Mercury stocks in discontinuous permafrost and their mobilization by river migration in the Yukon River Basin

The YRB spans more than 330 000 km² in regions of northwestern Canada and central Alaska and is underlain by areas of continuous and discontinuous permafrost [30]. The Yukon River has the highest flow-weighted annual THg concentration out of the six major Arctic rivers [29], and the YRB is one of the six major freshwater contributors to the Arctic Ocean, supplying 3–32 times more THg to the oceans than the 8 other major northern hemisphere river basins [19]. This makes the YRB an important focus of study in the context of riverine THg inputs in a changing Arctic.

Yukon river waters contain a range of sediment sizes, which are expected to influence organic carbon (OC) and Hg contents. To capture some of this variability, we focused on two regions in the YRB underlain by discontinuous permafrost with distinct riverbed sediment characteristics (figure 1). Our sites were chosen near Alaska Native communities that are at different risk levels for erosion, flooding, and permafrost thaw [31, 32] to coincide with overarching collaborative efforts to understand the effects of erosion in the YRB. At both sites, the river channel migrates laterally through cutbank erosion and point bar deposition at rates of meters per year [33].

Zoom In Zoom Out Reset image size

2.2.1. Field sampling procedures

2.2.2. Lab analysis

2.2.3. Stock calculations

We calculated THg stocks for the most complete stratigraphic sections sampled (Huslia: 15 cutbanks and 11 bars; Beaver: 13 cutbanks and 16 bars, supplementary, table S1). Near-surface stocks were determined by integrating over 1 meter and 3 m depth to compare with previously published datasets [15–17]. Total stocks that can potentially be reworked by river lateral migration were determined for the full column depth (∼10–15 m), defined as the distance from the top of the cutbank (CB) or point bar (PB) to the bottom of the thalweg (the deepest part of the river). However, incomplete bank exposure and inability to dig below the thawed active layer meant we could not sample below the top ∼20%–50% of this sedimentary column. Thus, we estimated full column stocks for PB and CB by the sum of the sampled and inferred stocks for each stratigraphic layer in the column $\left( i \right)$ (equation (1), supplementary, table S1)

$$\begin{align}{S_{{\text{PB}}\;{\kern 1pt} {\text{or}}\;{\kern 1pt} {\text{CB}}}} = \sum\limits_{i = 1}^n {{\rho _{{\text{dry}},i}}} *{h_i}*{\text{TH}}{{\text{g}}_i}.\end{align} \tag{ 1 }$$

For exposed sections of bank and bars, sampled stocks were directly calculated using measured layer thicknesses (${h_i}$, km) from each identifiably stratigraphic layer, dry density of bank material (${\rho _{{\text{dry}}}}$, kg dry sediment km⁻³) and THg mass fraction (${\text{TH}}{{\text{g}}_i}\,$, kg Hg kg dry sediment⁻¹) from collected paired samples. Any missing stratigraphic information was supplemented with an average value from sediments of similar substrate composition from the same field location (supplementary, tables S2 and S3).

To calculate THg stocks for the unsampled sections (the 'inferred' portion in figure 2), we determined unsampled column heights and inferred associated sediment properties. Total column heights, independent of river stage height, were determined based on bathymetric and elevation data (supplementary, table S1). Bathymetry was mapped via SONAR surveys at the time of sample collection, referenced by RTK-GPS (real-time kinematic geographic positioning system). Topography data were from National Center for Airborne Laser Mapping Light Detection and Ranging datasets from flights over Huslia on 21–23 August 2022, and over Beaver on 2–5 August 2021 (figure 2). The sampled sections (h₂, h₃) were subtracted from total column height (H_CB, H_PB) to determine the unmeasured section (h₁, h₄) heights.

Zoom In Zoom Out Reset image size

To infer sediment properties, we used our most complete stratigraphic sections (∼5–10 m thick), measured in late fall when the Koyukuk (Huslia) and Yukon (Beaver) Rivers were at low stage. We determined that 3 m was a characteristic maximum thickness for fine-grained overbank sediments at both sites (supplementary, figure S3). We then bootstrap resampled all measured beds below 3 m depth from the modern floodplain surface to estimate sediment properties of all unmeasured beds. We found that lower beds (>3 m) were predominantly sand in Huslia and a mixture of gravel and sand in Beaver (supplementary, figure S4). Our findings of grain size fining upward is typical of river lateral accretion deposits [42]. We calculated inferred section stocks using an average dry density and THg content (supplementary, table S4) based on substrate composition for each location.

2.2.4. Flux calculations

To calculate the flux of sediment-bound THg going into and out of the river, we created a one-dimensional mass balance box model (supplementary, figure S5) representative of erosion and deposition along our study reaches within the Koyukuk and Yukon Rivers following prior work on OC [43] in the same region. For our most complete stratigraphic sections, we calculated THg fluxes for individual banks and bars (supplementary, table S5) following:

$$\begin{align}F = L*k*{S_x}\,\;{\text{where}}\;x = {\text{CB, PB}}.\end{align} \tag{ 2 }$$

Cutbank erosion and point bar deposition flux ($F$, ${\text{kg }}\,{\text{Hg }}\,{y^{ - 1}}$) for Huslia and Beaver were calculated using site-specific river migration rates ($k$, km y⁻¹) calculated from 10 m resolution Sentinel-2 satellite imagery over the period 2016–2022 [44], THg stocks (${S_x}$, kg Hg km⁻²), and river reach lengths ($L$, km) of the entire study areas. Huslia had river reach length of 58 km with migration rates ranging from 0.63–7.6 m y⁻¹, while Beaver had migration rates of 0.10–12 m y⁻¹ along the 55 km river reach (supplementary, figure S6, tables S4 and S5, [44]). Uncertainties were estimated via a bootstrapping resampling simulation (n = 10 000), selecting random calculated bank and bar fluxes to calculate a median net mercury flux; uncertainties are reported as 15th and 85th percentiles of the resulting distributions. The net flux (${F_{{\text{net}}}}$) from river migration was calculated based on the difference between cutbank (${F_{\,{\text{CB}}}})$ and point bar $({F_{{\text{PB}}}})$ fluxes, as:

$$\begin{align}{F_{{\text{net}}}} = - {F_{{\text{CB}}}} + {F_{{\text{PB}}}}.\end{align} \tag{ 3 }$$

This approach only quantifies loss or gain of Hg from a river reach due to floodplain erosion and deposition; it does not consider sediment imported from upstream and exported downstream, and as such does not capture all processes mobilizing Hg across the watershed.

The median THg content was 49⁺¹³/₋₂₁ ng Hg g sediment⁻¹ and 39⁺¹⁶/₋₁₈ ng Hg g sediment⁻¹ (figure 3, uncertainties reported as 15th and 85th percentiles) for Huslia and Beaver, respectively. The median RHgC for Huslia was 5.4^+2.0/_−2.4 µg Hg g⁻¹ C and 4.2^+2.4/_−2.9µg Hg g C⁻¹ for Beaver (figure 3). Where we collected paired samples of thawed and frozen material from cutbanks, THg contents were generally lower in the frozen material in Huslia and Beaver had no apparent trends (supplementary table S6). We cannot rule out contamination from the hole saw used to sample frozen material as contributing to the differences in Huslia, but effort was made to remove material that had come into contact with metal during sample collection.

Zoom In Zoom Out Reset image size

Across all of the samples analyzed, THg content showed no apparent trends with depth (figure S7) and a positive relationship to TOC (figures 4(A) and (D)) and substrate class (p-values < 0.0001, figures 4(B) and (E)). RHgC showed a negative relation to TOC (figures S8(C) and (F)) and substrate class (p-values < 0.0001, figures 4(C) and (F)). These results suggest that THg contents in these systems are strongly correlated to substrate class and OC content.

Zoom In Zoom Out Reset image size

The median heights of the complete stratigraphy from banktop to the channel thalweg at Huslia were 11 m for cutbanks and 10 m for point bars; depths at Beaver were 12 and 11 m, respectively. Median cutbank stocks at Huslia were 41 ⁺⁶/₋₂₀, 125⁺⁴⁴/₋₆₂, and 327⁺¹²³/₋₃₉ kg Hg km⁻², over 1 m, 3 m, and total depths, respectively (figure 5). Equivalent point bar THg stocks at Huslia were 33⁺⁴/₋₁₁, 90⁺⁴/_−11, and 250⁺⁷/₋₆₃kg Hg km⁻². At Beaver, cutbank stocks were 36⁺¹³/₋₈, 103⁺²⁸/₋₁₈, and 337⁺⁴⁴/₋₇₆ kg Hg km⁻², while point bar stocks were 29⁺⁶/₋₄, 92⁺⁴/₋₆, and 274⁺³⁷/₋₈₅ kg Hg km⁻². Point bar and cutbank stocks at both Huslia and Beaver overlapped within uncertainty for most depth intervals (figure 5). The bimodal distributions of THg stocks observed in figures 6(C)–(E) are due to the point bar elevation differences between sites.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The median fluxes at Huslia were 95 ⁺¹²/₋₄₇ and 60 ⁺⁴⁰/₋₂₉kg Hg yr⁻¹ for cutbank erosion and point bar deposition, respectively. Beaver had corresponding median fluxes of 26 ⁺¹⁵⁴/₋₁₃ and 10^+5.3/_−1.7kg Hg yr⁻¹. The net THg budget associated with river migration for Huslia (+32⁺²⁸/₋₂₉ kg Hg yr⁻¹, reflecting net deposition of THg in point bars) and Beaver (−17⁺⁹/₋₇, reflecting net erosion of THg into the river) were calculated as the difference between the median erosion and deposition fluxes, with uncertainties estimated by bootstrap resampling with replacement. Overall, both sites showed similar fluxes of THg release by cutbank erosion, overlapping within uncertainty. However, the THg flux from point bar deposition is significantly higher at Huslia than at Beaver, leading to different estimates of the net budgets.

This study presents new THg and RHgC measurements (supplemental dataset 1) from the YRB using a spatially dense sampling approach that allows us to determine regional THg stocks with accuracy not possible with prior, more generalized approaches. Median THg values for Huslia and Beaver are similar to those reported in prior studies from Alaska as well as other settings in the Arctic ([15–17]; figure 3).

THg stocks for the upper 3 m are similar to those predicted for our study region in the map produced by Schuster et al, who used OC datasets and the RHgC to determine Pan-Arctic THg stocks ([15], figure 5). The consistency of our stock calculations with their predictions may be unsurprising since their model was based on samples collected from interior Alaska. However, our RHgC values were higher than reported in their study and other Arctic studies (figure 3, [15–17]), primarily because of lower OC at our sites. Interestingly, lower carbon content is not reflected in lower THg content in these sediments despite the expected association of Hg and OC. The end result of similar stock values emerges by the coincidence of higher RHgC values and lower OC content in our samples, so even though the studies converge on similar final numbers, we find underlying sediment chemistry that is notably different than predicted.

We find variable RHgC both across sites and with depth (supplementary figure S8). Similar to another Arctic study [18], substrate composition plays an important role in soil THg content (figures 4(B) and (E)) and has an even stronger influence on RHgC values (figures 4(C) and 5(F)). We find few samples with very high OC, and variable soil composition could be a simple explanation for the variable RHgC values—particularly in the case of our peat samples. Our sites were dominated by low OC mineral soils and many of the peat samples were found as laminae in silty layers. It is possible that during formation, periodic river floods allowed water to deposit fine mineral sediment into the peat pore spaces. Thus, samples that appear to be peat may have high mineral soil composition, diluting the percentage of OC. Understanding the causes of varying RHgC in Arctic soils, e.g. via micro-analysis to look at mineral and OC-phase associations of Hg, could be a valuable target for future work. In any case, our results highlight that incorporating sedimentological controls on THg contents and RHgC ratios, and their spatial variability, in models will likely improve estimates of THg stored in permafrost.

Point bar and cutbank stocks in both Huslia and Beaver overlap within uncertainty for most depth intervals (figure 5), although in all cases the median values for cutbanks are higher than for point bars and in some cases the difference is statistically significant. For the full sediment column depth, the cutbank and point bar THg stocks are significantly different at both sites (Beaver p < 0.001, Huslia p < 0.001). These differences could be explained by the difference in elevation and age of the features: cutbanks have had more time to develop topsoil and accumulate peat in addition to fine grained sediments from overbank deposition, while point bars are lower in elevation and consist of coarser sediment in the top few meters.

Abrupt thaw events can rapidly mobilize meters-thick deposits of sediment, potentially releasing the large Hg stores in permafrost. For example, thaw slumps adjacent to a tributary of the Mackenzie River in Canada were shown to elevate suspended particulate Hg contents downstream, but river Hg loads decreased once the particles settled out of the water column [28]. Our results, based on riverbank stocks, reveal the integrated effects of erosion and sedimentation along multiple eroding bends of the Yukon and Koyukuk Rivers. The nearly balanced THg stocks between cutbanks and point bars in our study suggest that most THg eroded from the banks during river migration is redeposited with sediments in aggrading bars (figure 6), similar to the Mackenzie River slump study [28].

In principle, the sediment budget of cutbank erosion and point bar deposition should be balanced along the river if the river channel is maintaining a constant width over time [45]. Any imbalance would lead to widening (if erosion outpaces deposition) or narrowing over time (if deposition exceeds erosion). If we assume the flux of sediment into the river from erosion is balanced by the flux out of the river via deposition, then comparing the THg in sediment on eroding banks and depositing point bars reveals the net THg flux associated with riverbank erosion. This framework has been applied to OC fluxes along the Koyukuk River near Huslia [43]. With this approach, we find a net release of Hg from the floodplains to the rivers, because the full column stocks on eroding cutbanks have higher THg than aggrading point bars (figure 5).

We can also relax the assumption of equal rates of bank erosion and deposition, and instead base these fluxes on observed local migration rates from satellite imagery [44], as used to calculate the THg fluxes in figure 6. In this case, the apparently more rapid accretion of point bars at Huslia compared to eroding cutbanks leads to a calculated net deposition of THg from the river into sedimentary deposits. In contrast, high rates of cutbank erosion at Beaver (figure 6) lead to a net erosional release of THg to the river. This spatial difference—with one site exhibiting apparent net Hg erosion and the other deposition—emerges from different erosion rates, emphasizing the importance of quantifying such rates and their spatial variability for understanding biogeochemical responses in a changing Arctic. While satellite-based migration rates may capture a more accurate picture of recent changes over our study sites than assuming balanced erosion and deposition, satellite observations are inherently limited in their time and length scales [46]. Imbalances in erosion and deposition cannot be sustained indefinitely and may not hold over longer reaches of the Yukon and Koyukuk Rivers.

The fluxes per unit river length associated with bank erosion and deposition that we calculate for Huslia and Beaver are 0.6 and 0.3 kg Hg km⁻¹ yr⁻¹. The Yukon River delivers 4400 kg Hg yr⁻¹ to the Arctic Ocean [19], which would equate to 1.38 kg Hg km^–1 yr⁻¹ just based on the main stem length of ∼3200 km. While we recognize that many tributaries contribute to the Yukon (including the Koyukuk) and there are many other sources of Hg to the river across its watershed, this simple comparison reveals that the magnitude of Hg exchange between bank and bar sediments is significant in the context of Yukon River Hg transport. If exchange fluxes in our study areas (figure 6) are representative of the river as a whole, then our results imply that there could be complete exchange of particulate Hg between the river and floodplain over the length of the Yukon. Warming climate is expected to cause permafrost loss and change upland hydrological dynamics, which in turn may alter the pace of this exchange and potentially allow for erosion to outpace deposition. Given the magnitude of the floodplain exchange fluxes, such changes could lead to significant net Hg mobilization from floodplain deposits.

As we find that significant Hg is being eroded in some areas and deposited in others, understanding the extent of Hg mobilization to rivers and its impacts will depend on local sampling because monitoring at a small number of gauging stations may not capture evolving dynamics of Hg mobilization in a changing climate. For example, in the Rio Bermejo, a tributary of the Paraguay River in northern Argentina, water flows through the ∼1200 km channel in 14 d, while sediments take on average ∼8500 years [47]. During transport, sediments undergo ∼4.5 erosion-deposition events, each taking ∼1900 years [48]. In comparison, the Koyukuk spans ∼645 km and the Yukon ∼3200 km [30]. As both of these rivers are experiencing active erosion and deposition, it may take decades or longer for geochemical signals to make it to Pilot Station where most river chemistry observations are made on the Yukon River. Our results thus highlight the importance of accurately capturing the dynamics of erosion and deposition for understanding Hg transport in Arctic rivers and how they will evolve in a changing climate.