Results from Estimating the Circulation and Climate of the Ocean (ECCO)–Scripps Institution of Oceanography (SIO) global ocean state estimate, available over the 11-yr period 1992 through 2002, are compared with independent observations available at the Hawaii Ocean time series station ALOHA. The comparison shows that at this position, the estimated temporal variability has some skill in simulating observed ocean variability and that the quality of future syntheses could benefit from additional information available from the Argo network and from the time series observations themselves. On a decadal time scale, the influence radius of the station ALOHA T–S time series covers large parts of the tropical and subtropical Pacific Ocean and reaches even into the Indian Ocean through the Indonesian Throughflow.
Estimated changes in sea surface height (SSH) result largely from thermosteric changes; however, nonsteric (barotropic) variations on the order of 1–2 cm also contribute to SSH changes at station ALOHA. Moreover, changes of similar magnitude can be caused by changes in the salinity field because of a quasi-biennial oscillation in the horizontal flow structure and heaving of the mean salinity structure on seasonal and interannual time scales.
The adjoint modeling framework confirms westward-propagating Rossby waves (due to wind forcing) and subduction of water-mass anomalies (due to surface buoyancy forcing) as the primary mechanisms leading to observed changes of T–S structures at station ALOHA. Specifically, the analysis identifies surface freshwater fluxes along the wintertime outcrop of intermediate waters as a primary cause for salinity changes at station ALOHA and wind stress forcing east of the station position as another forcing mechanism of salinity variations around the Hawaiian Archipelago.
Ocean time series stations are among the few places in the world where observations of variables such as velocity, temperature, and salinity exist over many years. Those datasets are extremely valuable for describing changes in the ocean near the surface and especially at depth, observations that otherwise are difficult to obtain. Causes for those changes are manifold and include processes affecting water-mass formation, usually occurring in remote regions. As an example, Sturges and Hong (2000) described changes in sea level observed at the Bermuda Atlantic Time Series Station (BATS) in the western North Atlantic in terms of westward-propagating first-mode Rossby waves caused by wind changes in the eastern part of the basin off the coast of Africa.
The primary objective of the Hawaii Ocean time series (HOT) station ALOHA (Karl and Lukas 1996), located 100 km north of the island of Oahu, is to obtain long time series of physical and biogeochemical observations in the North Pacific subtropical gyre in support of the U.S. Global Change Research Program. During each of the monthly cruises to station ALOHA, CTD profiles are being measured down to a 1000-m depth at 3-hourly intervals over 36 h. Karl et al. (2001), Lukas (2001), Lukas et al. (2001), Finnigan et al. (2002), and Dore et al. (2003) discuss some of the major findings from research at station ALOHA since it began in 1988.
Water masses present at station ALOHA include the North Pacific Tropical Water (NPTW), the Eastern North Pacific Subtropical Mode Water (ESMW), the Shallow Salinity Minimum (SSM), the North Pacific Intermediate Water (NPIW), and Antarctic Intermediate Water (AAIW). The dominant time scales of water-mass variability at ALOHA increase, from interannual to multidecadal, with depth from the surface through the main pycnocline (cf. Lukas 2001; Nakamura and Kazmin 2003). This is consistent with the idea that deeper variations are influenced from more distant regions than those observed at shallower depths (e.g., Huang 2000).
In addition to the water-mass formation processes described above, two different coupled processes may be responsible for the changes in temperature and salinity variation observed at station ALOHA under the seasonal thermocline in the ocean. The first is the Rossby wave mechanism introduced first by Latif and Barnett (1994, 1996). The second is the mechanism of subduction described by Deser et al. (1996) and Gu and Philander (1997). Based on these two mechanisms, Zhang and Liu (1999) depicted two preferential pathways of water-mass anomalies around the North Pacific subtropical gyre: a subduction pathway that isopycnically transfers temperature anomalies from the midlatitude isopycnic outcrop region into a deeper layer and then moves southward into the subtropics and tropics from there. This subduction pathway is basically determined by time-mean thermocline flow; its efficiency mostly depends on the amplitude of surface buoyancy anomalies occurring around the midlatitude outcrop region (Liu and Pedlosky 1994; Schneider et al. 1999). The second is a subtropical pathway along which anomalies move predominantly westward. The associated westward-propagating signal at the sea surface and at thermocline depth resembles a first-mode baroclinic Rossby wave.
Among others, Qiu (2002) and Qiu and Durland (2002) discuss mechanisms of ocean variability observed around the Hawaiian Islands. Qiu (2002) showed, in particular, that the North Pacific Current (NPC) intensified steadily over the period from 1992 to 1998 and that much of this intensification is due to the persistent sea surface height (SSH) drop on the northern side of the NPC. A similar SSH trend is also found in the interior of the Alaska Gyre. Both of these SSH changes are shown to be the result of surface wind stress curl forcing, accumulated along the baroclinic Rossby wave characteristics initiated from the eastern boundary. Huang and Qiu (1994) showed that around the Hawaiian Islands, subduction rates are partially due to vertical pumping along a 1-yr particle trajectory and partially to the difference in the winter mixed layer depth over the 1-yr trajectory. Because the mixed layer is relatively shallow in the North Pacific, the vertical pumping term is very close to the Ekman pumping, while the sloping mixed layer base enhances subduction, especially near the Kuroshio Extension.
Roemmich et al. (2001) showed that the North Pacific heat engine is a shallow meridional overturning circulation that includes warm Ekman and western boundary current components flowing northward, balanced by a southward flow of cool thermocline waters (including subtropical mode waters). A near balance of geostrophic and Ekman transports holds in an interannual sense as well as for the time mean. The Hawaiian Islands create two wakes (Qiu and Durland 2002). A transport wake directly west of the island is characterized by an alteration of the zonal transport caused by the diversion of the interior flow incident upon the east coast of the island into zonal jets extending westward from the northern and southern tips of the island. A potential vorticity (PV) wake embedded in the second-layer streamlines westward and equatorward of the island is characterized by an alteration of the ventilated PV signature and of the baroclinic nature of the flow. The effects of the two wakes combine for a significant impact on both the transport and the baroclinic structure of the Hawaiian Lee Countercurrent (HLCC), indicating that effective modeling of the HLCC should include not only the forcing mechanism but also the influence of the large-scale Sverdrup flow to the east as modified by the Hawaiian Islands. Because of the complex reaction of the ocean near station ALOHA to local and remote forcing, a dynamical interpretation of ocean variability around the Hawaiian Islands cannot be done from the station data alone (as with any other time series data) but requires additional information about large-scale flow changes and nonlocal air–sea interactions.
A valuable instrument for interpreting time series station data is using numerical models that are driven by Numerical Weather Prediction Center winds and buoyancy forcing fields: such an instrument serves to put the local measurements into the large- scale context and can help identify sources of observed changes. However, a prerequisite for a respective study is that the model has skill in producing relevant aspects of the flow field. Because this generally cannot be assumed, one way to enforce the agreement of the model with observations is to constrain the model through large-scale datasets using data-assimilation approaches. If done in a dynamically consistent way, the results can then be used to interpret regional and large-scale changes in the flow field in terms of internal ocean dynamics and in terms of time-varying surface forcing fields. The global Estimating the Circulation and Climate of the Ocean (ECCO)–Scripps Institution of Oceanography (SIO; Stammer et al. 2002a, Köhl et al. 2007) synthesis is such a simulation of the time-varying circulation of the global ocean, which was brought into consistency with most available surface and subsurface observations of the ocean. Results are available on a 1° grid for the period 1992 through 2002. High-resolution modeling and assimilation runs of the tropical Pacific, performed by Hoteit et al. (2005) and Hoteit et al. (2008), were nested into the 1° ECCO solution described here. Douglass et al. (2006) used the ECCO solution for a comparison with and interpretation of high-resolution XBT data in the northeast Pacific.
Because the ECCO solution provides a unique description of the time-varying ocean, it should also be useful for a dynamical interpretation of the station ALOHA data. And because station ALOHA data were not used as constraints in the optimization procedure, they also provide a unique opportunity to test the ECCO global synthesis results with independent information available at the station position. While this clearly holds in the vicinity of the station location, the station data can potentially also be helpful for testing the model over a larger domain: the station data are essentially a measure of ocean changes upstream of the station position (see discussion below) and they represent an integral effect of the ocean circulation and forcing fields along the path of the water masses observed at the time series station. Any agreement between the station ALOHA data and the model can therefore also be interpreted as a stringent test of the model’s quality (or lack thereof) over larger scales of the Pacific.
Accordingly, one goal of our analysis is to compare interannual changes observed at the HOT station ALOHA site with ECCO model results. However, our emphasis is to use the ECCO estimates to infer the origin of major signals observed at station ALOHA. Based on the ECCO synthesis, a budget for salinity (freshwater) anomalies will be provided at station ALOHA to help interpret the local time series measurements. In addition, we will use the adjoint framework to identify the processes (local and remote) that in principle can lead to observed salinity—or, more generally, water mass—changes at station ALOHA. The corollary of this study is the identification of areas that potentially can be influenced by the HOT measurements in a dynamically consistent assimilation framework by spreading the information backward or forward in time and space.
The structure of the remaining paper is as follows: in section 2, we introduce the methodology. In section 3, hydrographic variations estimated at station ALOHA are described, and in section 4, underlying dynamics are discussed. A sensitivity study is provided in section 5 to identify processes and regions potentially affecting the hydrographic data at station ALOHA. Section 6 concludes with final comments.
The ECCO ocean synthesis is obtained by forcing the ECCO model to consistency, within a priori error margins, with ocean data. This is achieved by using the model’s adjoint (Marotzke et al. 1999) to modify the initial temperature and salinity conditions over the full water column and to adjust the time-varying meteorological forcing fields over the full estimation period. The approach used is that of fitting the ECCO general circulation model (GCM) to a large variety of input datasets by using Lagrange multipliers in the context of constrained least squares (see Wunsch 2006 for a summary). The fit is an iterative one as described by Stammer et al. (2002b). In the present application, the horizontal model resolution is 1° over the geographic domain ±80° in latitude with 23 levels in the vertical, and the estimation period is 11-yr, 1992–2002. The ECCO GCM is based on the MIT GCM described in detail by Adcroft et al. (2002), is coupled to a surface mixed layer [K-Profile Parameterization (KPP)] model (Large et al. 1994), and uses the eddy-parameterization scheme of Gent and McWilliams (1990).
The underlying model, data, and methodology are described by Köhl et al. (2007), who also provide further details on the specific model setup and input datasets used as constraints. Observed constraints include several satellite datasets [altimetry from Ocean Topography Experiment (TOPEX)/Poseidon; European Remote Sensing Satellite (ERS-1 and -2); scatterometer data; and Reynolds and microwave SST fields], time-mean surface drifter velocities, in situ hydrographic temperature and salinity profiles, as well as hydrographic sections. Specifically, the difference between the time-mean TOPEX/Poseidon SSH field and a Gravity Recovery and Climate Experiment (GRACE) geoid was used to impose constraints on the time-average absolute circulation. For further details, we refer to Lu et al. (2002), who described the space–time distribution of all input datasets, as well as their prior uncertainties.
As an example of the models’ performance in the Pacific Ocean, the mean wintertime depth of the density layer σt = 25.8 (roughly being a central density of the above salinity integral) is shown in Fig. 1 as it results from the 11-yr mean model March density field. The density surface outcrops along the Kuroshio axis and reaches into the eastern-northeastern Pacific. In the model, its mean depth near station ALOHA is located around 250–300 m; in the HOT data it is 280 m. The outcrop of the density layer σt = 25.3, defining the upper limit of the integral, is located farther south and east and, in the western basin, coinciding approximately with the location of the equatorward limit of the Kuroshio extension.
3. Hydrographic variations in station ALOHA data
Temporal changes of hydrographic fields at station ALOHA are displayed in Fig. 2 as a function of time. A seasonal cycle is clearly visible in the temperature profiles with a maximum amplitude near the surface and a secondary maximum around a 250-m depth. Both are shifted in time with the lower maximum occurring about 6 months earlier/later. For salinity, a seasonal cycle is less dominant in the near-surface layers. Instead, interannual salinity changes are more apparent near the surface. However, as for temperature, a seasonal cycle can be seen around a 250-m depth, suggesting that it is caused not by surface buoyancy forcing but by a different mechanism.
The model seems to simulate the near-surface seasonal cycle in temperature reasonably well, although during summer months simulated temperatures are warmer (by about 1°C) than observed (middle row of Fig. 2). This is potentially caused by the failure of the mixed layer (KPP) model to properly mix the summer surface heat input in the vertical through wind stirring during late summer and early fall. Consequently, the model also fails to reproduce the observed variability observed below a 50-m depth, illustrating its problem to mix deep enough at this specific location, even during winter months. Köhl et al. (2007) provide a rationale that a modification of the optical water type used in the KPP mixed layer model might help to remedy this problems through the absorption of solar radiation at greater depth. The respective destratification could support the wind-induced stirring to destratify the water column during late summer and early fall and thereby would help mixing heat downward.
Focusing on interannual salinity variations, the agreements between the observed and modeled salinity is surprisingly good near the surface, in that the model is capable of simulating a relatively fresh period during the first part of the 10-yr period and a salty event following the 1997/98 ENSO event. We note in this context that in contrast to temperature, not many salinity observations are available for assimilation. But while the model does simulate a seasonal cycle in the near-surface salinity, it seems out of phase with the observations, which tend to show minimum salinity in summer.
The agreement between the estimated and the observed fields is especially clear from the near-surface (down to about 300 m) interannual changes in salt content of the upper ocean water column. Below that depth, the model shows similar variations in its temperature and salinity field, but both do not seem to match the observed structures or trends. In particular, the model shows a maximum in temperature and salinity around 1998 at about a 600-m depth that was not observed. The water is getting colder near the surface after the 1997/98 ENSO event, and salinity increased at the same time. In contrast, the water is getting warmer but still saltier between 100- and 250-m depths. Long-term changes in temperature below 300 m are not pronounced; however, salt seems to decrease over the 11-yr period at this depth.
We note that the ECCO flow field (Fig. 2, bottom) suggests a quasi-biennial oscillation of the circulation in the vicinity of the Hawaiian Islands. The quasi-biennial oscillation is composed of simultaneous increases in the flow field in the zonal and meridional directions, which are especially obvious in the top 200 m. Those oscillations do exist also below that depth in the model but appear there a few months earlier, suggesting an upward propagation. Overall, these quasi-interannual oscillations seem to be more prominent than any ENSO-like oscillation that might have led to the somewhat increased negative zonal and meridional velocities during 1997/98. Investigating the cause of those flow variations on larger scales is beyond the scope of this paper and will be discussed in a separate study. Interannual temperature variations in deep and near-bottom waters at ALOHA were described previously by Lukas and Santiago-Mandujano (1996). Their Fig. 13 shows these variations in 100-dbar averages around 3500 and 4450 dbar. The character of the observed interannual variations is biennial near 3500 dbar and is closer to a period of 3 yr near the bottom. The authors concluded that these latter variations are likely associated with baroclinic Rossby waves excited along the coast of Mexico during ENSO events.
Estimated temperature and salinity changes at station ALOHA lead to changes in SSH (Fig. 3, top). Shown in the figure are ECCO estimates of SSH changes at station ALOHA together with the model’s dynamic topography 0/1000 m (a bias and a least squares trend were removed from the model results). It is obvious that the dynamic topography 0/1000 m does not explain all variations in SSH, and even a dynamic topography computed over the full water column (not shown) leads to residuals on the order of 1–2 cm, suggesting that SSH changes in the model at station ALOHA to some extent are caused by barotropic mass redistribution. Moreover, a splitting of the variability of the full dynamic topography into thermosteric and halosteric contributions (accomplished here by replacing the time-dependent salinity and temperature by their time mean) suggests that salt also leads to important contributions in sea level variations. The salinity-driven signal in the estimated SSH is on the order of ±2 cm, on time scales of several years. In particular, salinity leads to an increased sea level during the period 1995–99; in contrast, the periods 1992–94 and 2000–04 seemed to be characterized by a lower halosteric sea level. During several years this salt contribution to sea level changes is of similar magnitude as those due to temperature (e.g., 1997 or 2000); during other events, the signal is made up entirely by the halosteric change (e.g., 1992). In contrast, thermosteric variations of sea level dominate seasonal to interannual SSH changes.
Firing et al. (2004) discussed in detail the temporal variations in the observed dynamic topography 0–500-m depth, computed from CTD profiles at station ALOHA and Kahe Point and concluded that most of the changes presented in their results on interannual to decadal time scales are thermosteric in origin. This is not altogether confirmed by our analysis. To further illustrate the potential of salinity for changing sea level at station ALOHA, we show in the bottom of Fig. 3 the salinity contribution to changes of the dynamic topography 0–1000-m depth together with similar estimates obtained from ALOHA observations, as they result when the climatological T–S relation for station ALOHA was used instead of individual salt measurements (this provides a measure of the impact of salinity changes on sea level in this region). Although not correlated in detail, in both cases amplitudes are on the order of ±2 cm, which is about 1/3 of the original amplitude and as such cannot be dismissed as unimportant for sea level at this position in the ocean. We also note that the dynamic height differences using climatological T–S versus cruise T–S observations show a biennial oscillation similar to those found before in the velocity field (Fig. 2).
4. Dynamic interpretation of T–S variations
Bingham and Lukas (1996) discussed the vertical structure of the seasonal cycles in HOT temperature and salinity data. For temperature, the most significant seasonal cycle is surface intensified and caused by the seasonal heating cycle. In contrast, annual changes in salinity were found to be large in the core of the NPIW near 500-m depth, potentially due to local anomalous forcing at the water-mass source or to the annual cycle of along-isopycnal advection, or both. The extent to which advective processes are important for the seasonal and longer-period T–S variations in the upper pycnocline (and even in some deeper water masses, such as NPIW) can be tested using the ECCO output. For that purpose, individual terms of the salinity balance equation can be evaluated from the model fields (as for temperature) according to
Here, 〈S〉 stands for the salinity vertically averaged over a layer of thickness Δh = h1 − h2, u = (uih, w) represents the three-dimensional velocity field, and ∇ = (∂/∂x, ∂/∂y, ∂/∂z) is the nabla operator. Also here, HS = 35(E − P)/Δh is the surface flux in terms of salt, usually based on the net surface freshwater flux E − P, which needs to be taken into account if h1 = 0 m. The F represents the 3D mixing terms.
Splitting the salinity signal into time-mean and time-varying components, S = S + S′, and assuming a balance between the time-mean components, the above equations can be rewritten to describe changes in the time-varying salinity field as a function of time-mean and time-varying quantities according to
Monthly mean model fields of salinity and three-dimensional velocity were used to compute individual terms of Eq. (2). At the surface, we included the anomalous surface forcing H′S = HS − , that is, the deviation from the time-mean . Results are shown in Fig. 4 for the two depth levels of 0–200 and 200–400 m, respectively. All curves were smoothed with a 6-month running-mean filter to allow a better display of otherwise noisier curves.
Shown as a bold black curve in Fig. 4, top left, are the vertically averaged salt tendencies [i.e., lhs term of (2)], evaluated from the model’s salinity fields and averaged over the top 200-m depth range. The curve has to be compared with the bold dashed black curve, showing the sum of the rhs (except the residual term, F′, i.e., the sum of horizontal and vertical diffusion that has to be diagnosed from the residuals). Overall, both curves agree, suggesting that the mixing term is not of major significance in the vicinity of station ALOHA near the surface. Instead, a significant contribution to the salinity tendency in the upper level on seasonal time scales emerges from the local net surface freshwater flux anomaly (blue curve) and, on interannual time scales, from the correlation term u′ · ∇S′ (green curve).
This is further illustrated in Fig. 4, top right, showing again the salt tendency term (solid black curve) and the correlation term u′ · ∇S′ (green curve). In addition, we show the horizontal components of the first two rhs terms (the vertical terms are not of significant amplitude and are therefore not shown). An inspection of the figure clearly shows that not a single process dominates the salt tendency but that it is the interplay among all components that leads to the final balance. As an example, during 1992 and between 1997 and 1999, the term · ∇S′ (dashed red curve) correlates with the salt tendency, while during the latter period u′h · ∇S (dashed blue curve) and u′ · ∇S′ seem to cancel each other. Considering the entire period, the changes in the terms u′h · ∇S and · ∇S′ are of equal magnitude but tend to compensate for each other on interannual time scales, so that overall, u′ · ∇S′ remains the largest term. The figure also suggests some of the near-surface salinity tendencies present in the top 200 m that appear to be correlated with meridional velocity anomalies shown in the bottom row of Fig. 2. This is consistent with the above finding that the salt contribution to the dynamic surface height also shows an interannual variation (Fig. 3).
In contrast to the top layer, the next layer, 200–400-m depth (lower row of the figure), is, to first order, largely influenced by the vertical advection (heaving) term, w′ · ∂S/∂z (green curve). The term u′ · ∇S′ (not shown) is now small. Moreover, the horizontal terms u′h · ∇S (dashed red curve) and · ∇S′ (dashed blue curve) vary on interannual time scales as well and tend to compensate for some of the tendencies caused by the vertical heaving term, so that the net effect is smaller than a pure heaving would suggest. As an example, during early 1994 this is accomplished through the u′h · ∇S term, while several months later the uh · ∇S′ reduced the heaving influence. Nevertheless, the term w′ · ∂S/∂z remains an important mechanism for salt changes on seasonal and interannual time scales in this depth range, consistent with the understanding that upwelling annual Rossby waves passing the Hawaiian Islands are responsible for some fraction of the observed hydrographic variations there (Sakamoto et al. 2004). Although the previous analysis of the station ALOHA data does not confirm such a heaving signal, we recall that those computations were done with only 8 yr of noisy data and that a recomputation of the longer time series now available might also show a seasonal cycle of heaving that is significant.
5. Adjoint salinity sensitivity
To further understand which mechanisms and associated pathways might actually be responsible for the changes observed at station ALOHA, we used the adjoint sensitivities to investigate what causes salt to vary in our model solution in the depth range of 200–600 m at the position of station ALOHA. For that purpose, a target function was defined as
where (S − S0) represents salt anomalies. This depth range, over which averaged salinity sensitivity was computed, corresponds to roughly 25.3 through 26.8 σt in density for station ALOHA and ECCO results (this density range includes NPIW, which is created in the far western North Pacific, and also the SSM waters of the eastern North Pacific).
To evaluate processes impacting the above salinity integral J, the ECCO adjoint model was run again on the same 1° grid as the previous ECCO-SIO optimization, but this time only once forward and backward for 5 yr. The adjoint variables were analyzed during the full backward run, starting from the last time step of the model τf . These adjoint variables specify the gradients (sensitivities) of the target function with respect to the model-state variables, including surface forcing fields. [See, e.g., Marotzke et al. (1999) and Köhl (2005) for discussions of adjoint sensitivities dealing with the heat transport and the meridional overturning circulation at 30°N in the North Atlantic, respectively.]
a. Sensitivity to hydrographic changes
Figure 5 shows the sensitivity field ∂J/∂Xi, where J is specified as above and Xi is the time-varying salinity at 222-, 610-, and 2450-m depths, respectively. Shown are monthly mean fields for periods 6 months, 2 yr, 3.5 yr, and 5 yr before τf , respectively (time goes backward). Initially (τf − 6 months), a vertically coherent sensitivity signal occurs east of the HOT position with a shape that is typical for a Rossby wave. Further backward in time, this signal moves eastward in the sensitivity field, consistent with Rossby waves in the forward simulation radiating westward from the eastern Pacific. As can be seen, this signal takes about 3 yr to cover most of the region of the Pacific east of Hawaii, suggesting that a positive salt anomaly emerging along the coast of California takes about 3 yr to reach station ALOHA in the form of first-mode Rossby waves (the phase velocity cp = 5.5 cm s−1 is in good agreement with theoretical and observational results; Chelton and Schlax 1996).
As time progresses backward, an advective/diffusive signal emerges toward the northwest and north, especially at the upper (shallow) depth level. This signal aligns well with the stream lines of the time-mean circulation in the model, shown in the right column of the figure for each depth level. As can be seen from Fig. 5, top right, the shallow water-mass branch (shallow salinity minimum waters of the eastern North Pacific) reaches northwest into the winter outcrop from where it emerged about 2 yr before arriving at the HOT site, while deeper levels (NPIW) were in contact with the winter outcrop out 3–5 yr backward in time, following advective pathways from there to station ALOHA. This is in good agreement with independent estimates of travel times from the outcrop of isopycnals to ALOHA of 3, 4.8, and 6.9 yr for the 25.0, 25.5, and 26.0 σθ surfaces, respectively, provided by R. Lukas (2006, personal communication) on the basis of geostrophic speed along streamlines computed from Levitus et al. (1998) hydrographic data.
Also shown in the bottom row of Fig. 5 is the sensitivity ∂J/∂T600 m, that is, the sensitivity of vertically averaged salt to temperature changes at the 600-m depth. For an easier comparison, the ∂J/∂T600 m field was scaled by −β/α ≈ −10 so as to represent the equivalent salt sensitivity. Because the scaled temperature sensitivity leads essentially to the same result as found for ∂J/∂S600 m, much of what emerges in terms of ∂J/∂S600 m outside the immediate vicinity of Hawaii seems to be a dynamic signal, with kinematic effects probably limited to the near surface.
To aid in the dynamical interpretation of the sensitivity signal, Fig. 6 shows vertical sections of ∂J/∂S(z), the top two sections crossing the HOT site in the meridional and zonal directions, and two additional zonal sections being located north and south of ALOHA at 15° and 35°N, respectively. The propagation of information along isopycnals toward the HOT site is especially obvious for the upper ocean in the meridional section, where a positive anomaly can be traced back residing between the two isopycnals σt = 25.8 and σt = 26.8 during the year τf − 5 yr around the latitude of 35°N. This result confirms the mechanism of ventilation (subduction + advection) as one of the primary drivers changing salinity above the 600-m depth at station ALOHA.
Perhaps less expected is the sensitivity of salinity above the 600-m depth emerging from underneath and from the south below the density surface σt = 26.8 with contributions reaching as far back as 15°N during τf − 5 yr. A possibility is that this could be due to the fact that Antarctic Intermediate Water (centered around 800–1200 m) mixes with the NPIW near 500 m, so that there is an influence from below in the 200–600-m range. AAIW enters the North Pacific along the western boundary and extends eastward near 12°–15°N (Reid 1965; Kennan 1993). Whether the negative sensitivities present in the figure west of the longitude of station ALOHA can be associated with this process has to be further investigated.
In contrast to the meridional section, the zonal sections reveal pronounced vertically coherent sensitivities east of station ALOHA, which in the meridional section are present only during the beginning of the backward run. The signal can be interpreted as being caused by baroclinic Rossby waves of varying vertical modes, traveling with different phase speeds. Some of this signal is again bottom intensified and appears to originate from steep topography, suggesting the influence of bottom slopes on vertical motion, which can influence salinity between 200 and 600 m through heaving. The sensitivity of Rossby waves to bottom topography was already studied by Barnier (1988) in the context of a numerical study on the influence of the mid-Atlantic ridge on nonlinear first-mode baroclinic Rossby waves generated by seasonal winds. It suggests here an additional mechanism for salinity changes near Hawaii through the flow interaction with the bottom topography and the resulting vertical movement of the water column. The extent to which this is a real mechanism—given the crude representation of topography near Hawaii in the model—remains to be investigated.
A summary of Figs. 5, 6 is that on time scales of up to 1 yr, regions that can influence the model salinity at the HOT site via Rossby waves are confined to the area east of Hawaii. The primary mechanism for changing salinity below the surface mixing layer on those time scales is a changing flow field, advecting different water masses past the time- series position (vertically or horizontally). As time moves backward, Rossby wave energy can still be present, but sensitivities aligning with the background flow field, suggesting that the advection of water-mass anomalies along isopycnals seems to be the principal mechanism for water-mass changes at the HOT station. We note that because the salinity specifies values per model grid cell, the sensitivities shown above appear somewhat enhanced in deeper layers because the model’s vertical resolution is lower there.
b. Sensitivity to forcing
The question remains: What causes both near-surface and deeper salinity changes at station ALOHA? From the above discussion two candidate processes are obvious: 1) surface buoyancy flux forcing upstream on streamlines passing the station position and 2) wind forcing to the east, generating Rossby waves. To further elaborate on the hypothesis that surface forcing is an important mechanism for T–S changes at station ALOHA, we show in Fig. 7 the sensitivity of salt at station ALOHA between a 222- and a 610-m depth to surface salt flux (i.e., negative freshwater) forcing FW, and heat flux forcing, HF (both being positive into the ocean), that is, ∂J/∂HF, ∂J/∂FW, and to zonal and meridional wind stress, that is, ∂J/∂τx,y, as a function of time over the same 5-yr period as considered before.
From the figure, wave activity is apparent over the last 1–2 (τf − 2) yr with ventilation/advective processes being dominant on longer time scales. Lukas (2001) previously discussed the impact of decadal changes in rainfall on the watermass properties of the North Pacific and potential impacts on the salinity changes observed at station ALOHA. The author also shows density-compensated salinity anomalies, so that HF and FW should be negatively correlated in the region that is producing those anomalies.
Processes involved in the sensitivity of wind forcing on the salinity near Hawaii include Rossby waves and the change of the large-scale (i.e., time-varying) Sverdrup circulation. Both are processes acting east and upstream of the HOT site. Accordingly, the sensitivities of salt to wind stress appear east of the Hawaiian Ridge and show Rossby wave–like structures. With increasing time scales, the sensitivity tends to show larger zonal scales for the zonal wind stress as would be required to change the Sverdrup circulation. This can again lead to changes in salinity at the HOT station through the advection of different water masses. In contrast, the sensitivity to the meridional wind stress component also moves farther eastward (suggesting the creation of Rossby waves there) and becomes less important as indicated by the decaying signal on longer time scales.
Because the wind effect leads primarily to barotropic changes of the circulation through linear vorticity dynamics, the effect of the wind can result in near-bottom sensitivities (through flow-bottom interaction/form drag). The short-time lead pattern in ∂J/∂τx could also be due to Ekman pumping, as the ∂τx/∂y maximum appears to be at the same latitude as ALOHA. Downwelling would increase salinity between the salinity maximum (about 160 m) and the salinity minimum (about 550 m) as discussed above. The wind stress sensitivity pattern near 25°N may therefore be related to increased downwelling in the region of the subtropical front, where the salinity maximum waters are created.
c. Forward and backward particle trajectories
The identification of mechanisms for creating all of those water-mass anomalies, embraced by the above salinity integral, and their pathways to station ALOHA is not possible from our coarse (in the vertical) sensitivity results. However, it would be desirable for future experiments to gain further information about water-mass pathways in our solution. For that purpose we introduced particles at every vertical level of the model at the location of station ALOHA between 200- and a 600-m depths and used the model’s velocity field to advect the particle backward and forward in time, thus mimicking the backward and forward pathways of information (sensitivities). The results are shown in Fig. 8 in terms of backward trajectories for the years τf − 4, τf − 6, and τf − 8. In each panel, the actual particle depth is coded in color.
The contact with the surface of the upper most particles (isopycnals) within the initial depth range after 4 yr northeast of station ALOHA is obvious, essentially following the streamline pattern of the mean circulation, shown in Fig. 5. After 7–8 yr, all particles released in the upper depth range (about 200–400 m) occupy essentially the entire outcrop region. Particles released on deeper isopycnals, roughly within the depth range of 400–600 m, reach farther east and northward from there, again consistent with the stream lines of the background flow field. However, they never outcrop during the 8-yr period considered here. This suggests that in terms of surface freshwater forcing and ventilation of new water masses, it is really the upper part of the considered depth range that matters. In contrast, the deeper layers are dynamically active in that they change the advection pathways of water masses at the HOT site, and therefore should be sensitive primarily to wind forcing and large-scale flow changes.
Suga et al. (2000) analyzed the correlations of variations in NPTW between station ALOHA and locations to the west, for example, showing the influence of advection in the North Equatorial Current (NEC). This suggests that the observations at ALOHA, if provided over long enough a time, could also constrain the analysis to the west as well as to the north and east, one region for prior states, and the other region for subsequent states. To demonstrate this we also computed forward trajectories, shown in Fig. 9, for the years 4, 6, and 8, but now in forward mode. The figure reveals the quite distinct advection pathway between the upper and middle water masses at station ALOHA, the latter following essentially the gyre circulation westward at that depth, while the upper branch gets advected eastward by the North Pacific current before it recirculates westward south of the Hawaiian Islands.
The particles from the middle branch reach the western boundary and the Indonesian Throughflow after about 5 yr; after 6 yr, they already populate the tropical Pacific and start to invade the Indian Ocean as well; after 8 yr, some particles of the lower branch can be found in the Kuroshio Extension and in the Indian Ocean, while the particles from the upper branch are quite dispersed over the tropical Pacific. Particles released near 600-m depths do not seem to move much over the time scale considered. Figures 8, 9 suggest that the typical time scale for information, propagating around the subtropical gyre of the Pacific, is about 16 yr.
A comparison of the ECCO global ocean-state estimate is performed with independent observations available at the Hawaii Ocean time series station ALOHA. The comparison shows that at this location, ECCO has a realistic vertical structure in its hydrographic fields. The temporal variability, notably that of salinity in the upper 200 m, also shows skill in simulating observed changes. However, deficiencies are obvious, which need to be improved upon in future estimation experiments that make use of more recent in situ data available from Argo.
Changes in ECCO sea surface height can largely be explained through changes in the density field, but deviations from the pure steric component on the order of 1–2 cm suggest that barotropic mass redistribution contributes to observed SSH changes. The steric changes are dominated by the thermal components, but on interannual and longer time scales, salinity changes are an important contributor to observed sea level changes.
Based on the ECCO synthesis, a balance of salinity (freshwater) tendencies is provided at the HOT station for the upper 400 m. There is a clear correlation of salinity tendencies with a quasi-biannual oscillation in the horizontal flow structure at this location. Between 200 and 400 m, heaving of the mean salinity structure contributes significantly to seasonal and interannual changes in salinity (and temperature). Data from station ALOHA are an important element for testing long-term variations in the ECCO-state estimates against independent data, which for this reason are currently excluded from the estimation procedure.
At the position of station ALOHA, the advection of water-mass properties is an important element for shaping long-term changes there. On time scales of up to 1 yr, all changes observed at the HOT site below the surface mixed layer are induced from the east and appear to be primarily induced through anomalous advection of the background hydrography. Longer time scales reveal enhanced influence of water-mass anomalies being advected along isopycnals by the background flow field with increased influence from the north, where the isopycnals are ventilated. While internal model errors need to be improved upon in future estimation approaches, the sensitivities of the estimated salinities at the HOT position shown in Fig. 8 clearly demonstrate the large impact of surface forcing on the salinity (and temperature for that matter) simulated at station ALOHA; this suggests that further improvements in surface forcing fields are a prerequisite for further improvements of ECCO results in the North Pacific. At the same time, our sensitivity results support the hypothesis that decadal changes in salinity observed at station ALOHA result from changes in surface buoyancy forcing, especially freshwater forcing in the ventilation region as suggested by Lukas (2001). Heaving through local Ekman pumping or through Rossby waves seems another plausible mechanism for creating measurable salinity changes near Hawaii.
Until now we have considered model–data differences (or agreements) between the ECCO simulations north of the Hawaiian Archipelago and also mechanisms that lead to observed changes in watermass properties there. However, our investigation also describes the region of the Pacific basin that can be influenced in state estimations by observations taken at a single time series station over a sufficiently long time period. Data from station ALOHA, if included in a state-estimation context (e.g., the ongoing ECCO ocean syntheses over a 50-yr period; Köhl and Stammer 2008a, b), can influence the estimation of the circulation and water masses over large parts of the North Pacific. Even when performed over only a 10-yr period, it is already obvious that some benefit from the station ALOHA data can be expected, in a state-estimation context, to reach as far west as the Kuroshio western boundary region or even into the Indian Ocean. On the other hand, time series stations, in the presence of ample other datasets now available since the 1990s [thanks to the global ocean observing system that emerged out of World Ocean Circulation Experiment (WOCE)], have their very specific value: because of their long duration and their associated information content of large-scale variability, those data are uniquely suited as independent datasets required for testing ocean and climate models. They need to be maintained as part of a long-term ocean and climate observing system.
SP thanks the hospitality of the Scripps Institution of Oceanography, where he spent two years as visiting scientist. R. Landsberger helped with some computations. Helpful comments from two anonymous reviewers are gratefully acknowledged. Reanalysis surface forcing fields from the National Center for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) were obtained through a computational grant at NCAR. Computational support from the National Partnership for Computational Infrastructure (NPACI) and the National Center for Atmospheric Research is acknowledged. Supported in part through ONR (NOPP) ECCO Grants N00014-99-1-1049 and N00014-99-1-1050, through NASA Grant NAG5-7857, through NSF Grant OCE 9730071. RL and FSM were supported by the National Science Foundation through Grants OCE-0117919 and OCE-0327513. This is a contribution of the Consortium for Estimating the Circulation and Climate of the Ocean (ECCO) funded by the National Oceanographic Partnership Program.
Corresponding author address: Detlef Stammer, Institute für Meereskunde, Universität Hamburg, Bundestrasse 53, 20146 Hamburg, Germany. Email: email@example.com