Strong seasonal temperature variation below the thermocline is detected by Argo profiling floats. The variation is associated with a long baroclinic Rossby wave. The results of this study show that seasonal signals can be captured at 1000–2000 dbar not only in basin boundary regions or deep convection regions but also in the interior ocean. A clear, westward propagation signal is observed. The phase speed is of the same order as a theoretically derived speed. Below the thermocline, the first mode of empirical orthogonal function analysis of potential density variations in the strong signal region shows a clear seasonal variation. Even in the middle layer, the amplitude of the temperature anomaly is 0.05–0.12° C at 1200 dbar, which is equivalent to the annual standard deviation from the WOA01. These results are obtained using the dense spatial and temporal distributions provided by the Argo observations.
 Oceanic temperature variation plays a key role in climate change due to the large heat capacity of water. The timescale of oceanic variation has both intra- and interannual components. Seasonal variations repeated yearly are basic components in the atmosphere–ocean climate system. In order to understand oceanic variability, it is important to understand the dominant air-sea interactions that occur on seasonal timescales.
 Generally, seasonal changes of hydrographical parameters, such as temperature and salinity, are caused by atmospheric seasonal variations at the sea surface [e.g., Gill, 1982]. Previous studies examined seasonal variations at the ocean surface and thermocline layers using two methods: observation of temperature variation based on expendable bathythermograph (XBT) data [White, 1977; Meyers, 1979] or satellite observations of sea level height variations [Chelton and Schlax, 1996; Killworth et al., 1997]. Both types of observations detected the westward propagation of long Rossby waves associated with intra- and/or interannual variations. Other than some observational studies conducted in the western boundary region or deep convection regions [e.g., Dickson et al., 1982], few studies detailed seasonal variations below the thermocline in the interior ocean. It is generally suggested that seasonal variation mainly occurs above the thermocline, except in areas of deep convection or coastal regions [e.g., Knauss, 1996]. However, this suggestion may be based on a lack of observations and insufficient temporal and spatial data, especially below the thermocline. Although such observations are difficult using either moorings or ship-based CTD observations, new technology should allow for continuous observations below the thermocline.
 Since 2000, many profiling floats capable of observations down to 2000 dbar have been deployed throughout the world ocean under the international Argo project [Argo Science Team, 1998, 2001]. The profiling floats measure temperature, salinity, and pressure data at accuracies of 0.005°C, 0.01 psu, and 5 dbar, respectively. Data are processed using real-time quality control (rQC) and delayed-mode quality control (dQC) [Wong et al., 2003; Kobayashi and Minato, 2005]. As of October 2005, more than 2000 observation floats are operating in the world ocean, realizing approximately two-thirds of the target density of one float per 3° × 3° square. This new observation system is able to monitor temperature and salinity fields for intra- and/or interannual variations below the thermocline, and also the Argo observing system continues to grow in density.
 Furthermore, the advantage of the unique float observation system is that observations are independent of weather and so provide year round cover even in difficult areas. With the Argo floats, operational mapping of temperature and salinity can now be conducted easily over an entire basin. The Argo float observation system makes it possible to clarify the characteristics of seasonal changes in the world oceans. The purpose of this research is to detect and clarify characteristics of signals below the thermocline using the Argo float data in the North Pacific.
2. Data and Methods
 We used temperature and salinity data from profiles obtained by the Argo floats, from the Triangle Trans-Ocean Buoy Network (TRITON) buoys deployed in the western tropical Pacific region, and from available conductivity, temperature, and depth (CTD) device measurements. The numbers of sampling levels on each device were about 115 (Argo float), 12 (TRITON buoy) and more than 1000 (CTD). To avoid conducting a vertical interpolation with larger error, we eliminated profiles with less than 12 sampling layers and temperature only, as well as layers with “bad” or “probably bad” flags. The rQC and/or the dQC had already been carried out for the Argo profile data. For the TRITON buoy data, we calculated monthly averaged profiles obtained from real-time quality controlled data at each buoy site [Ando et al., 2005]. The salinity profiles obtained from shipboard CTD observations were corrected using the salinity value of the sampled seawater. For all the profiles, we interpolated the available profiles using the method by Akima  at 10-dbar levels. Data from spikes or larger errors in the profiles were removed, and data were converted to standard pressure levels. To use the TRITON buoy data, we converted the data from depth to standard pressure levels.
 Monthly two-dimensional temperature and salinity mapping on pressure surfaces was performed using the optimal interpolation method (OI) [Mizuno, 1995; White, 1995] from January 2001 [Hosoda and Minato, 2003]. The analysis area of the map is the Pacific Ocean (60.5°S–60.5°N, 109.5°E–69.5°W). The horizontal resolution is 1° × 1° with vertical 25 layers at 10 to 2000 dbar; the deepest level is equivalent to the maximum depth of the Argo profiles. For monthly mapping, we assume that all data obtained in a month are observed at the same time. Spatial distributions of the decorrelation radius and error variances are based on White's  analysis method; i.e., estimations from historical observations were conducted only for temperatures in the upper 400 m but covered the entire basin. Both parameters are functions of latitude and depth, and the values at the surface become smaller than those below the surface. The decorrelation radius and error variances are considered the same for salinity and temperature maps. The parameter in a layer deeper than 400 m is given the same value as the parameter at 400 m for both the temperature and salinity maps. For the first guess of the map, we used gridded monthly mean temperature and salinity data from the World Ocean Atlas 2001 (WOA01) [Boyer et al., 2002; Stephens et al., 2002]. Here, we refer to the potential temperature distribution obtained from this map as, for example, “OI temperature.”
3. Characteristics of Seasonal Temperature Variation
Figure 1 shows the standard deviation (SD) of the OI temperature anomaly at 1200 dbar for 4 years (January 2002–December 2005) in the North Pacific Ocean. Through this period, the interpolation error of the OI map is small enough to realistically display the month-to-month temperature variation. In the southern part of 20°N, the SD is higher than in the northern part. Concerned with the distribution, there are a distinct meridional contrast of the SD in the North Pacific except in the Kuroshio Extension region, where there is deep structure of a strong current with temperature variations. The SD derived from the OI temperature is 0.05–0.12°C in the south of 20°N, and the value is equivalent to the annual SD obtained from historical observation data (about 0.08–0.15°C; WOA01).
 To investigate the propagation characteristics of the seasonal signal variation, the OI potential temperature anomaly time series at 1200 dbar along 24.5°N, 17.5°N and 10.5°N are shown in Figures 2 (top), 2 (middle), and 2 (bottom). Along 10.5°N, wavelike propagation of the temperature anomaly is captured clearly, and the signal from the eastern Pacific Ocean to western Pacific Ocean propagates without strong decay or contamination from any other signal. A large amplitude temperature anomaly appears in the east of 150°W in the source region, the temperature anomaly has a maximum in summer-autumn and a minimum in winter-spring. The zonal scale of the signal with a seasonal cycle is about 30–50°. As for along 24.5°N and 17.5°N, the signal propagation originated from eastern region are not so clearer than that along 10.5°N, the westward propagation of the seasonal signal, which is originated from the central region, are still appeared. The magnitude of the signal is weaker than that along 10.5°N, especially eastern region. This pattern of the distribution corresponds with the SD pattern (Figure 1). Further, in Figure 2, it seems that the westward propagation speed becomes slower at the higher latitude.
 Assuming that the periodic wave speed is constant and the propagation direction is westward, we estimate the propagation speed using the least squares method from the OI temperature anomaly in the North Pacific Ocean (Figure 3). Apparently, the wave speed decreases smoothly from low to high latitudes and is the same order as that of an unperturbed theoretical first mode baroclinic Rossby wave. Therefore, we compare the estimated propagation speed to the theoretical propagation speed of the first mode baroclinic Rossby wave, which is calculated from the annual averaged potential density profile obtained from the WOA01. The two propagation speeds are of the same order, whereas the speed is slower (faster) to the south (north) of 20°N than the theoretical speed. This result is nearly consistent with the tendencies found for sea level height based on satellite altimeter data analyses [Chelton and Schlax, 1996].
 Empirical orthogonal function (EOF) analysis is applied to the OI potential density profile at 10.5°N, 130.5°W (averaged at 8.5–11.5°N, 140.5–120.5°W), which is the region with the strong signal (Figure 4). Because the surface layer (upper 50 dbar) is influenced directly by the atmosphere, we remove the layer for the EOF analysis. While the variation depended strongly on the vertical gradient of potential density, we normalize the variation to the vertical gradient, which is nearly equivalent to the displacement of the potential density surface. The first mode EOF explains approximately 78% of the potential density variance and consisted of two portions with opposing phases separated at 400 dbar; the second mode explains approximately 8% of the variance. The EOFs, ranked by eigenvalues, show that the two modes contribute over 80% of the signal variances (data not shown). The maximum (minimum) value of the first mode EOF for seasonal variation corresponds with that in Figure 2 (bottom) and is displayed clearly in Figure 4a. From the eigenvector of the first mode in Figure 4b, a large amplitude appears below the thermocline and a small amplitude with the opposite sign is observed at the thermocline. Comparing with the dynamical mode structure (Figure 4c), the vertical structure of the first mode EOF below the thermocline corresponds nearly with that of the dynamical first mode. However, the small amplitude at the thermocline in the first mode EOF cannot be seen in the dynamical first mode. Instead, the depth of the small amplitude with opposite sign corresponds with the depth of the upper peak in the second dynamical mode (also the third mode). This suggests that the signal of the first mode EOF is made by combination of plural dynamical modes.
4. Summary and Discussion
 The deploying of numerous profiling floats by the Argo project has allowed hydrographical maps down to 2000 dbar to be produced. In this paper, we capture a seasonal signal below the thermocline based on Argo data and investigate the characteristics of the seasonal variation. The characters of the displayed signal propagation are nearly the same as those of the sea level height variation observed by satellite observations and for thermocline variations obtained by XBT/XCTD observations in previous studies except for seasonal variations. The Argo float system can capture these characteristics because the data are from good, quality controlled hydrographical profiles, the Argo observation system covered in the entire ocean, the cast is conducted down to 2000 dbar for each profile, and the observations are carried out every 10 days, a period shorter than a seasonal cycle.
 The large SD of the OI temperature anomaly, which the magnitude is equivalent to that from the WOA01, is observed widely in the lower latitude. Further, the westward propagation of the signal is captured clearly in the larger seasonal signal, which is difficult to detect by surface observations. In particular, along 10°N, a strong seasonal change appears in the middle layer in the eastern Pacific Ocean. This region coincides with a region of strong Ekman convergence associated with the Intertropical Convergence Zone (ITCZ) in the eastern Pacific Ocean [Meyers, 1979]. This suggests that once a strong signal is input in the ocean, the signal undergoes little decay while propagating. Also, we confirm the existence of inter-annual or local modifications of the seasonal variation, this issue will be discussed in the future. Anyhow, our analysis displays the spatial structure of propagation of a planetary Rossby wave using a powerful device: the Argo profiling float.
 Comparison of propagation speeds between the observational wave and theoretical baroclinic Rossby wave shows the same tendency as that associated with latitudinal variation of inter-annual sea level height variation obtained by satellites. The first mode of EOFs clearly shows the seasonal variation of the signal; additionally, it also indicate that the variations consist of pycnocline and middle layers in the vertical structure. This result is consistent with the dynamical first mode below the thermocline, but inconsistent at the thermocline, where it seems that the signal variation consists of plural dynamical modes. Also, some modification may occur by a thermal effect and horizontal advection in a surface and/or subsurface layer.
 The SD of the OI temperature variation at 1200 dbar is 0.05–0.12°C, which is equivalent to the annual SD obtained from the WOA01. Further, from raw float data, the variation magnitude of the signal amplitude is about a few times as much as that of the OI temperature variation (However, below the thermocline, water mass change associated with the seasonal variation does not exist within the accuracy limits of the float observations; data not shown). This result indicates that variation below the thermocline cannot be ignored even in the interior ocean, which has been considered an area less affected by strong variation. This demonstration of the existence of strong seasonal signals below the thermocline may call into question the interpretation of earlier results showing apparent long-term variations. For example, if interannual temperature variation is calculated using different monthly data, the variation results may be contaminated by both intra-annual and interannual variations. Therefore, analyses of interannual variation or heat transport change perform using CTD observations, for example, should be based on the denser spatial and temporal observations.
 We thank all the members of the Japan Agency for Marine–Earth Science and Technology (JAMSTEC) Argo Group for producing the high-quality data. We especially thank the technical staff, Mr. T. Ohira and Mr. H. Nakajima for helpful assistance with software usage. We are also grateful to Drs. I. Kaneko, T. Kobayashi, T. Suga, M. Nonaka, and G. Mizuta for useful discussions. The Argo profiling data are corrected and are available via the following Web site: http://www.jcommops.org. The OI map can be viewed and is routinely updated at the JAMSTEC Argo Web page: http://www.jamstec.go.jp/ARGO/J_ARGOe.html.