Seasonal fast ice thickness at the island of Hopen (Barents Sea) was monitored over 40 years. Sea ice thickness variability as a climate indicator provides more quantitative information on the state of the ice cover than solely sea ice extent. Usually, starting to form just before December Hopen fast ice reaches maximum thickness in May (on average 0.99 m), before the ice starts to decay. Swell, currents, and winds interrupt the fast ice development at Hopen during several of the winters observed, leading to ice removal and new ice formation. Since 2000, no ice thicker than 1.0 m was observed. We find a trend in the ice thickness anomalies of −0.11 m per decade, coinciding with decreasing seasonal maximum ice thickness, and an increase in local surface air and water temperatures. This is consistent with the decreasing sea ice extent in the Barents Sea and the entire Arctic.
 Part of the Svalbard archipelago, the narrow island of Hopen is 29 km long and between 0.5 and 2 km wide (Figure 1). The annual mean surface air temperature (SAT) measured at the Hopen meteorological station during 1966–2007 is −5.56°C, the annual mean surface water temperature (SWT), measured at the shore near the station while there was no fast ice, is +1.62°C (1972–2007). Hopen is located on top of Spitsbergenbanken, a shallow, less than 100 m deep bank stretching to Bjørnøya in the South West on the colder side of the Polar Front [Loeng, 1991]. The area around Hopen is very dynamic with different water masses circulating and mixing forced by wind, tides, and density-differences [Sundfjord et al., 2007]. Warm water of the North Atlantic Current (NAC) enters the Barents Sea from the south while another branch of the NAC follows the shelf break in the northeast Atlantic northward as the West Spitsbergen Current (WSC) and recirculates in the Barents Sea from the Arctic Basin. The pathways of the currents are strongly influenced by the seafloor topography [Pfirman et al., 1994; Gawarkiewicz and Plueddemann, 1995; Steele et al., 1995; Løyning, 2001].
1.3. Sea Ice in the Barents Sea
 The sea ice in the Barents Sea is predominantly seasonal ice, but multi-year ice is regularly advected from the Arctic Basin. The marginal ice zone (MIZ) in the Western Barents Sea extends south to 75°–78° N in early spring and retreats to about 80°–82° N in late summer, with a trend to higher latitudes during the last 10 years [Løset and Carstens, 1996; Tronstad et al., 2007]. The sea ice extent in the Barents Sea shows high interannual variability, overlaying a negative trend in summer ice extent 1966–1988 [Vinje and Kvambekk, 1991]. Vinje  found a decrease of sea ice extent in the Barents Sea when including older data back to 1864. Notably, the Barents Sea is the area in the Arctic with the largest reductions in sea ice extent observed in 1979–2006 [Meier et al., 2007] and with the highest increases in SAT projected in climate models for the Arctic [Christensen et al., 2007].
 Little is published on the thickness of Barents Sea sea ice. Abrahamsen et al.  present a pack ice draft time series, recording from November 1994 to November 1996 from an upward-looking sonar equipped mooring northeast of Hopen; monthly means of ice draft for the two seasons covered in these observations vary strongly, indicating influence of dynamical processes and possible advection of multi-year ice. Process studies in the MIZ of the Barents Sea show that the ice edge can change over short periods of time during spring [e.g. Ivanov et al., 2003]. From drill hole ice thickness measurements on a smaller number of ice stations (80 drillings) in the MIZ in May 1999, a modal ice thickness of 0.75 m was found, representing first-year ice, comparable to thicknesses in Svalbard fjords [Gerland and Hall, 2006; S. Gerland, unpublished data, 2008].
2. Measurements and Data Set
 Since 1946, a Norwegian permanently manned meteorological station has been in operation on Hopen's east coast (76°30′N, 25°01′E; Figure 1) where standard meteorological data are collected [Søreide, 1994]. From those data, we take air temperature and wind data into account in this study. Additionally, regular ice thickness measurements at approximately 100 m from the shore were initiated in the 1960s, resulting in 30 years of ice thickness measurements since winter 1965/1966 (Figure 2a). The shore at the research site is a gravel beach. The sea floor topography is not known in detail, but single measurements showed a water depth of 1.7 m and 3.0 m in a distance of 100 and 150 m from the shore, respectively. Most ice thickness data were collected from drill holes in shore-fast sea ice within 150 m of shore. The accuracy for an ice thickness reading is 1 cm (reading interval). Drill sites are not at the identical spot for each measurement throughout a season, resulting in apparent thickness changes due to local variability in ice thickness. Snow thicknesses were also measured, but less frequently than ice thickness. Therefore we focus here solely on the ice thickness data from the east coast. During ice-free periods, SWT is measured since 1972 in water collected in a bucket near the shore. The fast ice monitoring was recently updated and follows now a procedure similar to the Norwegian Polar Institute's (NPI) fast ice monitoring at Kongsfjorden at the western coast of Spitsbergen, Svalbard [Gerland and Renner, 2007].
 Due to the open setting of the coastline, the fast ice at Hopen is not always stable over an entire winter season as is the case for protected bays, such as e.g. in Kongsfjorden [Gerland and Renner, 2007]. In several of the observation years, the ice broke off and new ice formed later in the same winter. In some instances, thickness data indicate that advected ice was frozen in among the newly forming ice. Figure 3 shows all data for each of the observed ice growth seasons, with four winters highlighted. The earliest ice formation is in November, and the last stable ice is found in July. Maximum ice thicknesses (Figure 2c) observed are about 1.5 m (1967/1968, 1976/1977,1987/1988 and 1997/1998), and in one case 2 m (1993/94). For the other years, maximum thicknesses are around 1 m or less. The maximum ice thickness is usually reached in April or May. Seasons 1967/1968 and 1970/71 show almost undisturbed ice growth. Season 1981/1982 has an overall increase spanning over most of the season, but shows more variability on monthly and sub-monthly time scales. Season 1997/98 exhibits a disturbed ice evolution with early ice formation, but with two events of suddenly reduced thickness, probably due to breakup of fast ice. Winters with observations were classified according to disturbance of ice thickness evolution (Table 1). As criterion for disturbance we use the standard deviation (σ) of the differences in ice thickness between two consecutive measurements of a winter. Time series of winter seasons during which five or less measurements were done are too short to draw conclusions about the ice evolution as in most cases the reason for the low numbers was not documented. For 2005/2006 and 2006/2007 the situation is different: no or only one drilling could be performed due to unusually warm winters when fast ice developed only sparsely and for a short period of time.
Table 1. Criteria and Brief Descriptions for Classification of Winters According to Sea Ice Thickness Evolutiona
σ is the standard deviation of the ice thickness differences between two consecutive measurements of a winter. The distribution of classes since 1966 is shown in Figure 2b.
number of measurements ≤ 5
0 m < σ < 0.1 m
undisturbed ice evolution
0.1 m ≤ σ < 0.2 m
ice evolution with minor perturbations
0.2 m ≤ σ < 0.35 m
ice evolution with major perturbations
0.35 m ≤ σ
disturbed ice evolution
 The interannual variability in ice thickness evolution is high with no significant trend in the data towards more or less disturbed ice growth (Figure 2b and Table 1: classes 3,4 and 1,2 respectively). However the smooth evolution during the seasons 1967/1968 and 1970/1971 relate to the climatic conditions the Barents Sea experienced during that period while the higher variability with more frequent breakup events is caused by atmospheric conditions with winds from SW.
 To analyze the temporal development of sea ice thickness at Hopen over the last 40 years, we discuss (i) the maximum ice thickness per season, and (ii) statistics of the full time series.
4.1. Maximum Seasonal Ice Thickness
 The maximum ice thickness in a season ranges from less than 0.5 m to 2 m. Linear regression analysis for all seasons with data exhibits a negative trend of −0.098 m per decade for seasonal maximum ice thickness (Figure 2c). The season with no measurable ice (2005/06) is not considered. Although the ice thickness maxima vary highly interannually, the maximum thickness decreased by about a third over the observational period.
4.2. Statistical Analysis
 To eliminate the seasonal cycle which dominates the variability we calculate mean monthly ice thickness anomalies by subtracting monthly climatologies from the monthly averaged observational data. We use the statistical tool Significant Zero Crossings of Derivatives (SiZer) [Chaudhuri and Marron, 1999] to evaluate the presence and significance of trends in the ice thickness record. SiZer first smoothes data at various time scales, and then quantitatively determines which parts of the smoothed features are statistically significant at a given level α. A key idea in SiZer is that significant features are found at different scales, i.e. at different levels of smoothing. Significance is then a function of the level (here, α = 0.05), the size of the smoothing window (specified by a bandwidth) and the time of the signal.
 The SiZer analysis of the time series shows a statistically significant trend on longer time-scales suggesting a steady decline in fast ice thickness during the last 40 years. Due to the irregularity of ice thickness measurements, the number of significant features on shorter time scales is restricted: An increase of ice thickness in the second half of the 1980s is followed by a decrease in the first half of the 1990s. The long-term trend in ice thickness from 1966 to 2007 is −0.11 m per decade. This trend correlates with the development of seasonal maximum ice thickness (Figure 2c). Since the climatology calculated from all observations represents a not necessarily realistic mixture of the different observed ice evolution classes (Figure 2b and Table 1), the SiZer analysis was also tested using data from an undisturbed season (1970/71), scaled to the average maximum ice thickness. The result was very similar to the one using the climatology as described. The reduction in ice thickness coincides with increasing SAT (+0.85°C/decade in 1966–2007) and summer and fall SWT (+0.36°C/decade in 1972–2007 for the August–October average) at Hopen.
 The changes found both in the ice thickness maxima and from the statistical analysis are consistent with negative trends in ice thickness observed in the Arctic Ocean since the early 1990s [e.g., Haas, 2004], a reduction of sea ice extent in the Barents Sea [Meier et al., 2007] and the decrease of sea ice extent for the entire Arctic [e.g., Serreze et al., 2007; Stroeve et al., 2007].
4.3. Relationship of Fast Ice Thickness and Meteorological Data
 The development of fast ice at Hopen is highly influenced by air temperature and wind speed and direction. In years with air temperatures higher than normal, fast ice forms later or breaks up several times during winter, and ice thicknesses remain lower (e.g. years 1982, 1984, 1990, 1997–2007). The dominant wind direction at the meteorological station on Hopen is from the northeast. Ekman drift results in advection of ice at an angle of 30° to the right of the wind in the Arctic. Northeasterly winds therefore push the ice towards the shore, helping to create denser pack ice and to maintain the fast ice. When the wind direction changes to southwesterlies, the ice pack strain is away from shore. Also, as the surface water circulation in the Barents Sea is mainly wind-driven, southwesterly winds bring warmer surface water from the North Atlantic close to Hopen. If such a wind change coincides with a spell of high air temperatures, the fast ice thins and eventually breaks up. It regrows only when winds reverse again or stay calm and temperatures are low.
 During the most recent two seasons (winter 2005/2006 and 2006/2007), air temperatures were anomalously high with only few and short periods of temperatures below −5°C. At the same time, the wind direction changed more often than in previous years. This led to no ice thickness measurements at all in winter 2005/2006 and only one drilling in 2006/2007 due to too little and unstable fast ice.
 Sea ice thickness, measured in situ over 40 years at Hopen exhibit a negative trend, as revealed from both statistical analysis and selection of seasonal maxima. So far, only few similar datasets exist or are accessible. The scientific value of studies like the one presented here will increase once more similar datasets are combined and analyzed together. This and the start of new monitoring at selected locations are also part of studies within the ongoing International Polar Year 2007/08.
 The positive trends in observed SAT and SWT could explain the reduced ice thickness which was revealed by the data analysis. But further analysis and modeling will be necessary to quantify the role of all factors influencing the sea ice at Hopen. In general, changes in sea ice at a single spot might be currently best visible in locations such as Hopen, close to the MIZ of the Arctic sea ice cover. Other Arctic coastal sites with long-term sea-ice monitoring data in Siberia [Polyakov et al., 2002] show less strong trends in ice thickness with maximum thinning or thickening of 1.3 cm per decade (observations from the mid 1930s to 2000), but the local conditions with the Siberian landmass nearby and a setting further away from a deep, non-polar ocean might dominate and stabilize the sea ice evolution.
 The fieldwork conducted by the Hopen wintering teams and the support by Helge Tangen and co-workers of the Meteorological Institute of Norway (met.no) enabling the monitoring since the 1960s are highly acknowledged. Sea ice thickness monitoring at Hopen was initiated in the 1960s by Torgny Vinje (NPI). Hopen meteorological data are kindly supplied by met.no. We thank Victoria Lytle (NPI) for comments on the manuscript, and Olga Pavlova (NPI) for assistance with the map. We are grateful for constructive criticism by two anonymous reviewers. The work of the author and co-authors for this publication is financed by the Norwegian Polar Institute and the University of Tromsø, Norway.