RESEARCH HIGHLIGHT

Application of the variational data-assimilation technique for the study of the Bering Sea: Climatological studies including hindcast and forecast of the local circulation.

G. Panteleev, D. Nechaev, V. Luchin, P. Stabeno, N. Nezlin, M. Ikeda

February 15, 2006

Introduction

The circulation and distribution of the hydrophysical properties in the Bering Sea determine the heat and fresh water exchange between the North Pacific and Arctic Oceans. The circulation in the BS basin is driven by atmospheric forcing and the inflow/outflow transports through four primary passes: Kamchatka Strait, Near Strait, Amchitka Pass and Amukta Pass [Stabeno et al., 1999, 2005]. The currents in the Bering Strait are relatively well monitored by velocity moorings, while the water transport through the straits and passes of the Aleutian Arc have been mainly explored through the estimates of the baroclinic currents using a dynamical method. According to Stabeno et al., [2005], recent velocity measurements in Amutka Strait reveal the inflow of approximately 5 Sv, which is 4 times larger than previous estimates based on the dynamical calculations.

The direct velocity measurements are relatively expensive and it is hard to believe that detailed velocity observations will be carried out in the Aleutian passes and straits in the near future. At the same time, there are currently many historical mooring and surface drifter measurements in the Bering Sea region collected during the two last decades (http://www.pmel.noaa.gov/foci/FOCI\_data.html). In addition, there are a lot of historical meteorological measurements. By combining all historical velocity measurements and historical temperature/salinity data, a reliable estimate of the climatological BS circulation may be derived.

The natural way to combine temperature, salinity and velocity observation is to apply the variational data-assimilation approach based on an ocean general circulation model (OGCM). During the two last decades, this technique proved to be an extremely useful tool for the reconstruction of both climatological [Tziperman and Thacker, 1989] and real circulation of the world ocean [Stammer et al., 2003, Awaji et al., 2004]. We also would like to note that reconstruction of the climatological Bering Sea circulation gives at least two obvious perspectives:

1. First, we will have the possibility to analyze the sensitivity of climate circulation with respect to the inflow/outflow through the southern boundary. This analysis will help to define the passes, which give the largest influence on the BS circulation.
2. Second, the reconstruction of the climatological sea surface height (SSH) will give the referenced SSH distribution. This SSH distribution can be combined with the SSH altimetry anomaly measured from a number of satellites (Topex Poseidon, ERS, Jason-in order to derive the estimate of the absolute SSH in the BS. The obtained estimates of the SSH can be used for the hindcast and forecast of the surface circulation in the Bering Sea.

The preliminary results of application of the variational data-assimilation technique to the region shown in Figure 1 gave a new estimate of Kamchatka Strait transport and reveal the strong potential for monitoring surface circulation through variational data assimilation of anomaly altimetry data derived from the AVISO project.

Figure 1. Distribution of historical salinity data in August. The green line marks the region of study.

Data-assimilation technique

To find the optimal solution of the model we performed strong constraint minimization of the cost function measuring the distance between the model solution and data on the space of the control variables [Le Dimet and Talagrand}, 1986]. Control variables include the initial conditions, the model field values required to specify the open boundary conditions, and the surface fluxes of momentum, heat, and salt [Nechaev et al.}, 2005].

Forward and adjoint models
The primitive equation model utilized in this study was successfully used for the reconstruction of the evolution of summer climatological circulation in the Barents Sea [Panteleev et al., 2006] and for the hindcast of the circulation in the Tsushima Strait [Nechaev et al., 2005] for a period of 10 months. The forward model is a modification of the C-grid, z-coordinate OGCM designed by [Madec et al., 1999]. The model is implicit both for barotropic and baroclinic modes permitting model runs with relatively large time steps. The adjoint code was built analytically.

Data:
We utilize the following data sets:

1. 41,911 temperature/salinity profiles collected in the Chukchi and Bering Seas (between 55N and 69N). The data were collected by US, Japanese and Russian organizations during the period 1932—2004 (Figure 1).
2. 500 satellite-tracked drifters trajectories from Fisheries Oceanography Coordinated Investigations (NOAA) data base (http: //www.pmel.noaa.gov/foci/FOCI\_data.html). Only "reliable" drifter velocity estimates were assimilated. (Figure 2a).
3. Transport estimates of 1.1+-0.2 Sv were taken from [ Woodgate et al., 2005].
4. NCEP/NCAR wind stress and surface heat/salt flux climatology.

Figure 2. (a) - averaged summer velocities at 40 m derived from drifter data. The shaded areas show the spatial distribution of zonal velocity STD, which ranges from 5 cm/sec (green) to 20 cm/sec (red). (b) - the trajectories and 2-day mean velocities of the four ARGO drifters (http://www.usgodae.org) parked at 1000 m during 2002--2004. Circles and asterisks designate the initial and final location of the drifter.
Dashed line denotes 1000~m and 3000 m isobaths.

Preliminary results

Climatological circulation

The reconstruction of the climatological summer circulation was done using a quasistationary variational data-assimilation approach proposed by Tziperman and Thacker, [1989]. We assimilated all the observations outlined above, i.e.temperature and salinity climatologies, surface drifter data, the Bering Strait transport, and meteorological data.

The optimized velocity fields obtained are shown in Figure 3. The general circulation pattern agrees with conventional scheme of the BS circulation, which includes weak currents on the eastern shelf, the Bering Slope current (BSC), the intensive KC along the Eurasian continent, the intensive Navarine Current in the Gulf of Anadir and strong northward flow in the Bering Strait. The KC originates as a continuation of the BSC at approximately 175E and then flows clockwise around the Shirshov Ridge. The obtained circulation reveals strong topographic steering of the KC. According to our results, in the vicinity of the point 58N, 170E, the KC splits into two branches. One of these branches (we will call it "coastal branch") follows northward along the 1000m isobath, while, the other branch (the "off-shore branch") flows westward across the Kamchatka Basin and joins the coastal branch of the KC near Karaginsky Island. This is similar to the flow pattern in Stabeno and Reed, [1994]. These branches join northeast of t Karaginsky Island, resulting in the gradual increase of the KC transport from approximately 14 Sv in the Olyutorski Gulf to 24 Sv in Kamchatka Strait. The obtained estimate of the KC transport is almost 1.5--2 times higher than the "traditional" estimates derived by dynamical method.

The relative mean between modeled surface velocities (Figure 3a) and drifter velocities (Figure 2a) is 0.71. This relatively high error can be explained by the high STD of drifter velocities, which reaches 20 cm/s in the BSC (Figure 2a). Despite the high error, the absolute amplitude of the surface velocities in Bering Slope and Kamchatka Currents (Figure 3a) are close to the amplitude of drifter velocities shown in Figure 2a. A limited amount of available mooring velocities (Figure 3a, thick arrows) show very good agreement with optimized velocities. Unfortunately, most of these data are located in the eastern part of the Bering Sea and cannot confirm the reliability of the KC reconstruction. Because of that, we compare our results with the trajectories of four ARGO drifters (www.usgodae.org), which were launched at 1000 m depth (Figure 2b).

Three of these drifters were released in the southeastern part of the Bering Sea and were carried by the BSC up to 58-59N where they deflected from the continental slope and drifted across the Aleutian Basin to Shirshov Ridge, where they joined the KC. One of these drifters entered the KC and sailed clockwise around the Shirshov Ridge up to 60N, i.e. the trajectory of this drifter follows the "coastal" branch of the KC discussed above (Figure 3a,b). The fourth drifter entered the BS through Near Strait (Figure 2b). The drifter crossed the Bowers and Aleutian basins and joined the KC near the southern end of Shirshov Ridge (Figure 3b) and drifted clockwise around the ridge. This drifter deflected westward at 58N, 170E following the "off-shore" branch of the KC obtained in our results. The splitting of the KC at this point is probably caused by sharp bottom topography changes in the Shirshov Ridge region. Analysis of bottom relief (Figure 1) allows us to speculate that the off-shore branch initially follows the 3000 m isobath and then deflects westward near 58N, 170E, while the trajectory of the coastal branch of the KC coincides with the 1000 m isobath. The averaged speed of these four drifters was approximately 4.5 cm/sec. This value is close to the mean 3.5 cm/sec speed of the modeled currents at 1000 m (Figure 3b). That seems to be a good agreement with observations, because Lagrangian velocity estimates tend to be larger than Eulerian.

The obtained estimate of the KC transport estimate is in a good agreement with the 20Sv summer KC transport derived by combining the section hydrophysical data and surface floats data by Hughes at al., [1974], and two times higher than results based on the dynamical method [Stabeno et al., 1999]. Note that the recent measurements in Amutka Strait reveal the transport of 4 Sv, which is 4-5 times higher than the previous estimates based on the dynamical method.[Stabeno et al., 2005]. Therefore, we can state that the current knowledge of the water balance in the Bering Strait should be re-examined.

Figure 3. The optimized velocities at 17m (a) and 1000m (b). Thick red arrows denote the mean velocities observed at several moorings (a) and ARGO velocities at 1000m (b).

Forecast of the local circulation

The reconstructed mean SSH can be used as a reference level for recalculating the weekly SSH Anomaly distributed by the Archiving Validation and Interpretation of Satellite Data in Oceanography (AVISO) project (www.aviso.cls.fr). Figure 4 shows the evolution of the surface circulation reconstructed through the assimilation of the real SSH into the non-stationary OGCM. The comparison with available surface drifter velocities (black arrows at Figure 4) reveals very good agreement with reconstructed velocity fields.
The quantitative comparison between the absolute surface velocities provided by the AVISO project and surface circulation derived through the variational data-assimilation technique is shown at Figure 5. As it seen in Figure 5, the results of the 4Dvar approach (Figure 5, left panel) are in a better agreement with surface drifter velocities than the surface circulation provided by AVISO data.

Conclusions.
The preliminary results of the reconstruction of the climatological and real circulation in the BS reveals strong advantages in using the variational data-assimilation approach. Our results allowed us to derive a new estimate of the transport through the Kamchatka Strait and show the possibility of reliable hindcast/forecasts of the currents in the region.

Figure 4. Hindcast and forecast in the Bering Sea through the assimilation of the SSH data. Blue arrows – model results. Black arrows – surface-drifter velocities.

Figure 5. Left panel - hindcast and forecast of the Bering Sea circulation during summer 2002. Black arrows- the real surface drifter velocities. The mean relative model-drifter velocity data error - 0.77. Right panel -Surface velocities derived from the AVISO (Topex-Poseidon) absolute SSH data. The mean relative AVISO-drifter velocity data error - 1.6.