Surface Currents in British Columbia Coastal Waters: Comparison of ...

Surface Currents in British Columbia Coastal Waters: Comparison of Observations and Model Predictions William R. Crawford, Josef Y. Cherniawsky and Patrick Cummins1 Canadian Hydrographic Service Fisheries and Oceans Canada Institute of Ocean Sciences P.O. Box 6000, Sidney, BC V8L 4B2

[Original manuscript received 13 July 1998; in revised form 26 January 1999]

abstract Observations of the motion of ocean surface drifters are used to evaluate numerical simulations of surface currents in the region of Queen Charlotte Sound on the West Coast of Canada. More than 30 surface Argos drifters were deployed in the spring and summer of 1995, revealing daily average currents of 10 to 40 cm s–1 near the coast of Vancouver Island in summer, and less than 10 cm s–1 in mid-sound. Wind observations in this region are provided by a network of weather buoys. Comparison of daily average drifter velocities and winds shows that the drifters moved at 2 to 3% of the wind speed, and at about 30 degrees to the right of the wind. A complex transfer function is computed between daily wind and drifter vectors using least squares techniques. The ratio of variance in the least squares residual currents to the variance of observed drifter currents is denoted γ2. A percent goodness-of-fit is defined as g(γ2) = 100(1 – γ2), and is 42% for the case of daily winds and drifter currents. Drifter-measured currents are compared with two numerical simulations of surface currents: Fundy5, a steadystate baroclinic model based on historical water property measurements in summer, and the Princeton Ocean Model (POM), a prognostic, baroclinic model forced by the measured winds. Fundy5 by itself provides a goodness-of-fit of only 3%, whereas POM has g(γ2) = 42%. The combination of Fundy5 plus daily wind gives g(γ2) = 43%. Although the prognostic model performs only as well as the winds by themselves, it simulates the near shore currents more accurately and reproduces the speeds and veering in the surface Ekman layer on average without bias. Residual currents unexplained by POM are likely due to advection of water masses into this region and horizontal inhomogeneities in the density field that are not input to the model, as well as to Stokes drift of wind waves and to net Lagrangian tidal motion not represented by the model.

1 Ocean Science and Productivity, Science Section, Fisheries and Oceans Canada, Institute of Ocean Sciences, P.O. Box 6000, Sidney, BC V8L 4B2

ATMOSPHERE-OCEAN 37 (3) 1999, 255–280 0705-5900/99/0000-0255$1.25/0 © Canadian Meteorological and Oceanographic Society

256 / William R. Crawford et al. résumé Pour évaluer les simulations numériques des courants de surface dans la région du détroit de la Reine-Charlotte dans l’ouest du Canada, on a utilisé les observations du mouvement de dériveurs de surface océanique. En déployant plus de 30 dériveurs Argos de surface au printemps et à l’été de 1995, on a constaté qu’il y avait quotidiennement des courants moyens de 10 à 40 cm/s près de la côte de l’île de Vancouver en été, et des valeurs moindres que 10 cm/s au milieu du détroit. Les observations du vent dans cette région nous ont été fournies par un réseau de bouées météorologiques. La comparaison entre les valeurs moyennes quotidiennes des vents et des vitesses des dériveurs montre que les dériveurs se sont déplacés de 2 à 3 % à celle du vent, en environ 30 degrés à la droite du vent. Une fonction de transfert complexe est calculée quotidiennement entre les vecteurs vent et dériveur en utilisant la méthode des moindres carrés. Le rapport de la variance dans les courants résiduels des moindres carrés à la variance des courants observés des dériveurs est dénoté par γ2. Un pourcentage de validité d’adjustement est défini comme g(γ2) = 100(1– γ2) et il se chiffre à 42 % pour le cas des valeurs quotidiennes des vents et des courants des dériveurs. Les courants mesurés des dériveurs sont comparés avec deux simulations numériques de courants de surface : le Fundy 5 qui est un modèle barocline stable basé sur des mesures historiques prises à l’été à partir d’une station riveraine et le Modèle océanique Princeton (MOP) qui est un modèle barocline de prévision forcé par les vents mesurés. Le modèle Fundy 5 en lui-même fournit une validité d’adjustement de 3 %, tandis que le modèle MOP possède un pourcentage g(γ2) = 42 %. La combinaison du modèle Fundy 5 avec des vents quotidiens donne un pourcentage g(γ2) = 43 %. Quoique le modèle barocline de prévision ne fonctionne qu’avec les vents, il simule les courants près du rivage avec plus d’exactitude et reproduit, en moyenne sans biais, les vitesses et les courants dextrogyres dans la couche d’Ekman. Les courants résiduels qui ne sont pas expliqués par le modèle MOP sont probablement dus à l’advection de masses d’eau dans cette région. Ils sont dus aussi aux éléments horizontaux non homogènes dans le champ de densité ne faisant pas partie du modèle, ainsi qu’à la dérive de Stokes des vagues provoquées par le vent et le mouvement lagrangien résiduel de marée, non représenté par le modèle.

1 Introduction During the spring and summer of 1995 the Canadian Hydrographic Service undertook a major study of the surface currents in Queen Charlotte Sound and southern Hecate Strait, British Columbia. This region covers about 200 by 200 km, with a large open entrance to the Pacific Ocean. More than 30 surface drifters were launched, accumulating in excess of 500 drifter-days of data, and providing excellent coverage of flow in mid-strait as well as coastal currents within a few kilometres of the shoreline. We present the drifter observations here, together with detailed evaluations of numerical simulations of surface currents in these waters. This remote and windy region along the British Columbia coastline is the site of many vessel accidents and also has potential for oil exploration. Both activities require current predictions by the Coast Guard, who have ready access to real-time wind observations, and also to software that is able to imbed surface currents provided by diagnostic numerical models. However, it is a great leap for such desk-top software to implement baroclinic prognostic numerical models, and also for Coast Guard staff to manage such models. This paper reveals that for regions such as the central coast of British Columbia in summer, where prevailing currents are weak

Surface Currents in British Columbia Coastal Waters / 257 and local wind forcing is strong, surface currents may be predicted reasonably well using wind data alone. We have prepared several papers on surface currents in the shelf seas around the Queen Charlotte Islands. The first is a comparison of tidal currents measured by these drifters with currents simulated by numerical models (Crawford et al., 1998, hereinafter denoted C3F98). Results of this study reveal good agreement between these two fields, except where internal tidal motions limit the accuracy of the models. The second aspect, devoted to non-tidal motion is described by Crawford et al. (1996, hereinafter CCHB96). This paper examines the predictability of near-surface currents in the waters of Dixon Entrance and Northern Hecate Strait using diagnostic models that are based on measurements of density through the region, in a manner similar to that of Ballantyne et al. (1996). CCHB96 observe that model-defined seasonal currents improve predictions in regions where such currents are relatively strong. In this paper we examine surface current predictions in Queen Charlotte Sound and neighbouring waters. For this case, in addition to the seasonal current field used by CCHB96, we also use the Princeton Ocean Model (POM) (Blumberg and Mellor, 1987; Cummins and Oey, 1997) to simulate synoptic wind-driven flows, and to compare with the observed currents. The next section describes the geographic region, the type of drifters used and the seasonal extent of drifter coverage. Section 3 presents the dynamical numerical models and their application to this region. Section 4 describes the statistical analysis of predictions, providing insight to the relative usefulness of these models and wind forcing. Details of preparation of seasonal input data to the diagnostic model are presented in the Appendix. 2 Region and Measurements Queen Charlotte Sound and Hecate Strait are semi-enclosed seas bounded by the mainland of British Columbia, the north coast of Vancouver Island and the eastern shores of the Queen Charlotte Islands (Fig. 1). The western entrance to this sound opens to the Pacific Ocean at the latitude of weak prevailing eastward currents, which bifurcate several hundred kilometres to the west. One branch flows southeast along the west coast of Vancouver Island while the other flows northwest along the west coast of the Queen Charlotte Islands. In summer there is an intermittent shelf break current flowing to the southeast, often interrupted by plumes of water originating near the mainland coast of British Columbia and flowing westward out of the sound into the Pacific Ocean. Crawford et al. (1985) show that near-surface currents are strongest near shore and are correlated with local winds. Tidal currents are up to 1.5 m s–1 near Cape St. James (Crawford et al., 1995; C3F98) and have speeds between 1 and 2 knots over most of this region. However, we have filtered out tidal currents from the present analysis in order to focus on the non-tidal motion. We deployed more than 30 Argos drifters manufactured by Seimac Ltd. These are

258 / William R. Crawford et al.

Fig. 1

Queen Charlotte Sound and surrounding region. Depth contours are in metres.

Surface Currents in British Columbia Coastal Waters / 259

Fig. 2 Schematic of the Seimac drifter used in this study.

model SL-DMB units in cylindrical pressure cases, about 1 metre long by 10 cm diameter, designed to be launched from a fixed wing aircraft or helicopter (Fig. 2). Four drag-vane panels open after landing and act as near-surface drogues. A parachute cushions the landing in water and provides additional drogue area when floating. During operations in the spring of 1995, the transmissions from the drifters were received irregularly for 5 to 20 days, then ceased completely. It is believed that marine growth accumulated on these drifters and their drogues, eventually sinking them. For operations from June to September 1995 we removed the parachutes and launched the drifters from ships. Each drifter was set to turn its power off after 60 days. All transmitted for this period unless they washed up on shore and were damaged. The set of all drift tracks in spring and summer of 1995 is presented in Fig. 3. This figure reveals several significant features of the flow. Firstly, none of the drifters washed up on the shores of the Queen Charlotte Islands, nor the west coast of Vancouver Island. Secondly, none drifted into the region of Rivers Inlet on the central coast of British Columbia probably due to the fresh water outflow from this inlet.


Fig. 3

Tracks of Argos surface drifters between April and October 1995.

Surface Currents in British Columbia Coastal Waters / 261 The dense array of tracks in the interior of Queen Charlotte Sound provides excellent repeated coverage for statistical analysis. Drifter data were processed as follows. Positions were interpolated to hourly values using cubic splines. A low-pass Kaiser (1974) filter (1/4 power at 30-hour period) was applied to these hourly positions, and daily average velocities were computed from the filtered times series, and assigned to the daily mean location of the 24-hour track. An array of weather buoys (Fig. 4) along the west coast of British Columbia provide hourly wind measurements at 3.7-m height for these coastal waters (Cherniawsky and Crawford, 1996). Data from buoy C46181 near Kitimat, and buoys C46131 and C46146 in the Strait of Georgia do not represent winds in the region considered here, and were excluded from the study. The remaining thirteen buoys in offshore and coastal waters provide reasonable resolution of winds in mid-straits and shelf regions, but cannot resolve the flows around headlands, nor other wind adjustments very near shore in regions such as Cape Scott. Nevertheless, these measurements provide the best representation of winds available in these coastal waters. Hourly 10-m winds were computed at weather buoys as described in Cherniawsky and Crawford (1996). These were averaged over 24 hours to produce daily mean 10-m winds. Each daily mean wind component (x and y) was interpolated onto the daily mean location of the drifter using linear spatial interpolation between weather buoys, with each weighting coefficient given by a Gaussian function exp(–r2), where r is the non-dimensional distance, r2 = (∆x/rx)2 + (∆y/ry)2, ∆x and ∆y are dimensional distances to the weather buoy, and rx, ry are horizontal scales in x, y coordinates. Here, x and y are taken as the cross-shore and alongshore directions (308 rotation with respect to north) and (rx, ry) = (75,150) km respectively. The response of drifters to winds, waves and currents may have changed with the removal of the parachutes. To investigate this matter, drift rates in April and May 1995, when the parachutes were in place, were compared with drift rates in June to August 1995, when parachutes were absent. Figures 5a to 5c present the drifter speed, wind speed and percent speed ratio (drifter speed to wind speed) for the entire period, for daily mean wind speeds >4 m s–1. Near-shore observations were removed by eliminating tracks more than 20 km east of a line drawn through the weather buoys in Hecate Strait and Queen Charlotte Sound (Fig. 4). There are too few observations in April to provide significant results. In May the average speed ratio was 1.8%, whereas this value varied from 2.0 to 3.5% between June and September. Factors contributing to this increase in speed ratio are discussed below. Ocean conditions may have contributed to this change in response between spring and summer. A surface mixed layer was either weak or absent in spring. Thus, a portion of these stronger near-surface speeds in summer could be explained by a faster surface Ekman drift in a shallow surface mixed layer. Also, without the parachutes the drifters are likely to float closer to the surface, where both Ekman drift and Stokes drift due to wind waves are stronger. There have been no studies of the response of this type of drifter without a para-


Fig. 4

Network of weather buoys off the West Coast of Canada (reproduced from Cherniawsky and Crawford, 1996.)

chute, but it is somewhat similar to the Davis-type of surface drifter (Davis, 1985). The latter has four square vanes, each 90 cm high, with the top of each vane at 30 cm below the surface. The mid-drogue point of the Davis-type drifter therefore rides at about 50 cm further below the ocean surface than do our drifters. An example of how much shear can be expected in the upper 50 cm of the ocean is provided by a study of currents within 35 km of the California coast, by Fernandez et al. (1996), using a single HF radar beam at several frequencies. This study found that the near-surface currents rotate clockwise by 12 to 23 degrees over depth bins centred between 31- and 50-cm depth. In their study, lack of simultaneous wind measurements prevented a calculation of absolute rotation angle relative to the wind itself. Davis (1985) concludes that the Davis-type drifters “are accurate to about 3 cm s–1 even under strong wind conditions.” There may be a shear of about 15 cm s–1 between surface and mid-vane of the Davis-type drifter. A study by Allen (1995) finds a slippage of 0.6 cm s–1 in a wind of 5 m s–1 for a slightly smaller version of the Davistype drifter. Other factors may also be significant. For example, Mackas et al. (1989) report similar motion for two very different designs of drifters, drogued at 10-m depth


Fig. 5

Daily average drifter and wind speeds (a to b) and directions (d to e) for Queen Charlotte Sound, spring and summer 1995. Straight lines connect monthly averages of daily values. Thick (thin) bars represent standard errors (deviations) of monthly means. (c) Speed ratio between drifter and winds. (f) Direction difference between drifter and winds.

264 / William R. Crawford et al. and deployed close together in the southern portion of our study region. In their case, a Tristar drifter with very low windage stayed close to a Loran-C drifter with much higher windage, yet the wind speeds were up to 15 m s–1. They attribute this behaviour to a strong tendency of drifters to follow fronts in coastal waters. We therefore acknowledge that the presence of current shear and Stokes drift in the top metre of the ocean may cause the drifter behaviour to change with the removal of parachutes. In either case the drifters are an accurate measure of currents, although at different depths. Figures 5d to 5f present the wind and drifter directions, as well as the differences between the two. Despite high variability in absolute wind and drifter directions, monthly mean direction differences, plotted in Figure 5f, were almost always in the quadrant between 0 and –90 degrees, indicating drifter motion to the right of the wind. A different presentation of the data (422 drifter-days between April and September) is shown in Fig. 6a, where drifter speeds are plotted as a function of wind speeds. We fit both linear and quadratic functions to these data. The linear fit is for speeds >6 m s–1; the quadratic is for speeds >4 m s–1. A visual comparison indicates that both functions represent the data well for daily mean wind speed between 6 and 12 m s–1. The rms errors for linear and quadratic functions are 1.46 and 1.15 cm s–1 respectively in the speed range of 6 to 12 m s–1. The slope of the linear function shows a drifter speed of 2.4% of wind speed, but this slope increases to 2.9%, with an offset of –5 cm s–1, if the lower limit of wind speed is increased from 6 to 7 m s–1. Figure 6b presents drifter directions plotted against the wind directions for daily mean wind speeds between 4 and 12 m s–1. The mean direction is 29 degrees. The best fit straight line to observations at wind speeds greater than 6 m s–1 has a slope of 1.10, and therefore differs by 10% from uniform offset at these wind speeds. A separate analysis of the 384 drifter-days in summer, when the coastal waters were stratified and the parachutes were not attached to drifters, reveals insignificant changes in these factors. For the linear case, there is a mean direction difference of 31 degrees to the right of the wind, and a slope of 1.08. There are only 38 drifterdays of data during spring (April and May) and the scatter in these data is too large to determine average drift angles and speed ratios. Thus we cannot determine from these data above a significant difference in drift angle between stratified and unstratified conditions, or with parachutes and without parachutes. It is clear, however, that the drifters veer on average about 30 degrees to the right of the wind. Niiler and Paduan (1995) used drifters in the Northeast Pacific with mid-drogue depth of 15 m. In the 15- to 20-day band, they observed that a typical drifter moved at 0.5% of the wind speed, and at 65 to 85 degrees to the right of the wind. Longerperiod motion was influenced by mesoscale eddies, and shorter-period motion was eliminated by their tidal and inertial filter. The difference in behaviour between their drifters and ours can be attributed to the greater drogue depth on their drifters. 3 Current Simulations Our objective is to evaluate predicted currents based on numerical model simula-


Fig. 6

(a) Daily average drifter speeds plotted versus daily average winds speeds. Long-dashed line represents y1, the linear fit for wind speeds greater than 6 m s–1. Short-dashed line represents y2, the quadratic fit for wind speeds greater than 4 m s–1. Coefficients for these least-squares fits are in lower right. (b) Daily average drifter directions plotted versus daily average wind directions. Thin line represents no rotation angle. Thick line represents linear fit for wind speeds between 4 and 12 m s–1. Coefficients for linear fit are in lower right.

266 / William R. Crawford et al. tions of flow in these waters. These simulated currents are compared with observations provided by the drifters as noted above. Numerical simulations are provided by two models, a steady state, baroclinic diagnostic model, based on prevailing water properties, and a baroclinic, time-stepping, wind-forced model. Both provide Eulerian currents for these waters and although they are not directly comparable with the Lagrangian-based currents computed from drifter measurements, the small differences due to Stokes drift of tidal currents are not expected to be a significant problem (Foreman et al., 1992a). a Diagnostic Baroclinic Model To compute baroclinic currents, the steady state, baroclinic model Fundy5 (Lynch et al., 1992; Naimie and Lynch, 1993) was applied in a diagnostic mode. The overall procedure is described in Namie et al. (1994), Hannah et al. (1996), Han et al. (1997) and Loder et al. (1977). This model solves the linearized shallow water equations for a specified baroclinic pressure field and boundary conditions, using hydrostatic and Boussinesq approximations, constant vertical eddy viscosity, and linearized bottom friction. The baroclinic pressure field is computed from local density measurements. In this application, for water depths greater than 50 m, the vertical mesh has 21 unequally spaced nodes, with a minimum 2.5 m spacing near surface and bottom, and larger spacing in the interior. Levels are equally spaced in shallower water. Our approach uses the method and software developed by CCHB96, as applied to the waters of northern Hecate Strait and Dixon Entrance. This application requires historical density data that are computed from temperature and salinity profiles, archived by the Marine Environmental Data Service (MEDS) in Ottawa and at the Institute of Ocean Sciences in Sidney, B.C. These archives include observations from the three drifter deployment cruises to this region during spring and summer of 1995. Details of quality control of temperature and salinity data, and on preparation of temperature and salinity surfaces for calculation of dynamic heights for input to the model, are presented in the Appendix. Forcing in the diagnostic model has two components: the baroclinic pressure gradient, computed from the specified density field, and the surface elevation specified along the open boundary. Model domain is shown in Fig. 7, along with locations of water property profiles used to construct the summer mean density field. The baroclinic pressure gradient was computed on the same level surfaces that the density was estimated on. Surface elevation boundary conditions were specified along the southern and southeastern open boundaries (between points B and C in Fig. 7). The elevations were estimated from the density field along this boundary (the steric elevations of Loder et al., 1997), constrained to yield no normal geostrophic flow at the sea floor. A geostrophic outflow condition (Naimie and Lynch, 1993) was used along the open ocean boundary from A to B (Fig. 7), while a no-normal condition was used along the coastal boundary between points A and C. Figure 8 presents summer surface currents in this model, a time of year of prevailing northwest winds. Currents generally flow alongshore near the mainland coast


Fig. 7

Location of all historical summer temperature and salinity profiles used with the diagnostic model Fundy5. Only locations with both temperature and salinity measurements are plotted.

and Vancouver Island. Current vectors near Cape St. James (in the northwest corner of the plot) form a somewhat confusing pattern and are unlikely to be realistic. It is difficult to select the correlation scales in this region, where currents vary in direction over a distance of several kilometres. A more realistic pattern of currents for the Cape St. James area is presented in Crawford et al. (1995) based on a similar diagnostic model, but using Conductivity-Temperature-Depth (CTD) observations from a single cruise to this region in 1990. The general flow of currents for the Cape St. James region is described and modelled by Thompson and Wilson (1987). b Prognostic Baroclinic Model The second model used here is that of Cummins and Oey (1997), who apply the Princeton Ocean Model (Blumberg and Mellor, 1987; Oey and Chen, 1992) to the same region, but on a regular 5-km grid that is rotated 30 degrees toward the NW. This model solves finite difference analogs of the primitive equations, using the Mellor and Yamada (1982) level-2.5 turbulent closure scheme. There are 21 sigma levels in the vertical, with thinner levels near surface and bottom. Thickness of the


Fig. 8

Surface current field of diagnostic model Fundy5.

top layer is 0.15% of water depth, which is 0.15 m in 100 m of water. Tidal forcing is specified on the boundaries for constituents M2, S2, K1 and O1. The model is initialized with horizontally uniform stratification typical of the shelf break in Queen Charlotte Sound in summer. Surface temperature and salinity were relaxed to initial values over a 20-day timescale. A 5-day relaxation time was also tested, but this change had little effect on the surface current field. The forcing field for this simulation is the daily-averaged wind stress from the weather buoys (Fig. 4), spatially interpolated onto the model grid using the algorithms described above for direct comparison of wind and drifter motion, and linearly interpolated in time to match the model time step. The top model layer represents surface flows, not including Stokes drift due to wind waves. Figure 9a shows an example of a 5-day average of the spatially interpolated and gridded wind stress over the period starting at 1200 utc July 16 and ending at 1200 utc July 21. Figure 9b presents the POM surface currents averaged over this same period. The flows are generally along-shore near the mainland coast of Queen Charlotte Sound and along the northwest coast of Vancouver Island. However, surface flow over Cook Bank (the region directly north of Vancouver Island) is directed more toward shore than for the diagnostic case (Fig. 8).


a

Fig. 9

(a) Average wind stress field during the 5-day period beginning 1200 utc 16 July 1995. (b) Average surface currents simulated by the Princeton Ocean Model during this period.


b

Fig. 9

(Concluded)

Surface Currents in British Columbia Coastal Waters / 271 4 Analysis This section presents a description of statistical relations among winds, the two numerical simulations of surface currents and the daily average drifter tracks. Each drifter track contributes a minimum of five drifter-days of currents to a database. The velocities of each drifter are denoted here as a complex time series of the form Wd = ud + ivd, where ud and vd are eastward and northward speeds. Model currents, from either Fundy5 or POM, were interpolated to the drifter locations and are denoted as Wm = um + ivm. (POM currents were averaged in the same fashion as winds and observed currents; Fundy5 currents are steady state and need no averaging.) Direct wind forcing was provided by the weather buoy time series, as described above. Using hourly wind data from all buoys, we interpolated a daily mean wind at each drifter point and time, denoted Ww = uw + ivw. These three time series of velocities were combined into N complex equations of the form Wd = CwWw + CmWm + ε

(1)

where N is the number of drifter-days of data. Cw and Cm are complex coefficients of a least-squares fit between wind-driven response plus a model current, and the observed drifter vectors. The symbol ε denotes a residual current not explained by this fit, including an average bias (