Drift Characteristics of a Moored Conductivity Temperature Depth Sensor and Correction of Salinity Data

Transcription

1 282 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 22 Drift Characteristics of a Moored Conductivity Teperature Depth Sensor and Correction of Salinity Data KENTARO ANDO Japan Agency for Marine-Earth Science and Technology, Kanagawa, Japan TAKEO MATSUMOTO AND TETSUYA NAGAHAMA Marine Works Japan Co., Kanagawa, Japan IWAO UEKI, YASUSHI TAKATSUKI, * AND YOSHIFUMI KURODA Japan Agency for Marine-Earth Science and Technology, Kanagawa, Japan (Manuscript received 10 October 2002, in final for 11 June 2004) ABSTRACT The teperature and conductivity drift (tie change of the characteristics) of oored SBE37IM conductivity and teperature (CT) sensors was investigated by pre- and postdeployent calibration of the Triangle TransOcean Buoy Network (TRITON). This buoy network coprises the western portion of the basinwide (Tropical Atosphere Ocean) TAO/TRITON buoy array, which onitors phenoena such as El Niño and contributes to forecasting cliate change. Over the tie of deployent the drift of the teperature sensors was very sall, within 3 K of the postdeployent calibration data. The drift of the conductivity sensors was ore significant. After 1 yr of ooring, conductivity drift observed in the shallowest layer ( ) was positive and S 1 [equivalent to (PSS-78) at 30 C and 6S 1 ; here, 1Sis1 1 ]at6s 1 on average. Drift observed in the therocline layer ( ) was also positive and S 1 [0.034 (PSS-78)] at 6S 1 on average. Conversely, the drift of conductivity in the deepest layer ( ) was S 1 with a standard deviation of S 1 [ (PSS-78)]. Assuing a linear trend of conductivity drift with tie, the authors attepted to correct the conductivity data using the postdeployent calibration data. The corrected data for about 80% of the sensors exhibited saller differences than the uncorrected data when copared with the in situ conductivity teperature depth (CTD) data. However, the corrected salinity data becae worse than the uncorrected data for about 20% of the sensors. The reasons for these errors are also discussed in this paper. 1. Introduction * Current affiliation: Japan Meteorological Agency, Tokyo, Japan. Corresponding author address: Dr. Kentaro Anso, Japan Agency for Marine-Earth Sciences and Technology, 2-15 Natushia Yokosuka, Kanagawa , Japan. E-ail: andouk@jastec.go.jp Salinity easureents are iportant in the tropical oceans to understand general ocean circulation (Tsuchiya 1968), to deterine accurate dynaic height variability (Maes 1998), and to understand the barrier layer (Lukas and Lindstro 1991). The barrier layer is hypothesized near the surface layer preventing entrainent fro ixing with underlying therocline waters by salinity stratification. This hypothesis, in turn, suggests that salinity plays a role in controlling the therodynaic processes in the upper layer via changes in density stratification. Salinity variability near the surface is also iportant in characterizing hydrological circulation in the tropical Pacific. More than 2000 of rainfall a year over the western Pacific is indicated by the distribution of low salinity surface water (Fig. 1). During the El Niño Southern Oscillation (ENSO) condition the war and fresh (low salinity) surface water is displaced eastward (Picaut et al. 1996; Delcroix and Picaut 1998) associated with the eastward oveent of a high precipitation region (Ando and McPhaden 1997). A French group (Delcroix et al. 2000) began sea surface salinity (SSS) easureents in the 1980s by using Voluntary Observing Ships (VOSs) over three or four regions in the tropical Pacific. The Tropical Ocean Global Atosphere Coupled Ocean Atosphere Response Experient (TOGA COARE) ipleented tie series easureents of salinity (McPhaden et al Aerican Meteorological Society

2 MARCH 2005 A N D O E T A L. 283 FIG. 1. Location of the TRITON buoys deployed in the western tropical Pacific and the existing TAO array (McPhaden et al. 1998) with the SSS cliatology fro the World Ocean Database of 1998 (Boyer et al. 1998). 1990; Freitag et al. 1999) by installing teperature and conductivity sensors (odel SBE16 electrode-type conductivity sensors, anufactured by Sea-Bird Electronics, Inc.) along the ooring lines. Freitag et al. (1999) reported that the characteristics of the conductivity sensor drifted toward being positive in the shallower layer due to the scouring effect by strong currents. They corrected their COARE salinity data priarily using the linear trend calculated fro the postdeployent calibration data after recovery. They also reported that reoving the linear trend is insufficient for correcting all salinity data because of possible episodic changes in sensor characteristics in soe cases. Their studies helped to establish better quality salinity tie series easureents in the relatively shallower layer (0 200 ) by a oored buoy. Sixteen Triangle TransOcean Buoy Network (TRITON) project buoys are planned to be deployed in the western tropical Pacific west of 156 E (Fig. 1); these oorings in conjunction with the Tropical Atosphere Ocean (TAO) array (McPhaden et al. 1998) constitute a portion of the ENSO-observing syste. Twelve conductivity and teperature sensors (Type SBE37IM CT sensors) were installed on each buoy at depths fro 1.5 to 750 (Fig. 2). The buoys are designed to easure high quality salinity tie series data fro the surface down to 750 in the western tropical Pacific region (west of 156 E) to quantify war and freshwater pool variability and its relation to global cliate changes (Kuroda and Aitani 2001). However, we do not have sufficient knowledge regarding the tie drift of teperature and conductivity sensors. We will deterine how uch drift to expect in a yearlong ooring, investigate techniques of drift correction, and deonstrate iproveent of data quality. The drift of the conductivity and teperature sensors (type SBE37IM) used on the yearlong oorings will be investigated with a particular focus on the tie dependency (0 12 onths) and depth dependency (0 750 ) of the conductivity sensor drift. We provide the teperature and conductivity drift calculated fro the pre- and postdeployent calibration data in section 2 of this study. The Japan Agency of Marine-Earth Science and Technology (JAMSTEC) teperature and conductivity calibration syste is also described in section 2, and our calibration skills are evaluated. An in situ coparison with conductivity teperature depth (CTD) data is described in section 3 to confir the drift of conductivity obtained fro the calibration data. Finally, a correction is applied to the conductivity data in section 4, assuing a linear trend. The ultiate quality of the salinity data estiated fro the linear correction is also discussed in this paper. 2. Drift of the teperature and conductivity sensors a. Laboratory calibration at JAMSTEC The teperature and conductivity calibration systes in JAMSTEC were ade by Sea-Bird Electronics, Inc., and were calibrated by the anufacturer (see Matsuoto et al for details, and/or SBE Web site at All conductivity teperature (CT) sensors were recalibrated at JAMSTEC after delivery fro the anufacturer to confir that the differences between the anufacturer s calibration and ours were within 5 K in teperature and S 1 [( 1 S 1 (here, 1Sis1 1 )] in conductivity, and the sensors were in good condition. A further objective of this experient is to confir JAMSTEC s calibration skills. JAMSTEC s proficiency at calibration was evaluated as the difference between the anufacturer s calibration and ours, assuing that ost sensors will not drift during delivery to JAMSTEC. We perfored our calibrations for ost sensors within 3 onths of the original calibration. One hundred and eleven SBE37IM sensors fro S/N 0489 to S/N 0663 were used in the experient in The calibration differences are calculated as residuals by subtracting the true value of our reference sensors fro the calculated value using coefficients derived fro the original calibration and raw output data fro the sensor in the JAMSTEC cali-

3 284 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 22 brations. The true teperature values in a calibration bath are easured by using the reference sensors (SBE3). Our teperature reference sensors have been calibrated yearly at the anufacturer. The historical calibrations of the teperature reference sensors indicated a very sall tie drift, within 1 K yr 1, which is considerably less than the accuracy of the JAMSTEC calibration. Therefore, we decided not to correct the teperature data fro the reference sensors. The true conductivity values were estiated fro the conductivity easured by the reference sensor (SBE4). Reference sensor conductivity values were corrected for drift by ultiplying the by the ratio of the reference conductivity to the true conductivity derived fro a laboratory salinoeter (Autosal Guildline 8400B) easureent of water sapled fro the bath at 24 C. Therodynaic differences between the saltwater used in the bath and natural seawater are negligible because the calibration bath and the laboratory salinoeter are both operating at 24 C. The results of the coparison between JAMSTEC s calibrations and those of the anufacturer are shown in Fig. 3 as the enseble ean and standard deviation of differences between JAMSTEC s calibration and the anufacturer s calibration for each teperature and conductivity sensor. These coparisons were all perfored at 6S 1 for conductivity and at 30 C for teperature since the TRITON buoys are oored in a tropical region where the surface teperature is usually higher than 27 C, and half of the sensors are typically installed in the upper 150. The differences of teperature calibration at 30 were within 1 K on aver- FIG. 2. Configuration of a TRITON buoy. FIG. 3. Coparison of anufacturer s calibration and JAM- STEC s calibration for (a) the teperature sensor and (b) the conductivity sensor. Horizontal axis indicates the difference of teperature calibration (in K) for (a), and that of conductivity calibration (in S 1 ) for (b). Frequency at the ost right (left) indicate su of frequency ore (less) than axiu (iniu) value on the horizontal axis.

4 MARCH 2005 A N D O E T A L. 285 age with 2 K of standard deviation, and the differences of conductivity calibration at 6S 1 were 0.13 S 1 on average [equivalent to (PSS-78) at 6S 1 and 30 C] with 0.85 S 1 of standard deviation [equivalent to (PSS-78); Fig. 3]. These are sufficiently sall to easure salinity in the tropical upper ocean. The errors of our CT sensor calibrations were evaluated as the standard deviation of differences (i.e., 2 K in teperature and 0.85 S 1 in conductivity). These values are also thought to be the rootean squares of the anufacturer s errors and ours. If the errors fro the JAMSTEC calibrations are assued to be the sae value as the anufacturer s, that is, if the JAMSTEC calibration skill is the sae as the anufacturer s, the error of teperature calibration is estiated to be 1.4 K and that of conductivity calibration is 0.6S 1. These values are also equivalent to those reported by the anufacturer. We perfored two kinds of calibrations in the TRITON project, predeployent and postdeployent calibration. Predeployent calibrations are perfored at least twice for each sensor to confir that both calibrations agree within ranges of 2 K in teperature and1s 1 in conductivity. Sensors are postdeployent calibrated in an as is state after recovery. The bodies of the pressure cage of the sensors are usually washed; however, the conductivity cell is never washed until after postdeployent calibration is perfored. Postdeployent calibrations are usually perfored at least twice to confir that the calibration results agree. We send the sensor back to the anufacturer if the teperature drift is greater than 5 K copared with the predeployent results. We replatinize the conductivity cell if the conductivity drift is greater than 2 S 1, as instructed by Sea-Bird Electronics, Inc. b. Tie drift of teperature and conductivity estiated fro laboratory calibration The teperature and conductivity drifts were calculated using the pre- and postdeployent calibration data (hereafter referred to as the laboratory calibration). The data used are fro calibration of the sensors deployed in 1998, 1999, and 2000 and recovered in 1998, 1999, 2000, and 2001 (see Table 1). The calibration data were configured into a dataset of teperature and conductivity drifts at 30, 20, and 10 C, and at 6, 5, and 4S 1, respectively, fro the calibration curves by using original coefficients provided by the anufacturer. The drifts are defined in this analysis as the differences between the true value and the easured output during postdeployent calibration using the original coefficients obtained by the predeployent calibration. All buoys were recovered after only 3 onths of deployent in 1998, the first year that the TRITON buoys were deployed. TRITON buoys were next deployed along 156 and 147 E in March Two buoys were recovered after 4 onths, three buoys were recovered after 8 onths, and four buoys were recovered after 12 onths during Figure 4 illustrates the drift of the teperature sen- TABLE 1. Moored periods of TRITON buoys and CT sensors. The calibration data fro those sensors are used in this study. TRITON buoy sites Buoy ID Moored period Serial nuber of CT sensors installed on each buoy 8 N, 156 E 1001 Mar 1998 Jun Feb 1999 Mar Mar 2000 Feb N, 156 E 2001 Mar 1998 Jun Feb 1999 Mar Mar 2000 Feb N, 156 E 3001 Mar 1998 Jun Mar 1999 Nov Nov 1999 Mar E 4001 Mar 1998 Jun Mar 1999 Mar Nov 1999 Mar Mar 2000 Mar S, 156 E 5001 Mar 1999 Mar Mar 2000 Mar S, 156 E 6001 Mar 1999 Mar Mar 2000 Mar N, 147 E 7001 Feb 1999 Jun N, 147 E 8001 Feb 1999 Jun E 9001 Feb 1999 Oct Oct 1999 Sep Oct 2000 Dec N, 138 E Oct 1999 Sep E Oct 1999 Sep Oct 2000 Oct

5 286 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 22 FIG. 4. Drift of the teperature sensors used on the TRITON buoys. The horizontal axis indicates the oored periods in days in reference to the deployent date. The square with the solid bar indicates the average and standard deviation of drift of the teperature sensors oored in the shallowest layer ( ), the triangle with the dashed line indicates the sae except for in the therocline layer ( ), and the asterisk with the dotted line indicates the sae except for in the deepest layer ( ). sors. The tie axis is the oored period in days referenced to the deployent date. The teperature drift is very sall (3 K), and does not appear to be dependent on tie. The postdeployent calibration data indicated that 4 of the 160 sensors drifted ore than 5 K. Except for these sensors with notable drift, the average and standard deviations of the teperature drift after ore than 1 yr were 0.1 and 0.5 K, respectively. These values are of the sae agnitude as the difference in calibration fro that of the anufacturer, which indicates that the teperature drift is not significant. We can conclude that the teperature sensor is very stable over tie and is not dependent on pressure. The drift of the conductivity sensors recovered in 1999 and 2000 was calculated by classification into three regies of installed depth. The drift of conductivity was large and positive in the shallower layers (Fig. 5 and Table 2). The drift of conductivity in the shallowest layer ( ) was 10 S 1 [equivalent to (PSS-78) at 30 C and6s 1 ]at6s 1 on average after 1 yr of ooring, and that in the therocline layer ( ) was 5.3 S 1 [0.034 (PSS-78)] at 6 S 1 on average. The standard deviations also increased with tie in the surface and therocline layers, suggesting a large variance of drift for each sensor. The tie evolution of the drift of the conductivity sensors appeared to increase after 4 onths (120 days) of ooring in the surface (0 100 ) and therocline ( ) layers. The drift of conductivity appeared to increase especially rapidly in the surface layer after 8 onths ( 250 days). Conversely, the drift of conductivity in the deepest layer ( ) was very sall (0.02 S 1 with a standard deviation of 1 S 1 ) and had alost the sae aplitude as the results of the evaluation of our conductivity calibration. This indicates that the drift of conductivity in the deepest layer was not significant, and that the conductivity sensors were very stable. This iplies that the drift of the conductivity sensors in the shallower layers was not caused by the sensor itself but by its environent, such as biofouling and scouring effects relative to current speed. The drift in the group of sensors oored for 90 days was calculated fro sensors oored in A pinhole was found in soe teperature sensors after recovery. These sensors all had earlier serial nubers (before 250), and after the recovery, all the teperature sensors were replaced by the anufacturer. We assue that the negative teperature drift around 90 days does not reflect the real drift during ooring but rather is due to hardware-oriented drift. The drift of the conductivity sensors recovered in 2001 was also analyzed. The conductivity sensors oored in the layer between 125 and 250 exhibited a positive drift of 5.5 S 1 on average with a standard deviation of 3.9 S 1, which is alost the sae aount observed in the oored sensors. The standard deviation in the shallowest layer ( ) was larger (8.7 S 1 ) than its average (2.8 S 1 ), and a ratio of conductivity sensors revealed negative drift in a year. This ay be caused by the circustances differing fro the oored sensors.

6 MARCH 2005 A N D O E T A L. 287 FIG. 5. Drift of the conductivity sensors used on the TRITON buoys. The horizontal axis indicates the oored periods in days in reference to the deployent date. The square with the solid bar indicates the average and standard deviation of drift of conductivity sensors oored in the shallowest layer ( ), the triangle with the dashed line indicates the sae except for in the therocline layer ( ), and the asterisk with the dotted line indicates the sae except for in the deepest layer ( ). In fact, due to the interannual variability, the sea surface teperature in 2001 in the oored region is warer than that in Siilarly, surface current speed and biological productivity ay differ between the periods of and Such investigations will be necessary when enough calibration data will have been accuulated for statistical analysis. The conductivity sensors were very stable in the deepest layer ( ), as previously found in the oored sensors. 3. In situ coparison of salinity data The analysis of the laboratory calibration data indicated that the teperature sensors were stable in tie and had no depth dependency. The conductivity sensors tended to drift positively in the surface and therocline layers; however, we do not know if that drift, particularly the conductivity drift in the surface layer, was actually present while oored. Electrode conductivity sensors are very sensitive (N. Larson 2001, personal counication) to huan handling and to the conditions on land and on the ship before deployent and after recovery. Teperature sensors are very stable in tie (Fig. 4). Therefore, salinity data fro the TRITON buoys in this coparison were picked at the sae teperature as the shipboard CTD teperature, based on the assuption that the sae teperature salinity (T S) relations were easured by both instruents. Most of the CTD observations were perfored within 2 n i of TRITON buoy locations after deployent, during visits, and before recovery. The salinity data fro the shipboard TABLE 2. Conductivity drift of CT sensors (SBE37IM) installed on TRITON buoys calculated fro the pre- and postcalibration data. Theses data are also plotted in Fig. 5. Unit in this table is S 1,and1S 1 corresponds to (PSS-78) at 6S 1 and at 30 C. Boldface values are calculated with few nubers of data. Surface layer (0 100 ) Subsurface layer ( ) Deep layer ( ) Avg period oored in days Avg Std dev No. of data Avg Std dev No. of data Avg Std dev No. of data

7 288 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 22 CTD was checked with sapled water and corrected in cases of a substantial salinity difference [ore than (PSS-78)] fro the sapled water easureent. Table 3 shows the salinity offsets by bottled saple and the salinity fro the shipboard CTD. The offsets in all cruises were within (PSS-78), and the salinity fro the shipboard CTD was used as a reference for the TRITON salinity. The errors in the coparison experients were estiated fro the data when the buoys had just been deployed, since the TRITON sensors were thought to have only a very sall drift. The errors were evaluated as the average and standard deviations of the differences between the TRITON salinity and the in situ shipboard CTD, which were (PSS-78) and (PSS-78). A standard deviation of (PSS-78) is therefore the liit (error) of this coparison experient. The result of the in situ salinity coparison with the shipboard CTD syste (SBE9/11plus) is shown in Fig. 6. Large positive drifts occurred in the surface and therocline layers (0 100 and ) and were clearly evident after 4-onth ooring periods. The order of the drifts and variations were coparable to those found in the conductivity data fro the laboratory calibrations. However, the deviations in the in situ coparison were greater and the changes in the tie drift also exhibited considerable variability, probably due to the substantial horizontal T S variability in the surface layer and therocline oveents. The drift of the salinity in the deepest layer below 250 cannot be clearly observed in Fig. 6, which is consistent with the result of laboratory calibrations (Fig. 5). Both analyses indicate that the conductivity drift is quite sall and insignificant in the deepest layer. 4. Correction of data The drift estiation fro the laboratory calibration data and the results fro the in situ coparison revealed a positive increase with tie. The drift appeared to exhibit weak nonlinear behavior over tie; however, assuing a linear drift of conductivity over tie was the only way to correct the conductivity data. A axiu likelihood ethod was used to select the ost suitable equation of conductivity drift aong the linear, quadratic, and cubic fits for the postrecovery calibration data referenced to the predeployent calibration to estiate the best correction as a function of conductivity (3 6 S 1 ). Once the ost suitable curve was established, it was applied to the conductivity tie series in a teporally linear anner. The drift was corrected as a linear tie function by hours starting fro the deployed tie. In addition to that, while oored, other sensor drift could occur in various situations, for exaple, in the event of rough treatent of sensors or if the sensors are stored in unclean conditions. However, our sensors were treated properly as instructed by the anufacturer. The fact that sensors in deepest layers did not drift (Fig. 4) verified that our treatent had been proper. Our correction was applied to the sensors with drift exceeding 0.5 S 1 ; drift below 0.5 S 1 is not significant in our calibration. We then repeated the coparison with the in situ CTD data as in section 3. An exaple of the coparison of data before correction and after correction with the in situ data at the 5 S,156 E buoy on 6 Noveber 1999 (idway between deployent and recovery) is shown in Fig. 7 as a T S diagra. Large positive drifts of salinity were observed in the upper 200 in the uncorrected data (indicated by an X) due to a large positive drift of conductivity. The drifts were greatly decreased after correction [( ) ark]. The salinity actually decreased toward the true T S relation at a depth of 200 ; however, the corrected salinity did not coincide perfectly with the true salinity. One likely reason for this is that the conductivity sensor ay not drift linearly in tie. The linearity in tie is probably dependent on each sensor, and this particular sensor did not show a linear drift in this case. A detailed discussion will be provided in the next section. A second possible reason is natural variability. The salinity coparison was ade with a CTD cast within 2 i fro the buoy at alost the sae tie and at the sae teperature. However, the error (difference) between the CTD and the TRITON CT caused by natural variability, particularly on a shorter tie and space scale such as necessitated by tide and internal waves, cannot be reoved. Figure 8 shows the frequency distributions of differences in the TRITON salinity inus the in situ CTD TABLE 3. Differences between sapled salinity by bottle and easured salinity by CTD at the sae site and sae depth. The in situ coparisons with shipboard CTD (SBE9/11 plus syste) were carried out during the TRITON buoy cruises listed below. Nae of ship (Cruise ID) Mirai (MR99K01) Feb Mar 1999 Mirai (MR99K06) Oct Nov 1999 Mirai (MR00K02) Feb Mar 2000 Kaiyo (KY9906) Nov 1999 Dec 1999 Kaiyo (KY0007) Oct 2000 Nov 2000 Mirai (MR01K01) Feb Mar 2001 Period of cruise Meridional line of TRITON 156, 147 E 156, 147 E 156 E 147, 138 E 156, 147, 138 E 156, 147 E buoy served during cruise Avg of differences (PSS-78) Std dev of differences (PSS-78)

8 MARCH 2005 A N D O E T A L. 289 FIG. 6. Results of an in situ coparison of salinity with that easured by the shipboard CTD (SBE9/11 plus) syste during The coparison was done in salinity unit. For rough coparison to Fig. 5, equivalent conductivity was estiated by using the conversion ratio at 30 C and6s 1 is indicated by the right vertical axis. salinity at the sae teperature for all data deployed in We used the data during visits and recovery in this coparison. There were two peaks, around 0 and around 0.03 (PSS-78), in the coparison with the uncorrected TRITON salinity (Fig. 8a), which corresponded to the nondrifting sensors installed in the deepest layer and around the positively drifting sensors in the shallower layers, respectively. A sall nuber of large drift sensors, ore than 0.08 (PSS-78) and less than 0.08 (PSS-78), were also found. The two peaks found in the uncorrected data coparison disappeared and the positive drifts greatly decreased after correction (Fig. 8b). The nuber of peaks around 0 increased and the distribution becae sharper than that of the uncorrected data, indicating that the correction worked properly. However, it appeared that the distribution also shifted to the negative side. One reason for this is probably due to the assuption of a linear drift of the conductivity sensor over tie. Another curious fact is that a negative tail was produced around 0.05 (PSS-78). We noted in this case that the applied correction worked in the wrong direction for this sensor group. This correction procedure indicates the possibility of an unrealistic salinity correc- FIG. 7. Exaple of a coparison on a T S diagra of uncorrected salinity and corrected salinity at 5 S, 156 E on 6 Nov 1999.

9 290 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 22 78)]. The corrected data should specify the sae salinity as the in situ CTD salinity during recovery if the integrity of the sensor is aintained after recovery until postdeployent calibration, even if the sensors indicated a large drift. Therefore, the postdeployent calibration data cannot be used for this procedure if the corrected salinity does not indicate the sae salinity as the in situ value. In this case, the salinity data fro the predeployent calibration coefficients should be adopted. The liit of correction in this case [0.04 (PSS- 78)] was deterined to be 2 [1 equals (PSS-78), which is the liitation of the in situ coparison, as described in section 3, plus 0.01 (PSS-78), which is the liitation of this correction procedure]. The salinity data quality was greatly iproved after applying this concept, and the ean and standard deviations becae (PSS-78) and (PSS-78; Fig. 8c). 5. Suary and discussion FIG. 8. (a) Frequency distribution of the differences between the uncorrected salinity data and the in situ CTD salinity. (b) Sae as (a) except for corrected salinity applied to all sensors. (c) Sae as (a) except for corrected salinity applied to selected sensors, for which the salinity difference fro the in situ CTD did not exceed 0.04 (PSS-78) at recovery. tion for about 20% of the sensors if it is routinely applied to all sensors. Thus, we used the coparison data with the in situ CTD during recovery in this procedure to avoid a iscorrection of salinity data and to deterine whether a correction is needed. The correction procedure is not applied if the difference between the corrected salinity and the in situ CTD salinity during recovery exceeds a certain liit [in this case, 0.04 (PSS- The drift of conductivity teperature (CT) ooring type (SBE37IM) sensors was investigated by laboratory calibration data easured before deployent and after recovery of TRITON buoys and through coparison with the in situ CTD data obtained near the buoys. The ethod of correction was investigated as well, and the corrected data was evaluated with the in situ CTD data. The results are suarized as follows. 1) The drift of the teperature sensors was very sall, within 3 K fro the predeployent calibration data. 2) The drift of conductivity sensors was classified by the installed depth into three regies: the shallowest layer ( ), the therocline layer ( ), and the deepest layer ( ). The drift of the conductivity sensors installed in the upper two layers was positive and increased with tie. The drift of the conductivity sensors in the shallowest layer ( ) was 10 S 1 [equivalent to (PSS-78) at 30 C and6s 1 ]at6s 1 on average after 1 yr of ooring, and that in the therocline layer ( ) was 5.3 S 1 [0.034 (PSS-78)] at 6S 1 on average. Conversely, the conductivity drift was very sall (0.02 S 1 with a standard deviation of 1 S 1 ) in the deepest layer ( ) and the conductivity sensors were very stable. 3) An in situ coparison of the TRITON salinity data with the shipboard CTD syste (SBE9/11 plus) confired a positive conductivity drift with tie. The positive drift of conductivity becae larger over tie. We attepted to correct the conductivity data fro postdeployent calibration data, assuing a linear trend of conductivity with tie. The corrected salinity data for about 80% of the sensors exhibited saller

10 MARCH 2005 A N D O E T A L. 291 differences than the uncorrected salinity data when copared with the in situ CTD data, which confirs that our correction ethod was properly applied. Freitag et al. (1999) reported the positive drifts of conductivity sensors in the surface and therocline layers fro their analysis of calibration of conductivity sensors (odel SBE16, the sae electrode type of conductivity sensor as SBE37IM), which were oored during the TOGA COARE. Their results also indicated the sae agnitude of drift in conductivity. Results both fro this study and Freitag et al. (1999) indicate siilar instruental perforance. The results of very sall drift in the deepest layer below 250 iplies that an electrode conductivity sensor in a layer deeper than 250 will enable ore accuracy in autoatic long-ter salinity easureents, such as the profiling float used in the Argo project (Roeich and Owens 2000). An analysis of the conductivity sensors developed for the Argo float (Riser and Swift 2002, anuscript subitted to J. Atos. Oceanic Technol.) indicated that the conductivity sensors (type SBE41, the sae sensor type as SBE37IM) revealed a very sall drift of conductivity even after the sensors (floats) had been working for 2 or 3 yr. The salinity data fro SBE41 in Kobayashi et al. (2001) also deonstrated very good agreeent with the salinity fro the cliatological database (Hyrdobase; McDonald et al. 2001), even after several onths. Neither the conductivity sensors on Argo floats, which usually drift at around 2000 for ost of the operating period, nor the conductivity sensors of TRITON installed at a greater depth (below 250 ) will experience bad affects fro environental factors such as biological activity or strong currents, which would change the characteristics of electrode conductivity sensors. Thus, the sensor can easure salinity with greater accuracy under those two conditions. Acknowledgents. We are particularly grateful to Ken Lawson, Norge Larson, and David Murphy at Sea- Bird Electronics, Inc., for their helpful coents and discussions. The quality of TRITON buoy salinity data would not be iproved without the support of Sea-Bird Electronics. We also thank the operating personnel of the TRITON buoy project at the Japan Agency for Marine-Earth Science and Technology and at Marine Works Japan Co., as well as the captains and crews of R/V Mirai, R/V Kaiyo, and R/V Kairei. Their skillful work on deck during preparation of the buoys resulted in recovery of undaaged data fro the CT sensors. Discussions with Michio Aoyaa in the Meteorological Research Institute of the Japan Meteorological Agency and with Takeshi Kawano in the Ocean Observation and Research Departent in JAMSTEC were beneficial to this study. We also thank two anonyous reviewers whose helpful coents have iproved the presentation of the aterial. Finally, we wish to acknowledge use of the Ferret progra for the analyses and graphics in this paper. Ferret is a product of NOAA s Pacific Marine Environental Laboratory (ore inforation is available online at ferret.noaa.gov). REFERENCES Ando, K., and M. McPhaden, 1997: Variability of surface layer hydrography in the tropical Pacific Ocean. J. Geophys. Res., 102, Boyer, T. P., S. Levitus, J. Antonov, M. Conkright, T. O Brien, and C. Stephens, 1998: Salinity of the Pacific Ocean. Vol. 5, World Ocean Atlas, NOAA Atlas NESDIS 31, 166 pp. Delcroix, T., and J. Picaut, 1998: Zonal displaceent of the western equatorial Pacific Fresh Pool. J. Geophys. Res., 103, , C. Henin, F. Masis, and D. Varillon, 2000: Three decades of in-situ sea surface salinity easureents in the tropical Pacific Ocean. Institute de recherché pour le developpeent, CD-ROM. Freitag, H. P., M. E. McCarty, C. Nosse, R. Lukas, M. J. McPhaden, and M. F. Cronin, 1999: COARE SEACAT DATA: Calibrations and quality control procedures. NOAA Tech. Meo. ERL PMEL-115, 89 pp. Kobayashi, T., and Coauthors, 2001: Correcting ethod of Argo data based on HydroBase (I) Introduction of Potential Conductivity (in Japanese). Rep. Japan Mar. Sci. Technol. Center, 44, Kuroda, Y., and Y. Aitani, 2001: TRITON: New ocean and atospheric observing buoy for onitoring ENSO (in Japanese with English abstract). Ui no Kenkyu, 10, Lukas, R., and E. Lindstro, 1991: The ixed layer of the western equatorial Pacific Ocean. J. Geophys. Res., 96, Maes, C., 1998: Estiating the influence of salinity on sea level anoaly in the ocean. Geophys. Res. Lett., 25, Matsuoto, T., T. Nagahaa, K. Ando, I. Ueki, Y. Kuroda, and Y. Takatsuki, 2001: The tie drift of teperature and conductivity sensors of TRITON buoy and the correction of conductivity data (in Japanese). Rep. Japan Mar. Sci. Technol. Center, 44, Mcdonald, A. M., T. Suga, and R. G. Curry, 2001: An isopycnally averaged North Pacific cliatology. J. Atos. Oceanic Technol., 18, McPhaden, M. J., H. P. Freitag, and A. J. Shepherd, 1990: Moored salinity tie series easureent at 0, 140 W. J. Atos. Oceanic Technol., 7, , and Coauthors, 1998: The Tropical Ocean-Global Atosphere observing syste: A decade of progress. J. Geophys. Res., 103, Picaut, J., M. Ioualalen, C. Menkes, T. Delcroix, and M. J. McPhaden, 1996: Mechanis of the zonal displaceent of the Pacific War Pool: Iplications for ENSO. Science, 274, Roeich, D., and W. B. Owens, 2000: The Argo project: Global ocean observations for understanding and prediction of cliate variability. Oceanography, 13, Tsuchiya, M., 1968: Upper Waters of the Intertropical Pacific Ocean. Johns Hopkins Oceanographic Series, No. 4, 50 pp.