Distributions of runoff, sea-ice melt and brine using δ18O and salinity data — A new view on freshwater cycling in Hudson Bay

Distributions of sea-ice melt (SIM) and runoff (RW) were determined in Hudson Bay (Canada) using salinity and oxygen isotope (δ18O) tracers. In late summer/early fall, stations throughout the Bay exhibited a summer surface mixed layer (SSML) 30–60 m deep, which contained the seasonal freshwater inputs. The SSML overlies a cold subsurface water, extending up to 125 m deep in the water column, indicative of the previous winter's surface mixed layer (WSML). Our data show that most of the RW remains in a nearshore coastal regime in summer, with limited exchange into the interior of the Bay. At depth, brine (negative SIM) accumulation is closely associated with excess RW, implying that deep water formation occurs when sufficient brine is rejected from the growing ice to overcome RW buoyancy in surface water. The inventory of RW at depths below the WSML is equivalent to almost one year's runoff inputs which, given a residence time of deep water of 4–14 years, implies that 6–16% of the annual river discharge is shunted into deeper waters each winter. The balance between the RW buoyancy in surface waters and densification due to ice formation appears to control where and when deep-water formation can occur. Although not directly measured, regions forming deep water are likely small (< 10% of the Bay area) and associated with flaw leads extending around the perimeter of the landfast ice edge in the Bay. The processes evident in δ18O and salinity data must be incorporated into ocean models to obtain a realistic view of freshwater cycling and its sensitivity to climate perturbations.

Research Highlights► Hudson Bay freshwater cycling was studied using salinity and δ18O tracers. ► Majority of runoff is kept within coastal current in summer. ► Runoff is shunted to deeper waters by brine rejection reaching at least 125 m depth. ► Deep waters also contain excess runoff, indicating local deep water formation. ► Models need improvement to describe these processes realistically.