Distributions of freshwater (sea-ice melt and runoff) were investigated along inshore-offshore sections in southwestern Hudson Bay for fall conditions. Conductivity-temperature-density profiles and bottle samples collected for salinity, oxygen isotope (δ18O), and colored dissolved organic matter (CDOM) analyses were used to discriminate between contributions of river water (RW) and sea-ice melt (SIM). Stations had a fresh summer surface mixed layer 5–25 m thick overlying a cold subsurface layer indicative of the previous winter's polar mixed layer (PML). The fraction of RW decreased strongly with distance from shore, while the opposite was true for SIM. The majority of RW was constrained in a coastal domain within 100–150 km from shore, which, because of high alongshore velocities, accounts for the majority of freshwater and volume transports. On the basis of freshwater inventories and composition, brine and RW accumulate in the PML over winter because of ice formation and downward mixing. The summer surface circulation results in an annual net export of SIM from the region. Residence times for freshwater components in the southwestern sector of the bay, based on currents derived from a 3-D ocean model for Hudson Bay, are about 1–10 months, implying rapid transit of freshwater. Despite the short residence time for RW (1–3 months), CDOM is significantly photobleached and provides an unreliable tracer for RW. Photobleaching represents an important sink for dissolved organic carbon entering from rivers and could, in part, explain why Hudson Bay is only a minor sink for atmospheric CO2 in the open water season.
 Hudson Bay (Canada) (Figure 1) is the world's largest inland sea (0.83 × 106 km2) and has a drainage basin that is almost four times larger by area (∼3.1 × 106 km2) [Prinsenberg, 1984, 1986a]. The watershed contributes a cumulative annual discharge of about 700 km3 [Prinsenberg, 1986a; Dery et al., 2005] equivalent to a freshwater yield of 0.8 m, which is more than twofold the yield the Arctic Ocean receives. The river water (RW) yield is accompanied by an even larger seasonal freshwater input from sea-ice melt (SIM), estimated at 140 cm [Prinsenberg, 1988]. It is likely that the contribution from SIM may be even higher, at least locally, because of ice volume enhanced by ridging [Prinsenberg, 1988] and ice advection which tends to redistribute ice from northwest to southeast [Saucier et al., 2004]. The total seasonal freshwater yield is estimated at ∼3 m in central Hudson Bay [Prinsenberg, 1988]. These large freshwater fluxes to the surface of Hudson Bay produce a characteristic two-layer circulation and a strong summer pycnocline located generally between 5 and 25 m [Prinsenberg, 1986a; Harvey et al., 1997; Granskog et al., 2007].
 Clearly, in considering any process related to freshwater in Hudson Bay, both RW and SIM need to be considered. The distribution, location of entry, and timing (Figure 2) of these two freshwater sources differ. Riverine inflow to the perimeter supports a large-scale cyclonic estuarine circulation which commences with fresher, less dense waters entering northwestern Hudson Bay from Foxe Basin [Tan and Strain, 1996]. Longshore wind stress also contributes to the nearshore circulation, at times opposing or supporting the estuarine forcing, but the net surface flow around the margin of Hudson Bay forms a robust, persistent cyclonic pattern. As the water transports, it mixes with RW from numerous local rivers and incorporates SIM before eventually exiting the Bay at the northeast corner. In winter the circulation slows [Saucier et al., 2004] in response to the reduced effect of the winds, decreased inflow and withdrawal of freshwater by accumulating sea ice. The brine rejected from the ice contributes to a deepening of the polar mixed layer (PML) which may extend to depths of over 90 m by the end of winter in the west and possibly 60 m in the southeast [Prinsenberg, 1986a]. The deep waters also circulate generally in a cyclonic pattern but more sluggishly, with a roughly estimated residence time between 4 and 14 years [Pett and Roff, 1982], compared to what is likely <1 year for water in the surface layer [Harvey et al., 1997] (estimated using Prinsenberg's [1986b] current measurements). The input of RW and SIM to the southwestern sector of Hudson Bay both peak in June (Figure 2), but SIM exceeds runoff at this time (assuming the ice melts in situ and is evenly spread over the entire surface). Whether RW or SIM dominates at a given location or time depends on the initial conditions (after winter mixing), the seasonal additions, and how each freshwater component spreads over the surface. However, historical measurements of salinity [e.g., Prinsenberg, 1988] permit only a view of the net effect of runoff and sea-ice melt or freezing.
 The hydrological cycle and the ice growth/melt cycle both contribute to the freshwater budgets of polar estuaries [Macdonald, 2000], and thus ultimately control stratification, mixing and biological production. This is highlighted in Hudson Bay, where stratification and colored dissolved organic matter (CDOM) provide important controls on where primary production can occur [see Granskog et al., 2007 and references therein]. Clearly, understanding the sources, transport and residence times of RW and SIM is important in the context of productivity and biogeochemical functioning of the Bay.
 A full understanding of freshwater cycling requires the ability to distinguish between SIM and RW. Few studies have tackled this problem in Hudson Bay, and those have been limited to the northern Bay [e.g., Tan and Strain, 1980, 1996; Jones and Anderson, 1994], a region where large-scale advection rather than local processes contributes to freshwater dynamics. Stable oxygen isotopic composition of water (δ18O) is well understood in its application to distinguish freshwater sources in coastal polar environments [Macdonald et al., 1995]. Recently, Granskog et al.  proposed that riverine CDOM might provide an additional tool with power to discriminate freshwater sources in the Bay.
 In this paper we use salinity, and oxygen isotope (δ18O) data collected on three transects across the southwestern nearshore of Hudson Bay in October 2005 to estimate the distribution of SIM and RW and their relative contributions to total freshwater. With the aid of a 3-D model [Saucier et al., 2004] we examine the fluxes and residence time of these freshwater components into and out of the region bounded by the transects. Finally, on the basis of δ18O-derived estimates of RW and its residence time, we evaluate whether or not CDOM behaves conservatively as proposed by Granskog et al. .
2. Material and Methods
2.1. Field Sampling
 Data were collected in southwestern Hudson Bay during the ArcticNet (http://www.arcticnet-ulaval.ca) 0502 expedition, onboard the Canadian research icebreaker CCGS Amundsen, between 6 and 16 October 2005. Three cross-shelf sections A, B, and C were sampled (Figure 1). Section B (10–11 October), extending offshore from Nelson River, and section C (6–7 October), extending offshore from Winisk River were densely spaced, while section A (13–16 October), off Churchill River, had only four stations because of time constraints (Figure 1). Three to six bottle depths were sampled, with two to three targeted in the mixed layer (surface, 10 and 25 m). Where water depths were 50 m or less, three or four depths were sampled. Section C included shallow, nearshore stations at the mouth of the Winisk River (<10 m) sampled by Zodiac, out to offshore waters (depth >180 m). Section B started in the wide and shallow Nelson River estuary, such that the station nearest the mouth of the river (at 21 m depth) was 50 km from the river's point of entry. The station closest to shore along section A was located about 30 km north of the mouth of the Churchill River.
 The runoff in the region between Churchill River and James Bay (see Figure 1) is available from the Hydroclimatological Data Retrieval Program (HYDAT) database (Environment Canada) for different rivers for different time periods (summarized in Figure 2). It is important to note that in 2005, the Winisk River had above average monthly discharges in the latter half of the year. For Churchill and Nelson Rivers the monthly discharges from July to October 2005 were all-time highs on record, almost double the long-term average.
 The ship's rosette was equipped with 12-L Niskin bottles and a SBE-911plus (Sea-Bird Electronics, Inc.) conductivity-temperature-depth (CTD) sensor, from which data were averaged over 1 m intervals. Before each hydrocast, and particularly prior to disturbance of the mixed layer by the ship, a surface water sample was collected with the use of a prerinsed bucket from the foredeck of the ship. Salinity, δ18O, and CDOM were drawn directly from the Niskin bottles on the rosette or the bucket into bottles that were rinsed three times with sample water before being filled [Granskog et al., 2007]. For work from the zodiac, samples were collected into amber acid-rinsed 2-L HDPE bottles, rinsed three times with sample water before being filled. These bottles were kept refrigerated in the dark and subsampled for salinity, δ18O, and CDOM within a few hours of collection. For details on sampling procedures see Granskog et al. .
2.2. Analytical Methods
 Samples for oxygen isotopic composition were collected from the same sample volume directly into 20 mL borosilicate vials (rinsed three times before filling), closed tightly and sealed with flexible Parafilm to minimize evaporation. Samples were stored at +4°C. Samples were analyzed using a Gasbench attached to a DeltaPlus XP isotope ratio mass spectrometer (ThermoFinnigan, Germany) at the G.G. Hatch Isotope Laboratories (University of Ottawa). Subsamples (0.6 mL) were pipetted into an Exetainer, and, along with internal standards, flushed with a gas mixture of 2% CO2 in helium using the Gasbench. Exetainers were left to equilibrate at +25°C for 18 h minimum. Values are expressed in standard δ18O notation with the V-SMOW (Vienna Standard Mean Seawater) as reference value. Analytical precision was ±0.15‰.
 Samples for colored dissolved organic matter (CDOM; absorption at 355 nm (a355 (m−1)) and spectral slope coefficient for 250 to 400 nm (SCDOM (nm−1)) were determined as described by Granskog et al. . Salinity was measured onboard the ship with a Guildline Autosal 8400 salinometer with a precision better than 0.002. Samples were standardized against IAPSO Standard Sea Water.
2.3. Determining Water Composition Using Salinity and δ18O
 The use of paired δ18O-salinity measurements to quantify freshwater components in polar oceans (RW, SIM) is well known (see Östlund and Hut  and Macdonald et al.  for detailed descriptions). Briefly, the technique involves solving a set of three linear equations that incorporate three end-members (primary water types), in this case, seawater, river water, and sea-ice melt
where the subscripts denote the primary water types, F is the fraction of each primary water type, and S and δ refer to the salinity and δ18O values, respectively, of the subscripted end-members. The equations can be solved for any sample with salinity Smeas and δ18O value δmeas. Positive solutions for SIM reflect the addition of sea-ice melt whereas negative solutions reflect the loss of freshwater due to freezing (i.e., addition of brine). Appropriate assignment of the properties to each of the primary water types is required and may vary from region to region [e.g., Macdonald et al., 1995]. The basis of choice for the data presented here is outlined below.
2.4. Ice-Ocean Model Fluxes
 As our observations provide only three snapshot sections, we opted to set these observations in a wider context by running a three-dimensional ice-ocean model specifically developed for Hudson Bay using appropriate forcing for the year 2005 (see Saucier et al.  for details). Here, our objective is to provide modeled currents which, together with the freshwater composition data, can be used to derive RW and SIM volume fluxes across the three sections sampled (Figure 1). From the model, hourly currents (u, v) are calculated on an Arakawa-C grid (10 km horizontal resolution and 10 m vertical resolution). These currents, which represent the mean current of the 10 × 10 km cell over the thickness of the cell, have then been interpolated to align at 10 km horizontal spacing along the sections that were sampled. From u and v the current normal to the section has been calculated and filtered with two iterations of a low-pass fourth-order Butterworth filter (30 h) [Roberts and Roberts, 1978]. All signals with a frequency higher than or equal to 30 h have been filtered out because we focus on the low-frequency circulation. Volume fluxes are derived from the current normal to the section and cross-section area of the corresponding cell (taking into account water level and bathymetry for the surface and bottommost cells). Volume fluxes are multiplied by the observed fields (fractions of SIM and RW), which have been interpolated to the model output along the sections, to derive estimates for freshwater fluxes across the sections.
3. Results and Discussion
3.1. Hydrography Along Sections
 The distribution of salinity along sections (Figure 3) shows that nearshore waters were substantially impacted by freshwater, most likely by runoff that has been constrained to the perimeter of the bay because of the strong alongshore cyclonic circulation [Prinsenberg, 1986a; Saucier et al., 2004; Granskog et al., 2007]. Density (not shown) closely followed the isohalines, suggesting that salinity dominated density stratification in Hudson Bay, which is typical for polar waters. Lowest salinities were found at the nearshore end of sections, with considerable freshening (<29) occurring to depths of 40 m along section C (Figure 3). A distinct summer mixed layer (SML) was present at all stations. The SML initially become thicker toward the offshore along sections B and C (Figure 3), but thereafter leveled off at a depth of 16–25 m, in agreement with observations in eastern Hudson Bay [Harvey et al., 1997]. A temperature-salinity diagram (Figure 4) indicates that surface water temperature increased with decreasing salinity, with temperatures up to 7.0°C in the nearshore of section C and around 5.5°C in section B. Coldest nearshore waters, about 4.3°C, were found in the SML of section A. Cold subsurface water, at or close to freezing, was observed at intermediate salinities below the SML (Figures 4 and 5) indicating layers recently affected by sea-ice formation.
 Remnants of the PML were observed as subsurface layers with near-freezing temperatures at salinities from about 32.3 to about 33.3 (Figure 4), at depths ranging from about 50 to 90 m (Figure 5). These depths correspond well with the observations of winter pycnocline depth in western Hudson Bay, which can reach depths of at least 93.5 m, and observations of subsurface cold layers in summer [Prinsenberg, 1986a]. Cold subsurface layers have also been observed in recent years at the northern edge of our study area [Harvey et al., 2006]. Along section C, remnants of the PML were clearly present at all stations except those in water shallower than 90 m. The absence of the PML along the southwest inner shelf area may be due to the large winter runoff from the Nelson River (Figure 2), which is “upstream,” with the freshwater input preventing brine convection, as observed over the inner Mackenzie shelf in the Beaufort Sea [Macdonald et al., 1995] and off the Churchill Estuary [Kuzyk et al., 2008]. PML waters in the nearshore may have also been transported out of the region with the strong alongshore current (shown later) by the time of our sampling. Remnants of the PML were also clearly evident at a few midsection stations along sections A and B (Figures 4 and 5). At the offshore stations on these sections, the PML was apparently replaced by advection of intermediate waters, as the cold subsurface layer was observed earlier in the same year (August 2005; M. Harvey, Fisheries and Oceans Canada, personal communication, 2007), but was no longer present in October 2005. Prinsenberg [1986a] observed that the PML was intermittently replaced by warmer water in early spring in the Churchill area. In October 2005, the salinity of the PML layer along section A was about 33.2, and for sections B and C, about 32.8 and 32.3, respectively (Table 1). In August 2005, salinity in the subsurface PML ranged from 32.8 to 33.4 (40 to 100 m depth) in northwestern Hudson Bay (M. Harvey, Fisheries and Oceans Canada, personal communication, 2007), corresponding well to our observations in October 2005 off Churchill. However, we have only a limited number of water samples within the layer available to define PML composition (Table 1). The decrease in PML salinity from northwest to southeast is also apparent in the compiled salinity data for the Bay presented by Barber . Observations in spring 2005 off Churchill showed that extensive freezing in the flaw lead on the western shore resulted in surface salinities as high as 33.8 (at freezing) in February–March 2005 [Kuzyk et al., 2008]. Changes in PML salinity therefore reflect the general pattern that more ice is produced in the northwest [Saucier et al., 2004] and less ice formed and/or more freshwater accumulated into the PML in the southeast.
Table 1. Properties of Primary Water Types, Seawater, Sea-Ice Melt, and River Water Used in This Studya
The SW type represents the composition of the polar mixed layer present at the sections.
section A (n = 3)
33.2 ± 0.2
−2.0 ± 0.2
section B (n = 3)
32.8 ± 0.1
−2.2 ± 0.3
section C (n = 10)
32.3 ± 0.2
−2.7 ± 0.1
section B and C
 Our observations indicate that in some regions of Hudson Bay, here exemplified especially along section C (Figure 5), the previous winter's PML can remain intact at depth throughout summer until cooling and freezing commences in the fall. It is likely that the strong summer stratification, initiated by ice melt and runoff, prevents vertical mixing and keeps the subsurface PML intact. In the northern part of our study area the previous winter's PML has been partly replaced by or mixed into intermediate waters that likely enter the Bay from the northern channels [e.g., Tan and Strain, 1996]. We also note that northwestern Hudson Bay is more prone to mixing because of smaller local inputs of runoff and greater ice production in the flaw lead/polynya system that forms along its western shore [e.g., Saucier et al., 2004].
3.2. Total Freshwater in the Water Column
 To estimate the freshwater content in the water column we need to choose an appropriate reference salinity. As in other similar circumstances [Macdonald et al., 1999, 2002], we assume here that the water identified as remnant from the previous winter's PML represents the conditions of the water column from surface down to the bottom of the PML in late winter before ice melt and spring freshet initiate stratification. Furthermore, we assume that the brackish surface water observed during sampling is a product of mixing between this PML water and the two freshwater sources that have entered the system consequent to freshet and sea-ice melt. In this model, local freshwater additions during spring/summer are evaluated against the PML salinity baseline. Using the PML baseline salinity for each section (Table 1) together with a PML depth of 90 m, we estimate the net input (equivalent height) of freshwater at the time of sampling, which can include in situ melting during summer, the addition of local river runoff and the advection of seawater containing these components (if salinity is above our selected reference salinity).
 A larger freshwater inventory is present at stations in the nearshore areas (Figure 3), with maximum freshwater content found at the 50 m isobath. All cross-shelf sections can be divided into a nearshore domain (distance < 100–150 km to shore) with more freshwater, and an offshore domain (>100–150 km) with less freshwater. In the nearshore domain the total freshwater equivalent height is on average >4 (A section), 3.5 (B section), and 5.2 m (C section). In the offshore domain corresponding values are ∼3, 2.4, and 2.5 m, respectively. In the offshore the freshwater inventory agrees fairly well with earlier observations and the expected seasonal input from RW and SIM combined [Prinsenberg, 1986a, 1988]. However, in the nearshore there is far more freshwater than expected from simple Bay-wide averages [cf. Prinsenberg, 1988], suggesting that the runoff constrained or accumulated here is an important, but overlooked, component of the freshwater transport.
3.3. Components of Freshwater (River Water and Sea-Ice Melt)
 For the data collected in early October 2005 (Figure 6), salinity and δ18O strongly covary, especially at lower salinities, indicating that mixing between isotopically light river water and isotopically heavy marine water explains much of the δ18O-salinity distribution. However, the scatter in the data normal to the major axis of variation reveals the importance of sea-ice melt or freezing. Accordingly, Figure 6 can be viewed simply as the mixing of three water masses (RW, SIM and SW), where SIM may exhibit positive values where sea ice has melted, and negative values where sea ice has formed, withdrawing freshwater and leaving brine behind [cf. Macdonald et al., 1995]. Therefore, samples offset upward from the general mixing line contain SIM, whereas those offset downward have had salt added to them (negative SIM). Sections A and B have samples that are significantly affected by SIM, which may be contributed partly by SIM-rich surface inflow from Foxe Basin [Tan and Strain, 1996], and partly by melting of ice within the Bay. In the salinity range 25–29.5 along section C, fresher samples are clearly dominated by RW as they lie closer to a mixing line between river water and PML water. Along section B, several samples with 28.5–30 salinity are apparently affected by brine addition as are samples at high salinity along section C, which are offset downward from the mixing line.
 As noted in the methods, to determine freshwater components using equations (1)–(3) requires a reasonable assignment of appropriate primary water masses (Table 1), for which we have little precedent in Hudson Bay. Here we choose end-members internally consistent with our HB database for the purpose of examining the change in composition against the PML background [see, e.g., Macdonald et al., 2002], assumed to have been pervasive down to 90 m at the end of winter (∼April–May).
3.4. Saline End-Member
 For the SW water mass (equations (1)–(3)), we use properties representative of the previous winter's PML, i.e., subsurface water masses lying close to the freezing point (Figures 4 and 5 and Table 1), for each respective section. Using the PML composition as the saline end-member disregards any RW and SIM that have been mixed into the PML during previous years, thereby neglecting the effects of variable residence times and processes that maintain the PML. Presumably the properties of the PML will differ slightly from year to year and place to place depending on water column properties prior to freezeup, the amount of ice grown in winter [cf. Macdonald et al., 1995, 2002], and the amount and composition of freshwater incorporated in fall/winter, but this does not affect the seasonal change in freshwater that we estimate here.
3.5. Sea-Ice Melt End-Member
 SIM δ18O values are higher than those of the water from which the ice formed by 2–3‰ because of fractionation during freezing [see Macdonald et al., 1995]; for Hudson Bay, the only published data show a fractionation of 2.2‰ for landfast sea ice in the Churchill Estuary [Kuzyk et al., 2008]. Here we assume most of the sea ice has formed in western Hudson Bay [Saucier et al., 2004] predominantly from water of the PML composition (SW in Table 1), which together with a fractionation of 2.2‰ would imply a SIM δ18O composition of −0.5 to +0.2 ‰. Under these assumptions, a value of 0.0‰ therefore describes sea ice formed in Hudson Bay (Table 1).
3.6. River Water End-Member
 For the RW end-member, observations are very sparse, and there is likely to be variation between rivers and between seasons. In our sector of Hudson Bay the main rivers are, from north to south, the Churchill, Nelson, Hayes, Severn and Winisk. Along section A, waters from Chesterfield Inlet north of Churchill may also contribute. By itself, the Nelson River contributes between 50 and 80% of the combined discharge of these rivers, the contribution being highest in winter (Figure 2) [see also Dery et al., 2005]. On the basis of samples our group has collected in 2005–2006 in Churchill, Nelson, Hayes and Winisk Rivers, the δ18O value for runoff in the region lies between −11 and −13‰ (Table 1). Our samples come almost exclusively from the June–October period. However, the Nelson River drains a large watershed (∼1 × 106 km2), with large reservoirs due to hydropower development resulting in little seasonal variation in δ18O composition between May and November (monthly δ18O values from −11.2 to −10.2‰; L. Wassenaar, Environment Canada, personal communication, 2008). Similarly, data from Churchill River between May and September show little variation (δ18O from −13.2 to −12.3‰; L. Wassenaar, Environment Canada, personal communication, 2008), but lower values have been observed during snowmelt [see Kuzyk et al., 2008]. It is likely that, given strong cyclonic circulation, the freshet signal (low δ18O) has been transported east of the region and we consider the June–July values to be representative for the time of year we have sampled. The observed δ18O- salinity relationship (Figure 6) implies that the zero intercept is about −11.3‰ (−11.7 to −11.0 in the 95% confidence interval) for section C, further implying that the Nelson River contributes the bulk of the river water southeast of the Nelson River estuary. For sections A and B a value of −13‰ is plausible for the runoff at the time of sampling, based on samples collected in rivers upstream in the cyclonic circulation.
3.7. Distribution of Freshwater Components Along Sections
 RW is strongly concentrated into the nearshore waters, especially for sections B and C (Figure 7), although for section A, a nearshore concentration may be simply not well resolved because of the less frequent sampling not resolving nearshore features. The majority of RW is constrained within 100 km of shore, where it is also mixed more deeply into the water column than freshwater in the offshore. This may indicate stronger mixing or continued RW inputs to the PML during winter. SIM becomes the more significant freshwater source in the offshore domain and is also more constrained to the surface layer than runoff in the nearshore (Figure 8), indicating less mixing in the offshore regions as proposed by Griffiths et al. . Note that the sum of freshwater presented in Figures 7 and 8 is not necessarily identical to the total freshwater content (Figure 3), as the latter was estimated from CTD casts with much better vertical resolution than data using bottle samples which have been fitted to match the salinity profiles from the CTD.
 The difference in the SW end-member (i.e., PML) composition between sections A, B, and C (Table 1) can be considered as a change in SIM and RW content as the subsurface waters traverse cyclonically around the Bay, but with a longer residence time than the surface waters [Pett and Roff, 1982; Harvey et al., 1997]. The SW end-member compositions on sections B and C lie to the right hand side of mixing lines with RW (if section A SW end-member is considered the starting point; see Figure 6), implying that the PML accumulates RW and brine (negative SIM) along the cyclonic transport path. This is consistent with a decoupling of the sea-ice formation and melt processes, with surface melt being partly or completely carried out of the region before the following freezeup the next fall. The loss of SIM by this transport process is also implied by equivalent amounts of RW and SIM in the offshore at freezeup in October 2005 even though SIM should govern the offshore freshwater budget according to Prinsenberg . The result is that while some of the SIM has escaped by that time, there are still about equal amounts of RW and SIM available to be entrained into the PML during fall freezeup. Given that brine formation leads to convection whereas sea-ice melting together with river input leads to stratification and advection, it is likely that, on average, ice formation is more strongly preserved in the PML and sea-ice melt is more strongly advected away. Therefore sea-ice formation and melt do not simply balance over the season [cf. Prinsenberg, 1988] producing no net change annually but, rather, this salt pump contributes significantly to the separation of salt and freshwater products of sea ice, which then supports net stratification (buoyancy) and thereby reinforces the estuarine circulation in the Bay. We note that the cyclonic circulation of sea ice [Saucier et al., 2004] also controls where the sea ice melts (e.g., in southeast Hudson Bay), and therefore ice dynamics likely play a role on the export of SIM.
 The estimates presented above are sensitive to the values chosen for end-member composition (Table 1). For example, varying the RW δ18O assignment between −11 and −13‰, or SW within the range shown in Table 1, induce changes in the RW and SIM inventories that are <0.5 m. This variation is less than 15% for RW inventories, but significantly higher for SIM estimates (sometimes accompanied with a change in sign), as the SIM inventories are much lower. Variation in SIM composition, within the estimated range, has only a marginal effect on all estimates. Despite the variations related to end-member composition, the general patterns of the results, such as runoff-dominated nearshore domains, and similar amounts of runoff and ice melt in the offshore, remain valid. However, for SIM estimates the respective changes can be significant, especially in nearshore regions where the absolute values of SIM content are near zero (Figure 8). Here SIM values can become either positive or negative, although always close to zero, suggesting that ice melt and brine addition are both possible. Nevertheless, the SIM values are generally much lower than Bay-wide averages of gross ice melt, much of which is thought to occur in the southeastern side of the Bay (in June–July) [Prinsenberg, 1988; Saucier et al., 2004]. We infer from this that the density-driven circulation together with long shore wind stress which results in the cyclonic transport of freshwater, combined with spring/summer sea ice advection from the area prior to melt, effectively removes a large component of the summer's SIM by October and even more before fall freezeup sets in.
 How do the inventories and distributions of components of freshwater displayed in Figures 7 and 8 compare with other locations? For the Beaufort Sea and Mackenzie Shelf of the Arctic Ocean, where comparable efforts have been made to examine RW and SIM inventories along transects [Macdonald et al., 1989, 1995, 1999, 2002], similar amounts of RW (2–5 m) and SIM (0–2 m) were found in ice-free conditions at the end of summer. Distributions show some similarity between the two regions with higher amounts of RW near to shore and higher amounts of SIM away from shore. However, the nearshore transport system and its impact on freshwater distributions, as discussed here, seems to be uniquely important to Hudson Bay. Our interpretation is that this transport leads to a rapid flushing of freshwater from the Bay and an entrainment of SIM into that transport which then allows for a deeper PML to be produced in winter. In the Beaufort Sea and on the Mackenzie Shelf, the inventories of freshwater appear sufficient to restrict the PML to about 50 m for much of the region. Where divergence over the middle (Mackenzie) shelf favors large amounts of ice production in winter, dense water may form, but this then feeds into the Arctic halocline through export along the shelf bottom rather than producing a deep PML.
3.8. Fluxes and Residence Time of River Water and Sea-Ice Melt
 The freshwater inventories and distributions (Figures 7 and 8) can be used to derive residence times of the freshwater components in the study area provided we can establish volume fluxes across the sections. It is important to note that separation of the freshwater components is crucial to derive valid residence times in polar regions where freshwater contents can be manipulated both by RW and SIM [Östlund, 1982]. To estimate volume fluxes, we apply the three-dimensional ice-ocean model developed for Hudson Bay by Saucier et al. .
 We would ideally like to arrive at a bulk residence time for freshwater components in the study area in late fall for the water column down to the bottom of the assumed PML (0–90 m). Because we have referenced our estimates against the remnant PML, which occurs at intermediate water depths (generally 60–90 m), we are in reality estimating the residence time in the 0–60 m range, where virtually all of the SIM and RW components reside. Furthermore, our observations apply only to late summer (October) and residence times might vary given the strong seasonality of RW and SIM inputs as shown in Figure 2.
 For modeling, we have used a 2-week average for the period prior to sampling; we cannot assume that our observations of freshwater represent earlier periods. Modeled currents (Figure 9a) combined with the observed freshwater component fractions (Figures 7 and 8) then provide estimates of the fluxes of freshwater across each section (Figures 9c and 9e), as well as across from inshore to offshore which closes the sectors between the transects. Results from section C show what is typical for all sections: a strong nearshore current (Figure 9a) that transports almost all of the RW (Figure 9c) and is an important component of the net SIM transport (Figure 9e). In the offshore, the model shows a weaker current, which in places opposes the direction of transport in the nearshore.
 The total fluxes of SIM and RW in the study area (Figure 10) together with inventories of freshwater within the areas delimited by our sections (Figures 7 and 8) imply residence times of 125–300 days for SIM and 30–90 days for RW in late September. While total volume fluxes are balanced in the model, the freshwater components are not in exact balance (Figure 10), indicating transient buildup or loss from the sectors due to uneven distribution of properties within boxes.
 The observed inventories of SIM (Figure 8) also suggest that there has been sufficient time by early October 2005 to transport a large portion of ice melt products out of the region, given that the Bay became completely ice free (and hence production of SIM ceased) by the end of July (Figure 2). In contrast, RW inputs remain high throughout the summer (Figure 2), especially as 2005 was a high discharge year in the region, and might therefore be expected to exert a greater influence than SIM on fall conditions. The increase in the relative importance of RW to total freshwater inventories as summer progresses is also supported by our observations of equal SIM and RW inventories in the offshore where, previously, SIM has been thought to govern the freshwater inputs [Prinsenberg, 1984]. Advection and surface circulation play crucial roles in the distribution of freshwater in Hudson Bay, but we clearly require observations over the complete seasonal cycle to resolve questions of freshwater transport and interactions between freshwater sources. As year 2005 was exceptional in terms of RW inputs into the southwestern sector of the Bay, the extremely high RW transports across section C could partly be a result of this accumulation of RW earlier in the season.
3.9. Dynamics of Colored Dissolved Organic Matter (CDOM)
 The above estimates of RW distribution and transport in SW Hudson Bay allow us to evaluate whether or not CDOM, measured along the same sections, behaves sufficiently conservatively to provide a RW tracer as proposed by Granskog et al. . CDOM is clearly not sufficiently conservative to provide a reliable tracer over large distances/times, and its absorbance decreases when exposed to ultraviolet radiation (photobleaching) or degraded microbially [see Vodacek et al., 1997]. In Hudson Bay, Granskog et al.  observed an almost linear relationship between salinity and a355, especially within estuaries, suggesting at least limited application of CDOM to trace runoff in the ocean is possible. Furthermore, CDOM-based estimates for contribution and distribution of RW and SIM on a Bay-wide scale appeared reasonable, given the general oceanographic controls of freshwater and sea-ice pathways in the Bay [Granskog et al., 2007; Figure 6].
 A linear salinity-a355 relationship, however, is not an appropriate test of conservative behavior in a system where more than two major end-members (e.g., RW, SIM, and SW) are mixing. Because SW and SIM have basically the same a355 [see Granskog et al., 2007], the fraction of runoff FRW (Figure 7) provides the key control on absorption of CDOM in a sample. Deviation from a mixing line for FRW versus a355 (Figure 11) thus provides an appropriate way to evaluate whether CDOM behaves conservatively in the study area. Depressed values of a355 evident on this plot are consistent with a sink for CDOM within the mixing zone [cf. Vodacek et al., 1997]. The magnitude of the deviation from the conservative mixing line, 20–80%, is a measure of the absorptive loss of CDOM. At FRW values of 0.05–0.10 (i.e., 5–10% river water contribution to the mixture), a 50–80% loss of absorption has taken place. This large loss of CDOM during early mixing of RW agrees well with losses of 60–70% observed in the Middle Atlantic Bight [Vodacek et al., 1997]. Within the 1–3 month residence time of the RW fraction, over half of the CDOM carried by rivers into SW Hudson Bay has been lost, presumably through photobleaching. In addition to absorptive loss, the optical characteristics of CDOM changed in a manner consistent with photobleaching, as both the spectral slope S (nm−1) and the slope ratio increased as FRW decreased [cf. Vodacek et al., 1997; Helms et al., 2008]. Alternatively, optical properties could change in response to alteration and replacement of CDOM by in situ production, although the latter is likely not significant given the low productivity in Hudson Bay [e.g., Anderson and Roff, 1980a]. Part of the photobleached organic material is converted to inorganic carbon [Vodacek et al., 1997], which could contribute to the positive relation between CDOM and pCO2 in Hudson Bay surface waters [Else et al., 2008]. Although ice cover likely protects CDOM early in the year [Granskog et al., 2007], conditions are clearly sufficient to photobleach CDOM effectively within Hudson Bay on a seasonal basis.
 The significant photobleaching of CDOM, demonstrated in this study, therefore limits its utility as a RW tracer, but highlights the importance of this mechanism to convert organic carbon to inorganic carbon and thus contribute to CO2 evasion to the atmosphere. It might be possible to apply CDOM under limited circumstances (short time periods, self-shading or shading by snow and ice cover) where cumulative photobleaching is limited because of less exposure to radiation. An application of great interest might be to infer late winter convection where water made dense enough to sink by brine production entrains RW and its components in the nearshore flaw lead system [e.g., Kuzyk et al., 2008]. Climate change as it affects the interaction between runoff (timing, organic matter quantity and quality [e.g., Guo et al., 2007]) and ice cover (season length, ice thickness and dynamics, snow load) has the potential to alter the significance of photobleaching as a pathway in Arctic coastal seas.
 Whereas the salinity profiles collected for our study indicate freshwater distributions in October 2005, similar to previous late season surveys [Anderson and Roff, 1980a, 1980b; Prinsenberg, 1984, 1988; Harvey et al., 1997], the discrimination of freshwater sources through δ18O measurements reveals large differences in the distributions of river water and sea-ice melt in the southwestern sector of Hudson Bay. River water is tightly constrained to the nearshore where much of its transport occurs; in contrast the ice melt is distributed more toward the offshore in late fall, although the coastal corridor appears important for the net transport of sea-ice melt earlier in the season. In combining the source distributions with modeled transports, we see that the cyclonic circulation, a product of estuarine circulation and wind forcing, provides the means to transport river water and ice melt through the region and, presumably, out of Hudson Bay within the same season. This process provides a rectification that removes surface ice melt, which is a rather short-lived input in early summer, but is much less effective at removing brine produced during previous winters, which is injected deeper and therefore less affected by the strong summer surface circulation. An interesting question is how important this process might be for the deep convection observed in this system (penetrating at least to 90 m depth) relative to the shelves and interior regions of the Arctic Ocean, where a shallower PML (∼50 m) is traditionally observed, even though the freshwater yield (i.e., buoyancy flux) in the interior Arctic is much lower than in Hudson Bay. Is the strong summer surface circulation partly responsible for the deep PML in Hudson Bay by virtue of its role in exporting much of the summer buoyancy flux out of the Bay by the time of the fall freezeup? Alternatively, the high buoyancy flux may be balanced by more annual ice production in Hudson Bay. Or are the negative and positive buoyancy fluxes decoupled in space, such that ice formation occurs mainly in the west while ice melt and runoff input occurs mainly in the east? Notably, our results strongly imply that freshwater components are exported rapidly from Hudson Bay by summer surface circulation whereas brine is accumulated in deeper waters. These processes are relevant in understanding interannual variability and future changes in freshwater export from the Hudson Bay system and its impact downstream, e.g., in the Labrador Sea [e.g., Sutcliffe et al., 1983; Straneo and Saucier, 2008].
 Residence times based on freshwater inputs and inventories are approximately the same as the length of the open water season suggesting that much of the runoff entering our study region during freshet (June) had likely been exported eastward out of the area by the time of sampling, to be replaced partially by smaller, late season inflows. Sea-ice melt, on the other hand, has a longer residence time because it is stored farther offshore where the circulation is weaker. Eventually, sea-ice melt is exported out of the region but is not replenished as ice disappears. The imbalance in the input/output of freshwater components in the region bounded by our transects implies that freshwater is not in steady state on a seasonal time scale; rather, pulses of freshwater travel along the cyclonic circulation pathway and pass through the region. Our results also imply that freshwater is not likely in steady state on an annual basis. However, in this respect, year-round observations on the freshwater content and composition would be vital to get closure on the dynamics of freshwater in Hudson Bay.
 The inventories and residence times for river water provide the basis to determine whether CDOM is conservative in the coastal zone of Hudson Bay. Our data (October to the end of open water season) show large losses of CDOM early in the mixing between river water and seawater. Therefore, CDOM does not supply a reliable tracer for river water in Hudson Bay during summer. Nonetheless, CDOM might still provide tracer capability under circumstances where photobleaching has limited opportunity to operate (within estuaries, under ice with snow cover). The large losses observed at the end of the open water season indicate that there is a strong seasonal change in the photoscreening capacity of CDOM, as it is likely that at the end of the winter, when the shielding by ice and snow abruptly ends, CDOM has not undergone any major losses. Therefore the distribution of CDOM [see Granskog et al., 2007] and the seasonality of its photoprotective capacity likely affect photoprocesses in the Bay. The high load of dissolved organic matter to Hudson Bay, with most of the spring–summer input subject to photodegradation, suggests this to be an important process in the conversion of terrigenous fixed carbon back into CO2. Evasion of CO2 to the atmosphere from the photobleaching process likely contributes to explaining the observation that nearshore waters are a net source to the atmosphere of CO2, and that Hudson Bay is at best a minor CO2 sink during open water in contrast to many other midlatitude seas [see Else et al., 2008].
 We extend our special thanks to the captain and the crew of the CCGS Amundsen, for their inexhaustible support throughout the cruise. Several people assisted with sampling, including Brent Else, Alex Hare, Monica Pazerniuk, Pierre Larouche, and Rémy Coulombe. Many thanks to Marie-Emmanuelle Rail, Pascal Guillot, and Yves Gratton for operating the rosette/CTD, postprocessing, and quality check of the data. Pierre St. Laurent and Virginie Sibert aided with modeling. The authors are members of ArcticNet, funded in part by the Networks of Centres of Excellence (NCE) Canada, the Natural Sciences and Engineering Research Council (NSERC), the Canadian Institutes of Health Research, and the Social Sciences and Humanities Research Council. This work was supported by the Canada Research Chairs (CRC) program with grants to D.G.B.; M.A.G. was also supported by research grants from the Academy of Finland (107708 and 108150).