Low-level jets, orographic effects, and extreme. events in Nares Strait: a model-based mesoscale. climatology

Low-level jets, orographic effects, and extreme events in Nares Strait: a model-based mesoscale climatology R. M. Samelson and P. L. Barbour COAS, 14 COAS Admin Bldg Oregon State University Corvallis, OR 97331-553 USA rsamelson@coas.oregonstate.edu Submitted to Monthly Weather Review March 11, 28

Abstract A mesoscale atmospheric model, nested in operational global numerical weather prediction fields, is used to estimate low-level winds and surface wind stress through Nares Strait, between Ellesmere Island and Greenland, during two years from August 23 through July 25. During most of the year, the model low-level winds are dominated by intense, southward along-strait flow, with monthly mean southward 1-m winds reaching 1 m s 1 in winter. Summertime flow is weak, and distributions of hourly along-strait winds during the two year period are strongly bimodal. The strong southward low-level winds are associated with ageostrophic, orographically channeled flow down the pressure gradient from the Lincoln Sea to Baffin Bay, and are highly correlated with the pressure difference along Nares Strait. Two-year means and leading EOFs of monthly mean 1-m winds and wind stress place the strongest winds and stress in the southern parts of Smith Sound and of Kennedy Channel, at the openings to Baffin Bay and Kane Basin, at known sites of polynya formation, including the North Water polynya in Smith Sound, suggesting that the locally intensified winds may cause these persistent polynyas. An intense wind event observed in Nares Strait by a field camp, with surface winds exceeding 3 m s 1, generally follows the typical pattern of these low-level flows. Based on the model correlation of winds and pressure difference, a 51-year time series of estimated winds in Nares Strait is reconstructed from historical surface pressure measurements at Thule (Pituffik), Greenland, and Alert, Canada. The pressure difference and reconstructed wind time series are correlated with the Arctic Oscillation at annual and longer periods, but not on monthly periods. 1

1 Introduction Nares Strait, a channel roughly 5 km long and 4 km wide, bounded by steep topography on both sides, connects the Lincoln Sea of the Arctic Ocean with Baffin Bay and the northern Labrador Sea, and divides the northwest coast of Greenland from Ellesmere Island in the Canadian Arctic Archipelago (CAA). The narrow, steep channel and frequent large alongstrait pressure gradients combine to generate a unique meteorological environment that is predisposed toward extreme, persistent wind events. These winds may play an important role in the ice and freshwater balance of the Arctic Ocean, a balance that, through its influence on the maintenance of perennial Arctic pack ice, is a critical element of the global climate system. Nares Strait is the the main eastern channel of the CAA, and while little is known about the control on freshwater and ice flux through it, these fluxes may be a significant component of the corresponding total Arctic-Subarctic fluxes (Melling, 2; Melling et al., 27; see also: Agnew, 1998; Kwok, 25). Samelson et al. (26) have presented evidence that local atmospheric forcing can control the motion of sea-ice through Nares Strait. Unfortunately, because Nares Strait is a remote and logistically challenging region, no long-term, in-situ measurements of meteorological variables in the strait are available. Ito (1982) reports on the most extensive such in-situ measurements, which were made in the southern part of Kane Basin during one month in late spring 1975. In the present study, a regional mesoscale atmospheric model (Section 2) is instead used to simulate local meteorological conditions during August 23 through July 25, in support of a wider program, which 2

includes oceanographic moorings and remote-sensing analysis, to measure fluxes of sea-ice and freshwater through the strait. From these model results, a climatology of mesoscale winds and surface pressure is obtained (Section 3). In addition, a case study is presented of an extreme event that destroyed a field camp in April 25 (Section 4). The wind record is extended to a 51-year year time series using a combination of model results and long-term in-situ surface pressure measurements from regional meteorological observing stations (Section 5). 2 Regional atmospheric model The Polar MM5 (Bromwich et al., 21) version of the Pennsylvania State University/National Center for Atmospheric Research MM5 mesoscale atmospheric model was used in the present study. This is a non-hydrostatic, primitive-equation, terrain-following model with full moist physics, which has been optimized for the polar environment. The Polar MM5 has previously been used successfully to simulate meteorological conditions in both polar regions (e.g., Cassano et al., 21; Guo et al., 23). As described by Samelson et al. (26), it was implemented here in a triply-nested configuration, with 54- km, 18-km, and 6-km grid resolutions in the outer, intermediate, and inner (Fig. 1) nests, respectively. It was run in a daily 36-hr forecast mode, with initial and time-dependent lateral outer-nest and surface boundary conditions from the operational global National Center for Environmental Prediction Global Forecast System (GFS) model. The GFS model can be expected to provide an accurate estimate of large-scale atmospheric conditions on these timescales. For example, for the nested Polar MM5 model in the Nares Strait region, Samel- 3

son et al. (26) report correlations of observed and modeled hourly surface pressure at Alert (Fig. 1) of.98. Sea-ice is treated as described in Bromwich et al. (21; note that the seasurface temperature threshold for ice cover is 271.4 K rather than the stated 271.7 K): the GFS surface boundary conditions include a surface temperature field that, over water, determines whether sea-ice is presumed to be present or not at each point; if it is present, then a climatology is used to estimate the local fraction of sea-ice cover. In the region of interest, radiational cooling and surface fluxes cause frequent development of a stable planetary boundary layer, in which wind direction and speed are strongly affected by the steep coastal orography. Similar mesoscale models, nested in operational global atmospheric forecast models, have previously been shown to be capable of reproducing observed low-level wind structure arising through related dynamical processes at mid-latitudes, involving the interaction of a stable lower atmosphere with coastal orography along the U.S. west coast (e.g., Perlin et al., 24). As with Samelson et al. (26), it is presumed here that the Nares Strait regional implementation of the Polar MM5 will reproduce similar orographic effects in this region with similar accuracy. Previous successful MM5 simulations of flow through a narrow channel with steep sides include those of Colle and Mass (2). For this analysis, hourly model fields from daily simulations during August 23 through July 25 were concatenated into a single continuous time series covering the twoyear period. The two-year model climatology developed below is based on this two-year model time series. 4

3 Two-year climatology The dominant low-level flow pattern in the Nares Strait region during most of the August 23 through July 25 period consists of strong southward flow down the pressure gradient from the Lincoln Sea to Baffin Bay. The monthly mean 1-m wind speed, surface stress, and sea-level pressure during January 25 are representative of this pattern (Fig. 2). The mean January 25 pressure difference from the Lincoln Sea to Baffin Bay is roughly 8 hpa, and the mean 1-m wind and wind stress reach roughly 1 m s 1 and.5 N m 2 at their respective maxima in Smith Sound. Similar winds and stresses are found midway through Nares Strait, in the southern end of Kennedy Channel and the northern part of Kane Basin. The direction of the wind and wind stress throughout Nares Strait is nearly parallel to the channel. The general pattern of strong along-strait flow is consistent with the previous in-situ measurements in Kane Basin during April-May 1975 (Ito, 1982). The steep orography bounding Nares Strait, with topographic heights of 5-2 m within a few tens of kilometers or less of the coastline (Fig. 1), severely constricts the lowlevel flow. Consequently, the along-strait flow within the strait is strongly ageostrophic, with flow down the pressure gradient from the Lincoln Sea to Smith Sound. Sea-level pressure isobars in the strait are aligned at an oblique angle to the along-strait direction, consistent with approximate geostrophy in the cross-channel momentum balance. The maximum wind speed and stress are found where the channel widens into Kane Basin and Baffin Bay, in regions of intensified mesoscale pressure gradients. This pattern is characteristic of stable marine-layer flows past orography found in other coastal regions, in which mesoscale pressure gradients 5

are generated by shallowing of the stable boundary layer as the flow spreads to fill the wider downstream region (e.g., Winant et al., 1988; Samelson, 1992; Samelson and Lentz, 1994; Burk et al., 1999), leading to local ageostrophic acceleration of the flow; friction in the shallow downstream region limits the downstream extent of the intensified low-level winds. The southern Kennedy Channel location is of particular interest, as it is the main site for the moored oceanographic measurements of ocean currents and sea-water properties (Münchow et al., 25), in support of which the mesocale atmospheric modeling discussed here was undertaken. Frequency distributions of hourly low-level winds at the southern end of Kennedy Channel (model grid indices i=37, j=83) are bimodal (Fig. 3), with one peak centered at a strong southward flow regime, similar to the January 25 monthly mean, and one peak centered at weak or slightly northward flow. The weak flow regime dominates during the summer months, and the strong flow regime dominates during the remainder of the year. In contrast to the bimodality within the channel, the distribution of 1-m winds in the Lincoln Sea, away from the Nares Strait channel, is unimodal and centered around zero flow, with standard deviation of only 3.8 m s 1, and extreme values of ± 1 m s 1 (Fig. 3a, dash-dot line). The core of the wintertime southward jet in this region is located near 3 m height, where monthly mean winds are roughly twice 1-m values (Fig. 4). The bimodality of the hourly wind distribution is particularly pronounced at the lowest levels, where the orographic channeling is strongest, but is still apparent at the height of the low-level jet core. Approximately 7% of the hourly wind values at the 268-m model level at the southern Kennedy 6

Channel location exceed 25 m s 1 southward (Fig. 3). Potential temperature profiles indicate that the lower atmosphere is stably stratified in both January and July (Fig. 4). Even in January, with the intense mean southward flow centered near 3 m altitude, there is no indication in the model profiles of the formation of a substantial turbulent mixed-layer adjacent to the surface boundary, either in potential temperature or momentum. Evidently, the combination of strong surface cooling and small surface roughness is sufficient to maintain stability despite the strong vertical shear of the low-level jet. Cross-channel cross-sections of the mean and standard deviation of along-strait winds at the southern Kennedy Channel location during January and July 25 (Fig. 5) show the orographic channeling clearly, as the intense flows occur adjacent to the steep orography of Ellesmere Island, on the west side of the channel. Standard deviations in both seasons are comparable to the means, consistent with distributions (Fig. 3) showing hourly wind speeds that frequently reach twice the means. The channeling occurs in summer as well as winter, though wind speeds are much smaller in summer. Comparison of the mean January and July 25 cross-sections in (Fig. 5) with the profile of mean January and July 24-25 winds in (Fig. 4) illustrates the several m s 1 variability of the monthly mean flows from year to year. The two-year mean 1-m wind field (Fig. 6) has a structure similar to that of the January 25 mean: strong southward, along-strait winds, with speed maxima in Smith Sound and at the southern end of Kennedy Channel. The first EOF of the monthly mean fields, computed from fields that were constructed, for computational convenience, from every third 7

model grid point in each direction (and neglecting all points over land), captures 57% of the total variance in the corresponding monthly mean fields, and has a similar pattern to the mean (Fig. 6). The product of the amplitude (7) and EOF, added to the mean, describe monthly mean winds that fluctuate seasonally from weak winds to winds near twice the mean, as might be anticipated from the frequency distributions described above. There is an indication in the two-year mean of persistent, localized offshore winds from Greenland into Baffin Bay, perhaps due to katabatic flow, off the southern side of the Hayes Peninsula (Fig. 6). The two-year mean SLP field (Fig. 8) has a structure that is also similar to that of the January 25 mean: a 6 hpa gradient from high pressure in the Lincoln Sea to low pressure in Baffin Bay. Because the SLP fields are dominated by synoptic variability, most of the total variance in monthly mean SLP is captured in the first EOF, which has a nearly uniform spatial structure. The second EOF of SLP (Fig. 8), which has a structure similar to the mean, describes the mesoscale variability associated with the orographic channeling, and the corresponding amplitude time series is strongly anti-correlated with the first EOF of the 1-m winds (Fig. 7). The ratio of the second SLP and first 1-m wind EOF amplitudes can be compared with the linear regression of 1-m wind at the southern end of Kennedy Channel on the alongstrait pressure difference computed by Samelson et al. (26). This ratio is.15-.2 (m s 1 ) Pa 1, somewhat larger than the regression slope. The direct regression should give the more accurate value, since the wind and pressure EOFs are independent and may describe the corresponding fluctuations with different efficiencies. In the Lincoln Sea and northern Baffin Bay, the two-year mean 1-m winds are aligned 8

approximately along isobars, and so are nearly geostrophic (Figs. 6,8). The along-strait momentum balance of the orographically controlled flow within Nares Strait, however, is clearly ageostrophic, even in this two-year mean, with flow down the pressure gradient from the Lincoln Sea to Smith Sound. Comparison of the strongly anti-correlated (Fig. 7) first 1-m wind EOF (Fig. 6) and second SLP EOF (Fig. 8) shows that the monthly mean 1-m winds have similar dynamics, with strongly ageostrophic flow down the pressure gradient through the strait. In the Lincoln Sea and northern Baffin Bay, the monthly-mean 1-m flow appears to be less geostrophic than the two-year mean, with noticeably more cross-isobar flow. The two-year mean stress and the first EOF of monthly mean stress have structures that are similar to those of the 1-m wind, but with more intense concentration of the largest stress in the southern Kennedy Channel and Smith Sound regions (Fig. 9). These local stress maxima are four to five times larger than the stresses in the Lincoln Sea and Baffin Bay regions, and in the eastern and southern parts of Kane Basin. The stress variability, both spatially and temporally, is controlled primarily by variations in the 1-m winds. The spatial patterns of winds and stress are similar; the greater concentration of extreme stresses, relative to extreme winds, is due to the roughly quadratic dependence of the parameterized surface stress on the model 1-m wind. The temporal variations are also similar: the first stress EOF amplitude time series describes 74% of the monthly-mean variance is strongly correlated with the first 1-m wind EOF amplitude time series and, like the wind EOF time series, strongly anti-correlated with the second SLP EOF amplitude time series (Fig. 7). 9

4 Kennedy Channel wind event of 13-14 April 25 An intense wind event occurred in Nares Strait on 13-14 April 25. This event is of particular interest because it destroyed a scientific field camp in Lafayette Bay, on the eastern shore of the south end of Kennedy Channel. This camp was the first winter camp known to have been established in Kennedy Channel. The following quotation is taken from the informal report (Falkner et al., 25) on the field effort: The eight residents of the Lafayette camp...were wakened at about 4: a.m. [9 UTC] on April 13 by brief intense bursts of wind interspersed with long (15 minutes) intervals of calm...by mid-morning the average speed of wind had reached 2 kt, with extreme gustiness... Winds increased though the day. In late afternoon, [a 12 ft x 2 ft] storage tent came down in wind gusts exceeding 5 kt. One octagonal tent...followed soon after. Many camp items streamed downwind...personnel mustered in one of the remaining four [octagonal tents] to hold it down. As winds strengthened, two more were blown away... These extreme winds were not associated with weather excepting a little thin cloud and snow. The air pressure was steady... By mid morning of April 14, the wind had decreased to 35-4 kt and was much less gusty... At the climax of the event, the average wind exceeded 45-5 kt and gusts were off the top of the scale at 6 kt. Winds of this strength have never been encountered by DRDC [Defence Research Development Corporation] personnel who have worked in the vicinity of Alert (15 miles to the north of Lafayette Bay) for more than 3 years. Examination of large-scale pressure fields and mesoscale model simulations during this period indicate that this event generally followed the pattern of intense, orographically 1

controlled, ageostrophic, down-gradient low-level flow through Nares Strait that is evident in the means and EOFs described above. Sea level pressure fields from the operational National Center for Environmental Prediction (NCEP) global GFS forecast model during 12 UTC 13 April through 12 UTC 14 April show a surface low developing and moving into the Labrador Sea, with a surface high over the western Arctic Ocean (Fig. 1). At 12 UTC on 14 April, the GFS fields show a Labrador Sea low below 976 hpa, and an Arctic high above 132 hpa, a difference of roughly 6 hpa between these regions (Fig. 1). The corresponding pressure difference across the length of Nares Strait, which drives the low-level jet along the strait, was approximately 16 hpa (Fig. 11). This is a large but not extreme value in the context of the simulated record (see, e.g., Fig. 16 below). The model 1- m winds at the southern Kennedy Channel location during this period were 1-15 m s 1. The 13-14 April 25 event appears to have been unusual in the extent to which intense low-level flow was found over the Greenland landmass east of Kennedy Channel. Wind speeds over 2 m s 1 were found in a broad area over 7 km wide, above 2 m altitude and extending east of the strait, with maximum winds over 35 m s 1 in a shallow jet 3 km east of the strait (Fig. 12). Although the deep region of strong southward flow was still centered over the strait, this structure differs somewhat from the typical jet, which is more narrowly confined to the strait (Fig. 5). The eastward extension in this event may be related to the tongue of low pressure that extended northward over Greenland from the surface low in the Labrador Sea (Fig. 1). The extreme surface winds observed at the field camp, which were comparable to the peak low-level jet speeds in the model, likely occurred when the low-level jet touched the 11

surface, in response to small-scale orographic forcing or as the result of internal instabilities of the jet. The 1-15 m s 1 model surface winds during this event are typical for the winter season, suggesting that the conditions similar to those observed at the field camp probably occur frequently in Nares Strait. 5 A 51-year reconstructed Nares Strait wind time series 5.1 Observed pressure time series The analysis of the two-year model climatology described in the preceding section indicates that there is a robust relation between the SLP gradient and the alongstrait winds in Nares Strait. Samelson et al. (26) previously presented evidence for such a relation, based on analysis of a short subset of the two-year model record. The existence of this relation makes possible a reconstruction of long time series of estimated winds in Nares Strait from historical pressure records that are available from regional meteorological stations. For this reconstruction, 3-hourly SLP observations were obtained for the period January 1955 through December 25 from meteorological stations at Alert, on the southern coast of the Lincoln Sea, roughly 2 km west of the northern end of Nares Strait, and Thule (Pituffik weather station, Thule Air Force Base) on the west coast of Greenland, roughly 2 km southeast of the southern end of Nares Strait at Smith Sound (Fig. 1). The Alert and Thule SLP records were obtained, respectively, from Environment Canada and the National Climate Data Center. A time series of Alert-Thule SLP pressure differences (Fig. 13) was constructed 12

by differencing each pair of values from these two records at each time. The time series were essentially continuous except for several lengthy gaps, during the years 1971, 1972, 1992, and 1993, and the months October 1983 through September 1984, September through December 1991, and January 1994, which were missing. The seasonal cycle of monthly means was computed by averaging the monthly means for months with more than 15 3-hourly observations; binning all observations by month and averaging changes the values by typically less than 1%. The resulting seasonal cycle with amplitude 6.3 hpa and mean difference 4.7 hpa (Figs. 14,15). Year-to-year variability of monthly means has standard deviations of 2-3 hpa. The largest monthly mean pressure difference was 14.9 hpa during December 1998. The SLP time series at both Alert and Thule, and their monthly-mean departures from the seasonal cycle, are correlated with time series of the Arctic Oscillation (AO) and North Atlantic Oscillation (NAO) indices, obtained from the on-line databases managed by the Climate Prediction Center of the National Oceanic and Atmospheric Administration. However, neither the monthly mean SLP difference time series nor the time series of departures of the monthly mean SLP differences from the seasonal cycle is correlated with either the AO or the NAO indices (Table 1). On longer time scales, a correlation of SLP difference with the AO index does emerge (Fig. 15): for the annual means, the correlation is -.34, and for three-year running averages of the annual means, the correlation is -.57 (Table 1). This increased correlation with the AO index on the longer timescales suggests that large-scale variability has substantial influence on mesoscale conditions at annual and longer timescales, while the relative influence of regional variability increases for timescales shorter than annual. The standard 13

deviation of the annual means is roughly 1 hpa (Fig. 15), substantially smaller than the 51- year record mean SLP difference of 4.7 hpa. For all of the these correlation computations, the monthly mean values for months with fewer than 15 3-hourly observations were replaced by the corresponding seasonal cycle values. 5.2 Reconstructed wind time series Samelson et al. (26) noted that the daily-averaged along-strait 1-m wind at the southern Kennedy channel location was highly correlated with the sea-level pressure difference between Alert and Thule during the two periods examined in 25, with an approximately linear dependence of wind on pressure difference. For the two-year time series of hourly 1- m winds considered here, a qualitatively similar correlation holds (Fig. 16). Due to the more sensitive dependence of wind speed on pressure difference for moderate positive differences than for large or negative differences, the simple linear regressions considered by Samelson et al. (26) can be improved by considering a piecewise linear regression on three intervals. The intervals and the corresponding regression relations were found iteratively as follows. First, two appropriate first-guess interval boundaries were chosen, P a and P b, for example, P a = hpa and P b = 1 hpa. Linear regressions were performed separately in each of the three intervals P < P a, P a < P < P b, and P > P b, yielding the regression relations: a 1 + b 1 P, P < P a, V 1 = a 2 + b 2 P, P a < P < P b, (5.1) a 3 + b 3 P, P > P b, 14

Here V 1 is the 1-m wind in units of m s 1, and P is the Alert-Thule SLP difference in units of Pa. For these regressions, the two intersection points near P a and P b were then computed, and P a and P b were replaced with the corresponding P values of these intersection points. The procedure was repeated until the values of P a and P b converged to constants. This iteration was found to converge to a unique solution P a =.66 hpa and P b = 7.38 hpa for all first-guess values considered. For this solution, with the above units, a 1 =.41,b 1 =.73,a 2 =.5,b 2 = 1.29,a 3 = 5.75, and b 3 =.5. The corresponding continuous piecewise-linear regression fit (Fig. 16) reproduces the more sensitive dependence of wind speed on pressure difference for moderate positive differences than for large or negative differences noted above. The residual variance is reduced by the piecewise linear fit, relative to the simple linear regression, by the ratio.94. An important additional advantage of the piecewise fit is the more accurate representation of extreme events, the amplitudes of which are systematically overestimated by the simple linear regression. From the linear regression relations (5.1), with constants from the iterative solution described above, a 51-year time series of reconstructed 3-hourly 1-m along-strait winds at the southern Kennedy Channel location was obtained from the Alert-Thule SLP pressure difference time series. The linear regression relation gives roughly 1 m s 1 of southward wind for each 1 hpa of greater pressure at Alert relative to Thule. For pressure differences of large magnitude, the relative increase in wind speed with pressure difference is smaller. Since the piecewise-linear fit is non-linear, the mean and anomaly statistics of the resulting wind series, discussed below, were computed directly rather than from the statistics of the Alert-Thule SLP 15

difference series. The seasonal cycle of monthly mean winds from this 51-year reconstructed time series has an amplitude of roughly 5 m s 1, with minimum and maximum southward winds of 2 m s 1 in July and 7 m s 1 in November and January, respectively (Fig. 17a). The cycle is asymmetric, with a rapid onset of strong northward winds between August and October, and a more gradual decline toward weak winds from March to June. December pressure differences and reconstructed winds are slightly smaller in magnitude than the corresponding means for November and January. A seasonal cycle of wind profiles can be computed from this 51-year wind reconstruction by scaling the monthly mean model profiles for August 23 through July 25 by the ratios of the monthly mean reconstructed 1-m winds during 1955-25 to the corresponding monthly mean model 1-m winds. The resulting cycle of profiles shows monthly mean southward winds in excess of 1 m s 1 at 2 m height from November through February, with the 1 m s 1 contour extending from 15 m to 7 m in January (Fig. 17b). Individual monthly mean wind values during the 51-year period varied by as much as ±5 m s 1, but no northward monthly means were recorded during the months of October through February. The maximum reconstructed southward monthly mean wind was 12 m s 1 during February 1982. The monthly mean winds for the August 23 through July 25 period are slightly smaller than, or equal to, the long-term means (Fig. 17). The frequency distribution of the 3-hourly reconstructed wind values has a bimodal structure (Fig. 18). It is not as clearly defined as the bimodality of the hourly model winds discussed above (Fig. 3). There are peaks in the frequency distribution near 1 m s 1 southward 16

and a secondary peak near 1 m s 1 northward, but the intermediate wind regime between these peaks is relatively more populated for the reconstructed winds than for the model winds. For the individual 3-hourly reconstructed wind values, the southward and northward maxima were 28.2 and 26.1 m s 1, respectively, corresponding in turn to Alert-Thule pressure differences of 44.8 and -35.3 hpa on 14 February 1996 and 18 January 1958. The corresponding extrema for the hourly 1-m model winds at the southern Kennedy Channel location during August 23 through July 25 were 24.1 m s 1 southward and 15.4 m s 1 northward, corresponding to pressure differences of 3.9 and -22.6 hpa, respectively. The 51-year record mean reconstructed wind is 5 m s 1 southward, and the standard deviation of the annual means is roughly 1 m s 1 (Fig. 15). The annual mean reconstructed winds are, of course, highly correlated with the pressure difference time series from which they were derived (Fig. 15), and thus are also correlated on long timescales with the Arctic Oscillation index. 6 Summary A two-year climatology of Nares Strait meteorological variables has been constructed from daily mesoscale model simulations nested in an operational global numerical weather prediction model during August 23 through July 25. The results indicate that Nares Strait is an extreme meteorological environment, with roughly 7% of hourly southward wind values at the peak low-level jet altitude near 3 m exceeding 25 m s 1 in the central region of the channel. Low-level winds in Nares Strait are bimodally distributed, with dominant peaks near 17

1 m s 1 southward during October through June, and near 1 m s 1 northward during July through September. Two-year means and leading EOFs of monthly mean 1-m winds and wind stress place the strongest winds and stress in the southern parts of Smith Sound and of Kennedy Channel, at the openings to Baffin Bay and Kane Basin. These locations are known sites of polynya formation, including especially the North Water in Smith Sound, suggesting that the locally intensified winds may cause these persistent polynyas. The strong southward low-level winds are associated with ageostrophic, orographically channeled flow down the pressure gradient from the Lincoln Sea to Baffin Bay. The alongstrait winds are highly correlated with the pressure difference along Nares Strait. Based on the model correlation of winds and pressure difference, a 51-year time series of estimated winds in Nares Strait is reconstructed from historical surface pressure measurements at Thule, Greenland, and Alert, Canada. Although the Thule and Alert pressure records are correlated with indices of the Arctic Oscillation and the North Atlantic Oscillation, the monthly mean pressure differences and reconstructed winds are not correlated with either of these largescale indices. On annual and longer timescales, a correlation of these pressure differences and reconstructed winds does emerge, suggesting that a portion of the interannual variability of ice and freshwater flux through Nares Strait may be controlled by large-scale atmospheric variability, mediated by the mesoscale dynamics of the low-level winds in the strait. With a previous analysis that provided evidence for wind-forcing of sea-ice motion through Nares Strait (Samelson et al., 26), these results suggest that regional atmospheric dynamics can influence long-term variations in Arctic-subArctic sea-ice and freshwater fluxes. 18

Acknowledgments This research was supported the National Science Foundation, Grant OPP-23354. Useful comments on the manuscript were provided by H. Melling. 19

References Agnew, T. A. Drainage of multiyear sea ice from the Lincoln Sea. Canadian Met. Ocean. Bulletin, 26(4), 11-13, 1998. Bromwich, D.H., J.J. Cassano, T. Klein, G. Heinemann, K.M. Hines, K. Steffen, and J.E. Box, 21: Mesoscale modeling of katabatic winds over Greenland with the Polar MM5. Mon. Wea. Rev., 129, 229-239. Burk, S., T. Haack, and R. M. Samelson, 1999. Mesoscale simulation of supercritical, subcritical, and transcritical flow along coastal topography. Journal of the Atmospheric Sciences, 56, 278-2795. Cassano, J.J., J.E. Box, D.H. Bromwich, L. Li, and K. Steffen, 21. Evaluation of Polar MM5 simulations of Greenland s atmospheric circulation. J. Geophys. Res., 16, 33,867-33,89. Colle, B. A., and C. F. Mass, 2. High resolution observations and numerical simulations of easterly gap flow through the Strait of Juan de Fuca on 9-1 December 1995. Mon. Wea. Rev., 128, 2398-2422. Falkner, K., H. Melling, A. Münchow, R. Samelson, M. Torres, and K.-C. Wong, 25. Report on the spring 25 Nares Strait field effort. Unpublished technical report, available at: http://newark.cms.udel.edu/~cats/healy_25/science/reports.html. Guo, Z., D.H. Bromwich, and J.J. Cassano, 23: Evaluation of Polar MM5 simulations of Antarctic atmospheric circulation. Mon. Wea. Rev., 131, 384-411. 2

Ito, H., 1982. Wind through a channel - surface wind measurements in Smith Sound and Jones Sound in northern Baffin Bay. J. Appl. Met., 21, 153-162. Kwok R., 25. Variability of Nares Strait ice flux, Geophys. Res. Lett., 32, L2452, doi:1.129/25gl24768. Melling, H., 2. Exchanges of freshwater through the shallow straits of the North American Arctic. In, The Freshwater Budget of the Arctic Ocean. NATO/WCRP/AOSB, Kluwer Academic Publications, Amsterdam. 479-52. Proceedings of a WCRP/AOSB/NATO Advanced Research Workshop, Tallinn, Estonia, April 1998. Melling, H., T. Agnew, K. Falkner, D. Greenberg, C. Lee, A. Münchow, B. Petrie, S. Prinsenberg, R. Samelson, and R. Woodgate, 27. Freshwater fluxes via Pacific and Arctic outflows across the Canadian polar shelf. In: Arctic-Subarctic Ocean Fluxes: Defining the role of the Northern Seas in climate, R. Dickson, J. Meincke, and P. Rhines, eds. Springer, to appear. Münchow, A., H. Melling, and K.K. Falkner, 25: Observational estimates of volume and freshwater fluxes leaving the Arctic Ocean through Nares Strait. J. Phys. Oceanogr., submitted. Perlin, N., R. M. Samelson, and D. B. Chelton, 24. Scatterometer and model wind and wind stress in the Oregon - northern California coastal zone. Mon. Wea. Rev., 132, 211-2129. 21

Samelson, R. M., 1992. Supercritical marine layer flow along a smoothly-varying coastline. Journal of the Atmospheric Sciences, 49, 1571-1584. Samelson, R. M., T. Agnew, H. Melling, and A. Münchow, 26. Evidence for atmospheric control of sea-ice motion through Nares Strait. Geophysical Research Letters, 33, L256, doi:1.129/25gl2516. Samelson, R. M., and S. J. Lentz, 1994. The horizontal momentum balance in the marine atmospheric boundary layer during CODE-2. J. Atmos. Sci., 51(24), 3745-3757. Winant, C., C. Dorman, C. Friehe, and R. Beardsley, The marine layer off northern California: An example of supercritical channel flow. J. Atmos. Sci., 45, 3588-365, 1988. 22

Table 1: Correlations of Alert and Thule (Pituffik) SLP, and Alert-Thule SLP difference, with Arctic Oscillation (AO) and North Atlantic Oscillation (NAO) indices during 1955-25, for the time series of monthly means and of departures of monthly means from their 51-year monthly mean values. For the Alert-Thule SLP difference and the AO, the correlations are also shown for the annual means of the SLP differences and the AO index, and for a threeyear running average of the annual means and the AO index. In all cases, missing SLP difference values during 1971-1972 and 1992-1993 were replaced by the corresponding long-term means. 23

SLP record Correlation with AO Correlation with NAO Monthly means Alert -.56 -.5 Thule -.51 -.53 Alert-Thule -.11.3 Departures of monthly means from 51-year means of monthly means Alert -.7 -.6 Thule -.66 -.62 Alert-Thule -.3.7 Annual means of SLP differences Alert-Thule -.34 Three-year running average of annual means of departures Alert-Thule -.57 24

List of Figures 1 Inner-nest (6-km grid) model domain and terrain height (m, contour interval (CI): 5 m). Nares Strait connects the Lincoln Sea of the Arctic Ocean (upper left) with Baffin Bay (lower right), and separates Greenland, to the east (right), from Ellesmere Island, to the west (left). The strait is oriented roughly NNE by SSW, and the model grid is aligned along the channel; in the text, the northward alongstrait direction is parallel to the channel and toward the northern opening into the Lincoln Sea. From north to south, Nares Strait is comprised of Robeson Channel, Hall Basin, Kennedy Channel, Kane Basin, and Smith Sound. The locations of the Alert and Thule (Pituffik) weather stations, and of a model time series in southern Kennedy Channel that is discussed in the text, are indicated (solid circles).................. 3 2 (a) Mean sea-level pressure anomaly (SLP - 1 3 ; hpa, CI: 1 hpa), (b) magnitude of vector mean 1-m wind (m s 1, CI: 2 m s 1 ), and (c) magnitude of vector mean surface stress (N m 2, CI:.1, maximum contour.4), from hourly model values during January 25.................... 31 3 Frequency distribution of (a) 1-m and (b) 268-m hourly northward model winds in 1 m s 1 bins during August 23 through July 25 for the southern Kennedy Channel location (Fig 1; model grid indices i=37, j=83). The distributions are shown for all months (thick solid line), and for the October through June (thin solid) and July-September seasons (dashed). In (a), the 1- m northward wind distribution for all months for a location north of Alert in the Lincoln Sea (grid indices i=3, j=135) is also shown (dash-dot)...... 32 25

4 Vertical profiles of mean (a) along-strait wind speed and (b) potential temperature for January and July 24 and 25, at the southern Kennedy Channel location (Fig. 1). In (b), potential temperatures are shown relative to 1-m values of 245.3 K for January (thick line) and 273.5 K for July (thin)..... 33 5 Cross-channel cross-sections, through the southern Kennedy Channel location (Fig. 1a), of along-strait mean (left) and standard deviation (right) of hourly northward along-strait model winds (m s 1 ) during January (upper) and July (lower) 25. Note that the channel occupies only kilometers 21 through 24; the orography immediately east of the channel is unusually low at this location...................................... 34 6 (a) Mean (m s 1, CI: 1.5 m s 1 ) and (b) EOF #1 magnitude (CI:.2) for 1-m wind, from monthly means during August 23 through July 25. In (a), the contours are of the magnitude of the mean wind speed. In (b), the product of the EOF and the corresponding EOF amplitude (Fig. 7) gives the dimensional physical variability. In both panels, selected vectors, with length proportional to amplitude, are shown to indicate flow direction and structure. In (a), the isolated 1.5 m s 1 contour at lower right is adjacent to the Hayes Peninsula, on the southern shore of the portion of Greenland depicted........... 35 7 EOF amplitudes from monthly means during August 23 through July 25, for (a)1-m wind EOF #1, (b) SLP EOF #2, and (c) wind stress EOF #1. The products of these amplitudes and the corresponding EOFs (Figs. 6b, 8b, 9b) give the dimensional physical variability..................... 36 26

8 (a) Mean sea-level pressure anomaly (SLP - 1 3 ; hpa, CI: 1 hpa) and (b) SLP EOF #2 (CI:.2), from monthly means during August 23 through July 25. In (b), the product of the EOF and the corresponding EOF amplitude (Fig. 7) gives the dimensional physical variability................ 37 9 (a) Mean (N m 2, CI:.5 N m 2 ) and (b) EOF #1 (CI:.2) for wind stress, from monthly means during August 23 through July 25. In (a), the contours are of the magnitude of the mean stress. In (b), the product of the EOF and the corresponding EOF amplitude (Fig. 7) gives the dimensional physical variability. In both panels, selected vectors, with length proportional to amplitude, are shown to indicate flow direction and structure.......... 38 1 Sea-level pressure (hpa) from the NCEP GFS operational forecast model at (upper panel) 12 UTC 13 April 25, (middle) UTC 14 April 25, and (bottom) 12 UTC 14 April 25. Nares Strait is at upper center, with the Arctic Ocean and Greenland at upper left center and upper right center, respectively, and the Labrador Sea and Greenland Sea at center right and upper right (land areas are shaded.). The sequence shows the movement of a developing surface low (L) into the Labrador Sea, with a surface high (H) over the Arctic Ocean, resulting in a strong low-level pressure gradient through Nares Strait....................................... 39 11 Sea level pressure (hpa, contours; CI: 4 hpa, max: 12 hpa, min: 14 hpa) and model 1-m wind speed (barbs, full barb 5 m s 1 ) from the 21- hour mesoscale model forecast valid at 21 UTC 13 April 25. Land areas are shaded.................................... 4 27

12 Cross-channel cross-section, through the southern Kennedy Channel location (Fig. 1a), of model wind speed (m s 1 ) from the 24-hour mesoscale model forecast valid at UTC 14 April 25................... 41 13 Monthly mean Alert-Thule sea level pressure (SLP) differences (hpa) during 1955-25. Means for months with fewer than 15 3-hourly observations are not shown..................................... 42 14 Seasonal cycle of monthly mean Alert-Thule SLP differences (hpa) during 1955-25. The means over 1955-25 for each month are shown (squares), along with the means plus and minus one standard deviation (triangles).... 43 15 Annually averaged (a) Alert-Thule sea level pressure (SLP) differences ( P, hpa) and (b) corresponding reconstructed Kennedy Channel 1-m winds (V 1, m s 1 ) during 1955-25. The reconstructed winds were computed from a piecewise-linear regression model, as described in the text. In (a), the annually averaged SLP differences (thick dashed line) and a three-year running mean of those anomalies (thick solid line) are shown, along with the twice the dimensionless annual Arctic Oscillation index (thin dashed line) and a threeyear running mean of that index (thin solid line). In (a) and (b), the full-record mean SLP difference and reconstructed winds, respectively, are also indicated (dotted lines)................................... 44 16 Hourly 1-m model winds (V 1, m s 1 ) at the southern Kennedy Channel location vs. Alert-Thule sea-level pressure difference ( P, hpa) during August 23 through July 25 (dots), and piecewise-linear regression fit (solid line). The points P a =.66 hpa and P b = 7.38 hpa determining the piecewiselinear fit intervals are indicated (vertical dashed lines)............. 45 28

17 Seasonal cycle of monthly mean reconstructed (a) 1-m and (b) loweratmosphere northward along-strait winds (m s 1 ; (b): CI: 2 m s 1, minimum contour -11.5 m s 1 ) at the grid 83 southern Kennedy Channel location during 1955-25. In (a), the means over 1955-25 for each month are shown (squares), along with the means plus and minus one standard deviation (triangles), and the corresponding monthly mean model 1-m winds during August 23 through July 25 (solid line)....................... 46 18 Frequency distribution of 3-hourly reconstructed 1-m winds during 1955-25 (solid line) and 1-m model winds during August 23 through July 25 (dashed) in 1 m s 1 bins for the southern Kennedy Channel location... 47 29

8 Lincoln Sea Alert Robeson Channel Hall Basin 25 7 Kennedy Channel 2 6 5 Ellesmere Island Greenland 15 4 Kane Basin 5 1 15 1 3 Smith Sound 2 5 1 Thule Baffin Bay 1 2 3 4 5 Figure 1: Inner-nest (6-km grid) model domain and terrain height (m, contour interval (CI): 5 m). Nares Strait connects the Lincoln Sea of the Arctic Ocean (upper left) with Baffin Bay (lower right), and separates Greenland, to the east (right), from Ellesmere Island, to the west (left). The strait is oriented roughly NNE by SSW, and the model grid is aligned along the channel; in the text, the northward alongstrait direction is parallel to the channel and toward the northern opening into the Lincoln Sea. From north to south, Nares Strait is comprised of Robeson Channel, Hall Basin, Kennedy Channel, Kane Basin, and Smith Sound. The locations of the Alert and Thule (Pituffik) weather stations, and of a model time series in southern Kennedy Channel that is discussed in the text, are indicated (solid circles). 3

2 Distance (km) 8 7 6 5 4 3 2 13 9 8 1 a) 25 5 Distance (km) 7 6 SLP 1 3 (hpa) 5 Distance (km) 8 7 6 5 4 3 2 2 1 b) 25 5 Distance (km) 6 8 4 Wind speed (m s 1 ) 2 Distance (km) 8 7 6 5 4 3 2.1.2 1 c) 25 5 Distance (km).1 Wind stress (N m 2 ) Figure 2: (a) Mean sea-level pressure anomaly (SLP - 1 3 ; hpa, CI: 1 hpa), (b) magnitude of vector mean 1-m wind (m s 1, CI: 2 m s 1 ), and (c) magnitude of vector mean surface stress (N m 2, CI:.1, maximum contour.4), from hourly model values during January 25. 31

PDF: fraction/(1 m s 1 bin).12.1.8.6.4.2 a) 4 3 2 1 1 2 3 4 1 m northward wind (m s 1 ) PDF: fraction/(1 m s 1 bin).12.1.8.6.4.2 b) 4 3 2 1 1 2 3 4 268 m northward wind (m s 1 ) Figure 3: Frequency distribution of (a) 1-m and (b) 268-m hourly northward model winds in 1 m s 1 bins during August 23 through July 25 for the southern Kennedy Channel location (Fig 1; model grid indices i=37, j=83). The distributions are shown for all months (thick solid line), and for the October through June (thin solid) and July-September seasons (dashed). In (a), the 1-m northward wind distribution for all months for a location north of Alert in the Lincoln Sea (grid indices i=3, j=135) is also shown (dash-dot). 32

1 8 a) Height (m) 6 4 2 January July 15 1 5 5 Northward wind (m s 1 ) 1 8 b) Height (m) 6 4 January 2 July 5 1 15 Potential temperature relative to 1 m ( o K) Figure 4: Vertical profiles of mean (a) along-strait wind speed and (b) potential temperature for January and July 24 and 25, at the southern Kennedy Channel location (Fig. 1). In (b), potential temperatures are shown relative to 1-m values of 245.3 K for January (thick line) and 273.5 K for July (thin). 33

Figure 5: Cross-channel cross-sections, through the southern Kennedy Channel location (Fig. 1a), of along-strait mean (left) and standard deviation (right) of hourly northward along-strait model winds (m s 1 ) during January (upper) and July (lower) 25. Note that the channel occupies only kilometers 21 through 24; the orography immediately east of the channel is unusually low at this location. 34

Distance (km) 8 7 6 5 4 3 2 1.5 3 6 4.5 1.5 1 a) 25 5 Distance (km) 3 1.5 1.5 Distance (km) 8 7 6 5 4 3 2.4.8 1 b) 25 5 Distance (km).6.4.2 Figure 6: (a) Mean (m s 1, CI: 1.5 m s 1 ) and (b) EOF #1 magnitude (CI:.2) for 1-m wind, from monthly means during August 23 through July 25. In (a), the contours are of the magnitude of the mean wind speed. In (b), the product of the EOF and the corresponding EOF amplitude (Fig. 7) gives the dimensional physical variability. In both panels, selected vectors, with length proportional to amplitude, are shown to indicate flow direction and structure. In (a), the isolated 1.5 m s 1 contour at lower right is adjacent to the Hayes Peninsula, on the southern shore of the portion of Greenland depicted. 35

m s 1 /EOF#1 6 3 a) 3 6 6 12 18 24 Months (from July 23) hpa/eof#2 N m 2 /EOF#1 4 2 b) 2 4 6 12 18 24 Months (from July 23).4.2 c).2.4 6 12 18 24 Months (from July 23) Figure 7: EOF amplitudes from monthly means during August 23 through July 25, for (a)1-m wind EOF #1, (b) SLP EOF #2, and (c) wind stress EOF #1. The products of these amplitudes and the corresponding EOFs (Figs. 6b, 8b, 9b) give the dimensional physical variability. 36

8 16 8.8.6 Distance (km) 6 4 2 14 13 Distance (km) 6 4 2.4 12 a) 25 5 Distance (km).6 b) 25 5 Distance (km).8 Figure 8: (a) Mean sea-level pressure anomaly (SLP - 1 3 ; hpa, CI: 1 hpa) and (b) SLP EOF #2 (CI:.2), from monthly means during August 23 through July 25. In (b), the product of the EOF and the corresponding EOF amplitude (Fig. 7) gives the dimensional physical variability. 37

.5 8 8 Distance (km) 6 4 2.5.2.1 Distance (km) 6 4 2.2.8.6 a) 25 5 Distance (km).4 b) 25 5 Distance (km).2 Figure 9: (a) Mean (N m 2, CI:.5 N m 2 ) and (b) EOF #1 (CI:.2) for wind stress, from monthly means during August 23 through July 25. In (a), the contours are of the magnitude of the mean stress. In (b), the product of the EOF and the corresponding EOF amplitude (Fig. 7) gives the dimensional physical variability. In both panels, selected vectors, with length proportional to amplitude, are shown to indicate flow direction and structure. 38

Figure 1: Sea-level pressure (hpa) from the NCEP GFS operational forecast model at (upper panel) 12 UTC 13 April 25, (middle) UTC 14 April 25, and (bottom) 12 UTC 14 April 25. Nares Strait is at upper center, with the Arctic Ocean and Greenland at upper left center and upper right center, respectively, and the Labrador Sea and Greenland Sea at center right and upper right (land areas are shaded.). The sequence shows the movement of a developing surface low (L) into the Labrador Sea, with a surface high (H) over the Arctic Ocean, resulting in a strong low-level pressure gradient through Nares Strait. 39

9 72 Distance (km) 54 36 18 25 5 Distance (km) Figure 11: Sea level pressure (hpa, contours; CI: 4 hpa, max: 12 hpa, min: 14 hpa) and model 1-m wind speed (barbs, full barb 5 m s 1 ) from the 21-hour mesoscale model forecast valid at 21 UTC 13 April 25. Land areas are shaded. 4

2. Height (km) 1.. 18 21 24 27 Distance (km) Figure 12: Cross-channel cross-section, through the southern Kennedy Channel location (Fig. 1a), of model wind speed (m s 1 ) from the 24-hour mesoscale model forecast valid at UTC 14 April 25. 41