Since December 2019, a comprehensive in situ measuring campaign has been conducted using YSI Exo CTD+ sondes to investigate the AB water properties. It comprised of vertical measurements of water conductivity, temperature, pH, turbidity, and dissolved oxygen, dissolved organic matter, chlorophyll A, and phycoerythrin content at 31 sites across four years. The openly accessible data up until January 2022 can be found in the PANGAEA repository (Osińska et al., 2022). Detailed information regarding the scope and methodology of data collection is described in Osińska et al. (2023). Measurements conducted using an unaltered methodology have continued up until February 2023, and their findings have been analyzed in this study. For the present analysis, 23 measurement sites with depths exceeding 10 m were chosen ( Figure 1A ; green points) and divided into four zones ( Figure 1A ; green boxes):

Measurements with missing salinity records and those from the depths above 0.5 m have been excluded from the analysis (due to high uncertainty of near-surface measurements It was found that several salinity records had abnormally high mean values of >35.5, which raised suspicions. Consequently, it was decided that extracting outliers from the dataset was appropriate. A time-averaged salinity profile [ S x ( z ) ] was calculated for each site from all measurements at that site s n x ( z ) , where z denotes depth, x denotes a specific site and n is an index of individual measurement at that site. The following records have been classified as outliers and removed from the dataset:

After this procedure, the remaining dataset consisted of 1830 profiles from 136 days and all seasons of the year.

Freshwater thickness (FWT) was determined for all profiles using the Holfort et al. (2008) formula:

where S_ref is a reference salinity value. S_ref was determined for each measurement day as the mean salinity value from all measurements from that day below 60 m. The decision to use records from below 60 m was based on modeling results that showed glacial water spreading maximally to this depth (details in Section 4.1.2).

Figure 2 provides a comprehensive depiction of the fluctuations in AB water properties over four years of hydrographic measurements (see also this data presented in vertical profiles of salinity in Supplementary Figure 1 ). Overall, the water temperature varied in a range of −2 to 2°C, while the salinity ranged from 33.3 to 34.6 ( Figures 2A–H ).

The surface freezing line seen in TS diagrams ( Figure 3 ) shows that most of the time AB water properties were well above freezing conditions during all seasons of the year (details in Osińska et al., 2022). This is the reason for the absence of winter sea ice coverage in AB over the course of the measuring campaign.

The freshening of AB’s surface water during austral summer ( Figures 2B, D, F, H ) and the corresponding peaks in FWT ( Figure 4 ) indicate the presence of GMW. This is because marine terminating glaciers are a primary source of freshwater in the northern WAP region, as established by Powell and Domack (2002). Additionally, it has been determined that the contribution of sea ice and precipitation to AB’s freshwater content is limited. No evidence of fresher water plumes in subsurface layers was detected ( Figure 2H and more details in Osińska et al., 2023). Hence, the GMW continuously exhibits the highest buoyancy among the water masses in the AB region, a finding that has been corroborated by prior investigations conducted in the area (Monien et al., 2017; Meredith et al., 2018; Osińska et al., 2021).

The FWT variations exhibit a similar seasonal pattern across the entire AB. However, FWT values are lowest in the west and central AB, with a mean of 0.09 m and a median of 0.07 m ( Figure 4 ). The FWT mean and median values in the east AB are 0.15 m and 0.14 m, respectively; in Ezcurra Inlet, they are 0.12 m and 0.10 m; and in Lange cove, they are 0.14 m and 0.12 m. This would suggest that the presence of GMW is lowest in the western regions of the main basin of AB and noticeably highest in its eastern region, even surpassing that of regions directly adjacent to glacial fronts, such as sites in the Lange cove.

TS diagrams, as shown in Figure 3 , are used to differentiate between water masses inside of the AB. AB waters during the winter are generally homogenous, with temperature and salinity marginally rising as depth increases. During the spring season, a fresher and warmer layer of water is formed on the surface, overlaying waters characterized by increasing salinity and temperature with depth. Two layers in the AB water column are also present during summer and autumn. The summer surface layer experiences maximum freshening (average salinity dropping to 33.2) and warming (mean temperature ranging from 1 to 1.5°C). During the autumn, the upper layer, in comparison to the summer season, exhibits lower temperatures and higher salinity.

Observations indicate that AB contains up to two characteristic water layers throughout the course of a year. The distribution of these layers’ salinity and temperature values can be largely attributed to atmospheric and glacial influences. The water mass found below the surface layer during the seasons of spring, summer, and autumn, as well as the principal water mass observed during winter, shall be referred to as ambient water (AW). This water mass is primarily impacted by the waters of the Bransfield Strait and by atmospheric forcing. AW exhibit relatively small variability throughout the year and display typical patterns of seasonal fluctuation commonly observed in estuarian deep waters (Cottier et al., 2010). Fresh surface waters found in spring, autumn, and particularly during the summer are classified as GMW. GMW consists of a mixture of AW and glacial water that originates from subglacial discharge, submarine melting, glacial creeks, and icebergs. These waters are heated and cooled to varying extents through atmospheric forcing. The lowest summer surface temperatures were recorded in Langel cove since the freshly formed GMW surface layer has a limited duration of atmospheric exposure. Notably, there is a possibility of external freshwater entering AB at the surface, which may be indistinguishable from GMW using solely salinity records.

The presence of warm and highly saline Atlantic Waters in Greenland (Straneo et al., 2011; Sciascia et al., 2013; Slater et al., 2018) and CDW in the Antarctic (Moffat et al., 2009; Cape et al., 2019) has been shown to directly stimulate glacial melting and play an important role in shaping the hydrodynamics of glacial bays. The hydrographic data analyzed here does not support the existence of such warm external water masses in AB. The measurements conducted in this investigation were limited to a maximum depth of 100 m. Consequently, it is possible that distinct water masses could infiltrate deeper AB waters and remain undetected. Nevertheless, the probability of such an event and its substantial influence on AB’s glacial-oceanic boundary is low. Because water depth near AB glaciers seldom exceeds 100 m ( Figure 1C ), any warmer and more saline water intrusions would be unable to reach glacial fronts unless their presence were recorded at shallower measurement sites. Additionally, earlier measurements conducted in AB over a wider vertical range also did not find any signs of the presence of such water masses (Carbotte et al., 2007). Finally, studies of regional ocean circulation concluded that CDW intrusions into AB are unlikely (Hofmann et al., 2011; Sangrà et al., 2011).

The general two-layered stratification enables the determination of the internal Rossby radius ( r i = c i f , where c i 2 is the internal wave speed and f is the Coriolis parameter) which serves as a metric for evaluating the relative significance of water column stratification in comparison to rotation (Cottier et al., 2010). In the AB, depending on conditions, the internal Rossby radius varies between 0.41 and 11.86 km (its average values are 0.91 in winter, 1.00 in spring, 1.39 in autumn and 1.83 km in summer). Therefore compared to the ~8 km wide opening, it indicates that the AB can be classified as a “broad bay”, where the presence of cross-bay circulation has substantial importance. This is valid for all seasons, even the period of enhanced glacial melting, when freshwater influx strengthens the water column’s stratification.

The presence of a two-layered stratification in AB, where the surface layer consists of the most buoyant layer of glacial meltwater (GMW), is reminiscent of the conditions outlined in the small-fjord single-cell circulation model proposed by (Motyka et al., 2003). However, due to the “broad” character of the bay, the AB model must be three-dimensional.

Modelling of AB hydrodynamics has been performed using the open-source Delft3D-Flow model, developed as part of a Delft3D suite created specifically for coastal, river, and estuarine hydrodynamics (Deltares, 2020). The calculations were performed on a high-resolution curvilinear grid of over 30,000 points, thus, an average grid cell corresponds to an area of approximately 55 m². Figure 1C shows the entire model domain. The analysis was conducted in 3D, with fifty layers utilizing a vertically scaled σ-coordinate system, with more densely spaced layers toward the domain’s bottom and top. The bathymetric map shown in Figure 1C was used with a single smooth, ~10 km long open boundary between AB and Bransfield Strait.

The model was driven by tides, and temperature and salinity gradients. The tidal water level at the open boundary was calculated using the CATS2008 Antarctic tides model (Padman et al., 2002). Temperature and salinity data reanalysis by Dotto et al. (2021) was used to determine temperature and salinity values at the open boundary since it is the most robust data source for water properties in the northern WAP region, combining the majority of available in situ measurement records from 1990 to 2019. Seasonally averaged (for spring and summer) reanalysis values were extracted from a grid point closest to the model’s open boundary and interpolated in time and space to create varied vertical salinity and temperature profiles. Dotto et al. (2021) results show that Bransfield waters are weakly stratified ( Supplementary Figure 2 ) with seasonal mean temperature and salinity variations of -1.23–0.51°C and 34.22 –34.27. The in situ measurement results from Osińska et al. (2023) were used to determine initial values of water salinity and temperature inside AB, which were uniformly set throughout the domain. It has been found during preliminary model testing that after less than three days of simulations, the salinity stratification in the whole bay was predominantly influenced by the open boundary input, therefore no variation in the initial conditions setting was necessary.

To capture the variability associated with the entire range of tidal patterns in this region, calculations lasted 58 days (from 1.12.2021 to 28.01.2022), consisting of 3.5 days of model warm-up followed by two full lunar cycles.

Following Deltares recommendation, bottom roughness was calculated using the 3D Chézy formula (Deltares, 2020), and assumed spatially homogenous due to lack of information on bottom roughness variations in AB. During model testing, it was discovered that unreasonably high values of kinetic energy dissipation rate (>1000 m²/s³) were obtained close to the open boundary after approximately two days of calculations and persisted throughout the simulation length. It was determined that this was caused by inappropriately assessed bottom roughness. Through several additional test runs it was experimentally found that uniform 3D Chézy bottom roughness coefficients of 40 m^1/2/s, in both U and V directions, is the highest coefficient value which does not result in unrealistic energy dissipation anomalies, which cause a rapid increase in flow velocities near the open boundary, and consequent model destabilization. The energy dissipation rates had a reasonable median value in the order of 10^-8 m²/s³ (comparable values were found in Andvord Bay by Lundesgaard et al., 2020). Although they were greater in the bottom layer they were still within the realistic range of 10^-6 -10^-4 m²/s³ (see Supplementary Table 1 , Inall and Rippeth, 2002). Therefore, it was determined that a bottom roughness coefficient of 40 m^1/2/s was suitable and used all subsequent calculations.

Test runs were carried out to investigate the impact of boundary conditions on the model domain. In general, its impact was not significant. The only part of the model domain where the results were affected by the boundary conditions is the outermost, inflow region in the west (see Section 4.1.2). This implies that the results in this area should be interpreted with caution. The Reynolds number, which is a measure of turbulence, was in the range of 10⁴–10⁶ close to the open boundary and was lower than 100 close to the inner inlet heads.

Additional information regarding the model configuration can be found in Supplementary Table 2 . Importantly, as indicated by the aforementioned description of the model configuration, no atmospheric forcing was considered, i.e., ocean-atmosphere momentum, heat, and moisture fluxes were set to zero. This decision is justified by the fact that, first, the salinity differences between the oceanic and glacial waters dominate the density structure and gradients in the domain of study, and second, volume fluxes associated with tidal currents dominate those generated by wind, particularly over the time scales of several tidal cycles considered here. Such simplification is not unusual in studies at this scale (Straneo et al., 2011).

The typical density anomaly of water entering through the open boundary and that of the meltwater is σ=27.4 kg/m³ (at S=34.1 and T=−0.2°C) and σ=0 kg/m³ (at S=0 and T=0°C), respectively. The highest recorded value of surface water temperature observed in AB in the summer was 3.54°C (at 1.15 m depth), which was exceptionally high (Osińska et al., 2023); the corresponding density anomaly at S=32.56 is σ=26.0 kg/m³. Therefore, the contribution of temperature to the net variability of water density in AB is minor. Accordingly, the core of the analysis and discussion in the following sections is considering factors driven by salinity fluctuations. Since seasonal salinity variations derived from Dotto et al. (2021) dataset are small (<0.1 difference between mean seasonal values) model setup accurately replicates AB open boundary conditions throughout the year.

For model validation purposes, an RBR wavemeter was moored within Admiralty Bay (location indicated with blue dot in Figure 1A ) logging water level at 2 Hz frequency during the period from 6.12.2021 to 21.12.2021. The standard deviation of differences between Delft3D model data at this location and in situ RBR measurements is 0.08 m, the bias is 0.03 m, and their correlation coefficient is 0.99, i.e., the modeling results correspond very closely to the real water level changes in that part of AB. Analogously, CATS2008 compared with RBR measurements has a 0.08 m standard deviation of differences, a bias of 0.01 m, and correlation coefficient of 0.99.

The representation of interactions between glaciers and oceans is a crucial component in establishing the framework for glacial bay hydrodynamical modelling. The description of oceanic dynamics near marine terminating glaciers often relies on the buoyant plume theory (BPT). The BPT explains how freshwater discharged from underneath the glacier upwells along the glacial front, entraining and mixing with ambient waters to form a GMW plume. This plume then induces the submarine melting of the glacier’s front (Jenkins, 2011). The submarine melt rate is influenced by subglacial discharge volume and ambient water temperature; however, this relationship varies depending on the study location (Kimura et al., 2014; Xu et al., 2012; Sciascia et al., 2013). When GMW reaches its depth of neutral buoyancy, which may occur at or below the ocean surface, it forms a layer of distinct properties within the water column (Jenkins, 2011). The influence of glacial water on ocean hydrodynamics is contingent upon the distribution of subglacial discharge points, namely whether they are channelized or uniformly distributed along the glacial front, and the momentum of the discharge (Cowton et al., 2015; Slater et al., 2018).

The Buoyant Plume Model (BPM) coupled with the general circulation model (GCM) is currently considered the most sophisticated method for investigating the hydrodynamics of glacial bays (Cowton et al., 2015). However, its application may not always be necessary or practical. In an earlier investigation conducted by Chauché et al. (2014) observational data indicated that subsequent to channelized release, subglacial influx rapidly spreads laterally along the glacial front, effectively blurring the distinction between effects of localized and uniformly dispersed freshwater injection points. The study by Sciascia et al. (2013) demonstrated that the hydrodynamics of near-glacial waters is influenced to a greater extent by the volume of subglacial discharge than the momentum of its inflow. The usage of the BPM coupled with GCM for the purpose of modeling the hydrodynamics of bays with multiple ice-water boundaries is challenging. Firstly, such multiway coupling is computationally expensive. Furthermore, it requires detailed bathymetric and glaciological data, including discharge location points, volumes, and submarine melt rates (Carroll et al., 2016) which is currently unattainable in AB and, we argue, in the majority of glacial bays in Antarctica.

In light of the practical challenges involved, a question arises regarding the extent to which accurately reproduced vertical location and velocity of glacial water influx is significant for the understanding of general AB hydrodynamics. In order to address this question, several iterations of model tests were conducted, in which glacial water discharge locations and velocities were varied. The following are the identifiers and details of these test runs:

In order to emphasize the potential influence of glacial discharge velocity on AB hydrodynamics, a high value of 2 m/s was selected for testing (Xu et al., 2012; Cowton et al., 2015). The volume of the glacial discharge for all test runs was established at ~6 m³/s per 1 km of glacial front, a value that was deemed reasonable for the AB region during the summer melt season (see section 4.2).

Three measures were employed to examine disparities between test run results: FWT ( Figures 5A–D ), pycnocline depth ( Figures 5E–H ), and depth-averaged flow velocity ( Figures 5I–L ). These metrics serve as the foundation for further analysis of AB hydrodynamics, making them suitable instruments for determining if the results of test scenarios exhibit substantial differences between each other. FWT was calculated using Formula (1), where S_ref was determined as the mean salinity from below 60 m across the entire AB. Given the stratification of model open boundary waters, the utilization of this FWT calculation method shows the presence of freshwater influx from the Bransfield Strait into AB. Hence, in order to illustrate the distribution of freshwater originating exclusively from AB glaciers, the FWT values calculated for a scenario devoid of glacial water inflow (scenario 0 m³/s) were subtracted from the FWT results of the four test runs. The pycnocline depth was calculated as the depth at which dσ/dz< 0.025 kg/m³. The data that have been analyzed and presented in Figure 5 were averaged over a period from January 1st, 2022 to January 28th, 2022, which corresponds to a one complete lunar cycle.

All test run results show consistent patterns in the FWT, pycnocline depth, and flow velocity values distributions across the AB Supplementary Figure 3 ). On the other hand, discrepancies are visible when comparing maps of differences between test runs and the reference case (H0) results ( Figure 5 ).

For all test scenarios the FWT values are highest in the northwest region of AB, ranging from 0.35 to 0.55 m, ( Figures 5A–D ). In that area the three scenarios H2, S0, and S2 have slightly greater FWT values (<0.1 m) than the reference scenario H0. The overall FWT differences between scenarios range from −0.1 to 0.15 m ( Figures 5B–D ). In test runs with solely subglacial discharge, narrow regions of elevated FWT form along glacial fronts. The biggest differences in FWT and pycnocline depth are observed in scenario H2 ( Figures 5B, F ). For instance, in an area of a maximum pycnocline depth (~25 m) for H0, the pycnocline depth increases by up to 4 m in S0 and S2, and by over 6 m in H2. In S0 and S2 scenarios the presence of subglacial discharge and subsequent turbulent mixing prevents pycnocline formation in regions close to glacial fronts (blue areas in Figures 5G, H ). Crucially, the overall flow pattern remains consistent in all examined cases, characterized by a strong influx from Bransfield Strait along the AB’s western bank and an outflow in the east (see more details in section 4.1.2). The differences in flow velocities, shown in Figures 5I–L , are largest close to the AB opening. In scenarios H2, S0, and S2, the AB’s inflow and outflow have reduced velocities compared to the reference case results. This slowing down is largest in cases in which glacial waters are discharged with 2 m/s velocity (up to a −0.25 m/s decrease in H2 and −0.15 m/s decrease in S2).

The model test run results show that the freshwater content, the water column stratification, and the flow velocities in AB are locally impacted by changes in the location and momentum of glacial influx. In general, larger differences in the analyzed metrics were caused by variations in the velocity of glacial input rather than an alteration in its vertical position. Nevertheless, the overall circulation and glacial freshwater distribution patterns in AB have not changed as a result of employing any of the studied model configurations ( Supplementary Figure 5 ). This conclusion is further strengthened by the high correlation coefficients (r) and low σ D for all employed metrics across all scenarios ( Figure 5 ). Therefore, it i justified to conclude that for examining the overall impact of glacial water on AB hydrodynamics, a simplified methodology that disregards the influence of the vertical position and velocity of glacial injections is adequate.

Consequently, further model simulations were performed with glacial water discharged homogenously through the entirety of the glacial front, from all glaciers, with zero initial velocity. A total of fourteen scenarios with increasing volumes of glacial runoff were calculated: 0, 0.15, 0.3, 0.6, 0.9, 1.7, 3.0, 4.5, 6.0, 8.0, 11.0, 14.0, 28.0, and 60.0 m³/s of freshwater volume discharged per ~1 km of glacial front. Henceforth, these values will be employed as identifiers for the scenarios in order to enhance the conciseness and clarity of the text. Input of freshwater from the creeks was assumed to be vertically homogenous, was of equivalent volume to runoff from ~1 km of a glacial front in a given scenario and was introduced through a single grid cell.

Modeling results were analyzed through Principal Component Analysis (PCA) of water levels, using results from two full tidal cycles from 4.12.2021 12:00 to 28.01.2022 00:00. Each PCA mode consists of a spatial distribution (map) of PCA coefficients (also known as loadings), a time series of PCA scores showing the relative strength of that mode through time, and the overall percentage of the total variance of the dataset explained by that mode. Through the calculation of the squared correlation coefficients (r ²) between scores of PCA modes and time series of water level in all active grid points, maps of the spatial distribution of percentages of variance explained by the first four modes have been obtained ( Figures 6B–E ).

Tides are a primary driver of water level fluctuations in AB. This is demonstrated in Figure 6A where a comparison of in situ measurements collected by RBR wavemeter moored 9.5 km away from the AB outlet (location marked in Figure 1A ) with tidal data from CATS2008 at the open boundary of the Delft3D model is shown. The blue line corresponding to in situ measurements exhibits only small deviations from modeled data, presumably during periods of very strong winds. The yellow line represents Delft3D model results at the grid point closest to the wavemeter location. The very good agreement between the three curves shows that the water level in the whole AB reacts almost instantaneously to the open boundary forcing.

PCA analysis of water level in the 0 m³/s scenario further confirms almost instantaneous response of the whole AB to tidal shifts. Figures 6B–E shows maps of coefficients corresponding to the first four PCA modes and the percentage of water level variance explained by them, respectively. The first PCA mode (PCA 1), which represents homogenous changes in the water level of the whole AB, explains more than 99.8% of the variance in all studied scenarios. Accordingly, the PCA 1 score correlates almost perfectly with the time series of water level at the boundary and inside of the model domain (see time series in Figure 6A ). This indicates that anomalies from this pattern are of the order of a hundredth of a percentage, even in the 60 m³/s scenario, in which an additional 2000 m³ of water is pumped into AB every second. The predominance of tidal impact on water level changes is not surprising, since volume flux through the open boundary is of the order of 100,000 m³/s.

Although explaining small percentages of variance, other PCA modes of water level shifts are showing important characteristics of water level fluctuations in AB. Modes 2-4 represent standing-wave-like water level fluctuations with respectively one, two, and three nodes (Valle-Levinson, 2022). Each of the maps in Figures 6B–E emphasizes a region in central AB that corresponds to the location of a smaller circulation cell in the overall AB circulation pattern (Section 4.1.2 and Figure 7 ).

The most notable feature of AB general circulation is a strong northerly flow along its western boundary ( Figure 7A ). It is formed by the Coriolis force acting upon Bransfield Strait waters flowing northeast along the edge of the South Shetland Islands (Zhou et al., 2002). The existence of this current was recognized by prior modeling conducted in the AB by (Robakiewicz and Rakusa-Suszczewski, 1999). Following its initial development, the AB inflow current continues in a northerly direction and subsequently undergoes bifurcation. Part of it flows to the right in the central region of the main body of the AB, approximately 7 km from the AB outlet (around the location of main AB cross-section), and then exits the bay in close proximity to its eastern boundary. The second limb of the current penetrates deeper before reversing its course in the main embranchment of the bay (~13.5 km from the opening) and also flows back to the bay opening along its eastern coast. The clockwise (cyclonic) circulation cells formed by these two branches are crucial elements of the water exchange mechanism between the ocean and inner bay waters. A visualization of monthly average velocities across the main AB cross-section ( Figure 7B ) reveals that this exchange has the greatest magnitude in the surface layer. In the scenario without glacial influx there exists a state of equilibrium between the amount of water flowing into AB via its western half and the amount flowing out of it through the eastern half of the main AB cross-section ( Figure 7F ). At spring tide, the volume of water transported through each of the halves reaches 2·10⁵ m³/s. The quantities of water penetrating the three inner inlets of AB, Ezcurra, MacKellar, and Martel Inlet are two orders of magnitude smaller, with proportionally lower velocities observed across their respective cross-sections ( Figures 7C–E ).

When glacial influx is introduced, AB’s cyclonic circulation explains the development of distinct patterns in glacial water dispersal, illustrated by FWT and pycnocline depth maps ( Figure 8 ). In each of the model scenarios, following an initial warm-up period, a quasi-stationary state is reached, in which the distribution of FWT remains approximately constant ( Supplementary Figure 4 ). With rising glacial influx levels, freshwater accumulates in the northeastern region of AB, specifically in MacKellar and Martel Inlets. This freshwater is then transported to Bransfield Strait by the AB’s eastern outflowing current. In the accumulation zones, the FWT values range from 0 to 0.5 m. The FWT exceeds 1 m in larger areas only in the two strongest glacial influx scenarios, 28 m³/s and 60 m³/s. The increase of glacial input results in the expansion of the region where the pycnocline occurs, as well as in its deepening. The pycnocline depth is determined by the local bathymetry, resulting in the deepest pycnocline developing in the area of the main AB embranchment. In all scenarios, the depth of the pycnocline does not exceed 60 m.

The model results demonstrate circulation and freshwater distribution patterns that are consistent with the in situ measurement data. The average salinity values from the top 60 m of water at two sites in the western inflow area, ve6 and ez9, consistently exceed those reported at the outflow sites, ve4 and lv3. Despite the close proximity and similar distance from the glacial front and bay’s outlet between inflow and outflow sites, this salinity difference can reach 0.3 ( Figure 8U and Figure 8A for).

In all the model scenarios, the circulation pattern of two cyclonic circulation cells is preserved in AB ( Figures 9A–E ). The analysis of flow velocities and transport volumes across the main AB cross-section reveals that in scenarios ranging from 0 to 14 m³/s, the water exchange is consistently strongest near the surface and has a volume of ~10⁵ m³/s for both inflow and outflow ( Figures 9F–J , and Figure 9Z , analogous to Figures 7B, F ). However, in two highest glacial influx scenarios (28 and 60 m³/s), the water transport on both sides of the main AB cross-section decreases significantly to ~10⁴ m³/s ( Figure 9Z ). Similarly, in these scenarios, the flow velocities are reduced (see Figures 9E, J ). This observation suggests that a threshold value of glacial inflow volume exists limiting water interchange between the bay and the ocean. Specifically this threshold is observed to be between 14 and 28 m³/s ~1 km of glacial front, which adds up to 450 and 900 m³/s of overall freshwater input into AB.

In cross-sections located at the openings of inner AB inlets, Ezcurra, MacKellar, and Martel Inlets, the impact of increasing glacial influx is visible from relatively low glacial water inflow rates below 0.6 m³/s, ( Figures 9F–Y ). The surface outflow layer forms there, moving GMW out of the bay, most evidently in the Ezcurra and MacKellar Inlets ( Figures 9K–T ). Figures 9AA–CC shows the variability in water volume transported through the three inlet cross-sections in 14 model scenarios, in total, and split into layers above (surface layer) and below the pycnocline depth (calculated as in section 3.3.2). In AB inlets, surface outflow and deeper inflow increase with rising glacial influx, up to 14 m³/s scenario, when their values stabilize. This demonstrates how glacial influx drives vertical circulation, similar to the 2D glacial bay circulation of (Motyka et al., 2003). The drop in total transport values in Figures 9AA–CC indicates the importance of additional freshwater input for the water budget of AB inner inlets, which is barely visible in the flow transport sum up through the main AB cross-section, where overall values are 100 times higher ( Figure 9Z ).

Maps depicting the correlation coefficients between the glacial influx volumes and horizontal flow velocities and directions have been generated ( Figure 10 ). They show areas in which glacial bay buoyancy-driven vertical circulation can be a dominant flow pattern. The maps are shown in three versions: for the entire water column, for depths below the pycnocline, and for surface waters inside and above the pycnocline. In regions where pycnocline was not present, the entire column of water was treated as waters below the pycnocline. In order to acquire representative description of changes in waters above pycnocline ( Figures 10C, F ), correlations have been calculated for points in which pycnocline was present in at least six model scenarios. To reduce the possible influence of outliers and increase the robustness of the results, a bootstrap resampling of the data was performed (Trauth, 2010). The areas outlined with black borders in all of the maps in Figure 10 represent points where this analysis produced statistically significant results.

In three inner AB inlets, the whole of Ezcurra and MacKellar Inlets, and most of Martel Inlet, there is a strong correlation between horizontal flow velocity and glacial influx ( Figures 10A–C ). Overall, based on the evidence in Figures 9 , 10 , we conclude that in these areas glacial input can create vertical circulation, driving local water exchange.

In the entire water column and in the bottom layers, the distributions of correlation coefficients of flow direction changes versus glacial influx volumes do not show any discernible pattern ( Figures 10D–E ). However, in the surface waters, a distinct areas can be recognized where, with rising glacial input, water flow turns to the right in a broad area in the middle of AB and to the left in a smaller area in the east part of the main embranchment of AB ( Figure 10F ). This shows how the GMW surface layer deflects surface water following the general circulation pattern ( Figure 7A ), redirecting it toward the AB outlet and restricting its penetration of inner bay waters.