Monitoring Surface Velocity from Repeated ADCPObservations and Satellite Altimetry
KAORU ICHIKAWA1*, NORIAKI GOHDA2, MASAZUMI ARAI2 and ARATA KANEKO2
1Research Institute for Applied Mechanics, Kyushu University, Kasuga, Fukuoka 816-8580, Japan; also at Frontier Observational Research System for Global Change, Yokosuka, Kanagawa 237-0061, Japan2Division of Global Environment Engineering, Graduate School of Engineering, Hiroshima University, Higashi-Hiroshima 739-8527, Japan
(Received 20 May 2003; in revised form 18 October 2003; accepted 18 October 2003)
A method has been developed to monitor the surface velocity field by combining re-peated acoustic Doppler current profiler (ADCP) observations and satellite altimetrydata. The geostrophic velocity anomaly is calculated from the sea surface heightanomaly field estimated from the altimetry data by an optimal interpolation. It hasbeen confirmed that this accurately observes the smoothed velocity anomaly fieldwhen the interpolation scales are set according to the spatio-temporal sampling pat-tern of the altimeter used. The velocity anomaly obtained from the altimetry data issubtracted from the repeated ADCP observations to estimate temporal mean velocityalong the ship tracks. Regularly sampled, nine-year time series of surface velocitycan then be obtained by adding the computed mean velocity and the altimetry anomalycomponents. This clearly illustrates surface velocity fluctuations such as the move-ment of the Kuroshio axis due to its meandering and an increase of the interannualvariability of the Subtropical Countercurrent toward its downstream region.
Several researchers have nevertheless tried to recoverthe missing mean SSDT (e.g. Qiu et al., 1991; Ichikawaand Imawaki, 1994). One of the most promising methodsseems to be the combined use of the trajectories of sur-face drifters, which can support a wide coverage of thesurface velocity field (Uchida and Imawaki, 2003).Lagrangian observations such as drifters, however, maybe subject to the current field itself; i.e., the density ofobservations in a given area may depend on the conver-gence of the flow field. Some repeated Eulerian observa-tions of velocity field would therefore also be importantto establish monitoring systems.
Relatively wide temporal and/or spatial coverage ofin situ velocity observations can be obtained by acousticDoppler current profilers (ADCP’s) mounted on volun-teer ships (Hanawa et al., 1996; Kaneko et al., 1998, 1999,2001). These can provide not only frequent Eulerian sur-face velocity data, but also vertical current structures. Theuse of these data in a monitoring system, however, in-volves some difficulties due to the irregularity of theobservations, especially when a ship takes multiple routes;the routes of commercial ships may depend on economicdemands as well as weather conditions (Kaneko et al.,1998, 1999). In the present study, a method has thereforebeen developed to regularly monitor the surface velocity
1. IntroductionSince most large-scale signals in the ocean tend to
propagate westward, variations in mid-ocean areas mayalter the western boundary currents such as the Kuroshio(e.g. Ichikawa, 2001; Tanaka and Ikeda, 2004). It is there-fore crucial to monitor the signals in a wide area to studyand/or predict the western boundary currents (e.g.Kamachi et al., 2004).
Satellite observations are entirely suitable for moni-toring wide areas, although the observations are restrictedto surface phenomena. Satellite altimetry is especiallyuseful since it can provide observations of the sea sur-face dynamic topography (SSDT; deviation of the seasurface height from the equi-geopotential geoid surface)which is directly related to the surface geostrophic ve-locity field itself. However, since the available geoidmodels are not sufficiently precise, and will not be, untildrastic improvements are made, which are expected oncompletion of the satellite geodetic missions (NationalResearch Council, 1997), only the temporal variations ofthe SSDT can be accurately obtained from the altimetrydata so far.
366 K. Ichikawa et al.
field by combined use of repeated ADCP observationsand altimetry data. Use of the ADCP data provides amethod to estimate the missing temporal mean from theanomalies of the altimetry data, and the altimetry data, inturn, assure the regular monitoring of the surface veloc-ity field.
The remainder of the present paper is organized asfollows. The data and method used are described in Sec-tions 2 and 3, respectively; two independent data sets,ADCP velocity and the surface geostrophic velocity de-termined from the altimetry data, are carefully comparedin Subsection 3.2. The results obtained, including the tem-poral mean velocity field and the time series of the sur-face velocity field, are shown in Section 4, which alsoprovides detailed descriptions on variations of theKuroshio and the Subtropical Countercurrent. Discussionsof the results and concluding remarks are presented inthe following Section 5.
2. DataA research program to observe North Pacific veloc-
ity field by an ADCP mounted on a commercial mineraltransport ship “First Jupiter” started in 1997, led byHiroshima University with partial support from the CoreResearch for Evolutional Science and Technology of theJapan Science and Technology Corporation (Kaneko etal., 1998, 1999, 2001). The program has provided sur-face velocity observations in the upper 200-m layer along43 ship tracks during 1997–2001; these data are avail-able from our web site (http://www.ocean.hiroshima-u.ac.jp/).
Among these ship tracks, the tracks between Kashimaand the eastern side of Australia (KE), Kashima and thewestern side of Australia (KW), and Wakayama and thewestern side of Australia (WW), have been repeated 11,14 and 6 times, respectively, although they are irregu-larly spaced in time (Fig. 1; Table 1). At each point (ap-proximately 9.6 km apart along a track) on these threetracks, the most extreme outlier from the average is omit-ted in turn from the repeated observations until the stand-ard deviation of the zonal position is smaller than the in-ternal Rossby radius at a given latitude. If fewer than fiveobservations remain at the given point, we discarded allthe observations at that point to reduce the effect of high-frequency noise in the ADCP data. We used only the shal-lowest ADCP velocity data (70-m depth) to compare withthe surface geostrophic velocity determined from thealtimetry data; no clear spatial or temporal dependencyis found in the vertical shear of the observed ADCP ve-locity, suggesting that surface intensified ageostrophicflow such as the Ekman wind drift is not dominant in theADCP data.
Altimetry data sets of TOPEX/POSEIDON (T/P),European Remote Sensing satellite (ERS)-1 and ERS-2
used in the present study are distributed by AVISO (LeTraon et al., 1995; AVISO, 1996; Le Traon and Ogor,1998). From all these altimetry data, the temporal anomalyof the SSDT field, which is equivalent to the sea surfaceheight anomaly (SSHA) field, is determined on a 0.25°grid by an optimal interpolation every 9.92 days (a re-peat cycle of T/P) during a nine-year period betweenOctober 1992 and September 2001. The optimal interpo-lation used in the present study is the same as that de-scribed by Ichikawa (2001), with various spatio-tempo-ral smoothing scales (namely, the decorrelation scales ofthe Gaussian correlation function of the SSHA), as de-scribed in Subsection 3.2.
3. Method
3.1 FundamentalsThe surface velocity observed by the ADCP vship(r,
ti) at time ti and position r can be regarded as consistingof several components, as
v r v r v r v rship i i it t t, , , ,( ) = ( ) + ( ) + ( ) ( )′ ″ 1
where v indicates the temporal mean velocity, v′ repre-sents the temporal velocity anomaly which is regarded tobe in geostrophic balance, and v″ represents high-fre-quency ageostrophic fluctuations such as tidal currents,inertial oscillations and Ekman wind drift. The purposeof the present analysis is to monitor surface quasi-geostrophic velocity at a given position r by obtainingthe term { v (r) + v′(r, t)} at any time t.
In general, the velocity anomaly v′(r, ti) is expectedto be replaced by the geostrophic velocity anomaly v′alt(r,
KE
KW
WW
Kii Pen.
Fig. 1. Locations of the ADCP tracks. Thin horizontal barsindicate the zonal standard deviation at given latitude.
Monitoring Velocity by ADCP and Altimetry 367
ti) determined from the satellite altimetry SSHA data.Combined use of the repeated ADCP data and the satel-lite altimetry data thus provides several estimates of themean velocity v (r) with high-frequency noise v″(r, ti) ateach observation time ti; namely,
v r v r v r v rship i alt i it t t, , , .( ) − ′ ( ) ( ) + ( ) ( )� ″ 2
Provided that the high-frequency noises are reducedto a negligible magnitude by statistical averaging overmany ti, or ⟨v″⟩ � 0 where ⟨...⟩ denotes statistical averag-ing, the average of Eq. (2) should provide a good esti-mate of the mean velocity v (r). The fundamental con-cept to determine the mean velocity v by subtracting thealtimetry temporal anomaly component v′alt from the insitu velocity observations is same method as that de-scribed by Uchida and Imawaki (2003). Once the meanvelocity v (r) at the position r is obtained, the absolutegeostrophic velocity at any time t can be obtained by com-bining v (r) with the geostrophic velocity anomaly v′alt(r,t) obtained from the altimetry SSHA data at the givenposition and time.
3.2 Representativeness of v′alt for v′Altimetry measures the SSHA along the satellite
subtrack which is equivalent only to the geostrophic ve-locity anomaly component normal to the subtrack, so thatsome sort of interpolation is necessary to obtain the vec-tor velocity field v′alt. Since replacement of v′ and v′alt isthe most essential step in the present method, the similar-ity between v′ and the interpolated vector velocity v′altshould be carefully examined.
In the present study, the effect of interpolation onv′alt is investigated by comparing three altimetry SSHAdata sets with different smoothing scales of the optimalinterpolation. In two data sets, the SSHA at a point isestimated from data points on surrounding satellite
subtracks, all of which are assumed to be correlated toeach other. The spatio-temporal smoothing scales of theoptimal interpolation are thus set to the sampling con-figuration of the satellite (Ichikawa and Imawaki, 1996);we used smoothing scales of approximately 340 km forthe zonal distance, 223 km for the meridional distanceand 5 days for the temporal differences as typical T/Psampling scales in the study area, and also smoothingscales of 83 km, 110 km, 17.5 days, respectively, for typi-cal ERS sampling scales. Meanwhile, in the third dataset, the SSHA at a point is interpolated mainly with ref-erence to the closest data point on a single closest satel-lite subtrack so that the spatial smoothing scales are keptsmall to reproduce small-scale features (42 km, 42 km,17.5 days). These three data sets are hereafter called T/P-sampling SSHA, ERS-sampling SSHA, and closest-ref-erence SSHA, respectively.
The calculated v′alt from the SSHA field of these threedata sets should be compared with the observedgeostrophic velocity anomaly v′ itself. It is impossible,however, to extract the velocity anomaly v′ alone fromthe ADCP data. Therefore, in the present study, the ob-served ADCP velocity vship is instead compared with thecomposite velocity; the composite velocity is the combi-nation of the geostrophic velocity anomaly v′alt deter-mined by the altimetry SSHA data and the approximatedmean geostrophic velocity v obtained from theclimatological mean SSDT data (Ichikawa and Imawaki,1994; Ichikawa et al., 1995). Note that the comparisonbetween the ADCP velocity vship = v + v′ + v″ and thecomposite velocity v + v′alt includes an additional de-grading component due to the high-frequency noise v″and the inconsistency between the mean fields v and v ,so that the comparison in the present study is worse thanthe comparison between v′ and v′alt that was originallyrequired. The inter-comparison between the three SSHAdata sets, however, would not be affected by this degra-dation since effects due to v – v and v″ are common in
Table 1. Observation dates for each ship track shown in Fig. 1.
the inter-comparison. The approximate mean geostrophicvelocity v used in the present study was calculated fromthe climatological mean geopotential anomaly data de-termined from in situ observations referred to the 1000decibar surface (Teague et al., 1990).
Scatter diagrams of the zonal and meridional com-ponents of these three data sets are plotted in Figs. 2–4together with the number of points used, slope and inter-cept of the regression line, cross correlation coefficient,and root-mean-square (RMS) difference; each regressionline is determined so that the sum of distances betweenthe line and each data point is minimized. All ADCP datashown in Table 1 are used in the figures. For both regres-sion lines for zonal and meridional components, the slopedecreases as the spatial smoothing scale becomes larger;especially for v′alt calculated from the T/P-sampling SSHA(Fig. 2), the slope for the meridional component is only0.24. The decrease of the slope is caused by spatialsmoothing of the altimetry velocity v′alt induced by theoptimal interpolation of the SSHA, which generally re-
duces extreme velocities.On the other hand, the correlation coefficient is not
a simple function of the smoothing scales. Larger spatialsmoothing does decrease the correlation between the T/P-sampling and ERS-sampling data sets (Figs. 2 and 3),but the worst result is obtained from the closest-refer-ence SSHA data set with the smallest spatial smoothingscales (Fig. 4). This is probably because the slope of theSSHA at a given point has been taken into account dur-ing the optimal interpolation for the T/P-sampling andERS-sampling data sets by simultaneously referring tosurrounding data points, meanwhile only the SSHA at theclosest point, not the slope, is referred in the closest-ref-erence data set. These results indicate that, in order togive an accurate estimate of geostrophic vector velocity,interpolation of the SSHA should simultaneously refer todata points on surrounding satellite subtracks, although asmaller smoothing is preferable. In the present study,threfore, the geostrophic velocity anomaly from the ERS-sampling SSHA data set is hereafter used as v′alt.
Vship
a) b)
Uship
Numb=8680
slope=0.48
intercept=0.01
CorrC.=0.60
RMS diff=0.20
Numb=8680
slope=0.24
intercept=0.00
CorrC.=0.49
RMS diff=0.23
U' a
lt +
U~
V' a
lt +
V~
Vship
a) b)
Uship
Numb=8982
slope=0.59
intercept=0.01
CorrC.=0.65
RMS diff=0.21
Numb=8982
slope=0.49
intercept=0.00
CorrC.=0.54
RMS diff=0.23
U' a
lt +
U~
V' a
lt +
V~
Fig. 2. Scatter diagrams for zonal (a) and meridional (b) components between the ADCP velocity vship and the composite geostrophicvelocity v + v′alt determined from the T/P-sampling SSHA data set. Regression for each component is indicated by brokenlines, together with solid oblique lines as a reference of v′alt + v = vship; the unit of the axes, intercept, and RMS difference ism/s. Numbers of points used and cross correlation coefficient are also shown in each panel.
Fig. 3. As Fig. 2, but for the ERS-sampling SSHA data set.
Monitoring Velocity by ADCP and Altimetry 369
As seen in Fig. 3, the composite velocity v + v′alt isunderestimated compared with the ADCP velocity vship,and this is considered to be due to smoothing of thealtimetry data. This idea is confirmed in Fig. 5 in whichthe ADCP data is smoothed along the ship tracks with thesame decorrelation scales used in the ERS-sampling op-timal interpolation; the data points used in the figure arelimited to a separation of 90 km along the ship tracks sothat the comparison is independent after the smoothing.The slopes of the regression lines are found to be signifi-cantly improved, becoming almost unity. Taking accountof the possible contamination by v″, which may contami-nate the cross correlation coefficient and the RMS differ-ence but may not affect the regression line by assumingits randomness, this result confirms that altimetry canobserve the smoothed surface velocity field very accu-rately. Therefore, the ADCP data in the present analysisare hereafter smoothed over the ERS-sampling scales toaccord with the altimetry data, in order to maintain theassumption of the similarity of the velocity anomalies,v′ � v′alt, in Eq. (2).
4. Results
4.1 Temporal mean velocity vThe nine-year mean surface velocity v estimated by
averaging Eq. (2) from the smoothed ADCP data and theERS-sampling SSHA data is shown in Fig. 6(a). In orderto illustrate the improvement of the estimated mean ve-locity field v , a simple statistical average of the smoothedADCP velocity data ⟨vship⟩ is plotted in Fig. 6(b). Bothpanels show the westward velocity component along alltracks south of 16°N, and both exhibit a similar, strongKuroshio structure along the WW track south of the KiiPeninsula.
Significant discrepancies are also found in some ar-eas, though, especially south of the Kuroshio and around20°N. For example, a strong eastward velocity core around31°N along the KW and KE tracks is found only in Fig.6(b). The westward Kuroshio recirculation around 29°Nalong the WW track is clearly stronger in Fig. 6(a) thanin Fig. 6(b). Eastward velocity components related to theSubtropical Countercurrent (e.g. Hasunuma and Yoshida,
Vship
a) b)
Uship
Numb=8858
slope=0.64
intercept=-0.00
CorrC.=0.52
RMS diff=0.24
Numb=8858
slope=0.55
intercept=0.01
CorrC.=0.37
RMS diff=0.28
U' a
lt +
U~
V' a
lt +
V~
Vship
a) b)
Uship
Numb=878
slope=1.14
intercept=0.01
CorrC.=0.69
RMS diff=0.14
Numb=878
slope=0.99
intercept=0.01
CorrC.=0.60
RMS diff=0.17
U' a
lt +
U~
V' a
lt +
V~
Fig. 4. As Fig. 2, but for the closest-reference SSHA data set.
Fig. 5. As Fig. 3, but with the ADCP velocity smoothed along each ship track with the same scales used in the ERS-samplingSSHA data set. Data points are limited to a separation of 90 km along the ship tracks so that the comparison is independentafter smoothing.
370 K. Ichikawa et al.
mean velocity v itself since the statistical mean of theanomaly ⟨v′⟩ may not be negligibly averaged out, com-pared with the weaker mean velocity v .
4.2 Seasonal and interannual variabilityThe time series of the surface velocity v + v′alt along
each track can be determined at a 10-day interval. In Fig.8, the velocity anomaly v′alt is averaged over 89 days (nineT/P repeat cycles) in order to reduce variations caused bytransient mesoscale eddies so that seasonal andinterannual variations can be focused on in this subsec-tion.The Kuroshio
During the nine-year study period, the Kuroshio isknown not to have taken the so-called stationary largemeander offshore path, but it often took a large meander-ing path in 1999 and later. This can clearly be seen inFig. 8(b) as the Kuroshio axis is shifted southward alongthe KW track during that period. In Fig. 8(a), too, a sig-nificant eastward current between 30°–33°N is occasion-ally observed to the south of the strong northeastwardcurrent at 35°N (this is especially clear in 1999 and 2001),which is because the KE track passed the Kuroshio twice;
b)
0.5 m/s0.5 m/s
a)
Fig. 6. Mean surface velocity v along the ship tracks determined from the ADCP data and the altimetry data (a); as a compari-son, the statistical mean ADCP surface velocity ⟨vship⟩ is plotted in the same way (b). For convenience, the velocity vectors areplotted for every three points along a ship track (approximately 29 km apart). A thin ellipse at the top of each vector indicatesthe standard deviation in zonal and meridional components. Eastward velocity with 0.5 m/s speed is plotted at the bottom ofeach panel as reference.
1978; Qiu, 1999) are found in all three tracks (20°–23°Nfor WW; 20°–22.5°N for KW, and 18.5°–20°N for KE)in Fig. 6(a), while no eastward flow is present along theWW track in Fig. 6(b). Moreover, the eastward velocitycomponents in Fig. 6(b) along the KW track (18°–20°N)and the KE track (17°–19.5°N) are located to the southof those in Fig. 6(a), with larger standard deviation inFig. 6(b) than in Fig. 6(a).
These discrepancies are clearly confirmed in Fig. 7,in which the velocity difference between the two panelsin Fig. 6, v – ⟨vship⟩, is plotted over a distribution map ofthe difference between speed of the climatological meanvelocity | v | and the standard deviation of the altimetry
velocity anomaly ′( )valt2
. All the areas described above
as discrepancies are marked by larger vectors in Fig. 7,and most of them are associated with a mean speed | v |that is much smaller than the variability of the velocity
anomaly ′( )valt2
. This suggests that the velocity anomaly
v′ tends to be quite a lot larger than the mean velocity vin such areas, so that the simple statistical mean of theADCP velocity ⟨vship⟩ � v + ⟨v′⟩ would differ from the
Monitoring Velocity by ADCP and Altimetry 371
it intersected the Kuroshio at 35°N, and it also crossed anarea where a large meander of the Kuroshio interactedwith a cyclonic ring to the east, which is seen in the com-posite SSDT (see Ichikawa and Imawaki, 1994, as an ex-ample during the Geosat period). Note that these tracesof the Kuroshio meander remain in ⟨vship⟩ as if the south-ern branch of the Kuroshio were centered at about 31°N(Fig. 6(b)), since the dates of ADCP observations indi-cated by triangle marks in Fig. 8 concentrate in the laterhalf of the nine-year period.
Meanwhile, along the WW track in Fig. 8(c), theKuroshio took a relatively stable path, although slightsouthward shifts of the Kuroshio axis are found during2000–2001. This is consistent with Fig. 7, which indi-cates that the temporal mean velocity is larger than theanomaly in the Kuroshio region west of Kii Peninsula.The Subtropical Countercurrent
As described in Subsection 4.1, the region at 20°–25°N, where the Subtropical Countercurrent (hereafterabbreviated STCC) is present, is marked in Fig. 7 by alarger temporal velocity anomaly than the mean speed,similar to the downstream Kuroshio region east of KiiPeninsula. Actually, even in the 89-day averaged veloc-ity in Fig. 8, the STCC is so variable that its mean struc-ture, shown in Fig. 6, is not significantly resolved along
all tracks.Interannual variability of the STCC, however, is re-
vealed to depend on location by using the long-term timeseries. Figure 9 shows the time series of the 1-year aver-aged velocity at 16°–29°N with tilting of vector plots;the tilt angle for each panel is determined so that the meanvelocity v in Fig. 6(a) that corresponds to the STCC isapproximately plotted as a horizontal vector. The south-eastward STCC along the WW track (Fig. 9(c)) is con-sistently present at about 19°–23°N, although its width
[m/s]
0.2
0.1
0.0
-0.1
-0.2
0.2m/s
b) KW
c) WW
a) KE
0.5m/s
Fig. 7. Difference between the panels (a) and (b) in Fig. 6. Thereference eastward 0.2 m/s vector is plotted at the top ofthe panel. The vectors are plotted for every 3 points alongtracks as in Fig. 6. They are superimposed on the map ofspeed difference between the climatological mean velocityv and the standard deviation of the velocity anomaly v′alt.Positive speed difference (shaded with lines) indicates thatthe current is so variable that the speed of the velocityanomaly v′alt tends to be larger than that of the mean veloc-ity v . Negative values (shaded with dots) are observed onlyin the areas close to the southwestern coast of Japan and atlower latitudes.
Fig. 8. Time series of the surface velocity v + v′alt along theship tracks KE (a), KW (b) and WW (c). Northward veloc-ity is plotted as an upward vector; reference eastward 0.5m/s vector is plotted at the top of the panel (c). Each vectoris determined by averaging v′alt over nine T/P cycles (89days), and they are plotted for every three points as in Fig.6. Triangles at the bottom of each panel indicate dates ofthe ADCP observations shown in Table 1.
372 K. Ichikawa et al.
the WW track, but the eastward current is found to beseparated in 1996 to the south (18°–19°N) and to the north(23°–24°N). Similarly, the northeastward STCC at 18°–20°N along the KE track (Fig. 9(a)) is found, except in1994, 1998 and 2001, while another northern band of east-ward current at 22°–23°N is also found in 1994–1995 andin 1997. These would be related to the double subsurfacedensity fronts at 24°N and 18°N, recently reported by Aokiet al. (2002). It should also be noted that the interannualvariability of the STCC evidently becomes larger alongthe ship tracks in the downstream of the STCC.
Seasonality of the STCC has been discussed in pre-vious studies (e.g. Qiu, 1999; Kobashi and Kawamura,2002), which generally concluded that the STCC tends
a) KE
b) KW
c) WW
0.2m/s
0.2m/s
0.2m/s
0.2m/s
a) KE
b) KW
c) WW
0.2m/s
0.2m/s
Feb.-Apr. May-July Aug.-Oct. Nov.-Jan.
Feb.-Apr. May-July Aug.-Oct. Nov.-Jan.
Fig. 9. As Fig. 8, but each profile is determined by averagingover 37 T/P cycles (367 days), and the vectors are plottedbetween 16°N and 29°N with tilt by –45° (a), 30° (b) and60° (c). Reference eastward 0.2 m/s vector is plotted at theleft-hand corner of each panel.
Fig. 10. Seasonally averaged surface velocity v + v′alt alongthe ship track KE (a), KW (b) and WW (c). All nine-yeardata is sorted in periods between February–April, May–July,August–October and November–January. Tilting angles,interval of the vector plots, and reference eastward veloc-ity are as in Fig. 9, and thin ellipses are as in Fig. 6.
shows slight variations, such as the broader band in 1998.It is quite impressive that even this significant, steadySTCC would be easily dismissed in a short-term or sin-gle “snapshot” observation by overlaying the much largervelocity anomaly induced by transient mesoscale eddies.Indeed, the statistical average of the six ADCP observa-tions ⟨vship⟩ in Fig. 6(b) cannot represent the existence ofthe STCC. Note that there is also another stable south-eastward current at 25°–28°N, but this is considered notto be the STCC but accompanies the Kuroshiorecirculation at 27.5°–29.5°N.
Furthermore, the STCC along the KW and KE tracksexhibits clear interannual variations. The eastward STCCalong the KW track is frequently found in Fig. 9(b) at20°–22°N, the downstream area of the stable STCC along
Monitoring Velocity by ADCP and Altimetry 373
Fig. 11. Mean surface velocity determined from the altimetrydata and surface drifters (Uchida and Imawaki, 2003) ex-tracted along the ship tracks. Thin ellipses indicate the zonaland meridional errors. The reference vector and plottingintervals are as in Fig. 6, but no data are plotted if the meanvelocity is missing.
to be strengthened in summer and weakened in winter. InFig. 10, which shows the seasonally averaged vector pro-duced by annually folding the nine-year data, this ten-dency can be discerned, although its standard deviationis generally very large due to both interannual andintraseasonal variability. For example, all the STCCcentered at 20°N along the WW track, at 20.5°N alongthe KW track, and at 19°N along the KE track, becomesstronger and wider in May–July. Note that the northernband of eastward current along the KE track centered at23°N described above becomes significant not only inMay–July but also in November–January; this eastwardcurrent shows a quite pronounced semiannual variabil-ity.
5. Discussion and SummaryA method to regularly monitor the surface velocity
has been developed by combined use of repeated ADCPobservations vship and altimetry data. The mean velocityv along the ship track is first determined from vship bysubtracting surface geostrophic velocity anomaly v′altcalculated from the altimetry SSHA field. In order to ob-tain accurate vector velocity v′alt, it is found that the SSHAat a point should be estimated from altimetry data pointson surrounding satellite subtracks so that the slope of theSSHA at the point is accounted for in the interpolation.In addition, it is also necessary to smooth the ADCP vshipdata over the same scales as the SSHA data in order togive agreement with the interpolated altimetry data v′alt.In the present analysis, the best smoothing scales for thealtimetry SSHA data and ADCP data are the samplingpatterns of the ERS-1/2 altimeters; note that this conclu-sion may not hold in areas close to the Equator wherelarge-scale phenomena occur in short time scales, forwhich the T/P sampling would be more suitable.
As mentioned in Section 1, another mean velocityfield is available, which is determined similarly fromaltimetry data and surface drifter trajectories (Uchida andImawaki, 2003). The drifter-based velocity field is notlimited to ship tracks as in Fig. 6, so it can cover a widerarea. However, when the drifter-based nine-year meanvelocity vdrift is extracted along the ship tracks used inthe present study (Fig. 11), numerous missing areas arerevealed, where there are too few drifter observations todetermine the mean velocity vdrift accurately. The miss-ing field is especially serious in the Kuroshio recirculationarea around 30°N along the KW and KE tracks. More-over, in the STCC region the mean velocity field is sosegmented that the presence of the STCC is not clearlyrecognized, at least along these ship tracks. Meanwhile,in the Kuroshio region where plenty of drifters passed, acurrent structure similar to Fig. 6(a) is present with sig-nificantly larger maximum speed and horizontal velocityshear. It should be noted, however, that since the
geostrophic vector velocity v′alt from the altimetry datanever modifies any small-scale current structure even inUchida and Imawaki’s (2003) method, the small-scalestructure related to the large horizontal velocity shear ofthe Kuroshio is subject to the reliability of the statisticalaverage of intermittent observations of the drifter veloc-ity ⟨vdrift⟩ . Note too that the presence of a largeageostrophic component is indicated in Fig. 11 by con-vergence of the Kuroshio along the WW track and diver-gence along the KW and KE tracks. The ADCP velocityat 70-m depth used in the present study would representupper layer density structures more directly than the sur-face velocity determined from drifters with 15-m depthdrogues, which would be more affected by ageostrophicvelocity such as the Ekman wind drift. These advantagesensure that the present method produces a better moni-toring system along the given ship tracks, though it isspatially limited.
Although the standard deviation of the estimatedmean field v (Fig. 6(a)) is generally smaller than that ofthe simple statistical average of the ADCP data ⟨vship⟩ (Fig.6(b)), it still has an order of magnitude 0.1 m/s. One pos-sible reason for this is the introduction of errors by re-placement of v′alt and v′ in Eq. (2), which would be sig-nificant at lower latitudes below 13°N where the stand-ard deviation of the estimated mean v is larger than that
0.5 m/s
374 K. Ichikawa et al.
for the simple average ⟨vship⟩. Generally, an error ingeostrophic velocity v′alt calculated from the SSHA isexpected to be larger at lower latitude since the magni-tude of the inverse Coriolis parameter |1/f | increasescloser to the Equator. Another possible reason for the rela-tively larger standard deviation of the mean velocity v isthat the high-frequency noise v″ would be considerable,although the Ekman wind drift is expected to be not sostrong at 70-m depth, as mentioned above. This wouldfurther suggest that a single ADCP observation may becontaminated by high-frequency ageostrophic current, v″,so that it may induce unrealistic results unless treatedcarefully.
Using the estimated mean v and the anomaly v′altfrom the altimetry data, time series of the velocity alongthe ship tracks can be obtained at a 10-day regular inter-val. Since the downstream Kuroshio east of Kii Penin-sula and the STCC are very variable, as indicated in Fig.7, frequent, regularly-sampled data is quite useful to dis-cuss long-term variations which are never revealed by asingle “snapshot” observation. In the time series obtained,interannual variations of the Kuroshio and the STCC arewell described, as expected. In particular, the interannualvariation of the STCC is shown to become larger down-stream of the STCC. Meanwhile, even the relatively sta-ble STCC along the WW track is often dismissed in asingle “snapshot” observation by overlaying muchstronger mesoscale eddies, which suggests the difficultyof studying STCC variations based on a few “snapshot”observations.
AcknowledgementsDr. Hiroshi Uchida kindly provided data of the mean
velocity field. The authors also appreciate helpful sug-gestions from two anonymous reviewers. This study issupported by the Core Research for Evolutional Scienceand Technology of the Japan Science and TechnologyCorporation and partially by a Grant-in-Aid for Scien-tific Research from the Ministry of Education, Science,Sports and Culture of Japan.
ReferencesAoki, Y., T. Suga and K. Hanawa (2002): Subsurface subtropi-
cal fronts of the North Pacific as inherent boundaries in theventilated thermocline. J. Phys. Oceanogr., 32, 2299–2311.
Hanawa, K., Y. Yoshikawa and T. Taneda (1996): TOLEX-ADCP monitoring. Geophys. Res. Lett., 23, 2429–2432.
Hasunuma, K. and K. Yoshida (1978): Splitting of the subtropi-cal gyre in the western North Pacific. J. Oceanogr. Soc.Japan, 34, 160–172.
Ichikawa, K. (2001): Variation of the Kuroshio in the TokaraStrait induced by meso-scale eddies. J. Oceanogr., 57, 55–68.
Ichikawa, K. and S. Imawaki (1994): Life history of a cyclonic
ring detached from the Kuroshio Extension as seen by theGeosat altimeter. J. Geophys. Res., 99(C8), 15953–15966.
Ichikawa, K. and S. Imawaki (1996): Estimating sea surfacedynamic topography from Geosat altimetry data. J.Oceanogr., 52, 43–68.
Ichikawa, K., S. Imawaki and H. Ishii (1995): Comparisons ofaltimetry-derived geostrophic velocities and surface veloci-ties determined from drifting buoy trajectory south of Ja-pan. J. Oceanogr., 51, 729–740.
Kamachi, M., T. Kuragano, S. Sugimoto, K. Yoshita, T. Sakurai,T. Nakano, N. Usui and F. Uboldi (2004): Short-range pre-diction experiments with operational data assimilation sys-tem for the Kuroshio south of Japan. J. Oceanogr., 60, thisissue, 269–282.
Kaneko, A., M. Arai, N. Gohda, T. Sugimoto, H. Nakajima, Z.Yuan, H. Zheng, X. Zhu and S. Yamane (1998): ADCP ob-servation of the western equatorial Pacific by a commer-cial ship. Oceanogr. Japan, 7, 357–368 (in Japanese withEnglish abstracts).
Kaneko, A., N. Gohda, M. Arai, H. Nakajima and Z. Yuan(1999): Repeat ADCP survey of the Western Pacific sur-face current by a commercial ship. WOCE News Lett., 34,31–35.
Kaneko, A., Z. Yuan, N. Gohda, M. Arai, H. Nakajima, H. Zhengand T. Sugimoto (2001): Repeat meridional survey of thewestern North Pacific subtropical gyre by a VOS ADCPduring 1997 to 1998. Geophys. Res. Lett., 28, 3429–3432.
Kobashi, F. and H. Kawamura (2002): Seasonal variation andinstability nature of the North Pacific Subtropical Counter-current and Hawaiian Lee Countercurrent. J. Geophys. Res.,107(C11), 3185, doi:10.1029/2001JC001225.
Le Traon, P. Y. and F. Ogor (1998): ERS-1/2 orbit improve-ment using TOPEX/POSEIDON: the 2cm challenge. J.Geophys. Res., 103(C4), 8045–8057.
Le Traon, P. Y., P. Gaspar, F. Bouyssel and H. Makhmara (1995):Using TOPEX/POSEIDON data to enhance ERS-1 data. J.Atmos. Ocean. Tech., 12, 161–170.
National Research Council (1997): Satellite Gravity and theGeosphere: Contributions to the Study of the Solid Earthand Its Fluid Envelopes. National Academy Press, Wash-ington, D.C., 126 pp., ISBN 0-309-05792-2.
Qiu, B. (1999): Seasonal eddy field modulation of the NorthPacific subtropical countercurrent: TOPEX/Poseidon obser-vations and theory. J. Phys. Oceanogr., 29, 2471–2486.
Qiu, B., K. A. Kelly and T. M. Joyce (1991): Mean flow andvariability in the Kuroshio Extension from Geosat altimetrydata. J. Geophys. Res., 96(C10), 18,491–18,501.
Tanaka, K. and M. Ikeda (2004): Propagation of Rossby wavesover ridges excited by interannual wind forcing in a west-ern North Pacific model. J. Oceanogr., 60, this issue, 329–340.
Teague, W. J., M. J. Carron and P. J. Hogan (1990): A compari-son between the generalized digital environmental modeland Levitus climatologies. J. Geophys. Res., 95(C5), 7167–7183.
Uchida, H. and S. Imawaki (2003): Eulerian mean surface ve-locity field derived by combining drifter and satellite al-timeter data. Geophys. Res. Lett., 30(5), 1229, doi:10.1029/2002GL016445.
