scatterometer data at the JTWC. Preprints, 22nd Conference on Hurricanes and Tropical Meteorology, Ft Collins, CO, Amer. Meteor. Soc.,195–196.

^___________, M.A. Lander, C.E. Cantrell, J.L. Franklin, J.D. Hawkins, and P.S. Chang, 2002: Operational use of QuikSCAT over tropical cyclones. The 25^th Conference on Hurricanes and Tropical Meteorology. San Diego, CA, Amer. Meteor. Soc.,41–42.

Freilich, M.H. and R.S.Dunbar, 1993: Derivation of satellite wind model functions using operational surface wind analyses: An altimeter example. J. Geophys., Res., 98, 14633-14649.

Freilich, M.H, 2000: Evaluation and uses of QuikSCAT data. Slide presentation at the Naval Postgraduate School, Monterey, CA, June 2000.

Jones, W.L., I. Adams, J. D. Park, and S.S.Chen, 2002: Evaluation of seawinds wind speed measurements. The 25^th Conference on Hurricanes and Tropical Meteorology. San Diego, CA, Amer. Meteor. Soc.,555–556.

Long, D., 2000: Point-wise Wind Retrieval Wind Scatterometry Background. A QuikSCAT/Sigma-0 Browse Product, Ver 2.0. Earth Remote Sensing (MERS), BYU University, Provo, UT., 16 pp.

Naderi, F., M.H. Freilich, and D. G. Long, 1991: Spaceborne radar measurement of wind velocity over the ocean—An overview of the NSCAT scatterometer system. Proc. of the IEEE, pp. 850-866, Vol. 79, No. 6, June 1991.

Perry, K.L.,(Ed.), 2000: QuikSCAT science data product user’s manual, Ver. 2.0. Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, 90 pp. (Available from the JPL/QuikSCAT web site.)

Shaffer,S.J., R.S. Dunbar, S. V. Hsiao, and D.G. Long, "A Median-Filter-Based Ambiguity Removal Algorithm for NSCAT," IEEE Transactions on Geoscience and Remote Sensing, Vol. 29, No. 1, pp. 167-174, Jan. 1991

Smith, D., C. Mears, C. Gentemann, and F.Wentz, 2000: Validation of QuikSCAT tropical cyclone winds. . 24th Conference on Hurricanes and Tropical Meteorology, American Meteorological Society, Boston, MA, 193–194.

Stiles, B.W. 1999: Special wind vector data product: Direction Interval Retrieval with Thresholded Nudging(DIRTH), Product description, Ver 1.1. JPL Science Team report, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, 10 pp. (Available from the JPL/QuikSCAT web site.)

Stiles, B.W., and S. Yueh, 2001: Impact of rain on QuikSCAT. Proc. of the AGU 2001 Fall Meeting, 10-14 Dec, San Francisco, CA (also submitted to IEEE
Trans. on Geoscience and Remote Sensing).

Stoffelen, A.C. 1998: Scatterometry (PhD thesis). Royal Netherlands Meteorological Institute, Proefschrift Universiteit Utrecht, Nederlands,209 pp.

Voorrips, A.C. 1999: QuikSCAT status report, 20 Dec 1999. JPL Science Team report, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, 24 pp. (Available from the JPL/QuikSCAT web site.)

Weissman, D.E., M.A.Bourassa, and J. Tongue, 2002: Effects of rain rate and wind magnitude on Sea Winds scatterometer wind speed errors. J. Atmos. Ocean. Technol., 19,738-746.

Wentz, F.J, and D.K. Smith, 1999: A model function for the ocean normalized radar cross-section at 14 GHz derived from NSCAT observations. J. Geophys., Res., 104, 11499-11524.