Ice Draft and Current Measurements from the North-Western Barents Sea, 1993–96
Total Page:16
File Type:pdf, Size:1020Kb
Ice draft and current measurements from the north-western Barents Sea, 1993–96 Einar Povl Abrahamsen, Svein Østerhus & Tor Gammelsrød From 1993 to 1996, three oceanographic moorings were deployed in the north-western Barents Sea, each with a current meter and an upward- looking sonar for measuring ice drafts. These yielded three years of cur- rents and two years of ice draft measurements. An interannual variability of almost 1 m was measured in the average ice draft. Causes for this variability are explored, particularly its possible connection to changes in atmospheric circulation patterns. We found that the fl ow of Northern Barents Atlantic-derived Water and the transport of ice from the Central Arctic into the Barents Sea appears to be controlled by winds between Nordaustlandet and Franz Josef Land, which in turn may be infl uenced by larger-scale variations such as the Arctic Oscillation/North Atlantic Oscillation. E. P. Abrahamsen, British Antarctic Survey, Natural Environment Research Council, High Cross, Mading- ley Road, Cambridge CB3 0ET, UK, [email protected]; S. Østerhus, Bjerknes Centre for Climate Research, University of Bergen, Allégaten 55, NO-5007 Bergen, Norway; T. Gammelsrød, Geophysical Institute, Uni- versity of Bergen, Allégaten 70, NO-5007 Bergen, Norway. The Barents Sea plays an important role in the emerge from satellite altimetry observations climate system of the Arctic (Schauer et al. 1997; (Laxon et al. 2003). Maslowski et al. 2004, for example). About half Here we present two years of ice draft obser- of the heat loss to the atmosphere in the entire vations obtained from a mooring at 77° 55' N 28° Nordic seas takes place here (Simonsen & Haugan 20' E equipped with upward looking sonar (ULS). 1996), and the water masses in the Barents Sea are The time series is accompanied by three years of therefore strongly modifi ed by cooling, ice for- current meter data from the same position. The mation and brine release (Midttun 1985). When mooring location is indicated in Fig. 1. To the the modifi ed water leaves the Barents Sea in the authors’ knowledge this is the only multi-season- east, its increased density causes it to enter the al time series of ice thickness in the Barents Sea. Arctic Ocean as an intermediate water mass and Such datasets are essential for calibrating remote- infl uence the deep circulation within the Arctic ly sensed data, and will serve to help validate basins. Gerdes et al. (2003) have demonstrated numerical models (Budgell 2005, for example). that the ice extent exerts a strong control on the water mass transformation taking place in the Barents Sea. Sea ice extent data have been avail- Instruments and methods able for the Barents Sea over the last few decades from remote sensing (see, for example, Kvingedal ULS data & Sorteberg unpubl. ms.). Very few data, how- ever, are available on sea ice thickness, although A ULS was deployed in 1993–94, but unfortu- indirect measurements have recently begun to nately it did not work; there are no data from this Abrahamsen et al. 2006: Polar Research 25(1), 25–37 25 o N 80 2000 Kvitøya Victoria Island 500 o N Franz Josef Land Nordaust− 78 landet Kong Karls Svalbard Land Olga Basin Edgeøya o N B1 76 2000 500 o 200 N 74 mlya Bjørnøya 200 Novaya Ze o N Barents Sea 72 o N Norway 70 o 20 o 25 o 30o 35o o o E o E o E E E E E E 40 E 45 50 55 60 Fig. 1. Map of the Barents Sea. The 200 m isobath (from IBCAO v. 1) is drawn with a thicker line; other depths indicated are 100 m and 500 m and above with 500 m increments. The bold X marked B1 shows the location of the ULS and current meter mooring. year. A Christian Michelsen Research (CMR) ES- ysis project at the NOAA-CIRES Climate Diag- 300-IV ULS (Strass 1998) was deployed in 1994, nostics Center (NCEP 2003). The average density and was replaced in 1995 with an APL (Applied and speed of sound in the water column above the Physics Laboratory, University of Washington) ULS were calculated using temperatures meas- ULS Mark-2 (Drucker et al. 2003). The instru- ured in the ULS and in the current meter, along ments sampled the ice draft at intervals of 4 and 5 with climatological salinities (from Steele et al. minutes, respectively. 2001), enabling the pressure and return time to be converted to instrument depth and range, respec- Data processing tively. The target range is then subtracted from the instrument depth to obtain the initial ice draft The initial data consist of sonar return times and (or water level) estimate. These estimates were pressures. The pressures are corrected for var- then processed using a method involving satel- ying sea level pressure using quarter-daily sea lite-derived ice concentrations to correct the zero level pressure data for 77.5° N 27.5° E download- level of the data, and to classify the data into open ed from the website of the NCAR/NCEP reanal- water and ice. This process, more fully described 26 Ice draft and current measurements, north-western Barents Sea by Abrahamsen (2003), is summarized below. and 99 %, we sort the data in the block in order of Our source for ice concentration data was increasing draft. Since the lowest (100-n) percent the Defense Meteorological Satellite Program of our measurements correspond to open water, (DMSP) Special Sensor Microwave Imager we set the zero level to the median of the drafts (SSM/I) (Cavalieri et al. 1990). Available for in this range, and fl ag these points as open water. both hemispheres with 25-km resolution, these The use of the median is very robust at determin- data have previously been compared with ULS- ing the zero level, as the mode of the open water derived measurements (e.g. Harms et al. 2001). distribution should be centred on the mean water We note that in Fig. 8 of Harms et al. (2001), the level. The rest of the points are fl agged as ice, and satellite data appear to yield higher estimates of all points in the block are also marked as uncor- the ice concentration than the ULS. We experi- rected. We now interpolate the determined zero enced the opposite effect, with the satellite data level to all points previously fl agged as uncorrect- giving slightly lower estimates of the ice concen- ed. The resulting time series is fi ltered with a 10- tration. A reason may be that if both ice and water day low-pass fi lter and then subtracted from the are present in the ULS footprint—in our case this drafts. We have now determined the zero level for nominally covers a circle approximately 3 - 7 m in all the data points. The fi nal step is to force the diameter—the ULS would preferentially measure open water points to be distributed evenly around the ice, especially when the ice concentration is this zero level. Therefore we go through the data high. We found, empirically, that using the square blocks with ice concentrations above 0 % again, root as a transfer function for the satellite-derived and, in order of increasing drafts, we successively ice concentrations (thus skewing them towards reclassify data points originally fl agged as ice to higher concentrations), yielded better results in open water, such that the mean of the (corrected) the algorithm below, both visually and in terms of open water drafts is as close to zero as possible. the bias in the resulting mean ice drafts (as esti- mated in the next section). This may not be appro- Error estimate priate in other regions, particularly those that experience lower ice concentrations. Based on previous validated studies of sea ice Assuming that a fraction of the ULS measure- drafts using ULS, we estimate that the RMS error ments that corresponds to the satellite-derived ice of the individual measurements is on the order concentration is for ice-covered conditions, we of 10 cm, while there could be an overall bias know that the remaining points must correspond towards thicker mean ice drafts of up to 12 % for to open water. Our algorithm uses this to classify 1994–95 (Kvambekk & Vinje 1992; Strass 1998), each data point as either ice or open water, and to and probably around 2 % for 1995–96 (Drucker et correct the zero level of the data. al. 2003), the difference resulting from the slight- We fi rst go through the satellite-derived ice ly different sampling techniques within the dif- concentration time series for the pixel closest ferent instruments. This bias is most likely the to the mooring and fi ll in any missing points by dominant error in the ice thickness statistics, and, linear interpolation. We then take the square root as we have no ice thickness verifi cation data, it of the ice cover (for reasons described above). We cannot be reliably quantifi ed and corrected. now go through the ULS time series in blocks To determine errors resulting from, or remain- of 10 days. We found that 7 - 10 days works best; ing after the data processing, we went through the with longer blocks, the correspondence between time series and manually indicated the water level the percentage of open water points and the sat- for one-week blocks. Compared with this base- ellite-derived ice cover will improve, but the line, our data appear to have an overall bias on the longer averaging times will also smear out the order of 3 cm toward thicker ice, with RMS errors profi le, ignoring brief events in atmospheric/oce- around 6 - 8 cm.