The arctic ice thickness anomaly of the 1990s: A consistent view from observations and models

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. C3, 3083, doi: /2001jc001208, 2003 The arctic ice thickness anomaly of the 1990s: A consistent view from observations and models D. A. Rothrock, J. Zhang, and Y. Yu Applied Physics Laboratory, University of Washington, Seattle, Washington, USA Received 6 November 2001; revised 15 May 2002; accepted 18 November 2002; published 18 March [1] Observations of sea-ice draft from submarine cruises in much of the Arctic Ocean show that the ice cover was unusually thin in the mid-1990s. Here we limit our examination to digitally recorded draft data from eight cruises spanning the years 1987 to 1997 and find a decrease of about 1 m over the 11-year span. Comparisons of our modeled draft with observed draft show good agreement in the temporal change. Comparing average draft over entire cruises, the RMS discrepancy between modeled and observed draft is 0.3 m and the correlation is Agreement in the spatial patterns of draft is somewhat lower; the RMS discrepancy of 50-km averages of draft is 0.7 m and the correlation is We review reports of interannual variations of ice thickness or volume from other model studies. All models agree that thickness decreased by between 0.6 and 0.9 m from 1987 to Our model shows a modest recovery in thickness from 1996 to For the 1950s, 1960s, and 1970s, models tend to disagree on the size and to a lesser extent the timing or phase of interannual variations. INDEX TERMS: 1635 Global Change: Oceans (4203); 1863 Hydrology: Snow and ice (1827); 4207 Oceanography: General: Arctic and Antarctic oceanography; 4215 Oceanography: General: Climate and interannual variability (3309); 9315 Information Related to Geographic Region: Arctic region; KEYWORDS: sea ice, thickness, submarine, draft Citation: Rothrock, D. A., J. Zhang, and Y. Yu, The arctic ice thickness anomaly of the 1990s: A consistent view from observations and models, J. Geophys. Res., 108(C3), 3083, doi: /2001jc001208, Introduction Copyright 2003 by the American Geophysical Union /03/2001JC001208$09.00 [2] As the observational record of the Arctic Ocean has improved in the last several decades, a great deal has been learned about the variability of the arctic atmosphere, the ocean, and its ice cover. Changes during these decades are now well simulated by numerical weather prediction model reanalyses and well documented by data from numerous field and ship expeditions, drifting buoys, satellites, and submarines. From many of these data sets, a picture has emerged of interannual variability unimaginable 30 years ago. [3] Several authors have addressed the interannual changes in ice draft observed by submarines from the late 1980s until Analyses are complicated because data from submarine cruises are sparsely distributed by season, year, and region. Various authors have analyzed different data sets. Here the discussion is based solely on digitally recorded submarine data. Limiting our considerations to digitally recorded data avoids any uncertainty in comparing these data with (earlier) data recorded on paper charts and later digitized. These published submarine observations are summarized in Table 1 and in Figure 1. The data from columns 2, 4, and 5 of Table 1 are plotted for each region in Figure 2. Observations from mixed seasons (Figure 2, black) carry less weight in our opinion for studies of interannual draft variations, since they rely heavily on corrections for, or are intermingled with, an annual cycle of about 1 m. The differences in Figure 2 between observations within each region are generally attributable to differing observing seasons, not to discrepancies in the analyses of the data. [4] The observational evidence seems strong that ice draft in the western Arctic Ocean, on transects roughly following 150 W from the North Pole to the Beaufort Sea, and in a large portion of the central Arctic Ocean, has declined substantially during a decade of observation. The North Pole itself shows little change over two decades. Two other reported values (of an increase in the Nansen Basin and a decrease in the eastern Arctic) are based on data too sparse to be given much weight [Rothrock et al., 1999, Table 2]. [5] Simulated ice thickness reveals detailed long-term variability not captured by the existing observational record. Figure 3a shows the Arctic annual mean thickness for about 50 years, as simulated by our ice-ocean model (described below). Much about the figure is striking. The lowest ice thickness in the entire 50 years appears in the mid-1990s, near the end of the record, when the ice is 2.1 m thick. This is fully 1.4 m thinner than the 3.5-m maximum in 1966, a decrease of some 40%. There was another strong maximum of 3.0 m in Over 50 years the mean thickness is 2.9 m, with a standard deviation of 0.33 m. Changes of one-half meter within several years are common. The messages are that large changes in ice thickness are consistent with the known (modeled) physics and with the known variations in atmospheric forcing, and that there was a remarkable decline in modeled thickness during the period 1987 to This decline occurred in both the eastern and western longitudes of the Arctic Ocean (Figure 3b), more strongly in 28-1

2 28-2 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S Table 1. Observations of Interannual Change in Ice Draft From Digitally Recorded Submarine Sonars Source and Location Period Season a m Early Draft, b Later Draft, b m Draft Change, m North Pole to Beaufort Sea Transects Rothrock et al. [1999], Table 2 Beaufort Sea, Canada Basin S Winsor [2001], Figure W W and S c Tucker et al. [2001], Results W W North Pole Shy and Walsh [1996], abstract all Rothrock et al. [1999], Table S Winsor [2001], Table W and S Tucker et al. [2001], Results W SCICEX Box Rothrock et al. [1999], Table S a S, late summer (September and October); W, late winter (April and May). b In column headings, early and later refer to the beginning and end of the period for each investigation. c Where small trends (magnitude <0.03 m yr 1 ) were represented by the authors to be negligible, we have shown the trend as 0. Trend, m/yr the eastern sector. The decline continued more persistently in the western Arctic, whereas the eastern Arctic experienced some recovery after The thickness in the two basins tends to move in concert with the notable exception of 1973 to The decline from 1987 to 1996 is evident in both winter and summer (Figure 3c). Comparable results from other models are reviewed below. [6] Here we wish to illustrate the interannual changes in sea ice draft during 11 years of digitally recorded submarine observations from 1987 to 1997, to show the overall picture of interannual variability of ice thickness given by our iceocean model, and to examine the consistency between the ice thickness in models and in observations. A trusted consistency would allow the use of models to search for causes of ice thickness changes, which are difficult to identify from observations alone. Finally we discuss the interannual variation of thickness in several simulations and offer some conclusions. 2. Submarine Observations of Sea Ice Draft [7] Between 1958 and 2000 there have been about 63 cruises under sea ice by U.S. Navy submarines; many of the data and ship tracks are classified. In the 1990s, there were six unclassified SCICEX cruises (Scientific Ice Expeditions) with civilian scientists participating in cruise planning. The U.S. cruises have ranged widely over the Arctic Ocean. Data collected more than 200 miles from non-u.s. territories in a central Arctic region referred to as the SCICEX Box (Figure 1) are eligible for declassification and public release. Dates on some cruises are given only to within a 10-day window. The U.S. data, much of it still classified, are archived at the Arctic Submarine Laboratory in San Diego, CA. There are also data from a half dozen British naval submarine cruises, whose range includes the Greenland Sea to the North Pole and the Lincoln Sea. Several groups have worked and continue to work to prepare all these data for public release; data are available through the National Snow and Ice Data Center (NSIDC; see Acknowledgment). [8] To operate in ice-covered waters, these submarines require knowledge of the ice cover above them and are equipped with an upward-looking sonar that measures the distance to the bottom of the ice. A pressure sensor provides the distance to the sea surface. The difference between the two distances is ice draft. About 89% of the ice thickness is under water and seen as draft. There are many uncertainties in these observations of ice draft, arising from uncertain sound speed, changing sea level pressure, the finite sonar transducer beam width, and spurious returns. All data have been recorded on paper charts. Between 1976 and 2000 the U.S. Navy submarines also digitally recorded these data. Spurious data are removed and straight and level cruise segments selected for analysis. A single correction is performed to remove bias: open water segments are identified on the paper chart record by its grassy appearance, and an open water offset is interpolated into the record to provide the correct zero for the draft measurement. The precision (relative error) from ping to ping is estimated to be 0.3 m, and the overall accuracy of a 50-km mean draft, 0.15 m[mclaren et al., 1994]. [9] Here submarine ice draft data from eight cruises from the years 1987 to 1997 (Figure 4) are analyzed in comparison with modeled thickness. All these data are available at the NSIDC and were digitally recorded; here we have avoided comparing digitally recorded data with data recorded on paper charts and subsequently digitized. On Figure 1. Three regions for which interannual change in draft has been reported from digitally recorded submarine observations: Pole to Beaufort transects, North Pole, and SCICEX Box.

3 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S 28-3 consider this possibility. Taking Cruise Track 1, which occurred in 1987 (Figure 4, upper left), we compute an 11- year mean modeled draft D 1 supposing that Cruise 1 occurred along the same track in each of the 11 years ( ) on the same days of the year; we repeat this procedure for all eight cruises computing D 1 to D 8 and look for any decrease in the means D 1 to D 8, which would indicate that later tracks sampled thinner ice. There is no decrease evident for the five winter cruise tracks (D 1 to D 4, and D 6 ). The last summer value D 8 (for the 1997 cruise track) is about 0.4 m below D 5 and D 7, giving a fictitious downward summer trend of 0.10 myr 1, so the downward summer trend apparent in Figure 5a can be attributed to the fact that the 1997 cruise track sampled preferentially in regions of thin ice. 3. Comparing Modeled and Observed Draft [12] The ice-ocean model used here has the components listed in Table 2 and described in more detail in Appendix A. Surface forcing consists of daily surface air temperature and sea level pressure from the National Centers for Environmental Prediction (NCEP) reanalysis for 1948 to 1978, and, for 1979 to 1999, temperature and pressure data from the International Arctic Buoy Programme (IABP). Switching to the IABP data in 1979 does not significantly change the time history of average arctic ice thickness but does somewhat alter the spatial patterns of thickness. Figure 2. Summary of observed interannual changes in ice draft (see also Table 1). As in Table 1 and in Figure 1, the observations are grouped into three regions (separated by green lines). Winter observations are shown in blue, summer in red, and observations from mixed seasons in black. (RYM signifies Rothrock et al. [1999].) this basis, we do not consider here the earlier data of Rothrock et al. [1999] and the comparison of Wadhams and Davis [2000]. The spatial coverage varies from cruise to cruise (Figure 4); the cruise track lengths range from 1256 to 15,392 km. The North Pole and north-south transects between 140 and 150 W have been observed repeatedly, and there are many other data that it seems prudent to include in an analysis of the Arctic Ocean ice cover. The SCICEX cruises in 1993, 1996, and 1997 are all late summer cruises; each provides fairly broad spatial coverage. The data shown in the figure and the data we analyze are mean draft over nominally 50-km track segments (actually 25 to 65 km in length). The mean draft includes readings of open water. [10] The mean draft from each cruise is shown by the open symbols in Figure 5a. The data show a steady decline in ice draft, in both the winter and summer, of over 1 m during these 11 years. The trends are 0.16 m yr 1 in winter and 0.11 m yr 1 in summer. Similarly, the data (Figure 5b) reveal a gradual decrease by roughly half in the fraction of thick ice. [11] The fact that cruises in different years sample different regions of the Arctic Ocean raises the question of whether the trends seen in Figure 5, for instance, are the result of sampling variations rather than passing years. We use our model to Figure 3. (a) Modeled mean annual thickness for the Arctic versus year ( ). (b) Modeled mean thickness versus year for the western and eastern longitudes of the Arctic Ocean and for the Kara, Barents, and Greenland, Iceland, and Norwegian (GIN) seas. (c) Modeled mean thickness in May (annual maximum) and in September (annual minimum) versus year. (Disregard the first 4 or 5 years as an adjustment period to unknown initial conditions.) The domain includes the Arctic Ocean, Kara, Barents, and GIN seas. Mean thickness is defined as volume divided by extent.

4 28-4 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S 5a). As in the observations, the model shows a decline in both summer and winter. The agreement between observations and the model is compelling; the RMS difference is 0.28 m, whereas the range of the observations is 2.8 m. The model tends to overestimate the concentration of thick ice by about 0.1 (Figure 5b), yet the modeled decline in thick ice seems to strongly mimic the observed decline. [15] The same data comparing modeled with observed cruise means are shown as scatterplots in Figure 6. Both Figures 5 and 6 show that the model seems to underestimate draft for thicker (winter) ice and to overestimate it for thinner (summer) ice. They show that, although the model underestimates the range of mean draft, with whole-cruise (large-sample) averages, the model and data agree quite well. Figure 4. The observed ice draft (m) along eight submarine cruise tracks from 1987 to In chronological order, the submarines on these cruises were: HMS Superb, USS Archerfish, USS Pargo, USS Grayling, USS Pargo, an unnamed U.S. submarine, USS Pogy, and USS Archerfish. The SCICEX Box is shown in the 1988 panel. [13] How well do the model and the observations agree? Draft varies with location, time of year, and year. The basic method of comparison is this. First, modeled thickness is multiplied by 0.89 and called modeled draft. Then, from the daily modeled field, the mean draft for the same segment as the observational mean is computed. The effect is to reenact the submarine cruise under the modeled ice cover. The fundamental data set for comparison consists of records representing roughly 50-km means of draft from the submarine data and from the model along with location, date, track length, cruise identifier, etc. [14] The first comparison made is for an average of all data from each cruise: eight cruises, eight means (Figure Figure 5. (a) Draft and (b) concentration of ice drafts greater than 4 m averaged over each cruise track versus year, from observations (circles and squares) and from the model (+ and ). Note that these data are averages over entire cruise tracks and are not directly comparable with the modeled means over the entire Arctic Ocean shown in Figure 3.

5 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S 28-5 Table 2. Summary of Coupled Ice-Ocean Model Components Ocean model: baroclinic model with an embedded mixed layer; 21 vertical levels; ice-ocean coupling [Zhang et al., 1998] Ice model: 12 thickness categories each for undeformed ice, ridged ice, ice enthalpy, and snow depth [Thorndike et al., 1975; Flato and Hibler, 1995; Zhang and Rothrock, 2001] Ice thermodynamics: one snow layer and two ice layers [Winton, 2000] Ice dynamics: viscous plastic [Hibler, 1979]; momentum equation solved using Zhang and Hibler [1997] dynamics model Surface forcing: daily surface air temperature and sea level pressure from NCEP reanalysis (for ) and from the International Arctic Buoy Programme (IABP) (for ); geostrophic winds, downwelling radiation, and specific humidity following Parkinson and Washington [1979] Model domain: Arctic Ocean, Kara, Barents, Greenland, Iceland, and Norwegian seas with area of km 2 ;40km 40 km horizontal resolution [16] When the individual data points (the nominal 50-km means) are compared, the agreement is not as good (Figure 7). On a cruise-by-cruise basis (indicated by different colors and symbols), the real ice cover has more variance (abscissa) than the modeled ice cover (ordinate). The variance left in the record and the RMS discrepancy depend on the length scale of the averaging (Table 3). Thus, it appears that the spatial patterns of variability are not captured as well by the model as the arctic-mean temporal variability; the agreement is better when the spatial scale is increased by averaging. [17] Figure 8 shows the composites of all summer and all winter data: modeled and observed. In general, the modeled winter thickness is a bit thinner than observed; the modeled summer thickness is somewhat greater than observed, particularly in the Beaufort Sea. Neither modeled field shows as strong a gradient across the figures (from Alaska to Spitzbergen) as observed. This spatial pattern of the discrepancy is seen more clearly in Figure 9. The model overestimates ice draft (reds) in the Beaufort Sea and eastern Arctic Ocean. The model tends to underestimate draft (blues) near the pole and north of about 80 N inthe Canada Basin. The RMS discrepancy is 0.73 m (from Table 3). The same pattern of discrepancy occurs in the concentration of ice thicker than 4 m (not shown); the model shows more thick ice than observed both in the Beaufort Sea and in the eastern Arctic Ocean toward the Laptev Sea. 4. Discussion 4.1. Interannual Variation of Thickness in the Present Simulation [18] The modeled mass of ice in the Arctic Ocean is quite variable on decadal timescales (Figure 3). The range in thickness over 50 years is 1.4 m, from an Arctic Ocean annual mean of 3.5 m in 1966 to 2.1 m in There was a 50-year maximum in 1966, another strong maximum in 1987, and a strong decline from 1987 to The peak in 1987 and the subsequent decline seem to derive more strongly from the eastern Arctic Ocean but are evident in both eastern and western sectors. The interannual variability of temperature and of a pressure pattern index is captured in Figure 10. Of note are the minima in both the temperature and the North Atlantic Oscillation (NAO) index in and, in the latter decade of the record, the elevated temperatures and NAO index. The elevated temperature would be in equilibrium with a thinner ice cover. [19] To illustrate the recent change regionally, Figure 11 shows the trend in modeled thickness from 1987 to It is in agreement with that shown by Hilmer and Lemke [2000]. The trend is most strongly negative in the East Siberian Sea and in the corridor from there to Greenland, which are the regions most strongly affected by the changed ice circulation from 1989 to 1998 [Rigor et al., 2002]. During the 1980s, these regions were supplied with ice recirculating in the strong Beaufort anticyclonic gyre; when the Icelandic cyclonic circulation strengthened during the late 1980s and 1990s (high NAO index in Figure 10b), these regions were supplied with younger ice from the Laptev Sea. Tucker et al. [2001] suggest that the weaker Beaufort gyre and invasion of younger ice from the Laptev Sea caused the observed thinning along the submarine transects from the North Pole to the Beaufort Sea. High NAO also signals a more vigorous evacuation of ice from the central Arctic Ocean through Fram Strait, leaving younger, thinner ice behind Thickness in Other Simulations [20] A number of papers describing ice models and their behavior report the interannual change in ice amount Figure 6. (a) Draft and (b) concentration of ice drafts greater than 4 m averaged over each cruise, observed versus modeled.

6 28-6 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S Figure 7. Mean draft for roughly 50-km segments along cruise tracks from observations plotted against the same quantity from the model. (Table 4). Most authors give ice volume; Chapman et al. [1994] give mass, and Polyakov and Johnson [2000] show thickness. Here the given annual mean volumes are converted to annual mean thickness, dividing by the reported annual mean extent or our best estimate of it for the particular domain; where the annually varying extent is given, first seasonally varying thickness and then its annual means are computed. Our intention here is simply to illustrate the range of results found in the literature for this one variable, not to undertake a model intercomparison study. [21] The resulting estimates of thickness are shown in Figure 12 (upper). There are large discrepancies in the mean thickness that cannot generally be accounted for by different model domains. Most model domains are the same: the Arctic Ocean, and the Barents, Kara, and Greenland-Iceland-Norwegian seas. The domain of Holloway and Sou [2002] excludes part of the Greenland-Iceland-Norwegian seas. The domain of Hilmer and Lemke [2000] includes Baffin Bay and that of Chapman et al. [1994] includes Baffin and Hudson bays and the Bering Sea; clearly these larger domains include more areas of thin ice. [22] The wide range of model results shows that there are many differences in how each model has chosen and incorporated forcing data and in their internal representation Figure 8. Composite of mean draft for winter (a) and for summer (b) cruise tracks. Model mean draft for period of winter cruises (c) and of summer cruises (d). of the physics. From the model components and forcing data listed in Table 4, nothing is immediately apparent to explain some of the major differences among models except for the different surface air temperature data set (Jones et al. [1999] in Figure 10a) used by Chapman et al. [1994], Hakkinen [1993], and Flato [1995] (all three shown dotted in Figure 12). These three simulations show lower mean thicknesses than the others, which is likely due to the warmer mean temperatures of Jones et al. [1999]. There is no obvious correlation between mean thickness and the ice model components used or whether an ocean model is included. Of course the radiative formulations and treatment of albedo in each of the models would affect mean thickness strongly. [23] As for interannual variability, the one unanimous feature of the models response is a strong decline in thickness from 1987 to The present model shows the strongest decrease: 0.85 m, but even the smallest decrease, 0.60 m, is quite substantial. On other features of the interannual response shown in Figure 12 (lower), the models are in less firm agreement. The models all tend to have positive anomalies in thickness during the 1970s, negative anomalies in the early 1980s, and positive anomalies from 1986 to Agreement is poorest before about (Could this have to do with differences among data sets before the advent of the International Arctic Buoy Programme in 1978?) The present model and those of Table 3. Variability of Observed and Modeled Mean Draft Length of Average No. of Samples Observations SD, m Model SD, m RMS Discrepancy, m r a Observations as in 50 km Figure km km Whole cruise b b 0.75 b Figure 5a a r is the correlation between the observations and the model. b The standard deviation does not decrease as the averaging length is extended to whole-cruise averages because the cruises are of unequal length and contain different numbers of data points. These standard deviations use the divisor (n 1) 1/2.

7 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S 28-7 Figure 11. Trend in modeled ice thickness from 1987 to 1999 in m yr 1. Figure 9. Modeled minus observed mean draft (m) along cruise tracks from 1987 to Hilmer and Lemke [2000] and Holloway and Sou [2002] show an overall maximum in ; other models tend to show only a local maximum in the same period. The forcing in this period (Figure 10) has two features that contribute to a thick ice cover: the mean annual temperature was about 2 C below normal, and the NAO index was strongly negative, meaning that the polar surface anticyclone was strong and ice tended to recirculate longer in the Arctic Ocean before being exported. Evidently the present model is the most sensitive to these interannual differences in temperature and wind patterns. While several models show an overall decline from the mid-1960s to the early Figure 10. (a) The mean surface air temperature for each of two data sets versus year, averaged over our model domain. The horizontal lines are 52-year means. (b) The North Atlantic Oscillation (NAO) index versus year for 1948 to The bold curve in (b) is a 3-year running mean. 1980s, others show an increase. Those that show a decrease all make use of the NCEP surface air temperature as their thermodynamic forcing function; those that show an increase (Figure 12b) [Chapman et al., 1994; Flato, 1995] use the Jones et al. [1999] temperature record. There is, however, no obvious difference between the two temperature records during this period that would explain the different thickness variability. Hakkinen s [1993] simulation uses a seasonal temperature climatology without interannual variation. Holloway and Sou [2002] show five cases with different wind stress and downward longwave formulations. The 50-year-mean volumes for each of their five cases, V 50 j i, i = 1 to 5, vary by a factor of two. The interannual variations, however, are quite similar on a percentage basis for all five cases; the range as a percentage of the mean for each 50-year simulation varies narrowly from a high of 31% (in their case a) to a low of 26% (in their case e). 5. Conclusions [24] Models and observations show compelling agreement that ice thickness declined from the late 1980s through Our modeled trend during is 0.08 to 0.14 m yr 1 in the Transpolar Drift Stream and about half that elsewhere (Figure 11). The rate of decrease in other simulations is somewhat smaller. These decreases are the same size found in submarine ice draft observations (Table 1) in general but not at the North Pole. The draft observations in Figure 5a have a trend of 0.16 m yr 1 in winter; the downward summer trend apparent in the figure is likely due to bias from the location of the 1997 cruise track (section 2). These trends suggest that the bulk of the decrease of 1.3 m from the 1960s to the 1990s [Rothrock et al., 1999] occurred between the late 1980s and the 1990s [Tucker et al., 2001]. This view is supported by our modeled thickness in Figure 3a. [25] What is happening now? Modeled thickness in the western Arctic Ocean continued its decline through Our model indicates some recovery in the eastern Arctic Ocean and overall since We see no convincing argu-

8 28-8 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S Table 4. Model Simulations of Interannual Change in Sea Ice Amount, Converted to Thickness a Source Model Type b Ice Morphology Model Type c Rheology/Thermodynamics Coupled to 3-D Ocean Model Forcing Data d Wind/Temperature Period Chapman et al. [1994] mass in Figure 7 2-category VP/PW, 0-layer no NCEP! buoy SLP d / Jones et al. [1999] SAT m Hakkinen [1993] volume in Figure 4a 2-category V/PW, 3-layer yes NCAR SLP d / CM climatology SAT m Flato [1995] volume in Figure 6a 28-category TD VP/PW, 0-layer no NCEP! buoy SLP d / Jones et al. [1999] SAT m Thickness: Mean Over Simulation, m Thickness: Interannual Range of Annual Mean, m to 1.87 = to 1.65 = to 2.63 = 0.5 Arfeuille et al. [2000] volume in Figure 4 2-category G/0-layer no NCEP SLP d / NCEP SAT m to 2.68 = to 3.33 = 0.8 Hilmer and Lemke [2000] volume in Figure 1 Polyakov and Johnson [2000] thickness in Figure 5a Holloway and Sou [2002], Figure 7 (upper), case d 2-category VP/PW, 0-layer no NCEP surf. winds d / NCEP SAT d 6 thin-ice-category TD EP/PW, 3-layer yes NCEP SLP d / NCEP SAT d to 2.95 = category VP/PW, e 0-layer yes NCEP SLP d / NCEP SAT to 3.31 = 0.9 Present, volume in Figure 1 12-category TED VP/PW, 3-layer yes NCEP! buoy SLP d / NCEP! buoy SAT d to 3.52 = 1.4 a Rows are arranged by increasing size of the interannual range in thickness in far right column. b TD, thickness distribution, TED, thickness and enthalpy distributions. c Rheologies: V, viscous; E, elastic; P, plastic; G, granular. PW refers to surface heat balance parameterized following Parkinson and Washington [1979]. Layer refers to the resolution of heat content vertically through the ice thickness; 0-layer is the model in the appendix of Semtner [1976]. d NCEP, National Centers for Environmental Prediction. NCAR, National Center for Atmospheric Research. SLP, sea level pressure. SAT, surface air temperature. Climatology, monthly but no interannual variation. m, monthly mean. d, daily. NCEP! buoy means that NCEP data are used through 1978 and data from the International Arctic Buoy Programme buoys are used starting in e The downward longwave formulation differs from that of Parkinson and Washington [1979].

9 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S 28-9 errors in the fields produced by the conglomerate model and forcing data are a first step, but tests that isolate individual components of the model physics (e.g., atmospheric or oceanic drag coefficients, or ice rheology, or albedo or radiation formulations) or that directly evaluate the quality of forcing data will be the most useful in showing how to improve our model-based understanding of the ice cover. Figure 12. (Upper) Annual mean thickness from several ice models during the period 1951 to (Lower) The thickness difference from the mean over each simulation. ment that the decline through 1996 should be extrapolated as a prediction of future behavior. The most recent ice draft observations reported are more than 4 years old. Updating and filling in the observational record with satellite altimeter data, and very recent (1998 to 2000) and historical (from the 1960s and 1970s) submarine observations would help clarify the picture, as would a public archive containing all moored sonar observations of ice draft. [26] Although models agree that there was a strong decrease from the late 1980s through the mid-1990s, they disagree on the earlier record. In most models maxima appear in the mid-1960s and the late 1970s. [27] Our model shows greater interannual variability than other models but less than is seen in the observations. There is substantial disagreement between observations and our model in the spatial structure of the ice mass field; modeled ice is thicker than observations in the Beaufort Sea, and thinner toward the North Pole. [28] These results strongly illustrate the need for careful model intercomparison studies and for rigorous assessments of models vis-a-vis the growing data sets of ice draft, motion, extent, and concentration. Simple estimates of the Appendix A [29] The coupled ice-ocean model consists of a 12-category thickness and enthalpy distribution (TED) sea-ice model and an ocean model with an embedded mixed layer. The ocean model is described by Zhang et al. [1998]. The TED sea-ice model consists of five main components: a momentum equation that determines ice motion, a viscousplastic ice rheology [Hibler, 1979] with an elliptical yield curve that determines the relationship between ice internal stress and ice deformation, a heat equation that determines ice growth/decay and ice temperature, two (ridged and undeformed) ice thickness distribution equations that conserve ice mass [Flato and Hibler, 1995], and an enthalpy distribution equation that conserves ice enthalpy [Zhang and Rothrock, 2001]. The heat equation is solved, over each category, using Winton s [2000] three-layer thermodynamic model, which divides the ice in each category into two layers of equal thickness beneath a layer of snow. [30] Accompanying the ice model is a snow model described in terms of snow thickness distribution g s (h). The snow conservation equation, the treatment of the snow thickness distribution, and the treatment of the thermodynamics at the ice/snow/ocean surface, including surface albedo, follow Flato and Hibler [1995] with the snow and ice thermodynamic parameters of their standard case. The parameters governing the ridging process, such as the frictional dissipation coefficient, the ridge participation constant, and shear ridging parameter, also follow Flato and Hibler [1995]. [31] The model domain covers the Arctic Ocean, Kara, Barents, and Greenland-Iceland-Norwegian seas. It has a horizontal resolution of 40 km 40 km and 7349 cells covering an area of km 2. It has 21 ocean levels, and 12 thickness categories each for undeformed ice, ridged ice, ice enthalpy, and snow. The partition of ice thickness categories and the model domain and bottom topography is given by Zhang et al. [2000]. [32] Daily surface atmospheric forcing is used to drive the model from 1948 to The forcing consists of geostrophic winds, surface air temperature, specific humidity, and longwave and shortwave radiative fluxes. They are calculated following Parkinson and Washington [1979] using sea level pressure and surface air temperature fields provided by NCEP reanalysis project for and, for , using pressure and temperature from the IABP [see Rigor et al., 2000]. Model input also includes river runoff and precipitation as detailed by Zhang et al. [1998]. The wind stress denoted as a complex number is, t a = r a C a jgjge if, where r a is the density of air, the atmospheric bulk drag coefficient C a is * a C, and a C varies sinusoidally from a minimum of 0.50 on 1 January to a maximum of 1.00 on about 1 July [Ip, 1993, solid curve in Figure 6.7a]. The turning angle f, positive to

10 28-10 ROTHROCK ET AL.: ARCTIC ICE THICKNESS ANOMALY OF THE 1990S the left of the surface geostrophic wind vector G, varies from a maximum of 30 on 1 January to a minimum of 20 on about 1 July [Ip, 1993, solid curve in Figure 6.7b]. [33] Acknowledgments. The authors gratefully acknowledge the support of the National Science Foundation, Office of Polar Programs (grants and ) and of the National Aeronautics and Space Administration (grant NAG5-9334). These data, Submarine Upward Looking Sonar Ice Draft Profile Data and Statistics, were obtained from the National Snow and Ice Data Center, University of Colorado at Boulder (nsidc@kryos.colorado.edu). We thank D. Bentley of the Arctic Submarine Laboratory and W.B. Tucker of the Cold Regions Research and Engineering Laboratory for making these data available for public release to NSIDC and for consultation on data processing. We also thank M. Steele for a thorough review. M. Wensnahan kindly digitized the results from other model simulations. References Arfeuille, G., L. A. Mysak, and L. B. Tremblay, Simulation of the interannual variability of the wind-driven Arctic sea-ice cover during , Clim. Dyn., 16, , Chapman, W. L., W. J. Welch, K. P. Bowman, J. Sacks, and J. E. Walsh, Arctic sea ice variability: Model sensitivities and a multidecadal simulation, J. Geophys. Res., 99, , Flato, G. M., Spatial and temporal variability of Arctic ice thickness, Ann. Glaciol., 21, , Flato, G. M., and W. D. Hibler III, Ridging and strength in modeling the thickness distribution of Arctic sea ice, J. Geophys. Res., 100, 18,611 18,626, Hakkinen, S., An Arctic source for the Great Salinity Anomaly: A simulation of the Arctic ice-ocean system for , J. Geophys. Res., 98, 16,397 16,410, Hibler, W. D., III, A dynamic thermodynamic sea ice model, J. Phys. Oceanogr., 9, , Hilmer, M., and P. Lemke, On the decrease of arctic sea ice volume, Geophys. Res. Lett., 27, , Holloway, G., and T. Sou, Has arctic sea ice rapidly thinned?, J. Clim., 15, , Ip, C. F., Numerical investigation of different rheologies on sea-ice dynamics, Ph.D. thesis, 242 pp., Dartmouth Coll., Hanover, N. H., Jones, P. D., M. New, D. E. Parker, S. Martin, and I. G. Rigor, Surface air temperature and its changes over the past 150 years, Rev. Geophys., 37(2), , McLaren, A. S., R. H. Bourke, J. E. Walsh, and R. L. Weaver, Variability in sea-ice thickness over the North Pole from 1958 to 1992, in Polar Oceans and Their Role in Shaping the Global Environment, Geophys. Monogr. Ser., vol. 85, pp , AGU, Washington, D. C., Parkinson, C. L., and W. M. Washington, A large-scale numerical model of sea ice, J. Geophys. Res., 84, , Polyakov, I. V., and M. A. Johnson, Arctic decadal and interdecadal variability, Geophys. Res. Lett., 27, , Rigor, I. G., R. L. Colony, and S. Martin, Variations in surface air temperature observations in the Arctic , J. Clim., 13, , Rigor, I. G., J. M. Wallace, and R. L. Colony, On the response of sea ice to the Arctic Oscillation, J. Clim., 15, , Rothrock, D. A., Y. Yu, and G. A. Maykut, Thinning of the arctic sea-ice cover, Geophys. Res. Lett., 26, , Semtner, A. J., A model for the thermodynamic growth of sea ice in numerical investigations of climate, J. Phys. Oceanogr., 6, , Shy, T. L., and J. E. Walsh, North Pole ice thickness and association with ice motion history , Geophys. Res. Lett., 23, , Thorndike, A. S., D. A. Rothrock, G. A. Maykut, and R. Colony, The thickness distribution of sea ice, J. Geophys. Res., 80, , Tucker, W. B., J. W. Weatherly, D. T. Eppler, D. Farmer, and D. Bentley, Evidence for the rapid thinning of sea ice in the western Arctic Ocean at the end of the 1980s, Geophys. Res. Lett., 28, , Wadhams, P., and N. R. Davis, Further evidence of ice thinning in the Arctic Ocean, Geophys. Res. Lett., 27, , Winsor, P., Artic sea ice thickness remained constant during the 1990s, Geophys. Res. Lett., 28, , Winton, M., A reformulated three-layer sea ice model, J. Atmos. Oceanic Technol., 17, , Zhang, J., and W. D. Hibler III, On an efficient numerical method for modeling sea ice dynamics, J. Geophys. Res., 102, , Zhang, J., and D. A. Rothrock, A thickness and enthalpy distribution seaice model, J. Phys. Oceanogr., 31, , Zhang, J., W. D. Hibler III, M. Steele, and D. A. Rothrock, Arctic ice-ocean modeling with and without climate restoring, J. Phys. Oceanogr., 28, , Zhang, J., D. A. Rothrock, and M. Steele, Recent changes in arctic sea ice: The interplay between ice dynamics and thermodynamics, J. Clim., 13, , D. A. Rothrock, Y. Yu, and J. Zhang, Applied Physics Laboratory, University of Washington, Seattle, WA 98195, USA. (rothrock@apl. washington.edu)