An Intercomparison of TOPEX, NSCAT, and ECMWF Wind Speeds: Illustrating and Understanding Systematic Discrepancies

780 MONTHLY WEATHER REVIEW VOLUME 132

GE CHEN Ocean Remote Sensing Institute, Ocean University of China, Qingdao, China

(Manuscript received 18 March 2003, in ®nal form 19 September 2003)

ABSTRACT The availability of multiple satellite missions with wind measuring capacity has made it more desirable than ever before to integrate wind data from various sources in order to achieve an improved accuracy, resolution, and duration. A clear understanding of the error characteristics associated with each type of data is needed for a meaningful merging or combination. The two kinds of errorsÐnamely, random error and systematic errorÐ are expected to evolve differently with increasing volume of available data. In this study, a collocated ocean Topography Experiment (TOPEX)±NASA Scatterometer (NSCAT)±ECMWF dataset, which covers 66ЊS±66ЊN and spans the entire 10-month lifetime of NSCAT, is compiled to investigate the systematic discrepancies among the three kinds of wind estimates, yielding a number of interesting results. First, the satellite-derived wind speeds are found to have a larger overall bias but a much smaller overall root-mean-square (rms) error compared to ECMWF winds, implying that they are highly converging but are systematically biased. Second, the TOPEX and NSCAT wind speed biases are characterized by a signi®cant ``phase opposition'' with latitude, season, and wind intensity, respectively. Third, the TOPEX (NSCAT) bias exhibits a low±high±low (high±low±high) pattern as a function of wind speed, whose turning point at 14.2 m sϪ1 coincides well with the transitional wind speed from swell dominance to wind sea dominance in wave condition, suggesting that the degree of wave development plays a key role in shaping wind speed bias.

1. Introduction both time and space domains. It is obvious that such an effort would be meaningful only if the accuracy of all Since the 1970s, information on sea surface wind datasets involved is above a given level. A number of speed has been obtained from a variety of spaceborne studies have been carried out to compare wind estimates microwave instruments including scatterometer (e.g., from various sources and evaluate their consistency Naderi et al. 1991), altimeter (e.g., Fu et al. 1994), ra- (e.g., Halpern et al. 1994; Boutin and Etcheto 1990, diometer (e.g., Hollinger et al. 1990), and synthetic ap- 1996; Busalacchi et al. 1993; Rienecker et al. 1996; erture radar (SAR; e.g., Vachon and Dobson 1996). Among them, the satellite scatterometer is a dedicated Boutin et al. 1996, 1999; Bentamy et al. 1999; Quef- wind measuring sensor that provides both speed and feulou et al. 1999; Meissner et al. 2001). Most of the direction; other sensors usually make wind measurement comparison statistics indicate that the overall differenc- as a by-product. In the past two decades, it has been es among various datasets are within their measurement convincingly demonstrated that a near-synoptic, all- uncertainties. Meanwhile, they also suggest that the ob- weather view of marine winds from spaceborne micro- served deviations are geographically, seasonally, and wave instruments has greatly advanced our knowledge geophysically dependent. Such sensor-related system- in many aspects of meteorology and oceanography. atic discrepancies are poorly understood so far. Up until now, each type of the sensors mentioned In this study, the consistency of altimeter and scat- above, except SAR, has accumulated a decade-long ob- terometer wind speed measurements are examined, with servation database that allows wind climatology to be special attention paid to the spatial and temporal patterns constructed (e.g., Atlas et al. 1996; Bentamy et al. 1996; as well as the geophysical dependency of their system- Young 1999). Naturally, there is a growing interest in atic discrepancies. In doing so, a collocation dataset of combining or merging wind measurements from various ocean Topography Experiment (TOPEX) and National sources to achieve a better coverage and resolution in Aeronautics and Space Administration (NASA) Scat- terometer (NSCAT) is compiled together with European Centre for Medium-Range Weather Forecasts Corresponding author address: Dr. Ge Chen, Ocean Remote Sens- ing Institute, Ocean University of China, 5 Yushan Road, Qingdao (ECMWF) winds. These two sensors are selected be- 266003, China. cause the former represents the state of the art in satellite E-mail: [email protected] altimetry in terms of data quality and duration (Fu and

᭧ 2004 American Meteorological Society

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC MARCH 2004 CHEN 781

Cazenave 2001), and the latter is known to be one of stability. Since the ECMWF data used here are from the the most accurate spaceborne scatterometers ¯own to forecast rather than reanalysis product, neither NSCAT date (Freilich and Dunbar 1999). Therefore, the results nor TOPEX data have been incorporated (note that obtained here are expected to be largely representative ECMWF only started assimilating satellite wind data of the two kinds of sensors at their corresponding fre- operationally with QuikSCAT after January 2002). It quencies. The rest of the paper is organized as follows: has to be noted, however, that the ECMWF forecast Section 2 describes the collocated TOPEX±NSCAT± models experienced a change from a three-dimensional ECMWF dataset with general statistics. The spatial and variational data assimilation (3DVAR) system to a four- temporal patterns of the wind speed deviations among dimensional variational data assimilation (4DVAR) sys- the three data sources are examined in section 3. The tem in 1997. But this is not expected to cause systematic dependency of wind speed bias on wind intensity is biases in this analysis. illustrated in section 4 and is discussed in the context An optimal collocation scheme is desirable in order of swell and wind sea in¯uences in section 5. Finally, to achieve a reliable and meaningful comparison. The a summary with conclusions is presented in section 6. surface resolution of NSCAT is 25 km ϫ 25 km for each wind estimate, a so-called wind vector cell. To keep 2. Collocation data and general statistics sampling errors to a minimum, equal sensor surface coverage is sought. The high-resolution NSCAT product a. Collocation dataset is bene®cial for this purpose as the TOPEX wind resolution cell is relatively minisculeÐof the order of 2 The TOPEX/Poseidon satellite is a joint U.S.±French km ϫ 6 km for each footprint. To bring the TOPEX altimetric mission launched on 10 August 1992 (Fu et spatial resolution as near to NSCAT as possible, I in- al. 1994). It samples the ocean surface between 66ЊS clude an average over those TOPEX data points that and 66ЊN at a 1-s interval (corresponding to a 5.8-km fall within a given NSCAT wind cell. Thus the TOPEX resolution on the ground track) for each of the 254 as- ``wind cell'' characteristics become variable from 2 km cending and descending passes that make up a 9.9156- ϫ 6kmto2kmϫ 25 km (from one to four footprints). day cycle. The satellite has so far acquired more than The ECMWF marine wind output is on a 1.125Њ by 10 yr of sea level, signi®cant wave height, and radar 1.125Њ grid every 6 h. For our purpose the model output cross-sectional data. NSCAT was launched on 17 Au- is interpolated linearly in space and time to derive a gust 1996 as part of the Japanese Advanced Earth Ob- wind estimate collocated with our sensor observations. serving Satellite-I (ADEOS-I) mission. It has an array This results in a maximum time lag of Ϯ3 h. Ground of six antennas that scan two 600-km bands of the ocean resolution for ECMWF is then 125 km 60±125 km, on each side of the instrument's orbital path separated ϫ depending on the latitude. by a gap of 329 km. Four collocated backscatter measurements (three vertically polarized and one horizon- Some additional quality controls are performed to en- tally polarized) from three azimuth angles throughout sure a better consistency of the three wind speed da- each swath allow vector winds to be retrieved with ap- tasets. Since scatterometer winds will be used in section proximately 25-km resolution, and 90% of the global 5 as sea state independent measurements to distinguish ocean is covered by the swath within 2 days (Naderi et swell from wind sea, and there is evidence showing that al. 1991). NSCAT acquired vector wind data from mid- near-nadir scatterometer winds are affected by signi®- September 1996 until the spacecraft suffered a cata- cant wave height (Queffeulou et al. 1999), the crossover strophic failure of the solar panel on 30 June 1997. points where the NSCAT midbeam antenna has an in- Fortunately, the short-lived NSCAT instrument was ful- cidence angle less than 40Њ are eliminated from the da- ly overlapped with the TOPEX altimetric mission. The taset. For the altimeter data, an additional minimal ®l- legacy of the 10-month coincident scatterometer±altim- tering of outliers removes the points where the TOPEX eter data with unprecedented individual quality will be estimate of backscatter is below 5.0 dB or above 30.0 of great bene®t to a variety of geophysical applications. dB. Also, the TOPEX and NSCAT data are screened Taking advantage of this unique opportunity, a TOPEX± for rain contaminations using the schemes described in NSCAT collocation dataset is compiled together with Chen et al. (2003) and Hoffman et al. (1994), respec- ECMWF winds. tively. Some details of this collocation dataset are given The NSCAT wind speed data used for this study come in Table 1, and further information can be found in from the High-Resolution Merged Geophysical Data Gourrion et al. (2000). Product (Dunbar 1997). The TOPEX altimeter data are To provide a sense of the global distribution of the from the TOPEX/Poseidon Merged Geophysical Data compilation, Fig. 1 presents the data density of the TO- Records (Generation B) (Benada 1997) for the period PEX±NSCAT±ECMWF crossover samples for the 10- of the NSCAT mission. Global model outputs of surface month period. It is apparent that the dataset does contain wind vector from the ECMWF forecasts are also used. global observations out to the 66Њ latitude limit of TO- Such wind vectors are estimated from surface analysis PEX. A salient characteristic of the distribution is the for an altitude of 10 m above the ocean under neutral increased likelihood of high-latitude intersections, in

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC 782 MONTHLY WEATHER REVIEW VOLUME 132

TABLE 1. Some details of the TOPEX±NSCAT±ECMWF collocation dataset, where T, N, and E denote TOPEX, NSCAT, and ECMWF, respectively. Time duration 15 Sep 1996±30 Jun 1997 Spatial coverage 66ЊS±66ЊN No. of collocation data 97 613 Time window (h) T±N 1.0 T±E 3.0 Space window (km) T±N 12 T±E 60±125 contrast to the basically random nature of the geographical pattern in low and midlatitude areas. b. General statistics FIG. 2. Wind speed histogram based on the TOPEX±NSCAT± To have a ®rst overview of the collocation dataset, ECMWF collocation dataset. The thick, thin, and dashed curves rep- the histograms of the TOPEX, NSCAT, and ECMWF resent the TOPEX, NSCAT, and ECMWF results, respectively. wind speeds are shown in Fig. 2. It can be seen that the distributions of NSCAT and ECMWF winds are very close to each other (as a result of the fact that the NSCAT may result from the differences in 1) their measuring wind speed algorithm is partially determined by com- mechanisms (specular re¯ection for the former and parison with the ECMWF winds), while that of the TO- Bragg resonant scattering for the latter); 2) the de®- PEX is somewhat different. The TOPEX histogram ap- ciencies of the wind retrieving algorithms employed pears to be less smooth, with its peak shifted toward [note that the Witter and Chelton (1991) algorithm is high wind by 2 m sϪ1. Moreover, the TOPEX mea- used for both TOPEX and ERS-2 altimeters]; and 3) the surements seem to favor both high winds between 10 spatial scales represented by each dataset. and 20 m sϪ1 and low winds below 2 m sϪ1 compared To understand more details about the observed pattern to NSCAT and ECMWF. Given the large number of data in Fig. 2, scatter diagrams comparing the three sources used in generating the histograms, such departures are of wind measurements are plotted in Fig. 3, and their unlikely to be random. Coincidentally, in a similar com- general statistics are given in the ®rst three columns of parison between NSCAT and European Remote Sensing Table 2. Going through the three subplots in Fig. 3, it Satellite-2 (ERS-2) altimeter winds (Queffeulou et al. is evident that the degree of scatter is considerably less 1999), the features mentioned above appeared almost between the two types of satellite measurements (Fig. in the same way (see their Fig. 5). Therefore, it can be 3a) than with the model-predicted winds (Figs. 3b and argued that some systematic discrepancies exist between 3c), as also indicated by the overall variances in Table altimeter- and scatterometer-derived wind speeds, which 2 (1.35 m sϪ1 versus 1.82 and 1.83 m sϪ1). Meanwhile,

FIG. 1. Global distribution of spatial density of the TOPEX±NSCAT±ECMWF collocation data.

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC MARCH 2004 CHEN 783

a wind speed±dependent bias between the altimeter and scatterometer wind estimates is also visible (Fig. 3a). The deviation of the TOPEX histogram in Fig. 2 is con®rmed here by the systematic overestimation (underestimation) of its speed for low (high) winds, compared to the scatterometer winds. These characteristics imply that altimeter and scatterometer wind measurements are of higher consistency compared to model predictions, but the ECMWF winds are less biased as a whole compared to satellite measurements (see Fig. 3 and Table 2). As a result, a model-based wind product would be very useful in climate-related studies where global statistics and interannual or decadal variabilities are of major interest, while satellite observations would be most desirable when oceanic wind information with ®ne spatial and temporal resolution is needed. Of course, the value of absolute biases does not necessarily re¯ect the quality of the product. An absolute bias can be easily taken out by adding or subtracting a constant value to the retrieved wind speed. What really matters is whether this bias changes over time or if it is different in various geographical regions, as will be discussed in detail in the following sections. As far as the wind speed biases of NSCAT and ECMWF are concerned, Freilich and Dunbar (1999) showed that the former is biased low by 0.3 m sϪ1, while the latter is almost unbiased when validated against collocated buoy data. This implies that a Ϫ0.3 msϪ1 bias may exist between NSCAT and ECMWF, which agrees reasonably well with our estimate of Ϫ0.23 m sϪ1 (Table 2). In terms of rms difference between NSCAT and ECMWF, the result of Wentz and Smith (1999) is comparable to ours: 1.78 versus 1.83 msϪ1. A slightly larger value of our rms error is probably due to the fact that low-incidence (less than 40Њ) measurements are not included in this collocation dataset. As shown by Queffeulou et al. (1999), the standard deviation of scatterometer winds increases gradually with incidence angle. In the following analysis, we introduce a uni®ed reference wind ®eld that is de®ned as the mean value of the three kinds of collocated wind estimates. This reference dataset is used because an ideal reference dataset (the so-called ``sea truth'') with high accuracy, long duration, and global coverage simply does not exist. Some kind of ``surrogate'' has to be sought if wind comparisons are to be made on a global basis. As far as the present three data sources are concerned, there are basically two solutions: either use one of the datasets as a reference or use a combination of two or all of

←

FIG. 3. Scatter diagrams of sea surface wind speed based on the TOPEX±NSCAT±ECMWF collocation dataset: (a) NSCAT vs TO- PEX, (b) ECMWF vs TOPEX, and (c) ECMWF vs NSCAT. The color legend depicts the number of data within a 0.1 m sϪ1 ϫ 0.1 m sϪ1 grid box.

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC 784 MONTHLY WEATHER REVIEW VOLUME 132

TABLE 2. Comparison statistics of the collocated TOPEX±NSCAT± statistics are believed to be a realistic re¯ection of the ECMWF wind speeds, where T, N, E, and R denote TOPEX, NSCAT, relative quality and characteristics of the true data from ECMWF, and the reference wind ®eld, respectively. the three independent sources. Therefore the pseudo T±N T±E N±E T±R N±R E±R wind ®eld serves as a reasonable reference for the in- Bias (m sϪ1) 0.63 0.39 Ϫ0.23 0.34 Ϫ0.27 Ϫ0.05 tercomparison of the wind products concerned. Rms (m sϪ1) 1.35 1.82 1.83 0.89 0.88 1.13 3. Geographical and seasonal patterns of wind speed discrepancy them. It was decided to generate a pseudo wind ®eld, de®ned as the simple average of the three. This choice a. Zonal distribution is not ideal at all, but it has, at least, the virtue of being relatively ``equal'' to the three original datasets. More- The bias and rms error of the TOPEX, NSCAT, and over, it allows the TOPEX, NSCAT, and ECMWF winds ECMWF winds with respect to the reference wind are to be intercompared in a somewhat ``objective'' and plotted as a function of latitude in Figs. 4a and 4b, ``semi-independent'' sense. As an alternative, an eval- respectively. A striking feature in Fig. 4a is the ``phase uation using ECMWF wind as a reference dataset is also opposition'' between the TOPEX and NSCAT biases. performed for comparison purposes (see section 3a). The absolute value of the bias is smallest near the equa- The general statistics of the three types of wind es- tor for both products and gets positive (negative) for timates with respect to the reference wind ®eld are also TOPEX (NSCAT) near the poles. The general pattern included in Table 2 (see the last three columns). As of zonal wind bias is well correlated (positively for TO- expected, the ECMWF wind is least biased (Ϫ0.05 m PEX and negatively for NSCAT) to that of the zonal sϪ1) but most scattered (1.13 m sϪ1), while the TOPEX wind intensity (see Fig. 2 of Chen et al. 2002b), sug- and NSCAT winds are comparable in both mean bias gesting that the zonal bias is highly dependent on wind (0.34 versus Ϫ0.27 m sϪ1) and rms error (0.89 versus speed. Variations corresponding to the trade winds and 0.88 m sϪ1), with the latter being slightly better. These the horse latitudes can be clearly identi®ed in the TO-

FIG. 4. (a), (c) Bias and (b), (d) rms of sea surface wind speed with respect to latitude. The pseudodataset and the ECMWF dataset are used as reference wind ®elds in (a), (b) and (c), (d), respectively. The thick, thin, and dashed curves represent the TOPEX, NSCAT, and ECMWF results, respectively. The thin line in (a) and (c) denotes zero bias.

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC MARCH 2004 CHEN 785

PEX distribution but are less obvious in the NSCAT itive biases are scattered in the tropical oceans. Unlike distribution. Unlike the systematic biases associated the zonally banded and dramatically varying structure with satellite wind measurements, the zonal variation of of the satellite wind biases, the ECMWF bias has a the ECMWF bias exhibits a largely random nature, with somewhat meridional orientation with small and ho- narrow amplitude between Ϯ0.4 m sϪ1. This random mogeneous amplitude. uncertainty leads to a considerably larger rms error of The three maps of rms error in Fig. 6 appear to be the ECMWF winds at all latitudes compared to the TO- highly correlated in terms of regional features. Most of PEX and NSCAT winds (Fig. 4b). The TOPEX wind the features are characterized by an anistropic pattern appears to have the smallest rms error in the tropical that has a meridional orientation for low latitudes and areas between Ϯ20Њ. For the extratropical areas, how- a zonal orientation for high latitudes. Local maximums ever, the TOPEX and NSCAT rms errors are very close are found at 70Њ, 190Њ, and 285ЊE along 60ЊS, as well to each other. Note that for all three cases, the rms error as at 170Њ and 310ЊE along 60ЊN, where the wind speed has a signi®cant minimum around Ϯ20Њ where the trade estimates are ``most inaccurate.'' By contrast, ``more wind belts are located. Surprisingly to some extent, the reliable'' winds are obtained within Ϯ30Њ of the Atlan- equatorial area displays a signi®cant peak in all three tic, the east Paci®c, and the west Indian Oceans. Ob- rms distributions (Fig. 4b), although it is seen to be the least biased zone for both satellite and model winds (Fig. viously, the relative rms error of the ECMWF winds is 4a). A possible cause of the equatorial bumps in Fig. considerably larger than the satellite winds for almost 4b could be due to rain contamination, as their locations everywhere in the ocean. The TOPEX result is slightly seem to coincide with the intertropical convergence ``more accurate'' than the NSCAT one for low latitudes, zone (ITCZ) (Chen et al. 1997). TOPEX-measured wind while the reverse is true for high latitudes. It should be speed usually increases with the decreasing backscatter mentioned that some of the observed features in Figs. coef®cient caused by rain attenuation at Ku band, but 5 and 6 may result from the sampling mismatch among the reverse situation may also happen as a result of wave the three data sources, especially the nadir-pointing and damping by rain (Chen et al. 1998). For NSCAT, rain aliasing nature of the altimeter instrument as discussed can either raise or lower the wind speed, depending on by many previous investigators (e.g., Schlax and Chel- the incidence angle. Although the TOPEX and NSCAT ton 1994; Chen and Ezraty 1996; Freilich and Dunbar data have both been screened for rain contamination, 1999). the remaining rain effect may still introduce extra variability in wind speed rms around the ITCZ, as evi- denced in Fig. 4b. c. Seasonal variation In order to make sure that the above analyses are not misleading because of the inclusion of the pseudo ref- The short lifetime of NSCAT makes it impossible to erence wind ®eld, the wind speed bias and rms of TO- examine the full annual cycle of its bias and rms error. PEX and NSCAT against the ECMWF dataset as a func- But 10 months of collocation data have already per- tion of latitude are also produced (Figs. 4c and 4d). mitted us to identify the general trend of the seasonal Clearly, Figs. 4c and 4d con®rm nicely the major fea- variation. Figure 7 presents the wind speed bias and rms tures of Figs. 4a and 4b, implying that the results and as a function of month for the North Paci®c and North conclusions based on the reference wind ®eld will not Atlantic between 40Њ and 60ЊN. As can be seen, sea- be too far from the truth, at least in a qualitative sense. sonality is distinct for both wind speed bias and its rms. Large discrepancies are found in boreal winter for all b. Global distribution three biases. A low bias is already visible during summer months, although data from July and August are un- We now examine the geographical distributions of available. The seasonal departures of TOPEX and wind speed bias and rms for TOPEX, NSCAT, and NSCAT biases are also anticorrelated, with the former ECMWF, as shown in Figs. 5 and 6, respectively (note and latter being largely positive and negative, respec- that a 2D smoothing is applied to these ®gures, thereby tively. As far as the TOPEX bias is concerned, the re- lowering the extreme values compared to Fig. 4). As far as the mean bias is concerned, the three spatial pat- sults obtained here are consistent with a previous study terns are very different from one another (Fig. 5). The based on Japan Meteorological Agency (JMA) buoy TOPEX winds are most positively biased with maxi- data (Chen et al. 2000), in which a cosine-typed seasonal mum values appearing at Ϯ60Њ (Fig. 5a). Areas with a bias was observed for the North Paci®c using the Witter small negative bias are found in the tropical west Paci®c and Chelton (1991) algorithm. Given the well-de®ned and east Atlantic, as well as in the Arabian Sea and the phase opposition between TOPEX and NSCAT wind Bay of Bengal, and near the central America. Converse- speed biases, a combination of these two datasets may ly, the NSCAT bias is negative for the majority of the remove part of the seasonal discrepancy. In this sense, world's oceans with a poleward increase of its magni- altimeter and scatterometer winds could be rather com- tude (Fig. 5b). A few small areas with marginally pos- plementary for climatological studies.

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC 786 MONTHLY WEATHER REVIEW VOLUME 132 . 1 Ϫ . 6. Geographical distribution of wind speed rms based on the TOPEX±NSCAT±ECMWF IG F collocation dataset: (a) TOPEX,wind ®eld (b) is NSCAT, and used (c) to ECMWF generate results. the The subplots. pseudo The reference color scale is in m s . 1 Ϫ . 5. Geographical distribution of wind speed bias based on the TOPEX±NSCAT±ECMWF IG F collocation dataset: (a) TOPEX,wind ®eld (b) is NSCAT, and used (c) to ECMWF generate results. the The subplots. pseudo The reference color scale is in m s

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC MARCH 2004 CHEN 787

FIG. 8. (a) Bias and (b) rms of sea surface wind speed with respect to the reference wind speed. The thick, thin, and dashed curves represent the TOPEX, NSCAT, and ECMWF results, respectively. The FIG. 7. (a) Bias and (b) rms of sea surface wind speed with respect thin horizontal line in (a) denotes zero bias, and the two thin vertical to month for the North Paci®c and North Atlantic between 40Њ and dashed lines in (a) correspond to wind speeds of 10 and 15 m s Ϫ1, 60ЊN. The thick, thin, and dashed curves with triangles, squares, and respectively. circles represent the TOPEX, NSCAT, and ECMWF results, respectively. The thin line in (a) denotes zero bias. and ECMWF data are plotted against the reference wind speed in Figs. 8a and 8b, respectively. There are at least 4. Wind speed dependency two features in Fig. 8a that are noteworthy. First, the well-de®ned phase opposition between TOPEX and A number of previous studies have shown that the NSCAT reappears. Second, the variations of satellite biases of altimeter and scatterometer wind measure- biases are nonmonotonic, showing a low±high±low (L± ments are dependent on wind speed (e.g., Ebuchi et al. H±L) pattern for TOPEX and a high±low±high (H±L± 1992; Gower 1996; Freilich and Dunbar 1999; Quef- H) pattern for NSCAT. In fact, the TOPEX (NSCAT) feulou et al. 1999; Wentz and Smith 1999). But factors residual increases (decreases) with wind speed for low responsible for such dependencies are not altogether and medium winds (1 Ͻ U Ͻ 10 m sϪ1), a weakly clear. In this section, these dependencies are reexamined ¯uctuating band is observed between 10 and ϳ15 m sϪ1 using the collocation dataset and are discussed with re- before the trend reverses for high winds (U Ͼ 15 m gard to some of the published results. sϪ1). The turning point is somewhere between 14 and The wind speed bias and rms of the TOPEX, NSCAT, 15 m sϪ1. The ECMWF bias decreases with increasing

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC 788 MONTHLY WEATHER REVIEW VOLUME 132

TABLE 3. Summary of the characteristics of satellite wind speed bias based on selected references, where L, H, ϩ, and Ϫ denote low, high, positive, and negative, respectively. Figure numbers in the right column refer to those that appeared in the original references, cited in the left column.

Author(s) Dataset A Dataset B Wind range (m sϪ1) A±B (m sϪ1) Remark Gower (1996) TOPEX Buoy 0±5 L Fig. 8 5±15 H 15±20 L 0±20 ϩ10% Freilich and Dunbar (1999) NSCAT Buoy 0±3 ϩ Fig. 5 3±20 Ϫ 20±23 ϩ 0±23 Ϫ0.3 Wentz and Smith (1999) NSCAT Buoy 0±3 ϩ Fig. 20 3±12 Ϫ 12±23 ϳ0 0±23 Ϫ0.29 Queffeulou et al. (1999) NSCAT ERS-2 altimeter 0±7 ϩ Fig. 2 7±19 Ϫ 20 ϩ 0±20 Ϫ0.22 wind speed until 7 m sϪ1 before it levels off. For the ties associated with high winds, accumulating evidence rms error (Fig. 8b), the ECMWF shows a systematically provided by several previous works as well as the pre- larger magnitude as expected. The two satellite results sent result in Fig. 8 is clear enough to support the ar- are almost indistinguishable for moderate winds be- gument that a peak/trough of wind speed bias does exist tween 5 and 15 m sϪ1, but a divergent trend is observed for TOPEX±NSCAT between 10 and 15 m sϪ1. for both light and strong winds, with the scatterometer being ``better'' at the low end and the altimeter being 5. Discussions better at the high end. Note that the bias and rms estimates look noisy at high wind speeds in Fig. 8 because In this section, the observed systematic bias of sat- of the decreasing volume of available data. But the re- ellite wind speed measurement will be discussed in results are still expected to be statistically signi®cant given lation to swell and wind sea impacts. In doing so, ev- that the minimum number of collocations within each idence of swell- and wind-sea-induced biases in wind bin (0.1 m sϪ1 in width) is above 20, and their spatial/ speed measurement based on existing literature is pre- temporal distributions are basically random. sented in section 5a. A quantitative link between the Although the nonmonotonic feature of wind speed wind speed corresponding to the peak±trough bias and bias has not been explicitly mentioned for either altim- the frequency of swell±wind sea occurrence is examined eter or scatterometer in the references available to us, in section 5b. it is actually discernible, upon a close scrutiny, in the results of several previous studies, as summarized in a. Evidence of swell- and wind sea±induced Table 3. In the case of TOPEX, a comprehensive val- systematic wind speed bias idation of its wind speed was performed by Gower (1996). His Fig. 8 indicates that the altimeter under- The scatterometer and altimeter wind algorithms, estimates the wind speed for low winds (0±5 m sϪ1) though based on distinct physical mechanisms, share an and overestimates it for medium winds (5±15 m sϪ1), implicit assumption of a fully developed sea. In other while the degree of overestimation decreases consid- words, better accuracy of satellite wind measurement erably for high winds (15±20 m sϪ1), forming a L±H± can be expected under a mature sea state. This, however, L structure. In contrast, a H±L±H pattern is identi®able is rarely the case in the real ocean (Chen et al. 2002a). to some extent on similar plots for NSCAT, as shown The degradation of algorithm performance (especially in Fig. 5 of Freilich and Dunbar (1999) and Fig. 20 of those theoretically based) in the presence of either swell Wentz and Smith (1999). In the study by Queffeulou et or wind sea is, in a sense, inevitable. al. (1999), NSCAT winds are compared directly with For satellite altimeters, the poor quality of the data at ERS-2 altimeter winds. Interestingly, a H±L±H signature low winds is well known. Glazman and Pilorz (1990) of the bias as a function of wind speed is apparently showed theoretically that the degree of wave develop- visible in their Fig. 2. This may imply that the non- ment becomes an increasingly important factor of the monotonic feature of wind speed bias could also be radar backscatter measurement for wind speeds below 5 shared by altimeters other than TOPEX and scattero- msϪ1. Furthermore, based on a long-wave modulation meters other than NSCAT. Despite the larger uncertain- theory, Hwang et al. (1998) pointed out that it is basically

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC MARCH 2004 CHEN 789 the swell-induced surface tilting effect that causes the degradation of altimeter wind measurement. These ar- guments are con®rmed by both space and ®eld observations. For example, Queffeulou et al. (1999) found that the ERS-2 altimeter-measured wind speeds show a larger departure from NSCAT under swell or mixed swell and wind wave conditions. Hwang et al. (1998) showed a systematic overestimation of TOPEX-derived sea surface slopes (corresponding to an underestimation of wind speed) compared to ®eld optical observations. They further identi®ed that the range of low wind speed is where the major discrepancy between the radar and optical measurements is observed (see Fig. 6b of Hwang et al. 1998), and these data were found to be under strong in¯uence of swells (Hwang and Shemdin 1988). There are also reasons to speculate that the bias of scatterometer-derived wind speed is swell correlated, though it might be much smaller compared to the case of altimeter. In an attempt to evaluate the accuracy of the ERS scatterometer wind measurement, Quilfen et al. (2001) generated a plot of wind speed difference between ERS scatterometers and Tropical Atmosphere FIG. 9. A scatter diagram of sea surface wind speed and signi®cant Ocean (TAO) buoys during 1992±98 (see their Fig. 8). wave height based on the collocated TOPEX±NSCAT dataset. The In that plot, a well-de®ned area of large bias (over Ϫ1.0 wind speeds are extracted from NSCAT, and the signi®cant wave Ϫ1 heights are extracted from TOPEX. The grayscale depicts the number ms ) appeared in the eastern equatorial Paci®c with of data within a 0.1 m sϪ1 ϫ 0.1 m grid box. Also overlaid is the an observable northward preference that they attributed theoretical relationship between wind speed and signi®cant wave to the impact of near-surface current. But given the co- height for a fully developed sea according to the WAM model. incidence of this area with the core of the Paci®c swell pool (see Fig. 2a of Chen et al. 2002a), it is very likely timated by the altimeter while underestimated by the a re¯ection of the sea state effect. scatterometer, thus giving divergent predictions of wind Next, we examine the fetch-dependent biases in re- climate for those regions. This could be largely the case motely sensed oceanic winds. The study of Glazman et in reality, as can be understood by relating Figs. 5a and al. (1988) demonstrated that a systematic bias in Seasat 5b to Fig. 6 of Chen et al. (2002a), in which dominant scatterometer due to the fetch effect is well pronounced. wind wave regions are indicated. Since fetch (either long It was found that the scatterometer tends to overestimate or short) is always important in these regions, they are the wind speed at long fetch and vice versa. The esti- considered to be largely overlapped with the prevailing mated error trend is roughly 0.5 m sϪ1/(100 km) of the fetch zones discussed here. Of course, other factors such generalized wind fetch. as rain, current, and sea surface temperature may also Glazman and Pilorz (1990) are among the ®rst who contribute to the overall bias. pointed out that the bias of altimeter-derived wind speed is correlated to fetch. According to their result, the bias b. A proposed link between swell/wind sea conditions of Geosat wind speed in a short fetch can be experi- and speed-dependent discrepancies mentally expressed as To have an idea of the sea state condition in the real ÄÄ0.31 5 ␧ϭϪf 4.16 ϩ 0.15X (X Ͻ 10 ). (1) ocean, a scatter diagram of the coincident NSCAT wind speed (U) and TOPEX signi®cant wave height (H ) from Note that XÄ is the nondimensional fetch normalized as s the collocation dataset is presented in Fig. 9. The gray- XÄ ϭ gX/U 2, where X is the dimensional fetch in meters, scale legend depicts density level of the data. Also over- and g is the acceleration of gravity. Equation (1) sug- laid on Fig. 9 is the theoretical relationship between wind gests that the altimeter underestimates the wind speed speed and signi®cant wave height for a fully developed by0±5msϪ1 for short fetches (see Fig. 10 of Glazman sea based on the Wave Modeling Project (WAM) model and Pilorz 1990). When Eq. (1) was used to correct the (Wamdi Group 1988) which is expressed as fetch effect of Geosat for the Japan Sea by Ebuchi et Ϫ22 al. (1992), who averaged the wind speed along the whole Hs ϭ 1.614 ϫ 10 U fetch, a positive bias of less than 0.5 m sϪ1 was obtained. If fetch-related biases do exist, the global wind maps (0 Յ U Յ 7.5 m sϪ1 ), (2a) based on scatterometer and altimeter measurements H ϭ 10Ϫ22U ϩ 8.134 ϫ 10Ϫ43U could contain false climate trends. Speci®cally, wind s speed averaged in long fetch zones would be overes- (7.5 m sϪ1 Ͻ U Յ 50 m sϪ1 ). (2b)

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC 790 MONTHLY WEATHER REVIEW VOLUME 132

As expected, the swell (wind sea) frequency decreases (increases) monotonically with wind speed. Interesting- ly, the intersection of the two curves (i.e., the turning point from swell dominance to wind sea dominance) appears at 14.2 m sϪ1, which falls into the ¯at band of the TOPEX±NSCAT wind speed biases (Fig. 8a). Figs. 8a and 10 thus seem to suggest that predominant swell or wind sea conditions (i.e., U Ͻ 10 m sϪ1 or U Ͼ 15 msϪ1, respectively) have opposite impacts on wind speed bias for both the altimeter and scatterometer. The altimeter bias shows a positive trend under swell dominance and a negative trend under wind sea dominance. The reverse is true for the scatterometer. Consequently, the wind speeds between 10 and 15 m sϪ1 may represent a band of transition from swell (very long fetch) to wind waves with initially long and then progressively shorter fetches. To further understand this argument, separate plots of wind bias against the reference speed under swell or wind sea dominated conditions are shown in Figs. 11a and 11b, respectively. It is obvious that a FIG. 10. Frequency of swell (circles) and wind sea (triangles) oc- monotonically increasing (decreasing) trend is observed currences as a function of wind speed over the global ocean based for TOPEX (NSCAT) under swell dominance that exists on the TOPEX±NSCAT±ECMWF collocation dataset. The thin hor- for wind speed below 10 m sϪ1. In Fig. 11b, the in- izontal line denotes a 50% frequency, and the dashed vertical line indicates the intersection of the swell and wind sea curves at 14.2 m creasing (decreasing) portion of the TOPEX (NSCAT) sϪ1. curve can be related to wind wave, long fetch conditions, whereas the rest of the curves beyond approxi- mately 14.2 m sϪ1 are produced by wind wave, short A direct implication of this graph is that the theoretical fetch conditions. Thus, having a wind wave, long fetch relationship can be used as a dividing line for sea state condition is somewhat equivalent to having a predom- maturity. Measurements lying below the curve are most- inant in¯uence of swell. In this context, there seems to ly from a growing sea, while those above the curve are be a contradiction between the results of Glazman et al. probably swell dominated (Chen et al. 2002a). Of course (1988), who ®nd a wind speed overestimate by the Seas- this division is not supposed to be valid in an absolute at scatterometer for long wind fetches, and Quilfen et sense because of the complexity of the wind wave/swell al. (2001), who ®nd a negative bias for the ERS scat- coupling. But it is expected to give a meaningful clas- terometers with predominant swell conditions. My re- si®cation of the two regimes from a statistical point of sults appear to be in support of those of Quilfen et al. view. It is apparent that a large majority of the data (2001). Note that a few months of Seasat data used by Glazman et al. (1988) might be too short, in a statistical points are above the theoretical U±Hs line that corre- sponds to a mature sea state, implying a systematic swell point of view, to reach a reliable conclusion. dominance in the world's oceans. Figure 9 also suggests, however, that the possibility of encountering an under- 6. Summary and conclusions developed sea increases rapidly with wind speed beyond 10msϪ1. Based on a collocated TOPEX±NSCAT±ECMWF da- In order to quantify the occurrences of swell and wind taset, a detailed investigation of the characteristics and wave events, two wind speed-related frequency indexes possible causes of systematic discrepancies between al- are introduced as (Chen et al. 2002a) timeter- and scatterometer-measured as well as model- predicted wind speeds is carried out. The general sta- Fss(U) ϭ N (U)/N(U), (3a) tistics show that the TOPEX and NSCAT wind measurements have a larger overall bias but a much smaller F (U) ϭ N (U)/N(U), (3b) ww overall variance compared to the ECMWF winds (Table where Ns(U) and Nw(U) are the number of swell and 2), implying that errors associated with satellite data are wind wave events for a given wind speed, respectively. mostly systematic while those associated with model

Note that N(U) ϭ Ns(U) ϩ Nw(U), thus Fs(U) ϩ Fw(U) outputs are basically random. This is con®rmed by a ϭ 1. This means that, in a relative sense, any individual constant phase reversal between TOPEX and NSCAT sea state is classi®ed as either swell dominated or wind wind speed biases as a function of latitude, month, and sea dominated. Figure 10 shows the swell and wind sea wind intensity. Maximum departures are found at Ϯ60Њ frequencies as a function of wind speed. Also overlaid in terms of latitude (Figs. 4a and 5) and in the Northern is a thin straight line corresponding to a 50% frequency. Hemisphere winter in terms of season (Fig. 7a), with

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC MARCH 2004 CHEN 791

speed rms is found near the equator for all three types of data (Fig. 4b), which is likely to be a consequence of rain contamination over the ITCZ. The performance of TOPEX and NSCAT are very close to each other in terms of rms error for medium winds, but the latter (former) is seen to be considerably ``better'' for low (high) winds (Fig. 8b). The results obtained in this study suggest that the reference wind speed de®ned in section 2b is a useful surrogate. This can be expected to some extent, given the widely accepted ϳ2msϪ1 accuracy of satellite and model wind estimates. Considering an extreme case, if all the TOPEX, NSCAT, and ECMWF winds are biased high (or low) at a given site, the reference wind will also be biased high (or low), even though the relative magnitude and phase relationship among the three wind speed biases remain unchanged. This argument is sup- ported by the consistency of the results with respect to other published ones, as summarized in Table 3. The rms error, however, might be affected to some degree by the ad hoc nature of the reference dataset. But comparisons between the statistics in Table 2 and those obtained from direct validations against in situ measurements in speci®c regions suggest that the use of this reference dataset leads to quantitatively meaningful rms results in a statistical sense. To conclude, as a result of sea state effect and algorithm de®ciency, systematic bias is a common feature in satellite wind measurements, which needs to be kept in mind and taken into account when altimeter, scatterometer, and model-derived wind speeds are interpreted, compared, or integrated.

Acknowledgments. This work was cosponsored by the FIG. 11. Wind speed bias under (a) swell and (b) wind sea con- Natural Science Foundation of China (Project Numbers ditions with respect to the reference wind speed. The thick solid line, 40025615 and 40271083) and the Teaching and Re- thick dashed line, and thin dashed line represent the TOPEX, NSCAT, and ECMWF results, respectively. The thin solid curves depict the search Award Program for Outstanding Young Teachers number of collocation data, and the horizontal lines denote zero bias. in Higher Education Institutions of MOE, PRC. The vertical dashed line at 14.2 m sϪ1 in (b) separates regions of (left) long fetch and (right) short fetch. REFERENCES

Atlas, R., R. N. Hoffman, S. C. Bloom, J. C. Jusem, and J. Ardizzone, TOPEX and NSCAT winds biased positively and neg- 1996: A multiyear global surface wind velocity dataset using atively, respectively. Another new ®nding of this study SSM/I wind observations. Bull. Amer. Meteor. Soc., 77, 869± is the low±high±low (high±low±high) pattern of the TO- 882. PEX (NSCAT) wind bias with respect to wind intensity Benada, R., 1997: Merged GDR (TOPEX/Poseidon) Generation B (user's guide). Rep. D-11007, Jet Propulsion Laboratory, Pas- (Fig. 8a), which is speculated to be a consequence of adena, CA, 84 pp. sea state maturity. Swell and wind sea appear to have Bentamy, A., Y. Quilfen, F. Gohin, N. Grima, M. Lenaour, and J. opposite effects on wind speed bias for both the altim- Servain, 1996: Determination and validation of average ®elds eter and scatterometer, which shows a clear three-band from scatterometer measurements. Global Atmos. Ocean Syst., structure corresponding to swell dominance (U Ͻ 10 m 4, 1±29. Ϫ1 ÐÐ, P.Queffeulou, Y. Quilfen, and K. Katsaros, 1999: Ocean surface s ), mixed swell and wind sea conditions (10 Ͻ U Ͻ wind ®elds estimated from satellite active and passive microwave 15msϪ1), and wind sea dominance (U Ͼ 15msϪ1), instruments. IEEE Trans. Geosci. Remote Sens., 37, 2469±2486. respectively (Figs. 10 and 11). Boutin, J., and J. Etcheto, 1990: Seasat scatterometer versus Scanning As far as the rms error is concerned, low values are Multichannel Microwave Radiometer wind speed: A comparison on a global scale. J. Geophys. Res., 95, 22 275±22 288. observed around Ϯ20Њ latitude in space (Fig. 4b) and ÐÐ, and ÐÐ, 1996: Consistency of Geosat, SSM/I, and ERS-1 between July and September in time (Fig. 7b) for both global surface wind speedsÐComparison with in situ data. J. satellite and model results. A signi®cant peak of wind Atmos. Oceanic Technol., 13, 183±197.

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC 792 MONTHLY WEATHER REVIEW VOLUME 132

ÐÐ, L. Siefridt, J. Etcheto, and B. Barnier, 1996: Comparison of induced by the large-scale component of the wave ®eld. J. Geo- ECMWF and satellite ocean wind speeds from 1985 to 1992. phys. Res., 93, 1317±1328. Int. J. Remote Sens., 17, 2897±2913. Gourrion, J., D. Vandemark, S. Bailey, and B. Chapron, 2000: Sat- ÐÐ, J. Etcheto, M. Ra®zadeh, and D. C. E. Bakker, 1999: Com- ellite altimeter models for surface wind speed developed using parison of NSCAT, ERS-2 active microwave instruments, Special ocean satellite crossovers. IFREMER Tech. Rep. IFREMER- Sensor Microwave Imager, and Carbon Interface Ocean Atmo- DROOS-2000-02, 62 pp.

sphere buoy wind speed: Consequences for the air±sea CO2 ex- Gower, J. F. R., 1996: Intercalibration of wave and wind data from change coef®cient. J. Geophys. Res., 104, 11 375±11 392. TOPEX/POSEIDON and moored buoys off the west coast of Busalacchi, A. J., R. M. Atlas, and E. C. Hackert, 1993: Comparison Canada. J. Geophys. Res., 101, 3817±3829. of Special Sensor Microwave Imager vector wind stress with Halpern, D., A. Hollingsworth, and F. Wentz, 1994: ECMWF and model-derived and subjective products for the tropical Paci®c. SSM/I global wind speeds. J. Atmos. Oceanic Technol., 11, 779± J. Geophys. Res., 98, 6961±6977. 788. Chen, G., and R. Ezraty, 1996: Alias impacts on the recovery of sea Hoffman, R. N., F. Wentz, D. Long, and K. Arai, 1994: Atmospheric level amplitude and energy from altimeter measurements. Int. J. losses at 14 GHz. NSCAT SWT Attenuation Subpanel Report, Remote Sens., 17, 3567±3576. Jet Propulsion Laboratory, Pasadena, CA, 13 pp. ÐÐ, B. Chapron, J. Tournadre, K. Katsaros, and D. Vandemark, 1997: Hollinger, J. P., J. L. Pierce, and G. A. Poe, 1990: SSMI instrument Global oceanic precipitation: A joint view by TOPEX and the evaluation. IEEE Trans. Geosci. Remote Sens., 28, 781±790. TOPEX microwave radiometer. J. Geophys. Res., 102, 10 457± Hwang, P. A., and O. H. Shemdin, 1988: The dependence of sea 10 471. surface slope on atmospheric stability and swell conditions. J. Geophys. Res., 93, 13 903±13 912. ÐÐ, ÐÐ, ÐÐ, ÐÐ, and ÐÐ, 1998: Identi®cation of possible ÐÐ, W. J. Teague, G. A. Jacobs, and D. W. Wang, 1998: A statistical wave damping by rain using TOPEX and TMR data. Remote comparison of wind speed, wave height, and wave period from Sens. Environ., 63, 40±48. satellite altimeters and ocean buoys in the Gulf of Mexico region. ÐÐ, H. Lin, and J. Ma, 2000: On the seasonal inconsistency of J. Geophys. Res., 103, 10 451±10 468. altimeter wind speed algorithms. Int. J. Remote Sens., 21, 2119± Meissner, T., D. Smith, and F.Wentz, 2001: A 10 year intercomparison 2125. between collocated Special Sensor Microwave Imager oceanic ÐÐ, B. Chapron, R. Ezraty, and D. Vandemark, 2002a: A global surface wind speed retrievals and global analyses. J. Geophys. view of swell and wind sea climate in the ocean by satellite Res., 106, 11 731±11 742. altimeter and scatterometer. J. Atmos. Oceanic Technol., 19, Naderi, F. M., M. H. Freilich, and D. G. Long, 1991: Spaceborne 1849±1859. radar measurement of wind velocity over the ocean: An overview ÐÐ, R. Ezraty, C. Fang, and L. Fang, 2002b: A new look at the of the NSCAT scatterometer system. Proc. IEEE, 79, 850±866. zonal pattern of the marine wind system from TOPEX mea- Queffeulou, P., B. Chapron, and A. Bentamy, 1999: Comparing Ku surements. Remote Sens. Environ., 79, 15±22. band NSCAT scatterometer and ERS-2 altimeter winds. IEEE ÐÐ, J. Ma, C. Fang, and Y. Han, 2003: Global oceanic precipitation Trans. Geosci. Remote Sens., 37, 1662±1670. derived from TOPEX and TMR: Climatology and variability. J. Quilfen, Y., B. Chapron, and D. Vandemark, 2001: On the ERS scat- Climate, 16, 3888±3904. terometer wind measurement accuracy: Evidence of seasonal and Dunbar, R. S., 1997: NASA Scatterometer: High-resolution merged regional biases. J. Atmos. Oceanic Technol., 18, 1684±1697. geophysical data product (user's guide). Jet Propulsion Labo- Rienecker, M. M., R. Atlas, S. D. Schubert, and C. S. Willet, 1996: ratory, Pasadena, CA, 17 pp. A comparison of surface wind products over the North Paci®c Ebuchi, N., H. Kawamura, and Y. Toba, 1992: Growth of wind waves Ocean. J. Geophys. Res., 101, 1011±1023. with fetch observed by the Geosat altimeter in the Japan Sea Schlax, M. G., and D. B. Chelton, 1994: Aliased tidal errors in Topex/ under winter monsoon. J. Geophys. Res., 97, 809±819. Poseidon sea surface height data. J. Geophys. Res., 99, 24 761± 24 775. Freilich, M. H., and R. S. Dunbar, 1999: The accuracy of the NSCAT Vachon, P. W., and F. W. Dobson, 1996: Validation of wind vector 1 vector winds: Comparisons with National Data Buoy Center retrieval from ERS-1 SAR images over the ocean. Global Atmos. buoys. J. Geophys. Res., 104, 11 231±11 246. Ocean Syst., 5, 177±187. Fu, L.-L., and A. Cazenave, Eds., 2001: Satellite Altimetry and Earth WAMDI group, 1988: The WAM modelÐA third generation ocean Sciences. International Geophysical Series, Vol. 69, Academic wave prediction model. J. Phys. Oceanogr., 18, 1775±1810. Press, 463 pp. Wentz, F. J., and D. K. Smith, 1999: A model function for the ocean- ÐÐ, E. J. Christensen, C. A. Yamarone, M. Lefebvre, Y. MeÂnard, normalized radar cross section at 15 GHz derived from NSCAT M. Dorrer, and P. Escudier, 1994: TOPEX/POSEIDON mission observations. J. Geophys. Res., 104, 11 499±11 514. overview. J. Geophys. Res., 99, 24 369±24 381. Witter, D. L., and D. B. Chelton, 1991: A Geosat altimeter wind Glazman, R. E., and S. H. Pilorz, 1990: Effects of sea maturity on speed algorithm and a method for altimeter wind speed algorithm satellite altimeter measurements. J. Geophys. Res., 95, 2857± development. J. Geophys. Res., 96, 8853±8860. 2870. Young, I. R., 1999: Seasonal variability of the global ocean wind and ÐÐ, G. G. Pihos, and J. Ip, 1988: Scatterometer wind speed bias wave climate. Int. J. Climatol., 19, 931±950.

Unauthenticated | Downloaded 10/02/21 02:41 PM UTC