4.1. Dirty Ice Characteristics and Sources: Implications for the Entrainment Process
 Despite the wide range in the coarsest particles (95 percentile) in the dirty ice samples from 12 to 211 μm, all of the samples were finer than the maximum particle size for sea ice entrainment previously suggested as less than 250 μm [Reimnitz et al., 1993b, 1998; Nürnberg et al., 1994; Eicken et al., 2005]. Because anchor ice entrains whatever sediment is available on the seafloor, if coarser particles are present they should be entrained. The lack of sediment coarser than 250 μm however, suggests that such coarse particles cannot be used as criteria for anchor ice. Because the mean and coarsest particle sizes cluster near the finer sizes and include the one sample with shells (L4) that is most surely due to anchor ice, using the presence of coarse grains alone to distinguish anchor ice and suspension freezing is problematic (Figure 4). Principal component analyses of the <45 μm size analyses of these same dirty ice samples indicates a coarse silt-size component that was interpreted to represent anchor ice [Darby et al., 2009]. Perhaps the 250 μm size limit, which was really intended to distinguish sea ice from glacial iceberg IRD, is too large for distinguishing anchor ice and suspension freezing. Thus the presence of sediment coarser than about 100–150 μm might indicate an anchor ice over suspension freezing when glacial ice can be excluded, but the absence of such large grains alone cannot be used to indicate suspension freezing. This is primarily due to the fact that many shallow shelves in the Arctic probably lack coarse sediment and anchor ice would only entrain fine sediment.
 The lack of material coarser than 211 μm in the HOTRAX and LOMROG dirty ice samples however suggest that the entrainment did not occur in the beach or nearshore zone where wave activity during open water conditions would most always produce sandy sediment of this and coarser sizes. All but one of the LOMROG samples and four of the HOTRAX samples contained sediment coarser than 30 μm, and three of these HOTRAX samples were from off Alaska where all of the dirty ice floes contained low sediment concentrations and might be due to suspension freezing. Five of the HOTRAX samples and four of the LOMROG samples contain sediment coarser than 50 μm, including the one with the shells (L4). This coupled with the finding of a coarse silt component [Darby et al., 2009] might suggest that suspension freezing generally involves sediment less than 30–60 μm at the 95 percentile of the coarsest sizes and anchor ice entrains coarser particles than this. Size alone is an inadequate indicator of the two entrainment processes.
 The Fe grain fingerprinting indicates that most of the 2005 and 2007 dirty ice floes originated in the shallow coastal North American sector. The floes originating in northern Canada would most likely become entrained in the BG for three to as many as 20 or more years depending on the number of rotations they make before merging with the TPD and drifting toward Fram Strait. Although such multiple rotations in the BG are less likely today with the extensive melting experienced in the last decade [e.g., Kwok et al., 2009]. The drift trajectories based on drift buoys show that the BG was fairly large in diameter and was close to the TPD during the two to three years prior to 2005 (Figure 1). This is conducive for floes to move out of the BG and into the TPD within three years.
 Samples H1-6 and H1-7 near Alaska are from the Bering Strait according to Fe grain matches and this agrees with the surface drift in this area where the Alaska Coastal Current moves north along the west coast of Alaska transporting Bering Strait ice to the H1-6 site aided by offshore eddies once the floes reach the northern coast of Alaska (Figure 1) [Pickart, 2004; Pickart et al., 2005]. The low concentration of sediment in all floes off northern Alaska and the coarsest particles less than 60 μm (mean size <30 μm) suggest a suspension freezing origin.
 The central Arctic dirty ice samples sourced to Canada based on Fe grain fingerprinting (H3-5, H3-8, H3-10, H3-11 and all LOMROG samples) probably were initially drifting in the BG and then merged into the TPD. The net drift for these samples are indicated by the dashed yellow path (Figure 1) while those samples matched to Russian sources (H3-5 and H3-9) drifted east along the coast, through Vilkitski Strait past the Lena River before turning north near the New Siberian Islands as indicated by the net drift (red dashed line in Figure 1) and the buoy drift trajectories for the two years preceding the floes reached the sampling locations on the Alpha Ridge. This drift pattern has been documented for dirty sea ice floes in earlier studies [Pfirman et al., 1997; Wahsner et al., 1999; Eicken et al., 2000; Vogt and Knies, 2008].
 Although values up to 70% smectite in the <2 μm fraction have been reported for the mouths of the Ob and Yenisey Rivers [Wahsner et al., 1999], values above 25% are uncommon in sea ice samples [Dethleff et al., 2000]. All of the 2005 HOTRAX dirty ice samples show less than 26% smectite (summed and rescaled to 100%) except for one sample, H1-6 at 28% smectite collected off Alaska (Table 1). All of the <45 μm XRD LOMROG sea ice samples show very high smectite values (46–54% of clay and mica rescaled to 100% in Table 1) and higher vermiculite than any of the source area samples, including the Ob and Yenisey Rivers. Thus neither the bulk mineralogy, nor clay mineralogy clearly indicate sources in the data set. Yet the summed and rescaled clay mineral percents shown in Figure 1 for the HOTRAX dirty ice samples could be linked to several possible source areas from Alaska to the Ob River, but none that have the exact same mix of clay mineral species (Table 1). The LOMROG dirty ice samples (L1–3) contain similar illite as the Victoria/Banks Island samples to which they all matched according to the Fe grain fingerprinting, but contained higher muscovite as well as much higher smectite and kaolinite than this source area (Table 1). These sources are tills from important Late Pleistocene ice streams from the Laurentide ice sheet outlet to the Arctic Ocean [Stokes et al., 2005]. Thus large amounts of this sediment must have been transported west through McClure Strait and deposited on the Canadian shelves near Banks Island where it could be mixed with other sediments containing different fine-grained clays without changing the sand-size Fe grains. From here the sediment could be entrained by sea ice. Thus based on the Fe grain provenance and the drift patterns in place for the years preceding the sampling of the 2005/2007 dirty ice floes (Figure 1), the Banks Island area source for the LOMROG samples and the Canadian sources for the HOTRAX samples agrees with known drift patterns in this area.
 Flaw leads or polynya conditions occur throughout the circum-Arctic coastal areas and except for the Cape Bathurst Polynya near Banks Island are more common and more extensive in the Russian Arctic [Hannah et al., 2009; Stringer and Groves, 1991; Overland and Guest, 1991; Dmitrenko et al., 2005]. Thus the abundance of dirty ice from North America is unexpected if we assume that flaw leads and polynyas are the primary factor in sea ice entrainment. We can only be certain that one of these floes, L-4, is due to anchor ice, and this floe is sourced to the Banks Island area, probably the shallow (<60 m) areas around this island or fringing McClure Strait or Amundsen Gulf that coincide with the Cape Bathurst Polynya [Arrigo and van Dijken, 2004; Barber and Massom, 2007] (Figure 1). The presence of shells and seaweeds in the ice as well as the nearly 10 cm thick sediment concentration over more than 5–10 m2 are the primary indicators of an anchor ice origin for this floe. In addition to the sites sampled in 2005/2007, nearly continuous sightings of similar heavy concentrations of dirty ice were observed along the three areas of sampling (Figure 1). We cannot be certain that all of these originated as anchor ice, but the amounts and thicknesses observed in many of them would be difficult to account for by melt-pond concentration of dilute dirty ice from suspension freezing. The dispersed nature of this form of sea ice entrainment would require much larger melt ponds than generally observed. In order to produce the thick sediment concentrations observed, sediment from an estimated area of more than 100–500 m2 of low sediment concentration due to suspension freezing would have to flow into an area of about 5–10 m2 to produce the thick concentrations observed for the dirty ice in 2005/2007. The slopes for such drainage just do not exist on the packice. Thus the high concentrations of sediment in the dirty ice floes observed in 2005/2007 were probably the result of anchor ice entrainment. Yet, Reimnitz et al.  concluded that anchor ice only forms in sandy sediments where the particles are large enough to overcome buoyancy and drag effects as the ice grows and the ice can form in the interstitial pore spaces and thus entrain sediment to greater depths below the seafloor. We contend that there are insufficient observations to exclude the formation of anchor ice in silty sediments where the sediment cohesion and interstitial ice in the sediment pore space counters these lift effects until the ice is sufficiently thick to lift off large amounts of the bottom sediment. While only sandy sediments have thus far been observed to have anchor ice attached to the bottom, there is no reason to believe that the higher pore volumes in fine-grained silt and clay with up to 80%, which is more than twice that of sandy sediments, could not also support anchor ice.
 Some of the dirty floes suspected to be anchor ice are sourced to areas with rare or small flaw leads such as northern Ellesmere Island (Figure 5). If these floes are anchor ice, then the mechanism of initiating bottom water freezing in the absence of flaw leads during the winter needs to be explained. We are assuming that very little anchor ice will form in shallow water during the summer months due to the generally above freezing, open surface waters at this time. One scenario that does not require open water during the winter and the continuous advection of super-cooled water to the bottom is that of seed-ice formation. This scenario only requires the formation of ice within the bottom sediment at some time during the year, preferably in the fall or early winter before the pack ice forms so that bottom ice can grow all winter. If super-cooled seawater is not advected to the bottom for ice initiation, then the flux of fresh groundwater from the continent through the underlying sediments might be the cause of ice formation because this fresh water would freeze at the ambient seawater temperatures during the late fall and winter (−1°C to −2°C [Aagaard, 1984; Reimnitz et al., 1987]). Once ice forms in the sediment pore spaces it will provide a seed for further ice formation from seawater, which is at the freezing point. Unfortunately there is virtually no observation of groundwater flux into the Arctic Ocean so this process remains speculative.
 Another possibility for sediment entrainment is sediment frozen into the keels of sea ice pressure ridges as they become grounded. Pressure ridges can extend several meters below the pack ice and can become grounded or plough bottom sediment, although observations on this are very scarce to non-existent. There are no observations or studies of whether this process could entrain significant amounts of sediment although there have been speculation that it does [Forbes and Taylor, 1994, and references therein]. Unless the keels of such pressure ridges remain in contact for long periods of time to allow the ice to act as seed crystals and freeze pore water in the sediment, entrainment is unlikely. If ice is grounded for long periods in shallow water, there is the problem of whether it will become free before Spring melting occurs as open water could occur offshore of this ice, which could promote its melting before it can be incorporated in any ice pack drifting into the central Arctic [Reimnitz et al., 1994]. Another problem is that the sediment will be overlain by several meters of sea ice and all of this has to melt in subsequent summers before this sediment can make its way to the surface of the pack ice where it can be observed and sampled.
 The near continuous occurrence of dirty ice over hundreds of kilometers suggests that the dirty ice was entrained and introduced into the ice drift at about the same time. If these dirty floes originated as anchor ice, this also needs explanation because only suspension freezing has been observed to produce such widespread events [Eicken et al., 2000]. Anchor ice will become incorporated into the overlying packice only when it is sufficiently thick so that its buoyancy overcomes its attachment to the seafloor. There is no observation of anchor ice actually lifting off the bottom so we can only speculate that if the ice begins to grow under favorable conditions, that the conditions for ice growth should be fairly uniform for a particular coastal region regardless of whether the ice initiated due to groundwater flux into the offshore sediments or by seed crystals due to open water and convection of super cooled water to the bottom in the Fall. Thus the ice should reach a critical thickness at about the same time over rather large areas. The role of sudden events, such as Spring tides, internal waves, or currents in affecting this process has yet to be explored, but might also provide an explanation for near simultaneous lift-off of anchor ice.
4.2. Reworking of Sediment in Sea Ice
 The longer a floe is drifting, the greater the chance that it will merge with floes from other sources through the constant opening of leads and collision of floes forming pressure ridges [Darby, 2003]. When floes merge, the sediment can also mix due to summer melting and the flow of sediment into melt ponds. Based on the number of dirty sea ice samples containing multiple sources in this study and previous studies [Darby, 2003], a little more than a quarter to a third of floes result in mixing of entrained sediment. Of course there is another possibility that there are source areas as yet unsampled by us that contain sediment from multiple sources that would produce the results in Figure 5. Of the more than 450 circum-Arctic source samples we have never seen one with Fe grains that could be matched to more than one source above the minimum level of significance.
 Regardless of the Fe grain source, all of the dirty ice samples contained different mineral contents than the indicated or nearby source samples. Although for dirty ice samples with multiple Fe grain sources, a mix of two or more source samples combined with selective sorting of the clays could result in the measured clay mineral contents of these floes. None of the single-sourced Fe grain sources contained the same mix of clays or non-clays as any of the dirty ice samples, thus the sediment in these floes must have been altered during transport across the Arctic or the location of the actual entrainment was not analyzed for mineralogy. The fact that nearly all of the HOTRAX dirty ice samples contained a similar clay mineralogy suggests that the summer melting and meltwater flow into melt ponds as well as occasional partial drainage of these melt ponds into the ocean might be altering the clay mix. This might occur if certain sized clays are entrained by meltwater while others are not. Similarly, density and size differences in the non-clay fraction could cause hydraulic sorting as well. If so, this introduces another problem in the use of mineral abundances for sea ice provenance. This would not cause a change in the Fe grain composition because the Fe minerals are of similar size with densities above 4.5 and would respond similarly to hydraulic sorting.